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.1004276 | Syk Signaling in Dendritic Cells Orchestrates Innate Resistance to Systemic Fungal Infection | Host protection from fungal infection is thought to ensue in part from the activity of Syk-coupled C-type lectin receptors and MyD88-coupled toll-like receptors in myeloid cells, including neutrophils, macrophages and dendritic cells (DCs). Given the multitude of cell types and receptors involved, elimination of a single pathway for fungal recognition in a cell type such as DCs, primarily known for their ability to prime T cell responses, would be expected to have little effect on innate resistance to fungal infection. Here we report that this is surprisingly not the case and that selective loss of Syk but not MyD88 in DCs abrogates innate resistance to acute systemic Candida albicans infection in mice. We show that Syk expression by DCs is necessary for IL-23p19 production in response to C. albicans, which is essential to transiently induce GM-CSF secretion by NK cells that are recruited to the site of fungal replication. NK cell-derived-GM-CSF in turn sustains the anti-microbial activity of neutrophils, the main fungicidal effectors. Thus, the activity of a single kinase in a single myeloid cell type orchestrates a complex series of molecular and cellular events that underlies innate resistance to fungal sepsis.
| Multiple cell types bearing a vast array of immune receptors with different modes of signaling ensure that the host response to infection is both robust and reliable. For this reason, loss of a single signaling pathway in a given cell type is often not enough to impact host resistance. Here, we find, surprisingly, that this is not the case in a mouse model of systemic fungal infection with Candida albicans. We show that a single kinase (Syk) in a single cell type (dendritic cells, DCs) coordinates the entire host resistance network. We highlight Syk-dependent production of IL-23p19 by DCs as the key to protection and show that IL-23p19 acts on another white blood cell type, NK cells, to specifically induce production of another mediator, GM-CSF. The latter is key for yet another cell, the neutrophil, to be mobilized into action and kill Candida organisms. This study places DCs, best known for their role in priming T cells, at the center of a cellular relay of innate immunity to fungal infection. It highlights key nodes of antifungal immunity that could be targeted in combination with antifungal drugs to provide new ways to treat patients with fungal sepsis, who generally have poor outcomes.
| Candida albicans is the most prevalent fungal pathogen in humans causing local infections of skin, nails, oral cavity and genital tract [1]. In some instances, Candida can spread systemically via the bloodstream and lodge in the kidneys, which then act as the major site of fungal replication [2]. Despite the availability of several anti-fungal drugs, invasive candidiasis still has a high mortality rate ranging from 45 to 75% [3], highlighting the need to further understand host-pathogen interactions and mechanisms of immune resistance to fungal spread.
Despite its potential pathogenicity, C. albicans generally behaves as an innocuous commensal in immunocompetent individuals because it triggers host defense pathways that keep the organism in check. Host protection from infection ultimately depends on recognition of Candida by pattern recognition receptors (PRRs) and their associated signaling pathways that initiate immunity. Many PRRs recognizing Candida are expressed by myeloid cells and belong either to the Toll-like receptor (TLR) or the C-type lectin receptor (CLR) families. A role for TLRs in anti-fungal defense was first suggested by studies in mice deficient for the TLR adaptor MyD88, which are highly susceptible to systemic candidiasis [4], [5]. However, MyD88 additionally transduces signals from IL-1 and IL-18 receptors, which can impact innate anti-fungal immunity [4], [6]–[10], and human deficiency in MyD88 does not lead to loss of resistance to fungal organisms [11]. Therefore, the role of TLRs in protection from Candida infection remains unresolved [12]–[14].
In contrast, the role of CLRs in anti-fungal defense is becoming increasingly well-established. CLRs involved in fungal recognition include Dectin-1, Dectin-2, mannose receptor, MCL and Mincle, and mice or humans deficient in some of these receptors display enhanced susceptibility to candidiasis [15]–[19]. Dectin-1, -2 and Mincle all signal via tyrosine-based motifs that recruit the spleen tyrosine kinase Syk [20]–[23], leading to an NF-κB-dependent transcriptional program via CARD9 [24]. CLR/Syk signaling additionally promotes activation of NFAT, MAP kinase and PI3 kinase (PI3K) pathways [25], [26] and can also lead to production of reactive oxygen species (ROS) and activation of inflammasomes [6]. Notably, Syk- or CARD9-deficient dendritic cells (DCs) fail to produce certain cytokines in response to Candida and fungal cell wall components [6], [21], [27] and CARD9-deficient mice are highly susceptible to systemic infection with C. albicans [24]. Likewise, human deficiency in CARD9 results in severe forms of superficial as well as invasive candidiasis [28], [29]. Thus, Syk-dependent signaling by CLRs appears an important and non-redundant pathway for anti-fungal responses. It is presently unclear whether this reflects a dominant role for Syk in a given myeloid cell type or the additive effects of PRR signaling in multiple phagocytes.
PRR signaling can trigger both innate and adaptive immune mechanisms. Adaptive immunity is initiated by DCs and important for defense against mucocutaneous candidiasis [30] but does not play a prominent role in combatting disseminated C. albicans infection [31]. Instead, innate immunity acts as the major barrier to systemic Candida spread. Indeed, the candidacidal activity of neutrophils is the key mediator of immunity to systemic candidiasis and neutropenia is a major risk factor for invasive Candida disease [31], [32]. Macrophages and inflammatory monocytes also coordinate aspects of resistance to systemic Candida spread [33]–[36] while, recently, NK cells have been shown to be crucial for promoting neutrophil candidacidal activity during experimental systemic candidiasis in mice [37]. The collaborative impact of NK cells and neutrophils is also apparent in a model of invasive Aspergillus fumigatus where co-depletion greatly decreases survival compared to neutrophil depletion alone [38]. Thus, neutrophils, monocytes/macrophages and NK cells all mediate innate resistance to fungal hematogenous spread although whether all these cell types act individually or coordinately to provide host protection and which signals are involved in regulating their activity remains unknown.
Experimental systemic candidiasis in mice mimics human candidemia in that fungal replication occurs primarily in the kidneys and resistance is mediated by neutrophils independently of T and B cells [39]. In this work, we report that the coordination of innate immunity to systemic C. albicans infection in mice is critically dependent on Syk and not MyD88 expression in CD11c+ cells. We identify the CD11c+ cells in question as DCs by ontogenetic criteria, thereby ascribing DCs a key role in innate immunity that is much less appreciated than their function in adaptive immunity. We show that this is because in the absence of Syk signaling, DCs do not produce IL-23p19 in response to C. albicans, which is necessary for fungus-driven production of GM-CSF by NK cells in the kidney. The loss of GM-CSF-producing NK cells leads to a failure to sustain the candidacidal ability of neutrophils and results in high kidney fungal burden and decreased survival to infection. Thus, effective immunity to systemic C. albicans infection involves a precise chain of sequential cellular activation events that is initiated by Syk-dependent signaling in DCs, depends on NK cells and culminates in neutrophil fungicidal activity.
To assess the relative contribution of Syk- and MyD88-dependent pathways in CD11c+ mononuclear phagocytes (predominantly DCs) to immunity during systemic C. albicans infection, we crossed Sykfl/fl [40] or MyD88fl/fl [41] strains to CD11c-Cre [42] mice to generate CD11cΔSyk and CD11cΔMyD88 lines, respectively. When CD11c+ MHC II+ cells (henceforth called DCs – see below) in the spleens and kidneys of CD11cΔSyk mice were compared to those in littermate controls (CD11cCre− Sykfl/fl), DCs from CD11cΔSyk mice displayed a marked reduction in Syk mRNA, with a PCR signal barely above that obtained for T cells, which do not express the kinase (Figure S1A). No reduction in Syk mRNA was seen in B cells (Figure S1A) and measurement of Syk protein levels by intracellular staining (Figure S1B and S1C) or Western blotting (data not shown) confirmed that the kinase was specifically deleted in CD11c+ cells. Importantly, levels of Syk were not reduced in neutrophils, indicating restriction of the deletion to the mononuclear phagocyte system (Figure S1B and S1C). Likewise, in CD11cΔMyD88 mice, a reduction in MyD88 staining was observed specifically in CD11c+ MHC-II+ DCs and not in neutrophils or other leukocytes (Figure S1D and data not shown), as reported [41].
Remarkably, CD11cΔSyk mice succumbed rapidly to systemic infection with 5×104 CFU of C. albicans when compared to controls, of which the majority survived for up to 3 weeks (Figure 1A and data not shown). At higher inoculum doses, the mortality of control animals increased [43], although, importantly, the difference in susceptibility between control and CD11cΔSyk mice was maintained (data not shown). The kidneys of infected CD11cΔSyk mice showed a large number of fungal abscesses with prominent hyphae (revealed by periodic acid Schiff (PAS) staining) heavily surrounded by leukocytes (shown by hematoxylin and eosin (H&E) staining) (Figure 1B). Consistent with these observations, fungal burden in the kidneys of CD11cΔSyk mice was around 100-fold higher than in control littermates (Figure 1C). Reflecting the massive candidemia, fungus could additionally be recovered from spleen and liver of CD11cΔSyk mice, which additionally displayed liver lipolysis (data not shown). In contrast, selective ablation of MyD88 in CD11c+ cells in CD11cΔMyD88 mice did not result in enhanced susceptibility to systemic Candida infection even though MyD88-deficient mice (lacking MyD88 in all cell types) were extremely susceptible (Figure 1C). Thus, ablation of Syk but not MyD88 in CD11c-expressing cells greatly compromises innate resistance to systemic C. albicans infection in mice and leads to death from fulminant candidiasis.
CD11c is not an exclusive marker of DCs. To narrow down the CD11c+ cell type required to express Syk in this model, we made use of recently-developed Clec9a-Cre mice in which Cre activity is restricted to cells derived from non-monocytic conventional DC precursors (CDP) [44]. Notably, despite the incomplete penetrance of Cre-mediated recombination in such precursors [44], Clec9aΔSyk were nearly as susceptible as CD11cΔSyk mice to systemic Candida infection (Figure 1D and data not shown). Because Clec9aΔSyk mice were used as homozygotes in these experiments and therefore lacked DNGR-1 expression [44], we confirmed that DNGR-1 deficiency does not impact on susceptibility to candidiasis by assessing fungal burden in infected Clec9aegfp/egfp mice [45] (Figure 1D). These data therefore suggest a key role for Syk signaling by conventional DCs. Consistent with that conclusion, all kidney DC sub-populations in CD11cΔSyk mice showed loss of Syk independently of infection (Figure S1F). We conclude that Syk expression by DCs and, possibly, additional CD11c+ cells is a key determinant of innate immunity to systemic C. albicans infection.
We assessed the composition of the leukocytic infiltrate in kidneys of infected CD11cΔSyk mice to determine if susceptibility to candidiasis correlated with loss of any particular CD11c+ phagocyte subset whose development or recruitment to the site of infection might depend on Syk. Interestingly, there was little change in the total size of the CD11c+ MHC-II+ DC compartment after infection (Figure S2A), although its relative composition was altered: in kidneys from uninfected mice, CD11bINT F4/80+ DCs were prominent whilst in infected mice this population decreased in size and a CD11b+ F4/80INT population became more abundant (Figure S2B and S2C). Importantly, despite infection-induced changes, there was no difference between control or CD11cΔSyk mice. For example, the total number of CD11c+ MHC-II+ cells was the same in the two strains and there was only a marginal difference in percentage (Figure S2A). Similarly, the change in hierarchy of CD11c+ MHC-II+ populations following infection was largely equivalent between the strains (Figure S2C). Small differences observed for the percentage but not total number of CD11c+ MHC-II+ CD11bINT F4/80+ cells (Figure S2C) might reflect changes in other leukocyte populations even though there was no obvious change in B, T or NK cells (data not shown).
In contrast to the CD11c+ mononuclear phagocyte pool, the numbers and percentages of CD11c− MHC-II− neutrophils increased greatly in the kidneys following infection in both strains (Figure 2A), as expected [43], [46]. However, CD11cΔSyk mice displayed higher levels of kidney neutrophilia, correlating with the greater fungal burden (Figure 2A). Importantly, the phenotype of neutrophils in the kidneys but not the bone marrow of infected CD11cΔSyk mice was atypical, with a large fraction of the cells expressing only low levels of CD11b and CD11a (Figure 2B, 2C and data not shown). The cells were also less granular but did not appear apoptotic or stain for active caspase 3 (data not shown). We further assessed levels of myeloperoxidase (MPO), a major constituent of azurophil granules necessary for generation of reactive oxygen species (ROS), a key component of the neutrophil killing arsenal [47]. Kidney neutrophils from infected CD11cΔSyk mice had decreased levels of MPO when compared to controls (Figure 2D). As these phenotypic differences might suggest impaired functionality [48], we assessed the ability of neutrophils in CD11cΔSyk mice to kill C. albicans. We infected control and CD11cΔSyk mice with a strain of GFP-expressing C. albicans and measured GFP signal among kidney leukocyte populations. As expected, the majority of the GFP signal was found within neutrophils (Figure 2E). However, a greater frequency of GFP+ neutrophils were present in CD11cΔSyk mice than in controls suggesting that kidney neutrophils from the former strain are impaired in their ability to destroy the fungus. To explicitly test this hypothesis, we sorted GFP+ and GFP− neutrophils from the kidney, lysed them and plated the lysates to determine C. albicans growth. This analysis confirmed that GFP+ neutrophils derived from CD11cΔSyk but not from control mice contained live C. albicans (Figure 2F). We then evaluated if neutrophils could kill C. albicans ex vivo by sorting GFP− neutrophils and incubating them with live fungus. Consistent with their phenotypic differences, neutrophils from kidneys of infected CD11cΔSyk mice showed a decreased ability to kill C. albicans ex vivo when compared to their counterparts from control infected mice (Figure 2G). In contrast, bone marrow neutrophils from either uninfected or infected CD11cΔSyk mice showed equivalent ex vivo candidacidal capacity (Figure 2G and data not shown), which argues that neutrophil impairment occurs locally at the site of infection. We conclude that in CD11cΔSyk mice infected systemically with C. albicans there is undiminished recruitment of neutrophils to the kidney but the recruited cells display phenotypic alterations and are locally impaired in their candidacidal activity.
We searched for local alterations in the inflammatory milieu of the kidney that might connect diminished neutrophil function to loss of Syk in DCs. Homogenates of kidneys from CD11cΔSyk mice showed higher levels of IL-6, KC, MIP-1α, IL-1β, TNF, IL-1α and MCP-3 at days 1 or 2 post-infection when compared to control mice (Figure S3A and S3B). This likely reflects the contribution of cell types other than DCs and macrophages as some of those cytokines are known to be produced in a Syk-dependent manner by mononuclear phagocytes in response to stimulation with Candida albicans [6], [27], [49]. Because the interpretation of the data was marred by the large differences in fungal burden between the two strains, a broader analysis was performed early after infection (16 h) when fungal burdens are more equivalent. This analysis confirmed the discrepancy in IL-6 levels between infected mouse strains while revealing that many inflammatory mediators are in fact induced to similar levels in both control and CD11cΔSyk infected mice (Figure S3C). A notable exception is GM-CSF, which was found to be selectively lost in the kidneys of infected CD11cΔSyk mice when compared to controls (Figure 3A).
GM-CSF has been reported to be important for enhancement of neutrophil maturation and neutrophil oxidative responses in both mice and man [50]–[52]. We therefore tested whether exogenous GM-CSF could decrease the susceptibility of CD11cΔSyk mice to infection. Recombinant GM-CSF administration resulted in a marked decrease in fungal burden in the kidneys of infected CD11cΔSyk mice (Figure 3B). In contrast, the same GM-CSF treatment had only a modest effect in control mice and, importantly, did not impact the hyper-susceptibility of MyD88 KO mice, demonstrating selectivity (Figure 3B). Altogether, these data suggest that the susceptibility of CD11cΔSyk mice to systemic Candida infection stems from a deficiency in GM-CSF production in the kidney, which results in failure to locally sustain neutrophil microbicidal activity.
We have recently found that kidney-infiltrating NK cells serve as a non-redundant source of GM-CSF to promote the candidacidal activity of neutrophils during systemic Candida infection [37]. Therefore, we assessed recruitment of NK cells to the kidneys of infected control and CD11cΔSyk mice and measured their production of GM-CSF and IFN-γ. There was no difference between the two strains in the percentage or the total number of kidney NK cells either before or at different times after infection (Figure 4A and data not shown). However, as early as 16 h after infection, a marked reduction was observed in CD11cΔSyk mice in both the percentage and number of GM-CSF-producing NK cells (Figure 4B). In contrast, the percentage and number of NK cells positive for IFN-γ was equivalent between the two strains (Figure 4B). The production of GM-CSF by NK cells was transient as levels of the cytokine diminished after 16 h in contrast to those of IFN-γ, which continued to greatly increase (data not shown). A similar loss of GM-CSF+ but not of IFN-γ+ NK cells was seen in infected Clec9aΔSyk mice (Figure 4C), strengthening the notion that the phenotype stems from loss of Syk in DCs.
To determine if reduced GM-CSF production by NK cells and impaired neutrophil microbicidal activity are linked, we investigated if NK cells from control mice could restore the resistance of CD11cΔSyk mice to infection. Transfer of cell preparations enriched for NK cells from naïve control mice into CD11cΔSyk mice prior to infection had no impact on fungal burden (Figure 5A). However, when the same preparations were isolated from infected control mice, fungal control was restored in subsequently infected CD11cΔSyk mice (Figure 5A). The decrease in fungal burden conferred by adoptive NK cell transfer associated with an increased proportion of CD11bhi neutrophils (Figure 5B) and restoration of the ability of neutrophils to kill C. albicans ex vivo (Figure 5C). Similar results were obtained upon transfer of pure NK cell populations that were isolated by cell sorting to exclude any confounding effect of contaminants (Figure 5D, 5E). Protection was not observed when NK cells were isolated from mice infected 48 h earlier (Figure 5D and data not shown), consistent with the notion that GM-CSF production is transient (see above). Transfer of unsorted total spleen cells was also not protective even when the cells were taken at 16 h after infection (Figure 5D, 5E and data not shown). We conclude that the susceptibility of CD11cΔSyk mice to C. albicans infection is due to a defect in GM-CSF-dependent NK cell “help” for neutrophils and can be prevented by transfer of appropriately-primed NK cells from recently-infected wild type mice but not by unprimed wild-type NK cells.
Finally, we sought to identify the signal that links Syk signaling in DCs to the production of GM-CSF by kidney NK cells. IL-23p19 has been reported to be induced very rapidly yet transiently in the kidneys and lungs of Candida-infected mice [53], [54]. In addition, IL-23p19 is important for early resistance to candidiasis [55], [56] and can be synthesized by DCs in a Syk-dependent manner upon stimulation with CLR agonists [27]. Although IL-23R can be found on both NK cells and T cells [57], we noted that NK cells but not T cells expressed IL-23R in the kidney (Figure 6A). The expression of IL-23R on kidney NK cells was upregulated in control but not CD11cΔSyk mice following infection (Figure 6B). In addition, we detected strong induction of IL-23p19 mRNA in kidney CD11c+ MHC-II+ DCs from control mice but, importantly, not from CD11cΔSyk mice early after infection with C. albicans (Figure 6C). The increase in the proportion of IL-23R+ cells (Figure 6B) may therefore be a consequence of positive feedback signaling of IL-23R in response to ligand [57], [58].
To test the significance of this observation, we infected IL-23p19 KO mice and measured NK cell production of GM-CSF. Notably, IL-23p19-deficient mice resembled CD11cΔSyk mice in that NK cells taken from the kidneys of either strain displayed markedly reduced levels of GM-CSF but not IFN-γ mRNA and did not secrete GM-CSF protein upon short-term ex vivo culture (Figure 6D and 6E). Furthermore, purified NK cells produced GM-CSF in vitro when stimulated with recombinant IL-23 but not IL-17A/F, C. albicans, curdlan or zymosan (Figure 6F and 6G). GM-CSF production by NKs in response to C. albicans occurred only in the presence of DCs derived from wild-type but not CD11cΔSyk or IL-23p19-deficient mice (Figure 6H). Finally, IL-23p19 KO mice infected with C. albicans were undistinguishable from CD11cΔSyk mice in having massively increased kidney fungal burdens (Figure 6I) that could be reversed by GM-CSF therapy (Figure 6J). Together, these data suggest that Syk-dependent IL-23p19 production by DCs in response to C. albicans acts directly on NK cells to promote GM-CSF production and subsequent resistance to systemic candidiasis.
Multiple receptors on macrophages, monocytes, neutrophils, NK cells and innate lymphocytes, as well as on non-immune cells, mediate the recognition of microbes and are thought to act co-ordinately and somewhat redundantly to provide innate resistance to infection. Here, we demonstrate that DCs, a cell type chiefly known for its ability to initiate adaptive immunity, coordinate the entire innate immune control to systemic infection with C. albicans and show that this orchestration depends on a single kinase, indicating a remarkable lack of redundancy in innate immune pathways. We further unravel a hitherto unappreciated series of cellular interactions whereby DCs provide IL-23p19 to NK cells that allows for production of GM-CSF, which in turn maintains the microbicidal activity of neutrophils, the main candidacidal effectors. Disruption of this cellular relay in CD11cΔSyk or IL-23p19 KO mice causes susceptibility to systemic candidiasis and restoration of resistance can be achieved with GM-CSF treatment. Thus, our analysis reveals Syk mediated IL-23p19 production by DCs as a central and non-redundant node of innate immunity to fungal infection and an unexpected indirect regulator of neutrophil microbicidal activity via NK cells.
Although the central function of neutrophils in innate protection from disseminated candidiasis is undisputed, the role of mononuclear phagocyte populations is not well established. It is surprising that loss of Syk from CD11c+ cells in CD11cΔSyk mice causes such a dramatic phenotype. We show that this is not because Syk uniquely regulates the development of particular CD11c+ subsets that coordinate anti-fungal immunity or even their recruitment to the site of infection, as there were no gross alterations in the composition of CD11c+ populations in kidneys from infected CD11cΔSyk mice. As in spleen and many other organs, kidney CD11c+ cells are also MHC-II+ and would therefore traditionally be defined as DC. However, not all kidney CD11c+ MHC-II+ cells are derived from committed DC precursors, leading to debate as to whether they are best classified as DCs or macrophages [59], [60]. Taking advantage of a new Clec9a-Cre line to selectively target those cells derived from pre-DC/CDP [44], we show that deletion of Syk in the DC lineage (as defined hematopoietically) phenocopies deletion in total CD11c+ cells. This would suggest that the susceptibility of CD11cΔSyk mice to systemic candidiasis is primarily due to loss of Syk from DCs. This in turn adds to the emerging notion that DCs may act as central regulators of innate immunity to infection in some instances [61]. Loss of resistance to Candida was also seen in LysMΔSyk mice (data not shown) and we do not presently exclude a possible contribution of Syk on CD11c+ cells of monocytic origin (although we note that such a result is ambiguous as LysM-Cre activity is also found on conventional non-monocytic DCs [62]). Whichever their origin, the central role of Syk in DCs suggests that ablation of CD11c+ cells should also have a dramatic phenotype on resistance to Candida infection. Surprisingly, this was not the case as CD11c-DTR (diphtheria toxin receptor) mice treated with diphtheria toxin were actually more resistant to infection with C. albicans (data not shown). This apparent discrepancy can be explained by the recent observation that ablation of CD11c+ cells in CD11c-DTR mice is accompanied by marked neutrophilia, which provides a major barrier to bacterial or, in this case, fungal infection [63].
It is notable that deletion of MyD88 in CD11c+ cells had no impact on C. albicans infection even though it markedly impacts responses to TLR agonists in vivo [41]. This may suggest a primacy of Syk-coupled rather than MyD88-coupled receptors in fungal recognition by DCs [1], [22], [64]. Nevertheless, MyD88 remains an important component of anti-fungal resistance as we find, along with Villamon et. al. [65], that MyD88-deficient animals are very susceptible to systemic C. albicans infection. Unlike that of CD11cΔSyk mice, this susceptibility is not preventable by exogenous GM-CSF therapy and presumably involves MyD88 signaling in CD11c− cells. Whether this happens downstream of TLRs or receptors for IL-1 family cytokines remains to be determined.
Depletion of neutrophils dramatically increases susceptibility of mice to experimental systemic C. albicans infection [31] and neutropenia places patients at severe risk from systemic candidiasis [32], [66]. Previous work showed that IL-6-deficient mice are highly susceptible to systemic C. albicans [67], which was attributed to a lack of neutrophil recruitment without impairment of candidacidal capacity [68]. It was therefore surprising to observe the opposite phenotype, namely normal neutrophil recruitment but impaired activity in infected CD11cΔSyk or Clec9aΔSyk mice. The ample production of neutrophil recruiting proteins such as IL-6, KC and MIP-2 (CXCL2) in the kidneys of such mice might account for unabated neutrophil recruitment. In contrast, the lack of GM-CSF in the microenvironment appears to be responsible for the loss of neutrophil activity. Neutrophil activation triggers re-localization of intracellular pools of CD11b to the plasma membrane [69] allowing for adhesion, migration and phagocytosis [48]. Thus, we suggest that the presence of CD11blo neutrophils is indicative of poorly activated cells with decreased microbicidal potential, as highlighted by our killing assays. Interestingly, GM-CSF has been linked to neutrophil functionality and survival via physical coupling of Src family kinase Lyn to the GM-CSF receptor, resulting in down-regulation of pro-apoptotic factors and up-regulation of anti-apoptotic pathways such as PI3K/Ark [70]–[72]. While we have failed to observe obvious signs of neutrophil apoptosis in CD11cΔSyk mice, we cannot exclude that any apoptotic cells might be removed rapidly and that the CD11blo phenotype is indeed a prelude to cell death. Notably, intravenous GM-CSF infusion is curative in cases of severe drug-resistant chronic mucocutaneous candidiasis [73] and patients with oral pseudomembranous candidiasis resulting from radiotherapy for head and neck cancers have been successfully treated with a GM-CSF mouthwash [74]. In addition, human neutrophil activation and survival relies in part on NK cell-derived cytokines, including GM-CSF [75], and activated human NK cells enhance neutrophil survival and promote an increase in neutrophil CD11b expression and ROS production in a GM-CSF dependent manner [76]. Thus, GM-CSF, in part derived from NK cells, may underlie resistance to Candida infection not only in mice but also in Man. This is seemingly at odds with the fact that NK cell deficiency is associated primarily with viral rather than fungal infections [77]. However, the very few NK cell-deficient individuals studied so far may not have been exposed to the conditions predisposing to systemic candidiasis such as catheter insertion or deep tissue surgery. Alternatively, NK-cell independent mechanisms may compensate in these individuals for GM-CSF-dependent fungal control. In this regard, the requirement for NK cells in antifungal immunity even in mice may vary depending on the Candida strain in question [34], [78].
We have recently shown that the functional development of NK cells in mice requires cell-intrinsic IL-17RA-mediated signals [37]. NK cells that develop in the absence of such signals are impaired in their ability to produce IFN-γ, kill target cells, as well as produce GM-CSF to control Candida infection [37]. Here, we show that even in IL-17RA-sufficient mice, where NK cell functional development is unaffected, the response of NK cells to acute Candida challenge is under stringent environmental control and requires exogenous priming signals. Priming signals for GM-CSF but not IFN-γ production in turn require Syk signaling in DCs as demonstrated by the fact that transfer of resting NK cells does not restore resistance of CD11cΔSyk mice to Candida, yet resistance is achieved if the transfer involves activated NK cells that were primed in an environment in which DCs express Syk. Together with the fact that Clec9aΔSyk and CD11cΔSyk phenocopy each other, this argues against the possibility that the defect in CD11cΔSyk mice is due to deletion of Syk in the NK cells themselves (even if a small population of NK cells can express CD11c and show evidence of Cre activity in CD11c-Cre mice [79]). Supporting this contention, purified NK cells do not respond directly to C. albicans ex vivo but will readily do so in the presence of DCs or conditioned medium from Candida-treated DC cultures. This is consistent with the notion that accessory cells, such as DCs, monocytes, and macrophages are necessary for activation of NK cells in response to most pathogens (reviewed in [80], [81]). Nevertheless, it is possible that the anti-fungal activity of primed NK cells additionally requires signaling via Syk-coupled NK cell receptors and it will be interesting to study the phenotype of mice in which Syk is selectively ablated in NK cells as opposed to DCs.
It is well known that stimulation of DCs and macrophages by C.albicans yeast and hyphae induces the production of IL-2, IL-6, IL-12, IL-23 and TNF-α in a Syk-dependent manner [21], [27]. In searching for which one of these or other factors might be responsible for priming NK cells to produce GM-CSF we focused on IL-23p19. We show that IL-23p19 is not induced in Syk-deficient DCs during systemic candidiasis and that Candida-stimulated control but not IL-23p19 KO DCs induce GM-CSF production by NK cells. We further show that IL-23p19 KO mice are very susceptible to systemic candidiasis, as previously suggested [55], but can be protected by GM-CSF treatment. Together, these data suggest that IL-23 might be the key Syk-dependent cytokine driver of DC-mediated resistance to candidiasis, consistent with the fact that addition of recombinant IL-23 and not recombinant IL-17A/F to purified NK cells induces GM-CSF production. This sheds light on a novel regulatory mechanism of cytokine production by NK cells that selectively affects GM-CSF but not IFN-γ secretion. However, IL-23 is composed of both the IL-23p19 and the IL-12/IL-23p40 subunits and it has been reported that IL-12p40-deficient mice are resistant to candidiasis [82], [83]. We have been able to reproduce this finding (unpublished observations) and therefore, at present, we are forced to conclude that the key mediator of resistance is either a novel IL-23p19-containing cytokine (including, possibly, an IL-23p19 homodimer) or that IL-23 deficiency impairs resistance to candidiasis in an IL-12-sufficient but not IL-12-deficient background. While work to assess these possibilities is ongoing, our existing data nevertheless argue for a model (Figure 7) where Syk-mediated recognition of fungal particles by DC, possibly through Syk-coupled CLRs, leads to production of an IL-23p19-containing cytokine, which acts on NK cells in the kidney to induce GM-CSF production. In turn, GM-CSF acts on recruited neutrophils to sustain microbicidal function. This unusual cellular relay from DCs to NK cells and to neutrophils via IL-23p19 and GM-CSF, respectively, provides a key axis for protection from disseminated candidiasis in mice that may be worth exploring as a possible therapeutic target in the context of fungal sepsis in humans.
All animal protocols were carried out under the authority of a UK project license (number PPL 80/2309) approved by the CRUK London Research Institute Animal Welfare and Ethical Review Body in strict accordance with UK governmental regulations (Animal Scientific Procedures Act 1986) or under protocols approved by the Veterinary office of the Canton Zürich, Switzerland (license number 184/2009 and 201/2012) in strict accordance with the guidelines of the Swiss Animal Protection Law. All efforts were made to minimize animal suffering and ensure the highest ethical and humane standards.
Control mice ‘Control’ (including C57Bl/6 and littermate CD11cCre− [42]×Sykflox/flox [40]), CD11cΔSyk (CD11cCre+×Sykflox/flox), CD11cΔMyD88 (CD11cCre+×MyD88flox/flox [41]), Control MyD88 (CD11cCre−×MyD88flox/flox), MyD88 KO [84], Clec9a(egfp/egfp) [45] (DNGR1 deficient), Clec9aΔSyk (Clec9aCre/Cre [44]×Sykflox/flox), Clec9a control (Clec9a+/+×Sykflox/flox) and IL-23p19 KO [85], [86] were crossed and bred at Cancer Research UK and at the Institute of Laboratory Animal Sciences, University of Zürich, Switzerland, in specific pathogen-free conditions.
Candida albicans strains SC5314 and CAI4-pACT1 GFP (described in [87]) were grown by agitation overnight at 30°C in yeast peptone dextrose (YPD) or synthetic complete medium (SC) containing 2% glucose and Drop-out mix without Uridine. The cells were then washed twice with PBS before use as live yeasts. Heat-killed C. albicans (HKCA) yeast or hyphae were generated by boiling samples for 1 h.
Mice aged 8–20 weeks were infected intravenously with 2×105 live C. albicans yeast unless stated. The mice were killed 2 days post-infection except where indicated and perfused with cold PBS. Mice that received GM-CSF treatment had two intraperitoneal doses of murine GM-CSF (Peprotech) 5 µg/mouse at time of infection and 24 h later. The kidneys were removed and homogenized in 1 ml PBS using an IKA T25 digital Ultra-Turrax homogenizer or a Qiagen Tissue Lyser. Serial dilutions were plated on YPD agar plates and the total number of colony forming units was calculated.
Samples were fixed in 10% Neutral Buffered Formalin and processed by the histopathology laboratory at Cancer Research UK. Samples were dehydrated with ethanol and embedded in paraffin. Periodic Acid Schiff (PAS) and hematoxylin and eosin (H&E) were used to assess fungal invasion and leukocyte infiltration respectively.
Single-cell kidney, spleen and bone marrow (BM) suspensions were prepared from PBS perfused mice. Kidneys and spleens were chopped into small pieces and digested in RPMI 1640 medium supplemented with glutamine, penicillin, streptomycin, (all from Gibco), collagenase type IV (200 u/ml, Worthington), and DNase 1 (0.2 mg/ml, Roche) for 1 h or 30 min respectively at 37°C. Cells were then passed through a 70 µm cell strainer (BD bioscience), washed with RPMI 1640 supplemented with 10% fetal calf serum, glutamine, penicillin and streptomycin (RPMI complete medium). Single cell kidney samples were then placed onto a non-continuous isotonic Percoll (GE Healthcare) gradient of 78% and 37% and centrifuged for 30 min at 550 g. The interface was collected from these samples and washed in RPMI complete medium. For isolation of BM cells, the femur and tibia were collected from both hindquarters. Bones were flushed with RPMI complete medium and passed through a 70 µm cell strainer to obtain single cell suspensions. Splenic and BM erythrocytes were lysed with Red Blood Cell Lysis Buffer (Sigma) for 3 min at room temperature (RT). Single cell populations were subsequently used for either FACS staining, in vitro candidacidal activity or cell sorting.
Data were collected on LSR Fortessa, FACSAria or LSRII (all BD Biosciences) and analyzed using FlowJo software (Tree Star). The staining protocols used combinations of antibodies listed below. Antibodies purchased from BD bioscience included: anti-CD3e (145-2C11), anti-CD4 (RM4-5), anti-CD8 (53-6.7), anti-CD11b (M1/70), anti-CD11c (HL3), anti-CD16/CD32 (2.4G2, Fc block), anti-CD19 (1D3), anti-CD45R/B220 (RA3-6B2), anti-CD49b (DX5), anti-CD64 (X54-5/7.1), anti-IFN-γ (XMG1.2), anti-Ly6G (1A8) and Streptavidin-APC. The following antibodies were purchased from eBioscience: anti-CD3e (145-2C11), anti-CD11b (M1/70), anti-CD103 (2E7), anti-GM-CSF (MP1-22E9), anti-MHC-II (M5/114.15.2) and anti-NK1.1 (PK136). The following antibodies were purchased from Biolegend: anti-CD11a (M17/4), anti-CD11c (N418), anti-CD18 (M18/2), anti-CD45.2 (104), anti-F4/80 (BM8), anti-Syk (5F3, purified and conjugated to AF647 with AF647 Antibody labelling kit (Molecular Probes)). Additional antibodies used were Polyclonal Goat anti-Mouse/Rat MyD88 (R&D), anti-Mouse IL-23R (753317, R&D), Rabbit anti-Goat AF488 (Molecular Probes) and anti-MPO (8F4, biotinylated, Hycult Biotech).
Single cell suspensions were surface stained directly ex vivo or following 7 h incubation with brefeldin A (Sigma). Most cell staining involved dead cell exclusion by live/dead fixable violet dye (Invitrogen), followed by fixation with 2% paraformaldehyde (Electron Microscopy Sciences) for 20 min at RT. Cells were washed twice with FACS buffer (PBS, 1% FCS, 2 mM EDTA). For intracellular staining, cells were subsequently permeabilised and stained with saponin containing Reagent B (ADG Bio Research GMBH).
Kidneys were removed 1 or 2 days post-infection following PBS perfusion and homogenized on ice in 0.5 or 1 ml of PBS respectively. Chemokines and cytokines from homogenates and cell culture supernatants were analyzed according to manufacturer's instructions. Briefly, clarified samples were incubated with either BD cytometric bead array kits (IL-6, KC, MIP-1α, TNF, IL-1α), FlowCytomix Kits (IL-15/IL-15R, MCP-3 and IL-10), R&D Quantikine ELISA kit (IL-1β) or eBioscience Ready-Set-Go ELISA kit (GM-CSF). Bead based assays were assessed using a LSR Fortessa whilst ELISA samples were read at 450 nm with all concentrations determined relative to a standard curve.
For proteome profiling, kidneys were removed from naïve or 16 h post-infection mice following PBS perfusion and homogenized in 1 ml PBS with protease inhibitor cocktail (cOmplete Roche) with Triton ×100 added at a final concentration of 1% prior to a freeze thaw step. Samples were clarified prior to addition to the R&D Proteome profiler (Mouse cytokine array panel A) and developed as per manufacturer's instructions. Relative pixel density of each duplicate blot was measured using Image J software. The data are presented as fold change in signal from infected samples compared to naïve control samples.
Single cell suspensions of kidney and bone marrow were prepared as above prior to staining and sorting for neutrophils, identified by CD11b+ Ly-6G+ F4/80− and GFP+ or GFP−. Sorted neutrophils were incubated with C. albicans (10∶1) in serum free medium on ultra low attachment plates (Costar) for 1 h at 37°C. Wells were collected and cells lysed with water prior to plating on YPD agar. C. albicans colony formation from neutrophil-containing wells was compared to that from control neutrophil-free wells to calculate the percentage of C. albicans killed. Data are combined from three independent experiments with each data point representing an individual well. Neutrophils were also isolated from mouse blood using a density gradient of Histopaque 1119 and Histopaque 1077 (both Sigma). Blood neutrophil killing activity was assessed using 104 C. albicans yeast co-cultured with 104 neutrophils (usually >80% Ly6G+) in protein low binding tubes (Sarstedt) for 2 h. The percentage of C. albicans killed was assessed as above with data combined from two independent experiments.
Single cell suspensions of spleens from control mice were obtained as described above and NK cells were either enriched with anti-DX5 microbeads (Miltenyi Biotech) or purified by FACS based on DX5 and CD3 expression. 8×106 enriched NK cells or 4×106 FACS purified NK cells (>95% pure and viable) were adoptively transferred into recipient mice 1 h prior to infection.
DCs were differentiated from BM precursors in presence of GM-CSF for 7 days. 105 FACS-purified NK cells from naïve spleens were cultured alone or co-cultured with 5×104 DCs in presence of heat-killed C. albicans (10 M.O.I.), 100 µg/ml Curdlan, 50 µg/ml Zymosan, recombinant IL-23 (BD bioscience; 100 ng/ml) or recombinant IL-17A/F heterodimer (BD bioscience; 1 µg/ml). The culture supernatant was collected after overnight incubation and GM-CSF was quantified by ELISA (eBioscience) according to manufacturer's instructions.
RNA was extracted from whole organs disrupted using the Tissue Ruptor (Qiagen) using TRIzol (Invitrogen) according to manufacturer's instructions. RNA from FACS sorted cell samples was isolated using either TRIzol or the QIAcube (Qiagen). Isolated RNA was reverse transcribed into complementary DNA using random primers (Invitrogen). Quantitative PCR was performed using Taqman primer/probe sets (Invitrogen), Sykb (Mm01333035_m1 (exon boundary 1–2)), csf2 (Mm00438328_m1), ifng (Mm01168134_m1) and house keeping Gapdh (Mm99999915_g1) or SYBR Green (Qiagen) with primer pairs il23a (F-GCCAAGAAGACCATTCCCGA R-TCAGTGCTACAATCTTCTTCAGAGGACA) and Gapdh (F-CAGTATTCCACTCTGAAGAAC R-ATACGGCCAAATCTGAAAGAC) using either the Viia7 or 7500 Fast Real-Time PCR System (Applied biosystems).
Prism version 6a (GraphPad) was used for plotting data and for statistical analysis. Survival data are presented as a Kaplan-Meier plot with a log rank test used to compare significance between groups. Data was subjected to D'Agostino & Pearson omnibus normality test, Shapiro-Wilk normality test and Kolmogorov-Smirnov test to determine the subsequent statistical tests applied. Statistical significance of differences between two groups or groups with fewer than three samples was determined by 2-tailed t test. For experiments with more than 2 groups, significance of any differences was determined using a 1-way ANOVA with Tukey multiple comparison of all pairs for post-test analysis. If the data was assessed to be non-gaussian then a Kruskal-Wallis with Dunn's multiple comparison test was undertaken. The level of significance was defined as *p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001. The test used for statistical analysis is indicated in each figure legend.
|
10.1371/journal.ppat.1003700 | The Microvesicle Component of HIV-1 Inocula Modulates Dendritic Cell Infection and Maturation and Enhances Adhesion to and Activation of T Lymphocytes | HIV-1 is taken up by immature monocyte derived dendritic cells (iMDDCs) into tetraspanin rich caves from which the virus can either be transferred to T lymphocytes or enter into endosomes resulting in degradation. HIV-1 binding and fusion with the DC membrane results in low level de novo infection that can also be transferred to T lymphocytes at a later stage. We have previously reported that HIV-1 can induce partial maturation of iMDDCs at both stages of trafficking. Here we show that CD45+ microvesicles (MV) which contaminate purified HIV-1 inocula due to similar size and density, affect DC maturation, de novo HIV-1 infection and transfer to T lymphocytes. Comparing iMDDCs infected with CD45-depleted HIV-1BaL or matched non-depleted preparations, the presence of CD45+ MVs was shown to enhance DC maturation and ICAM-1 (CD54) expression, which is involved in DC∶T lymphocyte interactions, while restricting HIV-1 infection of MDDCs. Furthermore, in the DC culture HIV-1 infected (p24+) MDDCs were more mature than bystander cells. Depletion of MVs from the HIV-1 inoculum markedly inhibited DC∶T lymphocyte clustering and the induction of alloproliferation as well as limiting HIV-1 transfer from DCs to T lymphocytes. The effects of MV depletion on these functions were reversed by the re-addition of purified MVs from activated but not non-activated SUPT1.CCR5-CL.30 or primary T cells. Analysis of the protein complement of these MVs and of these HIV-1 inocula before and after MV depletion showed that Heat Shock Proteins (HSPs) and nef were the likely DC maturation candidates. Recombinant HSP90α and β and nef all induced DC maturation and ICAM-1 expression, greater when combined. These results suggest that MVs contaminating HIV-1 released from infected T lymphocytes may be biologically important, especially in enhancing T cell activation, during uptake by DCs in vitro and in vivo, particularly as MVs have been detected in the circulation of HIV-1 infected subjects.
| Dendritic cells (DCs) are vital for immune recognition of pathogens as they capture, internalise, degrade and present foreign peptides to T lymphocytes. It is thought that HIV-1 hijacks the DCs functions, such as migration and maturation, to increase contact with the major target cell CD4+ T lymphocytes leading to dissemination throughout the body. Currently there is still some controversy over the ability of HIV-1 to infect and mature DCs, which may be due to differences in the inoculum used. Here we examined the effect of contaminating microvesicles (MVs) identified in HIV-1 preparations on HIV-1 modulation of DC function. We show that when MVs are present with HIV-1, the inoculum induces greater DC maturation and adhesion probably via cellular HSP90α and β and viral nef within the MVs. The functional consequences are reduced de novo replication of HIV-1 but increased clustering with T lymphocytes, resulting in increased T lymphocyte alloproliferation and HIV-1 transfer. As MVs are produced in HIV-1 susceptible cells and would be present in vivo due to HIV-1 induced cell death and hence are physiologically relevant, these results also indicate that MVs present in HIV-1 inocula should be considered when assessing HIV∶DC interactions.
| Dendritic cells (DC), located throughout the body, but in particular in the male foreskin and the anogenital and cervical mucosa, are susceptible to HIV-1 infection [1]–[3]. They are also able to transfer HIV-1 to T lymphocytes, resulting in viral dissemination [4]. This ability to transfer HIV-1 to T lymphocytes is related to the function of DCs as antigen presenting cells, with the efficiency depending on the functional state of the cells. Immature DCs are highly endocytic; express low levels of major histocompatibility complex (MHC)-I and MHC-II complexes and co-stimulatory molecules but are considered to be poor processor and presenters to T lymphocytes. Mature DCs however are characterised by an up-regulation of surface expression of co-stimulatory molecules CD40, CD80 and CD86 and de novo expression of CD83 enabling efficient antigen presentation to and activation of T lymphocytes [5], [6]. Productive HIV-1 infection of T lymphocytes is much more efficient in activated cells than resting T lymphocytes [7], [8]. Therefore maturation of DCs following HIV-1 uptake is likely to be a key event in cell to cell spread. Whilst one group has found that HIV-1 does not induce maturation in either infected cells or uninfected bystanders [9], our group and others showed that HIV-1 induces a partial maturation of DCs [10]–[12]. Maturation genes were differentially expressed to a greater degree in cells treated with viable compared to non-viable virus indicating a role for replication over entry. Furthermore, this partial maturation was shown to be a result of p38 MAPK signalling [12]. The ability of DCs to form clusters with T lymphocytes and the subsequent formation of an immunological synapse is also important in HIV-1 transmission and is dependent on the interaction between ICAM-1 (CD54) on DCs and LFA-1 on CD4+ T lymphocytes [13] as evidenced by impaired DC mediated HIV-1 transmission to T lymphocytes in patients who lack LFA-1 on leukocytes [14]. Hence, the ability of HIV-1 to up-regulate the expression of adhesion and co-stimulatory molecules on DCs as part of their maturation may aid viral transfer to T lymphocytes and replication. HIV-1 transfer from DCs to T lymphocytes occurs within a viral synapse. Viral synapses are characterised by segregated supramolecular structures with a central cluster of envelope surrounded by a ring of ICAM-1 [15]–[17] to induce DC∶T lymphocyte adherence and allow for viral transfer.
Both immature and mature DCs are able to transfer HIV-1 to T lymphocytes by different mechanisms during 2 sequential phases of uptake, the first following vesicular uptake and the second after de novo infection of DCs transfer [18]–[21]. However mature DCs formed DC∶T lymphocyte conjugates more readily than immature DCs even in the absence of specific antigen [22]. An additional factor in HIV-1 transfer is the role of DC-derived microvesicle (MV)-associated HIV-1 particles. Protein-laden MVs are released from many different cell types, including the HIV-1 target cells T lymphocytes and DCs. A significant fraction of these vesicles are similar in size and density to HIV-1 virions, such that they sediment with HIV-1 even when vigorous methods to purify the virus are used [23]. In sucrose density gradient purified HIV inocula from PBMCs the ratio of virions to MVs is 1∶1–2 [24]. Vesicles and HIV-1 virions have similar, but not identical, protein compositions [23]–[27]. These vesicles are primarily shedding microvesicles (SMV) released from the cell surface, exosomes released from late endosomes/multivesicular bodies by exocytosis or apoptotic blebs (AB)s from dying cells, reviewed in [28]. The signalling molecule CD45 is expressed on all hemopoietic cells and is present in high concentrations on the MVs but specifically excluded from HIV-1 virions. It can therefore be used to purify viral stocks by removing contaminating vesicles [29]–[31]. Furthermore, the removal of vesicles by CD45-depletion is confirmed by the removal of other cellular proteins from the HIV-1 preparation, such as HLA-DR and actin [25].
In this study we have investigated the relative roles of HIV-1 and associated MVs on the maturation status of DCs and viral transfer to T lymphocytes using HIV-1 inocula containing or stripped of MVs. We have shown that MVs present within an HIV-1 inoculum enhance MDDC maturation which in turn induces more DC∶T lymphocyte clusters and results in higher levels of HIV-1 transfer to T lymphocytes. This induced maturation limits the productive infection of monocyte-derived DCs (MDDC). We propose that these MVs play a role in cell to cell spread of HIV-1 in vitro and possibly in vivo.
Previous work from our laboratory showed partial maturation of dendritic cells in response to HIV-1BaL, however we were unable to rule out the role of MVs in non-viral replication dependent DC maturation [10]. Therefore HIV-1BaL viral stocks with or without the CD45+ MVs depleted were prepared as outlined by Bess et al [23]. The SUPT1-CCR5-CL.30 cell line was infected with VSV-G pseudotyped HIV-1pBaL, the CD45 contaminating MVs were removed with microbeads and the virus preparation was concentrated and purified further by centrifugation. The total HIV-1 p24 concentration as determined by ELISA was comparable for the HIV-1BaL(pellet) and HIV-1BaL(CD45−) viruses. Western blot analysis confirmed successful depletion of CD45 from the HIV-1BaL(CD45−) stock.
MDDCs were infected with either HIV-1BaL(CD45−) or HIV-1BaL(pellet) at a MOI of 3 measured as TCID50. Levels of productively infected cells were assessed by flow cytometry for intracellular HIV-1 p24 antigen at 48 and 120 hours post infection (Figure 1A and B). By 24 hours post infection, uptake of HIV-1 (p24 antigen) into vesicular/cave compartments of DCs have dissipated [32]. Depletion of CD45+ MVs increased the infection of MDDCs at both 48 and 120 hours (4.75+/−1.5% SEM, p<0.01 and 20.46+/−11% SEM, p<0.05 respectively). The level of HIV-1 proviral DNA was determined by Q-PCR and confirmed the flow cytometry infection values with increased infection of MDDCs at 120 hours when MVs were depleted (4.8+/−2% SEM at 48 hours and 22.7+/−5.1% SEM at 120 hours, p<0.05, Figure 1C). These results indicated that the CD45+ MVs limit the ability of the HIV-1 stock to productively infect DCs. This effect was not mediated via SAMHD1 as levels were similar in MDDCs treated with MVs from activated and non-activated CD4+ lymphocytes by western blot and densitometry (ratios of 0.9 and 0.9 respectively, compared to mock treated MDDCs; Figure S1)
The susceptibility of DCs to become productively infected with HIV-1 is related to the maturation status of the DCs [18]. We therefore hypothesised that the increased infection rate in MDDCs infected with HIV-1BaL(CD45−) virus was a consequence of lower maturation in these DCs compared to those infected with HIV-1BaL(pellet). In order to show this, DCs were infected with HIV-1BaL(pellet) or HIV-1BaL(CD45−) virus for 48 hours and the level of adhesion, co-stimulatory and other maturation markers was determined by flow cytometry and confirmed by q-PCR (data not shown). In agreement with our previously published work, HIV-1BaL(pellet) induced partial up-regulation of CD80, CD83, CD86 and HLA-DR as well as a partial down regulation of DC-SIGN and MR compared to a standard mixture of maturation inducing cytokines and prostaglandins (Figure 2A and B). In view of the importance of ICAM-1 in DC-T lymphocyte adhesion through viral synapses [33] its expression following HIV-1BaL(pellet) treatment was also examined and found to be elevated above that induced by the maturation mix. In contrast, the removal of the CD45+ MVs resulted in a failure to up-regulate surface expression of ICAM-1 and maturation markers on DCs including CD80, CD83, CD86 and HLA-DR, as well as a lack of down-regulation of CLRs compared to HIV-1BaL(pellet) treated cells (Figure 2B), indicating that MVs in the HIV-1 preparation induce some of the partial maturation.
To assess the kinetics of ICAM-1 up-regulation, the protein expression was determined at 0, 16, 24, 48 and 96 hours post infection by flow cytometry. We found that while ICAM-1 expression peaked at 16 hours post infection, the expression was still above baseline levels at 48 hours post infection (Figure 2C).
When the MDDC population was divided into the infected (p24+) and exposed but uninfected (p24−) populations it became apparent that within the HIV-1BaL(CD45−) treated cells, maturation was restricted to the p24+ productively infected cells and very little bystander maturation of p24− cells was seen. This contrasted with the HIV-1BaL(pellet) treated DCs which showed increased CD83, CD86 and ICAM-1 expression in both infected and bystander cells (Figure 2D). Thus removal of MVs from HIV-1 preparations results in markedly reduced DC maturation and ICAM-1 expression.
To confirm that the differences induced by HIV-1BaL(pellet) compared to HIV-1BaL(CD45−) virus were a result of the MVs that were removed, we next devised a MV replacement experiment. The bead bound CD45+ MVs could not be used for these replacement assays as the CD45 microbeads caused DC maturation (data not shown). Additionally, it was not possible to remove the CD45 beads from the depleted fraction without damaging the MVs. We therefore generated MVs from SUPT1.CCR5-CL.30 cells used for propagation of HIV-1 stocks by stimulation with CD3 and CD28 antibodies for 3 days. The supernatant was subsequently concentrated as for HIV-1BaL virus stock generation. This method of activating SUPT1.CCR5-CL.30 cells resulted in similar cytopathic phenotypic changes to those cells infected with HIV-1BaL. MVs from untreated SUPT1.CCR5-CL.30 cells had no effect on HIV-1 induced DC maturation or ICAM-1 expression (Figure 3A).
Once MV preparations were generated they were added back to the HIV-1BaL(CD45−) virus stock and then used to infect MDDCs. HIV-1BaL(pellet) was used as a positive control and the amount of MVs added was matched to the CD45 concentration of the HIV-1BaL(pellet) virus (determined by western blot densitometry, Table 1). In these experiments, the addition of isolated MVs to the HIV-1BaL(CD45−) virus resulted in increased expression of adhesion, co-stimulatory, maturation markers and CLRs on MDDCs while MVs alone did not induce maturation (Figure 3A). There was a concentration dependent effect (Figure 3B) as determined by CD83 and CD86 expression (r = 0.9 and r = 0.87 respectively).
In addition to the enhanced maturation, adding back MVs resulted in significantly less HIV-1 infection of MDDCs at 120 hours post infection (3.72+/−1.1% SEM, p = 0.05, Figure 3B and C). These results were further evidence that the maturation status of DCs and level of infection closely correlate (CD83 expression versus p24; r = −0.99).
In contrast to MVs from activated SUPT1.CCR5-CL30 cells, MVs from primary CD4+ lymphocytes activated by anti-CD3 and anti-CD28 antibodies induced significant maturation of MDDCs alone, without accompanying HIV-1BaL(CD45−) (Figure 3D).
In addition, exposure of primary blood myeloid BDCA1+ DCs to HIV-1BaL(pellet) showed enhanced maturation (up-regulated CD83) but this was diminished when they were exposed to HIV-1BaL(CD45−) and restored when MVs from activated SUPT1.CCR5-CL.30 cells were added to HIV-1BaL(CD45−) (Figure 3E).
ICAM-1 was markedly up-regulated in the MDDCs infected with HIV-1BaL(pellet) compared to maturation mix or HIV-1BaL(CD45−) virus (Figure 2B). This was further supported by investigating the expression in the infected populations and showing increased expression in p24+ MDDCs compared to p24− MDDCs for both virus stocks (Figure 2D). The functional effects of increased ICAM-1 expression include increased and more stable DC∶T lymphocyte conjugates so we examined this by developing a DC∶T lymphocyte clustering assay using flow cytometry. Optimal conditions were obtained by co-culturing immature or mature MDDCs with autologous CD4+ T lymphocytes for 45 minutes at a ratio of 5 T lymphocytes∶1 DC. The cells were then labelled with CD1a (DC marker) and CD3 (T lymphocyte marker) antibodies. Within the DC population the proportion of cells that were also positive for CD3 were considered to be DC∶T lymphocyte clusters (gating strategy Figure 4A and B). This resulted in 44.5+/−2.9% SEM of mature DCs forming detectable clusters with T lymphocytes compared to 13.2+/−1.3% SEM of immature DCs forming clusters with T lymphocytes (Figure 4B). Results were validated by comparison between mature DCs involved in clusters by flow cytometry and by counting of those in contact with T lymphocytes using confocal microscopy (correlation >90%, data not shown).
DCs were then infected with either HIV-1BaL(pellet) or HIV-1BaL(CD45−) virus for 48 hours and co-cultured with autologous CD4+ T lymphocytes. The percentage of DCs that clustered with CD4+ T lymphocytes was significantly different between the two virus stocks (p<0.02). The HIV-1BaL(CD45−) treated DCs, which showed very little up-regulation of co-stimulatory markers or of ICAM-1, had levels of clustering with CD4+ lymphocytes (17.2+/−2.9% SEM) similar to the untreated negative control DCs. In contrast, the HIV-1BaL(pellet) treated DCs with higher ICAM-1 and co-stimulatory marker expression showed an increased cluster formation with CD4+ T lymphocytes (30.44+/−4.8% SEM), similar to mature DCs (Figure 4C).
After defining infected and bystander MDDCs separately, significantly higher clustering was seen in the p24+ DCs than in p24− DCs for HIV-1BaL(pellet) (Figure 4D). Finally, blocking experiments with an antibody to ICAM-1 which reduced clustering to baseline levels, indicated that the elevated clustering seen with HIV-1BaL(pellet) treated MDDCs was due to ICAM-1 expression on DCs (Figure 4E). Thus, elevated ICAM-1 expression on HIV-1BaL(pellet) treated MDDCs results in greater DC∶T lymphocyte cluster formation than in HIV-1BaL(CD45−) treated MDDC.
Next, the effects of differential DC maturation by the two HIV-1BaL stocks upon T lymphocyte proliferation were examined. MDDCs were exposed to HIV-1BaL(pellet) or HIV-1BaL(CD45−) virus for 48 hours and then added to CFSE labelled allogeneic PBMCs from HIV-1 seronegative subjects at a ratio of 1 DC∶10 PBMC and co-cultured for a further 5 days. Following culture, cells were assessed for proliferation by flow cytometry. CD4+ T lymphocyte and MDDC populations were identified by size gating and the level of proliferation was assessed by CFSE dilution (Figure 5A). The percentage proliferation for each treatment was determined by normalising the CFSE dilution percentage to the percentage of live cells for both the PBMC and the MDDC populations. The values were then added to account for all CD4+ T lymphocytes present (Figure 5B). Immature and mature MDDCs co-cultured with CFSE labelled PBMC were used as negative and positive controls respectively (7.1+/−1.5% SEM, 10.7+/−0.9% SEM) (Figure 5B). HIV-1BaL(pellet) exposed DCs induced greater proliferation (36.2+/−7.2% SEM) of CD4+ T lymphocytes compared to mature DCs, whilst HIV-1BaL(CD45−) exposed DCs were significantly impaired in their capacity to induce T lymphocyte proliferation (4.98+/−0.5% SEM) compared to mature and HIV-1BaL(pellet) exposed DCs (p = 0.01 and p = 0.05 respectively).
As DCs are very efficient at transferring HIV-1 to T lymphocytes, we investigated the capacity of MDDC infected with the two different virus stocks to induce T lymphocyte activation and to transmit virus to T lymphocytes. MDDCs were infected with either HIV-1BaL(pellet) or HIV-1BaL(CD45−) virus for 48 hours and added to resting autologous CD4+ T lymphocytes for up to 96 hours. The levels of CD69 expression on CD4+ T lymphocytes, as a marker for T lymphocyte activation (Figure 6A), and the percentage of p24+ T lymphocytes (Figure 6B) was assessed over time. Consistent with their effects on DC maturation, only the HIV-1BaL(pellet) exposed DCs were able to induce activation of CD4+ T lymphocytes at 24 and 48 hours post co-culture.
CD4+ lymphocytes co-cultured with HIV-1BaL(pellet) exposed MDDCs resulted in higher and more rapid kinetics of infected CD4+ lymphocytes than when co-cultured with HIV-1BaL(CD45−) exposed MDDCs. Thus suggesting that DC maturation induced by MVs as well as virus in HIV-1BaL preparations acts to enhance HIV-1 transmission.
We next characterised the MVs in the HIV-1 and activated SUPT1.CCR5-CL.30 stocks. Initially, key candidate proteins present in MVs from the HIV-1BaL(pellet) and the MVs obtained from CD3/CD28 activated and non-activated SUPT1.CCR5-CL.30 cells were investigated by Western blot and compared to HIV-1BaL(CD45−), parent SUPT1.CCR5-CL.30 cells as well as the anti-CD45 bead bound HIV-1BaL preparation (Figure 7 and Table 2). Both HIV-1 and MV preparations were positive for the tetraspanin/exosome marker CD81 but CD9 and Alix were undetectable. In addition, HIV-1BaL(CD45−), HIV-1BaL(pellet) and activated SUPT1.CCR5-CL.30 MVs contained the histone marker H2A indicative of ABs. Heat shock proteins (HSP) 90α and 90β (AB1) were also detected in both the viral and MV preparations (Figure 7).
To fully characterize the protein complement of the MVs from the two HIV-1 inocula and also the anti-CD45 bead bound material compared to those from activated and non-activated SUPT1.CCR5-CL.30 cells we separated the protein bands by gel electrophoresis and subjected them to Tandem Mass Spectrometry. After filtering for non-human contaminating proteins present in the fetal calf serum, 266 proteins with >0.01% share of the total were detected in HIV-1BaL(pellet), 255 in HIV-1BaL(CD45−), 274 in the anti-CD45 bead bound preparation, 462 in MVs from activated SUPT1.CCR5-CL.30 and 143 from non-activated SUPT1.CCR5-CL.30 MVs. The key proteins characteristic of MVs which were identified are shown in Tables 3 and 4. These included α-enolase, pyruvate kinase, actin and actin related proteins, the tetraspanins, CD9 and CD81, and several heat shock proteins especially HSP90α and β. HSP90α and HSP90β, were the only cellular proteins present in concentrations of >0.01% which were likely to act as DC maturation stimuli in HIV-1BaL(pellet) and activated SUPT1.CCR5-CL.30 MVs and were present at high concentrations in HIV-1BaL(pellet) (0.32% and 0.20% of peptide spectrum IDs respectively) and in the activated SUPT1.CCR5-CL.30 MVs (0.40% and 0.25% respectively) (Table 4). They were also in the top 20 of the 200–400 identified proteins. HSP90α and β were present at 3 fold and 6 fold higher concentrations respectively in MVs from activated than in non-activated uninfected SUPT1.CCR5-CL.30 cells, By mass spectrometry there was little difference in HSP90α concentrations in HIV-1BaL(pellet) and HIV-1BaL(CD45−) but HSP90β appeared to be present at lower levels in HIV-1BaL(pellet) (0.20%) than after depletion of MVs with CD45 antibody bound beads (0.63%) (Table 4), although this was not supported by western blot and densitometry (Figure 7). Here both HSP90α and β were evenly distributed between HIV-1BaL(pellet) and HIV-1BaL(CD45−) inocula and both were present on CD45+ beads. However both were markedly increased, especially HSP90α in MVs from activated verses non-activated SupT1 cells. HSP70/71 were present at very low concentrations in all preparations (by MS).
The HIV-1 accessory protein nef was present in HIV-1BaL(pellet) (at lower levels than gag or env), but completed depleted from HIV-1BaL(CD45−) virus. However, Gag/pol and env peptides were concentrated 1.5 to 3 fold in HIV-1BaL(CD45−) compared to HIV-1BaL(pellet) (Table 4). Direct comparison of the levels of HIV-1 proteins concentrated on the anti-CD45 beads showed a rank order of gag/pol (mean 2.16% share), env (mean 0.12%), nef (mean 0.033%), vpr (0.015%).
To directly examine the effect of HSPs 90α and β and nef on MDDC maturation, graded concentrations of recombinant HSP 90α (Figure S2) and 90β and of nef were added to MDDCs (Figure 8A and B). Each alone induced maturation and ICAM-1 expression in a concentration dependent fashion. HSP90β was more potent than HSP90α (minimal effective concentrations 2.5–5 nM, significant at 25 nM for HSP90β and for HSP 90α effective at 25 nM, significant at 50 nM). The lowest effective concentration of nef was 10 nM (Figure 8B).
The combination of recombinant nef at its lowest effective concentration (10 nM) with HSP90β at a concentration (0.5 nM) tenfold below its minimum effective concentration (5 nM) enhanced maturation above nef alone, especially for CD80 (Figure 8C). Thus, MVs contain viral and cellular proteins that are capable of potent stimulation of DC maturation and ICAM-1 expression.
Purification of HIV-1 preparations released from infected T lymphocytes by removal of the contaminating MVs can be achieved by CD45 depletion [23], [25], [30]. In this paper we have described experiments using HIV-1 inocula with and without depletion of CD45+ MVs and then restoration of MVs from activated uninfected SUPT1.CCR5-CL.30 or primary T cells to investigate the biological role of MVs in HIV-1 uptake by and infection of MDDCs. When compared, we observed that HIV-1BaL depleted of MVs infected DCs to a greater extent than HIV-1BaL containing MVs. However, HIV-1 depleted of MVs induced much less MDDC maturation, ICAM-1 expression and clustering with CD4+ T lymphocytes than HIV-1 containing MVs. This should also be applicable in the tissue microenvironment in vivo.
Several HIV-1 restriction factors have been described for DCs, including the constitutively expressed APOBEC3G [34], SAMHD1 [35] and the HIV induced interferon stimulated genes [36]. Our finding that MVs within the viral inocula restrict HIV-1 infection of DCs may reflect a further mechanism by which DCs are able to control viral infection, or more importantly, their induction of partial maturation may facilitate viral transfer to CD4+ lymphocytes. It has been well documented that immature MDDCs are more HIV-1 susceptible than mature MDDCs [37]–[40] due to differences in CCR5 expression which is required for HIV-1 fusion [41] as well as reverse transcription and/or post-integration blocks [42], [43]. The presence of MVs in the inoculums resulted in increased maturation (but not SAMHD1 levels) of both productively infected and bystander DCs. This enhanced maturation, particularly in the bystander cells may explain the lower levels of HIV-1 infection in DCs infected with HIV-1BaL(pellet).
In this study we show that productively infected DCs are more mature than exposed bystander cells, confirming our previous findings that the maturation was a result of HIV-1 replication, as well as exposure [10], [11]. The role of HIV-1 replication in DC maturation was strengthened by the observation that HIV-1BaL(CD45−) productively infected DCs exhibited some maturation. MVs present in the HIV-1BaL(pellet) preparation must have a role in uninfected bystander DC maturation as their removal eliminated this effect. The greater maturation seen in the productively infected DCs with HIV-1BaL(pellet) compared to HIV-1BaL(CD45−) suggested an additive effect from the MVs. The use of two different HIV-1BaL inocula may contribute to an understanding of the different routes of entry of HIV-1 into DC. At early time points post infection HIV-1 and MVs induce maturation of infected DC as well as uninfected bystander DC due to vesicular uptake (into ‘caves’ and late endosomes). At later time points (>48 hpi) however, the effects of replicating HIV-1 on DC maturation may dominate. It is of note that we have assumed in this paper that HIV-1 p24− DCs were uninfected. This is probably true of the majority of DCs, although a small proportion of these cells will be in the early stages of infection but do not yet express HIV-1 DNA+ p24 antigen at detectable levels. Furthermore while CD14+ monocytes are commonly used as a source of in vitro MDDCs, the use of MDDCs in this study is strengthened by the observations that in vivo equivalent cells in both mice [44], [45] and recently in humans can be generated during physiological stress [46], [47]. Nevertheless we also found similar effects of HIV-1 inocula with and without MVs on primary blood BDCA1+ myeloid DCs.
The adhesion molecule ICAM-1 was investigated due to its role in immunological and viral synapse formation [48] and was shown to be expressed to a greater extent than other maturation and co-stimulatory markers examined by HIV-1BaL(pellet) compared to matured DCs. This supports the findings that DC subsets expressing higher levels of ICAM-1 transmit HIV-1 more efficiently [49]. The functional consequences of DC maturation and ICAM-1 expression were assessed using a novel flow cytometry based assay to assess DC∶T lymphocyte clustering interactions. The results showed that HIV-1BaL(pellet) treated DCs cluster with T lymphocytes to a greater extent than with immature DCs although less than matured DCs. Furthermore productively infected cells cluster more than uninfected bystander DCs which were exposed to the inoculum. Previous work has shown that HIV-1 viral protein nef increases the capacity of DCs to form clusters with allogeneic CD4+ T lymphocytes which increased immunological synapse formation [50], [51] and blocking ICAM-1 has been shown to decrease HIV-1 transfer [13], [14]
As well as different levels of DC maturation seen when cells were treated with HIV-1BaL(pellet) or HIV-1BaL(CD45−), corresponding rates of T lymphocyte proliferation and HIV-1 transfer were observed. DCs treated with HIV-1BaL(pellet) led to greater CD4+ and CD8+ T lymphocyte proliferation than those treated with HIV-1BaL(CD45−). In addition, HIV-1BaL(pellet) led to transfer of HIV-1 to both activated and resting CD4+ T lymphocytes, while HIV-1BaL(CD45−) was unable to activate contacting CD4+ T lymphocytes and was only able to be transferred to activated CD4+ T lymphocytes, probably due to decreased ICAM-1 expression and clustering. This indicated a significant role for infected and especially bystander DC maturation due to presence of MVs, on HIV-1 transfer. These results together are important as, until now, it has not been possible to determine the individual roles of maturation and infection on the spread of HIV-1 from DCs to T lymphocytes.
The viral inoculum in vivo (e.g. in semen or blood) is derived from lysed infected CD4+ lymphocytes and therefore likely to contain MVs, so this is a more physiologically relevant viral inoculum. HIV-1 infected activated T lymphocytes burst release HIV-1 [52], [53] which also releases cell debris and MVs. Such MVs containing HIV-nef have been identified and quantified in vivo in (ultracentrifuged) plasma from HIV infected [54], [55].
Our results show that in vitro a low proportion of HIV-1 infected DCs, if accompanied by maturation, is more important in virus transfer to T lymphocytes than high levels of DC infection without maturation (Figure 9). Brenchley et al described the systemic activation of the immune system in relation to HIV-1 immunopathology [56] whereby CD4+ T lymphocytes are constantly activated by leaky gut products which support HIV-1 replication and lead to significant CD4+ T lymphocyte death [57]. It was shown that circulating LPS was significantly increased in chronically HIV-1 infected individuals and in simian immunodeficiency virus (SIV)-infected rhesus macaques, while this was not seen in non-pathogenic SIV infection of sooty mangabeys [58]. By analogy MVs released in parallel with HIV-1 should also induce DC activation of CD4+ T lymphocytes in vivo and this may contribute to HIV-1 replication and spread.
Recent studies have shown that many virally infected cells secrete MVs, including vaccinia virus [59], hepatitis C [60] and Epstein Barr virus (EBV) [61], [62]. The analysis of the protein complement of the MVs present in the anti-CD45 bead bound and activated SUPT1.CCR5-CL.30 preparations, especially by Mass Spectrometry, detected candidate maturation inducing proteins: the HSPs 90α and β, and HSPs 70/71 (the latter at very low levels) and the HIV-1 protein nef [63].HSPs function as chaperone proteins with varied cellular locations (see review: [64]) and have been shown to act as a source of antigen and may play a role in the transfer of peptides to APCs when released into the extracellular milieu in exosomes [65], [66]. When expressed on the cell-surface HSPs are capable of activating the immune system in vivo [67], [68]. Several heat shock proteins (HSPs 60, 70/71, 96) have been reported to induce maturation or activation of DCs to variable extents, independently of lipopolysaccharide,, via NFκB and the MAPK pathway [63], [69]–[73]. Here we have shown that HSPs 90α and β, are the major cellular HSPs expressed in MVs and are fairly evenly distributed between MVs and MV depleted HIV. Both purified recombinant HSP90α and β induce maturation of MDDCs alone with HSP90β being the most potent.
Nef was also detected in MVs but depleted in the MV stripped HIV inoculum. Pure reombinant nef was a potent inducer of MDDC maturation. Nef has previously been shown to be secreted in exosomes and induce bystander cell effects, [74] and to be taken up by DCs and induce IL-12 secretion and clustering with T lymphocytes. Effects on DC maturation and ICAM-1 expression were not studied [51]. In our experiments a combination of both proteins induced higher levels of DC maturation/ICAM-1 expression together than either alone. Nevertheless MVs from uninfected activated SUPT1.CCR5-CL.30 cells or primary T lymphocytes containing only HSP90α/β but at high concentrations, could substitute for the MVs from infected cells, containing both HSP90α/β, and nef. Thus both HSP90α/β are responsible for the effect MVs derived from uninfected activated primary T and SUPT1.CCR5-CL.30 cells in enhancing maturation of MDDCs.
The complexity of the HIV-1 and MV effects on DCs is dissected into 4 scenarios in panels A and B of Figure 9. In panel A the whole MDDC culture is exposed to an inoculum of HIV virions, containing HSP90α/β, and MVs, containing HSP90α/β and nef, leading to maturation of both infected and uninfected bystander DCs (which are the majority). Additionally infected DCs produce endogenous nef, possibly explaining the higher degree of maturation. In panel B the whole MDDC culture is exposed to HIV containing (a lower dose of) HSP90α/β but not nef, which is insufficient to induce maturation in bystander DCs. Exogenous MVs from activated uninfected T lymphocytes, bearing HSP90α/β, are sufficient to restore the maturing effect of the inoculum. In the infected MDDCs the production of endogenous nef in the infected DCs probably combines with exogenous HSP90α/β, to induce a low degree of maturation. These proteins may induce signalling either through the Vav pathway (for nef) or plasma membrane TLR2/4 (for HSPs) [51].
In conclusion, in vitro DC maturation and ICAM-1 up-regulation is induced by MVs and HIV-1 exposure in all cells of the culture and by HIV-1 replication as well as exposure to MVs in infected DCs. In turn this maturation and adhesion molecule up-regulation induces greater cluster formation and activation of CD4+ T lymphocytes and subsequent transfer of HIV-1. Hence the biological role of MVs in HIV-1 emerging from lysing CD4+ lymphocytes needs to be considered during DC infection in vitro and in vivo. Activation of T lymphocytes in vivo may contribute to the progressive immunodeficiency of HIV disease. Conversely exposure of uninfected bystander DCs in vivo could lead to endosomal then cytosolic uptake of MVs containing nef and HSP90α/β, maturation of DCs and induction of CD8 lymphocyte responses to nef, which is disproportionately recognised as a target HIV protein.
MDDCs were generated from CD14+ monocytes isolated from peripheral blood mononuclear cells (PBMC) of anonymous blood donors from the Australian Red Cross Blood Service, Sydney using CD14 magnetic beads (Miltenyi Biotech; Gladbach, Germany) as described previously [10].
When required, MDDC were matured for 48 hours in maturation mix consisting of (v/v) final concentration 50 pg/mL IL-1β (R&D Systems; Minneapolis, MN, USA), 5 U/mL IL-6 (R&D Systems), 50 pg/mL TNF-α (R&D Systems) and 5 ng/mL PGE2 (Sigma; Milwaukee, WI, USA) in 0.1% (w/v) bovine serum albumin (BSA, Sigma) PBS.
CD4+ T lymphocyte isolation (97% purity) was performed using magnetic beads (Miltenyi Biotech) according to the manufacturer's protocol.
Primary blood myeloid DCs were isolated from PBMCs collected from whole blood, using a negative selection magnetic bead myeloid DC isolation kit (Miltenyi), as per manufacturer's instructions. Isolated mDCs were cultured at 1×106 cells/mL in RF10 (RPMI supplemented with 10% FCS and no cytokines) and collected at 24 h post treatment.
Phenotype/purity was checked by Flow cytometry. Stained with Live/Dead Aqua (Invitrogen), Lin marker (BD), HLA-DR (Biolegend), BDCA-1 (Miltenyi) and BDCA-3 (Miltenyi). Maturation was checked by staining with Live/Dead, HLA-DR, BDCA3, CD83 (BD) and CD86 (BD). Cells were gated by size, Live, and HLA-DR+ before gating on BDCA-3+ and BDCA-3− (“BDCA-1”). CD83 and CD86 expression was analysed on the separate BDCA-1/BDCA-3 populations.
Recombinant HSP90α and β free of LPS, were purchased from Abcam, UK (at 2.7 mg/mL and 1.7 mg/mL respectively) and recombinant nef from BioAcademia Inc. Japan (at 0.48 mg/mL)
HEK293T cells (Human Embryonic Kidney cells, NIH AIDS Research and Reference Reagent Program) were transfected with pWT/BaL (NIH AIDS Research and Reference Reagent Program, contributed by Dr. Bryan R. Cullen) and pHEF-VsV-g (NIH AIDS Research and Reference Reagent Program, contributed by Dr. Lung-Ji Chang) plasmids using polyethylenimine (PEI, Polyscience; Warrington, PA, USA) to generate VsV-g pseudotyped pBaL. Purified high titre HIV-1BaL stocks in the order of 2×109 TCID50/mL were generated by infection of SUPT1.CCR5-CL.30 cells (Human Non-Hodgkin's T lymphocyte Lymphoma, contributed by Prof. James Hoxie at the University of PA) with VsV-g pseudotyped pBaL. HIV-1 infected supernatants were concentrated using tangential filter concentration using the Millipore Lab scale system (Millipore; Billerica, MA, USA) and 2× Pellicon filters connected in parallel (300 kDa) (Millipore). When required, CD45+ MVs were depleted from supernatant using CD45 magnetic beads (Miltenyi Biotech). Virus (18 mLs) was incubated at room temperature with 2 mLs microbeads for 2 hours before adding to the top of a LS column. CD45 depleted virus that flowed through the column as well as non-depleted supernatants were concentrated further by ultracentrifugation with 1 mL under-layed 20% sucrose cushion and centrifuged at 100,000×g (Beckman Optima XL-100K Ultracentrifuge with 70Ti rotor) at 4°C for 90 minutes. Virus content was determined by p24 gag ELISA as per manufacturer's instructions (Beckman-Coulter; Hialeah, FL). The 50% tissue culture infectious dose (TCID50) values were generated in TZM-BL cells (NIH AIDS Research and Reference Reagent Program, contributed by John Kappes and Xiaoyun Wu) measured by LTR β-galactosidase reporter gene expression after a single round of infection [75].
MVs from activated SUPT1.CCR5-CL.30 or primary T lymphocytes were generated using antibodies to CD3 (1 µg/mL, BD Pharmingen; Becton Dickinson; San Jose, CA) and CD28 (5 µg/mL, BD Pharmingen) added to 100×106 cells, cultured for 3 or 6 days respectively alongside unstimulated cells and the supernatant concentrated as above. The CD45 concentration for virus and MV was determined by western blot. Endotoxin levels for all virus and MV stocks were negative using the ToxinSensor Chromogenic LAL Endotoxin Assay Kit (GeneScript; Piscataway, NJ).
Immature day 5 MDDCs were infected with pelleted (HIV-1BaL(pellet)) or CD45depleted (HIV-1BaL(CD45−)) virus in 200 µL media with a MOI of 3 at 37°C for 2 hours before resuspending at 1×106/mL and incubating further as required. MVs were added at 20 µL/mL to match the CD45 concentration to HIV-1BaL(pellet) virus stocks. Recombinant proteins were added at various final concentrations to MDDCs cultured at 0.5×106cells/mL in RF10+cytokines. Harvested cells at 48 hours post treatment for FACS analysis.
Direct conjugated mAbs directed against ICAM-1-fluorescein isothiocyanate (FITC) (Beckman Coulter), CD80-Phycoerythrin-Cy5 (PE-Cy5), CD83-FITC, CD86-PE-Cy5, HLA-DR-Allophycocyanin (APC), MR-APC, CD1a-FITC, CD40-FITC, CD3-APC, CD8-PE-Cy5 and CD69-APC (BD Pharmingen) and CD4-PE (Sigma) were used for surface staining. Antibody staining was performed at 4°C for 30 minutes using fluorescence-activated cell sorter buffer (1% (v/v) Human Ab Serum, 2 mM EDTA and 0.1% (w/v) sodium azide made up in PBS). For intracellular staining HIV-1BaL or mock treated MDDCs were fixed and permeabilised in Cytofix/Cytoperm (BD), re-suspended in permwash buffer (1% (v/v) human AB serum (Sigma), 0.1% (w/v) saponin, 0.1% (w/v) sodium azide, made up in PBS at 4°C) and incubated for 2 hours with PE conjugated p24 (clone KC57-RD1, Beckman Coulter; Fullerton, CA). IgG isotype control antibodies were incubated with cells to control for nonspecific binding. Cells were then analysed with a FACS-Canto flow cytometer (Becton Dickinson) and FlowJo (Tree Star Inc., Ashland, OR).
Blocking of ICAM-1 (clone HA58, 5 µg/2×105 cells, BD Pharmingen) was performed at 37°C for 30 minutes.
2×105 cells were lysed at 60°C for 90 minutes in Q-PCR Lysis Buffer (10 mM Tris-Hydrochloride, 50 mM potassium chloride, 2.5 mM magnesium chloride, 0.45% (v/v) NP-40, 0.45% (v/v) Tween-20, 50 µg/mL Proteinase K (Sigma)) followed by denaturing at 94°C for 15 minutes. HIV-1 proviral DNA was detected using the HIV-long terminal repeat (LTR) gag primer probe set as previously described [76]. The cell number was normalised to albumin as previously described [77]. The HIV-1 assay reaction contained Quantitative PCR SuperMix-UDG mastermix (Invitrogen), 300 nM forward primer, 300 nM reverse primer and 50 nM dual labelled probe and the albumin assay contained 150 nM dual labelled probe. After initial incubations of 50°C for 2 minutes and 94°C for 2 minutes, 45 cycles of amplification were carried out at 95°C for 15 seconds followed by 64°C for 45 seconds for the HIV-1 assay and 62°C for 45 seconds for the Albumin assay. The reaction run on the Corbett 3000 Rotor-Gene machine (Corbett Life Science; Sydney, NSW, Aus) and analysed using Corbett Rotor-Gene 6 software (version 6.0).
In addition, the expressions of selected genes were assessed using reverse transcribed total unamplified RNA as previously described [10]. The cDNA was subject to Q-PCR using published primers [10] with the addition of ICAM-1 primers: (Forward: CGTGGGGAGAAGGAGCTGAA, Reverse: CAGTGCGGCACGAGAAATTG, Sigma).
For the clustering assay, treated MDDC were co-cultured with autologous CD4+ T lymphocytes at a ratio of 1 DC∶5 T lymphocytes and incubated at 37°C for 45 minutes. Cells were subsequently stained for CD1a, CD3 and p24 for flow cytometric analysis as described above.
The proliferation was determined by carboxyfluroescein succinimidyl ester (CFSE) dilution. Briefly, allogeneic PBMCs were stained with 5 µM final concentration CSFE (Molecular Probes; Eugene, OR) for 10 minutes at 37°C, rescued with equal volume of 100% FBS and washed with RPMI+10% FBS. These PBMCs were added to the HIV-1BaL infected MDDCs at a ratio of 1 MDDC∶10 PBMC and incubated at 37°C for 5 days. Cells were subsequently stained for CD3, CD4, CD8 and p24 for flow cytometric analysis as described above.
MDDCs infected with HIV-1BaL (pellet or CD45−) for 48 hours were co-cultured with resting CD4+ T lymphocytes for a further 24, 48, 72, 96 and 120 hours at 37°C before staining with p24 antibody for flow cytometry as described above.
Viral, MV and parent cell preparations were lysed for 1 hour at 4°C in SDS lysis buffer (10 mM HEPES, 150 mM NaCl, 1% (v/v) Triton-X-100, 1 µg/mL protease inhibitor cocktail (Sigma) at pH of 7.5), followed by centrifugation for 10 minutes at 16,000×g at 4°C. The protein concentration was determined using the DC Protein Assay (Bio-Rad) according to the manufacturer's instructions. The protein concentration was determined using the DC Protein Assay (Bio-Rad) according to the manufacturer's instructions.
30 µg of protein prepared with 1× NUPAGE lithium dodecyl sulfate (LDS) sample buffer (Invitrogen) containing 400 mM dithiothreitol was boiled for 10 minutes before loaded into NUPAGE 4–12% polyacrylamide bis-tris gradient SDS-PAGE gels (Invitrogen) alongside a 10–250 kilo Dalton (kDa) molecular weight marker (Bio-Rad). Electrophoresis was run at 200 V for 50 minutes in NUPAGE 3-(N-morpholino)propanesulfonic acid (MOPS) SDS Running Buffer (Invitrogen). The proteins were transferred to a nitrocellulose membrane (Amersham Biosciences) in transfer buffer (250 mM Tris (pH 8.3), 1.92M Glycine and 0.05% (w/v) SDS) overnight at 55 mA. Non-specific binding sites were blocked by incubating for 1 hour in 250 mM Tris (pH 8), 1.4M NaCl and 30 mM KCl (pH 8) containing 20% (v/v) polyethylene glycol tert-octylphenyl ether and 5% skim milk). Blots were incubated with the primary antibodies in 1% skim milk solution followed by the appropriate peroxidase conjugated secondary antibody (Cell Signaling). Primary antibodies were: CD45 (clone 69/CD45, BD Transduction labs), exosome markers: Alix (3A9, Biolegend), CD9 (EPR2949, Abcam), CD81 (JS-81, BD Pharmingen), SMV markers: CD11a (HI111, Biolegend), CD40 (polyclonal, Abcam), AB markers: H2A (poly 6194) and H3 (poly6019, Biolegend) and maturation stimuli markers: CD40L (polyclonal, Abcam), TNF-α (Mab1, Biolegend), HSP60 (polyclonal), HSP70 (3C6), HSP90AB1 (4C10) and HSP90B1 (EPR3988, Origene). The membrane was developed using Western Lighting Enhanced Chemiluminescence Substrate Reagent and the Oxidising Reagent (Perkin Elmer; Glen Waverly, Vic, Aus).
For Mass Spectrometry, 30 µg of each sample was prepared in SDS lysis buffer, run on 4–12% bis-tris gradient SDS-PAGE gel and stained with Brilliant Blue G (Sigma).
The 1D SDS-Gel lanes were sliced into 31 1 mm×5 mm bands using a disposable grid cutter (The Gel Company, USA) and in-gel digested with trypsin using an automated liquid handling procedure with a TECAN Freedom Evo liquid handling system (Männedorf, Switzerland). The 31 fractions were pooled (2 gel fractions per pool) and analysed by LC-MS/MS in duplicate.
Peptide separation was performed on an Eksigent Nano 2D plus system (ABSciex, USA) employing splitless pumps enabled for nanoflow rates. RP-HPLC Trap and separation columns were prepared in-house (Supplementary data for extended Materials and Methods). For each LC run, sample was injected on the trap and washed for 10 minutes at 2.5 µL/min with loading buffer (2% v/v acetonitrile and 0.1% v/v Formic acid). Sample was separated by a linear gradient changing from 97% A (0.1% v/v formic acid in water) and 2% B (0.1% v/v formic acid in acetonitrile) to 40% A and 60% B in 60 minutes at 0.3 µL/min. The in-gel digested samples were analysed with a Thermo-Fisher LTQ-Velos Orbitrap. MS1 data were collected over the range of 300–2000 m/z in the Orbitrap set at resolution 30,000. FTMS preview scan and predictive automatic gain control (pAGC) were enabled. The full scan FTMS target ion volume was 1×106 with a max fill time of 500 ms. MS2 data were collected in the LTQ-Velos with a target ion volume of 1×104 and a max fill time of 100 ms. The 20 most intense peaks from a preview scan of each full Orbitrap scan were selected (with a selection window of 2.0 Da) for collision-induced dissociation (CID) with wide-band activation. Dynamic exclusion was enabled to exclude an observed precursor for 180 seconds after two observations. The dynamic exclusion list size was set at the maximum 500 and the exclusion width was set at ±5 ppm based on precursor mass. Monoisotopic precursor selection and charge state rejection were enabled to reject precursors with z = +1 or unassigned charge state.
Mass Spectrometry analysis involved converting Thermo .RAW files to mzXML format using MSConvert [78] and searched with X!Tandem [79] version 2010.10.01.1 (Supplementary data for extended Materials and Methods).
The identified peptides and their inferred proteins and spectral quantities are reported in the appended Microsoft Excel file (Mercier et al peptide-protein proteomic data.xlsx). The raw data for this project (103 Thermo .raw files, 53.5 Gb total) has been deposited in the Tranche proteomics raw data repository [80], [81] and may be downloaded from the Peptide Atlas data repository http://www.peptideatlas.org/PASS/PASS00251 or the ProteomeCommons.org Tranche repository using the following hash: “69iC+dKFUFu7JMcZxCSdI3cS0q37GPeG3yuWB2h32wnh27xemcTBoEY75tDGOGt0fGFes8Jyxf+ST5lv7Uz7lfUY5UQAA AAAAAAy3A = = ”
The raw data for this project (103 Thermo .raw files, 53.5 Gb total) has been deposited in the Tranche proteomics raw data repository [6], [7] and may be downloaded from the PeptideAtlas data repository
http://www.peptideatlas.org/PASS/PASS00251 or the ProteomeCommons.org
Tranche repository using the following hash:
“69iC+dKFUFu7JMcZxCSdI3cS0q37GPeG3yuWB2h32wnh 27xemcTBoEY75tDGOGt0fGFes8Jy 69iC+xf+ST5lv7Uz7lfUY 5UQAAAAAAAAy3A = = ”
|
10.1371/journal.pntd.0001866 | Human Macrophage Response to L. (Viannia) panamensis: Microarray Evidence for an Early Inflammatory Response | Previous findings indicate that susceptibility to Leishmania (Viannia) panamensis infection of monocyte-derived macrophages from patients and asymptomatically infected individuals were associated with the adaptive immune response and clinical outcome.
To understand the basis for this difference we examined differential gene expression of human monocyte-derived macrophages following exposure to L. (V.) panamensis. Gene activation profiles were determined using macrophages from healthy volunteers cultured with or without stationary phase promastigotes of L. (V.) panamensis. Significant changes in expression (>1.5-fold change; p<0.05; up- or down-regulated) were identified at 0.5, 4 and 24 hours. mRNA abundance profiles varied over time, with the highest level of activation occurring at earlier time points (0.5 and 4 hrs). In contrast to observations for other Leishmania species, most significantly changed mRNAs were up- rather than down-regulated, especially at early time points. Up-regulated transcripts over the first 24 hours belonged to pathways involving eicosanoid metabolism, oxidative stress, activation of PKC through G protein coupled receptors, or mechanism of gene regulation by peroxisome proliferators via PPARα. Additionally, a marked activation of Toll-receptor mediated pathways was observed. Comparison with published microarray data from macrophages infected with L. (Leishmania) chagasi indicate differences in the regulation of genes involved in signaling, motility and the immune response.
Results show that the early (0.5 to 24 hours) human monocyte-derived macrophage response to L. (Viannia) panamensis is not quiescent, in contrast to published reports examining later response times (48–96 hours). Early macrophage responses are important for the developing cellular response at the site of infection. The kinetics and the mRNA abundance profiles induced by L. (Viannia) panamensis illustrate the dynamics of these interactions and the distinct biologic responses to different Leishmania species from the outset of infection within their primary host cell.
| Leishmania parasites cause a spectrum of diseases (cutaneous, visceral and the deforming forms—chronic cutaneous and mucocutaneous) known as leishmaniasis. The macrophage, a key cell in the immune system, is the cellular target of Leishmania parasites in the mammalian host. Previous studies showed the responses of monocytederived macrophages from naturally infected humans to infection with Leishmania (Viannia) panamensis were key to adaptive immune responses and clinical outcome. Consequently, an mRNA microarray approach was employed to assess the changes in macrophage gene expression over time (0.5 to 24 hours) induced by L. panamensis. The highest level of gene expression induction occurred early (0.5–4 hours); the early pathways (groups of genes) activated included those involved in the innate immune response (signaling, phagocytosis, TLR activation, and inflammatory). Early gene activation is presumed to be important for the developing cellular milieu at the site of infection. By 24 hours post-infection the dominant pathways involved metabolic functions. However, a comparison of the macrophage response to L. (V.) panamensis to that of L. (L.) chagasi (causative agent of visceral leishmaniasis) at 24 hours revealed a differential up-regulation of genes (cell adhesion, signaling, and inflammation) in response to these species. These observations underscore the distinct biology of different Leishmania species from the outset of infection.
| Leishmania species are obligate intracellular parasites, of extensive public health importance in tropical and subtropical regions of the world. Leishmania organisms produce a wide spectrum of diseases.Depending upon the species initiating infection and on the immunological status of the host, disease manifestations range from simple cutaneous to chronic cutaneous, mucocutaneous, diffuse cutaneous and visceralleishmaniasis [3], [4]. Overall control of leishmaniasis has been associated with amoderate type1 CD4 response (IFNγ); whereas excessive down-regulation of this response through the action of various type 2 associated cytokines (e.g. IL-13, IL-10, IL-4) has been related with disease exacerbation.
However, the mechanisms underlying pathogenesis remain to be fully understood.Notably parasites in the Leishmania (Viannia) subgenus, which are prevalent in Central and South America and are associated with cutaneous (limited and chronic) as well as mucocutaneous disease, have not been extensively studied. Consequently, the mechanisms involved in pathogenesis are not as yet well characterized. In the case of mucosal disease, a hyperinflammatory response (IFNγ, TNFα) is evident with a down-regulation of IL-10R expression [5]–[8]. In contrast, a mixed cytokine response is observed across the spectrum of cutaneous infection (asymptomatic to chronic) caused by L. (Viannia) parasites, both IL-10 and IL-13 have been implicated in disease [1], [7], . Recent studies utilizing a murine model of chronic infection havecorroboratedthe mixed cytokine response and role of IL-13 and IL-10 in the pathogenesis ofL. (V.) panamensis infection [13].
Differences in parasite gene expression patterns and differences in virulence factors (between various Leishmania species and between strains of the same species) undoubtedly contribute to the variation in leishmanial disease manifestations. However, genomic sequences [14] and indeed RNA expression profiles between the promastigote (insect stage) and amastigote (intracellular mammalian stage) have revealed a limited number of differences [15]–[17]. Whereas the relationships between developmentally regulated Leishmania genes and parasite virulence or disease associated pathology are of interest, an alternate approach involves the investigation of parasite interactions with host cells. Such studies can provide insight into the underlying mechanisms of the host-parasite dynamic.
Central to the host-Leishmania relationship is the macrophage. These cells are fundamental to host defense through their functions in phagocytosis, destruction of the parasites and antigen presentation. Leishmania evade the host immune response and replicate inside macrophages establishing a long-term-infection, influencing macrophage gene expression and consequently, the microenvironment in which they persist. The macrophage response to infection is clearly involved in determining susceptibility/resistance in murine leishmaniasis [18]–[22]. The NRAMP1 or SCL11A1 gene encodes a macrophage transporter [23]–[26], that regulates murine innate susceptibility/resistance to L.(L.)donovani and also L.(L.)mexicana [26]and has been linked to susceptibility in human visceral leishmaniasis [27]. Further, susceptibility of specific mouse strains to infection parallels the ability of their macrophages to control infection [21], [22], [28]. Studies of Leishmania-infected mouse and human macrophages indicate that extensive changes in mRNA abundance occur upon infection [29]–[31] and may be critical to determining the inflammatory and developing host responses.
Individuals with chronic or recurrent human cutaneous disease caused by Leishmania (Viannia) have increased in vitro susceptibility of monocyte-derived macrophages (MDMs) to parasite entry and survival [1], [32], [33]in comparison to individuals with asymptomatic infection. In vitro proliferation of T lymphocytes in response to intracellular amastigotes was found to be significantly lower among individuals with a history of recurrent disease compared with subclinically infected individuals [1]. Furthermore, linear regression analyses generally revealed a relationship between the production of IFNγ, IL-13 and IL-10 by PBMCs of infected individuals [13]; however, this did not occur for cases of recurrent disease [1]. Thus, there are differences in both the innate macrophage and acquired immune responses between individuals with different outcomes of infection.
Previous studies that have examined the response of human monocyte-derived macrophages to Leishmania infection have primarily focused on Old World species [2], [15], [29], [34], which lead to distinct immune responses upon infection in comparison to New World species [35]. The gene expression pattern of primary human monocyte-derived macrophages to species of the Viannia subgenus has not previously been examined. Herein, we examine the changes in gene expression profiles in response to Leishmania (Viannia) panamensis exposure/infection in human MDMs from healthy donors using human gene microarrays. The study indicates a rapid parasite-induced activation of an MDM profile that dampened with time. Pathway analyses indicated an intrinsic activation of signaling and inflammatory responses. In addition, comparisons at 24 hours post-infection to L.(L.)chagasi infected human PBMC-derived monocyte-derived macrophages showed thatmolecules involved in cell signaling, adhesion, and inflammation that are evident in L. (V.) panamensis infection are generally absent or reduced in the MDM response to L.(L.) chagasi. These data suggest the activation/expression profiles of human MDMs differ after interaction with different Leishmania species and support the hypothesis that distinct functional activitiesof the host cells may determine the course and outcome of infection with distinct Leishmania species.
Leishmania (Viannia) panamensis strain MHOM/COL/85/1166 was propagated and infective stationary phase promastigotes obtained as previously described [33]. Briefly, promastigotes were cultured in Senekjie's medium at 25°C, recovered on day 6 (stationary phase) by centrifugation at 900× g for 10 min., washed in buffered saline solution (PBS) and used for infection.
All procedures were conducted in accordance with good clinical practice and IRB approved protocols (CIDEIM and YaleUniversity; DMID, NIH). Signed informed consent was obtained according to international guidelines and national regulations. Blood (150 mls) was collected by venipuncture from 6 healthy donors. Donors originated from areas that were non-endemic for leishmaniasis and were non-responsive to leishmanial antigen, as assessed by in vitro proliferation [33]. As previously described [33], blood was defibrinated and mononuclear cells were recovered from a Ficoll-Hypaque gradient. Cells were allowed to adhere for 2 hours in flasks coated with gelatin. After washing, adherent cells were differentiated to macrophages for 5 days in culture (37°C, 5% CO2) in RPMI 1640 medium supplemented with 10% FBS [33]. Monocyte-derived macrophages (MDMs) were infected at 34°C with L. (V.) panamensis promastigotes (opsonized with inactivated AB+ serum) at a 1∶20::MDM∶parasite ratio. Uninfected monocyte-derived macrophages served as controls. Cells were washed at 0.5 hours and the medium replaced. For analysis, cells were harvested at 0.5, 4 and 24 hours and RNA isolated as described below. To assess the level of infection, monocytes from each donor were cultured, differentiated and infected in chamber slides. Infection was microscopically evaluated at each time point, as described previously [1], [33]. Organisms were observed within macrophages by 25 minutes; over the time period of the experiment infection levels did not significantly differ (varied from58–46% of total macrophages) and averaged values of 3.1, 2.8 and 3.2 parasites/macrophage (at 0.5, 4, 24 hours respectively).
At the time points indicated TRIzol (Invitrogen) was directly added to each flask containing the MDMs. RNA was recovered using an RNeasy Micro Kit (Qiagen) and purified according to the manufacturer's specifications. RNA was quantified using a Hitachi GeneSpec III spectrophotometer; quality was assesses using 2 to 3 pg of total RNA and an RNA Pico Chip run on a 2100 Bioanalyzer system (Agilent Technologies). RNA was stored at −70°C until used for microarray analysis.
Polyadenylated RNA was amplified from total RNA using the SenseAMP plus RNA Amplification Kit (Genisphere). Subsequently, cDNA was synthesized from poly (A)+ RNA (1 µg) using an Array 900MPX kit (Genisphere) following manufacturer's recommendations. Reverse transcription was performed for 2 hours at 42°C using SuperScript II (Invitrogen). The reaction was stopped with 0.5 M NaOH/50 mM EDTA; and DNA/RNA hybrids were denatured for 15 min at 65°C. The reaction mix was then neutralized and cDNA was purified using a Qiagen MinElute PCR Purification Kit. Results from three separate comparative microarray analyses (data not shown), indicated that data from amplified and non-amplified RNA samples were comparable; consequently, amplified RNA was used for analysis.
cDNA was ligated to either Cy3 or Cy5 dye; Cy3 was used for control samples (uninfected monocyte-derived macrophages at identical incubation time) and Cy5 for L. (V.) panamensis infected monocyte-derived macrophages. Each tagged (Cy3 or Cy5) cDNA was purified using Qiagen-MinElute PCR Purification Kit. After addition of 1 µg of denatured human Cot-1 DNA to the purified tagged cDNA, the mixture was hybridized to a 70-mer OHU28K Human oligo array (Yale University Keck Microarray Resource)using a 22×71 mm AdvaCard overnight at 55°C using an Advalytix Array Booster Autohyb Instrument. The OHU28K Human Oligonucleotide array was fabricated from a 70 mer oligonucleotide set consisting of 21,329 oligonucleotides representing Operon Human Version 2.0 set plus a Version 2.1 Upgrade set of 6,228 oligonucleotides that represent probes from Version 3.0 Human Genome Set. After hybridization, arrays were washed with 2X SSC buffer and/or 0.2% SDS. Arrays were then dried and incubated for 4 hours with theGenisphere3DNA Array 900MPX capture reagent mix at 65°C. After washingusing 2X SSC buffer containing 2% SDS, arrays were dried and analyzed, as indicated below.
Scanning of each array was performed according to manufacturer's instructions using the Axon GenePix 4000A laser confocal scanner at 532 and 635 nm. Genisphere dendrimer cy3/cy5.tiff images and .gal files were analyzed using GenePix Pro Software to grid the images and generate a spreadsheet data.gpr file.
GeneSpring GX 7.2 (Agilent Technologies, Inc., Palo Alto, CA) was used for gene expression data analysis. LOWESS normalization was performed [36] to eliminate dye-related artifacts. Gene expression comparisons were made between infected and uninfected at different time points. In each comparison, genes with the average intensity values less than twice background intensity, or fold change less than 1.5 fold (up or down-regulated) were excluded in the further statistical analysis. A two-sample t test using p-value cutoff 0.05 was applied to determine if a gene was statistically differentially expressed. Genes were annotated with KARMA (Yale University, http://medicine.yale.edu/keck/ymd/downloads.aspx) and classified into different GO SLIM categories. Pathway analysis was performed to identify significantly affected pathways using the pathway database from the Yale Center for Statistical Genomics and Proteomics, http://bioinformatics.med.yale.edu/), which includes pathways from KEGG (http://www.genome.jp/kegg/pathway.html), GenMAPP(http://www.genmapp.org/) and Biocarta(http://www.biocarta.com/genes/index.asp) pathway databases. The cutoff to determine genes that were differentially expressed or to cluster similar expressed genes in significant pathways were 1.5- fold difference between the fluorescence signal intensity from a non-infected control and the L.(Viannia) panamensis-infected monocyte-derived macrophage with a significance level of ≤0.05. Although amongst these pathways there are genes in common (such as NF-κβ, JUN, FOS, MYC, PLCB1), the pathways are defined by distinct subsets of genes.The raw data and processed data have been made available at Gene Expression Omnibus (GEO/NCBI) website (GEO ACCESSION NUMBER GSE25819).
For comparative analyses of the monocyte-derived macrophage responses, L. (L.) chagasi/infantum (16 hours after infection) was compared to the 24 hour response to L. (V.) panamensis. Macrophages in both cases were derived from human peripheral blood monocyte-derived macrophages, using standard methods and infection ratios were comparable. For comparative analyses the microarray fluorescence values from the study of Ettinger and Wilson using Affymetrix U133Plus2 microarray chips [2], adjusted for background intensities (normalized) using the gcRMA method, were employed. For each probe set, the fold change (infected/uninfected) was calculated for each individual in the study, and their geometric mean taken. The data were linked to the 24-hour dataset of the current study by Gene ID (LocusLink). Where more than one probe set or oligonucleotide was present for a single Gene ID, the most extreme of the values was taken. In comparisons with L. (L.) chagasi, genes were considered if, in either study, the multiplicity-adjusted p-value was less than 0.05 for comparing infected versus uninfected. The current study used the Benjamini and Hochberg adjustment [37]. This analysis was done using the R software (http://www.R-project.org). The 24 hour time point was employed, as given the variation in macrophage gene expression found with time post-exposure, would represent comparable experimental conditions. It should be noted that given the decrease in gene expression levels with time([30], this study), the number of genes with transcripts up-regulated would be lower for L. V. panamensis at 24 hours (in comparison to 16 hours(L. chagasi/infantum)).
RNA and cDNA prepared from donors for the microarray analyses was not available in sufficient quantity to conduct the qPCR validation. Therefore MDMs were derived from blood monocytes of 6 to 7 additional normal donors for the validation analyses. cDNA was generated from RNA obtained from MDMs infected with L.(V.) panamensis and used as template for the RT-qPCR in an ABI PRISM 7000 Sequence Detection System (Applied Biosystems, Foster City, CA). The reactions were performed using iQ SYBR Green Supermix (Bio-Rad Laboratories) with oligonucleotides at 100 nM. The RT-qPCR settings included an initial activation of DNA polymerase at 95°C for 10 minutes, followed by 50 cycles of denaturation at 95°C for 15 seconds and assay-specific annealing temperature and extension times (Table S1). These SYBR-green assays were used to determine the relative abundance of the following genes: interleukin-1 β (IL-1β), tumor necrosis factor α (TNFα), colony-stimulating factor 2 (granulocyte macrophage) CSF2/GMCSF), prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase – PTGS2), and interleukin 6 (IL-6). Following Vandesompele et al. [36], the gene expression was normalized by the geometric mean of two internal control genes selected from the microarray analysis that displayed no variation in gene expression, namely, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and the39S ribosomal protein L18 (RL18), mitochondrial precursor. Each amplicon was cloned in PCR2.1-TOPO (Invitrogen, Carlsbad, CA) with the TOPO TA Cloning Kit (Invitrogen); the generated plasmids were used to generate a standard curve to determine the absolute number of transcripts for each gene in the samples. The number of transcripts for each gene was then normalized to the geometric mean of number of transcripts of GAPDH and RL18. Following the same analyses method applied to the microarray data, for each gene the normalized number of transcripts in the infected MDMs was divided by the normalized number of transcripts in the non-infected MDMs and the geometric mean was calculated for the ratios obtained for the samples from different donors.
The infection of macrophages with a parasitic protozoan, such as Leishmania, is a complex event involving multiple cellular processes (including phagocytosisand the induction of cell signaling), that result in the expression of cytokines, and chemokines, and that ultimately impact upon the induction of the host immunological response (adaptive immunity). Furthermore, monocyte-derived macrophage physiological and metabolic responses change in response to the parasitism. In order to examine the development of the initial L. (Viannia) panamensis promastigote – monocyte-derived macrophage interaction, we studied the changes in gene transcripts at 0.5, 4 and 24 hours of infection/exposure using monocyte-derived macrophages from six healthy non-immune donors. Confounding adaptive immune response by residual lymphocytes was obviated by choosing donors known to live in areas non-endemic for leishmaniasis who were determined to be non-responsive to leishmanial antigen.
MDM transcripts that were significantly up- or down-regulated as a consequence of interaction with L. (Viannia) panamensis (Figure 1) included more than 500 genes. This response was attenuated at later time points. Notably, at the earliest observational points (0.5 and 4 hours of infection/exposure), a total of 523 and 587 genes, respectively, were found to be up-regulated, whereas by 24 hours this number had declined to 388. The geometric mean of fold-change in expression of the up-regulated transcripts also decreased with time going from 4.39 (0.5 hours) to 3.21 (4 hours) and 2.53 (24 hours). The number of transcripts that were significantly down-regulated remained relatively constant in response to infection/exposure with 43, 32 or 50 transcripts down-regulated at 0.5, 4 or 24 hours, respectively. In addition, the overall geometric mean-fold magnitude of transcript down-regulation was similar overtime (0.0636 (0.5 hours), 0.0366 (4 hours), 0.0486 (24 hours)). Whereas 26 gene transcripts were consistently down-regulated, the expressions of 55 transcripts were observed up-regulated at all three time points. Many of the consistently up- or down-regulated mRNAs encoded proteins involved in cell signaling or gene regulation (Table S2).
Consistent with the apparent dampening of the response to infection between 4 and 24 hours after infection, the number of highly up-or down-regulated transcripts (expression>5-fold) decreased with time. At 0.5 hours post-infection/exposure, 210 transcripts were found to be highly (≥5-fold) changed, whereas at 4 or 24 hours of infection the number of highly changed transcripts decreased to 151or 104, respectively. It was notable that amongst the highly regulated transcripts, many were changed at only one time point, with 105, 49 and 23 genes being unique to the 0.5, 4 and 24 hour time points, respectively, underscoring the changing dynamic of the macrophage response.
Although fewer transcripts (Figure 1) were down-regulated than up-regulated, a higher proportion of the down-regulated transcripts were consistently down-regulated (52 to 81%, dependent on the time point).Genes encoding these transcripts (Table S2)primarily encoded molecules involved in regulation of cell signaling (protein kinases, phosphatases) or molecular transport. Additionally, some of the genes whose expression was down-regulated in L. (Viannia) panamensis infected MDMs encoded proteins potentially affectingthe host's immunological control of the invading parasite. These include the receptor for MCP-1 (a known activator of macrophage mediated leishmanial killing [37]–[39]) and the extracellular matrix protein, spondin 2, which has been shown to be involved in the initiation of the innate immune response [40], [41].
During the first 24 hours of infection, the majority of changes were transient; the gene expression profiles reflected a changing dynamic of the host-parasite interaction over the first 24 hours of infection. Whether these differences are due to parasite molecules derived from rapidly killed/damaged parasites or living Leishmania remains to be firmly established.The patterns of gene regulation were further examined using pathway analysis (below).
In order to validate the microarray data, the expression of selected genes was also evaluated using quantitative RT-PCR (Table 1). The genes selected for examination (0.5 and 4 hours post-infection/exposure) were chosen because of their potential pertinence to the early macrophage innate response and/or development of the adaptive immune response. These genes are: interleukin-1β (IL-1 β), tumor necrosis factor-α (TNFα), granulocyte-macrophage colony-stimulating factor (GMCSF), prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase; PTGS2), and interleukin 6 (IL-6) (Table S2).
The increase in expression between 30 minutes and 4 hours detected by microarray was corroborated by qRT-PCR for 4/5 of the genes. Although the fold changes in gene expression were more pronounced for some transcripts in the microarray (i.e. IL1-β, PTGS2), this variation is likely attributable to either the differences between hybridization versus qRT-PCR based amplification methods or by the fact that the experimental samples used for qRT-PCR were generated using PBMCs from individuals different than those employed for the microarray analyses [42]–[44]. Notably, the range of the fold-increases (95% confidence levels) found was consistent for the two methods.
In most cases the direction of change at the different time intervals was the same by both methods, although the expression of PTGS2 was up-regulated maximally at the 0.5 hour time point by qRT-PCR, whereas the transcript level was higher at the 4.0 hour time point according to the microarray analyses. The reason for this discrepancy is not clear. The involvement of prostaglandin synthesis is consistent with other studies of macrophage infection/exposure with Leishmania parasites [45], [46] as well as our observations of the engagement of the Peroxisome Proliferators via PPARα (0.5, 4 and 24 hours) and Prostaglandin Synthesis and Regulation (GenMapp) pathways (4 and 24 hours)(Table S3). Overall, the qRT-PCR analyses corroborate the results from the microarray studies. The mechanisms involved (induction of non-coding miRNAs, regulation of RNA binding proteins) in conferring mRNA stability/instability remain unclear but are of obvious interest for future research.
To determine the biological processes at play, pathway analyses examining the co-ordinate regulation of specific gene groups were performed for each time point. The genes activated and the correlative pathways illustrate the dynamic nature of the L. (V.) panamensis-MDM interaction (Table S3).The outcome of pathway analysis at the three time points post-infection/interaction is consistent with sequential responses after phagocytosis. Thus, the pathways coordinately regulated are more similar at the 0.5 hour and 4.0 hours time points than at the 24 hour time point. The pathways induced early in infection/interaction participate in the induction of signal transduction due in part to phagocytosis. It should be noted that all pathways indicated were up-regulated, with respect to the transcripts analyzed.
Because there are known and pronounced biological differences between the disease presentations resulting from different infecting species of Leishmania, comparative analyses were undertaken of MDM gene expression after 16 hours of infection with L. (L.) infantum/chagasi (visceral leishmaniasis) or macrophage response 24 hours after infection with L. (Viannia) panamensis (Table S4). Notably, even at 24 hours post-infection the range of genes whose expression was increased by L. (V.) panamensis was higher than those increased for L. (L.) infantum/chagasi, although the relative expression of a set of 11 housekeeping genes (Table S4C) in infected versus uninfected macrophages (for both species) were comparable. This is significant, as the macrophage response diminishes with time and it would be expected that fewer transcripts would remain elevated in the case of L. (V.) panamensis. Indeed, the dominant response to L. (L.) infantum/chagasi was down-regulation of gene expression [2]. This predominant down-regulatory response is consistent with other studies comparing the macrophage (mouse and human; different time intervals) responses to other Leishmania (Leishmania) organisms [85]. None of the genes whose RNA abundance was found to be up-regulated by macrophages in response to L. (L.) infantum/chagasi (9 genes) was found to be significantly up-regulated in response to L. (Viannia) panamensis. Of the MDM mRNA levels down regulated by L. chagasi/infantum at 16 hours post-infection, most were not found to be significantly changed in response to L. panamensis (Table S4B).
However, the mRNA levels for two genes, WAS protein family, member 2 (WASF2) and Rap guanine nucleotide exchange factor (RAPGEF1) in contrast to down-regulation after infection with L. infantum/chagasi, were up-regulated in response to L. panamensis (Table S4A). Other mRNA levels unchanged in expression in monocyte-derived macrophages upon L. infantum/chagasi infection were also found to be up-regulated upon L. (Viannia) panamensis infection. The differentially regulated genes included those that were involved in cell functions related to cell signaling, endocytosis and vesicular trafficking, and cellular adhesion/motility (RASAL2, serine/threonine protein kinase, phospholipase D2, N-myristoyltransferase 2, IPP5, PDLIM5, MYCBP2, PHACTR2) as well as the developing immune response (TRAIL R1, GTPBP, Tec kinase, eotaxin, CD109, IL3R, integrin β3 (CD61), VASP).
The expression of Tec kinase, TRAIL-R1, MYC-binding protein 2, CD109 and eotaxin is consistent with what is known concerning the human inflammatory response to L. (Viannia) infection [6], [10], [86]–[88]. Tec kinases are downstream from the activation of TLR [89], [90] receptors and lead to the production of inflammatory cytokines such as TNFα, which is widely associated with pathogenesis of L. (Viannia) infection. The ongoing signaling and vesicular trafficking is in contrast to the relatively quiescent state found for human macrophages infected with L.(L.)infantum/chagasi. The selective up-regulation of genes modulating the macrophage response indicates a differential response to L. panamensis and suggests that the initial innate immune response may lead to and support the development of the non-resolving inflammatory response to infection that is observed clinically.
Prior studies of macrophage transcriptome indicate that changes in MDM gene expression in response to external stimuli (numbers of genes; inflammatory, anti-inflammatory) are highly variable and dependent upon the particle (inert or pathogen) ingested and nature of the exposure [91]–[94]. After interaction or infection with L. (V.) panamensis, results of the current report suggest that the early responses (0 to 4 hours) lead to up-regulation of transcripts contributing to an inflammatory state of the host macrophage. Specifically transcripts encoding proteins involved in cell signaling, inflammation and notably genes involved in the TLR pathways are co-coordinately up-regulated. Simultaneous increased expression of genes associated with anti-inflammatory responses including eicosanoid pathway and transiently IL-10 suggests that a critical balance may be required (between the inflammatory-anti-inflammatory pathways of the macrophage). It is possible that the level of these responses will determine the nature of the acquired immune response that develops and ultimately the outcome of infection or disease. These results substantiate significant differences between the macrophage responses induced by different Leishmania species. The macrophage provides critical functions throughout leishmaniasis, including antigen-presentation, parasite containment, and suppression or activation of nearby immune cells through released modulators. Consequently the observed differences in macrophage gene expression may contribute to the host-parasite dynamic in the different forms of leishmaniasis caused by L. panamensis and L. chagasi/infantum.
|
10.1371/journal.pcbi.1004180 | Interaction of the Antimicrobial Peptide Polymyxin B1 with Both Membranes of E. coli: A Molecular Dynamics Study | Antimicrobial peptides are small, cationic proteins that can induce lysis of bacterial cells through interaction with their membranes. Different mechanisms for cell lysis have been proposed, but these models tend to neglect the role of the chemical composition of the membrane, which differs between bacterial species and can be heterogeneous even within a single cell. Moreover, the cell envelope of Gram-negative bacteria such as E. coli contains two membranes with differing compositions. To this end, we report the first molecular dynamics simulation study of the interaction of the antimicrobial peptide, polymyxin B1 with complex models of both the inner and outer membranes of E. coli. The results of >16 microseconds of simulation predict that polymyxin B1 is likely to interact with the membranes via distinct mechanisms. The lipopeptides aggregate in the lipopolysaccharide headgroup region of the outer membrane with limited tendency for insertion within the lipid A tails. In contrast, the lipopeptides readily insert into the inner membrane core, and the concomitant increased hydration may be responsible for bilayer destabilization and antimicrobial function. Given the urgent need to develop novel, potent antibiotics, the results presented here reveal key mechanistic details that may be exploited for future rational drug development.
| Antimicrobial peptides have the ability to kill harmful bacteria through interaction with bacterial membranes. This manuscript describes the first reported computational study of antimicrobial peptide interaction with both membranes of a Gram-negative bacterium. While antimicrobial peptides have been the topic of many simulation studies, these studies have not incorporated the biochemical heterogeneity of natural membranes. Our simulations add the missing biochemical details and in doing so, reveal that the mechanisms of interaction of polymyxin B1 with the inner and outer membranes of E. coli, are really rather different. The peptides insert readily into the inner membrane, whereas the interaction with the LPS-containing outer membrane is more complex. In summary our results represent a key finding for future drug development that targets these bacteria.
| Antimicrobial peptides (AMPs) are small cationic membrane-active peptides; they can be found in most living organisms and play an essential part in innate immunity [1–3]. These peptides exhibit broad-spectrum antimicrobial activity against bacteria, fungi and viruses, making them of great biomedical interest, particularly in the field of novel antibiotic design.
Polymyxin B1 (PMB1) is a small antimicrobial lipopeptide first derived from the bacterial species Bacilus Polymyxa in 1947[2, 4, 5]. It is composed of a cyclic polypeptide ring and a branched fatty acid tail, and among the amino acids forming the peptide segment are the irregular amino acids D-Phenylalanine (DPhe) and α, γ-Diamino Butyric acid (DAB). Its full sequence is thus: DABC-Thr-Leu-DPhe-DAB-DABC-DAB-Thr-DAB-CO(CH2)4CH(CH3)CH2CH3, where DABC represents the cyclic linkage. The five non-cyclized DAB amino acids each carry a charge of +1, and thus the cationic peptide has a total charge of +5 [6]. PMB1 is a highly potent antimicrobial peptide and is selective predominantly towards all Gram-negative bacterial species, with the exception of the Proteus groups [7]. Unfortunately, treatment of patients with PMB1 has been shown to have adverse side effects on the renal and nervous system [8–10], and therefore clinical use of PMB1 has been limited to topical treatment as well as “last resort” therapy of patients infected with multidrug-resistant bacteria or with chronic conditions who suffer from recurring respiratory infections [11]. However, given the alarming rise in the number of bacterial strains exhibiting multidrug-resistance, there has recently been renewed interest in PMB1 [2].
PMB1 lipopolypeptides are known to permeate across the bacterial outer membrane (OM) by self-promoted uptake, while it is disruption of the inner membrane (IM) that subsequently leads to cell death. The peptides are thought to fulfil the initial stages of their bactericidal activity by anchoring themselves to the bacterial membrane via the DAB amino acids[12, 13]. While the precise mode of action subsequent to peptide anchoring to the membrane is still unclear, it has been established that the polypeptide ring is responsible for causing an increased permeability of the bacterial membrane[4]. It has been proposed that the observed permeabilization is caused by membrane insertion of the polypeptide ring and fatty acid tail, resulting in bilayer disruption and an outflow of intracellular components, followed by cell death [4].
Because of experimental difficulties associated with characterizing dynamic, heterogeneous systems such as membrane-bound AMPs, molecular dynamics (MD) simulations provide a complementary approach to studying their modes of action, in unprecedented detail [14, 15]. Here, we have used a series of MD simulations (Table 1) over microsecond timescales to study, the interaction of an AMP with accurate models of both membrane components of a Gram-negative bacterial cell, there has previously been only one report of a computational study of an AMP interacting with a model OM[16]. In particular, the former membrane is represented by a realistic mixture of phospholipids representative of the bacterial IM, whilst the latter is modelled as an asymmetric bilayer containing a phospholipid mixture in the inner leaflet and rough lipopolysaccharide (LPS) in the outer leaflet. For comparison, we also study the interactions of the AMP with a symmetric lipid A membrane. Computational work on AMPs is well documented [17–19] and simulations of complex models of the OM are also available [20–23], though we report one of the first combinations of the two aspects in atomistic detail. We thus investigate the molecular-level mechanisms of PMB1 binding, insertion, and bilayer disruption for both IM and OM models of the envelope of the archetypal Gram-negative bacterial species, E. coli.
To study the initial stages of AMP interaction with the envelope of Gram-negative bacteria, we simulated PMB1 in the presence of a realistic model of the asymmetric E. coli OM, composed of Re LPS in the outer leaflet and a mixture of phospholipids (including phosphatidylethanolamine, phosphatidylglycerol, and cardiolipin) in the inner leaflet. The molecular compositions of the simulated systems are given in Table 1.
We performed simulations of simplified OM model systems, containing lipid A in both leaflets (Sim_LipA). The reasons for doing this are two-fold; firstly this setup more easily enables the peptides to permeate into the lipid part of the membrane, allowing interrogation of these membrane-peptide interactions over a tractable timescale, and secondly it abrogates the problem of peptides moving across periodic boundaries to interact with the phospholipid portion of the asymmetric bilayer. While the lack of sugars in lipid A provides a simplified model, it does enable us to study the behaviour of the peptide at the lipid headgroup/tail interface in realistic detail. Two independent simulation systems, each comprising eight PMB1 placed ~0.5 nm above one leaflet of the membrane, were run for 3 microseconds. Here, we report only the behaviour that differs from the asymmetric bilayer studies described above.
We next turned our attention to the E.coli IM. We performed simulations of a symmetrical phospholipid membrane model (composed of phosphatidylethanolamine, phosphatidylglycerol, and cardiolipin phospolipids), and exposed it to PMB1 lipopeptides (Table 1).
Our results reveal contrasting behaviour of PMB1 in the presence of different bacterial membrane models. In the case of the LPS simulations, our observations support data from previous fluorescence studies suggesting that DAB residues are key to PMB1 binding and antimicrobial function [12, 13]. Strikingly, we observed aggregation of PMB1 on the LPS membrane surface (Fig 1 and S2 Fig), in which monomers arranged themselves in a micelle-like conformation to bury their hydrophobic tails from the polar sugar rings. This aggregation, coupled with the tendency for PMB1 to cross-link the sugar hydroxyl groups resulted in formation of an immobile, “protein membrane cluster”. The adoption of this fatty acid tail orientation and aggregation may represent a necessary prelude to pore formation, in a manner that reduces the energetic cost for translocation of the PMB1 “micelle” across the polar environment of the LPS sugar groups. Intriguingly, two attempts to speed-up the process of PMB1 insertion into the outer membrane firstly by loosening the membrane by replacing the LPS cross-linking divalent cations with monovalent ions, and secondly by performing steered MD simulations to 'pull' a peptide, also provided strong support for the hypothesis that insertion of PMB1 into the outer membrane is not a facile molecular process.
Because of the slow dynamics of the LPS system [23], we also studied a simplified OM model consisting only of lipid A. Unlike the LPS system, PMB1 appeared to have no effect upon intra-lipid A hydrogen bonding, but instead the DAB residues interacted with phosphate groups, pushing them apart, leading to local membrane deformation and creating a region of reduced membrane-surface charge density that may no longer inhibit fatty acid tail insertion. During this process, Mg2+ ions were intermittently displaced and rebound to phosphate groups, but no long-term ion displacement was observed. As a result, we only witnessed complete insertion of a single PMB1 molecule on the timescales feasible with atomistic simulations. Nevertheless, it has been widely documented that divalent cations are essential for maintenance of OM stability, and PMB1 has been shown to cause destabilisation via displacement of these divalent cations [6, 31, 32].
For the IM model, our results are again in agreement with previous experimental work [12, 13, 26, 33] highlighting the importance of the DAB residues in protein binding, confirming that electrostatic interactions between the DAB and the lipid headgroups are the initial driving force for PMB1 adsorption. However, in contrast with the OM, the distinguishing behaviour of the PMB1 molecules in the IM system was the lack of PMB1 aggregation as well as the striking result that the fatty acid tails of every PMB1 molecule spontaneously penetrated into the hydrophobic core of the membrane, some as far as the lower leaflet (Fig 3). Unlike the OM models, the IM exhibited a large increase in the disorder of the acyl tails upon fatty acid tail insertion. Indeed, PMB1 insertion seemed to be driven largely by hydrophobic interactions, not only due to fatty acid tails but also the D-Phe sidechain. The converged, inserted PMB1 conformation (S4 Fig) structurally resembles that observed in NMR studies performed by Mares et al [26] which focused on the interaction between LPS and PMB1. In both cases, the majority of the ring structure remained bound to the phospholipid headgroups on the membrane surface, while the hydrophobic portion of the ring (residues D-Phe and Leu) adopted an embedded orientation.
In the case of the IM, we also observed a further possible role of the DAB residues in antimicrobial action, namely attraction of water towards the membrane core. Unlike the OM models, upon loss of the surface-bound state and PMB1-membrane insertion, the lipid lateral diffusion rate increased beyond the PMB1-free rate, correlating with peptide/water membrane penetration. Concomitant with this, we observed a remarkable thinning effect of >0.7 nm and increase in area per lipid in the IM as a result of PMB1 insertion, whilst the resultant cross-leaflet interdigitation of acyl tails is consistent with data from spin-labelling and X-ray diffraction reported by Boggs et al [29] and Theretz et al [30]. Since the PMB1 does not aggregate within the IM, but instead inserts as monomers, we may hypothesise on the basis of our collective observations that PMB1 disrupts the IM not through traditional mechanisms of pore formation but through membrane insertion, bilayer thinning, and water penetration. Such effects are certainly indicative of AMP permeation and the early stages of membrane disruption [34].
The limitations of the current study arise from the slow lateral diffusion of LPS, such that the long simulation times required to witness complete PMB1 translocation within the asymmetric LPS membrane is not presently feasible with atomistic simulations. Indeed it has recently been shown that even for simple, phospholipid bilayers, the simulation times required for convergence have historically been seriously underestimated [35]). The slow reorganization of ionic interactions involving zwitterionic phospholipid headgroups when solutes penetrate the lipid-water interface are a particular problem, which is accentuated in the complex LPS headgroups we are studying. Other studies have suggested that atomistic molecular dynamics simulations of AMPs require multi-microsecond timescales [36]. Our simulations reported here show that even non-equilibrium methods such as steered MD are not able to access insertion behaviour into membranes and thus another approach is needed. The large molecular systems and extended timescales accessible to coarse-grain molecular dynamics (CG-MD) simulations provide an alternative and complementary route to studying antimicrobial peptides. Indeed, CG-MD studies have been shown to provide insights into the action of antimicrobial peptides in flat lipid bilayers and spherical vesicles [15, 37, 38]. To study the OM of Gram-negative bacteria using this approach, a coarse-grain model of LPS is urgently needed. Nevertheless, the simulations we have performed have helped to identify the initial stages of PMB1 action on the OM of Gram-negative bacteria, and in particular highlight the regions of the peptide that form strongest interactions with the LPS molecules of the outer membrane. In future, a multiscale simulation approach based on these studies may provide further insights regarding mechanism of action.
In conclusion, in this study we have shed light on the potential mechanisms for bacterial envelope disruption by PMB1. The aggregation witnessed in the OM models is suggestive of the possible early stages of self-regulated translocation / pore formation, whilst the fatty acid tail insertion in the lipid A environment also appears to be dependent upon aggregation in order to create a charge-free area to allow PMB1 penetration, providing further evidence for a pore model of self-regulated uptake. We may speculate that a ‘ladder’ type mechanism occurs in the context of the full LPS membrane, with the PMB1 molecules initially aggregating on the surface, prior to further penetration through the sugar region by the DAB residues, subsequently disrupting the high surface charge density of the counterion-cross-linked lipid A moieties and resulting in fatty acid tail penetration within the hydrophobic membrane core. A stepwise process for PMB1 disruption of the IM has also been established, beginning with DAB-based adsorption and followed by rapid fatty acid tail insertion within the bilayer, supported by the hydrophobic D-Phe. In this case, deep penetration of monomeric PMB1 molecules enables the DAB residues to drag water into the membrane, suggesting an alternative antimicrobial mechanism for IM destabilisation. Nevertheless, it is possible that other mechanisms (e.g. carpet model) may apply at higher concentrations of PMB1.
All simulations systems are summarised in Table 1.
All simulations performed in this work used the GROMACS molecular dynamics software [53, 54], version 4.5.1 [55]. Standard parameters taken from the GROMOS 53A6 force field [56] were used to model the polymyxin B1 molecule in its fully ionized state. The parameters for the LPS molecules were as described and used previously [27, 57] and the GROMOS-CKP (Chandrasekhar[58]-Kukol[59]-Piggot[27, 60]) parameters were used for the phospholipids. The SPC water model [61] was used in all simulations. During the simulations, the LPS, phospholipids and solvent (water plus counterions) were maintained at a constant temperature of 313 K using the Nosé-Hoover thermostat [62, 63] with a time constant of 0.5 ps. The only exception being the lipid A bilayer simulations which were performed at a temperature of 323 K. These temperatures were chosen as they are above the gel to liquid crystal phase transition temperatures of all the lipids used in the simulations [64–68]. A pressure of 1 bar was maintained using anisotropic pressure coupling with the Parrinello-Rhaman barostat [69, 70] and a time constant of 5 ps. Electrostatic interactions were treated using the smooth particle mesh Ewald (PME) algorithm [71] with a short-range cutoff of 0.9 nm. The van der Waals interactions were truncated at 1.4 nm with a long-range dispersion correction applied to the energy and pressure. The neighbor list was updated every five steps during the simulations. All bonds were constrained using the P-LINCS algorithm [72] allowing a 2 fs time step to be applied. All LPS-containing membrane systems were neutralized with Mg2+ ions, whereas the inner membrane model was neutralized with Na+ ions.
|
10.1371/journal.pgen.1006055 | Essential Roles of Cyclin Y-Like 1 and Cyclin Y in Dividing Wnt-Responsive Mammary Stem/Progenitor Cells | Cyclin Y family can enhance Wnt/β-catenin signaling in mitosis. Their physiological roles in mammalian development are yet unknown. Here we show that Cyclin Y-like 1 (Ccnyl1) and Cyclin Y (Ccny) have overlapping function and are crucial for mouse embryonic development and mammary stem/progenitor cell functions. Double knockout of Ccnys results in embryonic lethality at E16.5. In pubertal development, mammary terminal end buds robustly express Ccnyl1. Depletion of Ccnys leads to reduction of Lrp6 phosphorylation, hampering β-catenin activities and abolishing mammary stem/progenitor cell expansion in vitro. In lineage tracing experiments, Ccnys-deficient mammary cells lose their competitiveness and cease to contribute to mammary development. In transplantation assays, Ccnys-deficient mammary cells fail to reconstitute, whereas constitutively active β-catenin restores their regeneration abilities. Together, our results demonstrate the physiological significance of Ccnys-mediated mitotic Wnt signaling in embryonic development and mammary stem/progenitor cells, and reveal insights in the molecular mechanisms orchestrating cell cycle progression and maintenance of stem cell properties.
| Stem cell self-renewal has two essential elements, cell division and at least of one of the daughter cells retaining stem cell properties, so-called stemness. The interconnections between cell cycle and cell fate specification have been explored in embryonic stem cells. However, less is known about how cell cycle affects the cell fate decision in tissue stem cells. In this study, we explore the function of particular mitotic factors Ccny and Ccnyl1 in regulating the dividing tissue stem cells. The development of the mammary gland occurs mostly in postnatal pubertal stage. At the time, the robustly dividing stem/progenitor cells reside at the forefront of the mammary epithelium extension, underlining the mammary gland as a good model to study the interconnection of cell cycle and tissue stem cells. In this study, we show that in dividing mammary stem/progenitor cells, Ccny and Ccnyl1 enhance Wnt signaling activities in mitosis. The signaling enhancement in this time window is essential for the stem/progenitor cell property maintenance during division. Deletion of Ccnys results in diminishing their competitiveness and developmental potential.
| Stem cell self-renewal is tightly associated with cell cycle progression. In particular, active stem cells rapidly divide and coordinate with cell fate choices [1]. Active stem cells can be replenished by quiescent stem cells over time or upon injury [2]. Distinct from other somatic cell proliferation, the process of self-renewal needs to ensure stemness maintenance. The molecular mechanism overseeing stemness maintenance during division is poorly understood.
Wnt/β-catenin signaling plays a prominent role in adult stem cell self-renewal in many tissues [3]. The level of Wnt/β-catenin signaling in stem cells is meticulously regulated, while different activation levels of the signaling result in distinct fate decisions by stem cells [4–6]. Wnt signaling is initiated upon the binding of Wnt ligands to Frizzled and lipoprotein receptor-related proteins 5 and 6 (LRP5/6) receptors. Consequent abolishing or titrating away of the APC/Axin/GSK3 complex leads to β-catenin accumulation and translocation to the nucleus where it binds to Tcf/Lef family members, activating target gene expression [3]. A key step for stoichiometric regulation of Wnt signaling happens at the membrane, where Lrp6 receptor activation occurs in sequential steps before reaching its full competence. It is well established that Lrp6 is phosphorylated on PPPSP motif by GSK3 and on CK1 site by CK1γ [7,8]. Recently, an additional phosphorylation event of Lrp6 has been identified to precede Wnt ligand stimulation [9]. During mitosis, cyclin Y (Ccny) localized at the plasma membrane recruits Cyclin-dependent kinase 14 (Cdk14) for the phosphorylation of Lrp6 on PPPSP motif, which sensitizes Lrp6 for upcoming Wnt signals [9,10]. The finding of Lrp6 phosphorylation by Ccny/Cdk14 reveals a cell cycle dependent Wnt signaling activation mechanism, adding a new level of complexity to stoichiometric Wnt signaling activation [9,11]. Although enhancing the Wnt-receptor Lrp6 competence by Ccny is important for Xenopus development [9], whether a similar Ccny/Lrp6 regulatory event is physiologically significant in mammals and in stem cell biology is unknown.
The mammary gland is a bi-layered epithelial organ consisting of an inner layer of luminal cells and an outer layer of basal cells (myoepithelial cells). Mammary stem/progenitor cells are Wnt-responsive cells resided in the basal layer [12–14]. The mammary gland develops mostly in the postnatal stage. At the onset of puberty, at around 3 weeks of age in mouse, a rapid expansion of the rudimentary ductal tree begins. Branching ductal morphogenesis proceeds across the entire mammary fat pad, and is completed at approximately 7 weeks [15,16]. During puberty, at the growing tips of the ducts are the highly proliferative terminal end buds (TEBs), which are believed to house active mammary stem cells (or transient amplifying cells), actively cycling to fuel the growth spurt [17,18]. This notion is further supported by a lineage tracing study in which labeled TEB cells undergo massive clonal expansion, giving rise to ample progeny cells in development [13].
In this study, we applied a genetic approach to investigate the roles of Ccny and its paralogue Ccnyl1 in active stem/progenitor cells and to examine their functions in mammary development and regeneration. Our study establishes that mitosis-induced and Ccny family-mediated Wnt signaling activity is essential for keeping the developmental potential of dividing mammary stem/progenitor cells, shedding light in the molecular mechanisms coordinating cell cycle progression and undifferentiated state maintenance.
We first investigated the expression patterns of Ccny and Ccnyl1, hereafter referred to collectively as Ccnys. We found that both Ccnys, which share high similarity in amino acid sequence (S1A Fig), are expressed in many tissues, including the mammary gland (Fig 1A). We generated Ccny and Ccnyl1 polyclonal antibodies and validated their specificity (S1B–S1D Fig). Cell fractionation and Western analyses indicated membrane localization of Ccnyl1, similar to that of Ccny (Fig 1B) [10].
To investigate the function of Ccny, we generated Ccny conditional mutant mice, with two loxP sites inserted to flank exon 4 (Fig 1C and see Methods for details). To create Ccny+/- mutant mice, male Ccnyflox/+ mice were bred with female EIIa-cre mice, inducing recombination in germ cells and transmitting the Ccny deletion to progeny. The resulting Ccny+/- mice went through additional cross to generate Ccny-/- mice (Fig 1C). We confirmed the deletion of Ccny in various tissues by western analyses, and the deletion of Ccny did not affect the level of Ccnyl1 (Fig 1D). The Ccny-/- mice were grossly normal and their mammary glands displayed no discernable phenotypes (S2 Fig), suggesting a possible compensation by Ccnyl1. To address this, we utilized a Ccnyl1 knock-in mouse line (Ccnyl1lacZ/+) generated in EUCOMM, in which a LacZ cassette was inserted into the intron between exon 4 and 5 (Fig 1E). Although the insertion disrupted the Ccnyl1 transcription, Ccnyl1lacZ/lacZ mice were viable and also exhibited normal mammary gland morphology (S2 Fig). The deletion of Ccnyl1 was also validated in multiple tissues while the level of Ccny was not affected (Fig 1F). Interestingly, the Ccnys double knockout mice (Ccny-/-;Ccnyl1lacZ/lacZ; DKO) were embryonic lethal (Fig 1G). At E14.5, Ccnys DKO embryos appeared smaller in body size yet alive (Fig 1H). At E16.5, the DKO embryos harvested were lethal, infiltrated with blood and partially absorbed by the uterus (Fig 1I). Together, these data suggest that Ccny and Ccnyl1 have overlapping functions in development. As neither single mutant displays discernable mammary gland phenotype, functional redundancy likely persists during mammary development.
We examined the expression of Ccnyl1 in the mammary gland using Ccnyl1lacZ/+ mouse. Mammary glands were isolated from pubertal mice (5-week and 6-week old) for whole mount X-gal staining. At this stage, mammary epithelium undergoes active extension. Interestingly, Ccnyl1 expression was enriched at the forefront of the pubertal mammary epithelium extension where TEBs are located (arrows in Fig 2A and 2B). Ccnyl1 expression appeared mostly in basal cells and surrounding stromal cells, but rarely in the inner layer body cells (Fig 2C). It has been reported that several members of the Wnt family are expressed in the mammary gland at this stage [19–21], which could contribute to the proliferative state of TEBs. We examined the Wnt-responsiveness in pubertal mammary glands using Axin2lacZ/+ reporter mouse [22]. We found that Axin2-expressing cells are also enriched in the TEB area (Fig 2D and 2E), with robust staining in the basal cells and surrounding stromal cells (Fig 2F), exhibiting a similar pattern to the Ccnyl1-expressing cells. To address whether Ccnyl1 is expressed in Axin2+ cells, double in situ hybridization were performed, revealing that Axin2 and Ccnyl1 frequently co-localized in basal cells of the TEBs (S3A Fig).
We next investigated whether Ccny expression also has a TEB enriched pattern. We harvested mammary glands from 5-week-old Actin-GFP mice, in which the forefront of the epithelium has extended slightly past the lymph node. Guided by the green fluorescence of GFP, we separated the TEB region from the ducts (illustrated in Fig 2G). Basal (Lin-, CD24+, CD29hi) and luminal (Lin-, CD24+, CD29lo) cells were isolated by FACS from the two compartments for quantitative PCR (qPCR) analysis. We found that Ccny was evenly expressed in the ducts and TEBs, with little difference between luminal and basal cells (Fig 2H). By contrast, Ccnyl1 exhibited a higher expression in TEBs, especially in the basal cell of TEBs (Fig 2H), consistent with the observation in the Ccnyl1lacZ/+ reporter mice (see Fig 2A–2C). Double colored RNA in situ hybridization was then performed to validate Ccny and Ccnyl1 expression in TEBs. We found that, consistent with the qPCR results, Ccny mRNA was detected in both basal and luminal cells, whereas Ccnyl1 mRNA was predominantly distributed in basal cells (Fig 2I). In 8-week-old nulliparous mice, the mammary gland has ceased rapid proliferation and the TEB structure has vanished. At this stage, we detected very rare Ccnyl1 expression in mature mammary ducts (S3B Fig), similar to the Axin2-lacZ expression pattern at this stage (S3B Fig) [12]. Thus, Ccnyl1 is robustly expressed in the basal cell of TEBs, coinciding with Wnt/β-catenin signaling activation.
In light of the overlapping expression of Ccnyl1 and Axin2 in pubertal mammary gland, we set to address whether the expression of Ccnyl1 is induced by Wnt/β-catenin signaling. We cultured the basal cells in 3D matrigel as previously described [12] and found that neither Wnt3A nor Wnt4 (the endogenous Wnt in the mammary gland) was sufficient to induce Ccnyl1 or Ccny expression, while either treatment successfully increased Axin2 mRNA levels (Fig 3A). A gradient of lithium chloride (LiCl) was also used to activate Wnt signaling, yet it failed to stimulate Ccnyl1 or Ccny expression (Fig 3B). Thus, Ccnys are likely not Wnt signaling targets.
Previous study indicates that Ccny is enriched in G2/M phase [9]. We thus examined whether Ccnyl1 level is regulated in a cell cycle dependent manner. We monitored endogenous Ccnyl1 level in drug synchronized mammary epithelial Eph4 cells and observed a prominent elevation of Ccnyl1 in M phase (Fig 3C). For insights into the upstream regulation of Ccnyl1, we examined possible binding sites for transcriptional factors in the Ccnyl1 promoter region. The cell cycle related transcription factors E2Fs and proliferation related transcriptional factors, Egr1 and Elk1, were predicted to associate with the Ccnyl1 promoter. We tested the potency of various transcription factors in activating Ccnyl1 transcription by luciferase reporter assays. The -2.3kb to +0.7kb genomic region of Ccnyl1 was cloned to drive a luciferase reporter gene. We found that E2F1 is able to activate the Ccnyl1 promoter in Eph4 cells (Fig 3D). When similar luciferase assays were performed using the Ccny promoter, however, E2Fs could not activate Ccny-luciferase activities (Fig 3E). Furthermore, chromatin immunoprecipitation (ChIP) analysis confirmed direct association of E2F1 with the Ccnyl1 genomic region containing a putative E2F1 binding site (Fig 3F). Together, these results suggest that Ccnyl1 expression in mammary cells is cell cycle regulated.
Axin2+ cells are enriched for basal mammary stem/progenitor cells [2]. In light of the similar expression patterns of Ccnyl1+ and Axin2+ cells, we next investigated the function of Ccnys on mammary stem/progenitor cells. We utilized the Ccny mutant mice (Ccny-/-) and knocked down the expression of Ccnyl1 by shRNA in the Ccny-/- background. The knockdown efficiency of the shRNA (Sh-Ccnyl1) was validated by Western analysis (S4 Fig). Basal cells were FACS-isolated from Ccny-/- mammary glands and infected with control or Sh-Ccnyl1 lentivirus in suspension. Infected mammary cells were then cultured in 3D Matrigel to allow colony formation. We found that the colony sizes were significantly smaller when the Ccnys expression was inhibited (Fig 4A), with decreased cell proliferation as shown by reduced EdU incorporation (Fig 4B). Next we dissociated the primary colonies to single cells and replated them to examine serial colony formation. We found that knockdown of Ccnyl1 in Ccny-/- mutant cells results in drastically decreased colony numbers in each passage, while the control colony numbers continuously expanded (Fig 4C). We also examined whether Ccnys affect luminal cell colony formation using the same approach in the Ccny-/- background. No differences in luminal colony sizes were observed when Ccnyl1 was knocked down (S5 Fig). Together, these results show that Ccnys are important for the expansions of basal, but not luminal, colonies in vitro.
We next investigated the influence of Ccnys on the regenerative capacity of mammary stem/progenitor cells. We followed the in vivo knockdown method established in previous studies [23,24]. Mammary cells were isolated and infected with lentivirus to express both shRNA and GFP. Infected cells were FACS-isolated using the GFP tag, followed by transplantation into the cleared fat pad of immunocompromised recipient mice. The mammary outgrowths with GFP were examined at 8 weeks post transplantation (Fig 4D). In control experiments, we observed that Ccny+/-;Ccnyl1lacZ/+ mammary cells (loss of 2 copies of Ccnys) infected with a scramble shRNA readily generated outgrowths (Fig 4E). Ccny-/-;Ccnyl1lacZ/+ mammary cells (loss of 3 copies of Ccnys) infected with the scramble shRNA also generated new mammary glands though their outgrowths were smaller (Fig 4E). By contrast, Ccny-/-;Ccnyl1lacZ/+ cells infected with Sh-Ccnyl1 (loss of 3 copies of Ccnys plus RNAi) completely lost the regeneration capabilities and were not able to reconstitute any outgrowths (Fig 4E). Together, these results suggest that Ccnys are critical for mammary stem/progenitor cell self-renewal and regeneration capacity.
To investigate the impact of Ccnys loss in normal development, we generated K14-Cre;Ccnyflox/flox;Ccnyl1lacZ/lacZ;mTmG mouse model to delete both Ccnys using a basal cell specific K14-Cre [25], at the same time tracking the fate of the Ccnys-deficient cells using the mTmG reporter [26]. K14 is activated as early as E15.5. The expressed Cre recombinase in basal cells would result in excision of the stop cassette in the reporter, thus marking the basal cells and their progeny, including luminal cells, with membrane-bound GFP (mG). On the other hand, intact cells would express membrane-bound tdTomato (mT), a red fluorescence protein. In control mice (K14-Cre;Ccnyflox/+;Ccnyl1lacZ/+;mTmG) (Fig 5A), the mammary glands (8-week) were largely labeled with GFP, as evidenced by whole mount imaging as well as histological sections (Fig 5B). FACS analysis indicated that 58% of basal cells and 88% of luminal cells expressed GFP (Fig 5C and 5D), suggesting that the K14-Cre did not induce 100% recombination. By sharp contrast, the mammary glands of the Ccnys-deficient mice (K14-Cre;Ccnyflox/flox;Ccnyl1lacZ/lacZ;mTmG) (Fig 5E) were mostly positive for tdTomato, as indicated by whole mount imaging and sections (Fig 5F). FACS analysis confirmed that GFP+ cells comprises of only 2.5% of basal cells and 2% of luminal cells (Fig 5G). In light of the partial efficacy of K14-Cre (58%) used in this study, we propose that stem cells that had successfully recombinated and lost all copies of Ccnys failed to compete with the normal stem cells in the developing mammary gland (Fig 5H). Since the mammary glands of K14-Cre; Ccnyflox/flox;Ccnyl1lacZ/lacZ;mTmG mice exhibited normal morphology (S6 Fig), we postulated that the remaining normal stem cells are able to generate the whole mammary gland during the development and thus render the animals phenotypically silent. In fact, the replacement of cell population through cell competition can often be phenotypically silent as previous reported [27–29]. Together with the in vitro and transplantation data, these results suggest that loss of Ccnys impairs the function of mammary stem/progenitor cell, thereby affecting their contribution to differentiation during development.
We next investigated the molecular mechanism by which Ccnys control the activities of mammary stem/progenitor cells. Ccny has been implicated in Lrp6 phosphorylation at Ser1490 (pS1490) in HEK293T cells [8,30]. In basal colonies, we observed enriched pS1490 in M phase cells (Fig 6A), suggesting that the cell cycle induced Lrp6 phosphorylation occurs during basal stem/progenitor cell expansion. To address the contribution of Ccnys in this event, we knocked down Ccnyl1 by shRNA in Ccny-/- mammary cells in order to completely inhibit the expression of Ccnys. We observed that the levels of Lrp6 phosphorylation at S1490 and active form of β-catenin were reduced in these cells (Fig 6B). In embryonic fibroblasts (MEFs) isolated from the Ccnys DKO mice, similar reduction in pS1490 was also observed (S7 Fig). Thus Ccnys are important for the S1490 phosphorylation of Lrp6 and may function in mammary basal stem/progenitor cells through Lrp6 activation.
We tested whether activation of Wnt/β-catenin signaling can restore the phenotypes induced by the loss of Ccnys. As described in Fig 4, knockdown of Ccnyl1 in a Ccny-/- background resulted in decreased basal colony sizes in 3D culture. We found that inhibiting GSK3β by LiCl restore the size of basal colonies to a normal range, while Wnt3A is ineffective (Fig 6C), supporting the notion that Ccnys function at the level of Lrp6 phosphorylation [9,11].
Next, we assessed the rescue capability in regeneration assays. We induced constitutive Wnt/β-catenin signaling activation using the Ctnnb1flox(ex3)/+ allele [31]. Mammary cells from Ctnnb1flox(ex3)/+ mice were infected with Cre adenovirus to induce recombination (Fig 6D). Simultaneously, we used mCherry-tagged Sh-Ccny and GFP-tagged Sh-Ccnyl1 lentivirus to knockdown expression of both Ccnys (Fig 6D; S4 Fig). The appearance of the N-terminal truncated form of stabilized β-catenin (β-cateninΔex3) (Fig 6E) indicated that Wnt signaling is constitutively activated in the Cre-expressed mammary cells. By contrast, no β-cateninΔex3 band was detected in the cells infected with a control adenovirus (Fig 6E). In order to assess their capability in regeneration, the infected cells were FACS-isolated and transplanted into the recipient cleared fat pads. We found that Ctnnb1flox(ex3)/+ mammary cells infected with the shRNAs were not able to reconstitute any outgrowth (Fig 6F), due to loss of Ccnys. When Ctnnb1flox(ex3)/+ mammary cells were infected with both the shRNAs and Adeno-Cre, however, they efficiently generated outgrowths, often with hyperplasia formation (Fig 6F), a phenotype reminiscent of activation of Ctnnb1flox(ex3)/+ alone [32], indicating that the constitutively active β-catenin restores the regenerative capability in the Ccnys-deficient stem/progenitor cells. Together, these data suggest that Ccnys’ regulation of Lrp6 activation facilitates Wnt/β-catenin signaling in mammary stem/progenitor cells.
Our study demonstrates the physiological significance of Ccnys in embryonic development and developing mammary gland. While the detailed cause of the embryonic lethality requires further investigation in the future, our results establish that mitosis-induced Wnt signaling enhancement is essential for keeping the properties of dividing mammary stem/progenitor cells. Although Ccny and Ccnyl1 showed different expression patterns in developing mammary gland (Fig 2), they are functionally redundant. In addition to single knockout (S2 Fig), female mice with only one allele of either Ccny or Ccnyl1 were not only viable (Fig 1G) but also produced functional mammary glands. Only upon full knockout did the mammary cell fail to contribute to the mammary gland formation (Fig 5). Consistently, Ccny-/-;Ccnyl1+/lacZ basal cells still reconstituted mammary outgrowths upon transplantation until the residual Ccnyl1 was further depleted by RNAi (Fig 4D and 4E). These results indicate that in mammals not only the gene numbers (two genes) but also the copy numbers (four alleles) are redundant for Ccnys, at least in most events of the development. This possibly serves as a way to safeguard this important layer of Wnt regulation.
The interconnections between cell cycle progression and cell fate specification have been explored in embryonic stem cells (ESCs) owing to the robust in vitro culture systems. The pluripotent status and differentiation propensity of ESCs are determined by specific cell-cycle profiles [1,33–35]. The importance of the cell cycle towards adult stem cells has also been documented in a variety of organs [2,36,37]. Notably, CDKs, together with their regulatory subunits cyclins, are involved in coupling cell cycle with stemness. G1 Cyclins, such as cyclin D, can impact the tendency and capacity of neural stem cells and hematopoietic stem cells to differentiate [37,38], and CDK6 regulates hematopoietic stem cell quiescence exit [39]. Our study adds a new episode by showing that the Ccnys-enhanced Wnt signaling activities in M phase is essential for dividing mammary stem/progenitor cell to maintain their competitiveness and developmental potential.
In adult stem cells, Wnt signaling activation can impact the adult stem cells by promoting cell-cycle entry through induced expression of the G1 factor, c-Myc and cyclin D1, which function as switches between quiescence and division (in intestine) [40,41]. Our data indicate that the reverse is also true, particularly the cell cycle can impact the adult stem cells by enhancing Wnt signaling activities during mitosis, which is crucial in maintaining the stem/progenitor cell properties and the progeny cell fate decision (in mammary gland). Our data highlight a cell cycle and Wnt signaling feed forward mechanism in active stem/progenitor cells for their expansion.
In many in vitro expansion systems of adult stem cell, both Wnts and mitogenic growth factors are required for sustaining the culture [12,42–45]. Wnt proteins are not required for proliferation in these contexts, indicating that inhibition of differentiation is its main function in self-renewal [12,46]. This leads to a simplified view that mitogenic growth factors are responsible for pushing the cells into division. In light of the feed forward mechanism, mitogenic growth factors also partake in keeping the stem/progenitor cell properties by influencing the output of Wnt signaling during cell division.
Many parameters can trigger cell competition, including differences in protein synthesis rates, growth factor receptivity and the expression level of Myc [27]. The replacement of cell population through cell competition is phenotypically silent, because the competitor cells conform to size-control mechanisms [27–29]. In this study, we found that deletion of Ccnys in a subset of mammary stem/progenitor cells diminishes their capability in generation of progeny cells and contribution to development, due to loss in competition with wild type stem cells. Hence, the Wnt signaling enhancement mediated by Ccnys is critical for dividing stem/progenitor cells to retain their competitiveness and full potency. This process may occur naturally, as it might provide a mechanism for elimination of suboptimal stem/progenitor cells during development.
In conclusion, our findings here establish the importance of Ccnys in keeping the stem/progenitor cell properties and contribute to a better understanding of cell cycle control of Wnt signaling activation in cycling mammary stem/progenitor cells.
All procedures were carried out in accordance with the Chinese guidelines for the care and use of laboratory animals. Experimental procedures were approved by the Animal Care and Use Committee of Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (SIBCB-NAF-15-002-S335-005).
Experimental procedures were approved by the Animal Care and Use Committee of Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. All mice were maintained in the specific-pathogen-free animal facility. The Ccny flox mice were constructed in Shanghai Research Center For Model Organisms. The Ccny flox mice were maintained on a 129/B6 mixed background. The Ccnyl1 knockout-first (kof) mice were kindly provided by EMMA (Stain ID, EM:04396. Stain name, C57BL/6NTac-Ccnyl1<tm1a(EUCOMM)Wtsi>/H), and maintained on a C57BL/6 background. The Axin2lacZ/+ mice, K14-Cre mice, Rosa26mTmG/+mice and Ctnnb1fl(ex3)/+ mice were described previously [22,26,31,47]. The EIIa-Cre mice were maintained on a FVB background. Genotyping analyses were performed by PCR with genomic DNA extracted from tail tips. For embryos studies, pregnancies were obtained by natural mating and were timed from the day of the vaginal plug, which was defined as embryonic day (E) 0.5. Embryos were dissected from uterus and then photographed under a dissection microscopy (Olympus SZX16). Primary mouse embryonic fibroblasts (MEFs) were isolated from E14.5 embryos and cultured as previously described [48].
The full lengths of mouse Ccny (NM_026484.3) and Ccnyl1 (ENSMUSG00000070871) cDNAs were amplified by PCR. Ccny and Ccnyl1 cDNA was cloned into pcDNA-HA to express HA-Ccny and HA-Ccnyl1 fusion proteins. To express GST or 6×His tagged Ccnyl1, its cDNA was cloned into pGEX-4T-1 or pET28a vector. For lentiviral shRNA constructs, the annealed oligonucleotides were inserted into pLKO.1, modified by replacing the puromycin-resistance gene with a cDNA encoding GFP or mCherry. The sequences of shRNAs were as follows: Scramble shRNA: 5'-TCCTAAGGTTAAGTCGCCCTCG-3'; Ccnyl1 shRNA: 5'-GCTCATGCTCAACAATATTTC-3'; Ccny shRNA: 5'-GCAAGAGTCTCTTCATTAACCC-3'.
GST tagged Ccnyl1 (77–367 aa) and 6×His tagged Ccnyl1 (77–367 aa) were expressed in E.Coli BL21 codon plus strain for anti-Ccnyl1 antibody generation and purification. Anti-Ccny antibody was as prepared previously described [10]. Other antibodies used were as follows: anti-β-catenin antibody (C-terminus) (BD, 610153); anti-β-catenin antibody (N-terminus) (Abclonal, A2064); anti-active β-catenin antibody (Millipore, 05–665); anti-Cre antibody (Novagen, 69050–3); anti-GAPDH antibody (Proteintech, 10494-1-AP); anti-Flag antibody (Sigma, F3165); anti-β-actin antibody (Sigma, 5316); anti-Lrp6 antibody (Cell Signaling Technology, 3395S); anti-phospho-Histone H3 Ser10 antibody (Cell Signaling Technology, 3377S); anti-phospho Lrp6 Ser1490 antibody for western blot (Cell Signaling Technology, 2568); anti-phospho Lrp6 Ser1490 antibody for immunofluorescence (gift from Christof Niehrs lab); anti-E2F1 IgG (Santa Cruz, sc-193).
Mammary glands were dissected and stained with X-gal as described [49]. Briefly, mammary glands were dissected and washed once with PBS. Mammary glands were fixed at room temperature for 2 h with β-galactosidase fixative (0.2% glutaraldehyde, 1.5% formaldehyde, 5 mM EGTA, 2 mM MgCl2 in PBS), and then washed 3 times in wash buffer (2 mM MgCl2, 0.01% sodium deoxycholate and 0.02% Nonidet P-40 in PBS) for 15 min each time. Finally, mammary glands were stained overnight in staining solution (1 mg/ml X-gal, 5 mM K4Fe(CN)6, 5 mM K3Fe(CN)6 and 2 mM MgCl2 in PBS) at 30°C. After X-gal staining, mammary glands were washed with PBS for several times, stained with carmine alum, and then dehydrated in 50%, 70%, 95%, 100%, 100% ethanol and Histoclear. Whole mount analyses were performed under a dissection microscope (Leica). To localize lacZ+ cells easily, mammary glands embedded in paraffin were sectioned for 10 μm thick. The sections were de-waxed in Histoclear, rehydrated in 100%, 95%, 85%, 75%, 50%, 30% ethanol, and counterstained with nuclear fast red, and followed by a serial of dehydration in 30%, 50%, 75%, 85%, 95%, 100% ethanol, and cleared in Histoclear before sealed by coverslip. The stained samples were photographed with an Olympus BX51 microscope equipped with an Olympus DP71 cooled CCD camera.
For whole mount analyses in conditional knockout mice (cKO), observation was made under a fluorescence dissection microscopy (Leica). After analysis, the mammary glands were processed for frozen sections or FACS analysis.
The terminal end-buds of mammary glands from 5 week-old female mice were fixed with 10% neutral buffered formalin at room temperature for 36 h and then prepared for paraffin sections of 7 μm thick. In situ hybridization was performed using the RNAscope kit (Advanced Cell Diagnostics) following the manufacturer’s instructions. Axin2, Ccny and Ccnyl1 probes were ordered from Advanced Cell Diagnostics.
Mammary glands were isolated from 8- to 12-week-old virgin or other specified-stage female mice. The minced tissue was placed in culture medium (RPMI 1640 with 25 mM HEPES, 5% fetal bovine serum, 1% penicillin-streptomycin-glutamine, 300 U ml−1 collagenase III (Worthington)) and digested for 2 h at 37°C. After lysis of the red blood cells in NH4Cl, a single-cell suspension was obtained by sequential incubation with 0.25% trypsin-EDTA at 37°C for 5 min and 0.1 mg ml−1 DNase I (Sigma) for 5 min with gentle pipetting, followed by filtration through 70-μm cell strainers. For cell labeling, the following antibodies were used: FITC-conjugated CD31, CD45, TER119 (BD PharMingen); CD24-PE/cy5, CD29-APC (Biolegend). Antibody incubation was performed on ice for 15 min in HBSS with 10% fetal bovine serum. All sortings were performed using a FACSJazz (Becton Dickinson). The purity of sorted population was routinely checked and ensured to be more than 95%.
FACS-sorted cells were infected with lentivirus overnight, and resuspended at a density of 4 × 105 cells ml−1 in chilled 100% growth-factor-reduced Matrigel (BD Bioscience). The mixture was allowed to polymerize before being covered with culture medium [DMEM/F12, ITS (1:100; Sigma)] which was changed every 24 h. Primary colony numbers were counted and the diameters were measured after 5–7 days in culture. The colonies were mostly spherical, if colony was oval, the long axis was measured. LiCl (Sigma) was added to the culture medium from day 3. For Wnt treatment, 200 ng ml−1 Wnt3A or Wnt4 conditional medium (1:50 conditional medium from Wnt4-expressing EpH4 stable cell line) was added from day 1. For passaging colonies, the medium was aspirated, and Matrigel was digested by incubation in 500 μl of Matrigel recovery solution (BD Bioscience) for 1 h on ice. Colonies released from Matrigel were harvested after pelleting. Single cells were obtained through incubation in 0.25% Trypsin-EDTA for 5 min at 37°C followed by gentle pipetting. Single cells were then replated in Matrigel as described above.
Mammary cells were prepared, infected by indicated virus, and then cultured in monolayer. After 5–7 days, cells were digested with trypsin, sorted by FACS, and resuspended in 50% Matrigel, PBS with 20% FBS, and 0.04% Trypan Blue (Sigma), to be injected in 10-μl volumes into the pre-cleared fat pads of 3-week-old female nude mice. Reconstituted mammary glands were harvested after 8–10 weeks post surgery. Outgrowths were detected under a fluorescence dissection microscope (Leica). Outgrowths with more than 10% of the host fat pad filled were scored as positive.
HEK293T cells, AD-293 cells, and EpH4 cells were cultured in DMEM (high glucose, Hyclone) supplemented with 10% fetal bovine serum, 100 units/ml streptomycin, 100 units/ml penicillin, and 0.3 mg/ml L-glutamine at 37°C and 5% CO2.
Plasmids were prepared using UNIQ-500 (Sangong Biotech). Lentivirus was packaged in HEK293T cells as described [50]. Briefly, HEK293T cells were co-transfected with vesicular stomatitis virus G, packaging plasmid Delta8.9, and transfer vector, using the conventional calcium phosphate method. At 48 h post-transfection, the culture medium was harvested and prepared for ultracentrifugation. The pellet of lentivirus was resuspended in PBS. Adenovirus particles were prepared in AD-293 cells and concentrated using CsCl gradient centrifugation as described [51]. Purified virus particles were stored at -80°C.
EpH4 cells were synchronized to specific cell cycle stage by drug treatment. Briefly, cells were synchronized to late G1 phase by 800 μM mimosine treatment for 18 h or to S phase by administration of 2 mM thymidine for 18 h. Mitotic cells were harvested by shake-off after 200 ng ml−1 nocodazole treatment for 4 hrs.
Cell fractionation was performed using the Membrane and Cytosol Protein Extration Kit (Beyotime, P0033). Briefly, unsynchronized EpH4 cells were lysed, and nucleus was depleted by centrifugation for 10 min at 700 g at 4°C. The post-nuclear supernatant of EpH4 cells was fractionated by ultracentrifugation for 30 min at 14,000 ×g at 4°C into cytosol (C) and membranes (M). Equal-volume fractions of C and M were analyzed by western blotting.
Total RNA was extracted using RNAiso plus (Takara), and the PrimeScript RT Master mix kit (Takara) with oligo(dT) primers was used for the reverse transcription reaction. Quantitative RT-PCR (qPCR) was performed using an Applied Biosystems 7500HT sequence detection system with a FastStart Universal SYBR Green Master Mix kit (Roche). Gapdh served as internal control. The reaction mixtures were incubated at 95°C for 10 min, followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. qPCR primers used in this work:
Ccny-F: 5'-TCTCTTCATTAACCATCATCCTCC-3',
Ccny-R: 5'-AATTTGCTTCTGTTCTGGGT-3';
Ccnyl1-F: 5'-AGTGACGTTGGTTTACTTAGAG-3',
Ccnyl1-R: 5'-GCCTTTCCATCTCATTCATGTC-3';
Axin2-F: 5’-AGCCTAAAGGTCTTATGTGGCTA-3’
Axin2-R: 5’- ACCTACGTGATAAGGATTGACT-3’
Gapdh-F: 5'-AGGTCGGTGTGAACGGATTTG-3',
Gapdh-R: 5'-TGTAGACCATGTAGTTGAGGTCA-3'.
Tissues or cells were lysed with ice-cold RIPA buffer [50 mM Tris-HCl (pH7.5), 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxysodium cholate, 0.1% SDS, 5 mM EDTA, 10 mM NaF. Before use, add 1 mM PMSF, 3 mM dithiothreitol, 1 mM sodium vanadate, and protease inhibitors (Merck)]. Proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes or polyvinylidene fluoride membranes. Immunoblots were developed in chemiluminescence reagent (PerkinElmer Life Sciences) and exposed in a Fujifilm LAS 4000 imager.
Antibodies were diluted in PBS containing 2% BSA. Primary colonies were fixed with 4% paraformaldehyde at room temperature for 10 min. Colonies were then blocked in PBS containing 0.2% Triton X-100 and 10% goat serum for 1 h, followed by incubation with rabbit anti-phospho-Lrp6 Ser1490 antibody overnight at 4°C. The samples were washed with PBS containing 2% BSA for 3 times and incubated with secondary antibody for 1 hr. The samples were finally washed with PBS for 3 times and stained with DAPI. Whole mount fluorescent images of mammary glands were obtained using a Leica MZFLIII dissection microscope. RNA in situ images were acquired using a Zeiss A1-AXIO upright microscope. For confocal imaging, mammary glands were minced and then coverslipped. Confocal Images were acquired through a 40× or 63× oil immersion objective on a Leica TCS SP8 confocal microscope.
EpH4 cells were plated into 24 well tissue culture dishes. After 16–20 h, EpH4 cells were co-transfected with 0.33 μg firefly luciferase reporter plasmid (pGL4.17) containing the promoter region of Ccnyl1, 0.033 μg Renilla luciferase plasmid (pRL-TK) and 0.67 μg pcDNA3-HA-E2Fs or the other transcription factors in per well of the 24-well plate. After 48 h, the cells were lysed and luciferase activities were measured with Dual-Luciferase Reporter Assay System (Promega, Madison, USA).
ChIP analysis was performed as previously described [52]. Briefly, EpH4 cells were cross-linked with 1% formaldehyde for 10 min at room temperature. For each group, a 10-cm dish of EpH4 cells with 80% confluency were lysed and chromatin was sonicated in a sonicator for 30 min (with 7-sec sonication and 7-sec rest alternatively). Sonicated chromatin was then diluted and immunoprecipitated with anti-E2F1 IgG or rabbit normal IgG (Santa Cruz, sc-2027). Immunoprecipitation products and input were analyzed by quantitative PCR using the following specific primers:
Site a,
forward:5’-CAGCTCGAGATGAATGGAAACC-3’,
reverse: 5’-TAGCCAATCAGACCCGGACTTC-3’.
Site b,
forward: 5’-CAGCAATGTCTCCATGTCACAT-3’,
reverse: 5’-CCCATGAGCACAACACAATTTC-3’.
Results are presented as mean ± s.d., unless otherwise stated. Differences were considered significant when P<0.05 in an unpaired Student’s t-test. Three independent experiments were carried out for statistic results unless specified otherwise.
|
10.1371/journal.pcbi.1000644 | SnugDock: Paratope Structural Optimization during Antibody-Antigen Docking Compensates for Errors in Antibody Homology Models | High resolution structures of antibody-antigen complexes are useful for analyzing the binding interface and to make rational choices for antibody engineering. When a crystallographic structure of a complex is unavailable, the structure must be predicted using computational tools. In this work, we illustrate a novel approach, named SnugDock, to predict high-resolution antibody-antigen complex structures by simultaneously structurally optimizing the antibody-antigen rigid-body positions, the relative orientation of the antibody light and heavy chains, and the conformations of the six complementarity determining region loops. This approach is especially useful when the crystal structure of the antibody is not available, requiring allowances for inaccuracies in an antibody homology model which would otherwise frustrate rigid-backbone docking predictions. Local docking using SnugDock with the lowest-energy RosettaAntibody homology model produced more accurate predictions than standard rigid-body docking. SnugDock can be combined with ensemble docking to mimic conformer selection and induced fit resulting in increased sampling of diverse antibody conformations. The combined algorithm produced four medium (Critical Assessment of PRediction of Interactions-CAPRI rating) and seven acceptable lowest-interface-energy predictions in a test set of fifteen complexes. Structural analysis shows that diverse paratope conformations are sampled, but docked paratope backbones are not necessarily closer to the crystal structure conformations than the starting homology models. The accuracy of SnugDock predictions suggests a new genre of general docking algorithms with flexible binding interfaces targeted towards making homology models useful for further high-resolution predictions.
| Antibodies are proteins that are key elements of the immune system and increasingly used as drugs. Antibodies bind tightly and specifically to antigens to block their activity or to mark them for destruction. Three-dimensional structures of the antibody-antigen complexes are useful for understanding their mechanism and for designing improved antibody drugs. Experimental determination of structures is laborious and not always possible, so we have developed tools to predict structures of antibody-antigen complexes computationally. Computer-predicted models of antibodies, or homology models, typically have errors which can frustrate algorithms for prediction of protein-protein interfaces (docking), and result in incorrect predictions. Here, we have created and tested a new docking algorithm which incorporates flexibility to overcome structural errors in the antibody structural model. The algorithm allows both intramolecular and interfacial flexibility in the antibody during docking, resulting in improved accuracy approaching that when using experimentally determined antibody structures. Structural analysis of the predicted binding region of the complex will enable the protein engineer to make rational choices for better antibody drug designs.
| High resolution structures of protein-protein complexes are necessary for understanding mechanisms of protein-protein interactions, analyzing mutations, and manipulating binding affinity [1]. The large gap between the number of experimentally determined complex structures and the available sequences of pairs of protein complexes underscores the challenges, time required and cost of x-ray crystallography or nuclear magnetic resonance approaches. The paucity in complex structures can be alleviated by computational docking, i.e., the prediction of protein-protein complexes, which potentially provides a fast and efficient alternative route. To predict the structure of a protein-protein complex, computational docking requires the structures of the interacting partners. However, sometimes even the structures of the monomeric units are unavailable, forcing the use of a homology modeled structure for one or both partners [2],[3]. Given the inaccuracies in a homology model, current computational docking strategies can determine the gross structural features of a complex, but find it exceedingly challenging to successfully predict high resolution structures of such protein-protein complexes, pointing to the need to develop new docking algorithms which incorporate the necessary degrees of freedom to compensate for the inaccuracies.
Antibody-antigen (Ab-Ag) complexes provide a model system where much needed high-resolution computational docking predictions are challenged by inaccuracies in antibody homology models. The selection of antibodies for exploring homology model docking simplifies the problem by isolating the various degrees of freedom according to prior knowledge of the uncertainties in an antibody homology model, namely the conformation of the complementarity determining region (CDR) loops (L1, L2, L3 in the light chain, and H1, H2, H3 in the heavy) [4],[5], the hyper-variability of the CDR-H3 loop [6]–[9], and the relative orientation of the antibody light (VL) and heavy (VH) chains [6], [10]–[12]. A recent study by Narayanan et al found that the VL-VH relative orientation has a significant impact on the antigen binding properties of an antibody [10], suggesting that simultaneous optimization of the VL-VH relative orientation and antibody-antigen relative orientation might capture some of the intramolecular changes undergone by an antibody upon antigen binding.
An additional motivation for studying antibody-antigen complexes is that therapeutic antibodies are revolutionizing healthcare [13]. Oncology, arthritis, immune and inflammatory disorder treatments have benefitted from newly developed therapeutic antibodies [14]. Success of several therapeutic antibody drugs has relied on homology modeling. According to Schwede et al., homology modeling played an important role in the development of 11 of the first 21 marketed antibodies including Zenapax (humanized anti-Tac or daclizumab), Herceptin (humanized anti-HER2 or trastuzumab), and Avastin (humanized anti-VEGF or bevacizumab) [15]. High-resolution computational docking can aid in the design of antibody biologics by providing insights into the complex interactions between an antibody and an antigen [16]. The importance of antibodies combined with the detailed knowledge of the flexibility in the various segments of an antibody fragment variable (FV) region make them ideal candidates for the development of novel flexible docking algorithms.
Although there are currently no flexible docking algorithms tailored for antibodies, there have been several efforts to incorporate some of the relevant modes of internal flexibility during docking. Early [17],[18] and more recent approaches [19] use hinges to account for internal flexibility. Multi-body docking, which might be useful for optimizing assembly of VL and VH and antigen chains, has been explored in a few case studies [20],[21] including some targets in the blind prediction challenge known as the Critical Assessment of PRediction of Interactions (CAPRI) [22]. Another genre of docking algorithms like HADDOCK [23] allows flexibility in the side chains and backbones to allow for conformational rearrangements in the interaction surface. Wang et al.'s modifications to RosettaDock allow simultaneous gradient-based minimization of the backbone torsional angles and the protein-protein rigid-body orientation [24]. We recently developed a RosettaDock generalization called EnsembleDock [25] that follows the conformer-selection plus induced-fit model [26] to additionally enable docking of a pre-generated ensemble of structures.
We also recently developed RosettaAntibody [27], an antibody FV region homology modeling protocol which incorporates refinement of VL-VH relative orientation, CDR H3 loop modeling and minimization of all the CDR loops. RosettaAntibody generates ten antibody homology models for each input sequence, and this set of models can be used simultaneously with EnsembleDock. However, errors in the CDRs of RosettaAntibody homology models (particularly H2 and H3) can still frustrate docking, and only the ten pre-generated backbone conformations are sampled during ensemble docking [27].
In this paper, we discuss the development and implementation of SnugDock, a new antibody docking algorithm built upon RosettaDock using the sampling components of RosettaAntibody. The new protocol performs multi-body docking by allowing simultaneous structural optimization of the relative orientations of antibody-antigen and VL-VH. SnugDock simulates induced fit by additional simultaneous optimization of the binding interface by allowing flexibility of CDR loops and interfacial side chains. Moreover, we combine SnugDock with EnsembleDock to synthesize a new docking algorithm that encompasses conformer selection, multi-body docking, and a flexible paratope. We test SnugDock using antibody homology models obtained from two accessible public servers, namely RosettaAntibody [28] and Web Antibody Modeling (WAM) [29]. Our goal is to achieve reliable high-resolution antibody docking starting from only the antigen unbound structure and the antibody sequence.
Figure 1 summarizes the SnugDock algorithm as incorporated in RosettaDock. Like RosettaDock, SnugDock is divided into a low-resolution and high-resolution stage. In the low-resolution phase (shown in shades of green), SnugDock adds to RosettaDock by additionally perturbing and minimizing the CDR H2 and H3 loops. In each iteration of the high-resolution Monte Carlo-plus-minimization loop (shown in shades of blue), SnugDock randomly chooses a trial move from a set including the various degrees of freedom in both the antibody conformation space and the docking space. Specifically, the trial move set consists of: i,ii) RosettaDock-like rigid body transformations and minimization of either the antibody-antigen or the VL-VH orientation; iii) gradient-based minimization of the CDRs L1, L2, L3 and H1 backbones; and iv,v) perturbation and minimization of the backbones of CDRs H2 or H3. The high-resolution iterations also include side-chain rotamer packing steps before each minimization and Monte Carlo Boltzmann acceptance decision (see Methods).
For various tests of the docking algorithm, we input either the crystal structure of the antibody, the lowest-energy (lowest-scoring) RosettaAntibody homology model, the ensemble of the ten lowest-energy RosettaAntibody homology models, or the WAM model. The antibody is docked to the unbound crystal structure of the antigen when available. In the following sections, we first detail the results of a case study as we build up the algorithm, then we summarize the performance of different algorithms on the whole set of antibodies. Next, we delve into the structural details of the sampling. Finally we summarize the results starting from WAM antibody models and the results of global docking.
A longstanding source of homology models is the Web Antibody Modeling server (WAM) created by Whitelegg and Rees [29]. The WAM server grafts antibody components together and models the H3 loop de novo. SnugDock may be able to compensate for the model errors during docking. Ensemble docking is not possible since WAM returns only one model for a given sequence. Table 2 presents the accuracy of docking predictions obtained by using WAM antibody homology models using both standard RosettaDock and using SnugDock. The lowest-interface-energy docking decoy generated by standard rigid-body docking simulations using WAM homology models are almost all incorrect. Subjecting the WAM models to SnugDock resulted in six medium quality lowest-interface-energy docking decoys. Thus SnugDock recovers more accurate docking predictions. The original WAM homology models showed strain in the molecule reflected in high Rosetta scores. By subjecting the homology model to minimization on the paratope degrees of freedom, SnugDock was able to relieve intramolecular and inter-chain steric clashes. Interestingly, the SnugDock results with WAM models are comparable to those with RosettaAntibody models, while the use of EnsembleDock-plus-SnugDock with the RosettaAntibody models achieves higher accuracy.
Local docking is often of interest in antibody applications since epitope information can be obtained by a variety of other methods. However, global docking is a computational alternative for producing epitope information when it is unknown. Global docking can be significantly more challenging because of the larger conformational space to search. Further, flexible docking creates additional danger of creating an unrealistic induced fit at a non-native docking location, resulting in a false positive prediction [41],[42]. Global docking is considerably more computationally demanding, and thus we restricted our tests of global docking to five targets, and to simulate a practical docking application, chose only those targets for which unbound crystal structures were available for both the antibody and the antigen: 1MLC, 1AHW, 1JPS, 1WEJ and 1VFB. The starting structures consisted of the unbound crystal structure of the antigen and the lowest-interface-energy RosettaAntibody homology model. The EnsembleDock and the EnsembleDock-plus-SnugDock protocols used the ten lowest-interface-energy RosettaAntibody homology models. For each target, we generated 5000 candidate structures, with each prediction run beginning from a random global rotation of the antigen and a small perturbation of the antibody (to keep the paratope generally directed toward the antigen).
Global docking using standard rigid-body RosettaDock generated no acceptable quality lowest-interface-energy decoys for any targets, and a top-ten ranked acceptable decoy for only one target (Table 3). Using EnsembleDock or SnugDock independently produced marginal improvement with a few acceptable quality predictions. The combination algorithm of EnsembleDock-plus-SnugDock generated two medium quality lowest-interface-energy predictions exhibiting the synergy already established in the local docking simulations. Additionally, the most native-like decoy in the ten lowest-interface-energy decoys was of medium quality for three of the five targets and acceptable for one target. The results are comparable to global docking using standard rigid-body RosettaDock with unbound crystal starting structures of the antibodies, where one structure had a high quality (1JPS), two had medium quality, and the others had acceptable quality prediction for the most native-like decoy in the ten lowest-interface-energy decoys. The results can also be compared with local docking using EnsembleDock-plus-SnugDock around the known epitope (Table 1), where four of the five targets had at least a medium quality prediction for the most native like decoy in the ten lowest-interface-energy decoys, and the fifth target (1VFB) had an acceptable prediction. Thus one target (1VFB) which had succeeded in local docking failed in global docking due to low-scoring non-native decoys (see docking energy landscapes, Supporting Information Figure S2). In general, addition of SnugDock increases sampling of more native-like decoys, enabling near-natives to be energetically more favorable.
SnugDock is the first docking algorithm with targeted antibody flexibility. SnugDock models flexible loop conformations, backbone motions, and inter-chain (VH-VL) adjustments. The introduction of flexibility during docking is critical to overcome the inaccuracies inherent in a homology modeled antibody structure. Comparison of algorithms shows that increasing the degrees of freedom in local docking gradually increases the quality of predictions. Ultimately, EnsembleDock-plus-SnugDock with homology models achieves accuracy comparable to docking crystal structures with standard RosettaDock. While the algorithm is limited to antibody-antigen interactions, the results suggest that it is possible for computational predictions to use homology models to bridge the gap between the number of experimentally determined complex structures and the available sequences of pairs of interacting proteins.
In CAPRI rounds 1–18, eleven of the forty targets involved docking of at least one homology modeled partner (both partners were homology models for Target 35) [43]–[45]. For six of the eleven targets, none of the participating groups could predict any medium or higher accuracy structures. When the sequence identity was under 40%, the solutions were of acceptable quality at best (Targets 20, 24). High quality predictions were obtained only for two cases (Targets 14, 19) and in both cases the binding region of the template structure was structurally similar to the co-crystal structure [44], and the other docking partner was in the bound conformation. The poor performance of homology modeled docking partners in CAPRI highlights the need of docking algorithms like SnugDock which are robust to inaccuracies in a homology model. Targets 20 and 24 with only acceptable predictions had poorly modeled loop and C-terminal regions which were responsible for key contacts in the native binding interface [45], showing that using homology models with loops at the binding region makes docking with homology models even harder. SnugDock with its loop relaxations at the binding interface addresses the challenge toward accurate high-resolution predictions.
The CDR H3 loop of antibodies provides the most diversity and is thus a focus of the conformational sampling in the SnugDock algorithm. In our experience with the RosettaAntibody Server [28], there are antibodies with non-H3 loops which elude the traditional Chothia classification system [4] and may not fit into canonical CDR templates. The SnugDock algorithm is easily generalizable to include perturbations of loop conformations for any of the six CDR loops. Extra sampling however should be restricted to special cases for efficiency and to avoid issues with over-optimized, non-native induced-fit structures. For approaching non-antibody targets, the SnugDock methods would need to be adapted requiring knowledge of a binding site and appropriate choices of loops to target flexibility. The multi-chain docking methods can be applied to any multi-chain docking partner.
The flexible docking methods help in identifying the correct docked complex structure, but unfortunately they do not yet help in refining the monomer homology structures themselves closer to the crystal backbone conformations. This limitation likely arises from the vast conformational space of the backbone and the difficulties with high-resolution refinement of protein structures [46],[47]. In docking, some of the energetic issues are avoided through the use of the interface energy instead of the total energy. Further advancements in refinement techniques will be needed to address this shortcoming. SnugDock's advancements in the sampling problem also reveal continuing issues with the knowledge of nature's energy function. Missing water molecules affected the prediction of targets 1ZTX and 1VFB. Antibody interfaces in general are polar [48], and several targets with the most polar interfacial CDRs (1VFB, 1JHL, 1NCA) failed perhaps due to the challenges in modeling electrostatics [49].
Experimental techniques for epitope mapping like hydrogen-deuterium mass spectroscopy [50] can help to pre-orient an antigen for local docking. However, when such data are not available, one must resort to global docking where the docking simulations are started with random orientations of the docking partners. Our limited testing of global docking encouragingly suggests that the EnsembleDock-plus-SnugDock approach can successfully find epitopes. Global searches should still be performed with care as the large conformation space can frustrate the ability to find the native binding interface or obscure it through false positive non-native interaction.
One could envision a complete computational antibody engineering pipeline starting from the antibody sequence and ending with accurate predictions for optimized antibody-antigen interactions. In this paper we have been successful in reaching the second step by computational docking using computational models of the monomer antibody. The next stages may be additionally challenging. High-resolution complex structures might next be used for computational alanine scanning [51], computational affinity maturation [52] or alteration of binding specificity [53]. For antibody therapeutics, structures will help define drug mechanisms for regulatory approval [15], enable epitope mapping [54] and humanize constructs [55]. These applications require varying amounts of resolution and further testing will reveal the full utility of the SnugDock predictions.
To compare with prior work, we use the set of fifteen antibody-antigen complexes as tested in the original RosettaAntibody publication [27]. The antibody-antigen complex dataset was compiled to ensure: 1) a fair representation of unbound-unbound, unbound-bound and bound-bound antibody-antigen docking targets, 2) a spectrum of CDR H3 loop lengths (7 to 11 residues) and 3) both old and newly released crystal complexes (PDB release dates 1992–2006). The RosettaAntibody and the WAM structures are as reported previously [27].
SnugDock is implemented in the Rosetta biomolecular modeling suite. Fold trees objects [24] are used to guide the propagation of structural changes during docking with backbone flexibility. One fold tree uses flexible jumps for moving the VL-VH and antibody-antigen pairs relative to each other. A second fold tree for CDR loop relaxation had fixed jumps joining the loop stems, and cuts at the middle of the loops. Move map objects are used to select particular sets of residues for backbone and/or side-chain torsion angle flexibility.
Figure 1 depicts the flowchart for the steps in the SnugDock protocol, implemented as follows. Steps 1 and 2 describe the initial setup, Steps 3–6 describe the low resolution stage and steps 7–13 describe the high resolution stage.
Each decoy of an independent docking simulation begins from a different random starting position. In local docking, the set of starting positions comprises a diffuse cloud that covers a reasonable area (∼20 Å rmsd) with moderate density around the native ligand position. In local and global docking, 1,000 and 5,000 candidate docking structures are generated for each target respectively. In our empirical tests, 5,000 decoys were sufficient and results were similar for test runs of 10,000 decoys in a subset of targets. The energy function used during the course of the simulation is as described previously [35] with (i) the interfacial terms of the scoring function including both the antibody-antigen interface and the VL-VH interface, and (ii) chain break penalties for six CDR loops. Interface energy [24] is used to rank and discriminate the structures produced by the docking simulations.
Ligand rmsd is the deviation of the N, Cα, C and O backbone atoms of the antigen after superposition of the antibody backbone atoms. Interface rmsd is the deviation of the backbone atoms at the interface after optimal superposition of those same atoms, where the interface is defined as all residues within 10 Å of a non-hydrogen atom of the other docking partner. Interface energy is the component of the total docking score that arises from inter-molecular residue-residue interactions at the antibody-antigen interface. For fnat calculations, residue-residue contacts are defined when a residue is within 5 Å of a non-hydrogen atom from the other docking partner. The docking models are assigned CAPRI [22]-style “high”, “medium”, “acceptable” or “incorrect” rankings that depend on the rmsd-to-native of the ligand position, the interface rmsd to native and the fraction of native residue-residue contacts (fnat) that are recovered in the docked model [59]. Convergence of a docking simulation is indicated by the presence of a docking funnel, which is defined to exist if there are at least five medium quality predictions in the ten lowest-energy docking decoys.
The SnugDock method presented here is freely available for academic and non-profit use as part of the Rosetta structure prediction suite at www.rosettacommons.org. The distribution includes documentation and full source code. The Rosetta version numbers and command lines used to generate the data are provided in Supporting Information Text S1.
|
10.1371/journal.ppat.1005362 | Structural Based Analyses of the JC Virus T-Antigen F258L Mutant Provides Evidence for DNA Dependent Conformational Changes in the C-Termini of Polyomavirus Origin Binding Domains | The replication of human polyomavirus JCV, which causes Progressive Multifocal Leukoencephalopathy, is initiated by the virally encoded T-antigen (T-ag). The structure of the JC virus T-ag origin-binding domain (OBD) was recently solved by X-ray crystallography. This structure revealed that the OBD contains a C-terminal pocket, and that residues from the multifunctional A1 and B2 motifs situated on a neighboring OBD molecule dock into the pocket. Related studies established that a mutation in a pocket residue (F258L) rendered JCV T-ag unable to support JCV DNA replication. To establish why this mutation inactivated JCV T-ag, we have solved the structure of the F258L JCV T-ag OBD mutant. Based on this structure, it is concluded that the structural consequences of the F258L mutation are limited to the pocket region. Further analyses, utilizing the available polyomavirus OBD structures, indicate that the F258 region is highly dynamic and that the relative positions of F258 are governed by DNA binding. The possible functional consequences of the DNA dependent rearrangements, including promotion of OBD cycling at the replication fork, are discussed.
| A conserved feature of Polyomavirus T-antigens is a phenylalanine situated at the C-termini of their origin-binding domains (OBDs). Using the T-antigen encoded by JC virus, we have investigated why this residue is critical for viral DNA replication. The studies presented herein establish that the consequences of this mutation are limited to the interface formed by the docking of the phenylalanine containing C-terminal pocket region of the OBD with the multifunctional A1/B2 region. Related studies indicate that the conformation of the C-terminal region of the OBD is altered by DNA binding. These observations suggest a model whereby cycling of the OBDs within the hexameric spiral structure at the replication fork is promoted by DNA binding.
| There are presently fourteen known human polyomavirus family members [1, 2]. Reasons for interest in these viruses include the diseases they are associated with, particularly in immunocompromised individuals [3–5]. As examples, JC virus (JCV) causes the often fatal demyelinating disease Progressive Multifocal Leukoencephalopathy (PML) ([6]; reviewed in [7]); Merkel cell polyomavirus causes Merkel cell carcinoma, a rare but highly aggressive skin cancer [8, 9] and BK polyomavirus causes BK nephropathy [10, 11]. Further interest in polyomaviruses stems from the profound insights they have provided into basic cellular process, such as DNA replication (e.g., [12–17]) and the mechanisms that underlie cellular transformation (e.g., [3, 18–20]).
Polyomaviruses have small double stranded DNA genomes [14] that contain a regulatory region that is termed the non-coding control region (NCCR). The NCCR contains the origin of replication as well as the promoter and enhancer elements (reviewed in [21, 22]). An additional feature of polyomavirus genomes is the "early region" that encodes several proteins, including the large T-antigen (T-ag; reviewed in [15, 23, 24]). T-ag is the only virally encoded protein needed for replication; therefore, it has been the target of numerous studies designed to understand its multiple roles during the duplication of the viral genome (reviewed in [12, 15, 25]). For example, polyomavirus T-ag's have been the focus of a number of recent structural studies (reviewed in [25, 26]). These structural studies have provided critical insights into the initiation process, such as establishing how the GAGGC pentanucleotides in the polyomavirus replication origins are recognized by the origin binding domains (OBD) within T-ag [27–32]. Related studies, conducted with the SV40 T-ag OBD, suggest that following site-specific DNA binding, the OBDs undergo rearrangements, including "spiral formation" [32–35]. Additional experiments have revealed the multiple roles played by the T-ag helicase domains during origin melting [36–38], oligomerization [39, 40] and helicase activities ([39, 40]). Based on these studies, models depicting T-ag's multiple roles during the initiation of SV40 DNA replication have been proposed (e.g., [25, 26, 41]).
To further our understanding of the initiation of polyomavirus DNA replication, we recently initiated structural studies of JCV T-ag. In particular, we solved the structure of the JCV T-ag OBD [42]. One of the interesting findings of the JCV T-ag OBD structure was the presence of a C-terminal "pocket". This structure also revealed that the pocket serves as the binding site for T-ag residues from the A1/B2 loops situated on a neighboring OBD subunit [42]. The binding of the A1/B2 loops to the pocket was of interest given that the initial function of the A1/B2 loops is to site-specifically bind the pentanucleotides in the core origin ([43, 44]; reviewed in [45]). These observations provided further evidence that the A1/B2 loops are multifunctional (reviewed in [26]) and that the interaction of the A1 & B2 loops with the pocket is a critical step that takes place at later stages of the initiation process (e.g., during the oligomerization of T-ag on the core origin ([42]; reviewed in [26]).
Additional evidence for the hypothesis that the pocket in the OBD plays a critical role during JCV replication was derived from studies of the JCV T-ag F258L mutant [42]. The F258L "pocket mutation" had no effect on levels of T-ag expression, but for unknown reasons it inactivated T-ag dependent JCV replication [42, 46]. Therefore, given our interest in the JCV T-ag OBD pocket, and its role(s) during JCV DNA replication, we elected to examine the structural basis for the inactivation of JCV replication by the F258L T-ag mutation. The results of these experiments, presented herein, prompted us to examine the dynamic properties of the C-terminal regions of polyomavirus T-ag OBDs. These structure-based analyses indicate that the C-terminal region of the SV40 T-ag OBDs move as a function of DNA binding. The possible biological consequences of these DNA dependent movements are discussed.
An expression vector encoding the JCV T-ag OBD (residues 132–261), that was termed pGEX1λT JC-OBD, was previously described [42]. Using the Quikchange Kit (Agilent Technologies), JCV-OBD residue Phe 258 was mutated to Leu. The oligonucleotides used for the mutagenesis were 5’-GGCCTTAAGGAGCATGACCTTAACCCAGAATAATCG-3’ (the mutated base is underlined) and its complement. The resulting plasmid was termed pGEX1λT JCV-OBD-F258L. The same mutation had been previously introduced into full-length JCV T-ag [42], using the Quikchange Kit, plasmid pCMVneo JCVT-ag and the oligonucleotides listed above. The resulting plasmid was termed pCMV-JCVT-ag F258L. DNA Sequencing at the Tufts University Core Facility (TUCF) confirmed the sequence of the F258L mutants in plasmids pGEX1λT JCV-OBD-F258L and pCMV-JCVT-ag F258L.
Sequence alignments were performed with the program Clustal Omega at the EMBL-EBI website [47]. The aligned sequences were displayed using the program JalView [48].
The wild type JCV T-ag OBD was purified using a previously described protocol [42]. The JCV T-ag OBD F258L mutant protein was purified from BL21 cells using the identical procedures used to purify the wt JCV T-ag OBD [42]. Once purified, the JCV wt and F258L OBD proteins were stored at -80°C in storage buffer (20 mM Tris pH 8.0, 50 mM NaCl, 10% glycerol, 1 mM EDTA, 0.1 mM PMSF and 5 mM DTT).
The subcellular localization of the F258L JCV T-ag mutant within C33A cells was determined by immunofluorescence ([49] and references therein. A detailed protocol describing the steps needed to detect JCV T-ag within C33A cells was previously published [46]). T-ag was detected using the Pab 416 monoclonal Ab (Santa Cruz Biotechnology) and a secondary goat anti-mouse antibody, conjugated with Alexa 488 (Life technologies). The cells were visualized using a Zeiss Axiovert 200M microscope and the data were analyzed using the OpenLab software package from Perkin Elmer.
The ITC studies were conducted with a VP-ITC calorimeter (MicroCal, Northampton, MA). Prior to conducting the ITC studies, the dsDNA oligonucleotide and the JCV T-ag OBD proteins (both wt and the F258L mutant) were buffer-exchanged, using PD-10 columns (GE Healthcare), into 10 mM Sodium Phosphate, pH 7.0, 50 mM NaCl and 5 mM DTT. Protein and DNA concentrations were determined spectrophotometrically, using extinction coefficients calculated with the ProtParam web server and the Integrated DNA Technologies (IDT) website, respectively. The data were analyzed using the Origin software provided by the manufacturer.
Binding isotherms and KD measurements were performed essentially as described [61, 62]. The reactions were performed in 96-well plates (OptiPlate-96 F HB black microplate, Perkin Elmer), in a final volume of 150 μl containing 15 nM of fluorescein-labeled probe and the indicated concentrations of T-ag OBD in buffer containing 20 mM Hepes pH 7.4, 50 mM NaCl, 0.01% NP40 and 0.1 mM DTT. Fluorescence readings were taken on a Victor3V 1420 Multilabel HTS Counter (Perkin Elmer) using the 485 nm/535 nm filter sets. Background fluorescence from buffer was subtracted and polarization (P) and anisotropy (A) values were defined as P = (III−I⊥)/(III + I⊥) and A = (III−I⊥)/(III + 2I⊥), where III and I⊥ are the fluorescence intensities recorded in the parallel and perpendicular orientations respective to the orientation of the excitation polarizer. Fluorescein-labeled oligonucleotides were purchased with the fluorophore attached at the 5`end by a six-carbon linker (IDT). Duplex DNA probes were prepared by annealing each fluorescein-labeled oligonucleotide to a complementary oligonucleotide. Apparent KD values were obtained from direct binding isotherms by nonlinear least-squares regression of the data as previously described [62].
JCV T-ag containing the F258L mutation, which is located in the C-terminus of the OBD (Fig 1), was unable to support T-ag dependent JCV DNA replication [42]. Furthermore, all of the human polyomavirus T-ags, as well as SV40 T-ag, have a phenylalanine at the analogous position (Fig 2). Therefore, to address the function(s) of this highly conserved phenylalanine in polyomavirus DNA replication, we initiated studies of the JCV T-ag F258L mutant.
We previously characterized, via immunofluorescence, the sub-cellular localization of JCV T-ag in C33A cells; a cell line that supports robust levels of JCV DNA replication [46]. It was concluded that JCV T-ag is predominantly in the nuclei of these cells where it is distributed in a punctate manner. Given that T-ag localization to the nucleus is essential for the replication of polyomaviruses (e.g., [63, 64]), we elected to determine if T-ag containing the F258L mutation is also preferentially localized to the nuclei in C33A cells.
Inspection of Fig 3 establishes that as with wt T-ag, full-length JCV T-ag containing the F258L mutation is largely in the nucleus (83% vs ~88%; respectively). Similar to wt T-ag, a much lower percentage of cells contain the mutant in both the nuclei and the cytoplasm (~16.2 vs 12%; respectively) or exclusively in the cytoplasm (~ 0.833 vs 0.3%; respectively). Thus, the failure of the JCV T-ag F258L mutant to support viral replication is not due to defects in its subcellular localization.
The next step taken to establish the defect caused by the F258L mutation was the determination of the structure of the JCV T-ag OBD F258L mutant. The F258L mutant was purified using previously described methods (materials and methods section). As shown in Table 1, the F258L JCV T-ag OBD crystallized in the I41 space group, diffracted to 2.7 Å and contained one molecule in the asymmetric unit cell.
The structure of the JCV T-ag F258L OBD mutant is presented in Fig 4A, only residues 133–259 are visible in the crystal structure. As with the wt JCV OBD [42], the topology of the OBD F258L mutant is a five-stranded antiparallel B-sheet sandwiched, on either side, by two helices. The leucine at position 258 is in orange and shown in a ball and stick representation. The multifunctional A1 and B2 loops are shown in red and blue, respectively. The A1 loop is the "DNA free" or "apo-conformation" ([42]; reviewed in [26]). Inspection of Fig 4B establishes that the previously described C-terminal pocket [42] is also a feature of the JCV T-ag OBD F258L mutant. A superposition of the F258L JCV T-ag OBD mutant with the wt JCV OBD ([42]; 4LIF) is shown in Fig 4C; it is apparent that these two structures are nearly identical (RMSD of 0.43 Å over 127 C-alphas). The spatial overlap between residues F258 with L258 is indicated (Fig 4C; (green and orange residues, respectively)). Finally, the structure also revealed that, owing to the relatively small size of leucine, the F258L substitution created a small cavity that was filled by Leu 258 moving closer to the core of the OBD (Fig 4C insert).
The initial function of the A1 and B2 loops is site-specific DNA binding to the viral origin [44, 65, 66]. Spatially, the A1 and B2 loops are relatively close (~ 15 Å) to residue 258 (Fig 4A). Thus, while the two structures are nearly identical (Fig 4C), subtle but significant structural differences may alter the DNA binding specificity of the F258L mutant and thus account for the replication defect associated with this JCV T-antigen mutant. Therefore, we elected to determine if the F258L T-ag OBD is altered in terms of its ability to bind DNA in a site-specific manner.
The DNA binding studies support the conclusion that the F258L mutation does not significantly alter the structure of the DNA binding A1 and B2 loops. Therefore, we next focused on another prominent feature of the JCV T-ag OBD, the subunit-subunit interface it forms. In the previously reported wt JCV T-ag OBD crystal structures [42], we observed a relatively small interface (~550 Å2) between neighboring OBD molecules. The interface was formed by the A1 & B2 regions of one OBD interacting with the C-terminal pocket of a neighboring OBD; thus, the OBDs were arranged in a head-to-tail manner [42]. Furthermore, the wt JCV T-ag OBD interface was analogous to the OBD:OBD interface observed in previous SV40 crystal structures wherein the OBDs formed a crystallographic spiral having 6 molecules/turn [33, 34]. In light of the SV40 studies, we hypothesized that the interface observed in the structures of the wt JCV T-ag OBDs is similar to that formed during hexamerization of full-length T-antigen [42].
The JCV T-ag OBD containing the F258L mutation crystallized in the same space group as one of the wild type JCV OBDs, suggesting a very similar, but not identical interface as the wild type. Therefore, to refine our understanding of the defect in the F258L mutant, the differences between the interfaces formed by the wt and F258L T-ag OBDs were analyzed in detail.
The studies presented in Figs 7 and 8 provide additional evidence that in the apo form of the JCV T-ag OBDs, the residue 258 containing C-terminal region plays a critical role in forming the interface with the A1/B2 loops. The A1/B2 regions are, however, initially involved in site-specific binding to DNA in the viral origin [43, 44, 65]. Moreover, DNA is known to alter the conformation of the A1 loop in SV40 T-ag (e.g., [27, 67]). These observations raised the question of whether DNA binding is also altering the conformation of the C-terminal region of polyomavirus T-ag OBDs.
A superposition of residue 258 in the apo forms of the JCV OBDs solved to date, including residue 258L (in orange), is presented in Fig 9A (left). Extending these analyses, a superposition of both JCV residue 258, and SV40 residue 257 (equivalent to JCV OBD residue F258) in all of the apo forms of the JCV and SV40 OBDs is presented in Fig 9A (right). It is apparent that in the absence of DNA, the phenylalanines, and the single leucine present in the 258L mutant, adopt a highly conserved conformation. Regarding the question of whether these structures change upon DNA binding, co-structures of the JCV T-ag OBD bound to DNA have yet to be determined. Therefore, to ascertain whether structural changes occur within the C-terminal region of the OBDs upon DNA binding, the analyses were conducted with the previously determined co-structures of the SV40 T-ag OBDs. A superposition of the co-structures of the SV40 T-ag OBDs bound to DNA ([27, 28, 35]) is presented in Fig 9B (left). Of interest, in several of the DNA bound co-structures, the position of residue F257 is altered. Moreover, the co-structure of a larger fragment of T-ag (i.e., the OBD & helicase containing T-ag131-627 fragment) bound to DNA has also been determined [31]. Fig 9B (right) presents just the F257 containing regions in this larger co-structure. It is apparent that in this DNA bound co-structure, residue F257 is also in a "non-apo" position. To more clearly illustrate the DNA dependent shifts in the C-terminus of the OBDs, all of the structures from JCV and SV40 T-ags were superimposed (Fig 9C; same coloring schemes as in Fig 9A and 9B). It is apparent from this figure that DNA binding causes the region around SV40 F257/ JCV F258 to become highly dynamic. Indeed, the position of F257 varies by as much as 13 Å (see insert); nearly the same distance moved by the beta-hairpin in the helicase domain as a function of ATP hydrolysis [40]. Therefore, it is concluded that DNA binding to the A1/B2 motifs alters the conformation of the C-terminal region of the SV40 T-ag OBD.
Finally, an interesting feature of the SV40 T-ag OBD co-structures solved to date is that the F257 residues in the distal orientations are often positioned opposite aromatic residues; an indication that stacking interactions may help to stabilize the individual conformations (Fig 10). Given that it lacks an aromatic ring, the F258L mutant would fail to make these stacking interactions. Therefore, in addition to perturbing the wt interface (Fig 8), the failure of the F258L mutant to form the observed stacking interactions could be an additional reason why it is defective for viral replication.
The F258L mutation in JCV T-ag inactivated JCV DNA replication [42]. Since a conserved feature of all human polyomavirus T-ag's is a Phe at this position, we elected to further characterize this mutation. The immunofluorescence studies established that the F258L mutation does not change the sub-cellular localization of full-length JCV T-ag. Furthermore, equal levels of wt and F258L T-ag were detected 72 hrs post-transfection [42], a finding that established that the F258L mutation does not change the stability of JCV T-ag. Therefore, to gain further insights into the function(s) of residue F258, and the C-terminal region of the T-ag OBD, we determined the crystal structure of the JCV T-ag OBD F258L mutant.
The crystallography experiments establish that the overall structures of the apo forms of the wt and F258L JCV T-ag OBDs are quite similar and that the structural consequences of the F258L mutation appear to be localized to the pocket. That the changes are localized to the pocket, and not transmitted to other regions of the JCV T-ag OBD, is supported by the observation that the F258L mutation has little to no impact on A1 and B2 loop dependent site-specific binding to duplex DNA. A related issue is whether the structural alterations resulting from the F258L mutation can be transmitted to neighboring domains of T-ag. It is noted that flexible linkers separate the polyomavirus OBDs from the neighboring J and helicase domains (e.g., [31, 32]). Therefore, it is unlikely that the structural ramifications of the F258L mutation are transmitted to other T-ag domains.
Regarding the precise changes that occur in the OBD:OBD interface as a result of the F258L mutation; pocket residue F258 normally makes contacts with A1 residues F152 and A150 (Fig 8). While L258 makes contacts to both of these residues, the interactions are different. Moreover, L258 causes disruptions to many of the non-bonded interactions and the L258 mutation also causes the loss (e.g., N259-R155) and acquisition (e.g., D257-F152) of several hydrogen bonds. Collectively, these findings establish that in the apo structure, the F258L mutation results in many subtle, but important changes to the OBD:OBD interface. It is, however, not clear if inactivation of JCV DNA replication by the F258L mutation is the direct result of changes to the interface or if other disruptions play equally important roles (e.g., the failure to stack with neighboring aromatic residues (Fig 10)).
In the apo structures of the JCV [42] and SV40 T-ag OBDs (reviewed in [26]), the positions and orientation of the F258 (F257 in SV40) residues are nearly identical (Fig 9A; (right)). In contrast, in the presence of DNA, the F257 containing C-terminal region in the SV40 T-ag OBD adopts additional orientations (Fig 9B). Thus, it is apparent that DNA binding to the A1/B2 regions alters the conformation of the C-terminal region of the SV40 T-ag OBDs. Support for this hypothesis comes from our previous NMR studies of the SV40 T-ag OBD [67]. When single-stranded poly(dT)25 was added to a sample of the SV40 T-ag OBD, the three regions that had large chemical shift differences were the DNA binding A1 and B2 regions, and the C-terminus. Given the extensive homology between SV40 and JCV T-ag OBDs, it is likely that DNA binding will also modulate the conformation of the C-terminus of the JCV T-ag OBD.
There are several possible functional consequences of the DNA dependent conformational changes in the OBD C-termini; one being that they play a role during T-ag assembly on the origin. Support for this hypothesis includes the finding that SV40 T-ag residue F183 is needed for oligomerization ([62, 65]; in JCV T-ag, the analogous residue (i.e., F184) sits at the base of the C-terminal OBD pocket). DNA dependent conformational changes in the OBD might also play a role in the poorly understood melting of the central or “site II” region of the core origin [68].
In addition, DNA dependent conformational changes in the OBD C-termini may promote DNA replication at later stages, such as when T-ag serves as the helicase at replication forks. We previously proposed that when T-ag is functioning at replication forks, that the SV40 T-ag OBDs are proximal to the ds/ss fork (Fig 11A; reviewed in [26]). Studies of the closely related papillomavirus E1 hexameric helicase also place the N-terminal DNA binding domain (DBD) at the replication fork [69–71]. Among the advantages of having the T-ag OBDs arranged at the replication fork is that as with the nonplanar DnaB spiral assembly [72], the OBD subunits could engage the ds/ss fork via a "hand-over-hand" mechanism; a process that could promote DNA unwinding. Consistent with this proposal, we previously reported that the SV40 T-ag OBDs may assemble into nonplanar hexameric spirals ([33–35]; reviewed in [26]). An additional advantage of hexameric spirals is that in one terminal monomer of the spiral the A1/B2 region is free and thus available for interactions with DNA. In contrast, were the OBDs to adopt a planar flat ring assembly, all of the A1/B2 regions would be involved in interface formation and thus unavailable to bind to DNA.
When the studies summarized above are considered in terms of our current findings they suggest a mechanism for promoting cycling of the JCV T-ag OBD monomers at the replication fork. According to this model, when the A1/ B2 regions in OBD subunit A, the terminal monomer within the spiral that is colored purple, interact with DNA at the fork (Fig 11B; left side), the DNA dependent structural changes in the C-terminus of the OBD will be induced (Fig 9C). This in turn promotes the disruption of the interface formed by the proximal pair of monomers within the spiral (subunits A & B). Consistent with this proposal, when bound to DNA the SV40 T-ag OBDs do not form "apo" like interfaces (reviewed in [26]). Once the interface is disrupted and subunit A is released from the spiral, the A1/B2 region is exposed on what had been the penultimate monomer in the spiral (subunit B (colored in blue)). Following engagement of the ds/ss forked DNA by the newly exposed A1/B2 regions on subunit B (Fig 11B; right side), the DNA dependent cycle of interface disruption will be repeated. Thus, according to the DNA dependent "interface disruption" model, the splitting apart of the OBD/OBD interface is an active process that does not depend upon the relatively inefficient thermal breakage of the interface.
Validation of the DNA dependent "interface disruption" model will require additional structural studies. For example, co-structures of full-length JCV T-ag or T-ag domains bound to DNA are needed to confirm the predicted DNA dependent movements within the C-termini of the JCV T-ag OBDs. Related studies are required to determine if DNA dependent movements occur in other polyomavirus OBDs. Also warranted are structural studies of the OBD:OBD interface formed in the context of T-ag hexamers. It is noted, however, that analogous DNA dependent changes in the DBD of papillomavirus E1 have not been reported [73, 74]. Why these DNA dependent changes are detected in the SV40 T-ag OBD, but not in the papillomavirus E1 DBD, remains to be determined. Nevertheless, recent experiments with E1 have established that the fork proximal origin-recognition domains play critical roles in regulating helicase activity [69] and that strand separation takes place inside E1 in a chamber N-terminal to the helicase domain [71].
In addition, the "interface disruption" model makes certain predictions that need to be tested. For example, the model suggests that upon encountering a gap in duplex DNA, the DNA dependent conformational changes in the OBD will not be induced. This would promote maintenance of the interface and possibly pausing of the T-ag helicase at the gap. Termination at gaps and nicks has been previously reported for prokaryotic (e.g.,[75, 76]) and eukaryotic (e.g., [77, 78]) hexameric helicases. However, it is not yet known if gaps and other forms of DNA damage cause the polyomavirus or papillomavirus hexameric helicases to pause. When completed, these experiments will address the generality of the DNA dependent structural changes within the C-termini of the polyomavirus OBDs and establish the functional consequences of these movements. Finally, MCM complexes are also known to form left-handed lock washer structures [79]. Thus, it will be of interest to determine if fork DNA promotes the cycling of MCM subunits.
|
10.1371/journal.ppat.1006525 | Abundance and co-occurrence of extracellular capsules increase environmental breadth: Implications for the emergence of pathogens | Extracellular capsules constitute the outermost layer of many bacteria, are major virulence factors, and affect antimicrobial therapies. They have been used as epidemiological markers and recently became vaccination targets. Despite the efforts to biochemically serotype capsules in a few model pathogens, little is known of their taxonomic and environmental distribution. We developed, validated, and made available a computational tool, CapsuleFinder, to identify capsules in genomes. The analysis of over 2500 prokaryotic genomes, accessible in a database, revealed that ca. 50% of them—including Archaea—encode a capsule. The Wzx/Wzy-dependent capsular group was by far the most abundant. Surprisingly, a fifth of the genomes encode more than one capsule system—often from different groups—and their non-random co-occurrence suggests the existence of negative and positive epistatic interactions. To understand the role of multiple capsules, we queried more than 6700 metagenomes for the presence of species encoding capsules and showed that their distribution varied between environmental categories and, within the human microbiome, between body locations. Species encoding capsules, and especially those encoding multiple capsules, had larger environmental breadths than the other species. Accordingly, capsules were more frequent in environmental bacteria than in pathogens and, within the latter, they were more frequent among facultative pathogens. Nevertheless, capsules were frequent in clinical samples, and were usually associated with fast-growing bacteria with high infectious doses. Our results suggest that capsules increase the environmental range of bacteria and make them more resilient to environmental perturbations. Capsules might allow opportunistic pathogens to profit from empty ecological niches or environmental perturbations, such as those resulting from antibiotic therapy, to colonize the host. Capsule-associated virulence might thus be a by-product of environmental adaptation. Understanding the role of capsules in natural environments might enlighten their function in pathogenesis.
| Extracellular capsules protect bacterial cells from external aggressions such as antibiotics or desiccation, but can also be targeted by vaccines. Since little was known about their frequency across Prokaryotes, we created and made freely available a computational tool, CapsuleFinder, to identify them from genomic data. Surprisingly, its use showed that many bacterial strains, especially those with the largest genomes, encode several capsules. The frequencies of the different combinations of capsule groups depended strongly on the phyla and the groups themselves, suggesting the existence of epistatic interactions between capsules. Bacteria encoding capsule systems were found in many natural environments, and were frequent in the human microbiome. In contrast to their frequent association with virulence, we found many more capsules in non-pathogens or facultative pathogens than among obligatory pathogens. We suggest that capsules increase the environmental breadth of bacteria thereby facilitating host colonization by opportunistic pathogens.
| Extracellular capsules, hereafter named capsules, constitute the outermost layer of some prokaryotic cells where they establish the first contact between the microorganism and its environment. They fullfill a myriad of roles, often linked to colonization and persistence. Their physical properties prevent dessication by retaining moisture near the cell surface, enhance survival in harsh environments, and protect cells from phagocytosis by grazing protozoa [1–4]. Capsules also play an essential role during infection; they downregulate pro-inflammatory cytokines [5], protect cells against reactive oxygen species generated by the host [6], and help bacteria to evade phagocytosis by macrophages and complement activation [3]. Capsules also reduce the efficiency of antibiotics [7] and cationic antimicrobial peptides [8]. These medical implications have driven the research on capsules and their roles, leading to the widespread perception that they are mostly associated with virulence [9, 10]. This triggered the numerous studies on the genetic diversity of capsules in several prominent bacterial pathogens such as Streptococcus pneumoniae [11, 12], Escherichia coli [13], Klebsiella pneumoniae [14, 15], Campylobacter jejuni [16], and Acinetobacter baumanii [17].
Capsules can be synthesized through different genetic pathways (Fig 1 and reviewed in [18–20]). Most capsules are high molecular weight polysaccharides made up of repeat units of oligosaccharides. In capsules synthesized through the Wzx/Wzy-dependent pathway or Group I [20], the oligosaccharidic repeat unit is linked to an undecaprenyl phosphate acceptor in the cytoplasm by membrane-bound glycosyltransferases. This precursor is then transported across the inner membrane by the Wzx flippase and polymerized nonprocessively in the periplasm by the Wzy polymerase. In contrast, the nascent polysaccharidic chains of Group II and Group III capsules are polymerized in the cytoplasm and linked to a phospholipid acceptor before being transported across the inner membrane by the ATP-binding cassette (ABC) transporter. Group II and III capsules will be jointly referred to as ABC-dependent capsules. In spite of these differences, both the Wzx/Wzy- and the ABC- dependent pathways use homologous outer membrane proteins from the polysaccharide export family to transport the capsule across the outer membrane of diderm bacteria [21]. Both pathways are characterized by large operons that have a conserved region encoding the secretion machinery and a variable region encoding numerous polymer-specific enzymes. The latter defines the capsule serotype and includes enzymes for the synthesis of NDP-sugars, glycosidic linkages (mainly by glycosyl-transferases), and sugar modification (O-acetylation). Within-species serotype-diversity prompted the biochemical characterization of the oligosaccharide composition of capsules, ultimately leading to the development of serotype-specific vaccines [16, 22, 23], and serotyping schemes for epidemic strains [24, 25]. The synthesis of the polysaccharidic Group IV capsules relies on the Wzy polymerase but not on Wzx flippase, and depends on very diverse export machineries, including in certain cases proteins homologous to those of Group I [26, 27]. Polysaccharidic capsules can also be produced by the synthase-dependent pathway, where a unique processive enzyme is responsible for the all the steps of initiation, polymerization and translocation of the capsule [28]. Some capsules are proteic, instead of polysaccharidic, notably the poly-γ-d-glutamate or PGA capsules produced by Bacillus anthracis [29].
To date very few studies have characterized the frequency and diversity of capsules across bacterial phyla, presumably because they are difficult to identify. Capsular systems have many poorly characterized components and are subject to frequent variation by homologous recombination and horizontal transfer, resulting in rapid genetic turnover [30]. Furthermore, the genetic pathways leading to the synthesis of lipopolysaccharides (LPS), extracellular polysaccharides (EPS), and capsules have many key homologous components that are difficult to disentangle [31, 32]. Finally, there are few studies on the role of capsules in ecological settings other than the host, limiting the identification of new capsule secretion pathways.
The understanding of capsule distribution and evolution across Prokaryotes has been hampered by the lack of computational tools to identify capsule systems in genomes. In order to tackle this limitation, we have built protein profiles to identify the key components of the different capsule biosynthesis pathways and defined models describing their expected frequency and genetic organization. We used them within MacSyFinder, a computational tool that allows the detection of macromolecular systems [33], to identify capsule systems in more than 2500 complete prokaryotic genomes. We then searched for the presence of species with capsules in more than 6700 metagenomes. We aimed at answering the following questions: How many capsules are there in prokaryotic genomes? Do multiple capsule groups co-occur and, if so, are there any correlations between capsule groups? Which Prokaryotes encode capsules? Which are the genetic and life-history traits associated to capsule prevalence? What is the environmental distribution of Prokaryotes encoding capsules? Our results uncovered novel intriguing patterns in the distribution of capsules, which have important biological implications and provide new insights into the evolutionary and ecological role of capsules.
We defined independent and customizable models describing the genetic composition and organization of eight groups and subgroups of capsules (Fig 1), based on the literature of the best-described experimental capsule systems [18–21, 26, 27, 29, 34]. This information was complemented with exploratory analyses of the diversity of these systems in other genomes (see Methods). We identified 58 key components (protein families) involved in capsule synthesis. The majority of them regard the secretory and polymerization components of each capsule system, as well as the most common polymer-specific enzymes. Each component was associated with a hidden Markov model (HMM) protein profile, retrieved from PFAM (31) or built for the purpose of this study (27) (S1 Table). The resulting computational tool—CapsuleFinder—uses as input the protein sequences of a genome, searches for the components of capsule systems using the HMM profiles and then delimits the systems based on the information provided in the models.
There is no curated database with information on the organisms encoding and/or lacking capsule systems. The literature rarely mentions the absence (or presence) of a capsule for non-pathogens. Nevertheless, we sought to validate CapsuleFinder by comparing its results with those mentioned in two lists of some of the best-studied encapsulated Prokaryotes [19, 35]. We successfully identified capsules in all 11 species that were reported as encapsulated and for which a complete genome sequence was available. To validate a broader set of systems, we randomly picked 100 species from our complete genome database. We then checked the literature for information on the presence of a capsule in the 40 species where a capsule system was detected (S2 Table). There were 28 species for which we could find published reference to the presence or absence of a capsule. Among these, we found published experimental evidence for a capsule in 15 species and some positive information (from either bioinformatic analyses or evidence in closely-related species) for 10 others. The literature explicitly mentioned that no capsule had been observed for the remaining three species (details S2 Table). It is difficult to say if these are false positives, which would give a false positive rate of ~8%, or if capsules actually exist in the species and the respective strains or conditions of expression were not yet identified. We have not attempted to quantify the rate of false negatives—cases where we missed an existing capsule—since there have been very few experimental efforts to show that a species lacks a capsule in a variety of environmental conditions. Yet, the analysis of our data showed a small number of cases where we missed some capsule systems and obtained some false positives. These are indicated in S3 Table. Even in the worst case, CapsuleFinder is able to identify all the best-known capsules whilst fetching few putative false positives (S1 Text).
We detected 2182 capsule systems in 1304 out of the 2643 genomes (Fig 2). The complete list of genomes and capsule systems is available in S1 Dataset. Around half (49%) of the genomes, representing 52% of the species, encoded at least one capsule system. Group I capsules were the most frequent, representing ca. 70% of the total. ABC-dependent and synthase-dependent capsules were less frequent (nearly 10% each), and subgroup CPS3 capsules were the most frequent among the latter. Group IV capsules (8.8%), and PGA proteic capsules were rarer (3.1%) (Fig 2 and S1 Dataset).
We investigated the presence of capsule systems in all major taxonomic divisions of Bacteria and Archaea (Fig 2). The highly abundant Group I capsule was detected in all bacterial phyla represented by more than 20 available complete genomes (except Spirochaetes and Tenericutes). PGA capsules, even if rare overall, were also present in most phyla. They were particularly abundant in Synergistetes, Planctomycetes, Bacillales and Fusobacteria (Fig 2). On the other hand, Group IV capsules were almost exclusively identified in γ-Proteobacteria and some subgroups were only identified in the taxa in which they were first described, e.g., all Group IV_f capsules were identified in Francisella spp. (Fig 2). We identified at least five out of the eight capsule groups in α- and γ-Proteobacteria and in Actinobacteria. Following previous observations of capsule-like structures in Archaea [38–40], and even if no experimental evidence has yet been given for their existence, we detected 47 capsule systems in 40 archaeal genomes. They were all synthase-dependent (both subgroups) or PGA capsules. Taken together, our results show that capsules are prevalent in Prokaryotes, where their frequency depends on the capsule group and on the phyla.
The genetic loci encoding the experimentally studied capsule systems have remarkably different sizes. Since the number of genes in the capsule system is expected to have some impact on the complexity and evolution of capsules, we computed the number of genes of each system identified in our work (see Materials and methods). These values are only approximate, because capsule systems surrounded by genes encoding enzymes involved in sugar metabolism cannot be delimited without ambiguity in the absence of experimental work. The Group I and ABC-dependent capsules were encoded by significantly more genes than the other capsule groups (S1 Fig). Whereas the median Group I and ABC-dependent systems had between 19 and 16 genes, the synthase-dependent HAS (hyaluronic acid) capsule was encoded in three genes and the Syn_CPS3 in four (S4 Table). These differences may be affected by the abovementioned inaccuracies in capsule loci delimitation and by the definition of the models. Nevertheless, our results show that some groups of capsules have loci of almost invariable sizes (all Group IV capsules), whereas others showed very significant variation in the number of components (especially Group I and ABC, see lower slopes in S1 Fig). These results give statistical support to the idea that the number of capsule components differs markedly between groups.
We then searched to test if genome size was correlated with the number of genes encoding a capsule system. Genomes encoding capsule systems were generally larger than those lacking them (Wilcoxon rank sum test, P < 0.001), but the number of genes in the capsule loci showed no correlation with genome size when controlled for phylogenetic dependence (S4 Table). This suggests that constraints on genome size have no significant effect on the complexity (number of genes) of each capsule system.
We found that almost half of the genomes encoding capsules have more than one system (40%, Fig 3A). Strikingly, two environmental cyanobacteria encoded up to eight capsules, and 23 other species encoded between five and seven systems (S5 Table for details). Among these 25 species, all with large genomes (>4.5 Mb), we identified very few human-associated bacteria: a commensal Bacteroidetes, and some opportunistic pathogens of the Burkholderia cepacia complex. Instead, most of the 25 genomes were from mutualistic or environmental bacteria, including several α- and β-Proteobacterial rhizobia. The size of the genome was correlated with the number of capsules it encodes (Spearman’s rho = 0.16, P < 0.0001 after phylogenetic correction) (Fig 3B), and with the sum of all capsule components (Fig 3B, and S4 Table for phylogenetic corrections). Hence, while the number of genes in a capsule system is not associated with genome size, larger genomes tend to encode more capsule systems, and thus have more capsule-associated genes.
Nearly half of the genomes with multiple capsule systems encode several occurrences of the same capsule group (246 out of 537). We analyzed their sequence similarity to test if they could have arisen by recent large segmental duplications. The systems were typically very divergent: 97% of the intra-genomic comparisons showed less than 80% sequence similarity at the homologous proteins used to identify the group (see Methods). Systems of the same group were not found in tandem, as expected if they had resulted from recent duplications [41] and only eight (out of 1004) pairs of consecutive systems were less than 10 kb apart (S2 Fig). Furthermore, some genomes encoded two (238), three (50), and up to four (in E. coli strain REL606) different capsule groups (Fig 3C). Hence, multiple capsule systems do not seem to originate from recent segmental duplications.
Remarkably, more than half of the genomes encoding an ABC-dependent capsule also encode a Group I capsule (S6 Table), and all genomes coding for Group IV_s and Group IV_e capsules also code for at least one other capsule group. A non-random assortment of capsule groups would suggest epistatic interactions between capsules. To test this possibility, we analyzed the co-occurrence of capsule groups in the light of the underlying phylogenies (see Materials and methods). We used Pagel’s method [42, 43] to fit models of dependent evolution between capsule groups and compared them with models assuming independent evolution (see Methods). We observed significant co-occurrence of Group I capsules and most of the other capsule groups (Fig 3D and S7 Table), including ABC-dependent capsules and Group IV_s. We also observed frequent co-occurrence between PGA and Group IV_f capsules (Fig 3D and S7 Table). In contrast, several groups of capsules showed unexpectedly low co-occurrence patterns suggesting the existence of negative epistatic interactions. For example, we only identified two co-occurrences of Group I and Syn_HAS.
The family of Enterobacteria showed the most frequent co-occurrence of capsules from different groups and subgroups (Fig 4, see S8 Table for the complete list of genomes). Since it also includes several of the model organisms used to study the capsule—E. coli, S. enterica, K. pneumoniae–we analyzed these genomes more in detail. We detected seven out of the 24 different combinatorial possibilities offered by the four different capsule groups identified in the clade. In the line of the results mentioned in the previous paragraph, we observed a clear pattern of correlation between Group IV_s and Group I capsules in enterobacterial genomes (Fig 3).
We observed that closely related genomes often encode different capsule systems. For instance, within the phylogenetic group B1 of E. coli, the two enteroaggregative pathotypes (E. coli 55989 and E. coli O104) encode the same capsule groups, which differ from all the others of the same phylogroup. Similarly, the two only commensal strains (ED1a and SE15) of the phylogroup B2 share the same capsular combination, which is different from all other B2 genotypes (S8 Table). Finally, E. coli from phylogroup A, comprising a majority of commensal bacteria, often have at least three different capsule groups, which is significantly more than other clades including many pathogens, such as Shigella sp. and E. coli B2 (two capsule systems per genome, on average). These results revealed an association between capsule groups and bacteria-host interactions. To conclude, the rapid genetic turnover of capsule systems within closely-related genomes [44] suggests that they can rapidly change to face environmental or lifestyle changes.
The observation that multiple capsules are more frequently observed in commensals, mutualists or environmental bacteria seems at odds with the hypothesis of a tight association between capsules and pathogenesis. We classified bacterial species according to the degree of host-association they commonly exhibit (S1 Dataset, see Methods for criteria and [45, 46]) and found that the probability of encoding a capsule depends on the lifestyle of the bacteria (Fig 5A), even when accounting for genome size (S9 Table). We then first tested whether free living species were more likely to code capsules than pathogens. We found that, indeed, capsules were slightly rarer in pathogenic species as opposed to free living species (Fig 5A and S4 Fig). The lower frequency of capsules in pathogens remains qualitatively similar when commensals and mutualists (or both) are grouped together with free living species. Additionally, we observed no difference in genome size between pathogenic bacteria encoding a capsule system and the others, suggesting that the association between the presence of capsule and pathogenesis is independent of genome size.
Many of the pathogenic bacteria in our dataset are facultative or opportunistic. These bacteria typically have environmental reservoirs and larger genomes than obligatory symbionts (pathogens or mutualists) [47, 48]. We observed that many facultative pathogens encode capsules, in contrast to most obligate pathogens, independently of the differences in genome size between the groups (Fig 5B and S1 Dataset). The difference between obligatory and facultative pathogens remained statistically significant when controlling for phylogenetic structure (see Methods, Fig 5 and S5 Fig). Whereas very few obligate pathogens encoded a capsule, amongst which Shigella flexnerii and Mycoplasma mycoides, a small majority of the facultative pathogens encoded a capsule (Fig 5B). This result does not change qualitatively when only human pathogens are taken into account.
Facultative pathogens tend to start infections only at high infectious dose (ID50), to be motile, and to grow fast under optimal growth conditions [49]. These characteristics also tend to be associated with a lack of ability to kill professional phagocytes of the immune system or to survive in the intracellular milieu of these cells [49]. Since capsules may provide some resistance to phagocytosis, we enquired on the possible association between the capsule, minimum doubling time, and ID50 (measured in humans as available for only 39 species, [49]). We observed that bacterial species that encode a capsule system (Csp+), show significantly lower minimum doubling times (Fig 5C and S4 Fig), higher ID50, are more likely to be motile, and are less likely to be able survive phagocytosis than those that do not encode a capsule (Csp-) (Fig 5, S3 and S4 Figs). Whereas the first association was significant even when controlling for genome size and pathogenicity (S9 Table), and phylogenetic dependence (S5 Fig), the two latter associations were not statistically significant due to lack of statistical power (there is little data available for these traits). Overall, our results indicate that capsules are more readily associated with facultative pathogens with high infection doses and short minimal generation times.
We analyzed microbiome data to confirm that capsule systems are frequent in environmental bacteria and in facultative pathogens (that often have environmental reservoirs). Unfortunately, loci encoding capsule systems are too long and complex to be identifiable in the sequences of metagenomes. To circumvent this difficulty, we identified the presence of the species for which we had at least one complete genome in a large number of publicly available metagenomics datasets (16S rRNA). We used this information to quantify the abundance of each species and, using the species' complete genomes as a proxy, to predict the presence of capsules in these environments. Specifically, we searched for the presence of Csp+ in 16S datasets from four classes and numerous sub-classes of environments (Fig 6A). This allowed both the qualitative and quantitative identification of bacterial species in 6700 environmental 16S datasets (S8 Table, see Methods). We computed the abundance of Csp+ relative to Csp- species in the 16S datasets in qualitative (number of species) and quantitative (number of 16S sequences) ways (see Methods). The percentage of Csp+ was similar in the 16S (53% out of 1197 bacterial species) and in the genome (52%) datasets. Csp+ were more frequently present and quantitatively more abundant than Csp- in all four classes of environments, even if this trend was not always significant (Fig 6A and S6A Fig).
Capsules allow Prokaryotes to withstand a series of stresses, from environmental disruptions to protozoa grazing, and are expected to be associated with broader environmental ranges. Indeed, Csp+ species were present in significantly more environmental subclasses than Csp- (Fig 6B). Importantly, the number of environmental subclasses for a given species increased with its average number of capsules per genome (Partial Spearman test, P < 0.001, after correction for genome size). These results show that bacteria encoding capsule systems are able to colonize a larger variety of environments.
The vast majority of previous studies focused on capsules of bacterial pathogens. To disentangle the relation between capsule and pathogenesis, we analyzed the presence of Csp+ species in human-associated datasets. We first checked that we were able to identify well-known pathogens in the host-associated environments. Indeed, we detected pathogens with Group I capsules, such as K. pneumoniae and S. pneumoniae, as well as pathogens with ABC capsule systems, namely Neisseria meningitidis, in samples of the human microbiome, and sometimes also in other environments (S10 Table). The total abundance of species encoding capsules within the human host varied between body locations (Fig 6C), and was higher overall than within the complete genome database (57%, binomial test, P = 0.005). Csp+ species were more abundant than Csp- in all locations, and especially in the gut microbiota, which encompasses the largest fraction of bacteria in the human body. Likewise, clinical samples over-represented Csp+ species. Interestingly, we observed that the relative abundance of Csp+ and Csp- was strongly dependent on the human body sites (ANCOVA, P < 0.001, S6B Fig).
Taken together, our results show that even if capsules are relatively rare among obligatory pathogens, they are very frequent in human microbiota where they are frequently associated with clinical conditions.
Capsules play important roles in bacterial virulence, but their study has been hampered by the lack of computational tools to identify them in genomes. Our tool, CapsuleFinder, identifies the eight major groups and subgroups of capsule systems in bacterial and archaeal genomes and is thus complementary to software designed to analyze very specific capsule systems, e.g., the recently released Blast-based tool to identify capsular serotypes in Klebsiella spp. (Kaptive, [50]). The models in CapsuleFinder can be modified to either increase specificity (obtain systems closer to the experimental models) or sensitivity (to detect more distantly related systems). This can be done by changing the number, type and genetic organization of the components that are required to identify a system. Users can also add novel models and protein profiles to improve the tool, e.g., to account for novel experimental data. If enough experimental serotype data is available for a given species, then the models can be specified in order to infer a putative serotype for the strains.
The construction of our models was based on previous experimental studies restricted to a relatively small number of model organisms from Proteobacteria and Firmicutes. Capsules, like many extracellular structures [51], are subject to rapid evolution and reorganization via recombination, complicating their detect from a small number of taxonomically restricted reference systems. In spite of this, we were able to identify them in many phyla of Prokaryotes—even in Archaea—with few putative false positives. Hence, we expect to have identified the majority of capsules of known groups in the complete genome database. The entire collection of capsule systems can be consulted in our database (http://macsydb.web.pasteur.fr/capsuledb/_design/capsuledb/index.html and S1 Dataset). Further, the identification of capsule systems by CapsuleFinder opens the way for their comparative analysis, including the study of how horizontal transfer leads to serotype switching across bacteria [52, 53].
Our analysis showed that a majority of Prokaryotes encodes at least one capsule system (Fig 2). Group I, PGA and Syn_CPS3 are the most widespread across the Bacteria whereas other groups were restricted to a few taxa, namely Group IV and Syn_HAS. Importantly, we found capsule systems in all phyla for which more than ten genomes were available. Future work will be necessary to assess if poorly sampled phyla—Chrysiogenetes, Deferribacteres, and Elusimicrobia—are effectively devoid of known capsule groups or if they encode novel groups of capsules. It will also be interesting to analyse capsule prevalence in newly discovered uncultivable phyla characterized by single-cell genomics since they may reveal novel capsule groups (or variants of existing ones) [37, 54]. Given our results in the phyla with higher representation in the database, capsules might occur across all prokaryote phyla.
Capsule-like structures have been described in Archaea [38–40], where a previous bioinformatic study revealed the presence of proteins similar to those involved in the synthesis of the PGA capsule in one species [29]. We identified PGA capsule systems and also Syn_CPS3 systems in many genomes of Archaea. These two groups of systems have few components and we couldn't find data suggesting that they allow extensive serotypic variation. However, the lack of more complex capsule groups, should be subject to caution owing to the lack of experimental data. Furthermore, our tools to identify capsules were based on bacterial systems. Alternatively, the peculiarities of the cellular envelope of Archaea may explain the absence of certain capsule groups in the phyla. Most Archaea have a S-layer composed of glycans that might affect secretion or cell surface association of certain capsules.
Our method may underestimate the number of capsule systems of the same group co-occurring in a genome owing to strict localization rules in our models to avoid false positives. For example, not all Group I Bacteroides thetaiotaomicron were detected because some operons lacked the minimum mandatory genes required to identify the gene cluster as a capsule system (S3 Table). This suggests that some structural elements involved in capsule secretion might be shared between different systems. Yet, to date, the existence of genomes encoding multiple capsules of the same group had previously been documented in only a few species, namely Bacteroides spp [55, 56]. In B. fragilis, a key commensal of the gut microbiome, there are several Group I capsule systems, some of which are implicated in the formation of intra-abdominal abscesses [57]. This species encodes a DNA inversion mechanism that combinatorically switches the expression of the different systems [58], producing extremely diverse capsule structures that are thought to increase bacterial fitness in the intestinal milieu by virtue of their immunomodulatory properties [59]. In this case, capsule variation seems to evolve as a response to the rapid change of the human immune system [58].
Bacteria may also encode multiple capsules from different groups, as described for the PGA and Syn_HAS capsules encoded in different plasmids of Bacillus cereus biovar anthracis[60]. Co-expression of different capsule groups is thus possible, implicating that capsules will physically interact in the cell envelope. Our data suggests that capsule combinations can be even more complex, since this same strain encodes a Group I capsule in the chromosome, and some enterobacteria encode up to four different groups of capsules.
The non-random patterns of co-occurrence of different capsule groups observed in this study suggest that capsule repertoires are affected by epistatic interactions (Fig 3D). The nature of these interactions depends on whether the different capsules are expressed at the same time, thereby producing combinatorial diversity, or at different moments, e.g., in response to different environmental cues. Positive epistasis may result from the synergistic combination of the properties of the different capsules, e.g., different capsules may provide a broader range of environmental protections and capsule switching (or variation in the proportions of each capsule group) may facilitate escaping grazing protozoa, professional phagocytes of the immune system, or bacteriophages. Negative epistasis associated with co-expressed capsules may result from problems in accommodating different capsule structures in the cell envelope. Negative epistasis between capsules that are not co-expressed could be caused indirectly by the effects of the genetic background, e.g., because some groups of capsules are more compatible with certain membrane structures (pili, flagella, secretion systems) than others.
The mechanisms leading to the acquisition of multiple capsules will have to be studied in detail in the future, but our results already provide some clues. We observed that many genomes encode capsules of different groups, that capsules of the same group are very divergent in sequence and are encoded in distant regions in the genome (or in different replicons). This suggests that capsules were independently acquired by multiple events of horizontal gene transfer. This fits the abundant literature showing that capsules vary rapidly within species by recombination and horizontal transfer [61–63]. It also explains why most capsule systems are encoded in a single locus, since this facilitates transfer [64]. Finally, the outcome of capsule transfer is likely to depend on the environmental challenges faced by the bacteria and will be affected by the abovementioned epistatic interactions.
A substantial part of the previous literature on capsule systems has focused on bacterial pathogens and on the role of capsules as virulence factors. For instance, it has been shown that acquisition of certain capsule types by horizontal gene transfer in Neisseria meningitidis allowed the bacteria to increase in pathogenicity and going from non-pathogenic carriage to infectious state [52, 53]. It was thus surprising that non-pathogens are more likely to encode capsules, and that, among pathogens, the ones establishing obligatory antagonistic interactions with their hosts typically lacked a capsule.
The abundance of capsules across most phyla and environmental classes, and their rarity among obligatory pathogens, suggest they play important roles beyond pathogenesis. Indeed, the capsule also constitutes an advantage for commensal bacteria of the gut. To colonize the gut, the bacteria have to first withstand the harsh conditions of the stomach and then grow and multiply in the duodenum and colon, in the presence of bile salts. In Bifidobacterium longum, capsule expression would enhance survival in the stomach and allow growth under high concentrations of detergent-like bile salts in the duodenum [65]. Similarly, a study performed in yeast has shown that although capsules from environmental and pathogenic strains display similar composition and features, they fulfil different roles [66].
Capsules are an example of the ability of bacteria to evolve structures serving multiple purposes in different environments. Like other virulence factors, such as some iron capture proteins, while evolving as an adaptation to an environment they also confer an advantage during pathogenesis (exaption), either during colonization or transmission across hosts [47, 67].
Our data also shows that the presence of capsule systems, and especially multiple systems, is associated with broader environmental ranges. The ability to express different capsules, or combinations of them, can result in heterogeneity in the surface charge of bacterial cells which can in term influence important phenotypes such as cellular adhesion to tissues or surfaces, susceptibility to certain cationic peptides, etc. In the aforementioned B. cereus strain, the co-expression of the two capsules did not increase virulence in two different animal models, but rather favoured bacterial colonization and dissemination [60]. Similarly, previous studies in soil-borne nitrogen-fixing bacteria indicated that bacterial exopolysaccharides and lipopolysaccharides that can be similar to capsules are involved in species-specific interactions between the bacteria and the host [68]. This is consistent with our observation that capsule multiplicity increases environmental breadth, and suggests that it may also increase host range.
Taken together, our study revealed an unsuspected prevalence of capsules in Prokaryotes, especially in environmental bacteria and facultative pathogens. Our results are in line with the multitude of roles proposed for capsules and are not consistent with the idea that capsules evolved to facilitate pathogenesis. Instead, they highlight that capsules might have an important role in facilitating bacterial adaptation to novel or changing environments. Interestingly, we found many capsule systems in soil bacteria, from which probably originated capsulated opportunistic multi-resistant bacteria such as Klebsiella pneumoniae, Enterococcus faecium, and Acinetobacter baumanii [69–72]. Capsules may have thus evolved primarily as an adaptation to a range of different environments, and this facilitated subsequent ecological transitions towards host colonization and pathogenesis.
We built a model for each group of capsule with the information we could obtain from the literature. We specified models with mandatory (biologically essential components for a putative functional system, a majority of which, if not all, are required to identify and classify the systems), accessory (non-essential components used to improve the annotation of the system), and forbidden components (e.g., those found in other capsule groups and not in the focal one, thus helpful to discriminate between the capsule groups, see below the example of Group IV capsules). Of note, due to the low conservation of some mandatory elements, for example Wzy polymerases, in some instances a system could be validated even if a certain number of mandatory components were not detected. This is controlled by the option min_mandatory_genes_required. The parameters used for the minimum quorum of mandatory genes were set based on the analysis of experimental systems and on our previous experiences with the development of similar models for protein secretion systems and CRISPR-Cas systems [33, 36]. While these systems are very different, they have in common that certain components that are thought to be biologically necessary may not be identifiable by sequence analysis either because they evolve too fast, or because they can be replaced by analogues lacking sequence homology.
Additionally, we specified that components should be encoded in a single locus (defined as a series of genes respecting a maximal pre-specified distance between consecutive elements). When the available experimental data suggested that it was relevant to allow components to be encoded elsewhere in the genome, we defined them as loners in the models. Models were written in plain text, using a specific XML grammar, and can be modified by the user (see http://macsyfinder.readthedocs.io/en/latest/ for details). For simplicity, we named the components after the protein names in the species that served as a biological model for each group of capsule. The names of the homologs to these proteins in other species with experimentally validated systems are listed in S1 Table. Polymer-specific enzymes were regarded as accessory in the models because they can be homologous to enzymes of other cellular processes [18].
We used MacSyFinder to search for capsule systems [33]. This program takes as input a proteome, a set of hidden Markov models (HMM) protein profiles (one for each component of the system, see below), and models describing the number of components and their genetic organization (Fig 1). MacSyFinder identifies the individual components of each capsule system using hmmsearch from the HMMER package v3.1b2 [86]. A component was retained for further analysis when its alignment covered more than 50% of the length of the profile and obtained an e-value smaller than 0.001.
We used 58 different HMM protein profiles in our searches (S1 Table), 31 retrieved from the PFAM 28.0 database (http://pfam.xfam.org, [87], last accessed November 2015) and 27 built in this study. Each protein profile was constructed as follows (except when explicitly stated otherwise). We started from a well-described and experimentally-validated component of a system and used BLASTP v 2.2.28 [88] (default settings, -v 4000, e-value < 10−4) to search for homologs among complete genomes. To reduce the redundancy of the dataset (i.e., to remove very closely related proteins), we performed an all-against-all BLASTP v 2.2.28 analysis and clustered the proteins with at least 80% sequence similarity using SiLiX v1.2.9 (http://lbbe.univ-lyon1.fr/SiLiX, default settings) [89]. We selected the longest sequence from each family as a representative. The set of representative sequences was then used to produce a multiple alignment with MAFFT v7.215 using the L-INS-i option and 1000 cycles of iterative refinement [90]. The alignment was manually trimmed to remove poorly aligned regions at the extremities, using SEAVIEW [91]. The HMM profile was then built from the trimmed alignment using hmmbuild (defaults parameters) from the HMMER package v3.1b2 [86].
We validated the method to identify capsule systems using two published lists of capsulated bacterial pathogens [19, 35]. Since these lists were very short, and not necessarily meant to be exhaustive, we made a complementary validation on a random set of species from our dataset. We used the R function sample to randomly draw 100 species from a curated list of 1241 species in our database (this list did not include genomes for which a genus but not a species was defined, such as Glacieola sp.). We identified capsule systems in 40 of the 100 species. We then sought to confirm the presence of capsule in the latter (they include 52.5% of free-living, 30% facultative pathogens, 12.5% commensals and 5% of mutualists) by analyzing the primary scientific literature. For those species for which we did not detect a capsule system, we did not seek further validation as negative results are not systematically reported.
We identified the core genome of 131 enterobacterial genomes belonging mostly to E. coli and Salmonella spp., but also Shigella spp., Citrobacter, Cronobacter, Klebsiella, and Enterobacter (see S8 Table for the complete list of genomes). We followed a previously published methodology [92]. Briefly, orthologs were identified as bidirectional best hits, using end-gap free global alignment, between the proteome of E. coli K12 MG1655 and each of the 130 other proteomes. We discarded hits with less than 60% similarity in amino acid sequence or more than 20% difference in protein length. The list of orthologs for every pairwise comparison was then curated to take into account the high conservation of gene neighborhood at this phylogenetic scale [93]. We defined positional orthologs as bidirectional best hits adjacent to at least four other pairs of bidirectional best hits within a neighborhood of 10 genes (five genes upstream and five downstream). The core genome was defined as the intersection of pairwise lists of positional orthologs and consisted of 759 gene families.
To control for phylogenetic independence of data at the genome-level, we aligned the 16S rRNA using secondary structure models with the program SSU_Align v0.1 [94] of 2440 bacterial genomes. The alignment was trimmed with trimAl v1.4 [95] using the option -noallgaps to delete only the gap positions but not the regions that are poorly conserved. The 16S rRNA phylogenetic tree was infered using IQTREE v.1.4.3 [96] under the GTR+I+G4 model with the options–wbtl (to conserve all optimal trees and their branch lengths), and–bb 1000 to run the ultrafast bootstrap option with 1000 replicates. Two hundred and eleven genomes from our database were excluded from the final phylogenetic tree because identical 16S sequences were already present in the multiple alignment. When data was analyzed at the species level, a 16S rRNA gene per species was chosen by the Bash function RANDOM (from all the available genomes of the species) from the secondary structure alignment and a new phylogenetic tree constructed as above.
To build the core-genome phylogenetic tree of the Enterobacteria, we aligned each core gene family at the amino acid level with MAFFT v7.215 (default options) [90], trimmed non-informative positions with BMGE v1.12 (default options and—t AA) [97], and concatenated the alignments. The tree of the concatenate was built using IQTREE v.1.3.10 under the GTR+I+G4 model [96].
In both trees, the model used was the one minimizing the Bayesian Information Criterion (BIC) among all models available (option -m TEST in IQTREE).
All phylogenetic corrections were done using the 16S rRNA tree of Bacteria. We restricted our phylogenetic controls to Bacteria, because the inclusion of Archaea reduced very much the phylogenetic signal (resulting from a shorter multiple alignment) and clumped together many species’ 16S sequences.
The presence of phylogenetic signal in the evolution of traits was estimated with Pagel’s lambda using the phylosig function of the phytools package v.0.5–20 for R [42] and the aforementioned 16S rRNA phylogenetic tree. To estimate the phylogenetic signal across capsule groups, instead of using the 16S rRNA tree, we built new trees comprising only the 16S rRNA sequences of the genomes for which we detected the given capsule groups. To control for the effect of the uncertainty in phylogenetic inference on the key positive results, we produced 1000 bootstrap trees (options -wbtl -bb 1000 in IQTREE) and randomly selected 100 of those trees. We then ran each key analysis (those in the figures, either GEE, fitPagel or phylosig functions) using the different trees. The distribution of the 100 P values of each analysis is presented in S5 Fig.
We tested the significance of the co-occurrence of capsule groups, with the default method (fitMk) of the fitPagel function from the phytools package (v0.5–52 maps v3.1.0). This function assumes an ARD—all rates different, which allows different rates at all transitions- substitution model for both characters and gives the probability that they are independent (the rates of transitions of each character are independent of the other character).
We controlled the associations between traits for phylogenetic dependence whenever one of their lambda’s P values was less than 0.05. We used the pic function to make independent contrast analysis of continuous data and the compar.gee function to analyze associations between discrete and continuous variables using generalized estimation equations (GEE). Both were computed with the functions included in the ape v.3.5 package for R [98]. We also controlled associations for the effect of genome size by fitting linear regression models using aov from R.
We selected from MG-RAST the metagenomes matching at least one species of our complete genome database and obtained from four environmental categories (subclasses indicated in S8 Table): (i) water (fresh, marine and spring water), (ii) soil (agricultural, dessert, forest, tundra and grasslands), (iii) air (indoor, mammal), and (iv) host-associated (human, other mammals, arthropods, aquatic organisms and plant). These categories are broad and heterogeneous (they put together many different environments). They are used to provide a very coarse-grained classification of the type of environment of each species.
We used 16S rRNA assembled reads to identify and quantify the presence of species from the complete genome dataset in the environmental samples. All analyses were performed at the species level rather than at the strain level because 16S rRNA does not allow resolving phylogenetic structure below the species level. For consistency with previous analyses, Archaea were also excluded from the 16S environmental datasets. First, for each metagenome we identified the 16S matching each of the species in our database using BLASTN v 2.2.28 (selected hits with more than 97% sequence identity and with alignments covering at least 90% of the query sequence). The relative abundance of each species was then calculated by dividing the number of 16S rRNA sequences in each metagenome by the total number of sequences. This information was used to draw the frequency of species with capsule systems in each environmental category and subcategory. To validate the analysis, we searched for well-known pathogens and quantified the frequency in which they appeared across metagenomes of each environmental subcategory (S11 Table).
Sequence identities and similarities were calculated with needle function (default settings) included in the EMBOSS 6.6 package. Phylogenetic trees were produced with iTol v3.0 [99]. Statistical analysis and graphs were done with R version 3.2.0 and the packages ggplot2 and RColorBrewer, unless stated otherwise. PMCMR [100], stats and NCstats [101] packages for R were used for post hoc pairwise multiple comparisons of mean ranks and data manipulation.
We have made publicly available the methods to detect capsules. CapsuleFinder can be used locally using the program MacSyFinder [33], freely available for download at https://github.com/gem-pasteur/macsyfinder. We recommend the use of our models without the option "all' (as recommended in the documentation of the program). It can also be queried on a dedicated webserver within the Galaxy platform (https://galaxy.pasteur.fr/root?tool_id=toolshed.pasteur.fr/repos/odoppelt/capsulefinder/CapsuleFinder/1.0.2). The protein profiles and capsule models used in this study are accessible at https://research.pasteur.fr/fr/tool/capsulefinder/. The models are written in a simple XML grammar in plain text files to allow user modifications (see documentation in http://macsyfinder.readthedocs.io/en/latest/). The results of MacSyFinder can be visualized with MacSyView, available online at http://macsyview.web.pasteur.fr. The capsules detected in this study, their genomic localization and organization are collected in an accessible database, CapsuleDB, http://macsydb.web.pasteur.fr/capsuledb/_design/capsuledb/index.html.
|
10.1371/journal.pcbi.1001021 | Little Italy: An Agent-Based Approach to the Estimation of Contact Patterns- Fitting Predicted Matrices to Serological Data | Knowledge of social contact patterns still represents the most critical step for understanding the spread of directly transmitted infections. Data on social contact patterns are, however, expensive to obtain. A major issue is then whether the simulation of synthetic societies might be helpful to reliably reconstruct such data. In this paper, we compute a variety of synthetic age-specific contact matrices through simulation of a simple individual-based model (IBM). The model is informed by Italian Time Use data and routine socio-demographic data (e.g., school and workplace attendance, household structure, etc.). The model is named “Little Italy” because each artificial agent is a clone of a real person. In other words, each agent's daily diary is the one observed in a corresponding real individual sampled in the Italian Time Use Survey. We also generated contact matrices from the socio-demographic model underlying the Italian IBM for pandemic prediction. These synthetic matrices are then validated against recently collected Italian serological data for Varicella (VZV) and ParvoVirus (B19). Their performance in fitting sero-profiles are compared with other matrices available for Italy, such as the Polymod matrix. Synthetic matrices show the same qualitative features of the ones estimated from sample surveys: for example, strong assortativeness and the presence of super- and sub-diagonal stripes related to contacts between parents and children. Once validated against serological data, Little Italy matrices fit worse than the Polymod one for VZV, but better than concurrent matrices for B19. This is the first occasion where synthetic contact matrices are systematically compared with real ones, and validated against epidemiological data. The results suggest that simple, carefully designed, synthetic matrices can provide a fruitful complementary approach to questionnaire-based matrices. The paper also supports the idea that, depending on the transmissibility level of the infection, either the number of different contacts, or repeated exposure, may be the key factor for transmission.
| Data on social contact patterns are fundamental to design adequate control policies for directly transmissible infectious diseases, ranging from a flu pandemic to tuberculosis, to recurrent epidemics of childhood diseases. Most countries in the world do not dispose of such data. We propose an approach to generate synthetic contact data by simulating an artificial society that integrates routinely available socio-demographic data, such as data on household composition or on school participation, with Time Use data, which are increasingly available. We then validate the ensuing simulated contact data against real epidemiological data for varicella and parvo-virus. The results suggest that the approach is potentially a very fruitful one, and provide some insights on the biology of transmission of close-contact infectious diseases.
| A century after the first contributions giving birth to mathematical epidemiology, and after 20 years of fast growth since the first public health oriented contributions [1]–[3], infectious diseases modeling has recently received a further dramatic impulse from pandemics threats. The Bio-terrorism and SARS first, the fear of a potentially devastating pandemic of avian flu then, and finally the recent pandemic of A/H1N1 influenza, have all fostered the development of more and more detailed predictive tools. These range from traditional models to network analysis, to highly detailed, large scale, individual-based models (IBM) [4]–[17]. IBM are highly flexible tools for policy makers as they allow to define intervention measures at the finest possible levels (e.g., the contact network of single individuals during a specific activity). For the first time, a pandemic model on a continental scale has been proposed [17].
A critical aspect common to all such models, is the parameterization of social contact patterns, i.e. how people socially mix with each other [18]. Social contact patterns are the key factors underlying the transmission dynamics of directly transmitted close-contacts infectious diseases [18]. Different models, independently of their level of complexity or geographical scale, are sensitive to the parameterization of social contact patterns.
In a relatively simple case, where individuals are stratified by age only, contact patterns are represented in the form of contact matrices whose entries represent the average number of contacts that individuals in age group i have with individuals in age group j, per unit of time. Until recently, contact patterns were estimated “indirectly” by calibrating suitably restricted contact matrices using observed epidemiological data, such as serological or case notifications data. The two major examples of this indirect approach are the Who-Acquires-Infection-From-Whom (WAIFW) matrix [3], and the proportionate/preferred mixing approach [19]. Such approaches have important restrictions: in a population divided in n age groups, a contact matrix contains nxn = n2 unknown entries. Therefore, in order to estimate the n2 parameters from the n data points (e.g., serological data) some simplifying assumptions about the structure of the matrix are needed. In addition, indirect approaches can only estimate adequate contacts or transmission rates, i.e. composite parameters given by the product between a contact rate and the corresponding risk of infection per contact.
Recently, important progress has been made in this area through direct collection of contact data by means of sample surveys [20]–[25]. The direct approach is based on appropriate definitions of an “at risk event” (e.g., a face-to-face conversation). Survey respondents are then asked to record in a diary relevant characteristics (e.g., age) of all the individuals they had contact with during a randomly assigned day, or other factors such as the location where the contact occurred (e.g., home, school, public transportation). Standardized international survey data on social contact patterns in 8 European countries are currently available [24]. In addition, contact matrices, and “time in contact” matrices, have been estimated from secondary data sources such as transportation surveys [26] or time use data [27], which are increasingly available. In the case of time use data, the underlying hypothesis is that the amount of time people spend doing the same activity in the same place is relevant for the transmission of the disease.
A drawback of time use data is that they usually do not give direct information about the number of social contacts of respondents, or the time they spent in contacts. They only give “marginal” information on the time individuals allocated to the various daily activities [27]. Therefore, these data need to be augmented with other data and/or assumptions to produce reliable estimates of contact matrices [27]. A way to supplement time use data relies on socio-demographic sources (e.g., routine or census data) which provide information on the size and distribution of the “arenas” (e.g., school, workplaces, households) where contacts take place. For example, for school contacts we often know the average class size and the average pupils-teacher ratio for all compulsory grades. As for contacts within the household, we have information on household size and composition. For most other activities, however, there is little information. Assumptions, e.g. independency, are therefore necessary to give some coarse ideas of contact patterns [27]. However, this approach ignores the structure of the social networks where contacts are formed. A promising approach is then to reconstruct such networks by the simulation of appropriate artificial social networks. A first example is the social network generated by the “Portland” synthetic population [26]. In that case, “contact” and “time in contact” matrices by age are by-products of the social dynamics of the Portland model. These matrices have the standard expected features: population contacts cluster around children and adult, children interact most frequently with other children close to their own age, etc. However, such matrices were neither compared with other contact matrices, nor validated against empirical epidemiological data. Thus, no actual evaluation of their “goodness” in explaining transmission of infections is available.
In this paper, we follow the same line and aim to reconstruct contact and time-in-contacts matrices by simulating a suitable “minimalistic” socio-demographic individual-based model for Italy. The model is parameterized by integrating time use data from the Italian time use survey [28] and other official socio-demographic data [29]–[30]. In the model, each artificial agent is a “clone” of a real individual, i.e. there is a one-to-one correspondence between the diary of each “artificial” agent and the one of a corresponding “real” survey participant. Since the sample is representative of the Italian population, but the size of the model population is comparable to that of a small Italian city, we named the model “Little Italy”. From this point of view, our model resembles the Portland model [6], and the Eemnes (a small Dutch city) model [31]. In the Little Italy world, agents “physically” displace during the day in order to attend their various daily activities in the corresponding location. In these locations, agents “contact” other agents. We defined a contact as “having shared the same physical environment” (i.e. house, the same class at school, the same bus) during a given time slot.
With our approach we generate three different types of contact matrices, possibly informative of distinct aspects of the biology of transmission: (a) a matrix describing the time spent in contact (Type 1) [27], (b) a matrix counting the number of repetition of contact episodes (Type 2), and (c) a matrix counting contacts as the average number of different persons contacted, i.e. the number of different social partnerships, (Type 3) as in [24].
In addition, we extracted an adequate [19] contact matrix from the socio-demographic model underlying the Italian IBM for pandemic prediction and mitigation [15], that we named “Big-Italy”. The synthetic contact matrices computed by simulation of Little and Big-Italy are tested against recently collected Italian serological data on Varicella and ParvoVirus (B19). Their performances are compared with other contact matrices available for Italy, i.e. the “Polymod” and “time use” matrices.
The Italian Time Use (TU) survey was carried out by the Italian National Statistical Agency between 2002 and 2003 [28], with a sample of 55,773 individuals, grouped into 21,075 households. Respondents, with the exception of children less than 3 years old, were asked to fill in a questionnaire with a diary of the activities done during a randomly selected day. To take into account the differences between workdays and week-ends, the sample was divided into three groups. One group was asked to fill the diary on a given workday (18,085 diaries collected), one on a Saturday (16,828) and one on a Sunday (16,293).
A 24-hour day, starting from 4am, is divided into 144 time slots of 10 minutes each, called “ticks”. For each tick, the respondent's diary records the type of location where the person was, and the type of activity done. Due to privacy issues, records always refer to types of places and types of activities, instead of exact places and exact activities. Types of places and types of activity are given unique codes (i.e. 1 for home, 2 for office, etc. for locations; 1 for working, 2 for caring children, etc. for activities). However, these codes are identical for every individual. Therefore, if at the same chronological time two people are both working, each one in his/her own office, we have two records with the same codes, but this does not imply they are in the same office doing the same thing. This has some drawbacks. First, there is never any clue about the purpose of the undertaken activity. For example, if in a certain tick someone reports being on the public transportation network, there is no indication about the reasons for being there. For instance, it could be for going from home to office, or bringing children, if any, to school and then going to work, or anything else. The same applies for places, with a single remarkable exception: if at any time two individuals report that they are at home, and we know from other data that they both belong to the same household, we can infer that they are in the same place. This is the only case in which the partial information given by respondents can be correctly augmented.
Finally, routine socio-demographic data on a) family size and composition [28]; b) firms size by number of employees [29]; and c) school class size for any school grade [30], were used to inform our model.
To create an artificial society that matches the one that is revealed by the Time Use survey, some assumptions were made. Unlike other approaches (e.g., the Portland model), whose aim is to create artificial societies that are as close as possible to a real population, we opted for an artificial society based on a “minimally” complex set of rules, that is nonetheless representative of the Italian population. This seems to be a useful departure point: by considering a simple spatial structure and a minimal set of activities/locations (school and work, the household, and “other”, non-school, non-household contacts), which are those considered fundamental in basic epidemiological explanations, we avoid the need to include several extra-assumptions for model parameterization. Further activities and locations can nonetheless be easily included.
Let us list the assumptions adopted. First, we restricted our model to individuals followed over an average workday. This choice sets Little Italy's population to 18,085 (artificial) individuals. We chose to ignore week-end days because the groups of respondents to the surveys are different and therefore some additional assumptions would be necessary to link workdays and week-end days agendas.
During the day, agents move to and from different places. Most of the time, respondents reported to be at home, in the office or at school. For the rest, they either declared to be in more specific places (e.g., bakery, park, etc.) or that they were moving from one place to another (e.g., on foot, by car or by bus). We chose a square grid as Little Italy's “environment”, with grid's size 150×150, in order to allocate families in single cells representing houses, leaving appropriate space for schools and workplaces. Each square in the grid is identified by a pair of coordinates. We allocated one house for each household on a random cell on the grid. House cells can contain at most 5 families.
In order to host all students aged 3–18, and Little Italy's only university, we allocated schools at random on the grid.
The setting up of workplaces required a few more assumptions, since respondents only reported that they were at work during some ticks but gave no information about either the size of the company they were working for, or the number of colleagues (and in many situations, like, for instance, bus drivers, workers “share the environment” with people that are not necessarily colleagues). Therefore, we drew samples of firms from the workforce size distribution of Italian firms in cities having population size comparable to Little Italy, i.e. 10,001–20,000 inhabitants [29]. This yielded a number of alternative configurations for the number of firms, and for their sizes, which are representative of the real variability observed in small Italian towns having the size of Little Italy. We finally put each firm on a single random cell and assigned each worker to a firm.
Two aspects of the previous process are worth mentioning. First, each agent declares how much time she/he spends going, say, from home to office by car. This time is a proxy for the distance from home to office which must be respected all over Little Italy. It is not possible, for example, that agent A takes 1 time slot (10 minutes) to move by 20 cells on the grid while agent B covers the same distance in 6 time slots (1 hour) if they both declare using the car. This would mean that A's car moves 6 times faster than B's one, which is possible, but unlikely. We proceeded as follows. After workers are assigned to firms, a random re-assignment of houses is performed: two households exchange their houses if, in the new configuration, the actual distances between offices and houses are closer to the ones that can be inferred from their diaries. A large number of exchanges is carried out until the error cases are a negligible fraction of total workers.
Second, there are workers who declared not having a single workplace, like, for examples, a plumber. For them, we decided to set their moving workplace at random. Each time they are about to go somewhere, the simulation chooses a random square on the grid as their next workplace. Commercial places are created on the grid and their workers are assigned to them. Students are assigned to classes in schools, according to routine data [30], which prescribe an average number of students per classroom on a regional basis. These figures are very close to the observed ones [28]. The related numbers (averaged over Italy) are: 25 for individuals less than 2 years old; 23 for individuals from 3 to 5; 18 for kids from 6 to 10; 21, from 11 to 13, and 22 for teenagers (14–18). Higher education is represented by a university with about 700 students.
To run Little Italy, at every tick each agent must be put somewhere on the grid. This requires each agent's list of activities to be put in a one-to-one correspondence with a pair of coordinates. This, in turn, requires a detailed modelling of the agents movements over Little Italy. Details are reported in Text S1.
Little Italy was coded in Java, using Repast 3 libraries [32]. We first drew a large number of alternative configurations in the number of firms and their sizes. From this initial set we discarded those configurations which resulted to be inconsistent with Little Italy. From the consistent set, we selected at random a subset of 100 “worlds”. For each of these worlds, we ran 100 single-day (i.e. each one lasting 144 ticks) simulations. Results obtained from multi-day simulations are not considered because of the limited variability: most agents in Little Italy have small stochastic components in their daily life, the only random elements being the displacement of their house, their office and the paths they follow during the day.
To keep track of contacts between agents, a definition of contact was necessary. The adopted “marker” of contact was “having shared the same physical environment with someone else” (i.e. house, the same class at school, the same bus) during a given tick. This corresponds to a form of localized random mixing. Assume, for instance, that during a given tick there are 20 pupils aged 7 and one teacher aged 44 in a class-room. Based on our definition each pupil has 19 contacts with people of the same age, and one contact with adult people (aged 44), while the teacher has 20 contacts with 7 years old people.
By aggregating across time ticks, matrices reporting the total number of contacts between each pair of ages were computed for the following activity/locations: household, school, work, transport, other activities. Then, by summing through activities we computed overall (i.e. including contacts through all locations) contact matrices by age, whose elements Kij represent the total number of contacts between individual in age-groups i and j. In fact, three different types of contact matrices were computed (we call them Type 1,2,3 matrices). They represent, respectively, the time in contact, the number of episodes of contact, and the number of social partnerships. As an illustrative example, assume that two agents of age i and j, respectively, share the same square on the grid for 6 ticks, then they are elsewhere for some time, and finally they “meet” again for two further ticks. These agents will contribute to the element Kij of the Type 1 matrix with 8 units of “time in contact”. On the other hand they contribute to element Kij of the Type 2 matrix with only 2 contact episodes. Finally, using the definition of contact commonly adopted in surveys [21]–[24], our two individuals contribute to the element Kij of the Type 3 matrix with only 1 unit of contacts. Note that all Types of matrices are symmetric by definition (“If individual A has shared a given location with B, then also B must have shared the same location with A”).
From total matrices Kij and the number ni of individuals in each age group, we computed standard mean contact matrices, i.e. matrices whose entries are the mean numbers mij of contacts with individuals having age j per individual having age i, using the (symmetric) relation Kij = mijni = mjinj = Kji.
Since Little Italy matrices do not offer information on contacts for the age group 0–2 years (it is not included in the Italian Time Use Survey), we integrated our matrices using Polymod data from that age group. These computations were applied to each “world”, and then the average of the ensuing matrices was taken.
We note that the different types of contact matrices considered correspond to different views of the contact process, perhaps useful to capture different aspects of the biology of transmission. Type 1 matrix might be relevant for infections for which the time of exposure matters (for instance, for those infections with low transmissibility rates, where the probability of transmission cumulates over time). Type 3 matrix implies that what really matters is the number of social partnerships, independently of the number of repetition of contact episodes and of the time spent together [24]. Type 2 lies in between, i.e. the transmission depends on repetition of contacts but not necessarily on their duration.
The Big-Italy matrix was extracted from the IBM used to simulate the spread and control of an influenza pandemic in Italy [15]. In this model, differently from Little Italy, each Italian individual is explicitly represented by a model agent. This agent is characterized by age, household membership, and school/workplace membership. The ensuing synthetic population has been obtained by using official socio-demographic data only. In other words, differently from the Little Italy population, it does not include time use data. Since the Big-Italy agents do not physically displace among the different types of locations, only one “general” contact matrix could be computed from the simulation of Big-Italy, by counting the number of contacts among the model agents and then weighting each contact by considering the location where the contact took place (namely households, schools, workplaces, or general community). Details on the computation of the Big-Italy matrix are given in Text S2.
We compared the performances of the Little and Big Italy matrices with two other contact matrices available for Italy: a) the overall (i.e. including all reported contacts) Polymod matrix based on survey data collected in eight European countries [24]; b) the Time Use matrix obtained with the methodology described in [27]. The matrix in (b) relies on the same Time Use Survey as Little Italy, but does not use additional socio-demographic data.
Recently collected Italian serological data (age range 0–79 years, sample size = 2,517) on varicella-zoster-virus (VZV) [33], and ParvoVirus B19 [34] were used. For these infections no mass vaccination programme is in place in Italy, so that their observed immunity profiles may be assumed to represent pre-vaccination equilibrium.
Fitting contact matrices to serological data was performed using a standard approach [22], [27], i.e. by plugging mean contact matrices into a simple age-structured SIR transmission model at its endemic equilibrium. The equilibrium force of infection (FOI, the per-capita probability to acquire the infection per unit of time) is therefore constant in each age group i defining the contact matrix, with the form: , where denotes the infective prevalence at equilibrium in each age group (defined as the ratios between the number of infective people in group j at equilibrium, and the corresponding population size ), and q is a single age-independent transmission parameter. By formally solving the model at equilibrium, and letting D and denote, respectively, the average duration of the infective period and the size of the j-th age group, one gets , where is the susceptible fraction at exact age . The equilibrium FOI in each age group is then determined by solving the system of n nonlinear equations . Once the equilibrium FOI is available, for any given q, the predicted immunity profile at equilibrium z(a) at any given age a is computed as . Finally, the fitting was carried out by maximizing the likelihood of the transmission parameter q in the explanation of the observed age-specific proportions of people immune to VZV and B19 in Italy.
The one-q strategy is a clear way to compare contact patterns, since it implies that the infection process only mirrors contact patterns rescaled by a constant representing infection transmissibility. Since the different contact matrices considered have different scales, the corresponding q's have different units. For example q represents a transmission rate per single tick of time for Little Italy Type 1, while it is measured per single episode of contact for Type 2, etc. The difference in units makes the qs of little direct comparability. For this reason, the comparison of the performances of the various matrices in explaining serological data is not based on the actual q estimates, but only on goodness of fit measures.
Given the estimate of the transmission parameter q, we can compute the “next generation matrices”, NGij = qmij, from which the corresponding Basic Reproduction Number R0 can be obtained [35]. In these cases, R0 is a measure of the potential of invasion of an infection with transmissibility q in a community whose contact patterns are summarized by the contact matrix of elements mij.
In order to achieve a high degree of explanatory power of the data to be used as a benchmark of the goodness of fit of the various matrices, we also considered a flexible non parametric model, given by a constrained monotonically increasing P-splines model [36] (details in Text S3).
We measure assortativeness in the various matrices considered using two different indices. The first one is the Q index [37], defined as where P = [pij] is the matrix whose elements pij represent the fractions of total contacts of age group i with age group j: and denotes the Trace of the matrix. The Q index ranges between zero, corresponding to proportionate mixing, and one, under full diagonal dominance, i.e. fully assortative mixing. Therefore Q represents a measure of departure from proportionate mixing for groups defined on a qualitative scale. The second measure is the dissimilarity-type index [38], defined as the mean square deviation from perfect assortativeness of the contact distribution , normalized by its value under homogeneous mixing. This index is a normalized measure of disassortativeness, ranging between 0, when assortativeness is perfect, and 1, when mixing is homogeneous. For symmetric contact distributions, is related to the correlation coefficients ρXY of the contact distribution as: , where respectively denote the variances of the marginal distribution of contacts with age, and of the age distribution of the population.
Contour plots of Type 1, 2, and 3 average contact matrices based on 5-years age groups (0–4, 5–9, etc) are reported (Fig. 1). The three matrices are very assortative, i.e. the majority of contacts are on the main diagonal, meaning that individuals tend to have contacts with people of the same age. Assortativeness, however, varies significantly across age: it is very pronounced in children, according to the stylized fact that most contacts occur with school classmates, which essentially are of the same age. In particular, the three Little Italy matrices are largely more assortative than all the other matrices considered: for individuals who are less than 15 years old, the proportion of contacts that individuals in each age group have with other individuals in the same age group ranges between 75 and 85% in the three Little Italy matrices, whereas it ranges between 25 and 55% in the Polymod matrix, even less in the Big Italy matrix (Fig. 2). This larger assortativeness of Little Italy matrices is confirmed by the measures (Table 1).
With regards to contacts between parents and their children, well evidenced in [24], [27], these clearly appear in Type 1 matrix (the stripes above and below the main diagonal), whereas they are less sharply defined in Type 2 and 3 matrices. This is explained by the fact that Type 1 matrix takes into account the long time spent by children at home (in most cases with at least one parent) whereas Type 2 and 3 do not. Compared to the Polymod matrix, in all Little Italy matrices household contacts are quantitatively less important because of the stronger assortativeness, which dominates non diagonal contacts. In addition, the lack of appropriate information in the time use diaries probably prevented several contacts between parents and children to be accounted for, thus leading to under estimation. Overall, we can say that Little Italy matrices are dominated by school contacts as a consequence of the assumptions made. The activity-specific matrices used to compute the Type 1 matrix are reported in Text S4.
Table 2 reports the main output of the fit (optimal q estimates, deviance and Akaike Information Criterion, and the corresponding estimates of the Basic Reproduction Number R0) to VZV and B19 data, for all the matrices considered. Results from the non-parametric model are also included. Graphic comparisons between observed and predicted sero-profiles by age are displayed in Fig. 3 and Fig. 4.
For VZV, the Polymod matrix provides the best fit. The Big-Italy and the Little Italy Type 2 matrices perform better than the Time Use matrix but substantially worse than the Polymod matrix, whereas the Little Italy Type 1 and 3 matrices fit poorly. Note that the non-parametric model performs slightly better than the Polymod matrix in terms of deviance, but worse in terms of AIC, due to its larger parameterization. This suggests that the Polymod matrix definitely represents an excellent “explanans” for VZV transmission. Disregarding the Little Italy Type 1 and 3 matrices, which poorly fit, the ensuing values for are in good mutual agreement (ranging between 4 and 5), and higher than the R0 estimates reported for Italy in [33]. We also note that both the poorly fitting Little Italy Type 1 and 3 matrices lead to much smaller R0 values. This follows from the limited ability of these matrices to capture contact patterns relevant for VZV. As a result, we observe a compensation through anomalously small values of the infectivity parameter q.
Things are different for B19. The Type 1 matrix provides the best fit, and overall the three Little Italy matrices perform better than the other matrices. It is however to be acknowledged that the fit remains far from the one provided by the non-parametric model, suggesting that there is still room for large improvements in the explanation. In particular, the Big-Italy and the Time Use matrices, though clearly less performant than the Little Italy matrices, are not worse than the Polymod matrix. The ensuing values for R0 range between 1.6 and 2.6. An explanation of the differences in the fit of B19 and VZV is not easy since we do not dispose of tools to globally compare the differences between two arbitrary contact matrices. Assortativeness measures provide however some clue. The three Little Italy matrices predict a very steep immunity profile at low ages, which however suddenly flattens to a plateau later on. This sudden change in regime, which is a pattern known to occur in presence of strong assortativeness, allows the Little Italy matrices to better explain the B19 data, which show a sharp plateauing (though with large randomness). On the other hand, this behavior prevents the Little -Italy matrices to capture the observed VZV profile.
Finally, given that the large-scale (transport and shopping malls) contacts of the Little Italy model required several assumptions to be parameterized, it was important to check the influence of these activities/locations on the result of the fit. We therefore fitted the Little Italy matrices without taking into account such activities, i.e. relying only on households and school/workplaces contacts. The results in the fit of B19 by Little Italy Type 1 and 3 matrices are reported (Table 3). In the situation where the Little Italy model performs better, i.e. the fit of B19 by the Type 1 matrix, the exclusion of transportation and shopping malls worsens the goodness-of-fit very little, indicating that these components only marginally affect the structure of the matrices.
Substantial improvements have been achieved in recent times in our knowledge of social contact patterns [20]–[27], which are thought to be a key factor underlying the transmission dynamics of close-contact infections. In this paper, we have investigated the potentialities of IBM as a tool for the generation of contact data, with two distinct approaches. The first approach is a novel one, based on a simple socio-demographic IBM (“Little Italy”) strictly integrating time use and routine socio-demographic data. As for the second, we have extracted the contact matrix by age (“Big-Italy”) implicit in the socio-demographic model underlying the Italian IBM for pandemic prediction and control. Both models are based on the same routine socio-demographic data, but the “Little Italy” model also considers the agents' daily allocation of time through Time Use data. The Little Italy approach allows for the computation of different types of contact matrices, labelled Type 1,2,3, reflecting respectively (a) the average time in contact, (b) the average number of repetition of contacts, (c) the average number of different persons contacted.
The ensuing contact matrices by age were fitted, on the basis of simple transmission models, to Italian serological data for VZV and B19. Goodness-of-fit comparisons with other available contact matrices, such as the questionnaire-based Polymod matrix and the Time-use matrix, were also made. The main results show that for VZV the best fit is provided by the Polymod matrix, which performs excellently, and much better than artificial matrices. However, for B19, all Little Italy matrices fit the data quite well, and better than available concurrent matrices, including the Polymod one.
This paper represents, as far as the authors know, the first comparison on real epidemiological data of bottom-up approaches to the generation of contact data, with the approaches based on direct contacts estimation, such as the Polymod study. Our results on VZV provide further evidence on the merits of the Polymod study, which represents a great advancement in our understanding of contact patterns. However, the better fit to B19 provided by artificial matrices compared to questionnaire-based matrices, is indicative of the difficulty to find “universal” contact patterns that can explain in a satisfactory way many different infections. Therefore, though artificial matrices can not surrogate observed ones, they can certainly represent valuable tools to assist mathematical modellers in the formulation of alternative assumptions.
An important related question is why different infections are better explained by different types of contact matrices. May this be due to the characteristics of the contacts which matter to various infectious diseases? The traditional WAIFW [3] and proportionate mixing [19] approaches, which were strongly constrained by data availability, considered the various diseases separately, as if they were outcomes of fully independent processes. Recent approaches [20]– have promoted the different idea that for a large family of infections there might be a unique “core” of observable social contact patterns, mediated through a unique, or a few, infection-specific transmission parameters. These new approaches raise a number of questions: first of all, whether the transmission biology of different infections could selectively exploit different types of contact patterns. Though this is still unclear, there is evidence that the infection-specific hit probabilities per single viral or bacterial unit occupy a wide range. This would suggest that for some infections, such as measles, even very short single episodes of contacts might be sufficient for transmission. Therefore, it is likely that most adequate contacts are usually “wasted”. On the other hand, there might be infections (e.g., bacterial ones) characterized by a very low hit probability, for which many repetitions of contact episodes, or long exposure times, might be necessary for successful transmission. Our results, i.e. the fact that for a mildly transmissible infection such as B19, the best fit is obtained using a matrix counting time spent in contacts, as opposite to VZV, where the best fit follows from a matrix only counting encounters with different individuals, irrespective of time of exposure, might support this idea.
With regards to model parameterization, the Little Italy model uses real data to parameterize the small scale components (household sizes, schools, workplaces) of the contact network. On the other hand, assumptions were necessary to parameterize the large scale components of the network, e.g. travel and shopping malls patterns. Nonetheless, we could at least make such patterns fully consistent with the general design of the Little Italy model, i.e. the daily time spent on travelling, or in supermarkets, by each Little Italy agent, correctly matches, based on an optimization procedure, the time spent on travelling by a corresponding real agent. In order to appreciate the potential impact on data fitting of the ad-hoc assumptions on travel patterns, we also fitted the model by excluding contacts on transports and shopping mall, showing that in the most significant cases the results were essentially unaffected. This suggests that the “empirically robust” component of the model is sufficient for the main target of the paper, i.e. the generation of contact data. Obviously, given the lack of appropriate epidemiological data to validate travel assumptions, the possibility to use Little Italy for further investigations beyond those presented here, i.e. for example epidemic prediction and information of measures targeting social distance, certainly requires caution. Future work will be devoted to the analysis of the model robustness to the assumptions on its large scale components.
Given the simplicity of the adopted definition of contact, the current model cannot reproduce, unless resorting to further data and hypotheses, the richness of data obtained by Polymod survey, where further noteworthy information such as the intimacy and frequencies of contacts, were collected. This is clearly a shortcoming since these types of contacts are arguably important for most respiratory infections [24].
However the current Little Italy model can potentially be used to answer several important questions. For example, the model can be expanded to describe contacts in a rural-urban environment, given the representativeness of Italian Time Use data for rural and urban populations. Moreover, longer time simulations could address how contacts cumulate (a) during periods of time having a length comparable to the infectivity period, (b) between work-days and week-end days [39]. We indeed recall that, although in this paper we considered work-days only, the Italian Time Use data actually include three distinct samples, one for working days, the other two for Saturdays and Sundays. This provides information on how time spent in the various activities/locations cumulates through the different parts of the week. Obviously, studies of contacts accumulation are difficult, as they are necessarily conditional on the specific assumptions made on the larger-scale topology of the Little Italy network, e.g. contacts on transportations, shopping malls, and so on. Nonetheless, in recent times, the first empirical evidence on this issue has become available [25], and may provide a useful starting point for comparison of contact accumulation in different social settings.
Further, this paper would like to reinforce the perspective that contact data and time-use data provide useful complementary information. On the data-gathering side, major gains could be achieved by combining the two approaches. This could be achieved, for example, by supplementing time-use surveys with a few questions about people “contacted” (for example those with whom a conversation was held) during any given activity or time slot. This would provide data that consistently incorporate the relationship between time of exposure and contacts. With regards to studies of transmission, it would be important to better understand how to integrate the two types of data, for example by comparing time use data and Polymod data on durations of contacts.
A final point regards the information embedded in age specific serological data, which are the base for infection control strategies. As clear for example for VZV [33], these data show a fast monotonic increase during school ages, say up to age 10–15, then the trend becomes flat, or slightly increasing with age, but with large randomness. This suggests that these data have little discriminating power about infection patterns at higher ages, which are critically important when control measures are in place. Therefore, it would be important to improve our understanding of infection patterns among adults, for example by grounding stochastic models of age mixing against simulation derived matrices (and related seroprofiles by age). On a related topic, in our models we are still relying on the assumption of monotonic seroprofiles. This assumption follows from postulating an infection which (a) is at steady state, and (b) decouples from the underlying dynamics of the population. If these hypotheses are not met, seroprofiles can become non monotonic. Recent work [40]–[41] has aimed at considering infection dynamics in non-steady populations, or non steady contact networks. This work has suggested the importance of population structures in shaping contact patterns, and therefore the intrinsic instability of contact matrices over time. Time is ripe for bringing such non stationary approaches also in epidemiological data analyses.
|
10.1371/journal.ppat.1004963 | Activated Brain Endothelial Cells Cross-Present Malaria Antigen | In the murine model of cerebral malaria caused by P. berghei ANKA (PbA), parasite-specific CD8+ T cells directly induce pathology and have long been hypothesized to kill brain endothelial cells that have internalized PbA antigen. We previously reported that brain microvessel fragments from infected mice cross-present PbA epitopes, using reporter cells transduced with epitope-specific T cell receptors. Here, we confirm that endothelial cells are the population responsible for cross-presentation in vivo, not pericytes or microglia. PbA antigen cross-presentation by primary brain endothelial cells in vitro confers susceptibility to killing by CD8+ T cells from infected mice. IFNγ stimulation is required for brain endothelial cross-presentation in vivo and in vitro, which occurs by a proteasome- and TAP-dependent mechanism. Parasite strains that do not induce cerebral malaria were phagocytosed and cross-presented less efficiently than PbA in vitro. The main source of antigen appears to be free merozoites, which were avidly phagocytosed. A human brain endothelial cell line also phagocytosed P. falciparum merozoites. Besides being the first demonstration of cross-presentation by brain endothelial cells, our results suggest that interfering with merozoite phagocytosis or antigen processing may be effective strategies for cerebral malaria intervention.
| Cerebral malaria accounts for most of the deaths caused by Plasmodium infection. In the mouse model of cerebral malaria, CD8+ T cells are known to be the effector cells responsible for lethal neuropathology, but it was not clear how they disrupted the blood-brain barrier. Here, we show that brain endothelial cells cross-present parasite antigen at the onset of pathology, hence allowing recognition by parasite-specific cytotoxic T lymphocytes. This process did not occur in mice lacking IFNγ, whereas TNFα and LTα were dispensable. The proposed mechanism of pathogenesis was recapitulated in vitro: IFNγ-stimulated primary mouse brain endothelial cells cultured with parasite-infected red blood cells were subsequently killed by CD8+ T cells isolated from an infected mouse. The murine endothelial cells primarily phagocytose not infected red blood cells but instead merozoites, the form of the parasite that infects red blood cells. We show also that human brain endothelial cells phagocytose P. falciparum merozoites in vitro, suggesting that our findings with the mouse model may be relevant to human disease. Strategies to interfere with merozoite phagocytosis or antigen processing by endothelial cells may prove useful for treating cerebral malaria.
| Half the world’s population is at risk of malaria infection, which is estimated to kill half a million children under the age of 5 annually [1]. Most of these fatalities occur due to a severe complication of Plasmodium falciparum (Pf) infection called cerebral malaria, with clinical features of impaired consciousness, seizures and abnormal posturing. Autopsies frequently reveal brain swelling and petechial hemorrhages, and most characteristically, dense sequestration of parasitized red blood cells in many brain microvessels [2]. Mechanistic understanding of the etiology of cerebral malaria remains elusive, given the ethical limitations of research in human patients. The mouse model of experimental cerebral malaria (ECM) induced by P. berghei ANKA (PbA) infection recapitulates many features of the human disease including parasite accumulation in the brain, albeit controversially to a much less prominent degree [3]. Extensive evidence has emerged that ECM is an immune-mediated disease, with roles described for CD4+ and CD8+ T cells [4–6], γδ T cells [7], NK cells [8], NKT cells [9], neutrophils [10], monocytes [11], microglia [12], and splenic CD8+ dendritic cells [13,14]. Amongst these cell types, CD8+ T cells play a unique effector role in ECM pathogenesis as their depletion one day before neurological symptoms are expected prevents disease [5]. In contrast, CD4+ T cells [5], γδ T cells [7] and neutrophils [10] have to be depleted early to be efficacious, and NK cells and CD4+ T cells in particular were found to act by recruiting CD8+ T cells to the brain via IFNγ [8,15,16]. Adoptive transfer experiments revealed that the pathogenicity of CD8+ T cells was dependent on perforin and Granzyme B expression [6,17], suggesting that their cytolytic function was directly responsible for the loss of blood-brain barrier integrity observed in ECM.
In the past few years, we and others have identified a number of PbA blood-stage epitopes, confirming the pathogenic role of antigen-specific CD8+ T cells in ECM [18–21]. By transferring TCR-transgenic CD8+ T cells (PbT-I T cells recognizing the PbA epitope NCYDFNNI) into hosts depleted of endogenous CD8+ T cells, Lau et al. showed definitively that PbA-specific CD8+ T cells can induce lethal neurological damage [20]. However, several of the known blood-stage epitopes are also conserved in other rodent malaria species or strains that do not cause ECM, and we found to our surprise that cytotoxic CD8+ T cells specific to these epitopes are also induced during these infections [19,21]. We asked whether the difference between PbA and non-ECM-causing strains could lie in the prevalence of targets for CD8+ T cell-mediated cytolysis amongst the cells constituting the blood-brain barrier i.e. cells that were cross-presenting parasite antigens. We first addressed this question using NFAT-lacZ reporter cells transduced with a TCR recognizing the Pb1 epitope (SQLLNAKYL, from the GAP50 protein), which turn blue following X-gal staining if they encounter Pb1 in the context of H-2Db MHC class I molecules [19]. These LR-BSL8.4a reporter cells were incubated with brain microvessel fragments isolated from naïve mice and infected mice. Only brain microvessels from PbA-infected mice, but not those from mice infected with non-ECM-causing P. berghei NK65 (PbNK65) or P. yoelii 17XNL (Py17X), gave rise to elevated numbers of blue reporter cells. Further experiments with reporter cells recognizing two other epitopes gave similar results—only PbA infection led to brain microvessel cross-presentation of parasite antigen, supporting the proposition that such cross-presentation is a necessary step in ECM pathogenesis [21]. Another indication that CD8+ T cells need to act on cross-presenting brain microvessel cells to cause ECM comes from experiments in which mice were rescued from ECM by treatment with anti-malarial drugs one day before symptoms were expected. Compared to untreated mice, the treated mice had similar numbers of antigen-specific brain-sequestered CD8+ T cells, but brain microvessel cross-presentation was severely reduced [19].
The importance of proinflammatory cytokines in ECM pathogenesis has been a topic of considerable interest. A crucial role for IFNγ was demonstrated using mice deficient for either the cytokine [4] or its receptor [22]. The association between TNFα and human cerebral malaria and ECM has been extensively studied, but Engwerda et al. showed that TNFα knockout (KO) mice were susceptible to ECM while LTα KO mice were protected [23]. Here, we first investigate the cytokine requirements for cross-presentation during ECM and identify the cell type involved, then develop an in vitro PbA cross-presentation model to gain mechanistic insight.
We examined the roles of IFNγ, TNFα and LTα in brain microvessel cross-presentation during PbA infection by comparing mice deficient in each cytokine with wild type (WT) C57BL/6J mice. Brain microvessel fragments were isolated from infected and naïve mice and incubated overnight with the reporter cell line, LR-BSL8.4a, to measure Pb1 presentation. Brain microvessels from IFNγ KO mice yielded background numbers of blue spots (each corresponding to a responding reporter cell), showing that IFNγ is necessary for cross-presentation in the brain (Fig 1A). In contrast, both TNFα and LTα are dispensable for brain microvessel cross-presentation as the assay readouts in the knockout mice were not significantly different or even higher than in WT mice (Fig 1B and 1C). The higher cross-presentation levels in LTα KO mice may indicate impaired CD8+ T cell-mediated killing of cross-presenting cells or increased survival of the latter. We also investigated the timing of brain microvessel cross-presentation in PbA-infected WT mice. Cross-presentation was not observed 5 days post-infection, started to occur at intermediate levels on day 6 and reached high levels on day 7, when ECM signs typically emerge (Fig 1D).
In our earlier work, we were unable to identify the brain microvessel cell type that was responsible for cross-presentation, because of the difficulty in obtaining a single cell suspension of viable cells with collagenase digestion [19]. However, we noted that completely eliminating collagenase treatment led to much lower activation of reporter cells, suggesting that the cross-presenting cells may be endothelial cells or pericytes that are surrounded by collagen-rich basal lamina. We have continued to experiment with different digestion enzymes and protocols to allow brain microvessel cells to be sorted. Digestion of microvessels with Liberase DH (Roche) yielded viable single cells that could be separated by FACS, but none of the sorted populations were able to activate LR-BSL8.4a reporter cells.
Ultimately, we succeeded using a papain-based Neural Tissue Dissociation Kit (Miltenyi) with automated mechanical dissociation of the entire brain. Single cell suspensions from naïve and PbA-infected mouse brains were separated into four populations after antibody labeling. First, all CD45+ cells (microglia and leukocytes) were gated, then the CD45- cells were further subdivided into CD31+ endothelial cells, a CD140b+ population including pericytes, and triple negative cells such as astrocytes and neurons (S1A Fig). After overnight incubation with LR-BSL8.4a reporter cells and X-gal staining, only the CD45+ and CD31+ populations gave rise to elevated numbers of blue spots comparing infected mice to naïve mice (Fig 2A). The numbers of blue spots induced by sorted endothelial cells were much smaller than those typically arising from brain microvessel fragments (tens compared to hundreds). One reason could be the low yield of endothelial cells with the Neural Tissue Dissociation Kit, averaging 1.7 × 103 cells sorted per brain compared to 1.6 × 104 cells from Liberase digestion of isolated brain microvessels. Further, while papain was selected because it largely preserves CD31 staining, its ability to cleave the heavy chain of MHC class I molecules is well-known. Despite reducing the papain concentration, it is possible that we were only able to detect “new” Pb1-H-2Db complexes that were generated in vitro from intracellular antigen stores. Because endothelial cells were isolated with much lower efficiency than leukocytes, we have also plotted the numbers of blue spots normalized by the input cell number, making it evident that endothelial cells were by far the most efficient cross-presenters on a per-cell basis (Fig 2B).
Since the CD45+ population also gave a positive signal, we wished to clarify whether microglia were responsible. The ability of microglial cells to cross-present soluble antigens has recently been demonstrated in vivo [24], and evidence of malaria antigen cross-presentation by these cells would have important implications. In the follow-up experiments, brains were digested with collagenase (papain digestion being unnecessary and disadvantageous as noted above) for sorting into CD45intCD11b+ microglia and CD45hi peripheral leukocytes and perivascular macrophages (PVM) as shown in S1B Fig. Cross-presentation of the Pb1 epitope was detected from the CD45hi fraction but not from microglia from infected mice (Fig 2C). Thus, while microglial cells have the potential to cross-present, they do not do so appreciably during PbA infection, perhaps because of limited access to antigen.
To further understand how brain endothelial cells cross-present during ECM, we sought to recapitulate the process in vitro using primary cultures of murine brain endothelial cells (MBECs). Brain microvessels isolated from naïve mice were cultured on collagen-coated plates using puromycin to kill non-endothelial cells for the first 2–3 days [25]. In our hands, after 10 days of culture, >97% of the cells were CD31+ endothelial cells, with essentially no CD45+ cell contamination and <0.5% NG2+ pericyte contamination (Fig 3A). The cultures also expressed the endothelial marker von Willebrand factor (Fig 3B). We always used unpassaged primary MBEC cultures to retain the original phenotype as much as possible. Unstimulated MBECs and IFNγ-stimulated MBECs were incubated with frozen-thawed PbA mature iRBCs (late trophozoites and schizonts) for 24 h, after which they were washed and co-incubated with LR-BSL8.4a reporter cells overnight. After X-gal staining of the reporter cells, only MBECs that had been exposed to both IFNγ and PbA-infected iRBCs were found to have cross-presented (Fig 3C). Both the ability of endothelial cells to cross-present PbA antigen and the necessity of IFNγ stimulation agree with our ex vivo results from infected mice. Similar results were obtained with reporter cell lines that we created to detect cross-presentation of two additional PbA epitopes (S2A and S2B Fig). Frozen mature iRBCs were used as antigen in these and many subsequent experiments for convenience; we compared these to freshly isolated PbA mature iRBCs and found no significant difference in cross-presentation efficiency (Fig 3D). We also confirmed that uninfected RBCs gave readouts indistinguishable from the background (S3A Fig).
Neither TNFα nor LTα were able to induce MBEC cross-presentation, nor could they enhance cross-presentation when used in conjunction with IFNγ (S2C Fig). The ability of MBECs to cross-present appears to be general rather than specific to malaria antigen: cross-presentation of soluble ovalbumin occurred even without IFNγ stimulation but was enhanced by it (S2D Fig). Wheway et al. recently showed that a human brain endothelial cell line expressed high basal levels of MHC class I not further enhanced by IFNγ [26]. In contrast, we found MHC class I expression on MBECs to be highly IFNγ-inducible (S2E Fig), which may partly explain the importance of this cytokine in cross-presentation. In addition, IFNγ upregulated MBEC expression of ICAM-1 but not VCAM-1 (S2E Fig).
The prevailing model of ECM pathogenesis attributes severe disruption of the blood-brain barrier to cytolysis of cross-presenting endothelial cells by PbA-specific cytotoxic T lymphocytes. To test this model, instead of adding reporter cells to IFNγ-stimulated MBECs pulsed with PbA mature iRBCs, we added CD8+ T cells isolated from a PbA-infected mouse just prior to ECM. The MBECs were indeed almost entirely dead after 20 h of co-incubation (Fig 4A). When PbA antigen was omitted, the activated CD8+ T cells adhered to the endothelial cells but did not kill them. When CD8+ T cells from a naïve mouse were added to PbA-pulsed MBECs, the endothelial monolayer also remained largely intact, although there may have been some MBEC dysfunction caused by interaction with parasite molecules (reviewed by Razakandrainibe et al. [27]). Hence, this in vitro experiment confirms that MBECs cross-presenting PbA antigen can be recognized for killing by the parasite-specific CD8+ T cell response.
To observe interactions between CD8+ T cells and endothelial cells in situ, we performed immunofluorescence staining of olfactory bulb smears. The olfactory bulb is the site of early and severe damage in ECM [28]. Brain smears allow the endothelial cells to be visualized more efficiently compared to tissue sections cutting through blood vessels at random planes. In mice with ECM, CD8+ T cells were found not only within blood vessels, but also on the abluminal face of endothelial cells and in the parenchyma (Fig 4B). The perivascular CD8+ T cells were flattened and in intimate contact with the endothelial cells (Fig 4B inset), indicating cell-cell interaction. No CD8+ T cells were observed in olfactory bulb smears from naïve mice. It has to be kept in mind that during ECM, the number of CD8+ T cells for a whole mouse brain is less than 100 000, of which only a fraction may be specific for parasite antigens. Thus, the few CD8+ T cells observed in the smears are consistent with the paucity of these cells in the brain of mice with ECM. In addition, the strong killing as seen in vitro with high CD8+ T cell to endothelial cell ratio is unlikely to be observed in vivo.
During cross-presentation, the peptides to be loaded onto MHC class I molecules may either be generated by the proteasome for subsequent transportation by TAP (the cytosolic route) or by endolysosomal proteases (the TAP-independent, vacuolar route). To investigate the molecular mechanism of MBEC cross-presentation of PbA antigen, we first cultured MBECs from TAP1-deficient mice. In contrast to WT MBECs, MBECs lacking TAP1 were completely unable to activate LR-BSL8.4a cells after IFNγ stimulation and exposure to PbA (Fig 5A). We used the proteasome inhibitor lactacystin to study the role of proteasomes. We encountered some technical difficulty performing the experiment because prolonged lactacystin exposure was toxic to the MBECs, while cross-presentation required extended co-incubation with PbA mature iRBCs. We found that 6 h of exposure to 10 μM lactacystin did not affect the viability or MHC class I expression level of MBECs, as assessed by their ability to activate LR-BSL8.4a cells after pulsing with Pb1 peptide (S2F Fig). MBECs were therefore incubated with PbA mature stages for 12 h, with lactacystin present during the second half of this time, before the wells were washed and assayed for Pb1 cross-presentation. The number of blue spots obtained from lactacystin-treated wells was significantly reduced (by about 70%) compared to wells without inhibitor, revealing the proteasome-dependence of cross-presentation (Fig 5B). Conversely, chloroquine, an inhibitor of endosome acidification and thus the vacuolar route, did not reduce cross-presentation efficiency when present throughout the 24 h incubation with PbA antigen (Fig 5C). Instead, the number of blue reporter cells increased significantly, suggesting that endolysosomal proteases destroyed rather than generated the MHC class I epitope. MBEC cross-presentation of PbA mature iRBCs was sensitive to cytochalasin D (Fig 5D), consistent with but not conclusive for antigen uptake by phagocytosis. No cross-presentation was seen when PbA mature stages were separated from MBECs by a Transwell support with 0.4 μm pores, arguing against soluble proteins being the source of cross-presented antigen (S3B Fig). Taken together, the evidence points towards PbA cross-presentation by the phagosome-to-cytosol route, requiring proteasomes and TAP.
To seek more evidence of phagocytosis, we labeled needle-sheared PbA mature stages (a mixture of free merozoites, late trophozoites and less mature schizonts) with PKH26, a red fluorescent membrane dye. After MBECs were incubated with PKH26-labeled parasites for 24 h, they were washed and stained with LysoTracker Green DND-26 to identify the acidic compartments (endosomes, lysosomes and phagosomes that have fused with these). The red fluorescence of parasite material that had adhered to MBECs but that had not been internalized was quenched with trypan blue [29]. Uptake of parasite material was quite heterogeneous, with some MBECs containing many red particles and others none. More importantly, red fluorescence was very well colocalized with LysoTracker (Fig 6A), indicating that the parasite material had entered the phagolysosomal pathway. To our surprise, the vesicles with colocalized red and green fluorescence were generally only 1–2 μm in diameter, consistent with the size of merozoites. Although iRBCs comprised most of the PbA material added to the wells, we did not observe any phagosomes that were large enough to contain intact mature stages (≥5 μm). To confirm that the phagosomes contained merozoites, MBECs that had been co-incubated with thawed PbA mature iRBCs were fixed and stained with rabbit antiserum recognizing merozoite surface protein 1 (MSP-1). Many of the vesicles were indeed positive for MSP-1, although under light microscopy, other dark vesicles containing malaria pigment were also observed, indicating that both merozoites and digestive vacuoles had been phagocytosed (Fig 6B).
Although the descriptive evidence suggests that MBECs phagocytosed merozoites and digestive vacuoles from ruptured schizonts, an alternative explanation could be that whole schizonts were phagocytosed, and subsequently broken down into individual merozoites and digestive vacuoles that split off into separate phagosomes. To investigate further, we compared how efficiently MBECs cross-presented an equal number of PbA mature iRBCs versus individual merozoites. One million merozoites, produced by passing mature stages through a 1.2 μm syringe filter followed by magnetic depletion of hemozoin [30], led to twice as many blue spots as one million mature iRBCs, even though the latter likely contained more than a million nascent merozoites (Fig 6C). These results support the hypothesis that merozoites are the parasite form avidly phagocytosed by MBECs and the primary source of antigen for cross-presentation. We suggest that when schizonts are added, cross-presentation efficiency is reduced by the additional requirements of rupture of the parasitophorous vacuole membrane (by parasite proteases, freeze-thaw or degradation) and merozoite egress.
We have previously postulated that differences in parasite sequestration levels in the brain may explain why PbA is cross-presented by brain microvessels during infection while the non-ECM-causing parasites PbNK65 and Py17X are not [19,21]. Here, we asked if there were also intrinsic differences in how efficiently MBECs capture and cross-present iRBCs from the different parasites. The Pb1 epitope and surrounding amino acids are fully conserved in the GAP50 proteins of these three parasites, allowing this comparison to be made [19]. Percoll-isolated mature stages of the three parasites were carefully counted and added at three different doses to IFNγ-stimulated MBECs. Significantly less cross-presentation was detected from Py17X compared to PbA at all doses (Fig 7A). PbNK65 was cross-presented with intermediate efficiency but had a relatively flat dose response curve, such that the cross-presentation readout was about one-third that of PbA at the highest dose. We infer that differences in both per-iRBC cross-presentation efficiency and local parasite abundance in the brain may combine such that only PbA induces a pathogenic, detectable level of MBEC cross-presentation during infection. To further investigate whether the differences in cross-presentation efficiency arose from differential uptake of the parasite strains, equal numbers of PKH-labeled PbA, NK65 or Py17X mature stages were added to MBECs. After overnight incubation, the wells were washed and imaged in the presence of trypan blue to quench extracellular and quantify intracellular PKH26 fluorescence. The amounts of internalized NK65 and Py17X were significantly lower than that of PbA (Fig 7B), and the relative levels suggest that this may be the primary factor behind the reduced cross-presentation efficiencies. The uptake level of PbA in MBECs that had not been stimulated with IFNγ was also quantified. Unexpectedly, IFNγ stimulation did not increase parasite uptake and may even have decreased it slightly, although this did not reach statistical significance (Fig 7B).
Currently, the lack of known blood-stage Pf epitopes and cognate TCRs is a barrier to determining whether human brain endothelial cells cross-present parasite antigen. Instead, we asked if human brain endothelial cells can at least phagocytose merozoites. The hCMEC/D3 line was stimulated with IFNγ and co-incubated with PKH26-labeled Pf merozoites for 24 h. Confocal fluorescence imaging was performed after LysoTracker and Hoechst staining, with trypan blue quenching of unphagocytosed parasites. We observed that almost all the hCMEC/D3 nuclei (82/89) were associated with red fluorescent vesicles 1–2 μm in diameter that colocalized with LysoTracker (Fig 8). We conclude that merozoites were avidly phagocytosed by hCMEC/D3 cells, leaving open the possibility that brain endothelial cross-presentation of parasite antigen may play a role in the pathogenesis of human cerebral malaria.
More than a decade ago, we and others quantified the migration of CD8+ T cells to the brain during ECM and speculated that they may exert their cytotoxic effector functions on antigen-presenting endothelial cells [5,6]. Although the ability to acquire and present exogenous antigens on MHC class I molecules was originally thought to be limited to professional antigen-presenting cells, it has become increasingly clear that endothelial cells are conditional antigen-presenting cells that cross-present when activated by danger signals or cytokines (reviewed by Mai et al. [31]). Nevertheless, evidence of endothelial presentation of malarial antigens has been elusive, in part because of the lack of known MHC class I epitopes and methods to detect them. Some circumstantial evidence was provided by Jambou et al. demonstrating that a human brain endothelial cell line could acquire proteins from Pf iRBCs [32]. We recently developed reporter cells for detecting three immunogenic class I epitopes from PbA and found that microvessel fragments from the brains of ECM-afflicted mice presented them [19,21]. However, the question of whether endothelial cells were the cross-presenting cell type remained unresolved.
Here, we present definitive evidence that brain endothelial cells in PbA-infected mice cross-present PbA antigens during ECM. Whole mouse brains were dissociated using papain and sorted into different cell populations, of which only endothelial cells and CD45+ cells presented Pb1 peptide. We had also previously considered pericytes to be candidate cross-presenting cells. Pericytes are phagocytic conditional antigen-presenting cells [33] and are embedded within the basal lamina, in line with the importance of collagenase treatment for detecting cross-presentation [19]. In fact, we found that murine brain pericytes cultured in vitro had the capacity to cross-present Pb1 peptide from PbA iRBCs when stimulated with IFNγ (S4 Fig). However, we did not detect cross-presentation from pericytes isolated from PbA-infected mice, suggesting that pericytes had limited access to iRBCs in vivo. We cannot exclude the possibility that a low level of pericyte cross-presentation, below the assay detection limit, occurred at endothelium breach sites. Amongst the CD45+ population, peripheral leukocytes and/or PVM but not microglial cells were found to cross-present PbA antigen, pointing again to the importance of antigen accessibility. We have previously shown that total brain-sequestered leukocytes cross-presented an order of magnitude less than brain microvessel fragments [19]. Here, the similar numbers of blue spots elicited by sorted endothelial cells and leukocytes can be attributed to the low yield of unicellular endothelial cells, and perhaps reduced cross-presentation ability after disruption of cell-cell junctions. Normalizing the spot counts by the number of sorted cells added makes it clear that MBECs cross-present PbA antigen during ECM and are the main cell type responsible for the results of the brain microvessel cross-presentation experiments. Cross-presentation by brain-sequestered leukocytes is unlikely to play a major role during the effector phase of ECM, as mice depleted of macrophages and dendritic cells from day 5 onwards using the MAFIA (macrophage Fas-induced apopotosis) transgenic mouse system still succumb to ECM and exhibit undiminished brain microvessel cross-presentation [19].
Because of the crucial role of endothelial cells in forming the blood-brain barrier, which is disrupted during ECM, and evidence of their apoptosis in a perforin-dependent manner [34], endothelial cell killing has been proposed to be major pathogenic process in ECM. Endothelial cell apoptosis during ECM has since been confirmed by other approaches [35,36] but has also been disputed [37]. We show here that MBECs cross-presenting PbA antigen in vitro are killed by CD8+ T cells induced during infection. However, while this experiment demonstrates the capacity of such cytolysis to occur, the high CD8+ T cell to endothelial cell ratio used in vitro is not intended to reflect the small number of CD8+ T cells found in the brain in vivo [5]. Direct killing of endothelial cells may not be the only (or even major) mechanism of blood-brain barrier disruption by CD8+ T cells. While CD8+ T cells were previously thought to be present only within the vasculature [38], intravital microscopy has recently revealed that some CD8+ T cells extravasate during ECM [39]. Our olfactory bulb smears confirm that CD8+ T cells are found in contact with both luminal and abluminal faces of endothelial cells as well as in the parenchyma. This suggests an alternative explanation for the importance of perforin and Granzyme B during ECM. Granzyme B plays an important role in breaking down the basement membrane during diapedesis by cytotoxic lymphocytes [40]. Transgenic CD8+ T cells have been shown to cross the blood-brain barrier at sites where the cerebral endothelium presented their cognate epitope [41]. Perforin-expressing CD8+ T cells have also been shown to disrupt tight junctions between endothelial cells pulsed with their cognate peptide [42,43]. Therefore, we propose that endothelial cells cross-presenting PbA antigen in vivo may promote CD8+ T cell extravasation and thus remodeling of the basement membrane and tight junctions in a perforin- and Granzyme B-dependent manner. The relative importance of this proposed mechanism versus endothelial cell apoptosis during ECM remains unclear.
We have shown here that MBEC cross-presentation of malarial antigen requires IFNγ both in vivo and in vitro. IFNγ can be viewed as the central cytokine coordinating the key players in ECM pathogenesis, directing CD8+ T cell migration to the brain [15,44], mediating parasite sequestration [45,46], and now enabling endothelial cross-presentation of PbA antigen. These three phenomena develop coincidentally in the brain 6 days post-infection [19,46], just prior to and probably precipitating the development of neurological signs. Note that because parasite biomass in the brain is significantly reduced in IFNγ-deficient mice [45,46], it is not clear whether the lack of brain microvessel cross-presentation in these mice results from low antigen availability or low endothelial cross-presentation ability or both. Our in vitro experiments conducted in the presence of ample parasite material are consistent with IFNγ stimulation causing endothelial cells to gain the ability to cross-present phagocytosed material. The role of IFNγ in enhancing antigen processing and presentation has been well studied. In the context of class I antigen presentation, IFNγ has been shown to upregulate immunoproteasome subunits, proteasome activator 28, TAP, tapasin, MHC class I heavy chains, β2-microglobulin (reviewed previously [47]) and ERAP1 [38]. Microarray analysis has revealed that genes involved in antigen processing and presentation, including TAP and β2-microglobulin, are upregulated in the brains of mice with ECM compared to infected ECM-resistant BALB/c mice [48]. In addition, we speculate that IFNγ may play a role in enhancing endothelial uptake of malarial antigens in vivo by upregulating receptors associated with adhesion and/or phagocytosis. Parasite cytoadherence to brain endothelial ICAM-1 via Pf erythrocyte membrane protein 1 (PfEMP1) is strongly implicated in human cerebral malaria [49]. The picture is less clear in ECM as PbA lacks PfEMP1 orthologues, and a recent in vitro model of PbA adhesion to immortalized MBECs was unaffected by ICAM-1 blocking [50]. With primary MBEC cultures, we observed that although IFNγ stimulation did not enhance parasite phagocytosis under static conditions, it increased surface expression of ICAM-1 (S3C Fig). ICAM-1 also mediates leukocyte adhesion, and in particular, is a ligand for LFA-1 highly expressed by PbA-specific CD8+ T cells in the brain [21].
Unlike IFNγ, LTα was not required for brain microvessel cross-presentation during infection. While LTα-deficient mice have been shown to be ECM-resistant [23], the role of LTα in ECM remains poorly understood. In wild-type mice, we expect some of the cross-presenting brain endothelial cells to be killed by cognate CD8+ T cells, thus reducing the readout of the brain microvessel cross-presentation assay. Thus, both ECM resistance and the increased brain microvessel cross-presentation signal in LTα-deficient mice could be explained by a deficiency in the number or cytotoxicity of antigen-specific CD8+ T cells in the brain. This remains to be investigated, but since LTα-deficient mice lack peripheral lymph nodes and have spleens with disrupted microarchitecture and severely reduced numbers of dendritic cells [51,52], induction of the T cell response is likely to be affected.
By studying primary cultures of MBECs, we have determined that endothelial cross-presentation of PbA antigen occurs by the phagosome-to-cytosol route and requires both TAP and proteasomes. This is not surprising as aortic endothelial cells have been reported to cross-present an unrelated antigen by the same mechanism [53]. Elucidation of the cross-presentation mechanism may suggest targets for adjuvant therapy of cerebral malaria by reducing the vulnerability of brain endothelial cells to CD8+ T cell-mediated cytolysis. Since the in vitro experiments were largely performed with freeze-thawed iRBCs, and no cross-presentation was observed with a Transwell format, we can formally conclude that the antigen presentation was true cross-presentation: the MBECs were not invaded by parasites, nor were the peptide epitopes already processed in the parasite rather than the MBECs. As both platelets [54] and microparticles [55,56] have been reported to contain parasite-derived proteins during malaria infection, we asked if either could be alternative sources of antigen cross-presented by brain endothelial cells. Neither platelets nor microparticles purified from infected mouse blood induced MBECs to cross-present Pb1 in vitro (S3C Fig); however, we cannot rule out that platelets and microparticles may contain other antigens that can be cross-presented.
Unexpectedly, our results suggest that MBECs phagocytosed free merozoites and digestive vacuoles rather than whole iRBCs in vitro. Merozoites and digestive vacuoles may be released from schizonts by membrane rupture after freeze-thaw, or in the case of freshly isolated mature stages, by continued schizont maturation and rupture or degradation in vitro. Purified merozoites were more efficient at inducing cross-presentation of Pb1 than the same number of mature iRBCs (even though each mature schizont contains 12–18 merozoites), implying that free merozoites are likely to be the major source of cross-presented antigen. While Pb1 originates from a merozoite-expressed protein (GAP50), it is likely that digestive vacuoles also contribute other antigens, e.g. Pb2-containing bergheilysin [21], for cross-presentation as they are avidly phagocytosed. It remains to be seen whether merozoites rather than iRBCs are favored for phagocytosis by endothelial cells in vivo, in mice and humans. While we have not observed any unambiguous instances of MBECs phagocytosing whole PbA iRBCs in vitro, it has been reported that Pf iRBCs first transfer membrane material to and are then engulfed by a human brain endothelial cell line [32]. In our hands, the same cell line avidly phagocytosed Pf merozoites, suggesting that our findings with PbA may have relevance to human disease. Our observation that PbNK65 and Py17X iRBCs were less efficiently internalized and cross-presented than PbA suggests that merozoite phagocytosis by MBECs may be at least partially ligand-specific. It is unlikely that opsonins were responsible for differential phagocytosis of the parasite strains as all three were washed and cultured overnight in medium lacking mouse serum. Much effort has been dedicated to studying the interactions between schizonts and endothelial cells, but we propose that research into merozoite-endothelial interactions, particularly defining any endothelial receptors mediating phagocytosis, could open up new avenues for adjunct therapy of cerebral malaria.
All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC #110630) and complied with the guidelines of the Agri-Food and Veterinary Authority (AVA) and the National Advisory Committee for Laboratory Animal Research (NACLAR).
C57BL/6J mice (6–8 weeks old) were used for infection experiments, sex-matched to knockout mice in the same experiment. C57BL/6J mice up to 12 weeks old were used for MBEC culture. C57BL/6J mice deficient for IFNγ, TNFα, LTα and TAP1 originated from the Jackson Laboratory. All mice were bred and housed under specific pathogen-free conditions in the Biomedical Resource Centre, Singapore.
Mice were infected with P. berghei ANKA clone 15Cy1 (PbA) by intraperitoneal injection of 0.5–1 × 106 iRBCs. Details of this and other rodent strains were previously published [19]. To obtain mature stages, infected blood was diluted in PBS, passed through a Plasmodipur filter (EuroProxima) to remove leukocytes, centrifuged and resuspended at 2–4% hematocrit in RPMI medium containing 20% FCS and penicillin/streptomycin (P/S). The blood was cultured overnight at 37°C on a rocking platform in non-vented flasks flushed with gas mixture (5% CO2, 5% O2, 90% N2). Schizonts and late trophozoites were isolated from the interface formed by gradient centrifugation (1450 × g, 10 min) over 65% Percoll, washed, aliquoted and frozen in parasite medium at -80°C.
For most experiments, the cloned TCR-transduced cell line LR-BSL8.4a was used to detect presentation of the Pb1 epitope SQLLNAKYL in the context of H-2Db, with TCR ligation inducing lacZ expression [19]. We recently generated two other reporter cell lines, LR-BSL13.6b and LR-WH3.4, which recognize the Pb2 and F4 epitopes [21]. All reporter cells were cultured in RPMI complete medium supplemented with 10% FBS, 1 mM sodium pyruvate, 50 μM β-mercaptoethanol, P/S and Primocin (Invivogen).
We described this protocol in detail previously [19]. In brief, brain microvessel fragments were isolated from terminally exsanguinated mice by dextran gradient centrifugation and capture on a cell strainer. These fragments were digested with collagenase prior to overnight co-incubation with LR-BSL8.4a reporter cells. Cross-presentation was measured by counting blue spots following X-gal staining.
Antibodies against mouse CD45 (clone 30-F11, FITC), CD31 (clone 390, PE-Cy7) and CD11b (clone M1/70, PerCP-Cy5.5) were purchased from BD Biosciences, Biolegend, and eBioscience respectively and were used at 1 μg/ml. αCD140b-APC (Miltenyi) was used at 1:10 dilution. Rabbit polyclonal antibodies against NG2 chondroitin sulfate proteoglycan and von Willibrand Factor were purchased from Millipore and used at 2–5 μg/ml for flow cytometry and immunocytofluorescence. Unconjugated rat antibodies against CD8α (clone YTS169.4) and CD8β (clone 53–5.8) from Bio X Cell were used at 5 μg/ml for immunofluorescence. P. yoelii PyMSP-1/19 Rabbit Antiserum (cross-reactive with PbA MSP-1), MRA-23, deposited by JH Adams (University of South Florida, FL, USA), was obtained through the MR4 as part of the BEI Resources Repository, NIAID, NIH (Bethesda, MD, USA). Goat anti-rabbit secondary antibodies conjugated with Alexa Fluor 488 and Alexa Fluor 647 and goat anti-rat antibody conjugated with Alexa Fluor 555 were from Life Technologies and were used at 1:1000–1:5000 dilution.
A papain-based Neural Tissue Dissociation Kit (Miltenyi) was used in conjunction with a gentleMACS Octo Dissociator with Heater to obtain unicellular suspensions from brains of terminally exsanguinated mice. The papain volume was reduced to 15 μL per brain to reduce cleavage of MHC class I molecules and CD31. The resulting homogenate was diluted in DMEM complete medium (high glucose, with pyruvate, supplemented with 10% FBS and P/S) and passed through a 100 μm cell strainer, then centrifuged at 1900 × g for 10 min over a 30% Percoll gradient to remove myelin. The pelleted cells were washed and stained for CD45, CD31 and CD140b in 50 μL, and then red blood cells were lysed by dilution with 200 μL of ACK buffer. Washed cells were then sorted on a FACSAria II (BD Biosciences), after gating on DAPI-excluding singlet cells, into 4 populations: CD45+, CD45-CD31+CD140b-, CD45-CD31-CD140b+ and triple-negative. To isolate microglia and peripheral leukocytes from brains, papain was avoided and each brain was manually mashed and digested in 10 mL PBS with 5 mg collagenase 4 and 0.1 mg DNaseI for 30 min at room temperature. Myelin removal and staining were performed similarly, except that CD45 and CD11b antibodies were used to sort CD45intCD11b+ microglia and all CD45hi leukocytes. All the cells of each gated population per brain were pelleted, resuspended in 100 μL RPMI complete medium and seeded in one well of a 96-well round bottom plate. LR-BSL8.4a cells (3 × 104 cells in 100 μL medium) were added to each well and the plate was incubated overnight (for collagenase-digested cells) or 24 h (for papain-digested cells) before the cells were transferred to a 96-well filter plate for X-gal staining.
Brain microvessels were isolated from naïve C57BL/6J mice as for the ex vivo cross-presentation assay, but collagenase/DNaseI digestion was lengthened to 3 h at 37°C. Digested vessels were triturated with a 10 mL serological pipette and washed twice with DMEM complete medium. Culture conditions were adapted from Lu et al. [57]. The vessel fragments were resuspended in endothelial medium (DMEM complete with added MEM non-essential amino acids, 0.1 mg/ml heparin and 0.1 mg/ml endothelial cell growth supplement from Corning) with 4 μg/ml puromycin and seeded in a collagen-coated 48-well plate. Typically, 5 brains were used to seed 32 wells. The medium was replaced (without puromycin) in the morning 3 days later and every 3–4 days thereafter.
MBECs were stimulated with 10 ng/ml recombinant mouse IFNγ (R&D Systems) 24 h before parasites (3 × 106 freeze-thawed PbA mature iRBCs unless otherwise stated) were added. After a further 24 h, the MBECs were washed and 6 × 104 reporter cells in 0.4 ml RPMI complete medium were added. Following overnight co-incubation, the reporter cells were resuspended and moved to a 96-well filter plate for X-gal staining.
For immunofluorescence and live fluorescence imaging, MBECs were cultured in 8-well Ibidi tissue culture μ-slides that we coated with collagen I. For immunofluorescence assays, cells were fixed with 2% formaldehyde in PBS for 20 min, permeabilized with 0.1–0.5% Triton X-100 in PBS for 10 min, and blocked with 10% goat serum for 30–60 min prior to staining. The buffer used for washing and staining was PBS with 0.5% BSA, and 0.05% Tween 20 was also added to reduce non-specific staining with MSP-1 antiserum. Slides were stained overnight at 4°C with the primary antibody, stained for 1 h at room temperature with the secondary antibody, counterstained with DAPI and mounted in FluorSave (EMD Millipore). For live fluorescence imaging, freshly isolated PbA mature iRBCs were repeatedly forced through a 29G needle to yield a mixture of late trophozoites, early schizonts and merozoites. The parasite material was labeled with 2 μM PKH26 (Sigma-Aldrich) according to the manufacturer’s protocol, and added to IFNγ-stimulated MBECs for 24 h. The cells were then washed thrice before labeling with 50 nM LysoTracker Green DND-26 (Life Technologies) and 8 μM Hoechst 33342 for 5 min at 37°C. After washing, the cells were covered with medium containing 0.5 mg/ml trypan blue to quench extracellular PKH26 fluorescence [29] and imaged immediately. Images were acquired on an Olympus FV1000 confocal microscope with a 100x objective lens; ImageJ was used for overlaying and cropping only unless otherwise specified.
Confluent MBECs in 9 wells of a 48-well plate were stimulated with 10 ng/ml IFNγ, and 3 × 106 thawed PbA mature iRBCs were added to 6 wells 24 h later. The next day, CD8+ T cells were isolated by negative magnetic selection (Miltenyi) from the spleens of a naïve C57BL/6J mouse and one infected with PbA 6 days previously. The MBEC wells were washed and aspirated 24 h after parasite addition, and 106 CD8+ T cells were added in RPMI complete medium containing 50 U/ml IL-2. The wells were gently washed and imaged (10× objective, DIC) after 20 h co-incubation.
Olfactory bulbs were dissected from terminally exsanguinated mice and each lobe was smeared between two glass slides. After air-drying, the bottom slide was fixed in acetone at -20°C for 5–10 min. Blocking, incubation with primary antibodies (against von Willebrand factor, CD8α and CD8β) and secondary antibodies (anti-rabbit Alexa Fluor 488 and anti-rat Alexa Fluor 555) was performed as described above.
PbA mature stages (from 5 infected mice) isolated by Percoll gradient centrifugation were resuspended in 3 ml of parasite medium and passed through a 1.2 μm Acrodisc syringe filter (Pall) to release the merozoites [58]. The filtrate was passed twice through an LS magnetic column (Miltenyi) to deplete hemozoin-containing digestive vacuoles [30]. Merozoites in the flowthrough were concentrated by centrifugation at 4000 × g for 10 min, after which a small aliquot was stained with 8 μM Hoechst dye and 5 μg/ml dihydroethidium for quantification on a MACSQuant Analyzer (Miltenyi).
The immortalized human brain endothelial cell line hCMEC/D3 [59], a kind gift of Pierre-Olivier Couraud (Institut Cochin, Paris, France), was cultured in EGM-2 MV medium (Lonza) on collagen-coated plastic. Pf (clone 3D7) was cultured in RPMI-Albumax II as previously described [60]. A schizont stage 3D7 culture was enriched with an LD magnetic column (Miltenyi), treated with 10 μm epoxysuccinyl-L-leucylamido(4-guanidino) butane (E-64) for 6 h, then passed through a 1.2 μm Acrodisc syringe filter [58]. The released merozoites were labeled with 2 μm PKH26 and added to hCMEC/D3 cells that had been stimulated with 50 ng/ml recombinant human IFNγ (R&D Systems) for 20 h prior. After 24 h, the cells were washed, stained with LysoTracker and Hoechst, and imaged as above for MBECs.
Blue spots counts were analyzed by unpaired t-test (2 groups), 1-way ANOVA (3 or more groups with one independent variable) or 2-way ANOVA (2 independent variables) after logarithmic transformation to achieve homoscedasticity and normal distribution. Bonferroni’s post test was used to compare groups following ANOVA. The exceptions were the cross-presentation experiments using sorted cell populations (Mann-Whitney U test) and quantification of parasite uptake (Kruskal-Wallis test) where non-parametric tests were employed as the data were not normally distributed. Error bars represent standard deviations unless otherwise stated.
|
10.1371/journal.pgen.1004123 | Identification of Novel Genetic Loci Associated with Thyroid Peroxidase Antibodies and Clinical Thyroid Disease | Autoimmune thyroid diseases (AITD) are common, affecting 2-5% of the general population. Individuals with positive thyroid peroxidase antibodies (TPOAbs) have an increased risk of autoimmune hypothyroidism (Hashimoto's thyroiditis), as well as autoimmune hyperthyroidism (Graves' disease). As the possible causative genes of TPOAbs and AITD remain largely unknown, we performed GWAS meta-analyses in 18,297 individuals for TPOAb-positivity (1769 TPOAb-positives and 16,528 TPOAb-negatives) and in 12,353 individuals for TPOAb serum levels, with replication in 8,990 individuals. Significant associations (P<5×10−8) were detected at TPO-rs11675434, ATXN2-rs653178, and BACH2-rs10944479 for TPOAb-positivity, and at TPO-rs11675434, MAGI3-rs1230666, and KALRN-rs2010099 for TPOAb levels. Individual and combined effects (genetic risk scores) of these variants on (subclinical) hypo- and hyperthyroidism, goiter and thyroid cancer were studied. Individuals with a high genetic risk score had, besides an increased risk of TPOAb-positivity (OR: 2.18, 95% CI 1.68–2.81, P = 8.1×10−8), a higher risk of increased thyroid-stimulating hormone levels (OR: 1.51, 95% CI 1.26–1.82, P = 2.9×10−6), as well as a decreased risk of goiter (OR: 0.77, 95% CI 0.66–0.89, P = 6.5×10−4). The MAGI3 and BACH2 variants were associated with an increased risk of hyperthyroidism, which was replicated in an independent cohort of patients with Graves' disease (OR: 1.37, 95% CI 1.22–1.54, P = 1.2×10−7 and OR: 1.25, 95% CI 1.12–1.39, P = 6.2×10−5). The MAGI3 variant was also associated with an increased risk of hypothyroidism (OR: 1.57, 95% CI 1.18–2.10, P = 1.9×10−3). This first GWAS meta-analysis for TPOAbs identified five newly associated loci, three of which were also associated with clinical thyroid disease. With these markers we identified a large subgroup in the general population with a substantially increased risk of TPOAbs. The results provide insight into why individuals with thyroid autoimmunity do or do not eventually develop thyroid disease, and these markers may therefore predict which TPOAb-positives are particularly at risk of developing clinical thyroid dysfunction.
| Individuals with thyroid peroxidase antibodies (TPOAbs) have an increased risk of autoimmune thyroid diseases (AITD), which are common in the general population and associated with increased cardiovascular, metabolic and psychiatric morbidity and mortality. As the causative genes of TPOAbs and AITD remain largely unknown, we performed a genome-wide scan for TPOAbs in 18,297 individuals, with replication in 8,990 individuals. Significant associations were detected with variants at TPO, ATXN2, BACH2, MAGI3, and KALRN. Individuals carrying multiple risk variants also had a higher risk of increased thyroid-stimulating hormone levels (including subclinical and overt hypothyroidism), and a decreased risk of goiter. The MAGI3 and BACH2 variants were associated with an increased risk of hyperthyroidism, and the MAGI3 variant was also associated with an increased risk of hypothyroidism. This first genome-wide scan for TPOAbs identified five newly associated loci, three of which were also associated with clinical thyroid disease. With these markers we identified a large subgroup in the general population with a substantially increased risk of TPOAbs. These results provide insight into why individuals with thyroid autoimmunity do or do not eventually develop thyroid disease, and these markers may therefore predict which individuals are particularly at risk of developing clinical thyroid dysfunction.
| Autoimmune thyroid disease (AITD), including Hashimoto's thyroiditis and Graves' disease, is one of the most common autoimmune diseases, affecting 2–5% of the general population [1], [2], [3]. Thyroid dysfunction has been associated with osteoporosis, depression, atrial fibrillation, heart failure, metabolic syndrome, and mortality [4], [5], [6], [7], [8], [9], [10], [11]. High serum antibodies against the enzyme thyroid peroxidase (TPO), which is located in the thyroid and plays a key role in thyroid hormone synthesis, are present in 90% of patients with Hashimoto's thyroiditis [12], [13], the most frequent cause of hypothyroidism and goiter. Although TPO antibodies (TPOAbs) are a useful clinical marker for the detection of early AITD, it remains controversial if these antibodies play a causative role in the pathogenesis of Hashimoto's thyroiditis [14], [15], [16].
Interestingly, TPOAb-positive persons also have an increased risk of developing autoimmune hyperthyroidism (Graves' disease) [17], [18], which is caused by stimulating antibodies against the thyroid stimulating hormone (TSH) receptor [19]. Numerous studies have shown that Graves' hyperthyroidism and Hashimoto's thyroiditis show co-inheritance [17], [20], [21]. Finally, thyroid autoimmunity is the most common autoimmune disorder in women of childbearing age, and TPOAb-positive women have an increased risk of developing pregnancy complications such as miscarriage and pre-term delivery [17], [18], [22], [23], [24], [25], [26].
The prevalence of TPOAb-positivity in the general population ranges from 5–24%, but it is currently unknown why these people develop TPOAbs, nor is it known why not all individuals with thyroid autoimmunity develop clinical thyroid disease [27], [28]. It is estimated that around 70% of the susceptibility to develop thyroid autoantibodies is due to genetic factors [29]. In this context it is remarkable to note that little is known about the genetic factors that determine TPOAb-positivity and the risk of AITD.
We therefore performed a genome wide association study (GWAS) meta-analysis for TPOAbs in the general population in 18,297 individuals from 11 populations. Newly identified genetic variants were studied in relation to subclinical and overt hypo- and hyperthyroidism, goiter, thyroid autoimmunity during pregnancy and thyroid cancer risk.
Characteristics of the studied populations are shown in Table 1 and the Supplementary Material S1. Heritability estimates in the family-based cohorts SardiNIA, TwinsUK and Val Borbera were, respectively, 0.65, 0.66, and 0.54 for TPOAb-positivity, and 0.43, 0.66, and 0.30 for TPOAb levels.
See Table 1 and Supplementary Figure S1 for TPOAb measurements and Supplementary Table S1 for genotyping procedures. In most autoimmune diseases, both the presence and the level of autoantibodies are relevant for the disease onset [18], [30], [31]. Furthermore, different pathophysiological processes may be involved in the initiation and severity of the autoimmune response. We therefore performed a GWAS on TPOAb-positivity (including 1769 TPOAb-positives and 16,528 TPOAb–negatives), as well as a GWAS on continuous TPOAb levels (including 12,353 individuals) in stage 1. See Supplementary Figures S2 and S3 for QQ (quantile-quantile) and Manhattan plots.
In stage 2, we followed-up 20 stage 1 SNPs (P<5×10−6; 13 TPOAb-positivity and 10 TPOAb level SNPs, with 3 SNPs overlapping) in 5 populations, including up to 8,990 individuals for TPOAb-positivity (922 TPOAb-positives and 8068 TPOAb–negatives) and 8,159 individuals for TPOAb level analyses (see Supplementary Material S1). Results of the combined stage 1 and 2 meta-analyses, including heterogeneity analyses, are shown in Supplementary Tables S2 and S3. Regional association plots are shown in Supplementary Figures S4 and S5. In the combined stage 1 and 2 meta-analyses GWAS significant associations (P<5×10−8) were observed near TPO (Chr 2p25; rs11675434), at ATXN2 (Chr 12q24.1; rs653178), and BACH2 (Chr 6q15; rs10944479) for TPOAb-positivity, and near TPO (rs11675434), at MAGI3 (Chr 6q15; rs1230666), and KALRN (Chr 3q21; rs2010099) for TPOAb levels (Table 2 and Figure 1). The TPOAb level meta-analysis P-values for the 3 GWAS significant TPOAb-positivity loci were: TPO-rs11675434: P = 7.4×10−13, ATXN2-rs653178: P = 1.3×10−7, and BACH2-rs10944479: P = 2.0×10−4.
As the 3 GWAS significant loci for TPOAb levels also showed associations with TPOAb-positivity (TPO-rs11675434: OR, 1.21 [95% CI, 1.15–1.28)], P = 1.5×10−16; MAGI3-rs1230666: OR, 1.23 [95% CI, 1.14–1.33], P = 1.5×10−6; KALRN-rs2010099: OR, 1.24 [95% CI, 1.12–1.37], P = 7.4×10−5), we subsequently studied the (combined) effects of these 5 SNPs on clinical thyroid disease. Genetic risk scores were calculated as described in the Supplementary Material. The variance explained by these 5 SNPs was 3.1% for TPOAb-positivity and 3.2% for TPOAb levels. Subjects with a high genetic risk score had a 2.2 times increased risk of TPOAb-positivity compared to subjects with a low genetic risk score (P = 8.1×10−8) (Table 3).
Table S4 shows the stage 1 TPOAb-positivity and TPOAb level meta-analyses results for GWAS significant SNPs reported in previous GWAS on thyroid related phenotypes.
The associations between the 5 GWAS significant SNPs and the risk of abnormal thyroid function tests are shown in Table 4. MAGI3- rs1230666 was associated with an increased risk of overt hypothyroidism and increased TSH levels below the Bonferroni threshold (i.e., P = 0.05/5 = 0.01). Borderline significant signals were observed at BACH2- rs10944479 with a higher risk of increased TSH levels as well as overt hyperthyroidism (P = 0.011 and P = 0.012), and at the KALRN-rs2010099 SNP with a lower risk of decreased TSH levels (P = 0.010).
Furthermore, a higher genetic risk score was associated with a higher risk of increased TSH levels (Supplementary Table S5). No effects of the genetic risk score on the risk of overt hypothyroidism, hyperthyroidism or decreased TSH levels were observed.
Individuals with a high genetic risk score had a 30.4% risk of sonographically-proven goiter, compared to 35.2% in subjects with a low score (P = 6.5×10−4) (Table 5). None of the individual SNPs was significantly associated with goiter risk.
As autoimmunity significantly changes during pregnancy [25], we additionally studied these effects in an independent pregnant population. Pregnant women with a high genetic risk score had a 2.4 times increased risk of TPOAb-positivity compared to women with a low score (10.3% vs 4.8%, P = 0.03). These women did not have a higher risk of increased TSH levels. However, a borderline significant signal with a lower risk of increased TSH levels was observed at ATXN2- rs653178 (OR, 0.54 [95% CI, 0.34–0.87], P = 0.012).
Ingenuity Pathway Analyses (IPA; Ingenuity Systems, Ca, USA) and GRAIL analyses [32] were performed to identify potential pathways involved in AITD, the results of which are shown in Supplementary Tables S7 and S8, and Figure S6. The identified top pathways involved cell death, survival, movement, and OX40 signalling.
This is the first GWAS meta-analysis investigating the genetics of TPOAbs in the normal population in up to 18,297 individuals from 11 populations with replication in up to 8,990 individuals from 5 populations. We identified 5 GWAS significant loci associated with TPOAb-positivity and/or levels.
The most significant hit for both TPOAb-positivity and TPOAb levels was located near the TPO gene itself. TPO is a membrane-bound protein located on the apical membranes of the thyroid follicular cell, catalyzing key reactions in thyroid hormone synthesis [33]. Mutations in TPO have been found in patients with congenital hypothyroidism [34], [35]. Although TPOAbs are valid clinical biomarkers of AITD, they are generally considered to be secondary to the thyroid damage inflicted by T-cells.
The FOXE1 gene has been previously associated with hypothyroidism [36], [37] and is known to regulate transcription of TPO [38]. In this context it is interesting to note that we did not find any associations of the variant near TPO with hypothyroidism. Most genes that have been associated with AITD (predominantly Graves' disease) by candidate gene and GWAS studies so far are located in the HLA class I and II regions, or in genes involved in T-cell (i.e., CTLA-4, PTPN22) or other autoimmune responses [28], [39]. Until now, the TPO gene itself had not been associated with AITD, except in one recent candidate gene analysis in a small cohort (n = 188) without replication [40]. A variant near TPO (rs11694732), which is in LD with rs11675434 (r2 = 0.97 in HapMap2), has previously been associated with TSH levels by Gudmundsson et al [41]. However, various other GWAS on serum TSH and FT4 levels have not found any significant associations in or near this locus, including a recent similar sized GWAS by Porcu et al [42].
Three of the other four loci identified here are located in or are in linkage disequilibrium (LD) with genes previously associated with other autoimmune diseases. Rs1230666 is located in intron 9 of MAGI3, encoding a protein that modulates activity of AKT/PKB. AKT/PKB is expressed in the thyroid and regulates apoptosis [43], which seems to play an important role in the development of AITD [44], [45]. In addition, rs1230666 is in LD with rs2476601 (r2 = 0.70 in HapMap2), a variant causing a R620W substitution in PTPN22. PTPN22 is a lymphoid-specific intracellular phosphatase involved in the T-cell receptor signaling pathway. Variations in PTPN22, and specifically R620W, are associated with various autoimmune disorders including type 1 diabetes, rheumatoid arthritis, systemic lupus erythematosus and Graves' disease [46], [47], [48], [49]. The associations of the MAGI3 locus with TPOAb-positivity and Graves' disease may therefore also be explained by linkage with disease-associated variants in PTPN22 [50]. Of note, the association signal at rs2476601 is one order weaker than that of the top variant rs1230666.
The BACH2 locus has been implicated in the susceptibility to several autoimmune diseases, including celiac disease, type 1 diabetes, vitiligo, Crohn's disease, and multiple sclerosis [46], [51], [52], [53], [54]. A recent candidate gene analysis associated the BACH2 locus with an increased risk of AITD, including Hashimoto's thyroiditis and Graves' disease [55]. However, the associations were not significant when Hashimoto's thyroiditis and Graves' disease were studied separately. BACH2 is specifically expressed in early stages of B-cell differentiation and represses different immunoglobulin genes [56]. Interestingly, BACH2 can bind to the co-repressor SMRT (silencing mediator of retinoid and thyroid receptor), which may suggest a more direct effect on thyroid hormone secretion and action as well.
Polymorphisms in ATXN2 have been associated with multiple neurodegenerative diseases, including spinocerebellar ataxia and Parkinson's disease [57], [58], [59]. Different epidemiological studies have associated thyroid dysfunction with cerebellar ataxia [60], [61]. Furthermore, the identified SNP in ATXN2 has been previously associated with renal function, serum urate levels and blood pressure [62], [63], [64]. However, this SNP is in high LD with rs3184504 (r2 = 0.873), a variant causing a Trp262Arg substitution of SH2B adaptor protein 3 (SH2B3). SH2B3 encodes the adaptor protein LNK, a key negative regulator of cytokine signaling playing a critical role in hematopoiesis. This variant is associated with susceptibility to several autoimmune diseases, including celiac disease, type 1 diabetes, vitiligo, and rheumatoid arthritis [46], [51], [53], [65], suggesting more relevance for TPOAb levels than ATXN2. This is supported by a recent study which showed that variants in LD with SH2B3, BACH2, and PTPN22 are associated with TPOAb levels in patients with type 1 diabetes [66].
Whereas the above four loci are located in genes involved in the immune response or the autoantigen, the KALRN (Kalirin) gene encodes a multi-domain guanine nucleotide exchange factor for GTP-binding proteins of the Rho family. The relation of KALRN with levels of TPOAbs is unclear. This gene has recently been found to be associated with megakaryopoiesis and platelet formation [67], which may suggest a function in the immune system [68]. We furthermore performed pathway analyses on the stage 1 TPOAb-positivity and TPOAb level lead SNPs, and identified the cell death, survival and movement pathway as an important pathway for TPOAbs. This finding is supported by previous studies, which show an important role for apoptosis in the development of AITD [44], [45]. Another top pathway involved was the OX40 signalling pathway, and it is of interest to note that OX40 is a T-cell activator promoting the survival of CD4+ T-cells at sites of inflammation [69].
Our results have potential clinical relevance for several reasons. Genetic risk scores based on these novel common (risk allele frequencies: 9–40%) TPOAb-associated SNPs enabled us to identify a large subgroup in the general population with a two-fold increased risk of TPOAb-positivity (10.4% vs 5.4%). These individuals also have a higher risk of increased TSH levels and a lower risk of goiter, suggesting an advanced stage of destruction of the thyroid due to autoimmune processes. Furthermore, pregnant women with high genetic risk scores had a 2.4 times increased risk of TPOAb-positivity during pregnancy. In this context it is interesting to note that TPOAb-positive pregnant women have an increased risk of miscarriages and preterm births independent of thyroid function [70].
Associations with thyroid disease were also found on an individual SNP level. The MAGI3 SNP was associated with a substantially increased risk of hypothyroidism, and the BACH2 SNP showed a borderline significant association (P = 0.011) with a higher risk of increased TSH levels, which includes subjects with subclinical and overt hypothyroidism. Furthermore, both loci were significantly associated with an increased risk of Graves' hyperthyroidism in an independent population. To predict which patients with first or second degree relatives with documented Hashimoto's or Graves' disease will develop clinical thyroid disease, a clinical algorithm has been developed (i.e., the THEA score) [18]. Future studies should analyze if these genetic markers increase the sensitivity of the THEA score. Graves' hyperthyroidism and Hashimoto's thyroiditis co-segregate in families and subjects with TPOAbs have an increased risk of both diseases [17], [18], [20], [21], [22], [26]. The current study provides insight into this phenomenon by showing that specific loci associated with TPOAbs and (subclinical) hypothyroidism, i.e. MAGI3 and BACH2, are also associated with Graves' hyperthyroidism in an independent case-control study.
The prevalence of TPOAb-positivity in the general population is high (5–24%), but it is currently unknown why part of the individuals with thyroid autoimmunity develop clinical thyroid disease whereas others do not [27], [28]. In this context it is interesting to note that the TPOAb-associated SNPs located in TPO and ATXN2 were not associated with clinical thyroid disease. This suggests that the TPOAbs in these individuals may be of less clinical relevance, providing insight into why TPOAb-positive individuals do or do not eventually develop clinical thyroid disease.
Our study has some limitations. The validity of the results is restricted to individuals from populations of European ancestry. Future GWASs in populations from non-European descent will be required to determine to which extent our results can be generalized to other ethnic groups. Secondly, we did not perform conditional analyses to further identify secondary association signals within the identified loci, nor did we perform functional studies for the identified variants. Further research is therefore needed to unravel the exact biological mechanism behind the observed associations. The fact that various TPOAb assays were used across the participating cohorts could lead to bias. We therefore used TPOAb-positivity cut-off values as provided by the respective assay manufacturer, instead of using one fixed cut-off value. This is also of clinical importance as in clinical practice most institutions rely on the TPOAb-positivity cut-off as provided by the assay manufacturer. Furthermore, we did not detect heterogeneity in our results, supporting the fact that results obtained with different assays can be combined across cohorts using the z-score based meta-analysis. Finally, as AITD coincides with other autoimmune diseases, our results could be driven by indirect associations with other autoimmune diseases. However, AITD is the most common autoimmune disease in the general population. We furthermore show that carriage of multiple risk alleles is associated with an increased risk of thyroid dysfunction, which underlines the clinical importance of our findings.
In conclusion, this first GWAS for TPOAbs identified five newly associated loci, three of which were also associated with clinical thyroid disease. Furthermore, we show that carriage of multiple risk variants is not only associated with a substantial increased risk of TPOAb-positivity, but also with a higher risk of increased TSH levels (including subclinical and overt hypothyroidism) and a lower risk of goiter. These genetic markers not only help to identify large groups in the general population with an increased risk of TPOAb-positivity, but may also predict which TPOAb-positive persons are particularly at risk of developing clinical thyroid disease.
For the TPOAb GWAS stage 1 and 2 analyses, and the hypothyroidism, hyperthyroidism and goiter analyses, individuals were recruited from 16 independent community-based and family studies. For the Graves' disease analyses, cases were recruited from the United Kingdom Graves' disease cohort and controls from the British 1958 Birth Cohort. Thyroid cancer cases and controls were recruited from the Nijmegen and Ohio thyroid cancer cohorts. A detailed description of the original cohorts contributing samples is provided in Table 1 and in the Supplementary Material. All participants provided written informed consent and protocols were approved by the institutional review boards or research ethics committees at the respective institutions, and conducted according to the Declaration of Helsinki.
Serum TPOAb levels were determined with a range of assays. TPOAb-positives were defined as subjects with TPOAb levels above the assay-specific TPOAb-positivity cut-off, as defined by the manufacturer (Table 1). Serum TSH and free thyroxine (FT4) levels were determined using a range of assays (Table 1). Assay-specific TSH and FT4 reference ranges were used, as provided by the manufacturer (Table 1). Overt hypothyroidism was defined as a high TSH (i.e., a TSH level above the TSH reference range) and a low FT4. Increased TSH was defined as a high TSH, including persons with overt hypothyroidism or subclinical hypothyroidism (i.e., high TSH with a normal FT4). Overt hyperthyroidism was defined as a low TSH and a high FT4. Decreased TSH was defined as a low TSH, including persons with subclinical or overt hyperthyroidism.
The diagnosis of goiter is described in the Supplementary Material, and the diagnosis of Graves' disease and thyroid cancer in the respective cohorts have been described previously [41].
Samples were genotyped with a range of GWAS genotyping arrays (Supplementary Table S1). Sample and SNP quality control procedures were undertaken within each study. For each GWAS, over 2.5 million SNPs were imputed using CEU samples from Phase 2 of the International HapMap project (www.hapmap.org). Genotyping procedures in the stage 2, Graves' disease and thyroid cancer populations are described in the Supplementary Material.
The heritabilities of TPOAb-positivity and serum TPOAb levels were estimated, as described in the Supplementary Material.
In stage 1, we performed a GWAS on TPOAb-positivity as well as a GWAS on continuous TPOAb levels. Persons taking thyroid medication were excluded. Each SNP was tested for association with TPOAb-positivity using logistic regression analyses, adjusting for age and sex. For cohorts with family structure, we approximated the probability of being affected with a linear mixed model adjusting for age and sex. The produced model was used to predict the expected proportion of “risk” (effective) alleles in cases and controls, hence giving the means to estimate odds ratios. Only unrelated individuals were considered for the SardiNIA cohort. For the GWAS of continuous TPOAb levels, samples with a TPOAb level lower than the minimum TPOAb assay detection limit (Table 1) were excluded. TPOAb levels were natural log-transformed, and sex-specific, age adjusted standardized residuals were calculated. Each SNP was tested for association with these TPOAb level residuals using linear regression analyses (additive model), correcting for relatedness in studies with family structure. See Supplementary Table S1 for the software used for these analyses.
Before meta-analysis, SNPs with a minor allele frequency (MAF) <1% or a low imputation quality were excluded (Supplementary Material), after which the results of each GWAS were combined in a population size weighted z-score based meta-analysis using METAL [71]. Genomic control was applied to individual studies if λ>1.0.
In stage 2, we followed-up stage 1 GWAS significant SNPs, as well as promising SNPs not reaching GWAS significance, in an attempt to reach GWAS significant associations by increasing sample size (Supplementary Material). Results from stage 1 and 2 were combined in a population size weighted z-score based meta-analysis using METAL [71]. A z-score based meta-analysis was used to reduce bias that might be induced by different assays. As this method does not provide betas, and we wanted to provide a rough estimate of the actual effect sizes for convenience, we calculated betas using the fixed effects (inverse variance based) meta-analysis method. Heterogeneity was tested, applying bonferroni based P-value thresholds of P = 0.004 for the TPOAb-positivity analyses and P = 0.005 for the TPOAb level analyses.
All studies assessed and, if present, corrected for population stratification using principal-component analysis (PCA) and/or multidimensional-scaling (MDS), with the exception of SardiNIA and ValBorbera where the high isolation substantiates a lack of stratification (Table S1) [72], [73]. Lambda values were all ∼1, indicating that population stratification was overall properly accounted for (Table S1). To fully remove residual effects, we applied genomic correction to studies were lambda was >1. The final meta-analyses reported a lambda of 1.01 for both the TPOAb-positivity and the TPOAb level GWAS, thus no genomic correction was applied.
The variances explained by the GWAS significant SNPs were calculated. We subsequently studied the individual as well as the combined effects of the GWAS significant SNPs on the risk of clinical thyroid disease, as specified in the Supplementary Material. In short, to study combined effects, a genetic risk score was calculated for every person as the weighted sum of TPOAb risk alleles. The associations between the individual SNPs, genetic risk scores and the risk of abnormal thyroid function tests were studied using logistic regression analyses. Logistic regression analyses were used to study the associations with goiter, Graves' disease and thyroid cancer (Supplementary Material). The results of each study were combined in a population size weighted z-score based meta-analysis using METAL [71].
Various bioinformatic tools were searched for evidence for functional relevance of the GWAS significant SNPs and pathway analyses were performed on the Stage 1 lead SNPs (see Supplementary Material).
|
10.1371/journal.pcbi.1000628 | Order and Stochastic Dynamics in Drosophila Planar Cell Polarity | Cells in the wing blade of Drosophila melanogaster exhibit an in-plane polarization causing distal orientation of hairs. Establishment of the Planar Cell Polarity (PCP) involves intercellular interactions as well as a global orienting signal. Many of the genetic and molecular components underlying this process have been experimentally identified and a recently advanced system-level model has suggested that the observed mutant phenotypes can be understood in terms of intercellular interactions involving asymmetric localization of membrane bound proteins. Among key open questions in understanding the emergence of ordered polarization is the effect of stochasticity and the role of the global orienting signal. These issues relate closely to our understanding of ferromagnetism in physical systems. Here we pursue this analogy to understand the emergence of PCP order. To this end we develop a semi-phenomenological representation of the underlying molecular processes and define a “phase diagram” of the model which provides a global view of the dependence of the phenotype on parameters. We show that the dynamics of PCP has two regimes: rapid growth in the amplitude of local polarization followed by a slower process of alignment which progresses from small to large scales. We discuss the response of the tissue to various types of orienting signals and show that global PCP order can be achieved with a weak orienting signal provided that it acts during the early phase of the process. Finally we define and discuss some of the experimental predictions of the model.
| Epithelial tissues are often polarized in a preferred direction which determines, for example, the direction of hair growth on mammalian skin, the orientation of scales in fish, the alignment of ommatidia in the fly eye and of sensory hair cells in the vertebrate cochlea. This in-plane polarization, known as planar cell polarity, is one of the morphogenetic fields that play a role in tissue patterning during development. Here we focus on planar cell polarity in the fly wing, where protein localization and inter-cellular ligand-receptor interactions combine with an unknown orienting signal to establish planar cell polarity of the wing epithelium. We demonstrate an analogy between this process and models of ferromagnetism in physical systems that have been studied extensively using the tools of statistical mechanics. The analogy helps in understanding how local interactions between cells can lead to global polarization order and elucidate the role of global orienting signals and the dependence of the dynamics of the process on parameters. We demonstrate that in the absence of an external orienting signal swirling patterns should emerge due to random noise. We propose ways to test this prediction and ways to quantify the magnitude and spatial variation of the unknown external orienting signal.
| Epithelia in diverse tissues, in addition to their apico-basal polarization, acquire a polarization within the two-dimensional layer of cells – a phenomenon called planar cell polarity (PCP) [1]–[5]. In the developing wing of Drosophila, PCP determines the growth direction of small hairs that extend radially from cell boundaries. In a wild-type wing, where cells are approximately hexagonal and form a regular honeycomb lattice, all of these hairs point to the distal direction.
A series of recent experiments show that several key proteins [6], including the transmembrane proteins Frizzled (Fz) and Van-Gogh (Vang) and the cytosolic proteins Dishevelled (Dsh) and Prickled (Pk), localize asymmetrically on cell boundaries [7]–[12] - defining a direction in the plane within each cell and forming a characteristic zig-zag pattern of protein localization on the lattice (Fig. 1A).
Other experiments show that local PCP orientation depends on inter-cellular signaling. First, mutant clones in which fz or Vang activity is suppressed or amplified, cause characteristic and reproducible inversion of polarity in large patches of cells that are proximal or distal to the clone [13]. These observations are summarized in Figs. 1 C,D. Second, in fat mutant clones [14],[15] hairs do not all point correctly in the distal direction, yet, their orientation is strongly correlated between nearby cells and varies gradually across the tissue creating a characteristic swirling pattern.
Thus the experimental evidence suggests that an interaction between neighboring cells tends to locally align their polarity [1],[3],[14]. This local polarity need not point distally unless, in addition, there is a global orienting signal that picks out the distal direction throughout the wing (most likely originating with the Dpp morphogen gradient which defines the Anterior-Posterior axis of the wing in the larval stage of development [16]). Yet, aside from a clear involvement of protocadherin fat [17],[18] the molecular details of this pathway remains for now unknown. The swirling patterns in fat mutants [14] and recent evidence [15],[19], suggest that the orienting field is related to the presence of a “gradient” in the fat, four-jointed, and dachs pathway.
These observations evoke an analogy between PCP and the behavior of ferromagnets, extensively studied in physics and well understood in terms of statistical mechanics of relatively simple models [20]. In these models each atomic site is assigned a magnetic dipole – spin – which can assume a different orientation (analogous to the direction of polarization in an epithelial cell). The salient properties of ferromagnets arise from the opposing influence of an interaction between neighboring spins, which tends to co-align their orientation, and the influence of thermal fluctuations, which tend to randomize the spin direction. Ferromagnets typically exhibit two phases of behavior: a high temperature phase, where spins are disordered and a low temperature ferromagnetic phase, where the interactions dominate over thermal fluctuations – leading to a spontaneous polarization in an arbitrary direction. In this state even a small external magnetic field has a big effect on magnetic polarization as the spontaneous polarization aligns itself with the external field, yet the dynamics leading to global alignment can be quite slow.
An essential lesson from statistical mechanics is that the ordered and disordered states exist in a broad class of models and can be discussed in a general context, focusing on a classification of the different regimes as a function of a few parameters. We follow this lesson by focusing the study on the competition between the intercellular interaction and the disordering influence of the fluctuations introduced by the noisy molecular interactions. As in statistical mechanics we define a phase diagram which identifies different regimes of behavior in the space of the most relevant parameters. We then address the role of the global directional signal in the dynamics of global alignment.
A molecular model for PCP formation was recently proposed in Ref. [21], and was shown to reproduce a number of experimental findings. This model involves 38 parameters that were adjusted to successfully reproduce a set of wild-type and mutant phenotypes. Here we pursue an alternative approach and instead of moving on to more and more complex models develop a model with a smaller number of degrees of freedom and a smaller number of parameters. Instead of fixing a particular set of parameters by fitting the data we explore the generic behavior of the model as a function of parameters defining quantitative features characteristic of the different phases. In formulating the model we identify several essential ingredients, required to obtain the characteristic zig-zag pattern and the non-autonomy of fz and Vang mutant clones. We expect our simplified model to capture important properties of PCP, although it does not incorporate all the molecular details.
After discussing the essential ingredients of the model, we obtain a phase diagram describing its steady state properties. We then consider the dynamics of local polarization strength and orientation in the absence and in the presence of a global orienting signal. We show that global alignment can be achieved with a weak global orienting signal provided it is present throughout the tissue at the earliest stage of PCP dynamics. Finally we discuss the experimental predictions coming out of the model and the tools required to test these predictions.
Three essential ingredients are included in the model, to account for the characteristic zig-zag patterns of protein localization and for the non-autonomy of fz and Vang mutant clones.
There are several reasons why the dynamic equations are not deterministic. Even in the steady state, interfacial complexes not only bind and unbind due to thermal fluctuations, but like nearly everything else inside the cell are being constantly recycled and reassembled. Stochastic fluctuations arise from the molecular noise of reactions and the variability in the state of the cell defining the “intrinsic” and “extrinsic” noise [22]. It will suffice however to describe stochasticity of complex binding and unbinding as if it were a Poisson process. Equation (1) is thus replaced by a stochastic equation,(4)[and a similar modification applies to Eq. (2)] where the noise can be approximated as white Gaussian noise if the number of molecules per cell is not too small. Assuming that the dominant contribution comes from the finite number of molecules participating in the binding/unbinding dynamics, the variance of is inversely proportional to (see Methods), where is defined as the number of molecules per interface: where is the total concentration of molecules (bound and unbound) and is the area of an interface (about – see Fig. 1B). Since the variance of decreases with increase of , plays a role similar to temperature in a ferromagnet. If there are Fz molecules per cell [23], is of order resulting in the root-mean-square fluctuations of the order of (i.e. ) of the mean.
Other sources of intrinsic noise, in addition to the stochasticity of binding and unbinding events, may increase the noise variance beyond the above estimate. These additional noise sources include, for example, stochasticity in the signaling pathway that generates the non-local inhibition within each cell, or fluctuations in and . Such sources of intrinsic noise, acting upstream of and , are propagated to the PCP signaling dynamics through the dynamics of complex formation, and can thus be described qualitatively by the noise term in Eq. (4), with an effective value of that is possibly smaller than predicted from the number of and molecules alone.
What are the consequences of the model defined above when cells are arranged on a hexagonal lattice? Let us first consider the steady state in the deterministic limit. Fig. 3A shows a typical phase diagram on a two-dimensional plane dissecting our five dimensional parameter space (see Methods): the axis is the range of the non-local interaction in units of the cell lattice spacing, and the axis the coefficient which controls inhibition (see Methods).
In the region labeled there is a unique steady state in which there is no polarization of the protein distribution. In contrast, in region the stable steady state has the symmetry shown in Fig. 3B: Both and distributions carry a vector dipole moment that points towards the center of a side, and due to the lattice symmetry there are six equivalent states of this type. A uniform steady state exists as well, but it is unstable. Region differs from in the direction of the dipole, which points towards a vertex instead of pointing towards and edge (Fig. 3C).
The transition from the uniform state, , to the edge state, in the phase diagram is continuous: the dipole moment tends to zero when approaching the phase boundary from the side. A similar transition from a state to a state can exist as well, and is present on another two dimensional “slice” through the parameter space of our model. This transition is also continuous.
We next consider the effects of stochasticity, which were ignored in the discussion above by setting . When is finite (similar to a non-vanishing temperature in a spin model), we ask whether the steady state maintains long-range order: i.e. whether a particular orientation is singled out throughout the lattice and the dipole moment has a non-zero average. In the language of the analogy with magnetic systems this would be a ferromagnetic state. The latter disappears as the temperature increases above a certain critical value, giving way to a paramagnetic state where dipole moments point in random directions and the average polarization vanishes (an intermediate state with quasi-long range order may exist as well, in similarity to 2-dimensional clock models [24]–[27]). Hence, we expect an ordered state to be stable only when is sufficiently large, and this is indeed observed in our simulations (Fig. 4). Yet with a realistic number of molecules per cell, in the order of several thousands, the vertex and side states in our model are typically ferromagnetic.
It may thus appear that when takes realistic values the system is in an ordered state and stochasticity is altogether unimportant. However, as we discuss next, the steady state is not necessarily reached within the time scales of wing development, and stochasticity plays an important role in the dynamics of ordering.
Let us consider the dynamics of PCP formation, first in the absence of a global orienting signal. Fig. 5 shows results from a stochastic simulation, starting from a state where and are uniformly distributed in all cells.
We can identify two stages of the process. The first stage corresponds to a gradual build up of a dipolar polarization on the cellular level. The dipole initially points in a random direction, but as its amplitude increases with time (Fig. 5B) local polarization begins to re-orient. At the end of this stage, when amplitude saturates, there is no global choice of PCP direction, but the orientation of nearby cells is strongly correlated: as an example, Fig. 5A shows the configuration of dipoles shortly after saturation. The second stage, which follows amplitude saturation, exhibits slow coarsening dynamics [27]: polarity direction is approximately aligned within discrete domains, the size of which gradually expands by movement of their boundaries. Note also the existence of vortex-like defects [28] (Fig. 5A and Fig. S1). Coarsening ultimately leads to a spatially uniform steady state, but this process occurs over a long time scale compared to that of amplitude growth.
A quantitative theory of the early dynamics is obtained from the linear instability of the uniform steady state (described in detail in Text S1, part II). The variance of the local dipole amplitude increases exponentially in time with a characteristic time scale ,(5)where for simplicity numeric prefactors of order unity are omitted (see Text S1, part II). In this equation is the amplitude of noise in the unstable uniform steady state, and both and are found from the instability analysis (Text S1, part II). This prediction is shown in Fig. 5B (dashed line) for comparison with the simulation.
Two additional insights come from the analysis of early dynamics (Text S1, part II). First, PCP is initially isotropic, despite the discrete 6-fold symmetry of the hexagonal cell lattice. Consequently, the dipole moment initially has equal probability to point in any direction in the interval . Second, the spatial correlation established during the early dynamics typically has a longer range in the direction parallel to the dipole, compared to the perpendicular direction. These two properties of the dynamics lead to a characteristic swirling pattern before non-linearities set in. The range of correlation at this stage depends on the location in the phase diagram and increases logarithmically as a function of .
We next consider how various types of symmetry-breaking orienting signals influence PCP dynamics.
|
10.1371/journal.pgen.1002522 | Genome-Wide Association Study Identifies Chromosome 10q24.32 Variants Associated with Arsenic Metabolism and Toxicity Phenotypes in Bangladesh | Arsenic contamination of drinking water is a major public health issue in many countries, increasing risk for a wide array of diseases, including cancer. There is inter-individual variation in arsenic metabolism efficiency and susceptibility to arsenic toxicity; however, the basis of this variation is not well understood. Here, we have performed the first genome-wide association study (GWAS) of arsenic-related metabolism and toxicity phenotypes to improve our understanding of the mechanisms by which arsenic affects health. Using data on urinary arsenic metabolite concentrations and approximately 300,000 genome-wide single nucleotide polymorphisms (SNPs) for 1,313 arsenic-exposed Bangladeshi individuals, we identified genome-wide significant association signals (P<5×10−8) for percentages of both monomethylarsonic acid (MMA) and dimethylarsinic acid (DMA) near the AS3MT gene (arsenite methyltransferase; 10q24.32), with five genetic variants showing independent associations. In a follow-up analysis of 1,085 individuals with arsenic-induced premalignant skin lesions (the classical sign of arsenic toxicity) and 1,794 controls, we show that one of these five variants (rs9527) is also associated with skin lesion risk (P = 0.0005). Using a subset of individuals with prospectively measured arsenic (n = 769), we show that rs9527 interacts with arsenic to influence incident skin lesion risk (P = 0.01). Expression quantitative trait locus (eQTL) analyses of genome-wide expression data from 950 individual's lymphocyte RNA suggest that several of our lead SNPs represent cis-eQTLs for AS3MT (P = 10−12) and neighboring gene C10orf32 (P = 10−44), which are involved in C10orf32-AS3MT read-through transcription. This is the largest and most comprehensive genomic investigation of arsenic metabolism and toxicity to date, the only GWAS of any arsenic-related trait, and the first study to implicate 10q24.32 variants in both arsenic metabolism and arsenical skin lesion risk. The observed patterns of associations suggest that MMA% and DMA% have distinct genetic determinants and support the hypothesis that DMA is the less toxic of these two methylated arsenic species. These results have potential translational implications for the prevention and treatment of arsenic-associated toxicities worldwide.
| Exposure to arsenic through drinking water is a serious public health issue in many countries, including Bangladesh and the United States. Although there is substantial inter-individual variation in arsenic metabolism and toxicity, the biological basis of this variation is not well understood. Here, we have conducted the first genome-wide association study of arsenic-related traits within a unique population cohort of arsenic-exposed Bangladeshi individuals. Using data on 1,313 well-characterized individuals, we identify multiple association signals for urinary arsenic metabolite concentrations in the 10q24.32 regions, near the AS3MT (arsenite methyltransferase) gene. In a subsequent analysis of >2,000 individuals, we show for the first time that variants that influence arsenic metabolism can also influence risk for arsenical skin lesions (the classical sign of arsenic toxicity) through interaction with arsenic exposure. Using array-based genome-wide gene expression data, we show that several of our lead genetic variants are associated with expression of AS3MT and neighboring gene C10orf32, providing a potential mechanism by which 10q24.32 variants influence arsenic metabolism and toxicity. Knowledge of variation in this region and associated biological processes could be used to develop intervention and pharmacological strategies aimed at preventing large numbers of arsenic-related deaths in arsenic-exposed populations.
| Over 100 million individuals worldwide are exposed to arsenic through drinking water, including approximately 56 million people in Bangladesh [1] and 13 million in the United States [2]. Arsenic is a class I human carcinogen, and chronic exposure to high levels of arsenic (>300 µg/L) is associated with substantial increased risk for a wide array of diseases including cancers of the lung [3], bladder [4], liver [5], skin [6], and kidney [7], [8], as well as neurological [9], [10] and cardiovascular [11] diseases. Emerging evidence suggests that arsenic may have adverse effects on health even at concentrations as low as 10–50 µg/L, as recent studies in Bangladesh have observed dose-response relationships with mortality [12], [13] and arsenical skin lesion risk [14] in populations with low to moderate arsenic exposure over many years. Arsenical skin lesions are a classical sign of arsenic toxicity, an indicator of susceptibility to arsenic-related disease, and a precursor to arsenic-induced skin cancers [6]. Once individuals are chronically exposed to arsenic, risk for arsenic-related diseases and mortality remains high for several decades even after cessation of exposure [15], [16].
Consumed arsenic enters the blood as AsV and AsIII, known collectively as inorganic arsenic (iAs). Once consumed, iAs is methylated using S-Adenosyl methionine (SAM) as the methyl donor, producing monomethylarsonic acid (MMA) and then dimethylarsinic acid (DMA). MMA is believed to be the more toxic of these metabolites, with the DMA/MMA ratio showing an inverse association with arsenic toxicity in several studies [17]–[20] and DMA being more readily excreted in urine and expelled from the body. Arsenic metabolite concentrations are often expressed as percentages of all arsenic species present in urine (i.e., iAs%, MMA%, DMA%) or as ratios that reflect methylation efficiency (e.g., DMA%/MMA%, MMA%/iAs%).
There is considerable inter-individual variation in arsenic metabolism, as some individuals are able to methylate, and thus excrete, arsenic more efficiently than others [21], [22]. Similarly, because high inter-individual variability in toxicity is observed among individuals with similar levels of exposure to arsenic, genetic susceptibility factors for arsenical skin lesions are believed to exist [23].
In light of the enormous global health impact of arsenic exposure and the remarkable inter-individual variability in arsenic metabolism and toxicity, we performed the first genome-wide association study (GWAS) of common arsenic-related phenotypes. We identified multiple genetic variants in the 10q24.32 region near AS3MT (arsenite methyltransferase, previously known as CYT19) that show robust associations with urinary concentrations of arsenic metabolites, risk for arsenical skin lesions, and local gene expression, including transcript levels of AS3MT.
We assessed genome-wide associations for the three arsenic metabolites measured in urine (iAs%, MMA%, and DMA%) using high-quality data on 259,597 single-nucleotide polymorphisms (SNPs) from 1,313 individuals randomly selected from a large population-based cohort of Bangladeshi individuals exposed to a wide range of arsenic concentrations through drinking water. Associations were assessed using mixed linear models [24] to account for existence of related individuals in our sample (Figure S1). The strongest association signals, genome-wide, for both DMA% and MMA% were in the 10q24.32 region (P<5×10−8) (Figures S2 and S3), which contains the AS3MT gene and substantial LD spanning ∼1 Mb (Figure S4).
For DMA%, the strongest 10q24.32 association was for rs9527 (P = 2.7×10−9; Figure 1). After conditioning on rs9527, a strong residual association signal remained (rs11191527; P = 8.0×10−8), the strength of which was weaker without adjustment for rs9527 (P = 2.3×10−5) due to mild LD between these SNPs (D′ = 0.26, r2 = 0.03 in our data; D′ = 0.27; r2 = 0.03 in HapMap GIH). After conditioning on both SNPs, there was very little evidence of additional association in the region. Analyses of imputed and measured genotypes produced the same two association signals, but with imputed SNPs rs3740394 and rs17115073 showing slightly stronger association than rs9527 and rs11191527, respectively (Figure S5).
The strongest association observed for MMA% was rs4919694 (P = 2.9×10−8) (Figure 2). After conditioning on rs4919694, residual association was still observed (rs4290163; P = 7.0×10−5). This association is much weaker without adjustment for rs4919694 (P = 0.03) due to LD between rs4919694 and rs4290163 (D′ = 0.80, r2 = 0.09 in our data; D′ = 0.80, r2 = 0.04 in HapMap GIH). Aftern conditioning on both SNPs, residual association was observed for rs11191659 (P = 0.0009), a SNP in moderate LD with rs9527, the top SNP from the %DMA analysis (D′ = 0.66, r2 = 0.23 in our data; D′ = 0.82, r2 = 0.30 in HapMap GIH). Conditioning on all three SNPs eliminated the 10q24.32 association signal. Imputation of unobserved genotypes in the region did not reveal associations stronger than those observed for the measured genotypes (Figure S6). Multivariate models for %DMA and %MMA including all five of the above-mentioned SNPs are described in Table 1. Outside of the 10q24.32 region, there was no genome-wide significant (P<5×10−8) association signal for DMA% or MMA%.
The 10q24.32 association results for the DMA%/MMA% ratio (the “secondary methylation index”, log-transformed), were very similar to the MMA% results, as these phenotypes were strongly correlated (r = −0.84; Table S1). Associations for 10q24.32 SNPs with iAs% and MMA%/iAs% (the “primary methylation index” (PMI) log-transformed) were much weaker than for DMA% and MMA%; The strongest association in the 10q24.32 region observed for iAs% was rs9527 (P = 0.0009) and no association of P<0.001 was observed for log(PMI). In genome-wide analyses of iAs% and PMI, no SNP reached genome-wide significance (Figure S7).
Because variants influencing arsenic metabolism may alter susceptibility to arsenic toxicity, we investigated the roles of metabolite-associated SNPs in arsenic-induced premalignant skin lesions, the hallmark of chronic arsenic toxicity. For our five lead SNPs, we tested association with skin lesion status among 1,085 skin lesion cases and 1,794 population controls, using the ROADTRIPS method that was developed for case-control association testing in the presence of cryptic relatedness [25]. The rs9527 allele associated with decreased DMA% (A) was associated with increased skin lesion risk (P = 0.0005), consistent with the hypothesis that DMA is less toxic than MMA (Table 2). rs11191659 showed suggestive association (P = 0.02), also consistent with this hypothesis.
To confirm that these associations with skin lesions were due to gene-arsenic interaction, we tested the interaction between rs9527 and arsenic exposure using a subset of 69 incident skin lesion cases and 700 controls with prospectively-measured arsenic exposure (measured in both water and urine at baseline, prior to skin lesion incidence and arsenic mitigation efforts [26]). We found SNP-arsenic interaction for both rs9527 (multiplicative interaction P = 0.01; additive interaction P = 0.004) and rs11191659 (multiplicative interaction P = 0.02; additive interaction P = 0.001), where water arsenic exposure showed stronger association with skin lesions in the presence of the risk allele (Table 2 and Table S2). There were no significant main effects for either of these SNPs in the context of models that included SNP-arsenic interaction terms.
For the subset of individuals with available genotype, arsenic metabolite, and skin lesion data (82 cases, 1211 controls), DMA% showed evidence of partial mediation of the association between rs9527 and skin lesions (accounting for 13% of the observed association).
To investigate the role of our lead SNPs in gene regulation, used genome-wide expression data derived from lymphocyte RNA obtained at baseline for 950 participants (Illumina HumanHT-12 array) and examined SNP-expression associations for all 30 genes in the 10q24.32 LD region (Table S3). Several of our lead SNPs showed association with AS3MT expression at P<5×10−5 (rs4919694, rs9527, rs4290163). However, after examining associations for all SNPs in this region, C10orf32 intronic SNP rs7096169 showed the strongest association with AS3MT expression (P = 8×10−12; Figure 3 and Figure S8), and conditioning on rs7096169 eliminated the eQTL signal. rs7096169 was not one of our lead SNPs, but it was associated with DMA% (P = 0.001; MMA% P = 0.28). Interestingly, the rs9527 risk allele (A) was associated with decreased C10orf32 expression (P = 2.6×10−41; Figure S8), the strongest eQTL signal for C10orf32 expression in the region (Figure 3) and the strongest genome-wide eQTL effect for rs9527 (Figure S9). C10orf32 is ∼4 kb upstream of AS3MT, and these genes are involved in C10orf32-AS3MT read-through transcription, producing a transcript that is a candidate for nonsense-mediated mRNA decay. Thus, it is possible that the eQTL signal observed for C10orf32 represents a regulatory mechanism that influences read-through transcript production. After conditioning on rs9527, the residual eQTL signal was best represented by rs11083790 (P = 10−5). Conditioning on both SNPs eliminated the eQTL signal. Interestingly, C10orf32 expression was also associated with arsenic exposure (measured as total arsenic in urine, collected at the same time as blood; P = 0.001), while AS3MT expression was not (P = 0.37). None of our lead SNPs modified the association between arsenic exposure and C10orf32 expression.
Our lead SNPs were also associated with USMG5 expression (Table S3), a gene ∼500 kb downstream of AS3MT, but these associations appear to be due to moderate LD with downstream variants showing very strong association with USMG5 expression (e.g., rs12220267; P = 10−210; Figure S8).
The role of AS3MT in arsenic metabolism has been described [27], and several prior studies have evaluated associations between candidate AS3MT variants arsenic-related traits in Bangladesh and elsewhere [28]–[34]. A recent review [35] highlighted two AS3MT SNPs, rs11191439 (Met287Thr) and rs3740393 (intronic), as being consistently related to arsenic metabolism across diverse populations. The most recent and comprehensive Bangladeshi study of AS3MT SNPs [28] reported three association signals for arsenic metabolites, best represented by HapMap3 SNPs rs1046778 (for MMA%), rs11191439 (DMA% and iAs%), and rs3740390 (DMA% and iAs%), a proxy for rs3740393 (r2 = 0.91). After imputation, we were able to replicate rs11191439 (DMA% P = 4.2×10−6; MMA% P = 5.8×10−7) and rs1046778 (MMA% P = 8.9×10−7; DMA% P = 0.0002), which were strongly correlated with lead SNPs rs4919694 (r2 = 0.69) and rs4290163 (r2 = 0.63), respectively. After conditioning on our lead SNPs, these associations were no longer significant. The evidence for rs3740390 was less convincing (DMA% P = 0.54; MMA% P = 0.007), as this SNP was not strongly correlated with any of our lead SNPs (Figure S10). We identified two novel 10q24.32 association signals, represented by rs9527 and rs11191527, which were not strongly correlated with any previously-reported SNP (Figure S10). These SNPs were likely missed in prior studies due to limited coverage of the SNPs in this region.
The identities of the functional variants in this region remain unclear. rs9527 lies in the 5′ UTR of C10orf32, a transcription factor binding region (GATA-1 and TAL1 (SC-12984)) and a DNase hypersensitivity site. If causal, rs9527 could also exert its effects through regulation of AS3MT-C10orf32 read-through transcription. However, the LD block represented by rs9527 includes transcription factor binding site SNP rs12416687 and miRNA SNPs rs11191401, rs12573077, rs7904252, and rs9527. Detailed information on potential functional variants from HapMap3 (GIH) for each of the 5 SNPs identified is contained in Tables S4, S5, S6, S7, S8. However, genetic variation in this population has not been comprehensively characterized (especially rare variation), and the underlying functional variants may not be present in HapMap3. It is also possible that the underlying causal variants have implications for surrounding genes. For example, rs4919694 and rs11191527 are intronic SNPs within the CNNM2 gene, which is involved in magnesium reabsorption by the kidney [36]. It is possible that magnesium and iAs interact [37], [38], influencing the amount of free arsenic available for methylation.
To our knowledge, this study is the largest genetic association study of arsenic metabolites to date, the only GWAS of arsenic-related traits, the first study to implicate 10q24.32 SNPs in both arsenic metabolism and arsenical skin lesion risk, and one of the earliest GWAS conducted in the developing country setting. Our results suggest that MMA% and DMA% have distinct genetic determinants and highlight the importance of conditional analyses, as LD among alleles with opposing effects can mask associations in univariate analyses. The associations observed in this study are likely due to the effects of unmeasured, potentially rare variants in LD with the measured SNPs and/or substantial allelic heterogeneity, whereby multiple 10q24.32 variants influence arsenic metabolism.
Considering the substantial LD in this region [39], the variation in allele frequencies and LD patterns among the various arsenic-exposed populations under study [40], and the apparent allelic heterogeneity with respect to arsenic metabolism, future DNA sequencing studies are needed to help identify causal variants in the 10q24.32 region. Identifying these variants will help clarify the links between the association signals observed for %DMA, %MMA, and AS3MT/C10orf32 expression. These association signals appear largely independent in our dataset, but perhaps there are underlying causal variants that influence all of these phenotypes. Developing a better understanding the effects of functional variation related to AS3MT will also provide a more nuanced understanding of the biology of arsenic methylation, which can in turn help us better understand how variation in methylation efficiency affects health. Finally, knowledge of this causal variation and the methylation processes that they influence could potentially be exploited for intervention strategies that aim to prevent large numbers of deaths arsenic-exposed populations, by defining susceptibility subgroups and exploiting the biological processes uncovered by genomics for developing pharmacological treatments.
The DNA samples genotyped in this study were obtained at baseline recruitment from individuals participating in one of the following studies: The Health Effects of Arsenic Longitudinal Study (HEALS) [41] or the Bangladesh Vitamin E and Selenium Trial (BEST) [42]. GWAS analyses of arsenic metabolites were conducted using urinary arsenic metabolite and SNP data on 1,313 individuals randomly selected from the HEALS study. Analyses of skin lesion data were conducted using genotype data from 1,085 skin lesion cases and 1,794 controls drawn from both studies, including the 1,313 HEALS individuals with metabolite data. Skin lesion cases included individuals with keratosis, melanosis, and leukomelanosis. Gene expression analyses were based on lymphocyte RNA extracted at baseline recruitment for the first 950 BEST participants. A summary of these overlapping sets of samples is provided in Figure S11.
The Health Effects of Arsenic Longitudinal Study (HEALS [41]) is a prospective investigation of health outcomes associated with arsenic exposure through drinking water in a cohort of adults in Araihazar, Bangladesh, a rural area east of the capital city, Dhaka. Between October 2000 and May 2002, we recruited healthy married individuals (age 18–75 years) who were residents of the study area for at least five years and primarily consumed drinking water from a local well. We enumerated 65,876 individuals residing in Araihazar, from which we identified a sampling frame of 14,828 eligible residents. Of these 14,828 individuals, 11,746 men and women were enrolled. During 2006–2008, additional recruitment of 8,287 participants from the same underlying source population expanded the cohort size to over 20,000 individuals. All 5,966 wells in the study area were tested for arsenic using graphite furnace atomic absorption spectrometry and individuals reported the primary well from which they drank. At baseline, trained study physicians, blinded to the arsenic measurements, conducted in-person interviews and clinical evaluations and collected spot urine and blood samples from participants in their homes using structured protocols. Similar in-person follow-up interviews were conducted biennially for the entire cohort during the following periods: follow-up 1 during September 2002 to May 2004, follow-up 2 during June 2004 to August 2006, and follow-up 3 during January 2007 to February 2009. At baseline and each follow-up interview, a structured protocol was used to ascertain skin lesions by the study physicians, who had undergone training for the detection and diagnosis of skin lesions [43]. The study protocol was approved by the Institutional Review Boards of The University of Chicago, Columbia University, and the Bangladesh Medical Research Council. Informed consent was obtained from all participants.
The Bangladesh Vitamin E and Selenium Trial (BEST) is a 2×2 factorial randomized chemoprevention trial evaluating the long-term effects of vitamin E and selenium supplementation on non-melanoma skin cancer (NMSC) risk. BEST participants are residents of Araihazar (the same geographic area as HEALS participants with 132 overlapping participants), Matlab, and surrounding areas. BEST uses many of the same study protocols as does HEALS, especially arsenic exposure assessment and biospecimen collection protocols. All participants were required to have existing arsenic-related skin lesions to be eligible. A total of 7,000 individuals have been randomized to one of the four treatment arms: vitamin E only (100 IU/day), L-selenomethionine only (200 µg/day), both vitamin E and selenium, and placebo. Participants have been actively followed for 6 years and systematic ascertainment of histopathologically-confirmed NMSC has been conducted (including BCC and SCC). For all participants, biological samples, including all fractions of blood including DNA and RNA, urine, toenails, and tumor samples have been collected at baseline, along with clinical and covariate data, creating a biological and data repository that is available for research purposes. The study protocol was approved by the Ethical Review Committee of International Center for Diarrheal Disease Research, Bangladesh, the Bangladesh Medical Research Council, and the Institutional Review Boards of The University of Chicago and Columbia University. Informed consent was obtained from all participants.
In each study, urinary arsenic was measured using graphite furnace atomic absorption spectrometry in a single laboratory [44]. Urinary creatinine was measured by a colorimetric diagnostics kit (Sigma, St Louis, MO, USA). Total urinary arsenic concentration was divided by creatinine to obtain creatinine-adjusted total arsenic concentration (µg/g creatinine) [45]. Urinary arsenic metabolites (arsenobetaine, arsenocholine, arsenite, arsenate, monomethylarseonous acid, and dimethylarsenic acid) were distinguished as described by Ahsan et al. [17], using a high-performance liquid chromatography method for separation of arsenic metabolites, followed by detection using inductively coupled plasma-mass spectrometry with dynamic reaction cell. The percentage of iAs, MMA and DMA in total arsenic was calculated after subtracting asenobetaine and arsenocholine (i.e., nontoxic organic arsenic from dietary sources). Because these metabolites lie on the same biological pathway and are expressed as a percentage of arsenic species, their values show substantial correlation (Table S1).
For BEST samples, DNA extraction was carried out from the whole blood using the QIAamp 96 DNA Blood Kit (cat # 51161) from Qiagen, Valencia, USA. For HEALS samples, DNA was extracted from clot blood using Flexigene DNA kit (Cat # 51204) from Qiagen. Concentration and quality of all extracted DNA were checked by Nanodrop 1000. As starting material, 250 ng of DNA was used on the Illumina Infinium HD SNP array according to Illumina's protocol. Samples were processed on HumanCytoSNP-12 v2.1 chips with 299,140 markers and read on the BeadArray Reader. Image data was processed in BeadStudio software to generate genotype calls.
Prior to genotype QC, our genotype data consisted of 2,920 samples typed for 299,140 SNPs. First, we removed DNA samples with very poor call rates (<90%; n = 8) and SNPs that were poorly called (<90%) or monomorphic (n = 39,276). Individuals with gender mismatches were removed (n = 10), as were technical replicate DNA samples run to assure high genotyping accuracy (n = 21). No individuals had outlying autosomal heterozygosity or inbreeding values. After inspecting distributions of SNP and samples call rates, we excluded samples with call rates <97% (n = 2) and SNPs with call rates <95% (n = 103). SNPs with HWE p-values<10−7 were excluded (n = 164). This QC resulted in 2,879 individuals with high-quality genotype data for 259,597 SNPs. All QC was performed using PLINK [46].
RNA was extracted from mononuclear cells preserved in buffer RLT, stored at −86°C using RNeasy Micro Kit (cat# 74004) from Qiagen, Valencia, USA. Concentration and quality of all extracted RNA were checked on Nanodrop 1000. cRNA synthesis was done from 250 ng of RNA using Illumina TotalPrep 96 RNA Amplification kit. As recommended by Illumina we used 750 ng of cRNA on HumanHT-12-v4 for gene expression. The chip contains a total of 47,231 probes covering 31,335 genes.
Pair-wise kinship coefficients were estimated using PLINK [46] and their distribution is shown in Figure S1. To assess population structure that was unrelated to the relative pairs present in our dataset, we removed one individual from each related pair (kinship coefficient >0.05) and assessed population structure in this dataset of 403 individuals using principal components analysis as implemented in EIGENSTRAT [47]. We found very little evidence of population stratification (Figure S12), with the eigenvalues from the first ten principle components being between 1.123 and 1.184. All SNP association tests for urinary metabolites were conducted using a mixed model that accounted for cryptic relatedness as implemented in EMMAX [24] (rather than principle components), adjusting for water arsenic, sex, and age. All regional association plots were generated using LocusZoom [48]. Association testing for skin lesion status was conducting using PLINK [46] and the ROADTRIPS [25] software developed for case-control association testing in samples with unknown population and pedigree structure. We conducted local imputation for the 10q24.32 region using MACH, the GIH reference panel, and imputation parameters suggested by the developers [49]. The estimated genotype and allele error rates were 0.034 and 0.017, respectively. LD structure in the 10q24.32 region was visualized using Haploview [50]. Information on the potential functional consequences of SNPs in the 10q24.32 regions was obtained using the NIEHS's SNPinfo Web Server [51]. Interaction analyses was conducted using only HEALS incident cases (n = 69) and controls (n = 701). For BEST participants and some HEALS participants arsenic exposure (based on water and urine) was not measured prior to arsenic mitigation efforts [26], so the measured exposure status for these individuals is not likely to reflect long-term arsenic exposure status. Interactions were tested using the SAS 9.2 PROC MIXED procedure, using the “bn” matrix derived using EMMAX. To assess mediation of the association between SNPs and skin lesions, we used the “proportion explained” (PE) equation for odds ratios (PEOR = (ORxy−ORxy|m)/(ORxy−1) where x is an exposure, y is a binary outcome, and m is a potential mediating factor [52]). Genome-wide eQTL analysis for our five lead SNPs was performed using the significance of microarray method as implemented in BRB Array Tools. Promising eQTL effects were then examined using EMMAX as described above, treating the expression values as a quantitative trait. In a similar fashion, arsenic exposure was tested for association with expression traits of interest and for interaction with SNPs in relation to expression traits using PROC MIXED.
|
10.1371/journal.ppat.1007177 | HCV NS5A dimer interface residues regulate HCV replication by controlling its self-interaction, hyperphosphorylation, subcellular localization and interaction with cyclophilin A | The HCV NS5A protein plays multiple roles during viral replication, including viral genome replication and virus particle assembly. The crystal structures of the NS5A N-terminal domain indicated the potential existence of the NS5A dimers formed via at least two or more distinct dimeric interfaces. However, it is unknown whether these different forms of NS5A dimers are involved in its numerous functions. To address this question, we mutated the residues lining the two different NS5A dimer interfaces and determined their effects on NS5A self-interaction, NS5A-cyclophilin A (CypA) interaction, HCV RNA replication and infectious virus production. We found that the mutations targeting either of two dimeric interfaces disrupted the NS5A self-interaction in cells. The NS5A dimer-interrupting mutations also inhibited both viral RNA replication and infectious virus production with some genotypic differences. We also determined that reduced NS5A self-interaction was associated with altered NS5A-CypA interaction, NS5A hyperphosphorylation and NS5A subcellular localization, providing the mechanistic bases for the role of NS5A self-interaction in multiple steps of HCV replication. The NS5A oligomers formed via different interfaces are likely its functional form, since the residues at two different dimeric interfaces played similar roles in different aspects of NS5A functions and, consequently, HCV replication. In conclusion, this study provides novel insight into the functional significance of NS5A self-interaction in different steps of the HCV replication, potentially, in the form of oligomers formed via multiple dimeric interfaces.
| HCV NS5A is a multifunctional protein involved in both viral RNA replication and infectious virus production, and is a target of one of the most potent antivirals available to date. However, the mode of action of NS5A inhibitors is still unclear due to the lack of mechanistic detail regarding NS5A functions during HCV life cycles. In this study, we have provided evidence that surface-exposed NS5A residues involved in two different dimeric interactions in crystal structures are indeed involved in NS5A self-interactions in cells. We also showed that these NS5A residues play critical role in HCV RNA replication and infectious virus production by regulating NS5A hyperphosphorylation, its subcellular localization and its interaction with host protein CypA. Overall, our data support the functional significance of “NS5A oligomers” formed via multiple interfaces in HCV replication. We speculate that the NS5A inhibitors exploited the NS5A oligomer-dependent functions during HCV replication, rather than targeting individual NS5A, which consequently resulted in their high potency.
| Hepatitis C virus (HCV) is a main causative agent associated with chronic liver diseases including chronic hepatitis, cirrhosis and hepatocellular carcinoma [1, 2]. It is an enveloped, positive-stranded RNA virus belonging to the genus hepacivirus within the flaviviridae family [3]. A single polyprotein translated from the viral genome encodes structural proteins, including core, E1, and E2 at the N-terminal domain followed by the viral assembly accessory proteins p7 [4, 5] and NS2 [6–10]. The C-terminal domain encodes five different nonstructural proteins including NS3, NS4A, NS4B, NS5A and NS5B, which comprise viral replicase complexes [11] and regulate viral assembly [12–17].
NS5A associates with membrane through its N-terminal amphipathic helix (AH) domain [18]. Following the AH domain are three major domains called domain I (DI), DII, and DIII. These domains are separated by two low-complexity sequences (LCS) called LCSI and LCSII. In general NS5A-DI and DII were shown to play roles in HCV RNA replication [19–21], and DIII was associated with virus particle assembly [16]. NS5A is a phosphoprotein expressed as a hypophosphorylated form, which is further phosphorylated to a hyperphosphorylated form [22, 23]. The clusters of highly conserved residues at LCSI served as a target of casein kinase I-α (CKI-α)-mediated hyperphosphorylation, and blocking this inhibited the NS5A localization to the lipid droplets (LD)-associated, low-density membranes and impaired infectious virus production [17]. The casein kinase II-mediated NS5A-DIII phosphorylation was also shown to affect HCV particle assembly by regulating NS5A and core interaction [24, 25].
Highly effective anti-HCV therapies are composed of different combinations of antiviral compounds targeting viral enzymes, such as NS3/4A protease and NS5B polymerase, and NS5A [26]. Since NS5A lacks enzyme activity, NS5A inhibitors were discovered via high throughput screening (HTS) of chemical libraries by using HCV replicon systems [27, 28]. NS5A inhibitors are one of the most potent antivirals to date, inhibiting NS5A function during HCV replication at a picomolar range in cell culture-based systems (reviewed in [29]). NS5A inhibitors impaired HCV RNA replication by preventing the formation of double membrane vesicles (DMV) that constitute HCV RNA replication factories [30–32]. NS5A inhibitors also blocked intracellular HCV particle assembly [33]. Both AH and DI domains of NS5A were indispensable for DMV formation [19]. Therefore, it is no surprise that NS5A inhibitors, which were shown to inhibit DMV formation, target these two domains, as evidenced by the accumulation of drug-resistant mutations in these areas [34]. Interestingly, many NS5A inhibitors are bivalent (have two identical pharmacophores), and the bivalent form of compounds such as the iminothiazolidinone-based compound from Bristol-Myers Squibb (BMS-824) are much more potent than the corresponding monovalent forms [27, 35]. These observations fueled the speculation that bivalent NS5A inhibitors target the dimeric form of NS5A. In fact, multiple lines of evidence from different studies support that NS5A functions as a dimer or multimer. These include its dimeric crystal structures [36–38] and the detection of NS5A-NS5A interaction in vitro by using purified proteins [20, 39] or in vivo in the NS5A ectopic expression system [40, 41] and HCV-replication system [30, 42]. In addition, Lim et al. showed that mutations introduced to four cysteine residues in NS5A (C39, C57, C59 and C80), shown to be involved in Zinc binding and required for HCV RNA replication [43], inhibited purified NS5A self-interaction [20]. However, while these studies clearly demonstrated the correlation between NS5A dimerization and HCV RNA replication, the exact roles of NS5A dimerization in HCV life cycle is still unclear.
In this study, we performed a structure-function analysis of highly conserved, surface-exposed NS5A-DI residues located at two different dimer interfaces, predicted from NS5A-DI crystal structures, on NS5A protein interactions and HCV replication. Our data indicate that these dimer interface residues are involved in NS5A self-interaction in Huh-7 cells and that NS5A self-interactions through these residues are critical for different steps of HCV replication by regulating NS5A hyperphosphorylation, subcellular localization and interaction with host protein CypA.
The genotype 1b (gt1b) NS5A-DI crystal structures depicted in Fig 1, designated as 1ZH1 and 3FQM, suggested that the NS5A self-interaction occurs via at least two different interfaces [36, 37]. To determine the relevance of these two dimeric forms of NS5A in its self-interaction in cells, we performed alanine-scanning mutagenesis of selected dimer-interface residues in full- length NS5A derived from gt1a (H77 strain) and gt2a (JFH1 strain) and determined the impact of these mutations on NS5A self-interaction. NS5A-DI residues 36, 37 and 38 were chosen since they provide a continuous surface patch involved in 1ZH1-specific dimer interaction (Fig 1A, see also S1A Fig showing the 1ZH1 residue interaction networks for detail). Residues 112 and 148 were selected since they could form a salt bridge at the 3FQM-specific dimer interface (Fig 1B, see also S1B Fig showing the 3FQM residue interaction networks for detail). As shown in Fig 1C, residues 38 serine (38S), 112 arginine (112R) and 148 aspartic acid (148E) are completely conserved among 672 HCV sequences derived from all genotypes deposited in the Los Alamos HCV database. Residue 36 encodes major variant phenylalanine (F) or minor variant leucine (L). Residue 37 encodes hydrophobic residues, including valine (V), leucine (L), phenylalanine (F) or isoleucine (I).
We determined the NS5A self-interaction by using a mammalian two-hybrid system as previously described [40, 41] (Fig 2). In brief, NS5A was fused to the GAL4 DNA-binding domain (GAL4/BD) in pBIND vector and the herpes simplex virus VP16 activation domain (VP16/AD) in pACT vector. Then, these two vectors plus a third vector, pGL4.3[luc2P/Gal4UAS/Hygro], encoding five GAL4 binding sites upstream of the firefly luciferase (F-Luc) gene, were transfected to Huh-7 cells. Two days following the transfection of these plasmids, cells were lysed to detect NS5A self-interaction efficiency by measuring F-Luc activity normalized by that of Renilla reniformis luciferase (R-Luc) expressed from a pBIND vector to adjust transfection efficiency. Under these experimental conditions, stronger protein-to-protein interaction resulted in higher F-Luc/R-Luc ratio. These ratios derived from the basal vectors (pACT and pBind) and the known interacting-pair [pACT-MyoD (myogenic regulatory protein) and pBind-ID (negative regulator of myogenic differentiation)] were used as a protein-protein interaction negative (-) control and positive (+) control, respectively (Fig 2A). Robust interaction was detected between a wild-type (wt) NS5A pair derived from a gt1a HCV H77 (designated as H-wt), evidenced by the higher F-Luc/R-Luc value from H-wt pair than that from the (+) control. Interestingly, NS5A from a gt2a HCV JFH1 (designated as J-wt) showed an even stronger self-interaction than did H-wt (Fig 2A). Next, we determined the impact of alanine mutations in the residue lining the 1ZH1 and 3FQM NS5A dimer interfaces, including 36A/37A/38A (mutation group a) and 112A/148A (mutation group b), respectively. These H77 NS5A mutants, designated as H-a or H-b, showed a significantly reduced NS5A-NS5A interaction compared to that of H-wt (Fig 2B). The mutants having both mutations, designated as H-ab, showed a stronger reduction in NS5A self-interaction (Fig 2B). Similarly, JFH1 NS5A-NS5A interaction was also significantly reduced when these mutations were introduced separately or combined (Fig 2C). In summary, these data suggest that NS5A from gt1a H77 and gt2a JFH1 form dimers/oligomers via at least two different dimeric interfaces in Huh-7 cells.
To narrow down the critical residues involved in H77 NS5A self-interaction, we introduced individual alanine mutations to residues 36, 37, 38, 112 or 148 in H77 NS5A two-hybrid vectors and designated them as H-36A, H-37A, H-38A, H-112A or H-148A, respectively. Then the effects of these individual mutations on NS5A self-interaction were determined by measuring the dual luciferase (F-Luc and R-Luc) activities from Huh-7 cells transfected with these plasmids as described above. As shown in Fig 3A, all mutations, with the exception of 38A, significantly reduced NS5A self-interaction down to the levels between ~10 to 50% of wt H77 NS5A-NS5A interaction efficiency on average. Interestingly, the 38A mutation significantly enhanced NS5A self-interaction (~30% above wt level on average). We speculate that 38A mutation altered the non-covalent interactions of a residue 157 arginine (157R), which was a wt 38S interactor, inadvertently stabilizing the NS5A inter-molecular interaction (see S1 Fig). Next, we determined the effect of mutations modulating NS5A self-interaction on infectious gt1a H77D replication [44], following electroporation of wt and mutant H77D RNAs to Huh-7 cells, then, determining the expression of viral proteins and RNAs at different time points post electroporation by western blot and quantitative RT-PCR analyses (Fig 3B and 3C). Results indicated that altered H77 NS5A self-interaction inhibited viral replication (Fig 3B and 3C). In detail, the 37A and 148A mutations caused moderate-to-substantial reductions in viral protein expression and RNA replication (Fig 3B and 3C). The 36A and 112A mutations completely blocked viral replication as evidenced by undetectable viral protein expression and the lack of viral RNA increase over the replication-defective H77 mutant AAG [45] (Fig 3B and 3C). The 38A mutation, which moderately enhanced NS5A self-interaction, also reduced HCV RNA replication (Fig 3C, left panel). These data suggest that optimal mode of H77 NS5A self-interaction is critical for efficient gt1a virus RNA replication. Alternatively, 38A mutation could have disrupted other functions of H77 NS5A unrelated to NS5A self-interaction, or altered overall conformation of NS5A, resulting in H77D replication defect. To better understand the exact role of H77 NS5A residue 38 in HCV RNA replication, we have tried to obtain the revertant of H77D/38A mutant. However, our multiple attempts to obtain H77D/38A revertant were unsuccessful.
Next, to test the effect of H77 NS5A self-interaction-altering mutations on infectious virus production, we determined the intracellular and extracellular infectivity titers following electroporation of H77D RNA having these individual mutations to Huh-7 cells. As expected, replication-defective H77D/36A or H77D/112A mutants showed no evidence of virus production (Fig 3E). Other mutants, including H77D/37A, H77D/38A and H77D/148A showed severely impaired virus production ranging between ~10- to ~1,000-fold reductions in intracellular infectivity and ~3- to ~100-fold reductions in extracellular infectivity compared to those of the wt at the 72-h time point (Fig 3E). None of the virus-producing H77 NS5A mutants impaired the secretion of virus particles, since the percentage of intracellular infectivity per total (intracellular plus extracellular) infectivity was similar between the wt H77D and these mutants (Fig 3F). Supporting this, the ratio of extracellular and intracellular HCV RNA was similar between the wt H77D and its virus-producing NS5A mutants (Fig 3D). On the other hand, virus particle-assembly efficiency was significantly reduced in these H77 NS5A mutants, since the relative levels of total infectivity per total HCV RNA in these mutants were significantly lower than that from wt H77D (Fig 3G).
Next, we performed a density gradient centrifugation of extracellular virus present in the cell culture supernatant to determine whether H77 NS5A mutations, including 37A, 38A and 148A, impaired the H77D infectivity by modulating virus particle density or specific infectivity. The results shown in Fig 4 indicate that the density distributions of wt H77D and three different NS5A mutant viruses are similar, since the relative percentages of HCV RNA and infectivity present in different density fractions are almost identical between the wt and mutants (Fig 4B and 4D). We also did not detect any significant alteration in the specific infectivity of virus particles between the wt and mutants calculated as a ratio of infectivity per HCV RNA at different density fractions (Fig 4E). Interestingly, the majority of infectious particles (>95%) from both wt and mutants banded at a density between 1.060 and 1.200 g/cm3, centering around at 1.100 g/cm3 (Fig 4B), whereas the peak of specific infectivity was detected at a density of ~1.060 g/cm3 (Fig 4E). These data indicate that a small fraction of low-density gt1a virus particles is more infectious than the majority of virus particles, consistent with previous reports [46, 47].
In summary, these results suggest that H77 NS5A self-interaction is critical for infectious particle-assembly efficiency, but has little impact on specific infectivity and density of infectious virus particles as well as their egress efficiency.
We introduced individual alanine mutations to residues 36, 37, 38, 112 or 148 in JFH1 NS5A two-hybrid vectors and designated them as J-36A, J-37A, J-38A, J-112A and J-148A, respectively. We transfected these two-hybrid plasmid sets for each mutant to Huh-7 cells to determine the impact of these individual mutations on JFH1 NS5A self-interaction. It is interesting that most of the mutations introduced individually to JFH1 NS5A showed relatively moderate defects in NS5A self-interaction, retaining between ~60 to 75% of wt JFH1 NS5A self-interaction on average, except for an 36A mutation, which caused severe defects in NS5A self-interaction (~20% of wt interaction). Unlike the 38A mutation in H77 NS5A, which increased NS5A self-interaction, the same mutation in JFH1 NS5A reduced NS5A self-interaction. In the structural perspective, the key inter-domain interactor of wt residue 38S in H77 NS5A is 157R, which provides both van der Waals and hydrogen-bond interactions to 38S (S1A Fig). However, in JFH1, the corresponding 38S interactor is 157 glutamine (157Q), which shows substantial physico-chemical differences from 157R. We speculate that different properties of residue 157 in H77 and JFH1 contributed a genotype-specific difference in the intermolecular interaction phenotypes of 38A mutation. Also contrary to 112A mutation in H77 NS5A, which caused severe defects in NS5A self-interaction, the equivalent mutation in JFH1 NS5A did not impair NS5A self-interaction in cells (compare Figs 3A and 5A). The wt residue 112R in H77 NS5A is involved in a network of electrostatic interactions (“salt bridge” pattern) with the residues 48 arginine (48R) and 148 glutamic acid (148E) in the partnering NS5A resulting in intra/inter-molecular 48R-148E-112R interaction network (S1B Fig). We speculate that the disruption of this electrostatic network mediated by the 112A mutation was substantial enough to impair the intermolecular H77 NS5A self-interaction. However, since the residue 48 alanine (48A) in JFH1 NS5A does not support a similar kind of interaction network, it is reasonable to assume that a different kind of, genotype-specific interaction network surrounding the residue 112 may have negated the impact of 112A mutation on JFH1 NS5A self-interaction. Alternatively, R112 residue in JFH1 NS5A may not contribute to NS5A self-interaction unlike the same residue in H77 NS5A (see discussion).
To determine the effect of these individual JFH1 NS5A mutations on viral RNA replication, we introduced them to H77-JFH1 chimeric HCV (HJ3-5) encoding JFH1 NS3-NS5B proteins [6, 48] (Fig 5B) and then analyzed viral protein expression and HCV RNA levels following electroporation of these RNAs to Huh-7 cells. The results of these experiments could be summarized as follows. First, viral RNA replication was undetectable for HJ3-5/36A mutant (Fig 5B and 5C). Thus, a 36A mutation in NS5A, which impaired both H77 and JFH1 NS5A self-interaction, blocked both H77 and JFH1 RNA replication (Figs 3 and 5). We also detected reduced replication of an HJ3-5/148A mutant, similar to the case of an H77D/148A mutant. The relatively moderate impact of 148A mutation on HJ3-5 replication, compared to its more severe effect on H77D replication, correlates with its weaker impact on self-interaction of JFH1 NS5A compared to that of H77 NS5A (compare Figs 3 and 5). Overall, these data indicate that JFH1 NS5A self-interaction is also critical for JFH1 replicase-mediated viral RNA replication. Second, both 37A and 38A mutants moderately increased the relative levels of intracellular HCV RNA compared to that of wt HJ3-5 at 72 h post electroporation (Fig 5C, left panel). These results may indicate that these two mutations in JFH1 NS5A potentially enhanced HCV RNA replication. However, additional data indicate that this may not be the case. For example, we detected much lower levels of extracellular HCV RNAs from these mutants compared to those from the wt HJ3-5 (Fig 5C, right panel). In addition, relative extracellular/intracellular HCV RNA ratios from these mutants were over 10-fold lower than that from the wt at the 72-h time point (Fig 5D). Based on these data, we believe that decreased viral RNA secretion, rather than enhanced viral replication, has caused the relatively high levels of intracellular 37A or 38A mutant RNA accumulation. Third, an 112A mutation in JFH1 NS5A completely blocked HCV replication despite having no effect on its self-interaction. At a first glance, these data seem to contradict to the potential role of NS5A self-interaction in HCV replication However, previous study determined that HCV replication defect caused by 112A mutation could be attributed to the inhibition of NS5A-RNA interaction and dysregulation of NS5A’s role in HCV translation [49]. Fourth, the 37A, 38A and 148A mutations impaired the JFH1 NS5A hyperphosphorylation (Fig 5B). This phenotype was also observed from the corresponding H77 NS5A mutants, although the hyperphosphorylation efficiency of H77 NS5A was quite low compared to that of JFH1 NS5A (compare Figs 3B and 5B, see also [50, 51]).
In summary, these data indicate that JFH1 NS5A self-interaction also plays an important role in JFH1 replicase-mediated RNA replication, despite some genotype-specific differences.
As expected from reduced viral RNA replication, the HJ3-5/148A mutant showed significantly reduced intracellular and extracellular infectivity during the entire time course of experiments (Fig 5E). In the case of 37A or 38A mutants, while they also showed significantly lower intracellular and extracellular titers at the 24-h time point, by 48 h, their titers rapidly caught up with those from the wt HJ3-5. However, by the 72-h time point, extracellular infectivity from these two mutants was significantly lower than that from the wt, while intracellular infectivity remained similar to that of the wt (Fig 5E). The extracellular/intracellular RNA ratios of the 37A, 38A and 148A mutants were also significantly lower than that of wt HJ3-5 at the 48- and 72-h time points (Fig 5D). In addition, the percentages of intracellular infectivity per total infectivity of all three mutants were significantly higher than that of the wt at 72 h (Fig 5F). These results suggested a decreased egress of these mutant viruses compared to the wt virus. Interestingly, none of these mutants affected virus particle-assembly efficiency, since the relative ratios of total HCV infectivity per total HCV RNA were similar between wt and these mutants (Fig 5G). Overall, it is remarkable that same mutations introduced to highly conserved residues 37, 38 and 148 in NS5A showed genotype-specific effects on virus production in that H77 NS5A mutants impaired virus assembly, not viral egress, but JFH1 NS5A mutants impaired viral egress, not virus assembly (Figs 3 and 5).
Next, we determined the density and specific infectivity of extracellular HJ3-5 wt as well as those of 37A, 38A and 148A mutant viruses by performing density gradient centrifugation. In general, comparable density distribution patterns were detected between wt HJ3-5 and mutant viruses, as judged from the relative density distributions of viral RNA and infectivity (Fig 6). However, the percentage of wt HJ3-5 in high-density fractions (>1.201 g/cm3) was higher than those of mutants. In fact, ~12% of viral RNA from wt HJ3-5 was detected in these high-density fractions compared to ~3% from each of the JFH1 NS5A mutants (Fig 6D, right panel). However, infectivity of wt HJ3-5 in these high-density fractions was low accounting for less than 2% of total infectivity (Fig 6B, right panel). These results suggest that a significant portion of poorly infectious, high-density immature particles might have been secreted from wt HJ3-5-replicating cells, probably due to highly efficient virus egress (Fig 5F, see discussion). On the other hand, relative titers of mutant viruses at low-density fractions (<1.059 g/cm3) were 6- to 7-fold higher than those of the wt HJ3-5 (Fig 6B, right panels). Due to this, the specific infectivity of mutant viruses at low-density fractions was relatively higher than that of the wt HJ3-5 (Fig 6E). These results suggest that too efficient virus egress might have negative impact on infectious virus maturation (see discussion).
The interaction between NS5A and CypA is critical for HCV RNA replication [52]. Since most of the NS5A self-interaction mutants, especially those derived from gt1a H77D, significantly impaired HCV RNA replication, we asked whether reduced NS5A self-interactions in these mutants might have impaired NS5A-CypA interactions, resulting in decreased HCV RNA replication. To measure the interaction between NS5A and CypA quantitatively, we used a checkmate assay as this method successfully measured the interaction between NS5A and CypA in the previous study [41]. As shown in Fig 7A, the level of interaction between H77 NS5A and CypA was comparable to that of the positive (+) control. Interestingly, interaction between JFH1 NS5A and CypA was stronger than that between H77 NS5A and CypA (Fig 7A). Next, we determined the interaction between CypA and H77 NS5A dimer-interface mutants. The results showed that the mutations that significantly reduced NS5A-NS5A interaction, including H-36A, H-37A, H-112A and H-148A (Fig 3A), also significantly impaired the NS5A-CypA interaction (Fig 7B). These results indicate that effective H77 NS5A self-interaction is critical for H77 NS5A and CypA interaction. These data also suggest that reduced NS5A-CypA interaction in H77 NS5A mutants was responsible for defective viral RNA replication (Fig 3C). The NS5A self-interaction-enhancing H-38A mutant did not significantly reduce the NS5A-CypA interaction (Fig 7B), which correlates with the relatively moderate effect of this mutation on viral RNA replication (Fig 3C).
The results of NS5A dimer-interface mutations on the interaction between JFH1 NS5A and CypA could be summarized as follows. First, the J-36A mutant that showed the most significant defect in JFH1 NS5A self-interaction (Fig 5A) also had the most substantial defect in the JFH1 NS5A and CypA interaction (Fig 7C), which correlates nicely with the undetectable level of HJ3-5/36A RNA replication (Fig 5C). Second, both J-37A and J-38A did not show any significant effect on JFH NS5A-CypA interaction (Fig 7C), which also correlates with their relatively minor effects on NS5A self-interaction and viral RNA replication (Fig 5A and 5C). Third, despite having no impact on JFH1 NS5A self-interaction (Fig 5A), 112A mutation impaired the JFH1 NS5A-CypA interaction (Fig 7C), suggesting that the role of J112R on NS5A-CypA interaction could differ mechanistically from other residues involved in NS5A self-interaction. Fourth, the J-148A mutant showed a reduced NS5A-CypA interaction (Fig 7C), correlating with reduced J-148A self-interaction (Fig 5A) and impaired replication of the HJ3-5/148A mutant (Fig 5C).
In aggregate, these results suggest that NS5A self-interaction contributes to NS5A-CypA interaction, and that the impaired viral RNA replication observed from the majority of NS5A self-interaction-defective mutants could be due to a defective NS5A-CypA interaction.
Hyperphosphorylation of NS5A was shown to contribute to infectious HCV production by regulating NS5A recruitment to low-density membranes in the vicinity of lipid droplets (LD) and facilitating NS5A-core interaction [17]. Since NS5A hyperphosphorylation, as well as infectious virus production, were reduced in NS5A self-interaction mutants, we asked whether these phenotypes were caused by impaired NS5A subcellular localization and/or its interaction with core protein. To facilitate the detection of the NS5A in an immunofluorescence assay and the NS5A-core interaction in a co-immunoprecipitation assay, we used HJ3-5/NS5AYFP, which encodes YFP-tag within the NS5A-DIII and is capable of virus production (Fig 8A) [6]. First, we confirmed that NS5A dimer interface mutations in HJ3-5/NS5AYFP also impaired NS5A hyperphosphorylation and virus production (Fig 8A). To determine the NS5A subcellular localization, Huh-7 cells electroporated with either wt HJ3-5/NS5AYFP or its 37A, 38A and 148A mutants were subjected to confocal imaging analysis following a LipidTOX deep-red lipid staining to detect the LD. As shown in Fig 8B, we frequently detected a tight association between wt NS5A (measured by YFP fluorescence) and LD. However, in the case of NS5A mutants, a majority of NS5A was detected as the distinct foci in the cytoplasm without the tight LD association (Fig 8B). In fact, significantly lower degrees of NS5A-LD co-localization were calculated from the mutants compared to those from the wt, based on Pearson’s correlation measurements derived from the confocal images obtained from ~30 different cells (with the means of Pearson’s correlation coefficients for wt equaling 0.5206 versus 0.2598, 0.2424 and 0.3245 for 37A, 38A and 148A mutants, respectively) (Fig 8C).
The NS5A-core interaction was measured by two different methods: NS5A-core co-localization and co-immunoprecipitation (co-IP). NS5A-core co-localization was determined by performing confocal imaging analysis following immunostaining of core by using core-specific antibody in cells replicating HJ3-5/NS5AYFP (Fig 9A). As shown in Fig 9B, a strong degree of co-localization was detected between wt NS5A and core (with a mean of Pearson’s correlation coefficient equaling 0.7726). However, lesser degrees of co-localization between these two proteins were detected from the 37A, 38A and 148A mutants (with the means of Pearson’s correlation coefficients equaling 0.5742 0.6564 and 0.6100, respectively) (Fig 9B). Next, we determined the NS5A-core interaction by performing a GFP-pull down assay. As shown in Fig 9C, compared to wt, NS5A mutants showed reduced NS5A and core co-IP efficiency (~50% lower than wt) expressed as a ratio of co-IP-core level per immunoprecipitated (IP)-NS5AYFP. These NS5A-core co-IP results correlate well with their co-localization data (Fig 9A and 9B) and indicate that NS5A-core interaction in NS5A mutants was reduced compared to that in wt. These results indicate that NS5A self-interaction regulates subcellular localization of NS5A and NS5A-core interaction. Since previous study showed the core-dependent recruitment of NS5A to LD-associated membranes [53], it is possible that NS5A self-interaction is critical for its interaction with core, which then promotes NS5A localization to LD-associated membranes. Alternatively, NS5A self-interaction promoted the NS5A localization to LD-associated membranes, consequently enhancing the interaction between NS5A and core at these membranes. Interestingly, all three NS5A mutants defective in self-interaction also reduced the core localization to the LD (Fig 9D). These results are consistent with recent study by Yin et al., which showed the reduced core localization to the LD by using other NS5A mutants (V67A or P145A), also defective in NS5A self-interaction [42].
To better understand the role of NS5A self-interaction in HCV replication, we attempted to isolate revertants with primary- or second-site mutations that could rescue viral replication. This was done by continuously sub-culturing the Huh-7 cells electroporated with H77D or HJ3-5 encoding different NS5A interface mutations and monitoring the viral replication every 3 days. Among mutants that showed no evidence of transient replication, including 36A and 112A mutants in H77D or HJ3-5 background, only the HJ3-5/36A mutant showed strong evidence of viral replication and infectious virus production by day 10 post electroporation of this viral RNA to Huh-7 cells. Sequencing of the entire coding region of secreted HJ3-5/36A-derived revertant collected at 28 days post-electroporation cell culture supernatants reveled a single mutation in JFH-1 NS5A at residue position 36 to valine (36V). This was not a wt reversion since the wt JFH1 NS5A residue in this position is a phenylalanine.
Since, JFH1 NS5A/36A was defective in NS5A self-interaction (Fig 5A) and NS5A-CypA interaction (Fig 7C), which, we believe, has caused defective HJ3-5/36A replication (Fig 5B and 5C), we asked whether the 36V mutation is capable of restoring all of these defects associated with NS5A/36A mutation. To answer this question, we determined the effect of 36V mutation on the efficiency of JFH1 NS5A self-interaction and NS5A-CypA interaction by using checkmate assays as described above. As shown in Fig 10A and 10B, the 36V mutation in JFH1 NS5A significantly restored the 36A mutation-mediated impairments in NS5A self-interaction and NS5A-CypA interaction. Next, to verify that the 36V mutation in JFH1 NS5A was indeed responsible for the emergence of a replicable revertant from HJ3-5/36A mutant, we introduced the 36V mutation to HJ3-5. The results shown in Fig 10C to 10F indicate that HJ3-5/36V substantially restored viral RNA replication and virus production. HJ3-5/36V also showed reduced virus secretion compared to wt HJ3-5 (Fig 10E and 10G) without affecting virus assembly efficiency (Fig 10H). It is also notable that NS5A hyperphosphorylation in the HJ3-5/36V mutant was significantly lower than that in wt HJ3-5, probably due to only a partial restoration of NS5A self-interaction by a 36V mutation in NS5A (Fig 10A and 10C). In fact, all of these replication phenotypes of HJ3-5/36V revertant strikingly resemble those of HJ3-5/37A and HJ3-5/38A mutants. Overall, these results verified the concept that JFH1 NS5A self-interaction is critical for the NS5A-CypA interaction and NS5A hyperphosphorylation, which contribute to efficient HCV RNA replication and virus secretion.
Next, we determined the density and specific infectivity of extracellular HJ3-5/36V. Overall, density profiles of viral RNA and infectivity of this mutant closely resembled to other replication-competent, NS5A self-interaction defective NS5A mutants, including 37A, 38A and 148A mutants (compare Figs 6 and 11), reflecting the fact that 36V mutant is a partial revertant showing significantly lower NS5A self-interaction than wt (Fig 10A). Accordingly, similar to above three mutants, the relative titers of low-density (<1.059 g/cm3) 36V revertant were ~10 folds higher than those of the wt HJ3-5 (compare Figs 6B and 11B right panels). Due to this, the specific infectivity of 36V mutant viruses at low-density fractions was also higher than wt HJ3-5 (Fig 11E), similar to other mutants. Interestingly, however, proportions of 36V mutant RNA and infectivity, respectively, detected at high density fractions (>1.201 g/cm3) were much higher than those of other mutants, but comparable to those of wt HJ3-5 (compare Fig 6B and 6D and Fig 11B and 11D, right panels, see below and also Discussion).
To determine the effect of 36V mutation on NS5A subcellular localization, we introduced this mutation to HJ3-5/NS5AYFP, and then confirmed that HJ3-5/NS5AYFP /36V mutant is defective in NS5A hyperphosphorylation and virus production compared to wt, similar to the phenotypes of HJ3-5/36V (Fig 12A and 12B). Confocal imaging analysis revealed that a majority of NS5AYFP/36V was detected as the distinct foci in the cytoplasm without the tight LD association (Fig 12C), which is consistent with low degree of NS5AYFP and LD co-localization based on Pearson’s correlation measurements (Fig 12D). Also we detected reduced degree of NS5AYFP and core co-localization as well as core-LD association (Fig 12F and 12H). These NS5AYFP/36V phenotypes resembled those of other NS5AYFP mutants shown in Figs 8 and 9, probably because they shared a common defect in NS5A hyperphosphorylation, which was shown to facilitate NS5A localization to the LD as well as NS5A and core co-localization [17]. However, uniquely to 36V mutant, we consistently detected small fraction of large NS5A foci, co-localizing with LD and core (Fig 12C and 12E). Also, we detected wt level pull-down of core by NS5AYFP/36V (Fig 12G), which, apparently, is contradictory to the reduced degree of NS5AYFP and core co-localization detected from this mutant compared to wt (Fig 12F). Based on these data, we propose that 36V mutation in NS5A may have enhanced its affinity to core, partially compensating its LD localization defect caused by its defective hyperphosphorylation, resulting in fraction of NS5A recruitment to LD in core-NS5A interaction dependent manner. We speculate that LD-localized NS5A/36V may have contributed to near wt level of high-density particles detected from HJ3-5/36V mutant (Fig 11B and 11D, right panels). However, further study will be needed to verify this point.
HCV NS5A is a multifunctional protein involved in both viral RNA replication and virus production [24, 54, 55]. Our study provided mechanistic insights regarding the roles of, crystal structure-defined, NS5A dimer interface residues in these two critical steps in the HCV life cycle. Specifically, our data revealed that these residues regulate NS5A self-interaction, NS5A-CypA interaction, NS5A hyperphosphorylation, NS5A localization to LD and NS5A-core interaction, promoting HCV replication and infectious HCV production.
Among three domains of NS5A, DI plays a major role in NS5A self-interaction [39]. Currently three independent crystal structures of NS5A-DI, two from gt1b (Con1 strain) and one from gt1a (H77 strain), are available [36–38]. While their monomeric structures were similar to each other with an average Cα RMSD (root-mean-square deviation) equal or less than 1 Å [37, 38], four distinct dimeric forms were detected in different crystal packing conditions, including the two forms from gt1b shown in Fig 1 [36–38]. Interestingly, these NS5A dimeric forms are not mutually exclusive but have a potential to form NS5A oligomers via multiple different interfaces [37, 38]. Our data shown in Fig 2B support this possibility, since combining mutations located at two different interfaces additively reduced the level of NS5A self-interaction. Importantly, the fact that NS5A mutations located at different dimeric interfaces exhibited similar phenotypes, including their effects on NS5A hyperphosphorylation, NS5A-CypA interaction and NS5A subcellular localization, and, consequently, viral RNA replication and virus assembly/egress (Figs 3 to 9), strongly suggests the cooperative roles of different dimeric interactions within the same complexes. From a functional point of view, NS5A oligomerization is desirable for its role in promoting the formation of DMV [31, 32], which are sites of HCV RNA replication. In addition, the NS5A oligomerization model may be the best way to explain the high potency of NS5A inhibitors, which corresponds to one molecule of NS5A inhibitor impacting ~ 50,000 molecules of NS5A as in the daclatasvir example [56], and a synergistic activity of different NS5A inhibitors in re-sensitization of drug-resistant NS5A variants [56].
The structure of gt2a JFH1 NS5A is currently unknown. However, a substantial NS5A-DI sequence difference exists between gt2a JFH1 and gt1b Con1 (69% amino acid homology). Thus, it was remarkable that two-to-five mutations introduced to JFH1 NS5A-DI residues located at positions corresponding to two different gt1b NS5A-DI dimer-interfaces significantly inhibited its self-interaction at the levels similar to those of gt1a H77 NS5A-DI (compare Fig 2B and 2C), since gt1a H77 NS5A-DI is more homologous to gt1b NS5A-DI in sequence (82% amino acid homology) and structure (average Cα RMSD of 0.57 Å) [36–38]. Interestingly, the impacts of individual NS5A-DI dimer-interface mutations on NS5A self-interaction were different between H77 and JFH1-derived NS5A (Figs 3A and 5A). This difference was most apparent for R112A mutation, since this mutation severely impaired H77 NS5A self-interaction, while same mutation had no effect on JFH1 NS5A self-interaction. These results suggest that some intermolecular NS5A residue interactions might vary in these two HCV isolates due to differences in near neighbor residues. Supporting this interpretation, our preliminary data indicate that R112 residue in JFH1 NS5A may not participate in NS5A self-interaction, since neither (similarly charged) R112K nor (oppositely charged) R112E mutations affected this interaction (S2 Fig). On the contrary, JFH1 NS5A self-interaction was severely impaired by E148R mutation, but unaffected by E148D mutation (S2 Fig). These results suggest that E148 residue in JFH1 NS5A is involved in NS5A self-interaction via salt bridge formation, similar to E148 residue in H77 NS5A, but with different residue(s) instead of R112. However, JFH1 NS5A-DI structure determination will be necessary to identify the exact residues involved in NS5A self-interactions at its dimer interface(s).
The NS5A mutation-mediated impairment in NS5A self-interaction correlated with HCV RNA replication-defects driven by either H77 NS5A or JFH1 NS5A-containing viral replicases (Figs 3 and 5). These results suggest that NS5A self-interaction is critical for HCV RNA replication regardless of HCV genotypes. Interestingly, we detected a strong correlation between NS5A self-interaction and NS5A-CypA interaction that was shown to be critical for HCV RNA replication [52] (Figs 3, 5 and 7). Since the mutations we tested are located at NS5A-DI rather than at NS5A-DII that encodes the CypA interacting domain [57, 58], a direct role of these mutations in disrupting NS5A-CypA interaction is unlikely. Also, two mutations in NS5A (D316E/Y317N), which conferred the CypA-independent HCV replication phenotype [58], did not affect H77 NS5A self-interaction and slightly reduced JFH1-NS5A interaction (S3A Fig). These results support that NS5A self-interaction is driving NS5A-CypA interaction, and not vice versa. Interestingly, NS5A-CypA interaction was detected only from GAL4/BD-CypA and VP16/AD-NS5A pairs and not in reverse configuration (S4 Fig). We believe that these data support our interpretation that NS5A oligomers may interact with CypA, since we could easily envision a soluble VP16/AD-NS5A, not a DNA-bound GAL4BD-NS5A, forming a CypA-binding-competent oligomer. Now, how could NS5A self-interaction affect NS5A-CypA interaction? We propose that NS5A self-interaction may modulate the orientation of CypA-binding region in NS5A-DII, likely within the context of oligomeric NS5A structure, allowing efficient CypA binding.
Previous study by Lim et al. showed that mutating zinc-binding cysteines to alanine (C to A) within NS5A-DI (C39A, C57A, C59A and C80A) disrupted NS5A self-interaction and HCV RNA replication by using bacterially expressed and purified proteins [20]. These data support the role of NS5A self-interaction in HCV replication. However, NS5A-CypA interaction was not affected by any of eleven C to A mutations within full length NS5A (including the four zinc binding residues mentioned above), regardless of their impact on NS5A self-interaction [20]. These findings are different from our data, which showed positive correlation between NS5A self-interaction and NS5A-CypA interaction. We speculate that potential difference in NS5A oligomeric states between current and previous experimental systems might have altered the availability of CypA-interacting region in NS5A-DII for CypA binding leading to different outcomes.
Our data indicated that impaired gt1a H77 NS5A self-interaction resulted in virus particle assembly defects (Fig 3G), while that of gt2a JFH1 NS5A resulted in virus secretion defects (Fig 5F). Although this genotypic difference in regard to assembly versus secretion is difficult to understand, we could envision that more than 10 folds higher viral replication from HJ3-5 (encoding JFH1 NS5A), compared to H77D (encoding H77 NS5A), could have altered the relative roles of NS5A self-interactions in viral assembly/egress processes. Alternatively, H77 NS5A and JFH1 NS5A contributed to HCV assembly/egress via intrinsically different mechanisms. Regardless, this genotypic difference was not due to the chimeric nature of HJ3-5, which encode H77 core to NS2 in the background of JFH1, since NS5A dimer interface mutations introduced to full length JFH1/QL showed exactly same virus secretion defect as did HJ3-5 mutants (S5 Fig). Interestingly, while H77 NS5A mutants showed no effect on virus density or specific infectivity (Fig 4), JFH1 NS5A mutants increased the proportion of low-density, high-infectivity particles (Fig 6). It seems paradoxical that the specific infectivity of JFH1 NS5A mutants was higher than that of wt considering the reduced overall infectivity of mutants. However, it is possible that slowed viral egress in JFH1 NS5A mutants allowed their enhanced lipidation, resulting in low-density, highly infectious viruses, while this type of slow maturation of wt virus was relatively decreased due to efficient virus secretion (Fig 5D and 5F). Interestingly, LD localization of JFH1 NS5A dimer interface mutants was reduced, indicating that virus maturation into low-density particles may not strictly depend on a tight association of NS5A with the cytoplasmic LD. Alternatively, efficient LD localization of NS5A in wt HJ3-5-replicating cells could have enhanced virus secretion at a level to over-saturate the cells’ capacity for normal virus maturation, consequently, forcing the significant portion of viral particles to quickly egress as immature forms (Fig 6).
It is important to note that not all viral RNA secreted to the supernatant of HCV replicating cells is associated with infectious virus. In fact, Gastaminza et al. showed that HCV replicating cells released the low-density particles, including the exosome-like large vesicles, and high density particles, most likely representing non-enveloped core particles, in addition to majority of intermediate density particles corresponding to enveloped HCV particles [59]. Accordingly, low infectivity (relative to viral RNA) of both high- and low-density particles detected in our study could be attributed to these defective-particles, including exosomes and non-enveloped core particles. In this context, it is interesting to note that relative proportion of secreted, minimally infectious, high density viral particles from 36V revertant was significantly higher than those from 37A and 38A mutants (compare Figs 6 and 11), despite that most of other phenotypes of 36V revertant were similar to those of 37A and 38A mutants consistent with their similarly defective NS5A self-interaction. The second phenotypic difference between 36V revertant and these other mutants was the affinity of core and JFH1 NS5A, which was unchanged in the 36V revertant, but reduced in the 37A and 38A mutants, compared to that of wt (compare Figs 9C and 12G). Based on these data, we propose that high-affinity interaction between core and NS5A, independent from NS5A self-interaction, could promote the secretion of defective, high-density core particles.
Previously Miyanari et al. demonstrated that two different triple alanine mutations introduced to the residues 99–101 and 102–104 within NS5A-DI reduced NS5A localization to LD, establishing the role of NS5A-DI on NS5A localization to the LD [53]. Subsequently, Masaki et al. showed that NS5A hyperphosphorylation promote its LD localization [17]. Now, our data suggest that NS5A dimer-interface residues in NS5A-DI contribute to NS5A localization to the LD by regulating NS5A hyperphosphorylation (Figs 5 and 8). Interestingly, our extended study by using HCV polyprotein expression system indicate that all of NS5A dimer interface mutants that we tested, including 36A, 37A, 38A, 112A and 148A, regardless of their impact on NS5A self-interaction, could impair JFH1 NS5A hyperphosphorylation (S6 Fig). These data suggest that NS5A self-interaction per se may not be sufficient to promote its hyperphosphorylation. A recent study by Ross-Thriepland and Harris [23] indicated that JFH1 NS5A-DI residue 146 serine (146S) is a target of phosphorylation and mutating this residue to phosphomimetic aspartic acid (146D) led to decreased NS5A hyperphosphorylation. Since 146S is located near the 3FQM dimer interface in the vicinity of the 112R-148E salt-bridge, the authors predicted that phosphorylation at 146S might potentially regulate NS5A dimerization [23]. However, our data showed that 146A or 146D mutations did not significantly affect H77 or JFH1 NS5A self-interaction (S3B Fig). These data indicate that 146D mutation impaired NS5A hyperphosphorylation without affecting NS5A self-interaction similar to the phenotypes of 112A mutation (Fig 5A and S6 Fig). Consistent with these data, the NS5A inhibitors also reduced NS5A hyperphosphorylation [60], yet they did not affect NS5A self-interaction [40]. Intriguingly, NS5A inhibitors were implicated to affect intermolecular NS5A conformation [56, 61], and phosphorylation of NS5A residue 146 has a potential to alter local dimeric conformation [23]. Thus, these data may indicate that hyperphosphorylation of NS5A is dependent on its specific conformation that allows its interaction with kinases [such as casein kinase I-α (CKI-α)] involved in this process [17]. Accordingly, we propose that only a defined conformation of NS5A, mostly likely within the oligomeric complexes, permits the access of kinases to NS5A LCSI domain resulting NS5A hyperphosphorylation [17, 62], and disturbing the kinase-accessible conformation of NS5A either by different NS5A mutations or treatment with NS5A inhibitors impairs NS5A hyperphosphorylation.
It is possible that all or some of our mutants may have impacted HCV RNA replication and virus assembly by altering NS5A conformation or other functions of NS5A, in addition to altering NS5A self-interaction-mediated functions. However, the ultimate proof supporting the role of NS5A self-interaction in HCV RNA replication and virus production was provided by a 36 valine (36V) revertant mutation in JFH1 NS5A, which replaced the original alanine mutation that conferred a severe defect in JFH1 NS5A self-interaction. This JFH1 NS5A/36V mutation, not only significantly restored the NS5A self-interaction, but also restored NS5A-CypA interaction, and HJ3-5/36V RNA replication and infectious virus production. Interestingly, 36V mutation introduced to H77 NS5A also partially enhanced NS5A self-interaction as well as an NS5A-CypA interaction (S7 Fig). These data support the notion that 36V mutation in JFH1 NS5A was indeed selected to rescue the NS5A self-interaction-defect caused by 36A mutation. However, the H77D/36V mutant did not show detectable level of viral RNA replication. The exact reason for impaired replication of H77D/36V is unclear. However, it is interesting to note that H77 NS5A residue F36 points toward lipid phase in the 1ZH1 structure (Fig 1A), suggesting that F36 may contribute to NS5A and membrane interaction. Based on this, we speculate that 36V mutation may have failed to restore the F36’s additional function at the NS5A-membrane interface and, as a consequence, could not rescue H77D replication. This potential, additional role of F36 residue in H77D replication may also explain the reason H77D/36A mutant could not replicate when 148A mutant showed low level replication (Fig 3C), despite their similar NS5A self-interaction defects (Fig 3A).
As illustrated in Fig 13, we propose that high-order NS5A oligomerization stabilizes NS5A conformation at local domains, including the DII and LCSI, which would allow them to interact with host proteins including CypA [63, 64] and CKI-α [17, 65]. The CypA-induced modulation of NS5A would then promote DMV formation, which will harbor HCV replication complexes [19, 30–32]. This early event will be followed by NS5A hyperphosphorylation by CKI-α [17, 22, 66], which would be promoted by interaction between NS5A and other HCV nonstructural proteins in the replication complexes as demonstrated in previous reports [67, 68]. Then the hyperphosphorylated NS5A (in the replication complexes) will move to low-density membrane domains near LD [53], interact with core protein, and promote HCV assembly and/or virus egress [17]. A recent pulse-chase imaging study by Wang and Tai strongly supports this scenario, since they determined that the NS5A-associated organelles (replication complexes) are continuously generated de novo and NS5A in the aged version of these organelles tend to associate with LD, accompanied with increases in NS5A hyperphosphorylation [66].
In conclusion, our study provided novel insights indicating that NS5A may function as oligomers formed via multiple dimeric interfaces to promote HCV RNA replication and virus production. It is likely that NS5A functions requiring its oligomerization make it an excellent target of highly potent inhibitors [29]. Although NS5A inhibitors did not perturb NS5A dimerization per se [20, 40, 69], it is suggested that they might have modulated its conformation [70] or higher-order NS5A oligomerization state [56]. In the future, understanding the detailed mechanistic function of NS5A oligomers during HCV replication, combined with determining the exact mode of action of NS5A inhibitors, will provide insights into understanding HCV replication mechanisms and for improving/identifying potent antivirals against HCV and other agents that utilize proteins functioning in an equivalent manner.
The construction of H77D and HJ3-5 was described previously [44, 48]. To generate the pairs of pACT and pBind vectors expressing full-length NS5A from two different genotypes, H77 and JFH1 NS5A sequences were PCR amplified from H77D and HJ3-5 with the primer sets introducing SgfI and PmeI restriction enzyme sites at their N- and C-terminus, respectively, and then cloned into pFN10A(ACT) Flexi vector and pFN11A(BIND) Flexi Vector digested with SgfI and PmeI enzymes (Promega, WI, USA). NS5A mutations were introduced by using the QuikChange II XL site-directed mutagenesis kit (Agilent Technology, Santa Clara, CA). The sequences of regions manipulated within each plasmid were verified by DNA sequencing. Other plasmids used for vector control (pACT and pBind), positive controls (pACT-MyoD and pBind-ID) and pGL4.3[luc2P/Gal4UAS/Hygro] were provided via a Checkmate/Flexi Mammalian Two-Hybrid System (Promega, WI, USA).
Huh-7 cell lines used in this study including Huh7.5 (a clonal cell line of Huh-7, kindly provided by Dr. Charles M. Rice at Rockefeller University [71]) and FT3-7 (a clonal cell line of Huh-7 as described in [72]) were maintained in Dulbecco's modified Eagle medium (DMEM) (Invitrogen, Carlsbad, CA) containing 10% fetal bovine serum (Invitrogen, Carlsbad, CA) at 37°C in a 5% CO2 atmosphere.
The interactions between NS5A-NS5A or NS5A-CypA were evaluated by using a Checkmate Mammalian Two-Hybrid System (Promega, WI, USA). In brief, pACT and pBIND based plasmids along with the pGL4.3[luc2P/Gal4UAS/Hygro] reporter plasmid were co-transfected into FT3-7 cells by using TransIT-LT1 (Mirus, Madison, WI) reagent according to the manufacturer’s instructions at a ratio of 3 μl transfection reagent per 1 μg of plasmid DNA. At 48 h post transfection, cell lysates were prepared to assess the firefly and Renilla luciferase activities by using a Dual-Luciferase Reporter (DLR) Assay System (Promega, WI, USA) and GloMax DISCOVER instrument (Promega, WI, USA) according to the manufacturer’s instructions.
HCV RNA was transcribed in vitro from linearized HCV cDNA by using the T7 MEGAscript Kit (Life Technologies, Carlsbad, CA) and purified by using an RNeasy RNA isolation kit (Qiagen, Valencia, CA). In brief, 5 x 106 FT3-7 cells were mixed with 10 μg of HCV RNA in a 4-mm cuvette and pulsed once at 270 V and 950 μF by using a Gene Pulser System (Bio-Rad, Hercules, CA). Electroporated cells were transferred into 12-well plates for HCV RNA analysis and or 6-well plates or 6 cm dishes for virus titration and protein analysis.
Extracellular and intracellular HCV titers in clarified cell culture supernatants and 4-cycle, freeze-thaw cell lysates harvested from FT3-7 cells at different time points post-electroporation of HCV RNA were determined by performing an HCV core antigen immunfluorescence assay as described before [73].
HCV RNA in cell culture supernatants and gradient fractions (see below-density gradient ultracentrifugation) was harvested by using a QIAamp viral RNA mini kit (Qiagen, Valencia, CA). Cell-associated HCV RNA was harvested by using an RNeasy RNA isolation kit (Qiagen, Valencia, CA). To quantitate the level of HCV RNA, a real-time RT-PCR assay was performed by using a QuantiNova Probe RT-PCR Kit (Qiagen, Valencia, CA) and a CFX96 real-time system (Bio-Rad, Hercules, CA) with custom designed primer probe sets (Sense primer: HCV84FP, 5’-GCCATGGCGTTAGTATGAGTGT-3’; antisense primer: HCV 303RP, 5’-CGCCCTATCAGGCAGTACCACAA-3’; and probe: HCV146BHQ, FAM-TCTGCGGAACCGGTGAGTACACC-DBH1). Briefly, 10 μl 2x QuantiNova Probe RT-PCR Master Mix, 1 μl each of 20 μM sense- and antisense primers, 0.16 μl of 20 μM HCV-specific probe, 0.2 μl of 100x QuantiNova Probe RT Mix, 4 μl of template RNA and RNase-free water were combined to make 20 μl reaction mixtures. HCV RNA was reverse transcribed for 10 min at 45°C followed by 5 min incubation at 95°C to activate PCR polymerase, then PCR was performed for 30 cycles of 95°C for 5 seconds (denaturation) and 60°C for 30 seconds (annealing and extension).
Cell lysates were prepared in 1% CHAPS in PBS lysis buffer containing 1x protease- and phosphatase inhibitor cocktail mix (GenDEPOT, Katy, TX), separated by SDS-PAGE and transferred onto PVDF membranes. The membrane was blocked and probed with primary antibodies to core protein (1:2,000 dilution of C7-50, Thermo Scientific, Rockford, IL), NS3 (1:2,000 dilution of 9-G2, ViroGen, Watertown, MA), NS5A [1:15,000 dilution of 9E10 (kindly provided by Dr. Charles M. Rice at Rockefeller University) or 1:2000 dilution of 2F6, BioFront Technologies, Tallahassee, FL], and NS5AYFP (1: 2000 dilution of anti-GFP, Life Technologies) and tubulin (1:7000 dilution, EMD Millipore, Billerica, MA). Protein bands were visualized by incubating the membranes with IRDye Secondary antibodies (Li-Cor Biosciences, Lincoln, NE), followed by imaging with an Odyssey infrared imaging system (Li-Cor Biosciences, Lincoln, NE).
Approximately 1.5 x 107 FT3-7 cells were electroporated with in vitro-transcribed RNAs and seeded into 175cm2 flasks. The HCV containing cull culture supernatants were collected from 48 to 72 h for every 4–6 h, pooled and centrifuged to remove cell debris. The clarified supernatants were loaded onto Centricon Plus-70 (Millipore, Germany), concentrated by centrifugation at 3,500 x g at 4°C and subjected to discontinuous Optiprep gradient centrifugation (60, 45, 30, and 15%) for 16h at 120,000 x g at 4°C in a SW55Ti rotor (Beckman, Indianapolis, IN). Each of 450 μl fraction was collected by aspiration from the top of the gradient and analyzed to determine its density, infectivity and amounts of HCV RNA as described above (see also [12]).
Cell lysates were prepared in 1 ml of lysis buffer [0.5% Triton X-100, 10mM Tris-HCl (pH 7.5), 150mM NaCl] containing 1x protease- and phosphatase inhibitor cocktail mix (GenDEPOT, Katy, TX) and incubated on ice for 1h. Cell lysates were incubated with anti-GFP magnetic beads (Miltenyi Biotech, Auburn, CA) for 1h at 4°C with gentle mixing and applied to μ columns. Magnetic beads were washed 4 times each with lysis buffer, and wash buffer I (150mM NaCl,1%NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 50mM Tris-HCl, pH 8.0), respectively, followed by one wash with buffer 2 (20mM Tris-HCl, pH 7.5). Bound immune complexes were eluted from columns by applying a preheated SDS sample buffer.
HCV RNA electroporated cells were plated on 8-well chamber slides (BD Bioscience, Bedford, MA) at a density of 1x104 cells per well. Two to three days later, the slides were washed with PBS, fixed with 4% formaldehyde for 20 min at room temperature, and permeabilized with 0.2% Triton X-100 in PBS for 10 min, then incubated overnight at 4 oC with anti-core monoclonal antibody (1:2,000 dilution of C7-50, Thermo Scientific, Rockford, IL), followed by Alexa Fluor 405-conjugated goat anti-mouse antibody (1:1000 dilution, Invitrogen, Carlsbad, CA) for 1 h. Lipid droplets were stained with HCS LipidTOX deep red neutral lipid stain (1:1000 dilution, Molecular Probes Inc, Eugene, OR). The slides were examined with an Olympus FluoView FV10i confocal microscope (Olympus America Inc, Waltham, MA). Pearson’s coefficient was obtained by using FV10i-ASW 4.2 viewer software.
Student’s t-test (unpaired) was performed by using GraphPad Prism version 6 software to determine the significance in differences between paired values. A P value less than 0.05 was considered statistically significant.
|
10.1371/journal.pcbi.1003909 | Deducing the Kinetics of Protein Synthesis In Vivo from the Transition Rates Measured In Vitro | The molecular machinery of life relies on complex multistep processes that involve numerous individual transitions, such as molecular association and dissociation steps, chemical reactions, and mechanical movements. The corresponding transition rates can be typically measured in vitro but not in vivo. Here, we develop a general method to deduce the in-vivo rates from their in-vitro values. The method has two basic components. First, we introduce the kinetic distance, a new concept by which we can quantitatively compare the kinetics of a multistep process in different environments. The kinetic distance depends logarithmically on the transition rates and can be interpreted in terms of the underlying free energy barriers. Second, we minimize the kinetic distance between the in-vitro and the in-vivo process, imposing the constraint that the deduced rates reproduce a known global property such as the overall in-vivo speed. In order to demonstrate the predictive power of our method, we apply it to protein synthesis by ribosomes, a key process of gene expression. We describe the latter process by a codon-specific Markov model with three reaction pathways, corresponding to the initial binding of cognate, near-cognate, and non-cognate tRNA, for which we determine all individual transition rates in vitro. We then predict the in-vivo rates by the constrained minimization procedure and validate these rates by three independent sets of in-vivo data, obtained for codon-dependent translation speeds, codon-specific translation dynamics, and missense error frequencies. In all cases, we find good agreement between theory and experiment without adjusting any fit parameter. The deduced in-vivo rates lead to smaller error frequencies than the known in-vitro rates, primarily by an improved initial selection of tRNA. The method introduced here is relatively simple from a computational point of view and can be applied to any biomolecular process, for which we have detailed information about the in-vitro kinetics.
| The proverb ‘life is motion’ also applies to the molecular scale. Indeed, if we looked into any living cell with molecular resolution, we would observe a large variety of highly dynamic processes. One particularly striking aspect of these dynamics is that all macromolecules within the cell are continuously synthesized, modified, and degraded by complex biomolecular machines. These ‘nanorobots’ follow intricate reaction pathways that form networks of molecular transitions or transformation steps. Each of these steps is stochastic and takes, on average, a certain amount of time. A fundamentally important question is how these individual step times or the corresponding transition rates determine the overall speed of the process in the cell. This question is difficult to answer, however, because the step times can only be measured in vitro but not in vivo. Here, we develop a general computational method by which one can deduce the individual step times in vivo from their in-vitro values. In order to demonstrate the predictive power of our method, we apply it to protein synthesis by ribosomes, a key process of gene expression, and validate the deduced step times by three independent sets of in-vivo data.
| Life is based on the continuous synthesis, modification, and degradation of proteins and other macromolecules. These processes are performed by complex biomolecular machines that bind their ligands and transform them into product molecules. Examples are provided by the transcription of DNA by RNA polymerases, the translation of mRNA by ribosomes, or the degradation of proteins by proteasomes. Each of these processes involves several steps: the binding of the ligand molecules, chemical reactions catalyzed at the active sites, as well as specific conformational changes and directed mechanical movements of parts of the molecular machinery. In principle, the kinetics of such multistep processes can be understood in terms of the individual transitions and the associated transition rates, a well-established approach both for enzyme kinetics [1]–[3] and for free energy transduction by molecular motors [4], [5]. In practice, the values of the individual transition rates can be typically measured in vitro but not in vivo, and the in-vitro rates depend on the composition of the buffer. Because the cytosol represents a rather complex buffer, it is difficult to assess whether a certain in-vitro assay provides a reliable description of the process in vivo. One important tool that is missing for such an assessment is a simple measure by which we can quantitatively compare the kinetics of a multistep process in different environments.
Here, we develop a general method that provides such a measure and allows the deduction of the in-vivo rates from their in-vitro values. Our method has two basic components. First, we introduce the ‘kinetic distance’, i.e., a distance metric for the kinetics, by which we can describe the similarity or dissimilarity of multistep processes in vitro and in vivo in a quantitative manner. The kinetic distance depends logarithmically on the rates and has an intuitive interpretation in terms of the associated free energy barriers. Second, we minimize the kinetic distance between the in-vitro and in-vivo processes, imposing the constraint that the deduced rates reproduce a known global property such as the overall in-vivo speed. Computationally, this constraint defines a hypersurface in the multi-dimensional space of transition rates. In order to demonstrate the predictive power of our method, we apply it to the elongation cycle of protein synthesis, a key process of gene expression.
In all living cells, proteins are synthesized by ribosomes, which translate the codon sequences of mRNA into peptide chains of proteins. During the elongation cycle of this process, the ribosome translates one codon after another by binding a ternary complex consisting of aminoacyl-tRNA (aa-tRNA), elongation factor Tu (EF-Tu) and GTP. The amino acid is transferred from the tRNA to the nascent peptide chain, and the ribosome moves to the next codon with the help of elongation factor G (EF-G) allowing for the next elongation cycle. Translation elongation involves several individual states with rapid transitions between them [6], [7]. The different states have been studied by a variety of experimental techniques: chemical probing methods [8], pre-steady state kinetics [9]–[15], electron microscopy [16]–[19], X-ray crystallography [20]–[22], and single molecule methods [23], [24]. The kinetic measurements in vitro provided values for the individual transition rates but, so far, it has not been possible to measure the corresponding rates in the cell.
The different states and transitions of the elongation cycle are schematically shown in Fig. 1. When the ribosome dwells at a certain codon and binds a ternary complex, the tRNA within this complex can be cognate, near-cognate, or non-cognate to the codon, which implies that the elongation cycle contains three different reaction pathways corresponding to the three branches in Fig. 1. During each round of elongation, the ribosome typically explores all three pathways in order to select a cognate tRNA and to reject the near-cognate and non-cognate ones. The individual rates of these pathways were measured in vitro at 20°C and/or 37°C using the ribosomes and translation factors from Escherichia coli [6], [13], [25]. Here, we combine these results with new data on the overall elongation rates in vitro to first derive a complete set of individual in-vitro rates at both temperatures. We then minimize the kinetic distance between the in-vitro and in-vivo processes, taking into account two known properties of the in-vivo process: the overall elongation rates [26], [27] and the tRNA concentrations [28], both of which have been measured in E. coli for different growth conditions.
The in-vivo rates of the elongation cycle obtained in this way are then validated by three independent sets of in-vivo data [29]–[31]. First, we compute codon-specific elongation rates and show that these rates correlate well with relative translation rates as obtained experimentally by [29]. Second, we predict the time-dependent incorporation of radioactively labeled amino acids into proteins as studied in vivo by [30]. The time course of synthesis obtained theoretically is in excellent agreement with the experimental data. Third, using the same in-vivo rates, we also compute the missense error frequency and obtain good agreement with the experimental results of [31]. In all three cases, our computations do not involve any fit parameter and, thus, directly validate the derived set of in-vivo rates.
In order to introduce a quantitative measure for the (dis)similarity of the in-vivo and in-vitro kinetics, we consider a generic multistep process within the cell and first focus on one of the individual transitions from state to state . The corresponding transition rates have the values and for a certain in-vitro assay and for specific in-vivo growth conditions, respectively. Instead of the rates, we can equally well consider the associated transition times and . Thus, we require that the distance between the rates and is equal to the distance between the times and , i.e., that(1)
The simplest expression for that fulfills this requirement is provided by(2)with the logarithmic difference(3)between the in-vitro and the in-vivo value of the individual transition rate.
The single transition distance is dimensionless and does not involve any parameter apart from the two rates and . In addition, this distance satisfies the two scaling relations and for any rescaling factor . The first scaling relation implies that and have the same distance from , which agrees with our intuition. The second scaling relation implies that the distance does not depend on the units used to measure the rates. For small deviations of from , which are equivalent to small deviations of from , the distance becomes asymptotically equal to both and .
The in-vitro rate can be expressed in terms of the activation free energy or free energy barrier and the attempt frequency which leads to(4)where the thermal energy provides the basic free energy scale. When we combine this expression with the analogous expression for the in-vivo rate, the logarithmic difference between the two rates becomes
(5)Because the prefactors and are expected to have the same order of magnitude, the second term should usually be small compared to the first term which represents the shift of the free energy barrier between state and state , see Fig. 2A. Therefore, for each individual transition along one of the reaction pathways, the logarithmic difference can be interpreted as the shift of the free energy barrier that governs the transition from state to state . In the following, we will use the intuitive terminology ‘single barrier shift’ for the quantity . It should be noted, however, that, in spite of this terminology, changes in the attempt frequency as described by the term in Eq. 5 are included in the logarithmic difference and, thus, will be taken into account in all our calculations.
Next, we consider all individual transitions along the reaction pathways of the multistep process and regard the associated in-vivo rates as unknown variables that can be visualized as the coordinates of a multi-dimensional space. These coordinates are somewhat impractical, however, because they are restricted to positive values. In order to eliminate this restriction, we perform a coordinate transformation from the in-vivo rates to the single barrier shifts , which can attain both positive and negative values. This coordinate transformation is highly nonlinear but invertible with the inverse transformation given by .
The overall distance between the in-vitro and the in-vivo kinetics is now defined by(6)where the summation under the square root runs over all individual transitions along the reaction pathways. As illustrated in Fig. 2B, the distance represents the Euclidean distance within the multi-dimensional space defined by the single barrier shifts . Therefore, the distance provides a genuine metric in the mathematical sense, which implies that it satisfies the triangle inequality if we compare three different in-vitro and/or in-vivo conditions.
If all in-vivo rates are identical to their in-vitro values, apart from a single one, , Eq. 6 for the kinetic distance reduces to Eq. 1 for the single transition distance . Because the choice of the two states and is arbitrary, this property of the kinetic distance applies to all individual transitions that enter in Eq. 6. The latter property represents, in fact, a general requirement for any meaningful definition of the kinetic distance. Therefore, if we considered the more general expression with dimensionless weight factors , this requirement would imply that all weight factors must assume the unique values and that must be equal to the kinetic distance as given by Eq. 6.
If we consider two different in-vitro assays, say and , the corresponding transition rates and will, in general, be different and define two sets of single barrier shifts via and(7)with the logarithmic differences . The latter quantities determine the kinetic distance between the two in-vitro assays, which is given by . The two sets of barrier shifts, and , provide two different coordinates for the multi-dimensional barrier space. Because of the linear relations as given by Eq. 7, the primed coordinates are obtained from the unprimed ones by shifting the latter coordinates by the logarithmic differences . Therefore, the transformation from the unprimed to the primed coordinates, corresponding to a change from assay A to assay A′, represents a Euclidean translation of the coordinate system, which preserves the shape of any geometric object within the multi-dimensional barrier space.
Next, we combine the kinetic distance as given by Eq. 6 with a minimization procedure to predict the unknown in-vivo rates from their known in-vitro values. Even though the rates of the individual transitions are difficult to study in the cell, one can usually measure some quantity that characterizes the overall kinetics of the intracellular process. One such quantity is provided by the average speed of the process. Any such global kinetic quantity, , depends on the individual transition rates, . The in-vivo values of the individual transition rates must reproduce the experimentally measured value of the global quantity. This requirement implies the equation(8)which represents a constraint on the unknown in-vivo values . This constraint can be expressed in terms of the single barrier shifts using the inverse coordinate transformation with the known in-vitro values . As a result, the constraint in Eq. 8 defines a hypersurface in the multi-dimensional barrier space as illustrated in Fig. 2B. Each point on this hypersurface is compatible with the measured value of the global kinetic quantity. In addition, the Euclidean distance of such a point from the origin is equal to the kinetic distance between the (unknown) in-vivo and the (known) in-vitro values of the transition rates. Our prediction for the in-vivo values is then obtained by minimizing this kinetic distance, i.e, by the point on the hypersurface that has the shortest distance from the origin. For clarity, the coordinate values of this point will be denoted by in order to distinguish these values from the variable coordinates .
Our approach involves the following assumptions. First, we make the usual assumption that the states of the biomolecular system that have been identified in vitro are also present in vivo. The molecular conformations of the corresponding in-vitro and in-vivo states are expected to be somewhat different when viewed with atomic resolution, but the gross features of these conformations should be similar, in particular when the in-vitro assay is functional and has been optimized. It is then plausible to assume that the in-vitro and in-vivo values of the individual transition rates do not differ by many orders of magnitude, which implies that the point in the multi-dimensional barrier space that represents the true in-vivo rates is located ‘in the neighborhood’ of the origin of this space and, thus, characterized by a ‘small’ kinetic distance . If the kinetic distance satisfied , the true in-vivo point would be located within a sphere of radius around the origin. The smallest sphere that is compatible with the in-vivo constraint as given by Eq. 8 is the one that touches the hypersurface depicted in Fig. 2B, and the radius of this sphere is equal to the Euclidean distance of the hypersurface from the origin of the Δij-coordinates. The associated contact point between -sphere and hypersurface represents the predicted in-vivo point, and its coordinate values lead to the predicted in-vivo rates based on the known in-vitro rates .
For a general, nonlinear in-vivo constraint, the coordinate values of the predicted in-vivo point will be different for different individual transitions. The minimization procedure then predicts different scale factors and, thus, different effects of the in-vivo environment on the individual transitions of the system. Such differences are indeed obtained when we apply our minimization approach to the kinetics of ribosomes as described in the next subsection. It is important to note that this approach leads to different scale factors even though the expression for the kinetic distance (Eq. 6) does not include any bias for one of the individual barrier shifts Therefore, the different scale factors follow from the imposed in-vivo constraint (Eq. 8) alone and do not involve any additional assumptions or expectations about the in-vivo conditions.
The minimization procedure described above represents an extremum principle with constraints. Such principles have been successfully applied in many areas of science, in particular in the context of optimization problems. One important and useful feature of extremum principles is that they provide global solutions for nonlinear systems. Thus, in the present context, we would obtain a prediction for the in-vivo rates even if the in-vitro assay were rather different from the in-vivo conditions. Another advantage of extremum principles is that they typically lead to a unique solution without any additional assumptions (‘principle of least prejudice’). In some exceptional cases, one may find more than one solution, which then indicates that the system undergoes some kind of bifurcation. For the kinetics of ribosomes, see next subsection, we always found a unique solution and, thus, a unique set of predicted in-vivo rates.
The rates of the in-vitro assay are only known with a certain accuracy. As a consequence, the predicted in-vivo rates have some uncertainty as well. As explained in the Methods section, this uncertainty reflects both the accuracy of the measured in-vitro rates and the associated changes in the location of the predicted in-vivo point. Furthermore, the latter location will also depend, in general, on the rates of the chosen in-vitro assay. Indeed, the change from assay to assay corresponds to a Euclidean translation of the coordinate system (Eq. 7) while the shape of the hypersurface (Eq. 8 and Fig. 2B) remains unchanged. These two properties imply that the distance of the hypersurface from the origin of the -coordinates may differ from the distance of this surface from the origin of the -coordinates. Therefore, the validity of the predicted in-vivo rates is difficult to assess a priori, but can be checked a posteriori in a self-consistent manner: we first deduce the unknown in-vivo rates from the known in-vitro rates via the minimization procedure and subsequently validate the deduced rates by calculating some other quantities that have been experimentally studied in vivo. In the next two subsections, we will apply this two-step procedure to the kinetics of ribosome elongation based on the in-vitro assay developed in [6], [25].
Our minimization procedure becomes computationally simpler if we have additional knowledge about some of the in-vivo values of the individual transition rates. If we knew one of these rates, e.g., , we would restrict our minimization procedure to the subspace with constant . As a consequence, we would not vary the coordinate during the minimization and use the constant value of this coordinate in Eq. 6 for the kinetic distance . On the other hand, if we knew only that the in-vivo rate is located within the range , we would minimize the kinetic distance also with respect to but within the subspace defined by . The latter procedure may lead to a boundary minimum, i.e., to a predicted in-vivo point that is located at the boundary of the considered subspace. Another simplification is obtained if the rates of two individual transitions, say from state to state and from state to state , have the same values in vitro and in vivo, i.e., if and . We will then reduce the multi-dimensional barrier space to the subspace with , and the corresponding expression for the kinetic distance in Eq. 6 will now contain the term under the square root. The latter reduction will be used in the next subsection on the kinetics of ribosomes for which different individual transition rates have the same in-vitro values.
Our quantitative description of the translation elongation cycle is based on the codon-specific Markov process displayed in Fig. 3. This process can visit, for each sense codon , twelve ribosomal states, numbered from 0 to 11. After the ribosome has moved to the next sense codon, it dwells in state 0, until it binds a ternary complex with an elongator tRNA that may be cognate, near-cognate, or non-cognate to codon .
The genetic code involves 61 sense codons, which encode 20 proteinogenic amino acids and are decoded by a certain number of elongator tRNAs. The latter number depends on the organism but is always larger than 20 and smaller than 61 [32], [33]. For E. coli, 43 distinct species of elongator tRNA have been identified [28]. The corresponding codon-tRNA relationships can be visualized by the large matrix in Fig. 4 with 61 rows and 43 columns. As shown by the color code in this figure, each sense codon defines a different decomposition of the total set of tRNA species into three subsets of cognates, near-cognates, and non-cognates. The corresponding molar concentrations , , and of cognate, near-cognate, and non-cognate ternary complexes determine the association rates(9)for initial binding with the pseudo-first-order association rate constant . This constant is taken to be independent both of the codon and of the ternary complex as observed in vitro [10], [25]. The latter experiments also imply that all ternary complexes dissociate with the same rate from the initial binding site and that the cognate and near-cognate ternary complexes have the same recognition rate .
After initial binding of a non-cognate ternary complex, this complex dissociates without visiting any other state, so that the ribosome returns back to state 0 with an empty initial binding site. Initial binding of a cognate ternary complex leads to state 1, from which the ternary complex can be released with rate or can move into the A site to attain the codon recognition state 2 with rate . When the ternary complex is recognized as cognate, the ribosome undergoes a forward transition from state 2 to state 3, which corresponds to the combined process of GTPase activation of the cognate ternary complex and GTP hydrolysis, followed by the irreversible transition from state 3 to state 4, which describes phosphate release and conformational rearrangements of EF-Tu [6]. From state 4, the cognate ternary complex may either move to become fully accommodated into the A site via a transition from state 4 to state 5 or, with low probability, may be released from the A site via a transition from state 4 to state 0. After the cognate ternary complex has been fully accommodated, the ribosome/tRNA complex undergoes the final transition from state 5 into the empty state at the next codon . This transition describes the combined process of peptide bond formation and translocation, the corresponding processing rate is denoted by .
Initial binding of a near-cognate ternary complex leads to state 6, from which the ternary complex can be released with rate or move to the codon recognition state 7 with rate . When the ternary complex is recognized as near-cognate, it is rejected and the ribosome undergoes a backward transition from state 7 to state 6, which provides the initial selection step during the decoding process. With low probability, the near-cognate ternary complex undergoes an irreversible transition from state 7 to state 8, corresponding to GTPase activation and GTP hydrolysis, as well as from state 8 to state 9, which describes phosphate release and conformational rearrangements of EF-Tu. From state 9, the near-cognate ternary complex is typically released again via a transition from state 9 to state 0, which provides the proofreading step during decoding. Very rarely, the near-cognate ternary complex is fully accommodated via a transition from state 9 to state 10. After a near-cognate tRNA has been fully accommodated, it is further processed via peptide bond formation and translocation and undergoes the transition from state 10 to state with rate .
Apart from the association rate constant , the kinetics of the elongation cycle then involves 12 different transition rates for the 17 transitions along the cognate, near-cognate, and non-cognate branches of the Markov process. All of these transition rates have been determined in vitro for the high-fidelity buffer developed in [12], [25], [34]. The corresponding in-vitro values are reported in Table 1. A few individual rates were measured at both 20 and 37°C whereas most of these rates were obtained either at 20 or at 37°C. We used a variety of computational methods to obtain complete and consistent sets of individual rates at both temperatures as described in the Methods section. In addition, we measured the overall elongation rate in vitro for a model protein, aa/s for 20°C and aa/s for 37°C (Supporting Figure S1). As explained in the Methods section (Eq. 22), the measured value of the overall elongation rate was then used to compute, for both temperatures, the in-vitro value of the processing rate. The results of these computations are included in Table 1.
To predict the unknown in-vivo rates from the known in-vitro rates , we consider the multi-dimensional space of single barrier shifts as described by the coordinates Because several transition rates of the Markov process considered here have the same values (Fig. 3), we use the resulting equalities for the associated coordinates as given by and to reduce the 17-dimensional barrier space to a 12-dimensional subspace and restrict the minimization procedure of the kinetic distance to this subspace. After this reduction, the latter distance has the explicit form(10)where the sum contains all the remaining transition rates of the Markov process in Fig. 3.
Because the in-vivo experiments are typically performed at 37°C, we use the in-vitro values for the same temperature, see Table 1. Furthermore, we take into account the known in-vivo values of the overall elongation rate at different growth conditions [26], [27]. For each growth condition, the constraint in Eq. 8 now has the explicit form as given by Eq. 23 in the Methods section. As a result of the constrained minimization procedure, we find the in-vivo rates as given in Table 2 and the single barrier shifts displayed in Fig. 5A, where we have again omitted the subscript ‘min’ for notational simplicity.
Starting from the complete set of individual in-vivo rates (Table 2), we computed the codon-specific elongation rates as described in the Methods section (Eq. 21, Supporting Figure S3). We then compared the in-vivo rates calculated for a growth rate of 2.5 dbl/h to relative translation rates as estimated in Ref. [29] based on the frequencies of the measured +1 frameshifting vs. readthrough of different codons. As shown in Fig. 6A, we obtain reasonable overall agreement between both data sets with a Pearson correlation coefficient of 0.56. The deviations reflect both limitations of our model parametrization and uncertainties in the experimental method. First, the calculated elongation rates for CGA, CGC, and CGU appear to be overestimated. These codons are all read by tRNA, which does not form a Watson-Crick base pair with any of its cognate codons because it carries inosine at the wobble position of its anticodon ICG. The corresponding reductions in the transition rates are not included in the parametrization of our model because we use only two different sets of values for these rates, corresponding to an average over all cognate and over all near-cognate ternary complexes, respectively. Second, for the experimental setup in [29], the UUU, UUC, UUG, UCC, and CCC codons, when located between a preceding CUU codon and a subsequent CXX codon, generate potential slippery sequences, which can lead to frameshifting events. The latter events were not considered and, thus, not taken into account by [29], which implies that the frameshifting rates were underestimated and the translation rates were overestimated for the respective codons. When we exclude these two particular sets of codons, we obtain an increased correlation coefficient of 0.73 as shown in Fig. 6A. Thus, the deduced values of the individual transition rates in vivo lead to a reliable description for the majority of codons.
To further validate these deduced values, we used the computed values of the codon-specific elongation rates (Supporting Figure S3), to model the time course of protein synthesis measured by [30]. In those experiments, the lacZ gene was expressed in E. coli at a growth rate of about 0.7 dbl/h, the cells were exposed to a 10-s pulse of radioactively labeled methionine, and the radioactivity of the synthesized proteins was measured over time. The calculated time course is in excellent agreement with the experimental data (Fig. 6B). Furthermore, varying the values of the internal transition rates leads to significant deviations of the simulation curve from the data (Supporting Figure S4).
Another quantity that can be used to validate the deduced in-vivo rates is the missense error frequency arising from the accommodation of near-cognate ternary complexes with incorrect amino acids. The calculated error frequency depends on all individual transitions for the accommodation of a cognate or near-cognate aa-tRNA and, in particular, on the concentrations of cognate and near-cognate ternary complexes whereas it is independent of the concentrations of non-cognates, see Eq. 36 in the Methods section. Using the deduced in-vivo rates in Table 2 and the ternary complex concentrations as estimated from the measured tRNA concentrations for 0.7 dbl/h [28], we obtain an average missense error frequency of for tRNALys misreading codons, in good agreement with the measured value [31].
The theoretical approach described here involves two novel concepts. First, we introduced the kinetic distance to provide a quantitative measure for the similarity of the in-vitro and in-vivo kinetics. This distance has an intuitive interpretation in terms of the free energy barriers that govern the individual transition rates along the reaction pathways, and provides a genuine metric in the mathematical sense. Second, we constructed a constrained minimization procedure in order to deduce the unknown in-vivo values of the individual transition rates from their known in-vitro values.
It is instructive to compare our approach with flux control or sensitivity analysis, a widely used method for multistep reaction pathways [3], [35]–[37], which has also been applied to protein synthesis [38]. The latter method explores the local vicinity of a given kinetics and describes the linear response of the overall flux to small changes in the individual transition rates in terms of flux control or sensitivity coefficients. In contrast, the theoretical approach introduced here is not restricted to the linear response regime but explores the space of transition rates in a global manner via an extremum principle (Fig. 2B). Furthermore, both the coordinate transformation from the individual transition rates to the single barrier shifts and the constraint arising from the global in-vivo property make our approach highly nonlinear.
When we applied our computational method to translation elongation by ribosomes, we obtained predictions for the individual in-vivo rates that could be validated by three independent sets of data for codon-dependent translation speeds, codon-specific translation dynamics and missense error frequencies of protein synthesis. In all cases, we found good agreement between theory and experiment without adjusting any fit parameter.
Even for the largest growth condition of 2.5 dbl/h, most of the deduced in-vivo rates are similar to the measured in-vitro rates (Fig. 5B) but three in-vivo rates are significantly increased compared to their in-vitro values: the rejection rate for near-cognates, the dissociation rate after initial binding, and the recognition rate for cognate and near-cognate ternary complexes. The largest difference is found for the rejection rate , which is increased in vivo by a factor of 3.9, while the dissociation rate and the recognition rate are increased by a factor 3.3 and 2.2, respectively.
For all transition rates of the elongation cycle, we find that the deviations between the in-vivo and in-vitro rates correspond to relatively small shifts of the corresponding free energy barriers (Fig. 5A). In fact, all single barrier shifts are predicted to be smaller than . Because the cytosol represents a rather complex buffer, such small changes in the free energy barriers can be easily envisaged, arising, e.g., from changes in the hydrogen bond networks around the ribosome or from changes in the flexibility of some parts of this complex. On the other hand, our results also show that the high-fidelity buffer at 37°C, used here and developed by [25] represents a good approximation to the cytosol as far as the ribosomal kinetics is concerned, in contrast to earlier estimates in Ref. [39].
The free energy barriers considered here could be studied by Molecular Dynamics simulations. The latter method has been recently applied to explore the free energy landscape of tRNA translocation through the ribosome [40], [41]. From such simulations, one can estimate the attempt frequencies for barrier crossing which are difficult to determine by other computational methods. In principle, these simulation techniques could also be used to investigate how the energy landscape changes as one varies the ambient buffer conditions in the simulations.
Even though the predicted shifts of the free energy barriers are relatively small, the associated changes of the transition rates have an interesting consequence for the relative importance of initial selection and proofreading for the error frequency of protein synthesis. For the codon-specific Markov process depicted in Fig. 3, the efficiency of initial selection and proofreading are described by the coefficients and , respectively. The in-vivo value of the initial selection coefficient is increased by a factor of 7.7 compared to the corresponding in-vitro value whereas the proofreading coefficient is increased by a factor of 2.9. The combination of improved initial selection and proofreading leads to a reduction of the in-vivo error frequency by a factor of 6.7, a reduction that is primarily achieved by the improved initial selection of the bound ternary complexes.
In the present study, the codon-dependence of the elongation cycle arose from the initial binding rates that depend on the concentrations of cognate, near-cognate, and non-cognate tRNA, because we used the same transition rates along the reaction pathways for all cognate as well as for all near-cognate tRNAs. Thus, the values of the rates , , …of the cognate branch represent average values, obtained by averaging over all cognate tRNAs of all codons, and likewise for the internal rates , , …of the near-cognate branch. In vitro, the decoding rates of different cognate codons were observed to be rather similar [14], [25] whereas the GTPase activation rate was found to vary between 0.06/s and 1.3/s for different near-cognate codons of tRNA [25]. Likewise, recent in-vivo experiments provided evidence that the error frequency on 4 out of 14 near-cognate codons of tRNA is much higher than on the remaining 10 near-cognate codons [42]. Theoretically, it is straightforward to include codon-specific decoding and processing rates. Experimentally, it is, however, quite challenging to determine these rates in vitro for all codons and tRNA species.
Our theory for protein synthesis by ribosomes can be extended in a variety of ways. For example, one could study how the overall elongation rate or the missense error frequency vary with changes in the overall ternary complex composition or as a function of individual ternary complex concentrations. Likewise, one may investigate how changes in internal transition rates arising, e.g., from protein or rRNA mutagenesis, affect the speed and accuracy of translation elongation.
The computational method developed here to deduce the in-vivo from the in-vitro rates is relatively simple and can be applied, in general, to any multistep process or Markov model, for which one can estimate the in-vitro rates. Simple examples are provided by the folding and unfolding of proteins, the catalytic activity of enzymes with one active site, or the motility of molecular motors. More complex examples are transcription by RNA polymerase, protein refolding by chaperones, or protein degradation by proteases. Our method can also be applied to the large number of biochemical processes that have been studied by flux control or sensitivity analysis. Furthermore, the similarity measure provided by the kinetic distance could be useful in the context of systems biology, where the importance of detailed kinetics has been recently emphasized [43]. One important target in systems biology is to standardize the experimental data for such networks. Using the kinetic distance introduced here, one could, in fact, compare the kinetic data obtained by different groups in a systematic and quantitative manner.
Using the general theory of stochastic processes [44], [45], we derived explicit expressions for important dynamical quantities of the translation elongation cycle (Fig. 3) in terms of the individual transition rates. These quantities include the codon-specific accommodation times, i.e., the times that the ribosome needs to fully accommodate a cognate or near-cognate tRNA and, thus, to move from state 0 to state 5 or state 10 for the Markov process in Fig. 3. A straightforward but somewhat tedious computation leads to explicit, analytical expressions for these time scales in terms of the individual transition rates . These expressions can be decomposed into four different dwell times according to(11)a decomposition that directly reflects the state space of the Markov process in Fig. 3 and has the following intuitive interpretation.
The first dwell time represents the total time that the ribosome spends in state 0 during one complete elongation cycle at codon . Because of the different dissociation and backward transitions, the ribosome typically visits the state 0 several times before it is fully accommodated in the states 5 or 10, see Fig. 3. The second dwell time in Eq. 11 corresponds to the total time that the ribosome binds a non-cognate ternary complex and, thus, dwells in state 11 during one complete elongation cycle at codon . The third dwell time corresponds to the total time that the ribosome spends in the intermediate states 1, 2, 3, and 4 of the cognate branch during one complete elongation cycle at codon . Finally, the fourth dwell time in Eq. 11 represents the total time that the ribosome spends in the intermediate states 6, 7, 8, and 9 of the near-cognate branch.
These four dwell times can be expressed in a particularly compact and transparent manner if one uses the transition probabilities(12)
The dwell time , which the ribosome spends in state 0 during one complete elongation cycle at codon , then has the form(13)and, thus, depends on the concentrations and of free cognate and near-cognate ternary complexes as well as on the dimensionless, concentration-independent ratios(14)and
(15)The second dwell time for state 11 with a bound non-cognate ternary complex is given by the expression(16)and is, thus, proportional to the concentration of free non-cognate ternary complexes.
The third dwell time , which represents the sum of all dwell times for the intermediate states 1, 2, 3, and 4 of the cognate branch, can be written as(17)with the concentration-independent time scale(18)that depends only on transitions that emanate from the intermediate states of the cognate branch. Likewise, the fourth dwell time for the intermediate states 6, 7, 8, and 9 of the non-cognate branch has the form(19)with the concentration-independent time scale(20)that depends only on transitions that emanate from the intermediate states of the near-cognate branch.
The expression for the codon-specific accommodation time as given by Eq. 11 involves all individual rates apart from the processing rate . When we add the processing time , we obtain the codon-specific elongation time which the ribosome needs to complete a full elongation cycle at a certain codon . The codon-specific elongation rates are then given by(21)
One important global property of protein synthesis is the average speed of the ribosomes, which defines the overall elongation rate . The inverse of the overall elongation rate is equal to the average elongation time , which is obtained by averaging the codon-specific elongation times over all codons using the codon usage . For each codon , the quantity represents the probability that the ribosome encounters this codon. These probabilities are normalized and satisfy .
For the in-vitro assay, the relation between the overall elongation rate and the codon-specific accommodation times was rewritten in the form(22)and then used to calculate the processing rate from the measured value of the overall elongation rate and the measured values of the individual rates , which determine the codon-specific accommodation times .
In vivo, the overall elongation rate is given by the analogous expression(23)where the codon-specific accommodation times follow from the same expression as in Eq. 11 but with the in-vitro values replaced by the in-vivo values . When we insert the known in-vivo value of the overall elongation rate into Eq. 23, we obtain a constraint on the (unknown) in-vivo values of the individual transition rates. This constraint can be expressed in terms of the single barrier shifts when we replace in Eq. 23 by , see Eq. 2.
All in-vitro values of the individual transition rates as given in Table 1 have been obtained for the high-fidelity buffer as developed in [12], [25], and [34]. Most of these values are based on previous measurements as explained in the following paragraph. In addition, we also performed new experiments to measure the overall elongation rate , both at 20°C and at 37°C, see Supporting Figure S1, as well as the individual rates and at 20°C, see Supporting Figure S2, using the experimental protocols described previously [34], [46], [47].
The in-vitro value of the association rate constant was previously measured at 20°C [12]. Its value at 37°C was obtained assuming an Arrhenius temperature dependence and using the previously determined activation energy of 2.4 kcal/mol for initial binding [10]. The dissociation rate at 20°C was taken from [12]. The decoding rates at 20°C were obtained by averaging over previously published values as measured for different codons of tRNA. In particular, we averaged the rates as given in Table 1 of [25] for cognate as well as for near-cognate codons to obtain the rates , , , , and . The rate has not been measured but estimated under the assumption that it is not rate-limiting. The rate at 37°C was reported previously and was used to determine the rate , i.e., using an error frequency of 0.06 for the proofreading step [34]. The rate has been measured both for 20°C and for 37°C [25], [34]. The rate was calculated for both temperatures from the measured values of the overall elongation rate via Eq. 22.
Finally, we assumed an Arrhenius temperature dependence to estimate some of the in-vitro rates at 37°C from their values as measured at 20°C. These estimates are based on the following considerations. We start from Eq. 4 for the transition rates and use the decomposition of the activation free energy into the activation enthalpy and the activation entropy , which leads to(24)where the last expression involves the attempt frequency as obtained from transition-state theory [2]. In this way, any state-dependence of the attempt frequency has been absorbed into the activation entropies . If one plots the logarithms of the measured rates as a function of the inverse temperature (conventional Arrhenius plots), one finds linear relationships [10], [13], [48], [49], which imply that the two unknown parameters in Eq. 24, and , do not depend on temperature over the experimentally studied temperature range. However, the activation entropies as obtained from the behavior of for small vary significantly with the ribosomal states and [10], [13], [48], [49]. Possible molecular mechanisms for this variation have been recently discussed based on atomistic molecular dynamics simulations [50].
Using the expression in Eq. 24 with -independent enthalpies and entropies , we now consider the ratios at the two temperatures of interest, K and K. We take the accommodation rate as a reference rate because the value of this rate has been measured at both temperatures. For each individual transition, we then obtain two equations, corresponding to the two temperatures and , which can be combined to eliminate the enthalpy . As a result, we obtain the relation(25)
At present, the entropy differences are difficult to estimate for all individual transitions from the available experimental data. However, these differences are multiplied by the relative temperature difference which is rather small. Therefore, we used the approximate relation(26)to estimate the values of the rates , , , , , , , and at 37°C (Table 1) from the measured values of , , and .
The overall elongation rate as given by Eq. 23 also depends on the association rates for initial binding, which are proportional to the pseudo-first-order rate constant and to the concentrations of the ternary complexes as in Eq. 9. Therefore, in order to use Eq. 23 for the process in vivo, we had to estimate the corresponding values and the ternary complex concentrations in the cell.
The diffusion of ternary complexes and, thus, their binding to ribosomes is slowed down in vivo by molecular crowding. The time it takes a ternary complex to find a single ribosome depends on the cell volume, the diffusion constant of the ternary complex, and the ribosome size [51]. Using the diffusion constant of /s [52], [53] for a ternary complex in the cytosol, we found that the in-vivo value of the bimolecular association rate constant is about 54% of the in-vitro value , compare Table 1 and Table 2.
For the in-vivo concentrations of the ternary complexes, we used the values of the tRNA concentrations as measured by [28] in E. coli for the growth conditions of 0.7, 1.07, 1.6, and 2.5 dbl/h. In the latter study, the authors determined the concentrations of all 43 elongator tRNA species . These concentrations are then combined, for each codon , into the concentrations , , and of cognate, near-cognate, and non-cognate ternary complexes within the cell. Thus, for each codon , we started from the corresponding row in Fig. 4, and added all concentrations up that correspond to green (cognate), yellow (near-cognate), and purple (non-cognate) tRNA species, respectively.
To estimate the uncertainty of the predicted in-vivo rates , we first simplify the notation. In this section, the internal transitions with distinct transition rates will be distinguished by the subscript with . Thus, we now use the short-hand notation , , and for the in-vitro rates of a certain assay, for the unknown in-vivo rates , and for the predicted in-vivo rates , respectively. For ribosome elongation as described by the Markov process in Fig. 3, we distinguish internal transitions.
The inaccuracy or error of the in-vitro rates can be described by(27)with the absolute error and the relative error(28)of the in-vitro rate . Both the average values and the absolute errors are estimated from the experimental data for the in-vitro assay under consideration.
When we apply the minimization procedure to the average values of the in-vitro rates, we use the coordinates(29)for the multi-dimensional barrier space. We then determine the point that is located on the hypersurface defined by Eq. 8 and depicted in Fig. 2B and has the shortest distance from the origin of the -coordinates. The coordinate values of the point then lead to the predicted values for the in-vivo rates.
In order to estimate the uncertainty of these predictions, it is useful to consider an auxiliary ensemble of fictitious in-vitro assays that is constructed ‘around’ the given assay as follows. For each transition , we introduce the binary variable . The binary variables can assume different ‘configurations’ as described by the different -tuples(30)
Each of these configurations defines a fictitious in-vitro assay, again denoted by , with transition rates(31)
The rates of assay define the coordinates(32)for the multi-dimensional barrier space with(33)where the asymptotic equality applies to the limit of small relative errors of the in-vitro rates. Therefore, if the origin of the multi-dimensional barrier space is defined by the coordinates in Eq. 29, corresponding to the average in-vitro rates , the ensemble of the fictitious assays forms the corners of a multi-dimensional ‘error polyhedron’ around this origin. For each corner, again labeled by , we can apply our minimization procedure and minimize the kinetic distance of the point from the hypersurface as defined by Eq. 8 and depicted in Fig. 2B. The point on the hypersurface with the shortest distance from the corner has the coordinates .
The variations in the coordinate values of the predicted in-vivo point as obtained for different corners can be used to obtain an estimate for the absolute error of these coordinate values. We then write the coordinate values of the predicted in-vivo point in the form(34)where the values correspond to the average in-vitro rates . The predicted in-vivo rates are now given by
(35)Therefore the relative error of the predicted in-vivo rates reflects both the relative error of the in-vitro rates (Eq. 27) and the absolute error of the coordinate values for the predicted in-vivo point (Eq. 34).
The Markov process for ribosome elongation considered here, see Fig. 3, involves distinct transition rates, which implies that the corresponding barrier space has 12 dimensions. We first determined the coordinate values of the in-vivo point as predicted from the average value of the in-vitro rates. The largest coordinate values of the predicted in-vivo point were found for the three transition rates , , and (Fig. 5B). We then focused on the errors of these three in-vitro rates, which define 8 corners of the ‘error polyhedron’ around the origin of the -coordinates. For each of these corners , we determined the closest point on the hypersurface and the coordinate values of this point. We then estimated the absolute error of the coordinate values from the largest and smallest values of as obtained for different corners . The errors were finally used, together with the relative errors of the measured in-vitro rates, to determine the relative standard deviations (RSDs) of the predicted in-vivo rates as displayed in Table 2.
Consider a certain tRNA species and a codon that is near-cognate to . The missense error frequency for misreading the codon by the tRNA species is equal to the probability that is fully accommodated at . For the multistep process considered here, this probability is given by(36)which depends on the concentration of the near-cognate ternary complex species , on the concentrations and of all cognate and near-cognate ternary complexes as well as on the concentration-independent ratios and as given by Eqs. 14 and 15.
The experimental study in [31] determined the error frequency for all codons that are near-cognate to . The average error frequency for misreading one of these codons is then obtained from(37)where the set contains all codons that are near-cognate to and denotes the codon usage as before.
|
10.1371/journal.pbio.1002347 | FIH Regulates Cellular Metabolism through Hydroxylation of the Deubiquitinase OTUB1 | The asparagine hydroxylase, factor inhibiting HIF (FIH), confers oxygen-dependence upon the hypoxia-inducible factor (HIF), a master regulator of the cellular adaptive response to hypoxia. Studies investigating whether asparagine hydroxylation is a general regulatory oxygen-dependent modification have identified multiple non-HIF targets for FIH. However, the functional consequences of this outside of the HIF pathway remain unclear. Here, we demonstrate that the deubiquitinase ovarian tumor domain containing ubiquitin aldehyde binding protein 1 (OTUB1) is a substrate for hydroxylation by FIH on N22. Mutation of N22 leads to a profound change in the interaction of OTUB1 with proteins important in cellular metabolism. Furthermore, in cultured cells, overexpression of N22A mutant OTUB1 impairs cellular metabolic processes when compared to wild type. Based on these data, we hypothesize that OTUB1 is a target for functional hydroxylation by FIH. Additionally, we propose that our results provide new insight into the regulation of cellular energy metabolism during hypoxic stress and the potential for targeting hydroxylases for therapeutic benefit.
| Hypoxia is a commonly encountered physiologic and pathophysiologic stress to which mammalian cells have evolved an effective adaptive response. This response is governed by a transcription factor termed the hypoxia-inducible factor (HIF). The mechanisms linking the cellular sensing of oxygen levels to HIF activation have been elucidated and involve oxygen-dependent hydroxylation of HIF on proline and asparagine residues by a family of hydroxylases. A key question that remains unclear is the extent to which oxygen-dependent hydroxylation occurs as a functional post-translational modification outside of the HIF pathway. This is key to developing our understanding of whether hydroxylation is a general regulatory modification or one which has specifically evolved for the regulation of HIF. Here, we demonstrate that the deubiquitinase ovarian tumor domain containing ubiquitin aldehyde binding protein 1 (OTUB1) is a target for functional hydroxylation by the FIH hydroxylase. Hydroxylation of OTUB1 by FIH on asparagine residue N22 results in a restriction in its interactome, leading us to hypothesize a possible role for hydroxylation in substrate targeting. Of interest, interactions of OTUB1 with a number of proteins involved in metabolism are altered upon removal of the hydroxylation site—implicating OTUB1 as a possible link between oxygen sensing and the regulation of metabolism.
| Hypoxia is a common feature of the microenvironment in a number of pathophysiologic conditions and represents a significant threat to cellular metabolic homeostasis [1]. Eukaryotic cells have evolved the capacity to rapidly sense changes in intracellular oxygen levels through a family of hydroxylases that confer oxygen-dependence upon the key transcriptional regulator of the adaptive response to hypoxia, termed the hypoxia-inducible factor (HIF) [2,3].
Hydroxylases were first identified as oxygen sensors in the HIF pathway and belong to the Fe(II)- and 2-oxoglutarate-dependent dioxygenase superfamily [4]. These enzymes catalyze the hydroxylation of proteins, in a manner that is dependent on the availability of molecular oxygen (O2). Therefore, hydroxylase activity is decreased when O2 is low [5]. Four discrete HIF-hydroxylase isoforms have been identified to date, three of which are prolyl-4-hydroxylases (PHD1–3) (Uniprot accession numbers: Q96KS0, Q9GZT9, Q9H6Z9), which regulate the stability of HIFα subunits. In normoxia, the PHDs hydroxylate specific proline residues (P402 and P564 on HIF-1α), which promotes binding of the von Hippel-Lindau (VHL) (Uniprot accession number: P40337) protein followed by formation of an E3 ubiquitin ligase complex, HIFα ubiquitination and subsequent degradation [2]. In parallel with this, a second oxygen-dependent repression of HIF transcriptional activity is regulated by asparaginyl hydroxylation. The asparagine hydroxylase, termed factor inhibiting HIF (FIH) (Uniprot accession number: Q9NWT6), hydroxylates an asparagine residue within HIFα subunits (N803 on HIF-1α), resulting in steric inhibition of its interaction with the transcriptional co-activator p300/CBP, thereby inhibiting HIF-dependent transcription [2]. In hypoxia, when HIFα hydroxylation is reduced, HIFα subunits escape degradation, translocate into the nucleus, bind to the subunit HIF-1β, form a transcriptional complex with p300/CBP and activate gene expression [6].
A number of proteins other than HIF are also sensitive to regulation by hypoxia. However, the mechanism governing their oxygen-sensitivity is less clear [7]. A key question that remains is whether functional hydroxylation is specific for the regulation of HIFα or if other oxygen-sensitive proteins are also regulated by this post-translational modification. While the number of proteins identified as being targets for proline hydroxylation is low [8–10], FIH-dependent asparagine hydroxylation has been demonstrated for a larger group of non-HIF substrates, including ankyrin-repeat domain (ARD)-containing proteins such as tankyrase, notch-1, ASPP2, and IκBα [11–14]. However, the functional consequences of asparagine hydroxylation in general remain less clear [15,16]. Notably, a functional hydroxylation of the ion channel TRPV3 by FIH has recently been reported [17].
FIH homozygous knockout mice demonstrate a metabolic phenotype leading to the proposal that FIH is a key regulator of cellular metabolism [18]. However, a change of HIF activity alone due to FIH knockout could not explain all of the observed metabolic changes [18], indicating HIF-independent mechanisms. Because of this, a key question remaining is whether other FIH target proteins are involved in the observed metabolic phenotype in FIH-deficient mice.
In a previous study, we identified a number of putative hydroxylation substrates in the IL-1β signaling pathway that could account for the inhibition of IL-1β-induced inflammation observed in cells treated with pharmacologic hydroxylase inhibitors [19]. One identified candidate for asparagine hydroxylation was the deubiquitinase (DUB) ovarian tumor domain containing ubiquitin aldehyde binding protein 1 (OTUB1) (Uniprot accession number: Q96FW1) [19]. Interestingly, in a global proteomic analysis, OTUB1 was identified to interact with metabolic regulators [20]. Furthermore, mice deficient in OTUB1 demonstrate a lean body mass phenotype that is reflective of altered metabolic function (http://www.mousephenotype.org/data/genes/MGI:2147616).
In the current study, we provide evidence that N22 of OTUB1 is a bona fide target for enzymatic hydroxylation by FIH as demonstrated by a combination of mass spectrometric techniques and in vitro peptide hydroxylation. Furthermore, site-directed point mutagenesis of N22 enhanced the OTUB1 interactome, particularly with respect to proteins involved in metabolism. Finally, overexpression of mutant OTUB1 resulted in cellular bio-energetic stress (as reflected by enhanced AMP kinase activation) when compared to wild-type (WT) OTUB1, thus indicating a functional role for N22 hydroxylation in terms of regulating cellular metabolism. These data provide a mechanistic link between FIH-dependent hydroxylation of OTUB1 and alterations in cellular metabolism and contribute a further level of understanding to the vital link between cellular oxygen-sensing mechanisms and the control of cellular metabolism.
Previous studies have linked homozygous FIH deficiency in mice with a phenotype of disrupted energy metabolism. This is in part reflected by altered phosphorylation of AMPKα, which reflects a change in the cellular AMP:ATP ratio and therefore can be used as a surrogate marker of cellular metabolic stress. The molecular mechanisms underpinning this alteration in AMPKα phosphorylation remain unknown, but appear to be independent of the prototypic FIH substrate HIF [18]. In order to test whether FIH regulates energy metabolism at the cellular level, HEK293 cells were transfected with FIH-V5 for 24 h and cell extracts were generated either prior to the addition of fresh medium or 8–24 h later. Overexpression of FIH led to an increase in phosphorylation of AMPKα compared to cells transfected with an empty vector (Fig 1A and 1B). This effect was most prominent prior to the addition of fresh medium. Overexpression of the catalytically dead mutant FIH H199A did not change AMPKα phosphorylation (S1A and S1B Fig). These data support previous work linking FIH hydroxylase activity to the regulation of metabolic processes in cells.
In a previous global screen, we identified the DUB OTUB1 as a putative new substrate for asparagine hydroxylation by FIH [19]. Furthermore, mice deficient in OTUB1 demonstrate a phenotype that is consistent with altered metabolism (http://www.mousephenotype.org/data/genes/MGI:2147616). Also, OTUB1 was shown to interact with proteins involved in the regulation of cellular metabolism [20]. In support of this, we found that overexpression of OTUB1 in HEK293 cells was associated with an increase in phosphorylation of AMPKα 24 h following the addition of fresh media (Figs 1C, 1D and S2A). This indicates that the cells were experiencing metabolic stress as reflected by an imbalance in the cellular AMP:ATP ratio (and subsequent AMPK activation) when OTUB1 activity is increased. Furthermore, overexpression of FIH together with simultaneous knockdown of OTUB1 prevented the FIH-dependent increase in AMPKα phosphorylation (S1C, S1D and S2B Figs). Overall, these data led us to hypothesize that OTUB1 hydroxylation by FIH may be a link in the association of FIH with an alteration in cellular metabolism.
In order to determine whether OTUB1 is a bona fide FIH substrate, initially we used mass-spectrometry–based approaches. First, we tested whether OTUB1 is enzymatically hydroxylated (as opposed to this being a spurious chemical oxidation event). To do this, in one set of cells we maximized the asparagine hydroxylation capacity by overexpressing FIH along with OTUB1. A second set of cells also overexpressing OTUB1 (but not FIH) were treated with the pan-hydroxylase inhibitor DMOG to minimize hydroxylation capacity. Successful overexpression of OTUB1 is demonstrated by immunoprecipitation and quantitative mass spectrometric analysis (Fig 2A). We next investigated the hydroxylation status of immunoprecipitated OTUB1. Fig 2B shows an extracted ion chromatogram of an OTUB1 peptide containing the N22 residue. Mass spectrometric analysis revealed that the peptides with the shorter retention time corresponded with the hydroxylated form of this peptide. Decreased retention time has previously been demonstrated to be associated with FIH-dependent hydroxylation of peptides [11]. Tandem mass spectrometry (MS/MS) analysis of immunoprecipitated OTUB1 peptide demonstrated N22 hydroxylation in the sample where FIH was also overexpressed but not in the cells treated with DMOG (Figs 2B, 2C and S3). Importantly, the oxidation/hydroxylation of M31 (reflecting a nearby spurious oxidation event) in the same samples was independent of enzymatic hydroxylase activity (Fig 2D). Methionine residues are highly susceptible to spurious oxidation [21].
In order to investigate if the observed OTUB1 N22 hydroxylation was regulated by physiologically relevant changes in the cellular microenvironment, we incubated HEK293 cells overexpressing both OTUB1 and FIH in 0.2% oxygen for 8 h with and without subsequent re-oxygenation at 21% oxygen for one additional hour. The analysis of the OTUB1 N22 hydroxylation levels by mass spectrometry showed a significant reduction of OTUB1 N22 hydroxylation in hypoxia which was significantly reversed by re-oxygenation (Fig 2E and 2F). DMOG-dependent inhibition of hydroxylases led to a similarly reduced OTUB1 N22 hydroxylation level as hypoxia (Fig 2F). Of note, the DMOG-dependent inhibition of OTUB1 hydroxylation was partly reversed by FIH overexpression when compared to the effect of DMOG without FIH overexpression (Fig 2C and 2F). Nutrient starvation for 8 h with and without re-introduction of nutrients following for one additional hour also down-regulated OTUB1 N22 hydroxylation, although to a lesser degree than hypoxia (S4A Fig). Overall, these data strongly support the contention that N22 of OTUB1 is a bona fide substrate for enzymatic hydroxylation by FIH and that this is regulated by changes in the cellular microenvironment such as hypoxia.
We next investigated whether OTUB1 hydroxylation on N22 is FIH-dependent. Alignment of the amino acid sequence around N22 of OTUB1 with known FIH substrates and a recently published consensus sequence for FIH target proteins revealed that N22 of OTUB1 lies within a motif highly similar to the consensus sequence (Fig 3A) [22]. Protein sequence alignments indicated that the OTUB1 consensus sequence is evolutionary conserved within mammals (S4B Fig). We next overexpressed FLAG-HA-OTUB1 in HEK293 cells and either treated these cells with non-targeting siRNA (siNT) or siRNA targeting endogenous FIH (siFIH) (Figs 3B and S2C). OTUB1 was immunoprecipitated and the hydroxylation status of N22 was analyzed by quantitative mass spectrometry. The hydroxylation of N22 was decreased in cells treated with FIH siRNA supporting the concept that N22 of OTUB1 is a target for endogenous enzymatic FIH-dependent hydroxylation (Fig 3C). Furthermore, to demonstrate that N22 is directly hydroxylated by FIH, FIH activity was measured in an in vitro CO2 capture assay using purified FIH and either wild-type OTUB1 peptides containing N22 or OTUB1 peptides in which N22 was replaced by an alanine residue (N22A). Using this assay, which measures the turnover of 2-oxoglutarate into succinate and CO2 by FIH, we found a significant increase in FIH activity in the presence of wild-type but not mutant OTUB1 peptide (Fig 3D). Taken together, these data demonstrate that N22 of OTUB1 is a bona fide substrate for enzymatic hydroxylation by FIH.
We next investigated possible functional consequences of OTUB1 hydroxylation on N22. To do this, we used the N22A mutant in order to prevent FIH-dependent hydroxylation at this site. HEK293 cells were co-transfected with either FLAG-OTUB1 WT or FLAG-OTUB1 N22A, along with FIH (in order to maximize hydroxylation capacity; Fig 4A). Following immunoprecipitation of OTUB1 (demonstrated in Fig 4B) we identified the OTUB1 WT and the OTUB1 N22A interactomes by mass spectrometry. Initially, we confirmed the interaction of both OTUB1 WT and OTUB1 N22A with six previously described OTUB1 interacting proteins (S5 Fig) [20,23–27]. We next used the interaction of FIH with OTUB1 as additional positive control [19] and confirmed this interaction in the case of OTUB1 WT in the interactome experiment and also subsequently by western blot analysis (Figs 4C and S5G). Interestingly, this interaction was greatly reduced with OTUB1 N22A, indicating a key role for this residue in the interaction between FIH and OTUB1 (Figs 4C and S5G).
Qualitative analysis revealed that 147 proteins were associated with OTUB1 WT, while 318 proteins were associated with OTUB1 N22A (Fig 4D). Of the OTUB1 interacting proteins, 127 were associated with both OTUB1 WT and OTUB1 N22A, indicating that the core interactome is not affected by mutation of N22. However, when we compared the OTUB1 WT and the OTUB1 N22A interactomes, we found that while just 13 proteins were enriched in their association with OTUB1 WT over OTUB1 N22A, 147 proteins were enriched in their association with OTUB1 N22A over OTUB1 WT (Fig 4D). There were 86 proteins that had equivalent levels of interaction with both OTUB1 WT and N22A. This indicates that loss of hydroxylation on N22 leads to more than a doubling of the number of proteins in the OTUB1 interactome through the recruitment of new binding partners. Of note, the N22A point mutation of OTUB1 did not change its deubiquitinase activity, which indicates that this mutation does not significantly alter protein structure (as catalytic activity is retained) (S6 Fig). Based on these data, we hypothesize that hydroxylation of N22 on OTUB1 profoundly alters its interaction with other proteins and is therefore likely of functional consequence.
Ontological analysis of the proteins differentially associated with OTUB1 WT and OTUB1 N22A using the Panther database (www.pantherdb.org) revealed proteins associated with multiple biological processes (S7A and S7B Fig). However, metabolism-associated proteins were most highly represented, which is in agreement with previously published data for wild-type OTUB1 by Sowa et al., demonstrating that OTUB1 interacts with metabolic regulators [20]. Furthermore, OTUB1 N22A had increased numbers of metabolism-associated proteins when compared to OTUB1 WT (Fig 4E and S1 Table) indicating that loss of N22 hydroxylation may impact upon interaction between OTUB1 and multiple proteins important in the regulation of metabolism. In summary, we demonstrate that N22A mutation of OTUB1 profoundly alters its physical interactome. Of note, proteins associated with metabolic processes are heavily represented in this cohort.
We next investigated the impact of N22A mutation of OTUB1 on FIH-dependent regulation of cellular metabolism under conditions of energy starvation in cultured cells. Simultaneous glucose, glutamine, and pyruvate deprivation caused an increase in the phosphorylation of AMPKα likely as a result of ATP depletion (Fig 5A and 5B). Cells overexpressing both wild type OTUB1 and FIH (to maximize OTUB1 hydroxylation) showed similar levels of AMPKα activity as control cells, however, cells overexpressing both FIH and N22A mutated OTUB1 (to minimize OTUB1 hydroxylation) demonstrated robustly increased phosphorylation of AMPKα. These data are consistent with our hypothesis that FIH-dependent N22 hydroxylation of OTUB1 contributes to the regulation of cellular metabolism by FIH.
We next investigated the impact of the hydroxylation of N22 on the OTUB1 protein. We considered a potential change of OTUB1 protein levels and its half-life due to the hydroxylation of N22 similar to the described regulation of the HIF-1α protein by prolyl hydroxylation. We therefore established HEK293 cells stably overexpressing FLAG-OTUB1 WT or N22A, which, at the same time, also carried a stably integrated shRNA targeting the 3′UTR of OTUB1 to diminish endogenous OTUB1 protein levels (S8 Fig). We transiently transfected these cells with FIH-V5 to maximize FLAG-OTUB1 WT hydroxylation and analyzed OTUB1 WT and N22A protein levels for up to 48 h by western blot. No significant change between the protein levels of OTUB1 WT and OTUB1 N22A was observed (Fig 6A and 6B). We next investigated if endogenous OTUB1 protein levels change in response to an alteration of N22 hydroxylation levels. We transiently transfected HEK293 cells with either empty vector (control) or FIH-V5 and treated the control cells with DMOG and the FIH overexpressing cells with DMSO. This experimental set up was similar to the experiment performed in Fig 2A–2D, which lead to maximally hydroxylated N22 of OTUB1 in the FIH overexpressing sample and to diminished hydroxylation of N22 in the DMOG-treated sample. We then analyzed endogenous OTUB1 protein levels in a time course for up to 48 h by western blot. No difference was observed between maximally hydroxylated OTUB1 and minimally hydroxylated OTUB1 protein levels (Fig 6C and 6D). In order to investigate the half-life of OTUB1 depending on its hydroxylation status, HEK293 cells were incubated with DMOG for 16 h prior to the treatment with cycloheximide (CHX) to inhibit protein synthesis. Within a time frame of 6 h CHX treatment, in which HIF-1α protein levels significantly decreased, no change in OTUB1 protein levels were observed (Fig 6E and 6F). Overall, these data demonstrate that the hydroxylation of OTUB1 at N22 does not impact on OTUB1 protein stability.
Hypoxia is a common feature of a number of diseases in which metabolism is significantly altered, including chronic inflammation and cancer. The mechanisms by which hypoxia-dependent alterations in metabolism occur in such disease states have important implications for disease development and potential targets for future therapeutic intervention. In this study, we provide new insight into the regulation of metabolism by hypoxia, which is mediated through the DUB OTUB1.
The identification of HIF as a ubiquitous master regulator of the cellular adaptive response to hypoxia and its oxygen-dependent regulation by 2-oxoglutarate-dependent hydroxylases were key discoveries in our developing understanding of the oxygen-sensing mechanisms which operate in eukaryotic cells [21,28–33]. Because several pathways apart from HIF also demonstrate sensitivity to hypoxia, it was initially anticipated that post-translational hydroxylation would be a common modification resulting in the conferral of oxygen sensitivity on multiple targets. However, while it appears that hydroxylation is indeed a common protein modification, understanding the functional role of this outside of the HIF pathway has remained elusive.
Functional proline-hydroxylation of non-HIF proteins by the HIF prolyl hydroxylases has been proposed for a limited number of proteins, including FOXO3a, CyclinD1, ATF-4, and IKKβ [8–10,34,35]. However, asparagine hydroxylation by FIH appears to be a more commonly observed modification and has been clearly demonstrated for multiple non-HIF proteins, including several ARD-containing proteins such as tankyrase, notch-1, and IκBα [11–13]. However, the functional impact of this on cellular signaling pathways (if any) remains unclear. Therefore, the identification of new functional hydroxylation events is of key importance in developing our understanding of this oxygen-sensitive, post-translational modification.
It has recently been demonstrated that mice that are homozygously deficient in FIH demonstrate a metabolic phenotype characterized by (for example) reduced body weight, elevated metabolic rate, and hyperventilation [18]. In these studies, the cellular bioenergetic status was assessed by measurement of the activation of AMPK, a key gauge of cellular metabolic stress which becomes activated when ATP is depleted and the AMP:ATP ratio increases. Because these mice do not display a phenotype consistent with activated HIF, it appears that the mechanisms underpinning the metabolic phenotype are at least in part independent of the HIF pathway and depend upon other FIH-dependent pathways [18]. In this study, we identified OTUB1 to be a new FIH substrate that may be important in the regulation of cell metabolism (as also reflected by altered AMPK activation) and, as such, may provide mechanistic insight into the metabolic phenotype observed in the FIH knockout mouse. Of note, wild-type OTUB1 has been reported to interact with metabolic regulators [20]. Furthermore, while homozygous deletion of OTUB1 results in early lethality, mice heterozygously deficient in OTUB1 were reported to display a metabolic phenotype characterized by decreased lean body mass (http://www.mousephenotype.org/data/genes/MGI:2147616) [36–40]. Taken together, these data suggest the possibility that OTUB1 hydroxylation may at least in part provide a molecular explanation for some aspects of the observed phenotype in FIH-deficient mice.
Previous work has demonstrated a profoundly anti-inflammatory effect of pan-hydroxylase inhibitors (which have inhibitory activity against both PHDs and FIH) in multiple models of intestinal inflammation [41]; however, the full mechanism underpinning this remains unclear. Altered metabolism has recently been demonstrated to be a key regulator of inflammation [42]. Therefore, a possible contributory mechanism for the anti-inflammatory activity of hydroxylase inhibitors is through altered FIH-dependent hydroxylation of OTUB1, leading to differential metabolism at inflamed sites.
In our study, we found that N22, the site of OTUB1 hydroxylation by FIH, is located in a region of the protein that may be key to determining its activity. OTUB1 is unusual in that it hydrolyzes specifically K48 ubiquitin bonds but also inhibits the formation of K63 and K48 ubiquitin chains via a non-canonical, non-catalytic function through inhibition of E2 ubiquitin ligase activity [43–45]. In the OTUB1 apoenzyme, the residues N-terminal to the OTU catalytic domain are disordered (approximately amino acids 1 to 45), whereas upon binding of both distal ubiquitin and an E2 ubiquitin-conjugating enzyme, for example UBCH5B, the folding of a significant portion of the tail becomes stabilized as a structured alpha helix (amino acids 23 to 44) (Fig 7) [26,27,43]. N22 is located at the junction of the α-helix and the remaining unstructured region (amino acids 1 to 22). This is similar to the C-terminal transactivation domain (CAD) of HIF-1α, which is disordered when it is unbound but forms three distinct α-helices upon binding to CBP/p300, of which one helix includes N803, the HIF-1α asparagine residue targeted for hydroxylation by FIH [46,47]. In an attempt to investigate the implications of the hydroxylation of OTUB1 at this key hinge region, we found no impact of OTUB1 hydroxylation on OTUB1 protein levels or half-life (Fig 6). Also, OTUB1 enzymatic activity was unaffected by the N22A point mutation (S6 Fig). Therefore, a direct regulation of the OTUB1 interactome by the N22 hydroxylation resulting in differential substrate targeting seems likely. Consistent with this concept, protein:protein interactions have previously been demonstrated to be directly modified by asparagine hydroxylation in the HIF pathway. FIH-dependent hydroxylation of N803 disrupts HIF-1α interaction with the transcriptional co-activators p300/CBP regulating HIF-1α-dependent trans-activation of gene expression. Ongoing studies are investigating whether hydroxylation of wild-type OTUB1 at N22 impacts on its (non-)canonical activity.
In summary, in this study we provide evidence that OTUB1 is a target for functional hydroxylation by FIH. We propose that this modification may have important implications for the regulation of cellular metabolism by changing OTUB1 substrate targeting under conditions of hypoxia, such as those that occur during ischemia, chronic inflammation, and tumor growth.
Human embryonic kidney cells (HEK293) were cultivated under standard conditions and used for all experiments presented. Standard cell culture media was DMEM media containing 4.5 g/l glucose, sodium pyruvate and L-glutamine. As media for nutrient starvation experiments DMEM media without glucose, sodium pyruvate or L-glutamine was used. For the transient transfection of both siRNAs and plasmids Lipofectamine 2000 reagent (Invitrogen) was used according to the manufacturer’s description.
The plasmid encoding FIH-V5 was a kind gift of Dr. Eric Metzen (University of Duisburg-Essen, Essen, Germany), whereas the wild-type FLAG-OTUB1 coding plasmid was generously provided by Dr. Mu-Shui Dai (Oregon Health and Science University, Portland, Oregon, United States) [25]. The FLAG-HA-OTUB1 plasmid was a gift from Dr. Wade Harper (Harvard Medical School, Boston, Massachusetts, US) (Addgene plasmid # 22551) [20]. The plasmid encoding FIH H199A has previously been described [32]. Nontargeting siRNA (siNT) was purchased from Dharmacon (GE Healthcare) (ON-TARGETplus SMARTpool). The siRNA targeting FIH (siFIH) was produced by Eurofins Genomics according to a previously reported sequence [11] (sequence F1). siRNA targeting the 3′UTR of OTUB1 was produced by Eurofins Genomics according to a previously reported sequence [25] (siRNA-4).
The FLAG-OTUB1 N22A mutant was generated with the Quikchange II XL Site-Directed Mutagenesis kit (Agilent technologies) according to the manufacturer’s description, using the plasmid encoding FLAG-OTUB1 wild-type as template. Successful mutation was confirmed by sequencing of the targeted site in the obtained plasmid.
Protein concentrations of cell lysates for western blot analysis were determined using the Bio-Rad DC protein assay. Equal amounts of protein were separated by SDS PAGE, transferred to nitrocellulose membranes and detected using anti-OTUB1 antibody (Cell Signaling), anti-β-actin antibody (Sigma), anti-AMPKα antibody (Cell Signaling), anti-phospho-AMPKα (Thr172) antibody (Cell Signaling), anti-FIH antibody (Abcam), anti-α-tubulin antibody (Santa Cruz), anti-FLAG antibody (Sigma), anti-V5 antibody (Invitrogen), or anti-HA antibody (Roche).
Peptide hydroxylation was assayed by the hydroxylation-coupled decarboxylation of [1-14C]-2-oxoglutarate by human FIH (hFIH) as described previously [50]. Each 40 μl reaction contained 3.5 μM MBP-hFIH, 625 μM peptide substrate, 300 μM FeSO4, 40 μM 2-oxo[1-14C]glutarate (40,000 dpm), 4 mM ascorbate, 500 μM dithiothreitol, 0.4 mg bovine serum albumin, and 50 mM Tris-HCl (pH 7.0), and was incubated at 37°C for 60 min. Filters were dried, UltimaGold XR scintillatant added, and counted on a MicroBeta 2450 (Perkin Elmer).
Immunoprecipitation was carried out as previously described [19]. Briefly, cell lysates were incubated with the antibody-coupled beads anti-FLAG M2 affinity gel (Sigma) at 4°C for 1 h. Subsequently, the agarose beads were washed twice with lysis buffer (1% Triton X-100, 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM MgCl2) and twice with washing buffer (20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM MgCl2). This was followed by sample preparation for mass spectrometric analysis as previously described [51]. The samples were analyzed by a Q-Exactive mass spectrometer (Thermo Scientific) and searched with MaxQuant. MS/MS spectra were searched against the human UniProt database (www.uniprot.org). Variable modifications included (MYWNDEPK) hydroxylation/oxidation. For a more detailed description, see Scholz et al. 2013 [19]. The mass spectrometry proteomics data of the OTUB1 hydroxylation experiments have been deposited to the ProteomeXchange Consortium [52] via the PRIDE partner repository with the dataset identifier PXD002103.
The datasets of OTUB1 WT and OTUB1 N22A mutant co-precipitated proteins obtained by mass spectrometric analysis were first filtered for significant enrichment of proteins over control. Proteins were only considered as part of the interactome when the average of the six datasets obtained (three biological and two technical replicates) was at least 2-fold different to the negative control and, additionally, when this difference was statistically significant (Student’s t test was applied). In addition, the obtained lists were analyzed for differences between the wild-type and mutant interactome. Proteins were only considered to be changed between these two groups when the difference was at least 2-fold and when this difference was statistically significant. These analyses were carried out with Excel (Microsoft). The PANTHER database (www.pantherdb.org) was used for functional annotation of the obtained lists of proteins and to cluster the proteins according to the assigned gene ontology terms [53,54]. The protein interactions from this publication have been submitted to the IMEx (http://www.imexconsortium.org) consortium through IntAct [55] and assigned the identifier IM-23897. For some identified peptides in this experiment, it was not possible to assign the sequence to one specific protein. These results will not be shown in the IntAct database but are available in S2 Data.
1.5 x 108 HEK293 cells stably expressing empty vector (control), FLAG-OTUB1 WT or N22A were harvested by trypsinisation, washed 1x in cold PBS and lysed in 1 ml 1x TBS plus 1% NP40, 5 mM MgCl2, 1 mM PMSF and 1x Roche cOmplete EDTA-Free Protease Inhibitor Cocktail per sample on ice for 30 min. Lysates were clarified by centrifugation at 14,000 rpm at 4°C for 10 min and pre-cleared by incubating with 12.5 μl Sepharose 6 fast flow resin and 12.5 μl Protein G Dynabeads (prepared according to manufacturer’s instructions) for 30 min at 4°C with rotation. Immunoprecipitation was performed on pre-cleared lysates by adding 25 μl Protein G Dynabeads and 10 μl ANTI-FLAG M2 antibody per sample and incubating for 2.5 h at 4°C with rotation. Beads were then washed 3x in cold TBS + 0.2% NP40 with a final wash in cold TBS before resuspending in 60 μl TBS. Immunoprecipitated FLAG-OTUB1 WT and N22A (bound to Protein G Dynabeads, equivalent to approximately 3 x 107 cells) was incubated with 600 nM K48-tetraubiquitin (K48-Ub4) at 37°C. Samples were harvested at 0, 30, and 60 min. Empty vector IP beads were used as a negative control and 26S proteasomes as a positive control for DUB activity. Samples were mixed with SDS loading buffer, heated at 90°C for 5 min and separated on 4%–12% bis-tris gels. Proteins were transferred to PVDF membranes and blocked in PBS/0.2% Tween-20 plus 3% BSA for 1 h at room temperature. Proteins were detected with HRP-conjugated P4D1 anti-ubiquitin antibody and. ANTI-FLAG M2 antibody.
OTUB1 WT and OTUB1 N22A were amplified by polymerase-based chain reaction (PCR) using pFLAG-OTUB1 WT or N22A as template. The sequences were cloned into the entry vector pENTR4 (Invitrogen) using the restriction enzymes NcoI (Thermo Scientific) and XhoI (Thermo Scientific). Subsequently, pDEST15-OTUB1 WT and N22A (vector carrying the GST-tag) were generated using the gateway system (Invitrogen) with the LR clonase II (Invitrogen) according to the manufacturer’s description.
Escherichia coli BL21-AI (Invitrogen) were transformed with pDEST15-OTUB1 WT or N22A, respectively, and protein expression was induced by adding 0.2% L-Arabinose for 3h at 37°C. Bacteria were lysed using a Cell Disruptor (TS Series Bench Top, Constant Systems Ltd.) at 35 kpsi in two cycles. Lysates were cleared by ultracentrifugation at 162,000 xg and 4°C for 1 h (Sorvall WX100 Ultracentrifuge) and subsequently affinity purified with Glutathione Sepharose Fast Flow Columns (GSTrap FF, GE Healthcare) in the duo flow system (Bio-Rad). Successful protein expression and purification was verified by Coomassie staining and western blot against OTUB1.
Purified GST-tagged OTUB1 WT and N22A were dialysed in 20 mM Tris, 150 mM NaCl, 1 mM DTT to remove GSH. Samples were loaded into SnakeSkin dialysis tubing, sealed and incubated in dialysis buffer at 4°C overnight with gentle stirring. Samples were transferred to fresh buffer for a further 1 h at 4°C. Final protein concentration was assayed using a Nanodrop 2000 spectrophotometer and checked by running equal concentrations of GST-OTUB1 WT and N22A on a 10% SDS PAGE gel and staining with Coomassie brilliant blue.
Purified GST-OTUB1 WT and N22A were incubated with 600 nM K48-Ub4 at 37°C. Samples were harvested at 0, 30, and 60 min. K48-Ub4 alone was used as a negative control. Samples were mixed with SDS loading buffer, heated at 90°C for 5 min and run on 4%–12% bis-tris gels. Proteins were transferred to PVDF membranes and blocked in PBS/0.2% Tween-20 plus 3% BSA for 1 h at room temperature. Membranes were probed with HRP-conjugated P4D1 anti-ubiquitin antibody and anti-GST antibody.
For the functional analysis of metabolic proteins enriched in the OTUB1 N22A interactome over OTUB1 WT interactome the online tool DAVID Bioinformatics Resources 6.7 (http://david.abcc.ncifcrf.gov/) was used [56,57]. Functional annotation clustering was performed using default settings.
Control plasmids and plasmids encoding FLAG-OTUB1 WT or FLAG-OTUB1 N22A were linearized by digestion with PvuI (Thermo Scientific) and transiently transfected into HEK293 WT cells. Cells were selected with 1 mg/ml G418 for 4 wk.
An expression vector encoding a short hairpin RNA (shRNA) sequence targeting the 3′UTR of human OTUB1 (Sigma, TRCN0000273238) and a non-targeting shRNA (shControl) (Sigma, SHC016) were purchased from Sigma. Lentiviral particles for shOTUB1 and shControl were produced in HEK293T cells using the Vira-Power lentiviral expression vector system according to the manufacturer’s instructions (Invitrogen). HEK293 cells overexpressing FLAG-OTUB WT or N22A or control cells were infected with shOTUB1 or shControl lentiviral particles followed by selection with 2.5 μg/ml puromycin for 4 wk.
Primers designed for the site-directed mutagenesis of OTUB1 were as follows:
Forward primer 5′-AGGCCAGACAGGCAACACCTTCGGAGTCGCTGC-3′
Reverse primer 5′-GCAGCGACTCCGAAGGTGTTGCCTGTCTGGCCT-3′
Primers designed for the cloning of OTUB1 WT and N22A into pENTR4:
Forward primer 5′-ACGTCCATGGCGGCGGAGGAACCTCAGCA-3′
Reverse primer 5′-ACGTCTCGAGCTATTTGTAGAGGATATCGT-3′
Peptide sequences of peptides used in the in vitro hydroxylation assay:
Human HIF1 N803: DESGLPQLTSYDCEVNAPI
Murine Notch N2012: VEGMLEDLINSHADVNAVDD
Human OTUB1 N22: QQQKQEPLGSDSEGVNCLAYDEAIMAQQDRIQQE
Human OTUB1 N22A: QQQKQEPLGSDSEGVACLAYDEAIMAQQDRIQQE
For the analysis of statistical significance one-way ANOVA followed by Tukey test was applied for comparisons of more than two different datasets. For the comparison of two different datasets, unpaired Student’s t test was applied. P-values < 0.05 were considered statistically significant.
|
10.1371/journal.pgen.1005989 | An Integrated Genome-Wide Systems Genetics Screen for Breast Cancer Metastasis Susceptibility Genes | Metastasis remains the primary cause of patient morbidity and mortality in solid tumors and is due to the action of a large number of tumor-autonomous and non-autonomous factors. Here we report the results of a genome-wide integrated strategy to identify novel metastasis susceptibility candidate genes and molecular pathways in breast cancer metastasis. This analysis implicates a number of transcriptional regulators and suggests cell-mediated immunity is an important determinant. Moreover, the analysis identified novel or FDA-approved drugs as potentially useful for anti-metastatic therapy. Further explorations implementing this strategy may therefore provide a variety of information for clinical applications in the control and treatment of advanced neoplastic disease.
| Metastasis, the spread and growth of tumor cells from the original tumor to secondary sites throughout the body, is the primary cause of cancer-related death for most solid tumor types. The process of metastasis is very complex, requiring multiple individual steps and the cooperation of different cell types during the dissemination and proliferation steps. Many genes are involved in this process, but at present few have been identified and characterized. In this study, we have integrated multiple genome-wide analysis methods to try to identify large numbers of candidate metastasis-associated genes and pathways based on a highly metastatic mouse model. Using this strategy, we have identified a number of genes that predict outcome of human breast cancer. These genes implicate specific molecular and cellular pathways in the metastatic process that might be used to intervene in the process. Furthermore, this integrated analysis implicates pre-existing drugs that might be re-purposed to help prevent or reduce metastatic burden in patients. The combined results obtained from this analytical strategy therefore provide an important platform for further genome-wide analysis into the etiology of metastatic disease.
| Metastasis is an extremely complex process that involves not only tumor-autonomous events but also interactions with local microenvironment and distant tissues. Hundreds or thousands of genes are thought to be associated with metastatic progression [1]; however, the proportion of genes that contribute etiologically to tumor progression is unknown. Identification of genes that contribute mechanistically to metastasis will deepen our understanding of tumor progression and potentially provide novel targets for prevention or improving patient outcome. Because metastatic disease is the major cause of patient mortality and morbidity for patients with solid tumors [2], the ability to prevent or successfully intervene would be expected to demonstrate significant clinical benefit. Hence strategies to accelerate metastasis-associated gene discovery are particularly valuable for understanding terminal stages of neoplastic disease.
Our laboratory previously demonstrated that different inbred mice lineages (S1 Fig) possess different propensities for metastatic disease. The highly metastatic FVB/NJ-TgN(MMTV-PyMT)634mul model [3] (MMTV-PyMT) was bred to 27 inbred strains and significant suppression of metastasis was observed in the progeny of 12 inbred strains. Since all of the tumors were induced by the same transgene, this suggests that polymorphisms in the genetic background can influence metastatic progression [4] (S1 Fig). Interestingly, projecting the metastatic capacity of each strain onto the mouse phylogenetic tree demonstrated closely related strains can have distinct metastatic capacities (Fig 1A). This observation suggests that single nucleotide polymorphisms (SNPs) which distinguish these closely related inbred strains are enriched for factors causally associated with metastatic progression. As a consequence, genome-wide identification of these SNPs and their associated genes provide a rapid method for capturing significantly more metastasis-associated genes than previous single candidate gene genetic strategies (e.g. Ref. [5]). Herein we describe a novel integrative strategy to identify candidate metastasis-associated genes based on an integrated mouse-human genome-wide inherited susceptibility systems genetics screen.
To test this strategy, whole genome sequencing of NZB/B1NJ was performed. This strain was selected because of the availability of pre-existing linkage, expression, and validated metastasis susceptibility gene data, enabling internal validation of the screen [4, 5, 7, 8]. Comparison of metastatic capacity demonstrated a significant suppression of metastasis by the NZB/B1NJ genome compared to the original FVB/NJ background of the MMTV-PyMT animal (Fig 1B). (Raw data: ERP000927; polymorphism data available at http://www.sanger.ac.uk/resources/mouse/genomes/). After alignment of the reads to the C57BL/6J mouse reference genome (GRCm38), approximately 5 million SNPs that distinguish NZB/B1NJ were identified, including approximately 54,000 unique SNPs.
A number of observations suggest that a subtractive strategy integrated with epigenetic and transcriptional filters could identify SNPs potentially driving metastatic disease. First, analysis of SNPs associated with phenotypes in human GWAS demonstrated that 88% of these SNPs are intronic or intergenic [9] and 71% fall within DNAse hypersensitivity sites (DHS) [10] which delineate cis-regulatory elements (promoters, silencers, insulators, enhancers, locus control regions) [11]. Together these data suggest that more than half of GWAS-associated SNPs are associated with polymorphic DHS (pDHS). Consequently, DHS represent less than 1% of the mouse genome, and only 10% of pDHS are associated with nearby transcriptional variation [12].
As most inherited variation is thought to result from changes in gene expression rather than structural changes in proteins, we restricted our analysis to those SNPs within the DHS sites [13] in the well-defined mouse mammary adenocarcinoma 3134 cell line [14, 15]. This reduced the number of SNPs for consideration from more than 5 million to approximately 120,000. Further, the SNPs shared between NZB/B1NJ and FVB/NJ [16] (the host genome for the MMTV-PyMT transgene) were subtracted from the overall list to enrich for SNPs likely to be causally associated with the reduced metastatic susceptibility observed in the NZB/B1NJ strain. The total number of NZB/B1NJ SNPs that remained after the subtraction was 54,659 (Fig 1C, S1 Table). The SNPs were then mapped to genes using the Genomic Regions Enrichment of Annotations Tool (GREAT) [17] using the default association rules. A total of 7902 genes were associated with pDHS following this analysis.
Next, the genes were screened across two panels of mouse mammary tumors to enrich for those genes with pDHS that alter gene expression. These panels were generated from crossing the MMTV-PyMT transgenic mammary tumor model and the highly genetically diverse mouse genetic mapping panel, the Diversity Outbred (DO) population [18]. This population is a randomly bred population generated from 8 progenitor inbred strains, including wild-derived representatives of the major mouse subspecies, Mus musculus, Mus domesticus, and Mus castaneous. As a consequence, the DO population closely resembles natural populations like humans [19] and is therefore likely to capture much of the heritable expression level variation within the Mus genus.
The two populations of tumors were generated from crosses between the MMTV-PyMT and different subsets of animals from the 5th generation (G5 N = 131) [20] or 7th generation (G7 N = 159) of the DO population. Approximately 25% of the total DO population was used to breed with MMTV-PyMT for each generation (45 females out of 175 DO breeding cages). As a result, these two populations of DO mice were expected to carry distinct, yet overlapping combinations of SNPs. Consistent with this, the tumor phenotypes were significantly different between the two mouse populations (ex. G5: 74/129 mice with metastatic disease; G7: 132/161 mice with metastatic disease; p = 4.4x10-8; S2 Fig). We therefore chose to screen the populations separately for metastasis-associated genes.
To investigate the role of polymorphism on transcription, total RNA from the mammary tumors was assayed on Affymetrix ST v1.0 chips. The 7902 genes with polymorphic DHS were tested for significant expression variation across each of the DO populations. Genes that exhibited significant variation (p<0.05) within each DO x PyMT population were assumed to have polymorphisms that functionally affected transcription and were included for further analysis. Genes without significant expression variation across the DO populations were assumed to have SNPs that did not affect gene transcription and were excluded from further analysis. This filter reduced the metastasis-associated candidate gene list to 2810 genes for the DO G5 tumors and 3223 genes for the DO G7 tumors. These differentially expressed genes were then subjected to analysis using BRB ArrayTools survival or quantitative trait tools to identify genes from each of the DO data sets associated with metastatic disease. The resulting screen yielded 4 lists of potential metastasis susceptibility genes ranging from 358–1518 members (See Table 1 and S2 Table). Examination of the signatures indicated that only a minority of the genes were common between the DMFS (distant metastasis-free survival) and metastasis-correlated signatures (S3 Fig), consistent with the two DO populations comprising different combinations of metastasis-associated factors.
We next evaluated the performance of this strategy on a genome-wide basis. The prognostic ability of each of the gene signatures was tested on human breast cancer datasets. The signature hazard ratio (DMFS signature) or the correlation coefficient (metastasis-correlated signature) from the mouse data provided weight and direction to each gene to require identical functionality between the two species. The weighted gene signatures were then screened for their ability to discriminate outcome in human breast cancer using the Gene expression-based Outcome for Breast cancer Online (GOBO) tool [21]. Of interest, all four signatures were prognostic in estrogen receptor-positive (ER+) but not estrogen receptor-negative (ER-) tumors (Fig 2A) in the GOBO data set, suggesting that a significant fraction of genes in the signatures contribute in the same manner to metastatic progression in both species. For a second independent validation, the signatures were then tested on the METABRIC gene expression dataset [22]. The weighted metastasis-correlated signatures were also prognostic in the ER+ subset in this patient dataset (Fig 2B and S4 Fig). The DMFS signatures did not discriminate outcome in this dataset.
Of note, outcome in the GOBO dataset is distant metastasis-free survival (DMFS) and a large fraction of the data is from adjuvant treatment-naïve patients, similar to the treatment-naïve mouse populations. In contrast, the METABRIC outcome data is overall survival and includes response to neo-adjuvant, adjuvant, and salvage therapy, all of which can alter the clinical course of disease for a significant fraction of patients [23], and may subsequently change the relative weights of genes within the signature. The weight for the G5 DMFS signature gene was therefore recalculated using half of the METABRIC data (discovery set) and validated on the remaining samples (test set) (Fig 2C). This analysis indicates that the genes identified by the DMFS screen are associated with tumor progression in mouse and humans, although the relative weight of each contributing gene may not be preserved. Results showed both gene signatures can better discriminate outcome in ER+ versus ER- breast tumors, which is consistent with the MMTV-PyMT being a model for human luminal breast cancers [24, 25].
To confirm that these gene signatures were more prognostic then thos generated by chance, three different analyses were performed using the METABRIC data set. In the first set of analyses, the weights were held constant and random genes were assigned for the 656 genes of the metastasis-correlated signature and the 383 genes of the recalculated METABRIC DMFS signature. The process was permuted 1000 times and the performance of the randomly assigned genes was compared to the experimentally derived signatures. Both of the experimentally performed signatures were significantly better at discriminating outcome in the METABRIC data than the weighted signatures for randomly assigned genes (for ER+ tumors p = 0.035 for the metastasis-correlated signature, p <0.001 for the DMFS signature).
Next, the identity of genes within the metastasis-correlated and DMFS signatures were held constant, but the weights were randomly generated by the pseudo random generator following normal distribution. This process was also permuted 1000 times and compared to the performance of experimentally derived signatures. The experimentally derived signatures also outperformed the signatures with random weights (for ER+ tumors p = 0.048 for the metastasis- correlated signature, p < 0.001 for the DMFS signature). Finally, we generated random weights with normal distribution and randomly selected gene sets of size 656 and 383 respectively and tested the random signatures 1000 times. Once again, the experimentally derived signatures were significantly better than the permuted data (for ER+ tumors p = 0.028 for the metastasis- correlated signature, p < 0.001 for the DMFS signature). Of note, the experimentally recalculated DMFS signature was also significantly better than the permuted signatures under all three conditions for patients with ER- as well as ER+ tumors (S3 Table). These data suggest that the genes identified by the integrated mouse subtractive strategy were unlikely to have been implicated with metastatic disease by chance.
To evaluate the subtraction strategy, the gene lists were compared with existing data. Linkage analysis previously identified the presence of a metastasis modifier locus on NZB/B1NJ chromosome 9 [7, 8], located 16 to 67 megabases distal to the centromere [5] (Fig 3A), containing approximately 1300 genes: ~ 800 annotated genes and ~ 500 predicted genes. Limiting the subtracted gene lists to this interval further reduced the number of candidates to 6 to 25 genes (Table 2). Encouragingly, one of the DMFS genes was Cadm1, a previously identified metastasis susceptibility gene [5]. For further validation, the gene closest to the modifier peak, Pvrl1, was tested. shRNA knockdowns of Pvrl1 in two independent mouse mammary tumor cell lines was performed. Knockdown of Pvrl1 had inconsistent effects on tumor growth and in vitro cell proliferation, but consistently reduced metastatic disease (S5 Fig). Normalization of metastatic burden by tumor weight to account for differences in tumor weight in vivo still resulted in significant differences between control and knockdown cells (Fig 3B), consistent with Pvrl1 being a tumor progression gene.
To evaluate a candidate gene identified only in the G7 population screen, the gene Zbtb16 was selected (Fig 3A). Zbtb16 was previously implicated as a candidate due to its membership in a proliferation-associated gene network that is predictive of metastatic disease [26]. However, orthotopic implantation of cells with Zbtb16 overexpression did not show any differences in metastatic disease (Fig 3C) suggesting Zbtb16 might have a tumor cell non-autonomous effect. Zbtb16 knockout animals [27] were therefore bred to MMTV-PyMT animals to generate PyMT+/Zbtb16+/- or PyMT+/Zbtb16+/+ animals. Zbtb16 heterozygotes showed increased incidence of metastasis (p = 0.04; Fig 3D), but did not display a statistically significant change in metastatic burden (S6A Fig), likely due to high experimental variability. Tumor weights from Zbtb16+/- animals were not significantly different compared to tumors from their wild type littermates (p = 0.82; S6B Fig) indicating the effect of Zbtb16 was unlikely due to differences in tumor cell proliferation. Consistent with a tumor cell non-autonomous role, injection of wildtype mammary tumor cells into wildtype or Zbtb16+/- animals showed a significant increase in pulmonary colonization in the heterozygous animals (Fig 3E) without significant effect on primary tumor burden (p = 0.81; S6C Fig).
To further validate this method, the genome-wide gene lists were examined to try and identifying metastasis-associated genes outside of the metastasis-associated susceptibility peaks. Analysis discovered a number of genes previously identified as playing important roles in metastatic disease (e.g., Tpx2 [28], Angptl4 [29], Ezr [30], Txnip [31]), indicating this strategy is capable of identifying putative metastasis genes on a genome-wide scale.
To identify the cellular and molecular pathways contributing to metastatic disease, Ingenuity Pathway Analysis (IPA) was performed. The most significant canonical pathways for genes associated with metastasis in the G5 DO population was antigen presentation for the DMFS signature (p = 4.7x10-8) and mitotic pathways for the metastasis-correlated signature (p = 1.28x10-4–1.07x10-6; S3 Table). These results are consistent with our [5] and other laboratories’ [32–34] findings showing a significant role for immunity or cellular proliferation in breast cancer progression. In contrast, significant pathways associated with metastatic progression in the G7 DO population included IL-17 pathways for the DMFS gene signature and metabolic and rheumatoid arthritis pathways for the metastasis-correlated genes, including diabetes signaling (p = 7.23x10-4–2.3x10-4; S4 Table).
IPA analysis was also performed to identify potential upstream regulatory genes that contribute to prognostic signatures and tumor progression. All four gene sets implicated TGFβ1 as an important upstream regulator that suppresses metastatic disease (p = 3.7x10-3–9.3x10-18), consistent with the role of TGFβ in early tumor progression [35]. Estrogen receptor-α and -β, and the progesterone receptor were also identified as significant upstream regulators in both analyses (S5 Table). These results further support the utility of the pDHS strategy in identifying metastasis-relevant pathways and implicate additional pathways for investigation.
Ingenuity Pathway Analysis of the gene lists was then carried out to identify potential clinically actionable targets or drugs for metastatic therapy. Tamoxifen, a current standard of care therapeutic for breast cancer treatment, was associated with suppression of metastatic disease in both the DMFS and metastasis-correlated analyses (S1 Table), suggesting this strategy can identify clinically relevant therapeutics. In addition, the cannabinoid receptor 1 gene, CNR1, was found to be associated with metastatic suppression (S5 Table). Consistent with this possibility, an independent study showed that synthetic cannabinoids suppress tumor growth and metastasis in the MMTV-PyMT model [36].
Moreover, the data suggest that 8-bromo-cAMP, a cell permeable cAMP analog, is associated with metastatic suppression (S5 Table). Increased cAMP levels are a downstream consequence of caffeine (a non-specific phosphodiesterase inhibitor [37]) metabolism and interestingly cAMP was previously shown in our laboratory to be a metastasis-suppressing agent [38]. Treatment of a highly metastatic mouse mammary tumor cell line with 8-bromo-cAMP induced a gene signature that was an independent predictor of both DMFS and overall survival in GOBO datasets (S7 Fig), indicating this pathway may be a useful clinical target.
Finally, the IPA analysis not only implicated the diabetes signaling pathway as significantly associated with metastatic disease, but also a number of agents used to treat diabetes that might suppress metastatic disease. To test this prospect within our experimental systems, metastatic mouse mammary tumor cells were implanted into immunocompetent mice. The mammary tumors were permitted to grow until established (10 days post-injection) and then the mice were treated with rosiglitazone until euthanasia (28 days post-injection). As predicted from the integrated analysis, the rosiglitazone-treated group had 25% fewer pulmonary metastases compared to the control group (Fig 4; p = 0.028), consistent with previous reports for the LMM3 mammary tumor cell line [39].
We have integrated quantitative trait genetics, transcriptional epigenetics, gene expression, and computational biology tools in a mouse model of metastatic mammary cancer to identify factors that contribute to inherited susceptibility to breast cancer metastasis. This approach significantly enriched for genes associated with metastatic disease as measured by the generation of gene signatures with prognostic value in human patients. Interestingly, our analyses has resulted in signatures that are prognostic in both early relapse and late relapse clinical scenarios (Fig 2A). The majority of gene expression signatures currently available for human breast cancer patients are limited because they are only associated with early relapse, hence this type of integrated mouse-human analysis could complement existing data as it provides additional insights into the later stages of metastatic disease.
Importantly, all of the gene signatures described herein are weighted and must contain genes that transcribe in the same direction with equal relative potency in both mouse and human. The ability of the mouse signatures to maintain prognostic ability in human samples supports the assumption that the underlying biology of metastasis is very similar between these two species. If this were not the case, one would not expect a weighted signature to preserve its prognostic ability across species. These results therefore provide additional support for the application of metastatic mouse models to characterize and identify metastasis genes.
Our results also illustrate an important caveat to this hypothesis, described as such: because genetic and molecular analysis was performed using a treatment-naïve mouse model, the results should represent the natural course of the disease. In contrast, patients are commonly treated using surgical resection of the primary tumor combined with subsequent radiation and adjuvant therapies to reduce the risk of recurrence. Since adjuvant therapy reduces the number of patients who develop metastatic disease by 20–30% [40] and salvage therapy prolongs survival for patients who develop metastasis, the inability of the DMFS signature to predict overall survival in treated METABRIC patients without re-optimization is not surprising. Furthermore, the gene signature difference between treatment-naïve and treated patients may also contribute to the conflicting association results for SIPA1, which is associated with treatment-naïve DMFS samples [41, 42], but not overall survival [43]. Hence accounting for the two different phenotypes (treatment-naïve DMFS vs. overall survival) is an important consideration when translating mouse model data into clinical useful human observations.
Furthermore, the gene signatures and associated gene lists derived from this analysis should be enriched for metastasis-associated genes. Prognostic gene signatures are usually derived by correlating gene expression with metastatic disease without distinguishing whether the associated gene expression is etiological or a secondary effect. The requirement for candidate genes to have polymorphisms within putative enhancers and promoters should significantly enrich for genes with primary effects. The ability of this strategy to both re-identify known metastasis susceptibility genes (e.g., Cadm1) and validate novel tumor progression genes (e.g., Pvrl1, Zbtb16) is consistent with this hypothesis. The resulting gene lists will therefore provide greater utility for identifying molecular and cellular pathways associated with the metastatic process.
Finally, the integrated strategy described herein allows for the identification of novel agents with anti-metastatic therapy potential. In addition to re-identifying agents that are the current standard of care (tamoxifen) and validating experimental drugs (cannabinoids [36]), this technique also identified FDA approved anti-diabetic drugs as potentially useful therapeutics. Of interest, orthotopic implantation assays confirmed the ability of one of these drugs, rosiglitazone, to reduce metastatic burden [39]. Investigations of other agents used to treat type II diabetes is also supported by the results obtained in this metastatic mammary tumor model. Increased survival in ER+/HER2+ breast cancer patients [44, 45] and triple-negative breast cancer [46] treated with the anti-diabetic drug metformin have been reported, as well as investigations regarding its ability to increase survival in other forms of malignancy [47]. These data support the ability of this integrated analysis to identify potentially useful tools for clinical use and further support the complementary use of animal models systems to understand the complex biology of metastatic disease.
The research described in this study was performed under the Animal Study Protocols LPG-002 and LCBG-004, approved by the NCI Bethesda Animal Use and Care Committee. Animal euthanasia was performed by cervical dislocation after anesthesia by Avertin.
NZB/B1NJ and NZW/LacJ DNA was obtained from The Jackson Laboratory DNA Repository. Library preparation and sequencing on Illumina HiSeq instruments was performed following the manufacturers recommended protocols. Four lanes of 102 base paired end sequence per strain was performed to achieve approximately 40x coverage. Reads were aligned to the reference genome (GRCm38) using BWA version 0.7.5a-r406 [19451168]. All lanes from the same library were then merged into a single BAM file using Picard tools and PCR duplicates were marked using Picard 'MarkDuplicates' [19505943]. Local realignment was carried out using GATK-v3.0 using default parameters to generate the set of intervals for realignment [20644199]. SNP and indel discovery was performed with the Samtools v1.1 (samtools mpileup -t DP,DV,DP4,SP,DPR,INFO/DPR -E -Q 0 -pm3 -F0.25) and calling was performed with Bcftools call v1.1 (bcftools call -mv -f GQ,GP -p 0.99). Raw sequencing data is available under accession ERP000927.
DNase hypersensitive site (DHS) data for the mouse 3134 mammary adenocarcinoma cells was downloaded from the Encode project websites.
http://hgdownload.cse.ucsc.edu/goldenPath/mm9/encodeDCC/wgEncodeUwDnase/
http://hgdownload.cse.ucsc.edu/goldenPath/mm9/encodeDCC/wgEncodeUwDgf/
Genotype data for FVB/NJ was downloaded from the Welcome Trust Sanger Centre website: http://www.sanger.ac.uk/resources/mouse/genomes/ The NZB/B1NJ and NZW/LacJ genotype data is also available at this site.
All computations were performed on NIH helix/biowulf system, documentation of which is available at https://helix.nih.gov. We used R computing environment, perl scripts, bedtools, and ucsc liftOver for most of the analyses.
The workflow consisted of the following. The UCSC liftOver tool was used to convert between mm9 and mm10 as necessary when using bedtools intersectBed to intersect two bed files. 1) The Encode DHS data were filtered for the regions overlapping with polymorphic sites. Since the DHS data were generated in Genome Build mm9, we used UCSC mm9 snp128 data to restrict the DHS sites. 2) The mice NZB/B1NJ genotype data in vcf was filtered to retain the SNPs that overlap with the DHS present in the 3134 cells. 3) We then removed SNPs in the DHS that are present in the mouse FVB/NJ strain.
BED files containing the subtracted NZB/B1NJ specific, DHS associated SNPs were loaded into the GREAT tool website using the default settings for gene assignment. GREAT calculates statistics by associating genomic regions with nearby genes and applying the gene annotations to the regions. Association is a two-step process. First, every gene is assigned a regulatory domain. Then, each genomic region is associated with all genes whose regulatory domain it overlaps. The default association settings included assignment of basal regulatory elements 5 kb upstream and 1 kb downstream of transcriptional start sites (regardless of other nearby genes). In addition, the gene regulatory domain was extended up to 1 megabase in both directions to the nearest gene's basal domain but no more than the maximum extension in one direction [16].
The Diversity Outcross x MMTV-PyMT G5 CEL files have been deposited in the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo) under the accession number GSE48565. The G7 CEL files are deposited under the accession number GSE64522. Data was imported into BRB ArrayTools version 4.3.2 (http://brb.nci.nih.gov/BRB-ArrayTools/) and normalized using the median array as a reference. Batch correction was not performed before BRB ArrayTool analysis. Only genes with a Log Intensity Variation of p< 0.05 were considered for further analysis. Distant metastasis free survival analysis of these data was performed using the Find Genes Associated with Survival tool (DMFS genes). Genes were considered associated with DMFS if p< 0.05. Genes correlated with metastatic disease was performed using the Spearman Correlation Test option of the Find Genes Associated with Quantitative Trait tool (metastasis correlated genes). Genes with a p< 0.05 were considered to be significantly associated with metastatic disease.
2x105 Mvt1[48] mammary tumor cells were plated in 6 well dishes. 24 hours later the cells were treated with 500 uM 8-Br-cAMP (Sigma) dissolved in 0.1% DMSO for 24 hours, then RNA harvested using Trizol following the manufacturers recommended protocol. Vehicle control or 8-Br-cAMP samples were performed in triplicate. Transcriptome analysis performed by the NCI Laboratory of Molecular Technology using Affymetrix MOE430 v2 chips. The data was analyzed using the Class Comparison tool of BRB ArrayTools. For the gene signature analysis, the differentially expressed genes were filtered for those with greater than 10-fold change in expression, either up- or down-regulated, compared to the vehicle alone. The fold change in gene expression was used to weight the individual genes, and the ability of the weighted gene signature to discriminate breast cancer patient outcome assessed using the webtool GOBO (http://co.bmc.lu.se/gobo/).
The 6DT1 and Mvt1 cell lines[48] were obtained from the laboratory of Robert Dickson, George Washington University. Microsatellite genotyping validated that these cell lines originated from an FVB/NJ animal. Both cell lines have been demonstrated to be mycoplasma-free.
shRNA lentiviral vectors were purchased from Sigma-Aldrich (cat. # SHCLNG-NM_021424). Stable Pvrl1 knockdown cell lines were generated by lentiviral transduction into Mvt1 cells and knockdown validated by qRT-PCR. 8x105 Mvt1 or 105 6DT1 cells were inoculated into the four mammary fat pad of 6–8 week old FVB/NJ female mice, 10 animals per group. Animals were euthanized at 5 weeks (experiment 1) or 4 weeks (experiment 2) after implantation. One-way Anova with Dunnett’s correction for multiple testing was performed for each experiment using GraphPad Prism. The results of the replicate experiments were then combined using Fisher’s combined probability test. All procedures were performed as approved by the NCI-Bethesda Animal Care and Use Committee.
Epitope-tagged Zbtb16 over-expression was performed by lentiviral transduction into the cell line Mvt1. Orthotopic implantation was carried out as described above and analyzed using the Mann-Whitney test in GraphPad Prism. Autochthonous tumor assays were performed by breeding MMTV-PyMT male animals to Zbtb16+/- female animals to generate PyMT+/Zbtb16+/- female animals. 20 PyMT+/Zbtb16+/- animals was selected for analysis based on previous experience which suggested this as an appropriate group size to achieve statistical significance. Statistical differences were assessed using the Mann-Whitney test in GraphPad Prism. All procedures were performed as approved by the NCI-Bethesda Animal Care and Use Committee.
100,000 cells/mouse of the mammary tumor cell line 6DT1 were orthotopically implanted into the fourth mammary fat pad of female FVB/NJ mice (6–8 weeks old). 10 days post-implantation all of the animals were combined into a single cage then randomly assigned to treatment or control group by alternating assignment to new cages. Roziglitazone (100μM) or vehicle (DMSO, 0.17%) was added to drinking water 10 days post tumor implantation and available to mice ad libitum. Drinking water with Roziglitazone or DMSO was refreshed every week. Tumor growth was monitored and animals were euthanized 28 days post implantation. Tumors and lungs were evaluated for weight and surface metastases respectively. The experiment was performed twice and the results of the Mann-Whitney test combined using Fisher combined probability test. All procedures were performed as approved by the NCI-Bethesda Animal Care and Use Committee.
NZB/B1NJ sequence polymorphism data available data is available at http://www.sanger.ac.uk/resources/mouse/genomes/ Raw data is available at http://sra.dnanexus.com/studies/ERP000927. The Diversity Outcross x MMTV-PyMT G5 CEL files have been deposited in the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo) under the accession number GSE48565. The G7 CEL files are deposited under the accession number GSE64522.
|
10.1371/journal.pgen.1005420 | The Nucleosome Acidic Patch Regulates the H2B K123 Monoubiquitylation Cascade and Transcription Elongation in Saccharomyces cerevisiae | Eukaryotes regulate gene expression and other nuclear processes through the posttranslational modification of histones. In S. cerevisiae, the mono-ubiquitylation of histone H2B on lysine 123 (H2B K123ub) affects nucleosome stability, broadly influences gene expression and other DNA-templated processes, and is a prerequisite for additional conserved histone modifications that are associated with active transcription, namely the methylation of lysine residues in H3. While the enzymes that promote these chromatin marks are known, regions of the nucleosome required for the recruitment of these enzymes are undefined. To identify histone residues required for H2B K123ub, we exploited a functional interaction between the ubiquitin-protein ligase, Rkr1/Ltn1, and H2B K123ub in S. cerevisiae. Specifically, we performed a synthetic lethal screen with cells lacking RKR1 and a comprehensive library of H2A and H2B residue substitutions, and identified H2A residues that are required for H2B K123ub. Many of these residues map to the nucleosome acidic patch. The substitutions in the acidic patch confer varying histone modification defects downstream of H2B K123ub, indicating that this region contributes differentially to multiple histone modifications. Interestingly, substitutions in the acidic patch result in decreased recruitment of H2B K123ub machinery to active genes and defects in transcription elongation and termination. Together, our findings reveal a role for the nucleosome acidic patch in recruitment of histone modification machinery and maintenance of transcriptional integrity.
| Chromatin, a complex of DNA wrapped around histone proteins, impacts all DNA-templated processes, including gene expression. Cells employ various strategies to alter chromatin structure and control access to the genetic material. Nucleosomes, the building blocks of chromatin, are subject to a myriad of modifications on their constituent histone proteins. One highly conserved modification with important connections to human health is the addition of ubiquitin to histone H2B. H2B ubiquitylation modulates chromatin structure during gene transcription and acts as a master regulator for downstream histone modifications. The proteins that promote H2B ubiquitylation have been identified; however, little is known about how these proteins interface with the nucleosome. Here, we exploited the genetic tools of budding yeast to reveal features of the nucleosome that are required for H2B ubiquitylation. Our genetic screen identified amino acids on the nucleosome acidic patch, a negatively charged region on the nucleosome surface, as being important for this process. The acidic patch is critical for regulating chromatin transactions, and, in our study, we identified roles for the acidic patch throughout transcription. Our data reveal that the acidic patch recruits histone modifiers, regulates histone modifications within the H2B ubiquitylation cascade, and maintains transcriptional fidelity.
| In eukaryotes, transcription and other nuclear processes take place in the context of chromatin. The basic unit of chromatin is the nucleosome, which consists of approximately 147 base pairs of DNA wrapped around a histone octamer, containing two copies of each of the four core histone proteins: H2A, H2B, H3, and H4 [1]. Histones are decorated with posttranslational modifications, which can alter chromatin architecture and recruit a wide range of proteins to the genome, thus regulating all chromatin transactions [2]. In addition to their intrinsic effects on modulating the chromatin template, certain histone modifications can promote other histone modifications, either on the same histone (cis-regulation) or on a different histone (trans-regulation) in a process termed “histone crosstalk” [3].
The monoubiquitylation of H2B on lysine 123 (H2B K123ub) in S. cerevisiae is associated with active gene transcription, impacts global nucleosome occupancy and plays important roles in transcription elongation, telomeric silencing, DNA replication, and DNA repair [4]. In yeast, this modification is catalyzed by the ubiquitin-protein ligase Bre1 and the ubiquitin-conjugating enzyme Rad6 [5–7]. In humans, the analogous lysine, H2B K120, is ubiquitylated by RNF20/RNF40 and RAD6A/RAD6B [8,9]. In one of the best-studied examples of histone crosstalk, H2B K123ub is required for other histone modifications associated with active transcription: H3 K4 and H3 K79 di- and tri-methylation [10–12]. H3 K4 dimethylation, which is enriched at the 5'-ends of coding regions, and H3 K4 trimethylation, which is associated with active promoters, regulate histone acetylation patterns on genes by directing the recruitment of histone acetyltransferases and histone deacetylases [13]. H3 K79 methylation occurs across active genes, and dimethylation of this residue locally alters the nucleosome surface [14,15]. All of these histone modifications are conserved in higher eukaryotes, and disruption of these modifications can result in a range of human diseases, including cancer [16].
In addition to Rad6 and Bre1, several protein complexes that regulate transcription elongation and nucleosome dynamics are required for wild-type levels of H2B K123ub. These include the Bur1-Bur2 cyclin-dependent kinase complex and the FACT histone chaperone complex [17–19]. Additionally, the Polymerase Associated Factor 1 complex (Paf1C), which travels with RNA pol II and Spt5 during transcription elongation, promotes H2B K123ub through the Rtf1 subunit of the complex [20–23]. While protein complexes that promote H2B K123ub have been identified, little is known about how the nucleosome itself promotes H2B K123ub.
We previously reported that the ubiquitin-protein ligase Rkr1/Ltn1 is required for the viability of yeast cells that lack the RTF1 gene or harbor an amino acid substitution for H2B K123 that prevents ubiquitylation (H2B-K123R) [24]. Rkr1/Ltn1 associates with ribosomes and degrades nonstop proteins [25,26]. The genetic interactions between rtf1∆, H2B-K123R, and rkr1∆ suggest a requirement for the quality control functions of Rkr1 in the absence of an intact H2B ubiquitylation pathway. We reasoned that the negative genetic interactions between rkr1∆ and H2B-K123R could be exploited to identify histone residues that are required for H2B K123ub. Using a genetic screen, we identified H2A and H2B residues required for proper H2B K123ub and downstream histone modifications. Many of these residues map to the acidic patch on the surface of H2A. We found that amino acid substitutions in the acidic patch cause defects in the recruitment of the H2B K123ub machinery to active genes, an accumulation of read-through transcripts, and altered transcription elongation efficiency in vivo. Interestingly, the substitutions differentially impact histone modifications downstream of H2B K123ub. Therefore, while the H2A acidic patch residues functionally converge in regulating H2B K123ub, they diverge in regulating downstream histone modifications. Our data reveal a requirement for the nucleosome acidic patch in H2B K123ub and argue that this exposed nucleosome surface serves as an important protein docking site in which individual residues uniquely contribute to the regulation of histone modifications and gene expression.
To identify histone residues required for H2B K123ub in S. cerevisiae, we screened a comprehensive histone mutant library [27] for alanine substitutions in H2A and H2B that cause synthetic lethality or sickness when combined with a deletion of the RKR1 gene. We previously showed that rkr1∆ is synthetically lethal in strains carrying H2B-K123R as the only form of H2B [24]. Using a plasmid shuffle strategy, HIS3-marked hta1-HTB1 or HTA1-htb1 plasmids from the library were transformed into a rkr1∆ deletion strain, replacing a URA3-marked plasmid carrying wild-type copies of HTA1 and HTB1. The URA3-marked wild-type plasmid was counter-selected on medium containing 5-fluoroorotic acid (5-FOA). Relative to their effects on a strain containing a wild-type RKR1 gene, nine histone mutant plasmids caused enhanced growth defects in the rkr1∆ background (Fig 1A). Eight of the amino acid substitutions were located in H2A, and one was H2B-K123A (Fig 1A). Identification of H2B-K123A served as a validation of our screen.
Many of the residues identified in our screen cluster within the nucleosome acidic patch (Fig 1B and 1C). The acidic patch serves as a binding site for several proteins, including the H4 tail of neighboring nucleosomes [1,28–30]. In addition to those in the acidic patch, two residues, L86 and H113, reside near the docking domain of H2A [1].
To test if the amino acid substitutions in H2A cause H2B K123ub defects, we assessed global H2B K123ub levels by western blot analysis. Because the plasmids in the H2A and H2B mutant libraries encode FLAG-tagged H2B [27], we initially used anti-FLAG western blots to distinguish H2B K123ub from unmodified H2B as a super-shifted band. We subsequently turned to a commercial antibody against human H2B K120ub, which can specifically detect yeast H2B K123ub [31] (Fig 2). Surprisingly, this antibody did not recognize FLAG-tagged H2B K123ub to the same degree as untagged H2B K123ub in our strains, raising concerns that the FLAG tag could influence H2B ubiquitylation or our ability to detect this modification (S1 Fig). Therefore, we removed the FLAG tag from all of the plasmids carrying hta1-HTB1 mutations identified in our screen, and we continued with these constructs for all experiments in this study. The western analysis revealed that all of the H2A mutants have reduced global H2B K123ub levels compared to the wild-type control strain; however, the different substitutions affect H2B K123ub levels to varying degrees (Fig 2A). For example, there is a striking difference in H2B K123ub levels in strains harboring substitutions of the neighboring residues H2A-E65 and H2A-L66 (Fig 2A, lanes 4 and 5). Our result reflects the H2B K123ub defect previously observed for an H2A-L66A mutant [27]; however, with removal of the FLAG tag, we now detect a defect in H2B K123ub in the H2A-E65A mutant as well.
To measure chromatin-associated levels of H2B K123ub, we performed chromatin immunoprecipitation (ChIP) analysis of H2B K123ub and total H2B at active genes (PYK1 and PMA1) and, as a control, at a non-transcribed region (TELVI). We normalized levels of H2B K123ub to levels of total H2B to correct for any defects in H2B occupancy (Fig 2B). For these ChIP analyses, and most other experiments in this study, we focused our efforts on H2A residues E57, E65, L66, L86, and E93, because residues F26 and L94 are buried within the protein core of the nucleosome and could be impacting H2B K123ub levels indirectly (Fig 1A). We also chose not to focus on H113, because it is not conserved in higher eukaryotes. In agreement with the western analyses, the ChIP assays revealed reduced H2B K123ub levels on active genes in the H2A mutant strains (Fig 2B). However, gene-specific defects are evident. For example, the E57A substitution causes an H2B K123ub defect at PYK1 but not ADH1 or PMA1 (Figs 2B and S2). Together these data demonstrate that the nucleosome acidic patch promotes H2B K123ub globally and at specific genes.
Previous studies have shown that H2B K123ub is required for proper histone occupancy [32,33], that the docking domain of H2A is important for the association of H2A and H2B with H3 and H4 [33–35], and that the acidic patch lies at the interface of H2A and H2B [1,35]. Therefore, we examined global and local levels of histones by western analysis and ChIP, respectively (Fig 3). Global levels of H2B, H3, and H2A were unaffected in the mutants, with two exceptions (Fig 3A). The two exceptions, H2A-E93A and H2A-L94A, were detected at levels that were lower than wild-type H2A, indicating a potential defect in the expression, stability, or antibody recognition of these H2A mutant proteins. H2B, H2A, and H3 occupancy levels were assessed at both the highly transcribed gene PYK1 and a non-transcribed telomeric region using ChIP analysis (Fig 3B–3D). Four of the alanine substitutions in H2A resulted in lower occupancy levels of H2B at PYK1 (Fig 3B). H2A occupancy was not as drastically affected in the mutant strains; however, the signals for H2A-E57A and H2A-E93A enrichment were reduced at all loci tested (Fig 3C). For H2A-E93A, this could be due to reduced H2A protein levels or reduced immunoreactivity (Fig 3A). H3 occupancy levels at PYK1 were also slightly affected in some of the mutant strains, particularly at the 5’ end of the gene (Fig 3D). Importantly, the reduced histone occupancy levels do not account for the reduced H2B K123ub levels in the H2A mutant strains, as we have normalized the H2B K123ub levels to total histone levels in our assays (Fig 2).
As an alternative measure of chromatin integrity in the histone mutant strains, we used northern analysis to monitor transcription of the SER3 and FLO8 genes, which can serve as sensitive reporters of defects in chromatin structure [36–38]. In rich media, SER3 expression is repressed by transcription-coupled nucleosome assembly over its promoter via transcription of a noncoding RNA, SRG1 [39,40]. Mutations in the genes encoding the histone chaperones Spt6 and Spt16 lead to strong derepression of SER3 without decreasing SRG1 transcription [39]. Relative to the temperature-sensitive alleles spt6-1004 and spt16-197, the H2A substitutions identified in our screen do not cause strong derepression of SER3, suggesting that transcription-coupled nucleosome occupancy is largely intact over SRG1 (Fig 3E).
Cryptic initiation can occur when cryptic promoters within coding regions are unveiled by perturbations in nucleosome occupancy or histone modifications [36–38]. To assess cryptic initiation in the H2A mutants, we performed northern analysis of the FLO8 gene, using spt6-1004 and spt16-197 as positive controls for cryptic initiation (Fig 3E). Relative to these control strains, the H2A mutants generate only very low levels of cryptic transcripts at FLO8 (Fig 3E). Together, these data suggest that, although histone occupancy defects can be detected, chromatin structure is not grossly impaired in the H2A mutants.
H2B K123ub is required for downstream histone modifications, including H3 K4 di- and tri-methylation (H3 K4me2/3), catalyzed by Set1, and H3 K79 di- and tri-methylation (H3 K79me2/3), catalyzed by Dot1 [10,11,13]. We therefore asked whether the H2A substitutions also cause defects in modifications downstream of H2B K123ub, using western analysis. Surprisingly, although all of the H2A mutants identified in our screen have reduced H2B K123ub levels, we observed a range of defects in H3 methylation (Fig 4). For example, two substitutions, F26A and H113A, cause no apparent defects in global H3 K4 or K79 methylation, despite dramatically reducing H2B K123ub levels (Figs 2A and 4A). In contrast, the E65A and L66A substitutions greatly reduce H3 K4 methylation and partially reduce H3 K79 methylation even though their effects on H2B K123ub levels are quite different (Figs 2A and 4A). Substitution of residues E93 and L94 to alanine resulted in a strong H3 K79 methylation defect and only slight defects in H3 K4 methylation (Fig 4A). Thus, E93 and L94 appear to selectively impact H3 K79 methylation. To determine the levels of H3 methylation on chromatin, we performed ChIP analysis of H3 K79me2/3 and H3 K4me3 at PYK1, PMA1 and TELVI in the H2A mutant cells and normalized the data to total H3 occupancy levels (Fig 4B). The modification defects observed by ChIP mirror the global H3 methylation defects visualized by western analysis with slight differences being likely due to differences in histone occupancy levels, which were taken into account in the ChIP assay. Our results indicate that the H2A residues play unique roles in regulating histone modifications dependent on H2B K123ub.
To test whether the H2A substitutions confer other histone modification defects potentially through a general change in nucleosome structure, we performed western analysis of Set2-catalyzed H3 K36me2 and K36me3, modifications that are not strongly dependent on H2B K123ub [12,20]. None of the H2A mutants exhibited defects in H3 K36 methylation (Fig 4A). This is in agreement with previous work, which identified a distinct nucleosome surface required for H3 K36 methylation [41], and the idea that the H2A substitutions identified in our screen are largely specific to the H2B K123ub cascade.
Previous studies have shown that H3 K4me3 and H2B K123ub are required for proper transcription termination of small nucleolar RNAs (snoRNAs) through the Nrd1-Nab3-Sen1 pathway [42–44]. However, little is known about how these histone modifications or other nucleosome residues affect transcription termination. To assess transcription termination in our mutants, we performed RT-qPCR analysis on four snoRNA genes that are affected by histone modifications [44]. For these assays, we used probes that hybridize to the intergenic region between the snoRNA gene and the downstream gene. Detection of a PCR product is a measure of transcription in the region downstream of the snoRNA terminator (S3 Fig). The RT-qPCR analysis indicates that the H2A acidic patch residues are required for proper transcription termination at the four snoRNA loci (S3 Fig). Previous work described snoRNA termination to be differentially sensitive to disruption of H2B K123ub: SNR47 required H2BK123ub for proper termination whereas SNR48 was relatively insensitive to the absence of this mark [44]. The mutants identified in our screen, which all have abrogated H2B K123ub, have termination defects at both loci, indicating that the mechanistic basis for read-through of these terminators could be downstream of H2B K123ub (S3 Fig).
H2B K123ub is a transient histone modification; therefore one possible explanation for reduced H2B K123ub levels in the H2A mutants could be through decreased stability of the mark through the enhanced action of a ubiquitin-specific protease. The removal of H2B K123ub is due to the actions of two ubiquitin-specific proteases Ubp8 and Ubp10 [45–48]. To test whether the H2B K123ub deficiency observed in the H2A mutants is through decreased stability of the modification, we performed western blot analysis of H2B K123ub levels in strains that contain the H2A substitutions and are deleted for UBP8. Upon deletion of UBP8, the fold recovery of H2B K123ub levels was comparable to that of wild-type cells for the H2A-L86A and H2A-E93A mutants, suggesting that the H2B K123ub defect in these mutants is at least partially due to decreased stability of the mark (Fig 5A and 5B). For the H2A-E57A, H2A-E65A, and H2A-L66A mutants, deletion of UBP8 did not fully rescue H2B K123ub levels (Fig 5A and 5B). The most drastic effect was that of H2A-L66A, where little to no H2B K123ub was restored. Therefore, for these mutants, and especially H2A-L66A, the defect in H2B K123ub is likely due to a failure of the ubiquitylation machinery to fully establish the mark. An alternative, but not mutually exclusive explanation, is that E57, E65, and L66 could form a surface required for Ubp8 function or recruitment, as these three residues reside near each other on the nucleosome structure (Fig 1B).
To test the extent to which H2B K123ub and downstream H3 methylation events are coupled in the H2A mutant strains, we measured H3 K4me3 and H3 K79me2/3 levels in the presence and absence of UBP8. Upon deletion of UBP8, H3 K4me3 and H3 K79me2/3 increased in the wild-type strain and in the H2A-L86A mutant (Fig 5C). When normalized to total H3 levels, an increase in H3 K4me3 and K79me2/3 levels was not detected in the H2A-E57A mutant upon deletion of UBP8, which is consistent with the poor recovery in H2B K123ub in this strain (Fig 5). This result suggests there is a correlation between K123ub and downstream marks in the H2A-E57A mutant when UBP8 is deleted. Similarly, for the H2A-L66A mutant, no recovery of the methyl marks was observed in the ubp8∆ background, which corresponds to the severe defect in K123ub in this mutant. This observation is consistent with the idea that the establishment of H2B K123ub is the primary defect in this mutant. For the E65A mutant, H3 K4me3 levels were extremely low in both the presence and absence of Ubp8, even though H2B K123ub levels were substantially recovered in the ubp8∆ background. This observation suggests that the E65A substitution prevents proper H3 methylation possibly by disrupting a functional interaction with the Set1/COMPASS complex. Finally, in agreement with our western and ChIP results (Fig 4), the E93A mutant appears most defective in supporting H3 K79 methylation, as deleting UBP8 elevated H3 K4me3 levels to a greater extent than H3 K79me2/3 levels in this strain.
In addition to decreased stability of the ubiquitylation mark conferred by Ubp8, the reduction in histone modification levels in the H2A mutants could be due to impaired recruitment of the modification enzymes required for the H2B K123ub cascade, such as the ubiquitin-protein ligase Bre1. To analyze the effects of the H2A substitutions on recruitment of Bre1 to actively transcribed genes, we performed ChIP analysis of HSV-tagged Bre1 (Fig 6A). All five of the H2A mutants tested showed reduced recruitment of HSV-Bre1 to PYK1 and PMA1, particularly at their 5’ ends (Fig 6A). With the exception of the H2A-E57A mutant, Bre1 occupancy was also reduced at ADH1 (S2 Fig). As expected, HSV-Bre1 levels at the non-transcribed TELVI region were similar to those of the untagged control strain. Also in agreement with previous observations [7], Bre1 levels at the 5’ ends of PMA1 and PYK1 were higher than those at the 3’ ends of the genes. To determine whether reduced levels of HSV-Bre1 could account for the reduced HSV-Bre1 occupancy in the H2A mutant strains, we performed western analysis. Our results show that total HSV-Bre1 levels in the H2A mutants are similar to those in a wild-type strain (S4 Fig). These results indicate that residues in the H2A acidic patch are required for proper Bre1 recruitment to active genes.
The Paf1C subunit, Rtf1, has been implicated in the recruitment of the H2B ubiquitylation machinery during transcription [49]. We therefore used ChIP analysis to test whether the H2A residues that are important for Bre1 recruitment are also important for Rtf1 occupancy on active genes. Our ChIP results demonstrate a significant reduction in Rtf1 levels at PYK1, PMA1, and ADH1 in the H2A mutant strains (Figs 6B and S2). To rule out the possibility that the reduced Rtf1 occupancy is a result of lower protein levels, we measured global Rtf1 levels by western analysis. This analysis showed that Rtf1 levels are unaffected in the H2A mutants, indicating that reduced Rtf1 expression is not the cause of the H2B K123ub defect (S4 Fig). Overall, the occupancy levels of HSV-Bre1 and Rtf1 correlated with H2B K123ub levels in some cases but not others. For example, the H2A-E57A mutant shows reduced HSV-Bre1 and Rtf1 occupancy but normal levels of H2B K123ub at the PMA1 locus. It is possible that small levels of Bre1/Rad6 and Rtf1 are sufficient to promote H2B K123ub at PMA1 in this mutant. Alternatively, decreased Ubp8 levels or activity could compensate for reduced Bre1 recruitment. We attempted to test this idea by ChIP but were unable to reliably measure Ubp8 occupancy in our strains.
We previously demonstrated that recruitment of Paf1C to coding regions is mediated through a direct physical interaction between Rtf1 and the elongation factor Spt5 [50,51]. Therefore, it is possible that the lower Rtf1 and Bre1 occupancy levels in the H2A mutant strains reflect impaired recruitment of the transcription elongation machinery. To test this idea, we performed ChIP analysis of Spt5, Spt16, and Pol II occupancy at PYK1, PMA1, and TELVI in the histone mutant strains (Figs 6C and 6D and S5). We observed gene and allele specific defects in Spt5 occupancy, with the E57A, E65A, and L66A substitutions causing reduced Spt5 occupancy particularly at PYK1. However, the levels of Spt5 occupancy largely mirrored Pol II occupancy levels, suggesting that the effects of the H2A substitutions on Spt5 recruitment are likely to be indirect. We also assessed the effects of the H2A substitutions on recruitment of the FACT complex member Spt16 (Fig 6D), which is required for proper histone occupancy and H2B K123ub [18,19,52]. Interestingly the substitution within the docking domain, H2A-L86A, of the nucleosome exhibited increased Spt16 occupancy at all tested loci. In contrast, substitutions E57A and E93A led to reduced Spt16 occupancy, suggesting that, for these H2A mutants, a defect in Spt16 recruitment may be a contributing factor to the reduced H2B K123ub levels and lower histone occupancy levels (Fig 3B and 3C). Global levels of Spt5 and Spt16 were not strongly affected, as judged by western analysis (S5 Fig).
Because the histone mutants have defects in H3 K4 methylation (Figs 4 and 5C), the acidic patch residues may be required for recruitment of the H3 K4 methyltransferase Set1. To test this, we performed ChIP analysis of HSV-tagged Set1 in the H2A mutants (Fig 6E). With the exception of E57A, all of the substitutions affect occupancy of HSV-Set1. However, after normalizing the H3 K4me3 occupancy levels to H3 occupancy levels, only the E65A and L66A substitutions cause a strong defect in H3 K4me3 (Fig 4C). We thus conclude that HSV-Set1 recruitment may be impacted by the occupancy levels of H2B K123ub and H3. For the E65A mutant, the severity of the H3 methylation defect and lack of restoration of H3 K4me3 upon deletion of UBP8 suggests that E65 may play a more direct role in promoting H3 K4 methylation. We did not observe a reduction in HSV-Set1 levels in the H2A mutants (S6 Fig), as has been reported to occur when H3 K4 cannot be methylated [53,54]. It is possible that the H2A mutants lack the ability to regulate Set1 levels.
Because the H2A mutants exhibit reduced levels of transcription elongation-coupled histone modifications, we asked whether the acidic patch substitutions alter the efficiency of transcription elongation. To assess transcription elongation efficiency in vivo we used a well-established galactose-controlled system to shut off transcription of a gene and measure occupancy of Pol II during the last wave of transcription [55–57]. This system incorporates the GAL1 promoter upstream of the non-essential gene YLR454W. Cells were grown in 2% galactose to activate the gene and 2% glucose was added to the cultures to prevent further initiation events. Samples were taken at different time points to determine "snap-shots" of Pol II density at four regions of YLR454W by ChIP (Fig 7A). In wild-type cells, Pol II rapidly cleared the YLR454W coding region, as previously described [55–57] (Fig 7B). In the H2A mutants, however, the rate and/or processivity of Pol II elongation was reduced. The most dramatic effect was observed with the H2A-L66A mutant, where Pol II density persisted at the 4 Kb and 8 Kb locations relative to the wild-type kinetics (Fig 7C). The H2A-E65A mutant also exhibited a delay in Pol II passage, with occupancy persisting at multiple locations throughout the time course (Fig 7D). The H2A-E93A mutant exhibited a slightly different and more modest elongation defect (Fig 7E). Collectively these data reveal an important role for the nucleosome acidic patch in promoting efficient transcription elongation.
Because the H2A acidic patch mutants have defects in H2B K123ub, we wanted to determine whether the in vivo elongation defects correlated with the loss of H2B K123ub. To begin to address this, we performed a similar analysis on H2B-K123R cells (Fig 7F). Interestingly, Pol II elongation efficiency was reduced in the H2B-K123R mutant, as indicated by persistent enrichment toward the 3’ end of the gene. These data indicate that residues within the acidic patch, at least partly through their role in promoting H2B K123ub, are important for transcription elongation efficiency in vivo.
In this study, we exploited a genetic interaction between the H2B ubiquitylation pathway and the protein quality control factor Rkr1 to identify residues in H2A and H2B that are required for H2B K123ub. We identified eight residues in H2A that, when changed to alanine, cause defects in H2B K123ub (Fig 2). Most of these residues map to the acidic patch on the nucleosome (Fig 1B), which plays critical roles in several important nuclear processes. Indeed, as shown through structural studies, the acidic patch serves as a direct binding platform on the nucleosome for a variety of proteins that affect transcription, chromatin structure, and chromosome segregation. These proteins include the Latency-Associated Nuclear Antigen (LANA) peptide from Kaposi’s sarcoma virus, the Regulator of Chromatin Condensation 1 protein (RCC1), the Bromo-Associated Homology (BAH) domain of Sir3, and the centromere binding protein CENP-C [28,30]. Additionally, as shown through functional studies and a recently published structure of the Polycomb Repressive Complex 1 ubiquitylation module in complex with a nucleosome, the acidic patch interacts with ubiquitin-protein ligases that target H2A [58–60].
Despite the importance of H2B K123ub in regulating gene expression, nucleosome stability, and genic patterns of histone methylation and acetylation, little is known about how the enzymatic machinery for H2B K123ub interfaces with the nucleosome. In a recent study, a basic region of the RING domain of Bre1 was shown to be important for interacting with the nucleosome [61]. Here, we show that nucleosome acidic patch mutants have impaired chromatin occupancy of the ubiquitin-protein ligase Bre1 and the Paf1C subunit Rtf1. The mechanism by which Rtf1 is required for H2B K123ub is largely undefined, although a recent study indicated a role for Rtf1 in stabilizing Bre1 protein levels [31]. In our H2A mutant strains, global protein levels of Bre1 are similar to those in a wild-type strain. This observation, together with our ChIP studies on Bre1 and Rtf1, suggests that the nucleosome acidic patch plays an active role in promoting H2B K123ub. A previous study found that the N-terminus of H2A, the H2A repression (HAR) domain, is also required for H2B K123ub. However, recruitment of the H2B K123ub machinery was not affected in the H2A N-terminal tail mutant [62]. It is possible, then, that the acidic patch could recruit the H2B K123ub machinery to chromatin, potentially through a direct interaction with Bre1 and/or Rtf1, while the HAR domain stimulates enzyme activity.
In light of previous work showing that Paf1C recruitment is governed by a direct physical interaction between Rtf1 and the phosphorylated C-terminal region of the elongation factor Spt5 [50,51,63,64], we were surprised that the H2A substitutions identified in our screen caused a loss in Rtf1 occupancy without a corresponding loss in Spt5 recruitment. However, it was recently shown that the human homolog of Bre1, RNF20/40, promotes recruitment of PAF1 to chromatin in human cells [65]. In addition, binding of human Paf1 to histone-like proteins and nucleosomes has been reported [66,67]. These observations align with our results and indicate that multiple interactions can mediate or stabilize the interaction between Paf1C and chromatin. Alternatively, given the importance of Spt5 phosphorylation in mediating the interaction between Rtf1 and Spt5 [50,51,63,64], it is also possible that the H2A mutants are indirectly affecting Spt5 phosphorylation. Finally, we also note that Rtf1 recruitment defects could be due to the combined effect of the individual, and relatively modest, defects in Pol II, Spt5, and Spt16 occupancy (Figs 6 and S5).
The function of the ubiquitin-specific protease Ubp8 also appears to be affected by substitutions within the acidic patch (Fig 5). A recent study suggested that the acidic patch residue H2A-Y58 promotes H2B K123ub through regulating Ubp8, as deletion of UBP8 rescued H2B K123ub in an H2A-Y58F mutant [68]. The H2A-Y58A is a lethal substitution in yeast and could not be isolated in our screen [27]. In our study, deletion of UBP8 rescued H2B K123ub to some degree in most of our mutants, which suggests that these mutants have defects in both ubiquitylating H2B-K123 and in stabilizing the mark (Fig 5A). For the H2A-L66A mutant, the nearly complete absence of H2B K123ub in the presence or absence of UBP8 suggests that little ubiquitin is placed on H2B-K123 such that removal of UBP8 makes little to no difference in this mutant.
The H2A residues we identified are required for H2B K123ub-dependent H3 methylation (Fig 4). Interestingly, some mutants exhibited defects in only H3 K4 methylation or H3 K79 methylation, while others had defects in both, despite all having reduced H2B K123ub levels. These data suggest that individual residues within the acidic patch promote methylation through separate mechanisms. Substitution of neighboring residues, H2A-E65A and H2A-L66A, differentially impacted H2B K123ub levels, but both mutants had undetectable levels of H3 K4 methylation (Fig 4) [27]. It is possible that the methylation defects caused by the L66A substitution are largely due to a severe defect in the establishment of H2B K123ub in this mutant, similar to the effect of the H2B K123R mutant [69]. In contrast, the H3 K4 methylation defect of the H2A-E65A mutant may stem primarily from the reduced recruitment and/or activation of Set1. For the H2A-E65A mutant, we noted a lack of recovery of H3 K4me3 and H3 K79me2/3 when H2B K123ub levels were increased through the deletion of UBP8 (Fig 5C). This observation suggests that E65 is important for coupling H2B K123ub to downstream H3 methylation events. Interestingly, substitution of other residues near H2B K123 has been shown to uncouple H3 methylation from H2B K123ub. For example, H2B R119 and T112, when mutated, increase H2B K123ub levels but decrease H3 K4me3 levels [70].
The severe deficiency in H3 K79me2/3 observed in the H2A-E93A mutant (Figs 4 and 5C) presents the intriguing possibility that this residue may interact with Dot1 to promote H3 K79 methylation. It is unlikely that the H3 K79 methylation defect detected in the H2A-E93A mutant is solely due to its defect in H2B K123ub, because when H2B K123ub levels are increased in the absence of UBP8, the increase in H3 K79 methylation is very slight (Fig 5C). Interestingly, the basic patch in the H4 tail is required for Dot1 methylase activity, but not for Dot1 recruitment [71]. Since the H4 tail interacts with the acidic patch of the nucleosome [1,29], one explanation for the H3 K79me2/3 defect could be that E93 is required for recruitment of Dot1, while the H4 tail stimulates Dot1 activity.
Further supporting growing indications that chromatin structure is important for proper transcription termination through the NNS pathway, the H2A mutants tested exhibited transcriptional readthrough at four SNR genes (S3 Fig). The magnitude of the transcriptional defect does not correlate strictly with the loss of any particular histone modification, suggesting that this phenotype may be sensitive to the combinatorial loss of several modifications and possibly other factors, such as histone occupancy and Spt16 recruitment (Figs 3 and 6D). Regardless of the mechanism, the increased levels of aberrant transcripts in the H2A mutants could provide a rationale for the synthetic growth defects observed in H2A mutants lacking RKR1. Rkr1 is a protein quality control factor that is involved in the degradation of aberrant proteins, including those that extend past stop codons [25,26]. The elevated synthesis of aberrant proteins, potentially as a consequence of improper transcription in the H2A mutants, could have lethal consequences for the cell [72]. The negative genetic interaction between rkr1∆ and the histone mutants suggests that the consequences of disrupting the acidic patch extend beyond chromatin and transcription.
We assayed the effects of the H2A substitutions on transcription elongation through analysis of Pol II density during the last wave of transcription across GAL1-YLR454W. In this assay, the H2A-L66A mutant exhibited a strong defect in elongation efficiency and most closely mimicked the behavior of the H2B-K123R mutant. These data support the view that H2A-L66A phenocopies H2B-K123R for loss of H2B K123ub and its consequences. The H2A-E93A and H2A-E65A mutants also exhibited impaired elongation, although not to the same degree as the H2A-L66A and H2B-K123R mutants. Given the differential effects of the H2A substitutions on the histone modification levels in the cells, differences in Pol II elongation efficiency were not unexpected. Taken together, these data indicate that H2B K123ub and its effects on downstream histone modifications and nucleosome stability are important for efficient Pol II passage through chromatin.
Combined, our data support a new role for the nucleosome acidic patch in transcription, specifically through the proper recruitment and/or activation of proteins that control H2B K123ub and downstream methylation events on H3. The mutations that disrupt this patch impair several transcription-related processes, including the modification of histones, recruitment of transcriptional machinery, the efficient passage of Pol II through chromatin, and transcription termination (Fig 8). Many of these transcriptional defects likely stem from the pleiotropic effects of losing the critical H2B K123ub mark. Together with recent structural studies, our results strongly suggest that the acidic patch is an interaction platform for proteins that modulate numerous chromatin transactions in eukaryotic cells. An exciting goal for future studies will be to understand how cells regulate access to this important region of the nucleosome.
The S. cerevisiae strains used in this study are listed in S1 Table and are isogenic to the strain FY2, which is a GAL2+ derivative of S288C [73]. Yeast transformations were performed as previously described [74]). With noted exceptions, experiments were performed using the strain KY943 transformed with histone mutant plasmids. To replace wild-type histone plasmids with HIS3-marked mutant histone plasmids, transformants were sequentially passaged three times on SC-His medium containing 2% dextrose and 0.1% 5-FOA. Unless otherwise noted, for all experiments, yeast strains were grown in SC-His medium containing 2% dextrose. HSV-Bre1 and HSV-Set1 strains contain three chromosomally located HSV tags on the N-termini of the proteins [75]. These proteins were confirmed to have proper function and expression.
Cells were grown to saturation at 30°C and washed with sterile water. Beginning with a cell suspension at a concentration of 1 X 108 cells/mL, cells were diluted serially four times by a factor of ten in water. Two microliters of each dilution were spotted on SC-His medium and SC-His medium containing 5-FOA. Plates were incubated at 30°C for three days.
Site-directed mutagenesis (Agilent) with primers listed in S2 Table was performed to remove the sequence encoding the FLAG tag from plasmids obtained from the H2A and H2B mutant library [27]. Plasmid sequences were confirmed by DNA sequencing. Plasmid names are given in S3 Table.
For western analyses other than those that measure H2B K123ub, yeast cells were grown to log phase (2–3 X 107 cells/mL) and lysed by bead beating in trichloroacetic acid (TCA), as described previously [76]. To make whole cell extracts for H2B K123ub analysis, cells were lysed in SUTEB buffer (10 mM Tris-HCl, pH 8.0, 1% SDS, 8 M urea, 10 mM EDTA, pH 8.0, and 0.01% bromophenol blue) [43]. Proteins were resolved on SDS-polyacrylamide gels (15% polyacrylamide for histone westerns, 10% polyacrylamide for Rtf1 and HSV-Bre1, and 8% polyacrylamide for HSV-Set1, Spt5, and Spt16 westerns) and transferred to nitrocellulose membranes. For H2B K123ub western blot analysis, proteins were transferred to PVDF membranes. Membranes were incubated with primary antibodies and then with anti-mouse or anti-rabbit secondary antibodies (GE Healthcare 1:5,000 dilution). Antibodies that recognize the following proteins or histone modifications were used: total histone H3 (1:30,000 dilution) [43], trimethylated H3 K4 (H3 K4me3) (Active Motif 39159, 1:2,000 dilution), H3 K4me2 (Millipore 07–030, 1:2000 dilution), H3 K79me3 (note: this antibody has been reported by the manufacturer to cross-react with H3 K79me2, Abcam ab2621, 1:2,000 dilution), H3 K36me2 (Millipore 07–369, 1:1000 dilution), H3 K36me3 (Abcam ab9050, 1:1000 dilution), H2A (Active Motif, 39235, 1:5,000 dilution), H2B (Active Motif, 39237, 1:5,000 dilution), HSV (Sigma-Aldrich H6030, 1:350 dilution), Spt5 (gift from Grant Hartzog, 1:1000 dilution), Spt16 (gift from Tim Formosa, 1:500 dilution), Rtf1 (1:5,000 dilution) [77], and glucose-6-phosphate dehydrogenase (G6PDH) (Sigma-Aldrich A9521, 1:30,000 dilution). An antibody against a human H2B K120ub-containing peptide (Cell Signaling 5546, 1:1000 dilution) was used to detect the analogous modification in S. cerevisiae, H2B K123ub. Proteins were visualized using enhanced chemiluminescence substrate (PerkinElmer) and either a 440 CF digital imaging station (Kodak) or a ChemiDoc XRS digital imaging station (BioRad). For western blot analysis, signals were quantified using ImageJ software and normalized to the loading control specified in the figure legend. The relative signal from the wild-type strain was set equal to one. Error bars represent standard error of the mean for three biological replicates (SEM).
Chromatin immunoprecipitation (ChIP) assays were performed with 250 mL of log-phase yeast cultures (1–2 X 107 cells/mL) as previously described [78]. For histone ChIPs, sheared chromatin was incubated overnight at 4°C with antibodies specific to H2B, (0.5 μl, Active Motif, 39237), human H2B K120ub (2.5 μl, Cell Signaling 5546), H3 K4me3 (2.5 μl, Abcam ab8580), H3 K79me2/3 (2.5 μl, Abcam ab2621), or total H3 (5 μl) [43]. Chromatin prepared from an H2B-K123R strain served as a specificity control for the human H2B K120ub antibody (S7 Fig). For other ChIPs, chromatin was incubated overnight at 4°C with antibodies specific to Spt16 (1 μl, gift from Tim Formosa), Spt5 (1 μl, gift from Grant Harzog), or Rpb3 (2.5 μl Neoclone W0012). Following incubation with the primary antibodies, chromatin was incubated for 2 hours at 4°C with Protein A-conjugated sepharose for all ChIPs, with the exception of Rpb3 ChIPs, for which chromatin was incubated with Protein G-conjugated sepharose (30 μl, GE Healthcare). For ChIP of HSV-Bre1 and HSV-Set1, chromatin was incubated overnight at 4°C with an antibody specific to the HSV epitope (2.5 μl, Sigma-Aldrich H6030), followed by incubation as described above. For ChIP of Rtf1, chromatin was incubated overnight at 4°C with polyclonal antisera that recognizes Rtf1 [77]. DNA was purified (Qiagen) and analyzed by qPCR using Maxima SYBR (Thermo) and primers for the 5’ coding region of PYK1 (amplicon: +253 to +346 relative to ATG), the 3’ coding region of PYK1 (amplicon: +1127 to +1270), the 5’ coding region of PMA1 (amplicon: +214 to +319 relative to ATG), the 3’ coding region of PMA1 (amplicon: +2107 to +2194), or a telomeric region of chromosome VI (chromosomal coordinates, 269495 to 269598). Occupancy levels were calculated using the primer efficiency raised to the difference between input and immunoprecipitated Ct values. Presented data are an average of two technical replicates for each of three biological replicates. The error bars indicate the standard error of the mean (SEM).
Total RNA was isolated from log-phase yeast cultures (1–2 X 107), and 20 μg of RNA were subjected to northern blot analysis as described previously [79]. Radiolabeled DNA probes were generated through random-prime labeling reactions of PCR templates. Membranes (Gene Screen Plus, Perkin Elmer) were incubated with radiolabeled DNA probes from PCR fragments of SCR1 (amplicon: -163 to +284 relative to the TSS), SRG1 (amplicon: -454 to -123 relative to SER3 ATG), SER3 (amplicon: +111 to +1342 relative to ATG), and FLO8 (amplicon: +1515 to + 2326 relative to ATG). See S2 Table for primer sequences. Signals were quantified using ImageJ software relative to the SCR1 loading control, with wild type set to one. For quantification of all northern blot analyses, signals were averaged for three independent biological replicates. Error bars represent standard error of the mean (SEM).
Total RNA was isolated as described above and then DNase treated using the Turbo DNA-free kit (Ambion, AM1907) and RNase inhibitor (Ambion, AM2682). cDNA was generated using the RETROscript kit (Ambion, AM1710) with random hexamers and oligo(dT) primers. Quantitative PCRs were performed as described above using primers specific for the regions downstream of snoRNAs (S2 Table). Signals were analyzed using the ∆∆CT method with ACT1 used as the target gene [80]. For controls, reactions lacking reverse transcriptase or template were performed. The graphs show the results of three independent biological replicates.
|
10.1371/journal.pgen.1006660 | TDRD6 mediates early steps of spliceosome maturation in primary spermatocytes | Tudor containing protein 6 (TDRD6) is a male germ line-specific protein essential for chromatoid body (ChB) structure, elongated spermatid development and male fertility. Here we show that in meiotic prophase I spermatocytes TDRD6 interacts with the key protein arginine methyl transferase PRMT5, which supports splicing. TDRD6 also associates with spliceosomal core protein SmB in the absence of RNA and in an arginine methylation dependent manner. In Tdrd6-/- diplotene spermatocytes PRMT5 association with SmB and arginine dimethylation of SmB are much reduced. TDRD6 deficiency impairs the assembly of spliceosomes, which feature 3.5-fold increased levels of U5 snRNPs. In the nucleus, these deficiencies in spliceosome maturation correlate with decreased numbers of SMN-positive bodies and Cajal bodies involved in nuclear snRNP maturation. Transcriptome analysis of TDRD6-deficient diplotene spermatocytes revealed high numbers of splicing defects such as aberrant usage of intron and exons as well as aberrant representation of splice junctions. Together, this study demonstrates a novel function of TDRD6 in spliceosome maturation and mRNA splicing in prophase I spermatocytes.
| Very little is known about splicing and its regulation in germ cells, particularly during meiosis. This paper describes the role of a male germ cell-specific protein, Tudor containing protein 6 (TDRD6), in assembly of the spliceosome in spermatocytes. We show that in prophase I TDRD6 interacts with the key protein methyltransferase of the splicing pathway PRMT5. PRMT5 methylates arginines in substrate proteins. In a methylation dependent manner, TDRD6 also associates with spliceosomal core protein SmB in the absence of RNA, thus before an RNP-type spliceosome has been assembled. In Tdrd6-/- diplotene spermatocytes PRMT5’s association with SmB and arginine dimethylation of SmB are much reduced. Abrogation of arginine methylation impairs the assembly of spliceosomes and the presence of the spliceosomal RNA U5 is aberrantly increased. These deficiencies in spliceosome maturation correlate with decreased numbers of Cajal bodies and gems involved in later stages, i.e. nuclear snRNP maturation. To reveal functional consequences of these deficiencies, transcriptome analysis of primary spermatocytes showed high numbers of splicing defects such as aberrant usage of intron and exons as well as aberrant representation of splice junctions upon TDRD6 loss. This study reveals a novel function of TDRD6 in spliceosome maturation and mRNA splicing in spermatocytes
| Spermatogenesis is essential for the generation of haploid male gametes required for sexual reproduction in higher eukaryotes. Spermatogenesis in mice starts at app. day 6 postpartum (dpp) as spermatogonia undergo mitotic expansion, generate tetraploid cells (4N) during premeiotic S-phase and enter meiosis. Meiosis is composed of two successive nuclear divisions. In the first meiotic division (meiosis I), pairs of homologous chromosomes are segregated and primary spermatocytes are reduced in chromosome content to diploid secondary spermatocytes (2N). These 2N cells then undergo reduction to haploid spermatids (N) in the second meiotic division (meiosis II) through a mitosis-like division segregation of sister chromatids. Meiotic prophase I is by far the longest phase of meiosis, lasting approximately three weeks in most mammals. It is described by four sequential substages, i.e. leptotene, zygotene, pachytene and diplotene. Prophase I features unique chromosome properties and behavior such as the pairing of homologous chromosomes and formation of the synaptonemal complex (SC) in pachytene. The SC consists of two axial elements (AE) that form earlier during leptotene. Each AE–often visualized by staining for protein SYCP3 –supports the two sister chromatids of one homologue. In the diplotene stage, homologues desynapse and remain linked only at the sites of crossing-over/chiasmata which are resolved at metaphase/anaphase (reviewed in [1]). Two successive meiotic divisions follow prophase I to produce haploid spermatids leading to the last stage of spermatogenesis called spermiogenesis; i.e. the process of morphological differentiation of haploid round spermatids to elongated spermatids to motile sperm.
Chromatin compaction during the late steps of spermatogenesis results in silencing of transcription at this stage despite ongoing translation of mRNA. This is enabled through the temporal uncoupling of transcription and translation of mRNA during spermatogenesis [2]. Most mRNA is transcribed at earlier stages, i.e. late prophase I (late pachytene, diplotene) and early spermiogenesis (reviewed in [3]). Direct measurement of de novo synthesized RNA during spermatogenesis demonstrates a peak in global transcription in late prophase I, which meets the later demand and provides the mRNA needed at this stage [4–7]. This entails posttranslational regulation and storage of these mRNAs (reviewed in [8, 9]). Germ cells are endowed with special granules involved primarily in posttranscriptional regulation of mRNA. These granules include the fibrous-granular chromatoid body (ChB) type 1 in late prophase I and the single globular ChB type 2 in round spermatids (reviewed in [10, 11]).
The ChBs contain mRNA, miRNA and piRNA as well as proteins functioning in RNA metabolism such as the DEAD-box RNA helicase MVH [12–14]. Tudor domain (TDRD) containing proteins are prominent components of ChBs. The tudor domain is a conserved protein structural motif of ~60 amino acids and was initially found in proteins that associate with nucleic acids [15]. It is characterized by a barrel-like structure composed of anti-parallel ß-sheets forming a hydrophobic pocket surrounded by charged residues that constitute a protein-protein interaction surface [16–19]. A scaffolding function has been suggested for TDRD proteins [20, 21]. Mammalian TDRD proteins contain variable numbers of Tudor domains that are occasionally accompanied by other types of domains. The Tudor domain requires methylated or di-methylated arginines or lysines for binding target proteins. Among the TDRD proteins in male germ cells, TDRD6 is expressed only from mid prophase I spermatocytes onwards and localizes to the ChBs [20, 22, 23]. Mouse TDRD6 encodes 6 Tudor domains and is the homologue of Drosophila Tudor protein. It appears first in late pachytene spermatocytes (day 17.5 post partum) in multiple fine filamentous cytoplasmic granules surrounding the nuclei. In round spermatids, the TDRD6 signal appears as a single bright distinct perinuclear dot representing the ChBs. TDRD6 interacts in vitro with MVH, MILI and MIWI and loss of TDRD6 results in mis-localization of ChB components MVH and MIWI. In the absence of TDRD6 the ChB architecture is severely impaired and spermiogenesis is arrested at the round-to-elongated spermatid stage. Pre-miRNA and miRNA transcripts are mis-regulated in Tdrd6-/- testis [22]. Lately we reported the role of TDRD6 in the long 3’ UTR-stimulated pathway of NMD during spermiogenesis [13]. Overall, these findings implicate that TDRD6 functions in distinct RNA pathways during spermatogenesis, but functions in meiosis remained unclear.
Mass spectrometric analysis of potential TDRD6 associated proteins [22] identified PRMT5, a key protein methyl transferase of the splicing pathway. This suggested a role for TDRD6 in the pre-mRNA splicing pathway. The studies reported here revealed a central role for TDRD6 in the regulation of spliceosome maturation and mRNA splicing in spermatogenesis.
The spliceosome is a large and flexible macromolecular machine with more than a hundred RNA and protein components [24]. Small nuclear ribonucleoproteins (snRNPs) constitute the main building blocks of the spliceosome machinery. Different types of spliceosomal snRNPs all contain a set of seven common Sm proteins (SmB/B’, D1, D2, D3, E, F, G), which form a heptameric ring around spliceosomal snRNAs, commonly referred to as U-RNAs due to their high uridine content. The majority of pre-mRNAs is spliced by the major spliceosome, which contains a combination of U1, U2 or U5 or both U4/U6 snRNAs. A small group of introns called ‘ATAC introns’ are spliced by the minor spliceosome containing a combination of U11, U12, U4atac, U6atac or U5 snRNAs. Formation of the heptameric Sm ring and its assembly with individual U-RNAs is catalyzed by the SMN complex in an ATP-dependent manner in vivo although this assembly can also take place spontaneously in vitro [24–26]. The SMN complex is composed of the Tudor-domain containing protein SMN, seven Gemin proteins (Gemin 2–8) and unrip. The SMN complex cooperates with the PRMT5 “methylosome” complex, which is comprised of WD45/MEP5, pICln and the protein arginine methyltransferase 5 (PRMT5) [27]. SMN-positive snRNP bodies translocate to the nucleus and are further modified to mature. These modifications take place in a non-membraneous nuclear organelle called Cajal body (CB) [28]. CBs are characterized by the accumulation of coilin, a protein essential for CB integrity, as well as scaRNAs, which are small RNA species transcribed by RNA pol III and targeted specifically to CBs as RNPs [29, 30]. After undergoing these multiple steps of maturation, the spliceosomes are ready to act on the primary mRNA transcripts. Very little is known about spliceosome assembly and regulation of splicing in spermatocytes, and the present study reveals a male germ cell-specific pathway of spliceosome assembly.
A previous mass spectrometric analysis performed in our lab [22] revealed PRMT5 as a putative interactor of TDRD6. PRMT5 catalyzes the symmetrical arginine dimethylation of proteins including SmB [24, 31, 32], one of the core protein components of spliceosomal snRNPs (reviewed in [33, 34]). PRMT5 belongs to a family of protein arginine methyl transferases that are widely expressed and that di-methylate arginine in an either symmetric (sDMA) or asymmetric (aDMA) manner. PRMT5, a so-called type II arginine methyl transferase, generates sDMA marks (recently reviewed in [35]). PRMT5 associates with Sm proteins [24] and PRMT5 deficient somatic cells display splicing defects [36]. Dart5, the Drosophila ortholog of murine PRMT5 is essential for germ cell specification and maintenance [37]. In mice, PRMT5 is dispensable for primordial germ cell specification. Still, PRMT5 regulates gene expression by altering the spliced repertoire of RNAs as it promotes arginine methylation of the Sm proteins in primordial germ cells [38].
To validate the interaction between TDRD6 and PRMT5, we generated a transgenic mouse carrying a Localization and Affinity Purification ‘‘LAP”- tagged TDRD6 BAC construct [39]. The LAP tag contains an EGFP sequence for localization studies and for immunoprecipitation (IP) (S1A Fig). In immunoblots (IB) of total cell extracts from LAP-TDRD6-positive transgenic testes an α-EGFP antibody recognized two bands of 250 kDa (cleaved LAP-TDRD6) and 270 kDa (full-length LAP-TDRD6), indicating that LAP-TDRD6 is processed like the endogenous protein [22] (S1B Fig). Fluorescent microscope analysis of testis sections of adult transgenic mice demonstrated as expected a fibrous and granulated perinuclear EGFP (LAP-TDRD6) signal in primary spermatocytes of stage X-XI sections (S1Ci Fig) and one single lobulated, perinuclear EGFP signal in round spermatids of stage I-II-III sections (S1Cii Fig). The typical structure of wild-type tubules was also maintained (S1C Fig). Motile spermatozoa were observed in the testis of LAP-TDRD6-positive Tdrd6-/- males from all founders. All LAP-TDRD6-positive Tdrd6-/- males were fertile. This proves that the LAP-TDRD6 is fully functional as it can rescue the developmental arrest during spermiogenesis and the sterility of Tdrd6-/- males.
LAP-TDRD6 was immunoprecipitated with an α-EGFP antibody from RIPA extracts of total adult testes and probing the IB for PRMT5 detected the PRMT5 signal (Fig 1A). IB probing of the reverse IP with an α-PRMT5 antibody yielded LAP-TDRD6, confirming the interaction (Fig 1B).
TDRD6 has 6 Tudor domain [22], which mediate protein-protein interactions by recognizing and binding to symmetrically dimethylated arginine (sDMA) residues generated by methyltransferases like PRMT5 [18]. In primary (pachytene, diplotene) spermatocytes, identified by anti SYCP3 staining, the majority of TDRD6 localizes to filamentous ChB type 1 structures in the perinuclear space (Fig 1C, panel i). In later stages (round, elongated spermatids), the single TDRD6 signal localizes to the ChBs type 2 marked by the MVH signal as reported before [22], which also overlaps with sDMA signals (Fig 1C, panel ii). The lack of sDMA signals in precipitates of LAP-TDRD6 suggests that TDRD6 itself is not symmetrically dimethylated and excludes that the overlap of all TDRD6 with sDMA signal is due to the methylation of TDRD6 itself (S2 Fig). This data is consistent with previous reports showing the association of Tudor domains with proteins carrying sDMA. It appears likely that this interaction contributes also to the architecture of ChBs, which harbor many sDMA-proteins such as MILI, MIWI, UPF1 and UPF2 [12, 13, 22].
The assembly of spliceosomal proteins with U-RNAs is mediated partly through direct interactions between the Tudor domain of SMN and the sDMAs of the spliceosomal proteins SmB, SmD1, SmD3 [16]. We hypothesized that TDRD6 would associate with sDMA-containing spliceosomal core proteins. Staining of LAP-TDRD6 germ cells with α-Y12 antibody recognizing SmB and SmD revealed an overlap between the Y12 and LAP-TDRD6 signals in the nuclear periphery of diplotene cells but not in the TDRD6-positive ChBs in post-meiotic round spermatids (Fig 2A). The absence of Sm proteins in ChB type 2 was also confirmed by our recent proteomic analysis of ChBs in round spermatids [13].
Sm proteins bind to the U-RNAs in spliceosomes and RNA could bridge its interaction with TDRD6. Therefore, we precipitated Sm proteins with α-Y12 IgG both in the absence and presence of RNase from total testes of adult males. We used SmB detected as a 25 kDa band to represent spliceosomal proteins. The interaction was completely lost in the absence of RNase (Fig 2B). Successful degradation of RNA was confirmed by staining of the total RNA extracted from the flow-through of the IP (S3 Fig). The interaction of TDRD6 with Sm proteins exclusively in absence of RNA places TDRD6 upstream of the association of SmB with the SMN complex during spliceosome maturation, i.e. before U RNA binding to the complex.
We then tested whether the sDMA modification mediates the TDRD6-Sm interaction. We inhibited methylation with 5′-deoxy-5′-(methylthio)adenosine (MTA) in total adult testis cultures for 16 h prior to IP. In the presence of MTA the interaction between LAP-TDRD6 and SmB is undetectable as shown by both α-EGFP (Fig 2C) and the reverse α-Y12 IPs (Fig 2D), while the interaction was observed in control-treated cells. To exclude the possibility that the interaction is lost due to apoptosis upon MTA treatment, we analyzed MTA and control-treated cells after 16 h via FACS by staining them for Annexin V and propidium iodide. There was no significant change in the early or late apoptotic and dead cell populations of MTA versus control cells (S4 Fig). Thus, we conclude that TDRD6 requires sDMA to interact with SmB.
Previous studies in our lab revealed that about 20 kDa off the C-terminus of Tdrd6 are removed during the transition from meiosis I to meiosis II [22]. We can speculate that the app. 20 kDa C-terminal region supports the localization of SmB to ChBs. This may explain the restriction of Sm-TDRD6 co-localization to meiotic stages as detected by immunostaining. The eliminated C-terminal fragment does not contain a Tudor domain as the most C-terminal Tudor domain is at position 1630 of the total of 2135 amino acids. Removal of the C-terminal fragment might affect the 3D-structure of the protein and thereby its interactions.
PRMT5 functions in a complex of cofactors determining its substrate specificity [40–42]. Having shown that TDRD6 interacts with both PRMT5 and one of PRMT5’s substrates, SmB, we hypothesized that TDRD6 regulates methylation of SmB by PRMT5 and that the sDMA status of SmB is altered in Tdrd6-/- cells. Given that the TDRD6 and SmB signals overlap only in diplotene cells (Fig 2A), we isolated diplotene cells to IP SmB and assess its sDMA status by IB.
Diplotene cells were isolated using anti-hCD4 antibody-coupled magnetic beads binding to germ cells expressing hCD4 under control of the endogenous Tdrd6 promoter from males at 20 dpp [22]. At this time point the juvenile mice showed no difference in development (S5A Fig) and in apoptosis (S5A and S5B Fig) between Tdrd6+/- or Tdrd6-/- strains, and cell preparations containing app. 80% diplotene cells were obtained (S5A Fig).
We IPed SmB and observed a strong decrease of 76% (Fig 3B; SD = ±7.1%) in the sDMA signal on SmB in Tdrd6-/- diplotene cells (Fig 3A). Considering that TDRD6 and PRMT5, TDRD6 and SmB, as well as PRMT5 and SmB interact with each other, TDRD6 may act as a tissue-specific mediator of PRMT5 for SmB methylation in the testis. TDRD6 may bring PRMT5 and SmB in close vicinity to each other and thereby facilitate the methylation. Alternatively, TDRD6 may interact with another cofactor of PRMT5 and load it on the methylosome complex, thereby indirectly controlling its activity. We analyzed the interaction between the enzyme and its substrate. IPs with α-Y12 revealed a complete loss of PRMT5 co-precipitation with SmB in Tdrd6-/- diplotene cells (Fig 3C).
An sDMA mark, which needs to be generated by another PRMT, is a known prerequisite for binding of PRMT5 to SmD3 [43]. It is unclear which PRMT modifies PRMT5. PRTM9 would be another type II PRMT with di-methylation activity but is very lowly expressed in germ cells as we observed in our transcriptome data of primary spermatocytes (see below). We detected 3-fold higher expression of PRMT7 than PRMT5 in the transcriptome suggesting that PRMT7 functions in these cells. It is controversial, however, whether PRMT7 only mono-methylates or also di-methylates as has been reported for Sm proteins [43] and other proteins, at least in vitro [44]. Therefore we can only speculate that PRMT7 may account for the remaining methylated pool (24%) of SmB detected in TDRD6-deficient murine testis, and PRMT7 may also be responsible for PRMT5 methylation.
An additional means of contribution by TDRD6 to overall sDMA levels may be protection of sDMA from protein demethylases. While the existence of arginine demethylases is still debated, recent evidence argues in favor of such enzymes [45]. Binding to sDMA on other proteins like PRMT5, TDRD6 may protect sDMA marks. In absence of TDRD6, these sDMA marks–such as those on PRMT5 –may get lost. As a consequence the interaction of PRMT5 with SmB and its methylation by PRMT5 may be affected [43].
After methylation, Sm protein sub-complexes are loaded onto the SMN complex, which mediates their assembly with the U-RNAs [46]. In human cells, cytoplasmic snRNP assembly requires the activities of both PRMT5 and PRMT7 [43]. Since the sDMA status of SmB is diminished in Tdrd6-/- diplotene cells, we hypothesized an impaired assembly of U-snRNPs at this stage. Using the same Tdrd6+/- and Tdrd6-/- diplotene stage cell populations, we performed RNA IP with the α-Y12 antibody and compared the levels of selected U-RNAs incorporated into the spliceosomes via real-time RT-PCR. U2 is a component of major spliceosome snRNPs while U12 and U4atac are found in minor spliceosomes. U5 is common in both (reviewed in [33]). We observed a 3.5-fold (SD = ± 0.5, P < 0.05) increase in U5 RNA levels in the splicing snRNPs in the absence of TDRD6. There was no significant change in the levels of U2, U12 and U4atac bound to Sm proteins (P > 0.05) (Fig 3D). SmB IB of the RNA IPs with α-Y12 IgG confirms the equal pull-down of this protein in both Tdrd6+/- and Tdrd6-/- samples (S6 Fig).
This finding together with the requirement for RNA exclusion in SmB-TDRD6 interaction suggests that TDRD6 acts as an assembly factor for at least SmB and either competes with certain snRNAs like U5 for SmB binding or can only bind in an early step of spliceosome maturation prior to RNA binding.
Exclusion of a certain U-RNA by TDRD6 resembles the mode of interaction of pICln with Sm proteins. pICln acts as a chaperone by binding to the small, highly conserved Sm domain that is N-terminal to the methylated RG motifs and responsible for oligomerization and RNA binding of Sm proteins [47]. As elaborated above, TDRD6-SmB binding is likely mediated by Tudor-sDMA interaction. The precise mode and domain requirements for TDRD6-SmB interactions are subject to future studies. We speculate that by binding to SmB at the methylated RG box, TDRD6 may trap sDMA-marked SmB and sterically occlude the assembly with U5. The determinants for this specificity and also for each U-RNA binding to the Sm core are still unknown. U5 may have higher affinity to the Sm core than the other U-RNAs. Alternatively, TDRD6 might suppress a factor responsible for U5 loading on the Sm ring.
Once the U-snRNP assembly takes place in the cytoplasm, the SMN complex translocates into the nucleus to carry the snRNPs to Cajal Bodies (CBs), where the final steps of spliceosome maturation take place. The rate of spliceosomal snRNP assembly correlates with CB number [48] and SMN depletion results in a decrease in CB abundance [48, 49].
Given the disrupted U-snRNP assembly in Tdrd6-/- diplotene cells, we hypothesized that formation of both nuclear SMN-positive bodies and CBs would be impaired. Via immunofluorescence staining, we detected a significant decrease by almost 70% in the number of SMN-positive bodies per nucleus (Fig 4B, p<0.0001) with a median number of 3 per Tdrd6+/- nucleus versus 1 in Tdrd6-/- (Fig 4Ai). There were no SMN-positive nuclear bodies detected in round spermatids (Fig 4Aii)
In Tdrd6+/- diplotene cells, the number of CBs was reduced by 50% (Fig 4D; p<0.0001) with a median number of 6 bodies in Tdrd6+/- versus 3 in Tdrd6-/- (Fig 4Ci). In round spermatids the median number of CBs in round spermatids remained 2 in both genotypes (Fig 4Cii & 4E; p = 0.7245). Higher numbers of both SMN-positive bodies and CBs in meiotic cells correlates with the previous data on the increased mRNA transcription and hence splicing in late prophase I [4].
Several studies on somatic cells previously revealed improper pre-mRNA splicing upon disruptions at the upstream steps of the spliceosome maturation pathway [50–52]. Our data implicated changes in the snRNP profile of TDRD6-deficient diplotene cells by altering the normal proportion of endogenous snRNPs. Therefore, we hypothesized that TDRD6 deficiency would lead to impaired splicing in diplotene cells.
To test this hypothesis, we performed high-throughput sequencing of total RNA from purified diplotene cells of Tdrd6+/- and Tdrd6-/- mice at a depth of 47–51 million paired-end fragments per sample. After aligning the fragments to mouse reference genome (mm10) with GSNAP and annotating them with Ensembl v75, we acquired over 360 million RNA sequence reads in total with a very high unique mappability of more than 90% for each sample (S7 Fig).
We analyzed the splicing profiles using two different approaches. Differential intron usage analysis was performed via DEXSeq with a false discovery rate (FDR) of 0.1. This analysis revealed 1270 genes (of a total of around 20 k genes with detectable expression) with 1850 mis-regulated introns, i.e. 1.5 mis-regulated introns per gene on average (Fig 5A). More specifically, 824 introns showed up-regulation, i.e. retention, in the Tdrd6-/- samples, whereas 1026 introns showed significant down-regulation compared to the Tdrd6+/-.
Another approach to assess the fidelity of splicing was to test the differential exon and splice site usage. The reads at the exon-exon junctions as well as within the exonic sequences were compared using JunctionSeq. 889 genes demonstrated significant differential usage in 1,989 events indicating more than 2 perturbations at exon or junction level per gene (Fig 5B; FDR = 0.1). 1,509 of these events are exonic and of those, 863 were up-regulated in the Tdrd6-/- condition while 646 of them were down-regulated. The remaining 471 events represented significantly mis-regulated splice sites. 381 of them were up-regulated while 90 of them showed down-regulation in Tdrd6-/- primary spermatocytes. These two approaches yielded 240 common genes with mis-spliced mRNA.
We validated a subset of these common 240 mis-spliced mRNA found by both approaches via RT-qPCR. Genes with several isoforms were excluded since different isoforms may account for the differential usage of introns, exons or junctions detected by these different approaches. We detected the mis-regulation of an intron by designing primers such that one primer resides in the mis-regulated intron whereas the other primer resides in the adjacent unchanged exon to ensure that any difference revealed by PCR would be due to the mis-spliced intron. This way we also avoided unknown genes perhaps expressed within the introns, and accounted for the differences that could arise from differential expression of putative gene(s) lying within that intron (S7B Fig). Overall, we validated mis-regulated introns in 6 different randomly selected genes by RT-qPCR (Fig 6C).
Our data demonstrates slightly less intron retention (824 introns) than intron removal (1026 introns) among the mis-regulated cases. This suggests that upon TDRD6 loss the specificity of the intronic splicing site selection is affected more than the overall efficiency of splicing. This is plausible considering that loss of TDRD6 leads to an altered snRNP repertoire. U5 snRNP is a common component of both types of splicing machineries and therefore centrally important [33]. Increased levels of one type of core snRNP of the splicing machinery might disturb the correct stoichiometry of splicing machinery components.
Functional cluster analysis of the mis-spliced transcripts did not reveal any significantly enriched clustering of these transcripts at a particular cellular function. Also, the mis-regulated miRNAs reported earlier in TDRD6 deficient spermatids [22] are not overrepresented among the 1270 genes showing mis-regulated introns.
The finding that a wide diversity of genes was affected indicates a global role of TDRD6 in the regulation of pre-mRNA splicing.
The data reported here suggest a novel function for TDRD6 in the maturation of spliceosomal snRNPs during the transcriptionally highly active prophase I of spermatogenesis. This function adds to the roles of TDRD6 at later stages of spermatogenesis/spermiogenesis in formation of the ChB type 2 and in nonsense-mediated decay. One notable difference between TDRD6 protein in meiosis I and in later stages is the conversion of the early 250 kDa form into the later 230 kDa form, which lacks about 20 kDa off its C-terminus We speculate that the 250 kDa variant is the one acting in spliceosome assembly while the 230 kDa variant–the only form present after meiosis I–acts in ChB type 2 processes. It will be fascinating in the future to assign specific molecular properties to the C-terminal end of TDRD6, which does not bear Tudor domains, and in silico analysis of the amino acid sequence did not reveal prominent motifs.
For the role of TDRD6 in spliceosome assembly, we propose the model shown in Fig 6. TDRD6 promotes PRMT5 association with and methylation of SmB and likely other Sm proteins. To do so, TDRD6 itself also associates with both proteins. Thereby TDRD6 may also control SmB availability for SMN complex formation. The SmB methylation and/or its TDRD6 association is required to ensure proper composition of the SMN RNP complex and subsequently for presence of SMN-positive bodies as well as Cajal body formation in the nucleus. In absence of TDRD6, ChB type 1 do not form SmB methylation is largely decreased, SmB cannot be properly regulated by TDRD6 and spliceosome assembly becomes unbalanced with aberrantly overrepresented U5 snRNP and possibly improper composition of other components. This leads to substantially reduced SMN+ bodies and Cajal bodies in the nucleus and to aberrant splicing.
Why is TDRD6 expressed only in the male germ line and even here only from mid-pachytene onwards? Obviously, TDRD6 is not needed for the universal splicing apparatus found in almost all animal cells, and it is not even needed for oocytes at any stage. There is no definitive answer to this question, but there are no ChB in non-male germ cells or somatic cells. Other bodies or speckles harbor splicing factors in those cells, and some like Cajal bodies are shared between spermatocytes and other cell types. Cajal bodies are reduced in absence of TDRD6 because their assembly depends on prior successful assembly of SMN complexes. Late spermatocytes and spermatids are unique in that they stepwise change their chromatin organisation until the genome is packaged into a protamine-dense body that is largely transcriptionally silenced. Thus the need to store some RNA species in ChBs, i.e. ChB type 2. There may be testis-specific splice events to produce the appropriate RNA molecules and these are not produced in any other cell type, perhaps explaining the need for TDRD6. However, we find a general splicing deficiency in absence of TDRD6, not only impaired splicing of testis-specific RNAs. Still, TDRD6 may have acquired the role of a spliceosome assembly factor to make sure that all RNA splicing events, including those specific for spermatogenesis, are properly carried out. An additional challenge to spermatocytes stems from the great burst of transcription that happens when chromosome pairing is complete, i.e. when the SCs are formed in pachytene. To handle this burst, which includes testis-specific transcripts, a highly efficient splicing machinery may be required and TDRD6 helps providing that efficiency. Since little is known about splicing processes during mammalian spermatogenesis, this holds intriguing questions for future research.
Generation of Tdrd6-/- mice was described in detail previously [22]. Tdrd6-LAP mice were generated as described in the Supplemental Experimental Procedures. The use of mice was approved by the State of Saxony animal welfare officials, Az DD24-5131/ 339/6 and was performed according to the national and EU guidelines.
Antibodies used in several different applications in this study are listed in Tables 1, 2 and 3 in “Supplementary Experimental Procedures”.
Coverslips were boiled in 1M HCl, rinsed with water, dipped into poly-L-lysine solution (50μg/ml poly-L-lysine and 10mM Tris pH8.0 in H2O) for 10min at room temperature (RT) and dried at RT. Cells were incubated on coverslips for 1h at RT, fixed in 2% formaldehyde/PBS. From permeabilization/blocking onwards, the same staining protocol was followed as for the sections (see Supplemental Experimental Procedures).
Z-stacks of the single cell suspension stainings were taken with the Leica SP 5 confocal laser-scanning microscope. Orthogonal sections of the images were later obtained using the software Fiji.
Single cell suspensions were obtained as described in Supplemental Experimental Procedures and incubated at 32°C in Dulbecco's Modified Eagle Medium supplemented with 10% fetal calf serum with 40μg/ml leptomycin B or in 1mM 5′-Deoxy-5′-(methylthio)adenosine (MTA) for 16h. Control cultures were incubated with 70% (v/v) methanol/water for leptomycin B or with DMSO for MTA.
100–300μg RIPA extract (see Suppl Exp procedures) was diluted at a 1:2 ratio in IP buffer (50mM Tris-HCl at pH 7.4, 150mM NaCl, 0.25% Triton-X100, 1mM EDTA, 5mM NaF, 1mM Na2VO3, 1mM PMSF, 1x protease inhibitor cocktail EDTA-free mini), supplemented with the primary antibody and incubated at 4°C overnight with gentle shaking. 8μl Dynabeads-proteinA/G per 1μg of primary antibody was added and incubated at 4°C with gentle shaking for 2h. Beads were washed 3 times for 10 min with the same IP buffer, resuspended in 2x Laemmli buffer and boiled at 95°C for 5 min to elute the proteins. SDS-PAGE and immunoblotting procedures were followed as described in Supplemental Experimental Procedures. Wherever indicated, 100μg/ml RNase or 1U/μl RNaseOUT was added into the RIPA and IP buffers.
Immunoprecipitation for the interactions SmB/PRTM5 and LAP-TDRD6/PRMT5 were performed with the following changes: Single cells were fixed in 1% formaldehyde solution for 5 min and then quenched for 5 min with 200mM Glycine at RT. After the cells were pelleted and lysed in the extraction buffer (1%SDS, 10mM EDTA, 50mM Tris pH8.0, 1mM DTT, 5mM NaF, 1mM Na2VO3, 1mM PMSF and 1x protease inhibitor cocktail EDTA-free mini), the extracts were diluted at a 1:9 ratio in the IP buffer (0.01%SDS, 1.1% TritonX-100, 1.2mM EDTA, 16.7mM Tris pH8.0, 5mM NaF, 1mM Na2VO3, 1mM PMSF and 1x protease inhibitor cocktail EDTA-free mini) to quench the excess SDS in the extraction buffer. For elution, beads were boiled in 2x Laemmli buffer first at 70°C for 1h and then at 95°C for 5 min. Densitometric quantification of the blots was performed using Fiji.
Depending on experiment, total RNA was extracted from either whole testes or MACS-purified hCD4+ cells via Trizol (Invitrogen) according to manufacturer’s instructions. Concentration and purity of the RNA samples were determined by UV absorbance measurements using the NanoDrop 2000c Spectrophotometer (Thermo Scientific).
100ng to 1μg of total RNA was treated with RQ1 RNase-Free DNase (Promega) according to manufacturer’s instructions and reverse transcribed into first strand cDNA via SuperScript II reverse transcriptase (Invitrogen) and random primer mix (NEB) according to manufacturer’s instructions.
2μl of the RT reaction was used directly as a template in a total volume of 20μl of real time PCR reaction with Rotor-Gene SYBR Green PCR Kit (Qiagen) according to manufacturer’s instructions. All the primers were used at a concentration of 1μM and are listed in Table 4 in S1 Supplemental Experimental Procedures. Real-time PCR was carried out in qTower 2.0 system (Analytic Jena). Reactions were run in triplicate with an initial activation step for 5mins at 95°C followed by 40 cycles of 5sec at 95°C and 10sec at 60°C. For each reaction, a no-template as well as a no-RT control sample for each condition was also run in parallel. Comparative quantification analyses of PCR products and melting curve analyses were carried out using the qPCRsoft 2.1 software. Expression data were analyzed using the 2(-ΔΔCt) method and normalized to the expression of housekeeping gene Tbp1 for the variability in RNA levels in each sample.
We performed the same protocol for RIP as in [13] with the exception that 3μg of anti-Y12 antibody (ThermoScientific) (or control mouse IgG (Santa Cruz)) were added to the RIP reaction instead of anti-UPF1 antibody.
All statistical analyses for the immunofluorescence and RIP data were performed with two-tailed, unpaired t-test using the statistics tool of GraphPad Prism (GraphPad Software, Inc.).
Differential intron usage was assessed using DEXSeq (v1.16.10) R package [53] and differential exon and junction usage was analyzed via R package JunctionSeq (v0.6.16) [54]. For more details, see Supplemental Experimental Procedures.
|
10.1371/journal.pgen.1002024 | Rif1 Supports the Function of the CST Complex in Yeast Telomere Capping | Telomere integrity in budding yeast depends on the CST (Cdc13-Stn1-Ten1) and shelterin-like (Rap1-Rif1-Rif2) complexes, which are thought to act independently from each other. Here we show that a specific functional interaction indeed exists among components of the two complexes. In particular, unlike RIF2 deletion, the lack of Rif1 is lethal for stn1ΔC cells and causes a dramatic reduction in viability of cdc13-1 and cdc13-5 mutants. This synthetic interaction between Rif1 and the CST complex occurs independently of rif1Δ-induced alterations in telomere length. Both cdc13-1 rif1Δ and cdc13-5 rif1Δ cells display very high amounts of telomeric single-stranded DNA and DNA damage checkpoint activation, indicating that severe defects in telomere integrity cause their loss of viability. In agreement with this hypothesis, both DNA damage checkpoint activation and lethality in cdc13 rif1Δ cells are partially counteracted by the lack of the Exo1 nuclease, which is involved in telomeric single-stranded DNA generation. The functional interaction between Rif1 and the CST complex is specific, because RIF1 deletion does not enhance checkpoint activation in case of CST-independent telomere capping deficiencies, such as those caused by the absence of Yku or telomerase. Thus, these data highlight a novel role for Rif1 in assisting the essential telomere protection function of the CST complex.
| Protection of chromosome ends is crucial for maintaining chromosome stability and genome integrity, and its failure leads to genome rearrangements that may facilitate carcinogenesis. This protection is achieved by the packaging of chromosome ends into protective structures called telomeres that prevent DNA repair/recombination activities. Telomeric DNA is bound and stabilized by two protein complexes named CST and shelterin, which are present in a wide range of multicellular organisms. Whether structural and functional connections exist between these two capping complexes is an important issue in telomere biology. Here, we investigate this topic by analyzing the consequences of disabling the two Saccharomyces cerevisiae shelterin-like components, Rif1 and Rif2, in different hypomorphic mutants defective in CST components. We demonstrate that Rif1 plays a previously unanticipated role in assisting the essential telomere protection function of the CST complex, indicating a tight coupling between CST and Rif1. As CST complexes have been recently identified also in other organisms, including humans, which all rely on shelterin for telomere protection, this functional link between CST and shelterin might be an evolutionarily conserved common feature to ensure telomere integrity.
| Telomeres, the specialized nucleoprotein complexes at the ends of eukaryotic chromosomes, are essential for genome integrity. They protects chromosome ends from fusions, DNA degradation and recognition as DNA double-strand breaks (DSBs) that would otherwise lead to chromosome instability and cell death (reviewed in [1]). Telomeric DNA in the budding yeast Saccharomyces cerevisiae, as well as in nearly all other eukaryotes examined to date, comprise short TG-rich repeated sequences ending in a short single-stranded 3′ overhang (G tail) that corresponds to the strand bearing the TG-rich repeats. The addition of telomeric repeats depends on the action of telomerase, a specialized reverse transcriptase that extends the TG-rich strand of chromosome ends. Recruitment/activation of this enzyme requires the Cdc13 protein that binds to the telomeric TG-rich single-stranded DNA (ssDNA) [2]–[6]. The direct interaction between Cdc13 and the Est1 regulatory subunit of telomerase is essential for telomerase recruitment, and it is disrupted by the cdc13-2 mutation that leads to gradual telomere erosion and accompanying senescence [2], [4], [7].
The average length of S. cerevisiae telomeric 3′ overhangs is 12–14 nucleotides, although it can increase to ∼50 nucleotides during the late S/G2 phase of the cell cycle [8]–[10]. While single-stranded telomeric G-tails can arise after removal of the last RNA primer during lagging-strand replication, the blunt ends of the leading-strand telomere must be converted into 3′ overhangs by resection of the 5′ strand. This 5′ to 3′ nucleolytic degradation involves several proteins, such as the MRX complex, the nucleases Exo1 and Dna2 and the helicase Sgs1 [10], [11]. Cyclin-dependent kinase activity (Cdk1 in S. cerevisiae) is also required for generation of the extended single-stranded overhangs in late S phase [12], [13]. As Cdk1 activity is low in G1, telomere resection can occur only during S/G2 [8], coinciding with the time frame in which G-tails are lengthened and can serve to recruit telomerase.
Keeping the G tail in check is crucial to ensure telomere stability, and studies in budding yeast have shown that Cdc13 prevents inappropriate generation of ssDNA at telomeric ends [2], [14], [15]. This essential capping function depends on Cdc13 interaction with the Stn1 and Ten1 proteins to form the so-called CST (Cdc13-Stn1-Ten1) complex. This complex binds to telomeric ssDNA repeats and exhibits structural similarities with the heterotrimeric ssDNA binding complex Replication protein A (RPA) [16], suggesting that CST is a telomere-specific version of RPA. Loss of Cdc13 function through either the cdc13-1 temperature sensitive allele or the cdc13-td conditional degron allele results in telomere C-strand degradation, leading to activation of the DNA damage checkpoint [13], [14], [17], [18]. Similarly, temperature sensitive mutations in either STN1 or TEN1 genes cause telomere degradation and checkpoint-mediated cell cycle arrest [19]–[21]. Interestingly, Stn1 interacts with Pol12 [22], a subunit of the DNA polymerase α (polα)-primase complex with putative regulatory functions, while Cdc13 interacts with the polα catalytic subunit of the same complex [7], suggesting that CST function might be tightly coupled to the priming of telomeric C strand synthesis. In any case, it is so far unknown whether the excess of telomeric ssDNA in cst mutants arises because the CST complex prevents the access of nuclease/helicase activities to telomeric ends and/or because it promotes polα-primase-dependent C strand synthesis.
In addition to the capping function, a role for the CST complex in repressing telomerase activity has been unveiled by the identification of cdc13, stn1 and ten1 alleles with increased telomere length [2], [21], [23], [24]. The repressing effect of Cdc13 appears to operate through an interaction between this protein and the C-terminal domain of Stn1 [25], [26], which has been proposed to negatively regulate telomerase by competing with Est1 for binding to Cdc13 [4], [24].
A second pathway involved in maintaining the identity of S. cerevisiae telomeres relies on a complex formed by the Rap1, Rif1 and Rif2 proteins. Although only Rap1 is the only shelterin subunit conserved in budding yeast, the Rap1-Rif1-Rif2 complex functionally recapitulates the shelterin complex acting at mammalian telomeres (reviewed in [27]). Rap1 is known to recruit its interacting partners Rif1 and Rif2 to telomeric double-stranded DNA via its C-terminal domain [28]–[30]. This complex negatively regulates telomere length, as the lack of either Rif1 or Rif2 causes telomere lengthening, which is dramatically increased when both proteins are absent [30]. The finding that telomere length in rif1Δ rif2Δ double mutant is similar to that observed in RAP1 C-terminus deletion mutants [30] suggests that Rap1-dependent telomerase inhibition is predominantly mediated by the Rif proteins. However, Rif proteins have been shown to regulate telomere length even when the Rap1 C-terminus is absent [31], suggesting that they can be brought to telomeres independently of Rap1.
In addition to negatively regulate telomere length, Rap1 and Rif2 inhibit both nucleolytic processing and non homologous end joining (NHEJ) at telomeres [32]–[34]. Telomeric ssDNA generation in both rif2Δ and rap1ΔC cells requires the MRX complex [33], and the finding that MRX association at telomeres is enhanced in rif2Δ and rap1ΔC cells [33], [35] suggests that Rap1 and Rif2 likely prevent MRX action by inhibiting MRX recruitment onto telomeric ends. Interestingly, the checkpoint response is not elicited after inactivation of Rap1 or Rif2, suggesting that either the accumulated telomeric ssDNA is insufficient for triggering checkpoint activation or this ssDNA is still covered by Cdc13, which can inhibit the association of the checkpoint kinase Mec1 to telomeres [36]. Notably, Rif1 is not involved in preventing telomeric fusions by NHEJ [32] and its lack causes only a slight increase in ssDNA generation at a de novo telomere [33]. These findings, together with the observation that Rif1 prevents telomerase action independently of Rif2, indicate that Rif1 and Rif2 play different functions at telomeres.
As both CST and the shelterin-like complex contribute to telomere protection, we asked whether and how these two capping complexes are functionally connected. We found that the viability of cells with defective CST complex requires Rif1, but not Rif2. In fact, RIF1 deletion increases the temperature sensitivity of cdc13-1 cells and impairs viability of cdc13-5 cells at any temperature. Furthermore, the rif1Δ and stn1ΔC alleles are synthetically lethal. By contrast, the lack of Rif2 has no effects in the presence of the same cdc13 and stn1 alleles. We also show that cdc13-1 rif1Δ and cdc13-5 rif1Δ cells accumulate telomeric ssDNA that causes hyperactivation of the DNA damage checkpoint, indicating that loss of Rif1 exacerbates telomere integrity defects in cdc13 mutants. By contrast, deletion of RIF1 does not enhance either cell lethality or checkpoint activation in yku70Δ or est2Δ telomere capping mutants. Thus, Rif1 is required for cell viability specifically when CST activity is reduced, highlighting a functional link between Rif1 and CST.
Yeast cells harbouring the cdc13-1 temperature-sensitive allele of the gene encoding the essential telomeric protein Cdc13 are viable at permissive temperature (20–25°C), but die at restrictive temperature (26–37°C), likely due to accumulation of ssDNA at telomeres caused by the loss of Cdc13 capping functions [14]. As also the shelterin-like complex contributes to the maintenance of telomere integrity, we investigated its possible functional connections with Cdc13 by disabling either Rif1 or Rif2 in cdc13-1 cells. Deletion of RIF2 did not affect cdc13-1 cell viability in YEPD medium at any tested temperature (Figure 1A). By contrast, cdc13-1 rif1Δ cells showed a maximum permissive temperature for growth of 20°C and were unable to grow at 25°C, where cdc13-1 single mutant cells could grow at almost wild type rate (Figure 1A). The enhanced temperature-sensitivity of cdc13-1 rif1Δ cells was due to the lack of RIF1, because the presence of wild type RIF1 on a centromeric plasmid allowed cdc13-1 rif1Δ cells to grow at 25°C (Figure 1B). The synthetic effect of the cdc13-1 rif1Δ combination was not uncovered during a previous genome wide search for gene deletions enhancing the temperature-sensitivity of cdc13-1 cells [37], likely because that screening was done at 20°C, a temperature at which cdc13-1 rif1Δ double mutants do not show severe growth defects (Figure 1A). Our data above indicate that Rif1, but not Rif2, is required to support cell viability when Cdc13 protective function is partially compromised.
If the lack of Rif1 in cdc13-1 cells increased the temperature-sensitivity by exacerbating the telomere end protection defects of these cells, Rif1 overexpression might suppress the temperature sensitivity caused by the cdc13-1 allele. Indeed, high copy number plasmids carrying wild type RIF1, which had no effect on wild type cell viability, improved the ability of cdc13-1 cells to form colonies on synthetic selective medium at the semi-permissive temperature of 26–27°C (Figure 1C).
The function of Cdc13 in telomere protection is mediated by its direct interactions with Stn1 and Ten1, leading to formation of the CST complex (reviewed in [38]). In addition to the capping function, the CST complex is implicated in repression of telomerase action [2], [21], [23], [24]. This CST-dependent negative regulation of telomerase can be separated from CST capping function, as yeast cells either carrying the cdc13-5 allele or lacking the Stn1 C-terminus (residues 282–494) (stn1ΔC) display extensive telomere elongation but no or minimal growth defects [24]–[26]. We evaluated the specificity of the genetic interaction between rif1Δ and cdc13-1 by analysing the consequences of deleting RIF1 and RIF2 in cdc13-5 or stn1ΔC cells. Deletion of RIF1 turned out to reduce cell viability of cdc13-5 mutant cells at any temperatures, while deletion of RIF2 did not (Figure 2A). Furthermore, meiotic tetrad dissection of stn1ΔC/STN1 rif1Δ/RIF1 diploid cells did not allow the recovery of viable stn1ΔC rif1Δ double mutant spores (Figure 2B), indicating that rif1Δ and stn1ΔC were synthetic lethal. By contrast, viable rif2Δ stn1ΔC spores were found with the expected frequency after tetrad dissection of stn1ΔC/STN1 rif2Δ/RIF2 diploid cells (Figure 2C). The observed synthetic phenotypes suggest that both stn1ΔC and cdc13-5 cells have capping deficiencies and that the lack of Rif1 enhances their protection defects. Consistent with this hypothesis, cdc13-5 and stn1ΔC mutants were shown to accumulate telomeric ssDNA, although the amount of this ssDNA was not enough to invoke a DNA damage response [24], [25]. We conclude that Rif1, but not Rif2, is required to support cell viability when a partial inactivation of CST capping function occurs.
A Cdc13 specific function that is not shared by the other subunits of the CST complex is its requirement for recruitment/activation of telomerase at chromosome ends [2]–[6]. Cdc13-mediated telomerase recruitment is disrupted by the cdc13-2 mutation, which leads to progressive telomere shortening and senescence phenotype [4]. We therefore asked whether RIF1 deletion influences viability and/or senescence progression of cdc13-2 cells. Viable cdc13-2 rif1Δ spores were recovered after tetrad dissection of cdc13-2/CDC13 rif1Δ/RIF1 diploid cells (data not shown), indicating that the lack of Rif1 does not affect the overall viability of cdc13-2 cells. When spores from the dissection plate were streaked on YEPD plates for 4 successive times, the decline in growth of cdc13-2 and cdc13-2 rif1Δ spores occurred with similar kinetics (Figure 2D), indicating that RIF1 deletion did not accelerate the senescence phenotype of cdc13-2 cells specifically defective in telomerase recruitment. Taken together, these genetic interactions indicate that Rif1, but not Rif2, has a role in assisting the essential function of the CST complex in telomere protection.
The CST complex functionally and physically interacts with the polα-primase complex [7], [21], [22], [25], which is essential for telomeric C-strand synthesis during telomere elongation. Thus, we analyzed the genetic interactions between rif1Δ and temperature sensitive alleles affecting DNA primase (pri2-1) [39] or polα (cdc17-1 and pol1-1) [40], [41]. Both cdc17-1 rif1Δ and pol1-1 rif1Δ cells were viable, but their temperature-sensitivity was greatly enhanced compared to cdc17-1 and pol1-1 single mutants (Figure 2E). Similarly, the maximal permissive temperature of the pri2-1 rif1Δ double mutant was reduced relative to that of pri2-1 single mutant cells (Figure 2F). Moreover both pol1-1 rif1Δ and pri2-1 rif1Δ cells showed growth defects even at the permissive temperature of 25°C (Figure 2E and 2F). Thus, Rif1, like CST, functionally interacts with the polα-primase complex.
The synthetic effects of combining rif1Δ with cdc13 and stn1 mutations suggest that Rif1 might normally assist the Cdc13 and Stn1 proteins in carrying out their essential telomere protection functions. It is known that cdc13-1 cells undergo checkpoint-dependent metaphase arrest when incubated at the restrictive temperature [14]. Failure to turn on the checkpoint allows cdc13-1 cells to form colonies at 28°C [42], [43], indicating that checkpoint activation can partially account for the loss of viability of cdc13-1 cells. We then asked whether the enhanced temperature sensitivity of cdc13-1 rif1Δ cells compared to cdc13-1 cells might be due to upregulation of the DNA damage checkpoint response. Deletion of the checkpoint gene RAD9, which partially suppressed the temperature sensitivity of cdc13-1 mutant cells, slightly improved the ability of cdc13-1 rif1Δ cells to grow at 23–25°C (Figure 3A), indicating that the synthetic interaction between Rif1 and Cdc13 can be partially alleviated by checkpoint inactivation. Furthermore, when wild type, rif1Δ, cdc13-1 and cdc13-1 rif1Δ cell cultures were arrested in G1 with α-factor at 20°C (permissive temperature) and then released from G1 arrest at 25°C (non-permissive temperature for cdc13-1 rif1Δ cells), they all replicated DNA and budded with similar kinetics after release (Figure 3B and 3C). However, most cdc13-1 rif1Δ cells then arrested in metaphase as large budded cells with a single nucleus, while wild type, cdc13-1 and rif1Δ cells divided nuclei after 75–90 minutes (Figure 3D).
To assess whether the cell cycle arrest of cdc13-1 rif1Δ cells was due to DNA damage checkpoint activation, we examined the Rad53 checkpoint kinase, whose phosphorylation is necessary for checkpoint activation and can be detected as changes in Rad53 electrophoretic mobility. Rad53 was phosphorylated in cdc13-1 rif1Δ cells that were released from G1 arrest at 25°C, whereas no Rad53 phosphorylation was seen in any of the other similarly treated cell cultures (Figure 3E).
RIF1 deletion caused a checkpoint-mediated G2/M cell cycle arrest also in cdc13-5 cells. In fact, exponentially growing cdc13-5 rif1Δ cell cultures at 25°C contained a higher percentage of large budded cells with a single nucleus than rif1Δ or cdc13-5 cell cultures under the same conditions (Figure 3F). Furthermore, Rad53 phosphorylation was detected in these cdc13-5 rif1Δ cells, but not in the rif1Δ and cdc13-5 cell cultures (Figure 3G). Thus, the lack of Rif1 results in DNA damage checkpoint activation in both cdc13-1 and cdc13-5 cells under conditions that do not activate the checkpoint when Rif1 is present.
The lack of Rif1 is known to cause telomere overelongation [29]. Thus, we examined telomere length in cdc13-1 rif1Δ double mutant cells. The length of duplex telomeric DNA was examined after transferring at 25°C cell cultures exponentially growing at 20°C, followed by Southern blot analysis with a TG-rich probe of XhoI-digested genomic DNA prepared at different times after shift at 25°C (Figure 4A). As expected [29], rif1Δ mutant cells had longer telomeres than wild type and cdc13-1 cells (Figure 4A). Telomeres in cdc13-1 rif1Δ double mutant cells either at 20°C or after incubation at 25°C were longer than those of wild type and cdc13-1 cells, but undistinguishable from those of rif1Δ cells (Figure 4A). Not only RIF1 deletion, but also the cdc13-5 mutation is known to cause telomere overelongation [24] (Figure 4B). Interestingly, when telomere length was analyzed in cdc13-5 rif1Δ double mutant cells grown at 25°C, telomeres were longer in cdc13-5 rif1Δ double mutant cells than in cdc13-5 and rif1Δ single mutants (Figure 4B), indicating that the cdc13-5 mutation exacerbates the telomere overelongation defect caused by the lack of Rif1.
The finding that telomeres in cdc13-1 rif1Δ double mutant cells at 25°C were longer than those of cdc13-1 cells, but undistinguishable from those of cdc13-1 rif1Δ cells grown at 20°C (Figure 4A) suggests that the growth defects of cdc13-1 rif1Δ cells at 25°C are not due to rif1Δ-induced telomere overelongation. Telomere lengthening in rif1Δ mutant cells is telomerase-dependent [44] and requires the action of the checkpoint kinase Tel1 that facilitates telomerase recruitment [45], [46]. To provide additional evidences that loss of viability in cdc13 rif1Δ mutants occurs independently of rif1Δ-induced alterations in telomere length, we asked whether RIF1 deletion was still deleterious in cdc13-1, cdc13-5 and stn1ΔC cells in a context where telomeres cannot be elongated due to the lack of Tel1 [45]. We found that TEL1 deletion did not alleviate the growth defects of cdc13-1 rif1Δ cells (Figure 5A). Rather, cdc13-1 tel1Δ and cdc13-1 rif1Δ tel1Δ cells showed an enhanced temperature sensitivity compared to cdc13-1 and cdc13-1 rif1Δ cells, respectively, presumably due to the combined effects of loss of a telomere elongation mechanism and inability to protect telomeres from shortening activities. Furthermore, the growth defects of cdc13-5 rif1Δ double mutant cells were similar to those of cdc13-5 rif1Δ tel1Δ triple mutant cells (Figure 5B). Finally, viable stn1ΔC rif1Δ tel1Δ mutant spores could not be recovered after meiotic tetrad dissection of stn1ΔC/STN1 rif1Δ/RIF1 tel1Δ/TEL1 diploid cells (data not shown), indicating that stn1ΔC and rif1Δ were synthetic lethal even in the absence of Tel1.
As telomere lengthening is dramatically increased when both Rif1 and Rif2 are absent [30], we also investigated whether the absence of Rif2 exacerbates cdc13-1 rif1Δ growth defects. As shown in Figure 5C, cdc13-1 rif1Δ rif2Δ cells formed colonies at the maximum temperature of 20°C and behaved similarly to cdc13-1 rif1Δ cells. We therefore conclude that the synthetic interaction between rif1Δ and cdc13 alleles is not due to rif1Δ-induced alterations in telomere length, but it is a direct consequence of Rif1 loss.
It is known that cdc13-1 cells at 37°C accumulate telomeric ssDNA that triggers checkpoint-mediated cell cycle arrest [14]. Thus, we investigated whether cdc13-1 rif1Δ and cdc13-5 rif1Δ cells contained aberrant levels of single-stranded TG sequences at their telomeres that could be responsible for loss of viability in cdc13-1 rif1Δ and cdc13-5 rif1Δ cells at 25°C. The integrity of chromosome ends was analyzed by an in-gel hybridization procedure [9], probing for the presence of single-stranded TG sequences. Both cdc13-1 and rif1Δ single mutants either grown at 20°C (Figure 6A, lanes 2 and 4) or incubated at 25°C for 3 hours (Figure 6A, lanes 6 and 8) showed only a very slight increase in single-stranded TG sequences compared to wild type (Figure 6A, lanes 1 and 5). By contrast, cdc13-1 rif1Δ double mutant cells contained higher amounts of telomeric ssDNA than cdc13-1 and rif1Δ cells already at 20°C (Figure 6A, lane 3) and the amount of this ssDNA increased dramatically when cdc13-1 rif1Δ cells were incubated at 25°C for 3 hours (Figure 6A, lane 7). A similar telomere deprotection defect was observed also for cdc13-5 rif1Δ cells grown at 25°C (Figure 6A, lane 11), which displayed an increased amount of telomeric ssDNA compared to similarly treated wild type and cdc13-5 cells (Figure 6A, lanes 9 and 10).
Because the length of single-stranded G overhangs increases during S phase [8], the strong telomeric ssDNA signals observed in cdc13-1 rif1Δ cell cultures at 25°C (Figure 6A) might be due to an enrichment of S/G2 cells. We ruled out this possibility by monitoring the levels of single-stranded TG sequences in cdc13-1 rif1Δ cell cultures that were arrested in G2 with nocodazole at 20°C and then transferred to 25°C in the presence of nocodazole for 3 hours (Figure 6B). Similarly to what we observed in exponentially growing cell cultures, G2-arrested cdc13-1 rif1Δ cells at 20°C displayed increased amounts of ssDNA compared to each single mutant under the same conditions, and incubation at 25°C led to further increase of this ssDNA (Figure 6B). Taken together, these findings indicate that the lack of Rif1 causes a severe defect in telomere protection when Cdc13 activity is partially compromised.
If telomeric ssDNA accumulation contributes to checkpoint activation in cdc13-1 rif1Δ and cdc13-5 rif1Δ cells, then mutations reducing ssDNA generation should alleviate the arrest and relieve the lethality caused by the lack of Rif1 in cdc13-1 and cdc13-5 background. Because the Exo1 nuclease contributes to generate telomeric ssDNA in cdc13-1 cells [47], we examined the effect of deleting EXO1 in cdc13 rif1Δ cells. When G2-arrested cell cultures at 20°C were transferred to 25°C for 3 hours, cdc13-1 rif1Δ exo1Δ triple mutant cells contained significantly lower amounts of telomeric ssDNA than cdc13-1 rif1Δ cells (Figure 6B). A similar behaviour of the triple mutant was detectable even when G2-arrested cultures where kept at 20°C, although the quantity of telomeric ssDNA accumulated by cdc13-1 rif1Δ cells at this temperature was lower than at 25°C (Figure 6B). Furthermore, EXO1 deletion partially suppressed both the temperature-sensitivity of cdc13-1 rif1Δ cells (Figure 7A) and the loss of viability of cdc13-5 rif1Δ cells (Figure 7B), further supporting the hypothesis that reduced viability in these strains was due to defects in telomere protection.
Exo1-mediated suppression of the cdc13 rif1Δ growth defects correlated with alleviation of checkpoint-mediated cell cycle arrest. In fact, when cell cultures exponentially growing at 20°C were incubated at 25°C for 3 hours, the amount of both metaphase-arrested cells and Rad53 phosphorylation was reproducibly lower in cdc13-1 rif1Δ exo1Δ cells than in cdc13-1 rif1Δ cells (Figure 7C and 7D). Similar results were obtained also with cdc13-5 rif1Δ exo1Δ cells growing at 25°C, which accumulated less metaphase-arrested cells and phosphorylated Rad53 than similarly treated cdc13-5 rif1Δ cells (Figure 7E and 7F). Thus, both cell lethality and checkpoint-mediated cell cycle arrest in cdc13 rif1Δ cells appear to be caused, at least partially, by Exo1-dependent telomere DNA degradation.
The lack of Rif1 might increase the lethality of cells with reduced CST activity just because it causes a telomere deprotection defect that exacerbates the inherent telomere capping defects of cdc13 or stn1 mutants. If this hypothesis were correct, RIF1 deletion should affect viability also of other non-CST mutants defective in end protection. Alternatively, Rif1-CST functional interaction might be specific, thus reflecting a functional connection between Rif1 and CST. To distinguish between these two possibilities, we analyzed the effects of deleting RIF1 in Yku70 lacking cells, which display Exo1-dependent accumulation of telomeric ssDNA, as well as checkpoint-mediated cell cycle arrest at elevated temperatures (37°C) [47], [51]. Loss of Yku in est2Δ cells, which lack the telomerase catalytic subunit, leads to synthetic lethality, presumably due to the combined effects of telomere shortening and capping defects [48]–[50], [52]. As expected [53], yku70Δ cells were viable at 25°C and 30°C, but they were unable to form colonies at 37°C (Figure 8A). Similarly, yku70Δ rif1Δ double mutant cells grew well at 25°C and 30°C (Figure 8A) and did not show Rad53 phosphorylation when grown at 25°C (Figure 8B, time 0). Furthermore, similar amounts of phosphorylated Rad53 were detected in both yku70Δ and yku70Δ rif1Δ cell cultures that were kept at 37°C for 4 hours (Figure 8B), indicating that loss of Rif1 does not enhance the telomere protection defects already present in yku70Δ cells. Consistent with a previous observation [54], RIF1 deletion partially suppressed the temperature sensitivity (Figure 8A) and the telomere length defect (data not shown) caused by the lack of Yku70, suggesting that the elongated state of the telomeres could be the reason why yku70Δ rif1Δ cells can proliferate at 37°C.
Checkpoint activation can also be induced during telomere erosion caused by insufficient telomerase activity [55], [56]. Thus, we asked whether RIF1 deletion accelerated senescence progression and/or upregulated checkpoint activation in cells lacking the telomerase catalytic subunit Est2. Meiotic tetrads were dissected from a diploid strain heterozygous for the est2Δ and rif1Δ alleles, which are recessive and therefore do not affect telomere length in the diploid. After 2 days of incubation at 25°C (approximately 25 generations), spore clones from the dissection plate were both streaked for 4 successive times (Figure 8C) and propagated in YEPD liquid medium to prepare protein extracts for Rad53 phosphorylation analysis at different time points (Figure 8D). Similar to what was previously observed [44], RIF1 deletion did not accelerate senescence progression in est2Δ cells, as est2Δ rif1Δ clones showed a decline in growth similar to that of est2Δ clones (Figure 8C). Furthermore, est2Δ and est2Δ rif1Δ cell cultures showed similar patterns of Rad53 phosphorylation with increasing number of generations (Figure 8D). Thus, the lack of Rif1 does not enhance either DNA damage checkpoint activation or senescence progression during telomere erosion caused by the lack of telomerase.
Finally, because the telomerase machinery is known to be recruited to an unrepaired DSB [57], we ruled out the possibility of a general role for Rif1 in inhibiting checkpoint activation by examining activation/deactivation of the checkpoint induced by an unrepaired DSB. To this end, we used JKM139 derivative strains, where a single DSB can be generated at the MAT locus by expressing the site-specific HO endonuclease gene from a galactose-inducible promoter [58]. This DSB cannot be repaired by homologous recombination, because the homologous donor sequences HML or HMR are deleted. As shown in Figure 8E, when G1-arrested cell cultures were spotted on galactose containing plates, both wild type and rif1Δ JKM139 derivative cells overrode the checkpoint-mediated cell cycle arrest within 24–32 hours, producing microcolonies with 4 or more cells. Moreover, when galactose was added to exponentially growing cell cultures of the same strains, Rad53 phosphorylation became detectable as electrophoretic mobility shift in both wild type and rif1Δ cell cultures about 2 hours after HO induction, and it decreased in both cell cultures after 12–15 hours (Figure 8F), when most cells resumed cell cycle progression (data not shown). Thus, Rif1 does not affect the checkpoint response to an irreparable DSB. Altogether these data indicate that Rif1 supports specifically CST functions in telomere protection.
Both shelterin and CST complexes are present in a wide range of unicellular and multicellular organisms, where they protect the integrity of chromosomes ends (reviewed in [38]). Thus, the understanding of their structural and functional connections is an important issue in telomere regulation. We have approached this topic by analysing the consequences of disabling the shelterin-like S. cerevisiae proteins Rif1 or Rif2 in different hypomorphic mutants defective in CST components. We provide evidence that Rif1, but not Rif2, is essential for cell viability when the CST complex is partially compromised. In fact, RIF1 deletion exacerbates the temperature sensitivity of cdc13-1 mutant cells that are primarily defective in Cdc13 telomere capping functions. Furthermore, cells carrying the cdc13-5 or the stn1ΔC mutation, neither of which causes per se DNA damage checkpoint activation and growth defects [24], [26], grow very poorly or are unable to form colonies, respectively, when combined with the rif1Δ allele. By contrast, RIF1 deletion does not affect either viability or senescence progression of cdc13-2 cells, which are specifically defective in telomerase recruitment. This Cdc13 function is not shared by the other CST subunits, suggesting that Rif1 is specifically required to support the essential capping functions of the CST complex.
Cell lethality caused by the absence of Rif1 in both cdc13-1 and cdc13-5 cells appears to be due to severe telomere integrity defects. In fact, telomeres in both cdc13-1 rif1Δ and cdc13-5 rif1Δ double mutant cells display an excess of ssDNA that leads to DNA damage checkpoint activation. Deleting the nuclease EXO1 gene partially restores viability of cdc13-1 rif1Δ and cdc13-5 rif1Δ cells and reduces the level of telomeric ssDNA in cdc13-1 rif1Δ cells, indicating that cell lethality in cdc13 rif1Δ cells is partially due to Exo1-dependent telomere DNA degradation and subsequent activation of the DNA damage checkpoint.
Although Rif1 and Rif2 interact both with the C-terminus of Rap1 and with each other [29], [30], our finding that only Rif1 is required for cell viability when Cdc13 or Stn1 capping activities are reduced indicates that Rif1 has a unique role in supporting CST capping function that is not shared by Rif2. Earlier studies are consistent with the idea that Rif1 and Rif2 regulate telomere metabolism by different mechanisms [30], [31], [35]. Furthermore, while the content of Rif2 is lower at shortened than at wild type telomeres, the level of Rif1 is similar at both, suggesting that these two proteins are distributed differently along a telomere [59]. Finally, inhibition of telomeric fusions requires Rif2, but not Rif1 [32].
Noteworthy, although RIF1 deletion is known to cause telomere overelongation [29], the synthetic interaction between Rif1 and CST occurs independently of rif1Δ-induced alterations in telomere length. In fact, the lack of Tel1, which counteracts rif1Δ-induced telomere overelongation [45], does not alleviate the growth defects of cdc13 rif1Δ cells. Furthermore, deletion of RIF2, which enhances telomere elongation induced by the lack of Rif1 [30], does not exacerbate the synthetic phenotypes of cdc13 rif1Δ double mutant cells. Thus, loss of viability in cdc13 rif1Δ cells is not due to telomere overelongation caused by RIF1 deletion, but it is a direct consequence of Rif1 loss.
By analyzing the effects of combining RIF1 deletion with mutations that cause telomere deprotection without affecting CST functions, we found that the functional interaction between Rif1 and the CST complex is highly specific. In fact, the lack of Rif1 does not enhance the DNA damage checkpoint response in telomerase lacking cells, which are known to experience gradual telomere erosion leading to activation of the DNA damage checkpoint [55], [56]. Furthermore, RIF1 deletion does not upregulate DNA damage checkpoint activation in yku70Δ cells, which display Exo1-dependent accumulation of ssDNA and checkpoint-mediated cell cycle arrest at 37°C [47]–[51]. This is consistent with previous observations that comparable signals for G strand overhangs can be detected on telomeres derived from yku70Δ and yku70Δ rif1Δ cells [54], indicating that RIF1 deletion does not exacerbate the end protection defect due to the absence of Yku. By contrast, the lack of Rif1 partially suppresses both temperature-sensitivity and telomere shortening in yku70Δ cells (Figure 8A) [54], possibly because the restored telomere length helps to compensate for yku70Δ capping defects. Notably, although RIF1 deletion leads to telomere overelongation in cdc13-1 and cdc13-5 mutants, this elongated telomere state does not help to increase viability in cdc13-1 rif1Δ and cdc13-5 rif1Δ cells.
The simplest interpretation of the specific genetic interactions we found between Rif1 and CST is that a functional connection exists between Rif1 and the CST complex, such that Rif1 plays a previously unanticipated role in assisting the CST complex in carrying out its essential telomere protection function. Indeed, this functional interaction is unexpected in light of Rif1 and CST localization along a telomere. In fact, while CST is present at the very ends of chromosomes, Rif1 is thought to be distributed centromere proximal on the duplex telomeric DNA [59]. However, as yeast telomeres have been proposed to fold back onto the subtelomeric regions to form a ∼3-kb region of core heterochromatin [60], [61], this higher-order structure could place Rif1 and CST in close proximity, thus explaining their functional interaction.
The function of Rif1 in sustaining CST activity cannot be simply attributable to the Rif1-mediated suppression of ssDNA formation at telomeres, as rif1Δ cells show only a very slight increase in ssDNA at both native (Figure 6) and HO-induced telomeres [33] compared to wild type. Furthermore, although deletion of Rif2 leads to increased amounts of telomeric ssDNA [33], cdc13-1 rif2Δ, cdc13-5 rif2Δ and stn1ΔC rif2Δ double mutants are viable and do not display growth defects. Finally, other mutants defective in telomere capping or telomere elongation (yku70Δ and est2Δ) are perfectly viable in the absence of Rif1.
One possibility is that Rif1 physically interacts, directly or indirectly, with the CST complex. Indeed, human Stn1 was found to copurify with the shelterin subunit TPP1 [62], suggesting the existence of CST-shelterin complexes in mammals. Unfortunately, we were so far unable to coimmunoprecipitate Rif1 with Cdc13 or Stn1, and further analyses will be required to determine whether Rif1 and the CST complex undergo stable or transient association during the cell cycle.
Indeed, not only 5′-3′ resection, but also incomplete synthesis of Okazaki fragments is expected to increase the size of the G tail during telomere replication. The yeast CST complex genetically and physically interacts with the polα-primase complex [7], [22], [25] and the human CST-like complex increases polα-primase processivity [63], [64]. Furthermore, the lack of CST function in G1 and throughout most of S phase does not lead to an increase of telomeric ssDNA [13], suggesting that the essential function of CST is restricted to telomere replication in late S phase. Altogether, these observations suggest that CST may control overhang length not only by blocking the access of nucleases, but also by activating polα-primase-dependent C-strand synthesis that can compensate G tail lengthening activities. Based on the finding that Rif1 regulates telomerase action and functionally interacts with the polα-primase complex (Figure 2), it is tempting to propose that Rif1 favours CST ability to replenish the exposed ssDNA at telomeres through activation/recruitment of polα-primase, thus coupling telomerase-dependent elongation to the conventional DNA replication process.
The recent discoveries that human TPP1 interacts physically with Stn1 [62] and that CST-like complexes exist also in S. pombe, plants and mammals [65]–[68] raise the question of whether functional connections between the two capping complexes exist also in other organisms. As telomere protection is critical for preserving genetic stability and counteracting cancer development, to address this question will be an important future challenge.
Strain genotypes are listed in supplementary Table S1. Unless otherwise stated, the yeast strains used during this study were derivatives of W303 (ho MATa ade2-1 his3-11,15 leu2-3,112 trp1-1 ura3 can1-100). All gene disruptions were carried out by PCR-based methods. The cdc13-1 mutant was kindly provided by D. Lydall (University of Newcastle, UK). The cdc13-2 mutant was kindly provided by V. Lundblad (Salk Institute, La Jolla, USA). The stn1ΔC and cdc13-5 alleles carried a stop codon following amino acids 282 and 694 respectively [24], [25], and were generated by PCR-based methods. Wild type and cdc13-1 strains carrying either the 2 µ vector or 2 µ RIF1 plasmid were constructed by transforming wild type and cdc13-1 strains with plasmids YEplac195 (2 µ URA3) and pML435 (2 µ RIF1 URA3), respectively. The strains used for monitoring checkpoint activation in response to an irreparable DSB were derivatives of strain JKM139 (MATa ho hmlΔ hmrΔ ade1 lys5 leu2-3,112 trp1::hisG ura3-52 ade3::GAL-HO), kindly provided by J. Haber (Brandeis University, Waltham, MA, USA) [58]. To induce HO expression in JKM139 and its derivative strains, cells were grown in raffinose-containing yeast extract peptone (YEP) and then transferred to raffinose- and galactose-containing YEP.
Cells were grown in YEP medium (1% yeast extract, 2% bactopeptone, 50 mg/l adenine) supplemented with 2% glucose (YEPD) or 2% raffinose (YEP+raf) or 2% raffinose and 2% galactose (YEP+raf+gal). Synthetic complete medium lacking uracil supplemented with 2% glucose was used to maintain the selective pressure for the 2 µ URA3 plasmids.
Genomic DNA was digested with XhoI. The resulting DNA fragments were separated by electrophoresis on 0.8% agarose gel and transferred to a GeneScreen nylon membrane (New England Nuclear, Boston), followed by hybridization with a 32P-labelled poly(GT) probe and exposure to X-ray sensitive films. Standard hybridization conditions were used. Visualization of single-stranded overhangs at native telomeres was done by in-gel hybridization [9], using a single-stranded 22-mer CA oligonuleotide probe. The same DNA samples were separated on a 0.8% agarose gel, denatured and hybridized with an end-labeled C-rich oligonucleotide for loading control.
For western blot analysis, protein extracts were prepared by TCA precipitation. Rad53 was detected using anti-Rad53 polyclonal antibodies kindly provided by J. Diffley (Clare Hall, London, UK). Secondary antibodies were purchased from Amersham and proteins were visualized by an enhanced chemiluminescence system according to the manufacturer. Flow cytometric DNA analysis was determined on a Becton-Dickinson FACScan on cells stained with propidium iodide.
|
10.1371/journal.pgen.1005366 | TSPO, a Mitochondrial Outer Membrane Protein, Controls Ethanol-Related Behaviors in Drosophila | The heavy consumption of ethanol can lead to alcohol use disorders (AUDs) which impact patients, their families, and societies. Yet the genetic and physiological factors that predispose humans to AUDs remain unclear. One hypothesis is that alterations in mitochondrial function modulate neuronal sensitivity to ethanol exposure. Using Drosophila genetics we report that inactivation of the mitochondrial outer membrane translocator protein 18kDa (TSPO), also known as the peripheral benzodiazepine receptor, affects ethanol sedation and tolerance in male flies. Knockdown of dTSPO in adult male neurons results in increased sensitivity to ethanol sedation, and this effect requires the dTSPO depletion-mediated increase in reactive oxygen species (ROS) production and inhibition of caspase activity in fly heads. Systemic loss of dTSPO in male flies blocks the development of tolerance to repeated ethanol exposures, an effect that is not seen when dTSPO is only inactivated in neurons. Female flies are naturally more sensitive to ethanol than males, and female fly heads have strikingly lower levels of dTSPO mRNA than males. Hence, mitochondrial TSPO function plays an important role in ethanol sensitivity and tolerance. Since a large array of benzodiazepine analogues have been developed that interact with the peripheral benzodiazepine receptor, the mitochondrial TSPO might provide an important new target for treating AUDs.
| Alcohol use disorders (AUDs) affect millions of patients worldwide and result in high social and economic burdens. Although environmental factors are involved, there are clear genetic components to AUDs. Both the acute sedating effect of alcohol exposure and alcohol tolerance contribute to long term risk for alcohol dependence and addiction. Yet the genetic etiology of AUDs remains to be determined. The mitochondria play a central role in ethanol metabolism and are important in many aspects of cellular physiology such as REDOX and ROS regulation, and apoptosis. The mitochondrial outer membrane translocator protein 18 kDa (TSPO) binds the benzodiazepines and perhaps other addictive drugs, and thus may play a role in AUDs. Since Drosophila is a well-established model for ethanol-related behaviors, we have developed systems for manipulating the Drosophila tspo gene and protein. With these systems, we have discovered that neuronal TSPO controls sensitivity to ethanol sedation via ROS and caspase-mediated signaling and that systemic TSPO levels are important in the development of tolerance to repeated ethanol exposure. Given the variety of known TSPO ligands, and the common mechanisms of various abusive substances, our studies suggest that TSPO might be a promising target to combat alcoholism as well as addiction to other drugs.
| Alcohol is one of the most widely used drugs worldwide, but long term consumption leads to its abuse and dependence. An estimated 17.6 million adults in the United States have AUDs with associated health concerns of alcohol dependence, liver cirrhosis, cancer, and injuries. From 2006 through 2010, this generated an annual average of about 88,000 alcohol-related deaths and 2.5 million years of potential life lost [1,2].
To develop therapeutic strategies for alcoholism it will be necessary to determine the molecular and cellular mechanisms underlying AUDs. Considerable effort has been invested in determining the role of the central nervous system in the etiology of AUD [3–5] but many features of the AUDs remain unexplained.
Neuronal function is highly dependent on mitochondrial bioenergetics [6,7]. In addition to the direct metabolizing of ethanol, the mitochondria are central to a wide range of essential neuronal cell functions including ATP synthesis, ROS production and REDOX homeostasis, Ca2+ buffering, and the metabolic regulation of apoptosis [8–10]. In humans mitochondrial DNA (mtDNA) alterations have been correlated with alcoholism, involving both acute ethanol responses and chronic damage [11–16]. In rodents, hepatic mtDNA depletion is seen in alcohol exposed mice [17] and mtDNA complex I gene variants have been correlated with “non-drinker” versus “drinker” rat lines derived from the same founder strain [18]. Variation in the mtDNA genes have also been shown to have profound effects of nuclear gene expression [19].
In previous studies we showed that the nuclear DNA coded Drosophila translocator protein 18kDa (dTSPO, CG2789) is localized in outer mitochondrial membrane (OMM) and important for regulating mitochondria bioenergetics, ROS production, caspase activity, and apoptotic function [20]. In humans, TSPO ligands are widely used in neuroimaging for neurodegenerative diseases and neuronal injuries, both of which are associated with increased brain TSPO levels and distribution [21]. As the previous nomenclature (peripheral benzodiazepine receptor, PBR) implies, TSPO binds the benzodiazepines and other psychotrophic drugs associated with tolerance and addiction [22]. Thus we hypothesized the TSPO may be an important factor in addiction to ethanol.
Drosophila’s sensitivity and tolerance to ethanol are similar to humans and rodents. Ethanol results in biphasic locomotor alterations. At lower doses ethanol acts as a stimulant, but at higher doses it acts as a depressant [23]. After repeated alcohol stimulation, tolerance is developed, defined as acquired resistance. Tolerance is thought to be an intermediate step to alcohol dependence and addiction [24].
Here we report that in male Drosophila, neuronal inactivation of dTSPO sensitizes flies to ethanol sedation, mediated by increased ROS production and decreased caspase activation. Furthermore, systemic but not neuronal loss of dTSPO inhibits the development of tolerance. By contrast, females are constitutively more sensitive to ethanol sedation than males and they have much lower dTSPO mRNA in their brains. Therefore, the mitochondrial TSPO is an important mediator of ethanol sensitivity and tolerance and contributes to gender-specific differences in alcohol sensitivity.
Acute ethanol sensitivity was analyzed by placing flies in vials closed by cotton clogs soaked with varying concentrations of ethanol thus exposing the flies to ethanol vapor. During initial exposure the flies flew to the top of the vial, and exhibited hyperactivity for a few minutes. With longer exposure, the flies became sedated and remained at bottom of the vial without locomotion. Wild type flies became comparably sedated whether the ethanol-soaked clogs were at the top of the vials or the vials were inverted with clogs at the bottom (S1 Fig). Moreover, the Drosophila showed a dose-dependent response to ethanol using this protocol (Fig 1). Therefore, the ethanol effects observed in the following experiments were due to the ethanol concentration rather than an environmental factor such as hypoxia due to ethanol vapor exclusion of air.
The tspo[EY00814] mutant Drosophila has a P-element inserted into the tspo gene leading to loss of dTSPO expression [20]. Male tspo-/- flies exhibited higher sedation sensitivity than tspo +/+ flies when exposed to ethanol vapor from 34% ethanol solution (Fig 1A) while at 44% or 54% ethanol vapor both the tspo-/- and tspo +/+ flies exhibited the same sensitivity (Fig 1B and 1C). Post sedation, we tested for the recovery from ethanol sedation by replacing the ethanol-soaked clogs with normal clogs. This revealed that at 54% ethanol exposure tspo-/- males were slower to recover than the tspo +/+ male flies (Fig 1D). Thus, tspo-/- male flies are more sensitive to ethanol sedation than their tspo +/+ counterparts. While the rempA gene overlaps with the tspo gene, rempA-/- deficiency is not the cause of the ethanol phenotypes since rempA-/- flies exhibit comparable sensitivity to 34% ethanol vapor as wild type flies (S2 Fig).
Since the tspo mutation is present in all developmental stages of the fly, it could act through creating a developmental abnormality. However, Hematoxylin-Eosin histological comparison of the brains of male tspo-/- and +/+ flies did not reveal any gross anatomical differences (S3 Fig).
To determine whether the increased ethanol sensitivity was attributable to dTSPO function in neurons, we depleted dTSPO in neurons by inducing dsRNA (RNAi) to knockdown dTSPO mRNA in adult flies following eclosion (days after eclosion, dae). We used the Gal4-GeneSwitch/UAS system [25] in which Gal4 is activated within the flies when fed with mifepristone (RU486). The activated Gal4 binds to the UAS of the UAS-dTSPO-RNAi which induces the dsRNA expression and inhibition of the dTSPO mRNA. Since the Gal4 element is expressed under the neuronal specific elav promoter (elav-GeneSwitch), this switch was restricted to neurons. In this way, the flies were permitted to progress through larval and pupal development with normal TSPO activity, and following eclosion, the dTSPO RNAi was induced in neurons by exposure to RU486. Male flies harboring both elav-GeneSwitch and UAS-dTSPO-RNAi (elav-GS/+; TSPO-IR/+)(GS means ‘Gene Switch’ and IR means ‘Inverted Repeats’) cassettes that were exposed to RU486 post eclosion had reduced head dTSPO mRNA as quantified by RT-PCR (Fig 2A). Therefore, activation of the elav-GS/+; TSPO-IR/+ system with RU486 specifically depletes dTSPO mRNAs in the neurons.
In parallel with the whole body knockouts, the elav-GS/+; TSPO-IR/+ RU486 knockdown male flies exhibited faster ethanol sedation in the presence of 44% ethanol vapor than did flies who were not exposed to RU486 (Fig 2B). RU486 exposure of flies harboring only the neuronal elav-GeneSwitch (elav-GS/+) or the UAS-dTSPO-RNAi (TSPO-IR/+) cassette had no effect on the ethanol sensitivity (Fig 2C and 2D). Similarly, elav-GS/+; TSPO-IR/+ male flies exposure to 34% ethanol also showed increase sedation after RU486 induction relative to uninduced flies (S4A Fig). After 55% ethanol sedation (S4B Fig), the RU486-induced flies were slower to recover (S4C Fig). The difference between the RU486 induced and uninduced flies was not due to differential alcohol absorption or metabolism since after a brief exposure to 44% ethanol vapor both groups of fly heads (with and without RU486) had the same ethanol concentration (S4D Fig). Hence, dTSPO inactivation in adult neurons is sufficient to sensitize male flies to ethanol exposure.
Male and female flies exhibit sexual dimorphic response to ethanol exposure [4] and this sexual dimorphism was also observed in the brains of the TSPO knockout and knockdown flies. Male tspo-/- flies showed an increased sensitivity to ethanol sedation relative to tspo +/+ flies with 34% ethanol exposure and delayed recovery from 54% sedation (Fig 1A and 1D) while female tspo-/- and tspo +/+ flies showed no difference in their response to 34% ethanol exposure (Fig 3A–3C).
In elav-GS/+; TSPO-IR/+ female flies, after neuronal inactivation of dTSPO by dsRNA expression, there was no effect on the sedation rate with exposure to 34% or 44% ethanol solution (S5A and S5B Fig). Furthermore, only slightly delayed recovery was seen for female flies after 44% vapor sedation (S5C Fig). The marked difference between male and female elav-GS/+; TSPO-IR/+ flies’ sensitivity to ethanol following dTSPO inactivation by RU486 induction correlated with male fly heads having about four times the level of dTSPO mRNA as female heads. Moreover, neuronal knockdown of dTSPO reduced male head TSPO mRNA level but had no effect on female head TSPO mRNA level (Fig 2A). Therefore, the lack of sensitivity of female flies to neuronal inactivation of dTSPO is likely do to a gender-specific lack of TSPO in female fly brains.
To determine what might be the physiological basis of the neuron-specific effects of dTSPO deficiency on male ethanol sedation, we examined the effects on ROS production, which we previously found was increased in dTSPO deficient mitochondria [20]. ROS has been identified as modulator of neuronal activity [26]. Using Amplex Ultrared to determine the amount of H2O2 in fly heads, we found that H2O2 levels were higher in male elav-GS/+; TSPO-IR/+ flies treated with RU486 than untreated flies (Fig 4A). Hence, neuronal dTSPO knockdown increased fly head H2O2 production. When these flies were fed with N-Acetyl-L-Cysteine (NAC), an efficient antioxidant, the enhanced sedation effect of the dTSPO knockdown flies to 44% ethanol vapor was negated (Fig 4B). In tspo +/+ male flies, exposure of 44% ethanol vapor for 20 minutes resulted in sedation of most of the flies but did not significantly alter H2O2 content in fly heads (Fig 4C). Hence, the increase in ROS is not caused by ethanol exposure. Rather, dTSPO inactivation in neurons up-regulates ROS and the increased ROS is responsible for the enhanced ethanol sensitivity of the dTSPO-depleted flies.
Since ethanol sedation sensitivity was controlled by TSPO and TSPO expression declines in tspo +/+ flies to a minimum at 30 dae (S5B Fig of [20]), we determined whether ethanol sensitivity changes during aging. Male tspo +/+ and tspo-/- flies were tested with 44% and 54% ethanol sedation at different ages i.e. young (about 5 dae), mid-age (about 20 dae), and old (about 35 dae). Wild type (tspo+/+) flies exhibited increased sedation as they aged, with the effect already evident by 20 dae. tspo-/- flies also displayed and increased predilection to sedation with age (S6C Fig), but they were initially significantly more sensitive to ethanol. This is consistent with their higher level of oxidative stress as demonstrated by the marked reduction in their ROS-sensitive mitochondrial aconitase activity (Fig 6 of [20]).
Depletion of dTSPO in flies suppresses caspase activation and impedes apoptosis [20]. However, caspase also has cell death-independent functions which might be involved in neuronal control [27]. The activity of caspase 3/7, the most downstream caspase in the intrinsic apoptosis pathway, was decreased in heads of flies with dTSPO-depleted neurons (Fig 5A). Neuronal expression of the caspase inhibitor protein, p35, also reduced caspase 3/7 activity to a similar degree as induction of the TSPO dsRNA (Fig 5A). The level of caspase reduction in TSPO knockdown and p35 induced neurons is likely to be much greater than shown in Fig 5A where whole brain homogenates were assayed. Whole brain homogenates mix the enzymes of all cell types most of which are not neurons and thus not subject to dTSPO knockdown. Supporting this speculation, caspase 3/7 activity of whole body homogenate tspo-/- flies was tenfold lower than that of tspo +/+ flies (Fig 5, legend). Neuronal expression of the caspase inhibitor protein, p35, also increased sensitivity of male flies to ethanol sedation when exposed to 34% ethanol vapor (Fig 5B and 5C). This phenocopyied the dTSPO knockdown flies and confirmed the importance of reduced neuronal caspase 3/7 in ethanol sensitivity. Hence, both increased ROS production and decreased caspase activity in neurons are important in enhanced ethanol sensitivity.
To investigate the development of ethanol tolerance (reduced ethanol sensitivity following repeated ethanol exposures), we exposed flies to ethanol, allowed them to recover for 6 hours, and then exposed the flies to the same ethanol concentration again and monitored their sedation. In tspo +/+ male flies, the sedation for second exposure to 54% ethanol solution vapor was significantly delayed compared with first exposure (Fig 6A), indicating tolerance formation. However, tspo-/- male flies exhibited no diminished sedation sensitivity between the first and second ethanol exposure (Fig 6A). Hence, the systemic inactivation of dTSPO prevented male flies from developing tolerance. In contrast to male flies, female tspo-/- flies developed ethanol tolerance similar to tspo +/+ flies (S7 Fig). Hence, loss of ethanol tolerance in tspo-/- flies is also gender-specific.
To determine if the effect of dTSPO on tolerance is attributable to neurons, we compared elav-GS/+; TSPO-IR/+ flies with or without RU486 to induce TSPO dsRNA. Knockdown of dTSPO in adult male neurons had no effect on the development of tolerance following a second ethanol exposure to 54% ethanol vapor (Fig 6B). Hence the suppression of tolerance in dTSPO-depleting flies was not driven by neuronal dTSPO levels.
Since there might be other cell types in which dTSPO functions in tolerance formation, we isolated the heads and bodies of male tspo +/+ flies to examine the expression of dTSPO during tolerance. Within 4 hours after first exposure to 54% ethanol vapor, the amount of dTSPO mRNA in heads was decreased while the dTSPO mRNA in bodies was markedly increased. Both head and body dTSPO mRNA levels began to normalize at 6 hours post exposure (Fig 6C and 6D). Therefore, tolerance is associated with the induction of dTSPO in fly bodies, which is consistent with the loss of the capacity to develop tolerance in tspo-/- flies but not in elav-GS/+; TSPO-IR/+ induced flies.
We have found that TSPO is a mitochondrial modulator of ethanol sensitivity and tolerance in Drosophila. Inactivation of dTSPO in either the whole body or in adult fly neurons conferred increased sensitivity of males but not females to ethanol exposure. The increased ethanol sensitivity associated with dTSPO deficiency is a product of increased ROS production and decreased caspase activity in neurons. However, inhibition of the development of ethanol tolerance was related to systemic but not neuronal TSPO levels and this correlated with the induction of TSPO mRNA in fly bodies on ethanol exposure. Hence, our results show that TSPO is an essential mediator of alcohol sensitivity and tolerance, though not involving all the same tissue types.
The involvement of TSPO-mediated increased neuronal ROS production and decreased caspase activity in the sensitivity to ethanol sedation is consistent with reports that oxidative stress and caspase-mediated apoptosis contribute to brain pathology [28]. Since TSPO controls mitochondrial ROS production and caspase activation [20], it follows that modulation of ROS levels and caspase activity could mediate ethanol sensitivity.
Inactivation of tspo increased ROS production and NAC negated the enhanced sensitivity to ethanol demonstrating that increased neuronal ROS is related to increased ethanol sedation sensitivity. Given the short exposure period of the flies to ethanol, the ROS effect is most likely due to its second messenger action [26] rather than due to a cell death mechanism. This is consistent with the recent report showing that expression of oxidative stress genes can be altered by ethanol exposure and their functions are essential for ethanol sensitivity [29,30]. It is possible that TSPO-deficiency induced ROS production could also participate in development of tolerance, but this effect must be mediated by cells other than neurons.
Inactivation of dTSPO also inhibits caspase activity[20] and inhibition of neuronal caspase activity also sensitized flies to ethanol sedation. This was confirmed by expression of the caspase inhibitor p35 resulting in increased ethanol sensitivity. Since caspase has been shown to function in neuronal apoptosis-independent pathways to control neuronal activity in both developmental and adult stages[27], it is reasonable to conclude that dTSPO depletion in fly neurons activates such pathways thus altering neuronal activity and ethanol response.
The male-specific effects of TSPO inactivation were particularly striking. Previous studies have shown that male flies are more resistant to ethanol-induced sedation than females [31], which we also observed. Inactivation of dTSPO in males increased their sensitivity to ethanol, bringing their sensitivity close to that of females. Furthermore, female flies were found to have much lower dTSPO mRNA in their heads than males and knockdown of neuronal dTSPO in male heads reduced dTSPO mRNA about 20% while having no effect on the dTSPO levels of female fly heads. Thus, female flies have inherently low expression of dTSPO in their neurons and this may account for to their increased sensitivity to ethanol sedation.
In humans, men and women also exhibit different responses to acute and long-term ethanol exposure [32,33]. Men are at higher risk of AUD than women, but once AUD develops, women are more susceptible to ethanol-induced damage in multiple organs. Perhaps differences in TSPO expression contribute to human gender differences as well.
The molecular basis for the differences in dTSPO expression in flies is unknown. Male flies express a male specific splicing isoform of neuronal sex determination gene fruitless (fru), FruM. This may control the gender-specific production of neurotransmitters and neuropeptides [31]. Such a system might also regulate dTSPO expression. Additional environmental factors to which male and female animals are differentially exposed may also affect dTSPO expression.
A variety of genes have been reported to control fly brain development and impact ethanol responses [5]. Since the tspo mutation affects all developmental stages in fly, it’s deficiency could create a developmental abnormality that alter ethanol sensitivity. However, Hematoxylin-Eosin histological staining of tspo-/- brains did not reveal any gross anatomical defect compared with tspo+/+ brains. Furthermore, by using the RU486-inducible gene switch system to knockdown dTSPO only after eclosion, we avoided any alterations in fly anatomy demonstrating that only physiological changes were important in ethanol sensitivity of adults. Hence, the ethanol sensitivity induced in male flies by the knockdown of dTSPO cannot be due to developmental alterations, but must be the product of the physiology of the adult neurons. This means that physiological modulation should be able to treat alcoholism.
The knockdown of dTSPO in neurons demonstrates that neuronal expression of TSPO is important in determining ethanol sensitivity. This neuronal action of TSPO is at variance with reports in mammals that TSPO probes co-localize primarily with glial [34,35]. That dTSPO must be expressed in neurons is not only confirmed by the current ethanol studies but also by our previous observations that systemic depletion of dTSPO protects flies from toxicity of neuronally-expressed Aβ42 [20]. Unfortunately, our current data do not indicate if the ethanol sensitivity effects of dTSPO knockdown are related to a specific group of neurons. Mammalian TSPO has been reported to function in hippocampal neurons to affect long-term potentiation and learning[36]. Also, ethanol effects have been reported for the KCNQ channel expressed in dopaminergic neurons[37] and PKA expressed in insulin-producing neurons [38].
Benzodiazepines are widely used for treatment of anxiety, insomnia, seizures and other neural disorders, and are known to enhance the effects of GABA at the GABAA receptor. However, long-term use of these drugs is controversial due to decreasing effectiveness, physical dependence, and withdrawal [39,40]. TSPO is also a target of benzodiazepines and our results suggest that benzodiazepine-derived antagonists might increase sensitivity to ethanol and decrease neurological damage [20] while benzodiazepine-derived agonists could have the opposite effects. Consequently, the TSPO may provide an important drug target for treatment of drug abuse and alcoholism [36] which could be conveniently investigated with the current system.
Our data demonstrate that the mitochondrial TSPO protein, also known as the peripheral benzodiazepine receptor, is important in determining both ethanol sensitivity and the development of ethanol tolerance. Given the existing of a broad range of benzodiazepine analogues, these compounds may provide a novel approach for treating AUDs.
Flies were raised on standard cornmeal medium in narrow (25x95mm) vials at 25°C, with 12 hours/12 hours light/dark cycles. The tspo[EY00814] strain, obtained from the Bloomington Drosophila Stock Center (Bloomington, IN, USA), has a P-element insertion in the 3' regulatory region of tspo gene. The UAS-dTSPO-RNAi stock was obtained from Vienna Drosophila RNAi Center (VDRC, Vienna, Austria) and contained a transgene which can be transcribed into a dsRNA that targets the dTSPO mRNA. Pan-neuronal gene switch Gal4 driver, elav-GeneSwitch, was also obtained from Bloomington Drosophila Stock Center. These strains were all backcrossed to w1118 (isoCJ1) background. UAS-p35 stock was kindly provided by Dr. Nancy Bonini in University of Pennsylvania. The rempA[e02928] strain was also obtained from Bloomington Drosophila Stock Center.
To induce gene switch, flies combining elav-GeneSwitch with UAS-dTSPO-RNAi (elav-GS/+;TSPO-IR/+) or UAS-p35 (elav-GS/+;UAS-p35) were raised in regular food with 50 μl 4 mg/ml ethanol solution of mifepristone (RU486, Sigma-Aldrich, St Louis, MO, USA) added on the surface of food in vials for 3 days. As control, the flies were raised in regular food with 50 μl ethanol. The food vials were changed every 24 hours. In N-Acetyl-L-Cystein (NAC) experiments, 20 μl 500 mM NAC (Sigma-Aldrich) water solution or pure water was pre-mixed with 50 μl RU486 solution or ethanol solvent and then added on the surface of food in vials. In NAC experiments, the drug feeding was extended to 5 days.
Flies at 4–7 dae age were used for all sedation, recovery and tolerance assays, except for the NAC experiment where 6–9 dae flies were used. In the aging experiments, 19–22 or 34–37 dae flies were studied. Flies were sorted under CO2 and loaded into empty narrow (25×95mm) vials. Ten flies were loaded into a vial as a single trial and allowed to recover for at least 2 hours before use. Ethanol solutions of 34%, 44%, 54% (weight/vol) were made by mixing absolute ethanol (Sigma-Aldrich, catalog number E7023, for molecular biology) and ultrapure distilled water (Gibco, Grand Island, NY, USA) at the ratio (vol/vol) of 4:6, 5:5, and 6:4, respectively. For each vial, regular cotton clog was replaced with clog added with 1 ml ethanol solution at the vial-side surface. Recording to number of sedated flies started immediately. The interval for recording was 2 or 5 minutes.
To monitor the recovery, the ethanol-containing clog was replaced with regular clog immediately after all flies were sedated. The number of flies remaining sedated was counted every 2 or 5 minutes. For tolerance assays, the flies were transferred into regular food vial with regular clog after all flies were sedated. Four hours later, the flies were transferred back into empty vial and the recording for sedation was performed as same as in naïve flies.
Internal ethanol content was measured with Abcam Ethanol Assay Kit (Ab65343, Abcam, Cambridge, MA, USA). In brief, twenty flies at 4–7 dae age were CO2 anesthetized and loaded into empty narrow vial. After 2 hours recovery, flies were exposed to ethanol vapor from cotton clog soaked with 44% ethanol solution for 6 minutes when >90% flies were inactive. Then flies were quickly frozen in liquid nitrogen and homogenized in lysis buffer provided by kit and then centrifuged for 14000 g for 10 min in 4°C. The diluted sample together with standard ethanol samples were incubated in 96-well plate wells with ethanol oxidation reaction mix to produce H2O2 which further reacts with the probe in the mix to generate color. The absorbance at 570 nm was measured with a plate reader (SpectraMax Paradigm, Molecular Devices, Sunnyvale, CA, USA). The content of ethanol was calculated based on the standard curve, and finally normalized by the total protein concentration measured by the Bradford method.
Total RNA was extracted from bodies of 20–40 flies or 100 fly heads using RNeasy Mini Kit (Qiagen, Valencia, CA, USA). The RNA was converted to cDNA using oligo(dT)15 (Invitrogen, Grand Island, NY, USA) and SuperScript II reverse transcriptase (Invitrogen). After reverse transcription, PCR reactions were performed using a ViiA7 Real-Time PCR System (Applied Biosystems, Grand Island, NY, USA) with SYBR Green Master Mix (Applied Biosystems) and primers for rp49 (forward, 5- gctaagctgtcgcacaaatg -3, and reverse, 5- ccaggaacttcttgaatccg -3) or dTSPO (forward, 5- ctcttcgtaccctacgtcgc -3, and reverse, 5- ctggttcgataggtcggaaa -3). The PCR protocol involved denaturation at 95°C for 15 seconds and combined annealing and extension at 60°C for 1 min over 40 cycles. The melting curve was generated after these cycles to ensure that the amplification in each reaction was specific.
Isolated fly heads or whole bodies were homogenized in Homogenization Buffer (225 mM mannitol, 75 mM sucrose, 10 mM MOPS, 1 mM EGTA, pH 7.2) on ice, then centrifuged at 300 g for 5 min. The supernatant was collected and added in 96-well plate wells together with an equal volume of reaction buffer (ApoONE kit, Promega, Fitchburg, WI, USA). The plate was shaken gently for 5 min, and then incubated in dark for 15 hours in room temperature. Fluorescence was measured with a plate reader (SpectraMax Paradigm, Molecular Devices) with the excitation at 499 nm and emission at 521 nm. The fluorescent values were normalized by total protein concentration measured by the Bradford method, and the relative activity was calculated based on the ratio of normalized fluorescent signals between samples.
Isolated fly heads were homogenized in Homogenization Buffer on ice. The samples were then centrifuged at 14000 g for 10 min in 4°C to collect the supernatant. The standard reaction solution containing 0.1 mM Amplex UltraRed, Invitrogen and 0.2 U/L horseradish peroxidase (Thermo Scientific, Pittsburgh, PA, USA) diluted in Homogenization Buffer was placed in 96-well plate wells. Then the fly extract samples or standard H2O2 samples were added to the plates and incubated for 15 min in the dark at room temperature. The fluorescence was measured with a plate reader (SpectraMax Paradigm, Molecular Devices) with the excitation at 530 nm and emission at 590 nm. The H2O2 content was calculated based on standard and normalized to the total protein concentration measured by the Bradford method.
Fly heads were fixed in standard Bouin's Fixative, embed in paraffin blocks, and sectioned at a thickness of 6 μm. Sections were placed on slides, stained with haematoxylin and eosin (Vector), and examined by bright-field microscopy.
|
10.1371/journal.pntd.0007384 | Knowledge, attitudes and practices (KAP) towards rabies and free roaming dogs (FRD) in Panchkula district of north India: A cross-sectional study of urban residents | Canine rabies is endemic in urban India. A questionnaire was administered to 204 residents of the urbanised municipality of Panchkula in north India to assess the influence of gender, age, family size, social status and dog ownership, over the knowledge, attitudes and practices (KAP) towards rabies control and free-roaming dogs (FRD) in their locality. Bivariate analyses revealed significant knowledge gaps regarding crucial information on the control and transmission of rabies. Multivariable logistic regression models found that the respondents with a high/middle socio-economic status were likely to be more knowledgeable than those from low socio-economic levels (OR 3.03, 95%CI 1.5–6.0, p = 0.001). Households with children ≤14 years of age were likely to be lacking in knowledge about rabies compared to households with older or no children (OR 0.5, 95%CI 0.3–0.9, p = 0.04). The attitudes and practices of the respondents towards rabies control was positive in households with a high/middle socio-economic status (OR 3.4, 95%CI 1.7–7.2, p = 0.0008) but poor in older (≥ 35 years) participants (OR 0.4, 95%CI 0.2–0.7, p = 0.001). It is concluded that rabies awareness campaigns should be developed and conducted to target sectors of the urban community such as those belonging to lower socio-economic sections and schools to improve the residents’ knowledge and practices towards rabies. Educating dog owners about sterilising their pets is also recommended to alter the attitudes of the residents towards FRD population control.
| An enhanced level of awareness regarding rabies and management of dog-bite wounds is usually expected of urban residents owing to improved health care facilities and superior avenues for information dissemination. This perception can be misleading as gaps were found in the knowledge of residents of the Municipal Corporation Panchkula in north India regarding the disease, especially its modes of transmission and management of dog-bites. Also, there were gaps in the knowledge of the residents regarding the contribution free-roaming dogs make to keeping rabies endemic in the city. Such shortcomings in the knowledge, attitudes and practices towards rabies were prominent in older residents, those from low socio-economic status, and those families who had children ≤ 14 years of age. Dog owners were not aware of the importance of dog population management and responsible dog-ownership. It is recommended that the Municipal Corporation of Panchkula should encourage educational institutions to include material about rabies in the curriculum, and increase the frequency of awareness campaigns to alter the perception of the wider community, including dog owners, towards rabies, its control and dog population management.
| An estimated 59,000 human deaths occur annually due to rabies in the world, and 99% of this global mortality is attributed to the transmission of the virus through dog-bites [1]. The disease is endemic in Asia with India reporting the highest number of human deaths within the region, primarily amongst people from rural areas with poor socioeconomic backgrounds [2–4]. However, the true public health impact of rabies in India is unknown due to a lack of accurate data [5]. A gross lack of awareness about the disease is one of the prime factors that leads to under-reporting of human mortality due to rabies [6].
As most rabies prevention centres in India are located in urban areas [7], one would expect lower exposure and higher treatment seeking behaviour against rabies in the urban compared to the rural population. However, rapid urbanisation and inadequate garbage management systems within urban environments facilitates the emergence of rabies through the indiscriminate breeding of free-roaming dogs (FRD) on the city streets [8–10]. These FRD in the urban localities might range from semi-owned dogs that maintain some level of human interaction through the supply of food/shelter, to being completely unrestricted feral dogs that solely depend on scavenging for their existence. The dog population in India grew the fastest in the world during 2007–2012 [11, 12] and the human-FRD conflict in urban areas is evident through the increasing incidence of unprovoked dog-bites [13–18]. Vanak [19], stated that there were fewer cases of rabies in humans in urban India than in rural India due to better availability to post-exposure prophylaxis. However, as rabies is endemic in the FRD population of Indian cities, there remains potential risk of rabies exposure to urban residents through dog-bites [20, 21].
An efficient rabies control programme in urban areas is characterised by application of measures that would involve: mass vaccinations and control of the movements of FRD; control of their reproduction; initiating habitat control measures such as better garbage management; and remove unsupervised dogs [2, 22]. None of these are possible without the active participation of the people whose knowledge, attitudes and practices regarding rabies and FRD may be largely influenced by their religious, cultural or traditional beliefs. The availability and economics of preventive measures, such as anti-rabies vaccine (ARV), rabies immunoglobulins (RIG) and canine vaccines, also influence the uptake of control programmes.
In the wake of the increased FRD population in urban India it is important to assess the KAP of urban communities, not only towards rabies, but also towards FRD, before the disease can be effectively controlled [23]. KAP surveys can help to point out the inadequacies of the existing disease control programmes and help improve their effectiveness by managing the shortcomings. In the current study the KAP of a sample of residents of Panchkula Municipal Corporation in Haryana state, India was undertaken through a cross-sectional survey to assess: (1) the KAP of an urban community towards rabies and its control; (2) the KAP of an urban community towards FRD population management; and (3) the attitudes and practices of an urban dog owning population towards responsible ownership of dogs.
This study involved survey of residents of Panchkula Municipal Corporation administrated wards (ward 9 to 17) in the state of Haryana, India and the administrative approval of Panchkula Municipal Corporation was obtained for the study while ethical approval was obtained from the Murdoch University Human Ethics Committee (permission number: 20/2016).
A questionnaire was developed and administered to a sample of residents of Panchkula Municipal Corporation, Panchkula district, Haryana state in north India during September-October 2016. In the greater Panchkula district, 54.87% of the population is described as urban, of whom most reside in the wards under the administrative control of Panchkula Municipal Corporation [24]. The total number of households in wards approved for conducting the study was 13,627, with a population of 59,306 people (http://censusindia.gov.in/, as accessed in July 2016). The wards of the Panchkula Municipal Corporation administrated area are numbered 9 to 16 and comprise of highly organised residential, administrative and industrial habitats interspersed with unorganised slums and villages [25]. Ward 11 comprises of slums and urban villages with non-numbered houses, while the houses in the other wards are numbered. The number of households in the wards varies from 1,016 (ward 13) to 2,322 (ward 16) (Fig 1). Administrative approval for conducting the study in the eight wards was obtained from the Panchkula Municipal Corporation.
The target sample size for this study was based on the weighted average of three previous KAP surveys conducted in urban India by calculating the average number of participants with awareness of rabies from these studies [26–28]. 87.8% of the households were assumed to be aware of rabies, and for a 95% confidence and 5% error rate, a sample size of 172 was determined from the 13,627 present in Panchkula (http://epitools.ausvet.com.au, accessed 23 April 2016). We approached slightly more households (204) to cater for refusals (2) or partially completed surveys (7).
The number of households selected from each ward was in direct proportion to the total number of households in that ward and ranged from 18 to 38. The households were randomly selected from sectors of each ward using a random number generator, except for ward 11 where the houses were not numbered, and from where 20 households were selected through a rolling sampling method [29].
The head of the household was invited to participate; however if the household head was not present, the oldest adult member of the family/household (> 18 years of age) was surveyed in a face-to-face situation. Prior to the commencement of the survey, the study was explained to the participants, the confidentiality of their answers confirmed and oral consent to participate obtained.
The aim of the KAP questionnaire was to: identify gaps in the knowledge about rabies; assess the practices of the urban residents towards the disease that potentially contributes to the persistence of rabies; evaluate the attitudes of the respondents towards FRD; and assess the attitudes and practices of urban dog owners towards their pets. The questionnaire consisted of closed questions on: (a) the demographic characteristics of the household; (b) KAP regarding rabies (16 questions of which 11 pertained to knowledge and five to attitudes and practices towards rabies, respectively); (c) attitudes and practices towards FRD (seven questions); and (d) pet care practices adopted by the dog owners (15 questions). The questions were read out to the respondents in their local language (Hindi) by the interviewer and their answers were recorded in English (Appendix). The questionnaire was approved by the Murdoch University Human Ethics Committee (2016/20).
The responses on the questionnaire sheet were transferred onto an EXCEL spreadsheet (Microsoft Excel, Microsoft Corp., Redmond, WA, USA) and made compatible for subsequent analysis using the software R [30]. A matrix was developed to categorise the respondents into high, middle, and low socio-economic status on the basis of their educational qualification and occupation on a design based on www.praja.org (accessed 18 March 2016). Subsequently, the high and middle categories were merged to obtain a binomial distribution of respondents into two socio-economic divisions: low and high/middle (S1 Table). The age of the respondents and the family size of the households were each dichotomised into two groups based on the median age/family size.
A bivariate analysis of the responses of the participants to the individual questions pertaining to knowledge of rabies, attitudes and practices regarding rabies control and attitudes towards FRD was carried out using a χ2 or Fisher’s exact test. The residents were categorised as having adequate or inadequate knowledge of rabies; positive or negative attitudes and practices towards controlling rabies; and positive or negative attitudes towards FRD based on the median score to the responses to the questions pertaining to the relevant sections of the questionnaire. The associations between this outcome and the various descriptive variables were initially evaluated with the χ2 or Fisher’s exact test. All descriptive variables with a p ≤ 0.25 were then offered to multivariable logistic regression models. Reduced subset models were developed using backward elimination based on the AIC (Akaike Information Criteria) score. The final multivariable logistic regression models were evaluated using Pearson’s and Deviance residuals and goodness-of-fit was assessed by the Hosmer-Lemeshow test [31].
A total of 204 respondents completed the questionnaire and their descriptive characteristics are summarised in Table 1.
The median score for correct responses towards knowledge of rabies; attitudes and practices towards rabies control; and attitudes towards FRD were 8, 3 and 3 respectively.
The univariable analyses (χ2 test) of responses pertaining to knowledge of rabies that was asked of 195 (96%) respondents are displayed in Table 2. The variables gender (p = 0.2), socio-economic status (p = 0.0003), families with and without children ≤ 14 years of age (p = 0.01) and dog ownership (p = 0.14) were offered to the multivariable logistic regression model to assess the participant’s knowledge (Table 3). The model was shown to adequately fit the data with a Likelihood ratio (χ2) test of 17.6 (p = 0.00015) and a Hosmer–Lemeshow goodness of fit test value of 0.01 (p = 0.91).
In the bivariate analyses of responses about knowledge on rabies (S2 Table), respondents from the low socio-economic group were less likely to: have heard of rabies (OR 0.1 95%CI 0.02–0.4, p = 0.001); and know that the disease was fatal (OR 0.5, 95%CI 0.2–0.9, p = 0.02) or preventable (OR 0.2 95%CI 0.1–0.5, p = 0.001). They also were less knowledgeable about: the role dogs played in rabies transmission (OR 0.15, 95%CI 0.04–0.6, p = 0.008); the importance of administering human prophylaxis, such as PEP (OR 0.2, 95%CI 0.1–0.5, p = 0.001); and the control of rabies through vaccination of dogs (OR 0.3, 95%CI 0.1–0.5, p = 0.001). Younger respondents (18–34 years of age) were more likely to know of the possible transmission through licks/scratches from infected animals (OR 1.9, 95%CI 1.1–3.5, p = 0.02). Households with children ≤ 14 years of age were less aware of rabies transmission through animal bites or through licks or scratches from infected animals (OR 0.27, 95%CI 0.05–0.9, p = 0.04 and OR 0.57, 95%CI 0.3–1.0, p = 0.05, respectively). Dog-owners were more aware of the usefulness of vaccination of dogs to prevent rabies (OR = 3.0, 95%CI 1.4–7.1, p = 0.003) than non-dog owners.
The variables age of the respondents (p = 0.003), socio-economic status (p = 0.001), family size (p = 0.24) and households having children ≤ 14 years of age (p = 0.06) were offered to the multivariable logistic regression to assess participant’s attitudes and practices towards rabies control based on the univariable analysis (Table 4). The model was shown to be a good fit of the data (Hosmer-Lemeshow goodness of fit test = 0.04; p = 0.83) (Table 5).
The bivariate analyses of the responses regarding attitudes and practices of the respondents towards rabies that help its control are presented in S3 Table. Younger respondents (≤ 34 years) were more likely to use soap and water to wash dog-bite wounds (OR 2.7, 95%CI 1.4–5.1, p = 0.001) and inform the municipal authorities if they came across a dog displaying rabies-like signs (OR 2.7, 95%CI 2.7–5.0, p = 0.001) than older respondents (≥ 35 years). Respondents from the high/middle socio-economic class were more likely to: approach a hospital in the event of a dog bite (OR 3.8, 95%CI 1.7–8.6, p = 0.01); inform municipal authorities about sighting a dog showing rabies-like-clinical signs (OR 3.8, 95%CI 1.9–7.4, p = 0.001); and believe that restricting the population of FRD would help control rabies (OR 3.8, 95%CI 1.9–7.5, p = 0.001) than those from a low socio-economic class.
In the univariable analysis of the attitudes and practices towards FRD, none of the explanatory variables, when tested for association with the response variable, had p-values < 0.25 (Table 4), hence developing a multivariable logistic regression model was not attempted.
The bivariate analyses of responses regarding attitudes and practices of the respondents towards FRD are summarized in S3 Table. Dog-owners (OR 2.3, 95% CI 1.2–4.4, p = 0.008) and older residents (≥ 35 years) (OR 2.4, 95%CI 1.3–4.5, p = 0.006) believed that FRD were useful to society while younger respondents (≤ 34 years) (OR 2.01, 95%CI 1.1–3.7, p = 0.02), households containing children ≤ 14 years (OR 1.9, 95%CI 1.06–3.5, p = 0.03), and households with a low socio-economic level (OR 2.7, 95%CI 1.3–6.4, p = 0.01) considered that FRD were a problem to their society. Younger respondents (≤ 34 years) were less likely to take an injured dog to a veterinarian (OR 0.3, 95%CI 0.2–0.6, p = 0.002) than older respondents. More male (OR 2.2, 95%CI 1.1–4.4, p = 0.02) than female respondents considered FRD a threat to human health. In contrast, respondents from the high/middle socio-economic group (OR 0.3, 95%CI 0.1–0.9, p = 0.02) and dog owners (OR 0.5, 95%CI 0.2–0.9, p = 0.03) did not feel that FRD were a threat to human health. Details of the attitudes and practices of the urban residents towards FRD are presented in Table 6. A significant association was found between respondents who have good knowledge and those who have positive attitudes about rabies control (OR 3.2, 95% CI 1.8–5.8, p<0.001).
Seventy-four (36%) of the survey participants owned one or more dogs (total of 91 dogs owned). Most (67) owned one dog, three residents owned two, three residents owned four, and one resident owned 6 dogs. The details of the owned dogs are presented in Table 7. Respondents from the low socio-economic status were less likely to own a dog (OR 0.3, 95%CI 0.1–0.7, p = 0.004) and if they did it was less likely to be a pedigree dog (OR 0.2, 95%CI 0.07–0.5, p = 0.008).
We identified a number of factors of interest regarding the KAP of respondents in Panchkula, in particular: (a) the rabies awareness level of households from the low socio-economic level and those with children ≤14 years is significantly low; (b) the respondents in the higher age group (≥35 years) and households from the low socio-economic level have gaps in the attitudes and practices towards rabies control; and (c) dog-owning residents prefer a pedigree dog than a FRD, however, they would provide food and shelter to FRD due to compassion for them.
A high proportion (96%) of respondents had heard of rabies, an increase from that reported in urban localities in India (69 and 74%) by Ichhpujani, Chhabra [32] and Herbert, Basha [28]. More recent studies, report a higher proportion (84%), albeit lower to the present study [27, 33]. Elsewhere, international studies have reported comparable levels of awareness viz. Sri Lanka (90%); Bali, Indonesia (97%); and 94% in the Bohol Province, Philippines [34–36]. This rise has been largely attributed to wide dispersion of information via television and radio sources [36]. While these reasons apply to increased cognisance in India as well, it may also be linked to the Government of India prioritizing rabies as a disease of importance and its inclusion in the Nation’s recent five year plan [7]. In Panchkula, information on rabies is spread by the Municipal Corporation through awareness campaigns, rallies and awareness quizzes in schools (personal communication, Municipal Commissioner, Panchkula). Although these measures are intended to increase the awareness about rabies, they are also instrumental in more residents taking notice of the disease, even when their knowledge about various aspects of the disease remains incomplete.
In spite of Panchkula being one of the well planned and organised municipal towns in India [24], with a literacy rate higher than the national average (http://www.schooleducationharyana.gov.in/, accessed 30 June 2017); the respondents lacked understanding of the disease, especially regarding its mode of transmission, methods to prevent infection after dog-bites, and the possibility of disease transmission by animals other than dogs or by licks and scratches from a rabid animal (S1 Table). Such knowledge gaps not only contradicts the presumption that urban dwellers are well informed about the disease but also reflects the inconsistent reach of the awareness programmes in different sections of the population. Similar findings have also been reported in Dehradun (23.7%) and Delhi (42%) [27, 37]. The respondents in Panchkula were, however, better informed about the prophylactic rabies immunisation for dogs and PEP for humans compared to studies from elsewhere in the country [26, 27, 37]. Nonetheless, a lack of vital information on transmission of rabies virus in the residents may also be due to the prime focus of measures initiated by the Municipal authorities towards control of the FRD population rather than concerted efforts to enhance general awareness of the community about rabies as a disease (personal communication, Executive officer, Panchkula Municipal Corporation). A shift in focus is recommended to make the residents aware of the routes of transmission and also simple measures that can possibly prevent rabies, such as washing of dog-bite wounds with soap and water. Awareness of the disease and processes to adopt can be improved by increasing the visibility of information through posters, print and mass media as reported in villages near Bangalore in southern India following use of Information, Education and Communication (IEC) material to enhance knowledge about rabies [38]. Introduction of rabies information sessions in schools also helps improve knowledge and awareness about the disease as has been demonstrated in Sikkim, India [39]. As the present study reinforced the role of the socio-economic status on a participant’s knowledge (Table 2), awareness campaigns need adjusting to target disadvantaged groups [28, 40]. A smaller number of respondents (16) recalled the running of awareness campaigns on rabies, of which only two were from low socio-economic sections of the society implying that the awareness campaigns are neither far-reaching, nor targeting the low socio-economic sector.
Conversely, there are some positive outcomes from this study, such as no significant difference between the knowledge about rabies in males and females in Panchkula (S1 Table). This may be due to equal opportunity of males and females to acquire information on rabies, as opposed to the better opportunities offered to males to gather knowledge regarding rabies as reported in some studies such as in Ethiopia [41, 42]. The prospect of equality of knowledge dissemination among genders in Panchkula could be capitalised to spread information amongst women with children as it was found that families with children of vulnerable ages (≤14 years) lacked adequate understanding of rabies. As the currently employed tool for spreading awareness in Panchkula Municipality is primarily mass media, which has similar exposure opportunities to all groups in the community, irrespective of gender or age, a targeted approach to enhance knowledge, such as in educational institutions, is recommended to educate school children about the methods of rabies virus transmission [39].
Dog ownership was found to be a trait of the economically well off members of the urban society. The likelihood of a dog-owner belonging to a low socio-economic level in Panchkula was found to be low (OR 0.34, p = 0.04), and this explains why dog ownership is not an influencing factor for having a high knowledge score in the multivariable model.
Failure to wash dog-bite wounds with soap and water by a large proportion of the respondents (39%) reinforces the observation by others [43, 44] that a large section of society is unaware of a simple procedure that can help reduce the incidence of rabies substantially. It is not surprising that more than half of the respondents (113, 55.4%) favoured the use of traditional healing applications, such as chilli powder and turmeric, similar to studies reported elsewhere in urban India including Delhi (51%) [45] and Dehradun (57%) [27]. It is important that awareness campaigns should emphasise that, although turmeric may have antiseptic properties [46], it is not able to kill the rabies virus, which has higher chances of being destroyed if wounds are properly washed with soap and water.
The urban respondents favoured restricting the FRD population but lacked genuine concern for controlling rabies as demonstrated by their negative perception to questions pertaining to practices regarding the disease (S3 Table). In contrast, a KAP study in Bhutan by Dhand, Rai (29) found that most respondents who favoured FRD population control (99.7%) also reported cases of canine rabies to the authorities (98.8%) and practiced washing dog-bite wounds with soap and water (85.4%). In another recent study in Bhutan, the public demand for formulating legislation that could control the FRD population also implied a high level of awareness and a responsible attitude towards controlling rabies [47, page 99], which unfortunately was lacking in Panchkula. A positive aspect of the respondents in the present study, however, was their PEP seeking behaviour (85%), which was similar to a survey in Eluru district (85.5%) in Andhra Pradesh and far higher than the findings of participants from Delhi slums (27.6% and 26.5%), Pune (24%), and Dehradun (55%) [26, 27, 37, 45, 48]. This response is largely driven by easy accessibility to hospitals/clinics in the vicinity of Panchkula, compared to the previously mentioned areas that do not have close access to hospitals.
The socio-economically better off respondents were more likely to seek hospital treatment (OR 3.83, p<0.001), alert the municipal authorities of the presence of a rabid dog (OR 3.79, p<0.001) and support measures restricting the stray dog population (OR 2.49, p = 0.01, S3 Table). This is not unexpected, as better knowledge about the disease should translate into adoption of better practices, as we found a significant association between respondents with good knowledge about rabies and those with positive attitudes about rabies control (OR 3.21, p<0.001).
The older respondents (≥ 35 years of age) were found to have inadequate knowledge regarding the measures that can control/prevent rabies (OR = 0.37, p = 0.001). The low literacy level in the older generation and limited access to information about health and diseases via the internet may be reasons for this lack of knowledge and positive attitudes. This, once again, warrants targeting older people by formulating a structured and sustained information campaign in urban centres, if the knowledge and practices of older residents are to be improved to reduce the incidence of rabies throughout the country.
The majority of the respondents (68%) in Panchkula considered FRD a problem, which was consistent with recent findings in Bhutan [47, 70%] but was lower than that reported in Abruzzo, Italy (90%) by Slater, Di Nardo [49]. This difference is most likely because of the low socio-economic standing of participants included in the current study. However, in this study, over a quarter of the respondents felt that FRD were useful (27%) which may be due to the higher sense of security that the urban residents apparently desire, as guarding of the house premises was the most quoted utility (83.6%) by the respondents for the usefulness of FRD.
A prominent finding in this study was that 72.5% of the respondents admitted to feeding FRD, with most (80%) doing this out of compassion, probably linked with the perceived poor welfare of these dogs as only 19% of the respondents felt that FRD were of good health. We feel that factors, other than those included in this study, influence the attitudes of the residents towards FRD, as none of the predictors included in this study were significant. The spatial distribution of the FRD was concentrated in the vicinity of community shopping centres and it is likely that the distance of the households from the shopping centres could be a significant factor that may influence the residents’ attitudes. The city has seen many rabies awareness campaigns and dog population control interventions over the past decade, however, the programmes are often interrupted owing to failure of contracts or insufficient personnel or other resources (personal communication, Executive officer, Panchkula Municipal Corporation).
Most dog owners (68%) kept pedigree dogs and not surprisingly, more owned dogs were purchased (53%) than were adopted (29%) off the streets. The rise in dog ownership in India has resulted from the increased income of urban Indian residents [12]. The current study found that dog owners preferred to register their dogs with the Kennel Club India (KCI) but not their local municipality implying that registration is considered important for commercial purposes and eligibility for entering in dog shows rather than enforcing responsible dog ownership. As most dogs were pedigree and purchased, it was expected that the number of dog owners supervising their dogs when they were not restricted/confined would be high (66%). Although a majority (71%) of owned dogs were vaccinated against rabies, only 8% of them were sterilised and many dog owners (43%) could not cite any specific reasons for not having their dog sterilised. This highlights the need for spreading information on dog population control, as well as rabies, in Panchkula.
This KAP survey in Panchkula highlights that contrary to the expected belief that urban residents are better informed about rabies, significant gaps persist in their knowledge towards the disease, especially regarding the means of transmission through licks and scratches of rabid animals in residents of low socio-economic level and in families with children of vulnerable age (≤ 14 years). Inadequate practices regarding rabies prevention were found in the older urban respondents (≥35 years of age) and those from the low socio-economic status. We recommend that in addition to the holistic efforts to spread awareness about rabies, a targeted focus on the sections of the society such as slum residents, primary schools and unskilled workers of industrial sectors should be adopted by the Municipal Corporation to improve the knowledge of these community sectors. Implementing compulsory registration of pet dogs by Municipal Corporation will help in the monitoring of vaccination coverage of owned dogs. An incentive of free vaccination against rabies could also be started by the Municipal Corporation to reduce infection in dogs and hence the human community. Apart from wider participation of educational institutes, it is suggested that the frequency of awareness campaigns should be also increased to alter the perception of the wider community, including dog owners, towards rabies and its control and management of the dog population.
This KAP survey, however, had some short-comings. We suspect that some of the responses may not reflect the true picture, as people sometimes do not practice what they say, e.g. we feel a higher percentage of residents feed a FRD but did not acknowledge this due to the prevailing feeling by the community of their nuisance in this locality. Many dog-owners from high/middle socio-economic sections usually asked the household servants to answer the questionnaire on the pretext that in most cases it is the servant who takes care of their dog. This potentially introduced a bias as the servants do not necessarily represent the awareness and knowledge level of the dog owners. Also, the questionnaire did not explore the reasons for the 29% of dog owners failing to have their pets vaccinated against rabies which is an important aspect that should be included in future surveys. It is recommended that follow-up surveys are conducted including sampling more people per household to confirm the perceptions of urban residents towards FRD. Also we accept that comparisons made in this study with other surveys must be interpreted with caution as the questionnaires and statistical methods vary between studies. Enlarging the sample size and repeating the surveys in other urban areas may overcome such limitations for future KAP surveys.
|
10.1371/journal.pgen.1002258 | Elongator Complex Influences Telomeric Gene Silencing and DNA Damage Response by Its Role in Wobble Uridine tRNA Modification | Elongator complex is required for formation of the side chains at position 5 of modified nucleosides 5-carbamoylmethyluridine (ncm5U34), 5-methoxycarbonylmethyluridine (mcm5U34), and 5-methoxycarbonylmethyl-2-thiouridine (mcm5s2U34) at wobble position in tRNA. These modified nucleosides are important for efficient decoding during translation. In a recent publication, Elongator complex was implicated to participate in telomeric gene silencing and DNA damage response by interacting with proliferating cell nuclear antigen (PCNA). Here we show that elevated levels of tRNALyss2UUU, tRNAGlns2UUG, and tRNAGlus2UUC, which in a wild-type background contain the mcm5s2U nucleoside at position 34, suppress the defects in telomeric gene silencing and DNA damage response observed in the Elongator mutants. We also found that the reported differences in telomeric gene silencing and DNA damage response of various elp3 alleles correlated with the levels of modified nucleosides at U34. Defects in telomeric gene silencing and DNA damage response are also observed in strains with the tuc2Δ mutation, which abolish the formation of the 2-thio group of the mcm5s2U nucleoside in tRNALysmcm5s2UUU, tRNAGlnmcm5s2UUG, and tRNAGlumcm5s2UUC. These observations show that Elongator complex does not directly participate in telomeric gene silencing and DNA damage response, but rather that modified nucleosides at U34 are important for efficient expression of gene products involved in these processes. Consistent with this notion, we found that expression of Sir4, a silent information regulator required for assembly of silent chromatin at telomeres, was decreased in the elp3Δ mutants.
| Elongator is a conserved protein complex in eukaryotes. Studies in yeast, worms, and plants have revealed that Elongator complex is required for formation of mcm5 and ncm5 side chains at wobble uridines in a subset of tRNA species. The primary function of Elongator complex in yeast is to modify U34 in tRNAs. Lack of these tRNA modifications causes pleiotropic phenotypes in yeast Elongator mutants due to inefficient translation. In this report, we demonstrate that the defects in telomeric silencing and DNA damage response observed in yeast Elongator mutants are a consequence of a tRNA modification defect. We suggest that the requirement of Elongator complex in tRNA modification is conserved in all eukaryotes, and diseases linked to human Elongator mutations may involve impaired translation due to lack of tRNA modifications.
| Elongator complex, first identified in Saccharomyces cerevisiae, consists of a core complex, Elp1–Elp3 and a sub-complex, Elp4–Elp6 [1]–[3]. Orthologs of Elp1 to Elp4 has been identified in higher eukaryotes and a six-subunit Elongator complex has been purified from humans [4]–[5]. In yeast, Elongator mutants display pleiotropic phenotypes in multiple cellular processes including RNA polymerase II transcription and exocytosis [1]–[3], [6]–[9]. A crucial observation in understanding the role of the yeast Elongator complex was the discovery of its requirement for formation of 5-carbamoylmethyl (ncm5) and 5-methoxycarbonylmethyl (mcm5) side chains of wobble uridines [10]. In yeast Elongator mutants, the formation of ncm5 and mcm5 side chains were abolished in the 11 tRNA species that normally contain one of these two side chains [10]–[12]. Elongator complex in C. elegans and A. thaliana is also required for formation of ncm5 and mcm5 side chains at wobble uridines [13]–[14]. When the ncm5 and mcm5 side chains were eliminated, the corresponding tRNA species acted less efficiently in translation [12]. Although lack of modifications at position 5 affects the decoding properties of many tRNAs, it appears that the pleiotropic phenotypes of Elongator mutants are predominantly due to decreased translational decoding by hypomodified and [15]. Simultaneous over-expression of hypomodified and , which both have the mcm5s2U modification at wobble position U34 in wild type strains, compensated all phenotypes observed in Elongator mutants including those in RNA polymerase II transcription and exocytosis without restoring formation of ncm5 and mcm5 side chains in tRNA [15]. These observations not only argue against a direct involvement of Elongator complex in other cellular processes than tRNA modification, but they also suggest that the mcm5 side chain is important for efficient translation of mRNAs encoding gene products critical for the processes in which Elongator mutants generate phenotypes.
In eukaryotes, the whole genome is packed into a nucleoprotein complex known as chromatin through which the genetic material is processed to regulate cellular processes including transcription, cell division, DNA replication and DNA repair [16]–[17]. Chromatin properties can be altered by the posttranscriptional modifications of histones including acetylation, methylation, phosphorylation and ubiquitination [16]. The Elp3 protein of Elongator complex contains a tentative histone acetyltransferase (HAT) domain in the C-terminal region and the histone acetylation levels are decreased in elp3 mutants [7]. However, the reduced histone acetylation levels in the elp3 mutant were restored by increased expression of and , indicating that the involvement of Elongator complex in chromatin remodeling is indirect [15]. In addition to the HAT domain, Elp3 contains an N-terminal region with sequence similarity to the radical S-adenosylmethionine (SAM) enzymes [18]. A recent report showed that Elongator mutants have a partial loss of telomeric gene silencing and are sensitive to DNA damage agents [19]. It was also observed that strains with different point mutations in the ELP3 gene, resulting in amino acid substitutions in the radical SAM and HAT domains, displayed differences in telomeric gene silencing and DNA damage response [19]. The participation of Elongator complex in telomeric gene silencing and DNA damage response was linked to its interaction with proliferating cell nuclear antigen (PCNA), a protein involved in DNA replication and DNA repair [19].
In this report, we demonstrate that defects observed in DNA damage response and telomeric gene silencing of yeast Elongator mutants are caused by the absence of wobble uridine tRNA modifications. So far, all phenotypes observed in yeast Elongator mutants can be explained by their influence on tRNA modification. We conclude that the primary role of Elongator complex in yeast is in formation of ncm5 and mcm5 side chains at U34 of tRNAs.
In a recent report, Elongator mutants were shown to have decreased telomeric gene silencing, which was investigated by using an ura3-1 strain with a wild-type copy of the URA3 gene inserted near the left telomere of chromosome VII [19]. Cells with increased expression of Ura3 show reduced growth on plates containing 5-fluoroorotic acid (5-FOA) since the nontoxic 5-FOA is converted to the toxic 5-flurouracil by the URA3 gene product. In such a strain, 30–50% of the cell population are resistant to 5-FOA [20]. The URA3 gene was expressed in a population of cells in both wild type and elp3Δ strains (Figure 1A). However, the elp3Δ strain grew poorly on the 5-FOA containing plates compared to the wild type (Figure 1A), suggesting that telomeric gene silencing was decreased in the elp3Δ strain. Since we earlier showed that the primary function of Elongator complex is in formation of wobble uridine tRNA modifications, we investigated whether increased levels of hypomodified , and could suppress the defects in telomeric gene silencing of an elp3Δ strain. Over-expression of these tRNA species significantly improved the growth of the elp3Δ strain on 5-FOA plates (Figure 1B). The telomeric gene silencing defect of Elongator mutants was also investigated by using a color assay with the ADE2 marker inserted near the telomeric region. The elp3 mutant forms white color colonies due to loss of silencing of ADE2, which could be rescued by increased expression of , and (data not shown). This observation confirmed that Elongator mutants have a defect in telomeric gene silencing, which is caused by a translational dysfunction. The decreased telomeric silencing observed in other Elongator deletion mutants (elp1Δ, elp2Δ, elp4Δ, elp5Δ and elp6Δ) was also suppressed by elevated levels of , and (Figure 1C). Elongator mutants are also sensitive to DNA damaging agents, especially hydroxyurea (HU) [19] (Figure 2). Similar to the defect in telomeric gene silencing, the HU sensitivity of Elongator mutants was suppressed by elevated levels of , and (Figure 2). Collectively, these observations indicate that the reduced gene silencing in telomeric regions and the defect in DNA damage response of Elongator mutants is caused by inefficient translation due to lack of wobble uridine tRNA modifications.
To investigate which of the , and species most efficiently suppressed the defects in telomeric silencing and DNA damage response of the elp3Δ strain, we introduced plasmids encoding these tRNAs independently or in various combinations into the mutant. Increased expression of alone could efficiently suppress the telomeric silencing defect and the HU-sensitivity of an elp3Δ strain (Figure S1). Simultaneous over-expression of , and gave a minor improvement in suppression of the telomeric gene silencing defect compared to over-expression of alone (Figure S1A). In the HU sensitivity assay, increased expression of together with improved the suppression compared to that of and was as good as elevated levels of , and (Figure S1B). These results indicate that certain open reading frames, encoding gene products critical for telomeric gene silencing and DNA damage response, might be enriched in AAA, CAA and GAA codons. Of these three codons, translation of AAA codons by seems to be most affected by lack of the mcm5 side chain.
Asf1 functions as a histone chaperone to direct the histone acetyltransferase Rtt109 in substrate selection and stimulate its acetyltransferase activity [21]–[23]. The combination of elp3Δ asf1Δ or elp3Δ rtt109Δ mutations causes synergistic phenotypes to the strains, such as a more pronounced reduction in growth and increased sensitivity to HU (Figure 3 and Figure S2), which was suggested to be caused by loss of histone acetylation in the elp3Δ strain [19]. GCN5 encodes a histone acetyltransferase that acetylate H2B and H3 [24]–[25]. Previously it was shown that the elp3Δ gcn5Δ mutations generate a synergistic growth reduction [26]. However, increased levels of hypomodified tRNAs suppressed the synergistic growth reduction caused by the elp3Δ gcn5Δ mutations, and restore the histone acetylation levels in the elp3Δ mutant but not in the gcn5Δ strain [15]. When we over-expressed , and from a high copy vector in the elp3Δ asf1Δ or elp3Δ rtt109Δ double mutants, the growth reduction and HU sensitivity of the double mutants were similar to the defects observed in an asf1Δ or rtt109Δ strain, respectively (Figure 3 and Figure S2). These observations support the earlier conclusion that Elp3 is not directly required for histone acetylation [15].
Elp3 contains two conserved domains, a radical S-adenosylmethionine (SAM) domain in the N-terminal region and a putative histone acetyltransferase (HAT) domain located in C-terminal end (Figure 4A). Most strains expressing Elp3 proteins with amino acid substitutions in these two domains showed a reduction in telomeric gene silencing and HU resistance [19] (Figure 4). The elp3-C103A and elp3-G168R mutations did not influence telomeric gene silencing and HU sensitivity (Figure 4B and 4C) [19]. The elp3-Y540A and elp3-Y541A mutations partially reduced telomeric gene silencing and increased HU sensitivity but not as much as elp3Δ (Figure 4B and 4C) [19]. The remaining strains were similar as an elp3Δ null strain in telomeric gene silencing and HU sensitivity (Figure 4B and 4C) [19]. Moreover, all strains carrying individual mutations listed in Figure 4A except for elp3-C103A were resistant to Kluyveromyces lactis killer toxin (data not shown), indicating that these mutants have a defect in formation of wobble uridines tRNA modification [11].
To examine the status of wobble uridine tRNA modification in these elp3 mutants, total tRNAs from these mutants were isolated and analyzed by HPLC. The elp3-C103A and elp3-G168R mutants, which did not have defects in telomeric silencing and DNA damage response, had 96% and 51% mcm5s2U left, respectively (Figure 5, Table 1). Mutations in the HAT domain did not completely eliminate the formation of wobble uridine modifications, both elp3-Y540A and elp3-Y541A have 2 or 6% mcm5s2U left compared to the wild type (Figure 5, Table 1). In the rest of mutants, the mcm5 side chain formation was entirely abolished (Figure 5, Table 1). We conclude that phenotypes exhibited by elp3 mutants correlate with the levels of wobble uridine tRNA modification.
Our observations suggest that phenotypes of Elongator mutants are caused by an inefficient translation due to lack of tRNA modification. If our model is correct, reduction in modification levels in elp3 mutants should result in decreased translation efficiency. To analyze whether the modification levels of different elp3 mutants listed in Table 1 influence translation efficiency, we used a dual-luciferase reporter system (Figure 6A) [27] to measure the ochre stop codon read through by a suppressor tRNA encoded by the SUP4 allele. The SUP4 allele encodes a suppressor with a G34 to U34 substitution in its anticodon. The U34 of this suppressor tRNA is modified at position 5 with a mcm side chain [10]. Presence of this modification improves the ability of the suppressor tRNA to read UAA ochre stop codons [10], [12].
In the dual-luciferase construct, the Renilla and firefly luciferase genes are separated by an UAA ochre stop codon [27]. Read through of the ochre stop codon was determined by calculating the ratio of firefly luciferase activity to Renilla luciferase activity. This ratio was compared to the value obtained from a control construct in which a CAA codon replaces the UAA stop codon (Figure 6A). Due to lack of mcm5 side chain in the SUP4 tRNA, the stop codon read through in the elp3Δ strain is reduced to 46% of wild type (t-test, p = 0.001), supporting that the mcm5 side chain is important for efficient decoding (Figure 6B). In the elp3-G168R mutant, in which the mcm5 side chain is reduced to 51%, the level of read through was significantly decreased compared to that in wild type (t-test, p = 0.008), but is higher than that observed in strains carrying the elp3-Y540A, elp3-Y541A or elp3Δ alleles (t-test, p = 0.04 and 0.03 respectively) (Figure 6B). In the elp3-Y540A and elp3-Y541A mutants, a small fraction of total tRNA was modified (2–6%) (Figure 5, Table 1), which contributed to an improvement of stop codon read through by the SUP4 suppressor tRNA compared to the elp3Δ strain (t-test, p = 0.004 and 0.006 respectively) (Figure 6B). In mutant alleles eliminating formation of the mcm5 side chain, no differences were observed in stop codon read through by the SUP4-encoded suppressor tRNA compared to the elp3 null mutant (Figure S3). These data show that reduced mcm5 modification levels correlate with decreased translational efficiency.
Our findings that the defects in telomeric silencing and DNA damage response in Elongator mutants were bypassed by elevated levels of , and indicated that the mcm5 side chain in tRNA is critical for the expression of gene products in these two processes (Figure 1 and Figure 2). In addition to the mcm side chain at position 5 of U34, these three tRNAs also contain a 2-thio group forming mcm5s2U. Since the s2 group is also important for decoding [12], [15], [28], we hypothesized that strains deficient in formation of the 2-thio group might also display defects in telomeric silencing and DNA damage response as Elongator mutants. Tuc2 in yeast is required for the formation of the 2-thio group of the mcm5s2U nucleoside [15]. In a tuc2Δ strain, the formation of s2 group is abolished. As expected, telomeric gene silencing was decreased in the tuc2Δ strain (Figure 7A). This strain was also sensitive to 50 mM HU nearly to the same extent as observed in Elongator mutants (Figure 2 and Figure 7B). The defects in telomeric gene silencing and DNA damage response were completely suppressed by increased levels of , and (Figure 7). The phenotypes of Elongator and tuc2Δ mutants demonstrates that a translational dysfunction due to lack of U34 modifications in , and causes the defects in telomeric gene silencing and DNA damage response.
Among the three tRNA species responsible for the suppression of elp3Δ induced phenotypes, increased expression of gives the best suppression of the defect in telomeric gene silencing (Figure S1). Since decodes AAA codons, elimination of the mcm5 side chain from in the elp3Δ strain could influence the decoding efficiency of AAA codons. Therefore, we searched for open reading frames highly enriched in AAA codons (unpublished results). This analysis lead to the identification of SIR4, encoding a silent information regulator in yeast. Based on this observation, we hypothesized that the telomeric gene silencing defect of the elp3Δ mutant might be caused by decreased Sir4 expression. Accordingly, the Sir4 protein levels in the elp3Δ mutant were decreased to 34% of wild type (Figure 8A). The decreased Sir4 levels were restored to 80% of wild-type by increased expression of , and , and to 74% of wild-type by elevated levels of alone (Figure 8A and data not shown). We also observed that SIR4 mRNA levels were reduced to 76% of wild-type (Figure 8B), which cannot account for the decreased Sir4 protein levels. In addition, introducing the SIR4 gene on a high copy vector significantly suppressed the telomeric gene silencing defect of the elp3Δ strain, confirming that this defect seems to be caused by decreased Sir4 expression (Figure 8C). However, we do not exclude the possibility that there might be other open reading frames enriched in AAA codons whose translation is also affected and which might weaken silencing, directly or indirectly.
Elongator complex was initially identified by its apparent association with the elongating form of RNA polymerase II, implicating a role in PolII transcription [1]. However, its requirement in transcription was controversial based on its cytoplasmic localization and failure to detect this complex on actively transcribed genes [8], [29]–[30]. We discovered that Elongator complex was required for formation of mcm5 and ncm5 side chains at wobble uridines of tRNA [10]. The participation of Elongator complex in PolII transcription and exocytosis was indirect as elevated expression of hypomodified and could suppress previously reported phenotypes of Elongator mutants without restoring tRNA modification [15]. Recently, it was reported that Elongator complex modulates telomeric gene silencing and DNA damage response by its interaction with PCNA and its requirement for histone acetylation [19]. Since the histone acetylation defect of the elp3Δ mutant could be completely suppressed by increased expression of and [15], we assumed that Elongator complex indirectly participated in telomeric gene silencing and DNA damage response.
In this report, we show that the defects in telomeric gene silencing and DNA damage response in Elongator mutants were also suppressed by increased expression of hypomodified , and (Figure 1, Figure 2, and Figure S1). Thus, all phenotypes exhibited by Elongator mutants except the tRNA modification defect are overcome by elevated tRNA levels, indicating that the major function of this complex, at least in yeast, is in the formation of mcm5 and ncm5 side chains of wobble uridines. When , and were over-expressed in Elongator mutants, the HU sensitivity phenotype, but not the defect in telomeric gene silencing, was fully suppressed (Figure 1 and Figure 2). Since Elongator mutants affect the mcm5 and ncm5 side chain formation in 11 tRNA species, it is possible that poor translation of codons decoded by any of the other 8 hypo-modified tRNA species contributes to the defect in telomeric gene silencing, but not the HU sensitivity. In addition to the mcm side chain at position 5, U34 of , and are also thiolated at position 2. If our model is correct that the phenotypes observed in Elongator mutants are a consequence of inefficient translation, strains lacking the 2-thio group in , and will have similar phenotypes as Elongator mutants. We observed that the failure to form the 2-thio group in the tuc2Δ mutant resulted in defects in telomeric gene silencing and DNA damage response (Figure 7). These defects of the tuc2Δ mutant were completely suppressed by increased expression of , and . In addition, lack of the methyl ester in mcm5 side chain at wobble uridines in a trm9Δ strain has been linked to the defect of DNA damage response [31]. Thus, both mcm5 and s2 side chains of mcm5s2U containing tRNAs are important for efficient expression of gene products required for telomeric gene silencing and DNA damage response. These observations strongly suggest that Elongator complex influence these two processes by promoting efficient translation. Since increased expression of gives the best suppression of the telomeric gene silencing defect in Elongator mutants, we assumed genes encoding products important for this process are enriched in AAA codons. One such gene is SIR4. We demonstrate that Elongator mutants influence telomeric gene silencing by impairing efficient expression of SIR4. Even though we observed a slight reduction in SIR4 mRNA levels in the elp3Δ mutant, it cannot fully explain the decrease in Sir4 protein levels, and it is unclear if this reduction is caused by reduced transcription or increased decay of the poorly translated mRNA.
Recently, it was discovered that Elongator complex in C. elegans and A. thaliana is also required for formation of mcm5 and ncm5 side chains at wobble uridines of tRNA [13]–[14], indicating that this function of Elongator complex might be conserved in eukaryotes. In multicellular organisms, Elongator complex has also been linked to multiple processes including transcription, cytoplasmic kinase signaling and development [32]–[34]. Two recent articles suggested that Elongator complex was also required for α-tubulin acetylation and played a role in neurological processes in both mice and C. elegans [35]–[36]. In early developmental stages, C. elegans Elongator mutants have a decreased α-tubulin acetylation [36]. However, in adult Elongator mutant worms, normal levels of α-tubulin acetylation were observed, suggesting that Elongator complex is not absolutely required for acetylation of α-tubulin [13], [36]. Elongator mutants in C. elegans were also resistant to the acetylcholinesterase inhibitor aldicarb, indicating a reduced efficiency of synaptic exocytosis [13], [36]. However, a mutant allele of mec-12, which is completely missing α-tubulin acetylation, was not resistant to aldicarb, suggesting that the defect in synaptic exocytosis of Elongator mutants was not caused by reduced levels of α-tubulin acetylation [13]. Furthermore, mec-17 was discovered to be the α-tubulin acetylase in in Tetrahymena cells, C. elegans, zebrafish and mammalian cells, suggesting that Elongator might indirectly influence α-tubulin acetylation by modulating the expression of α-tubulin acetylase [37]. Based on these observations, it is tempting to speculate that the primary function of Elongator complex in multicellular organism is, as in yeast, in formation of wobble uridine tRNA modifications.
The Elp3 subunit in yeast has an N-terminal radical S-adenosylmethionine (SAM) domain and a C-terminal histone acetyltransferase (HAT) domain. In Methanocaldococcus jannaschii, the radical SAM domain of mjElp3 contains an iron sulfur cluster region and a region that binds SAM [38]. Cysteine residues at positions 96, 101 and 104 are critical for the FeS cluster formation in M. jannaschii [38]. When these corresponding cysteines at position 108, 118 and 121 in the yeast Elp3 were substituted with alanines, it eliminated the activity of yeast Elongator in formation of modified nucleosides at U34. In vitro, SAM can bind to M. jannaschii Elp3, but the binding of SAM to Elp3 from S. cerevisiae has not been detected [38]–[39]. However, when the conserved SAM binding sites (G180R G181R) in the radical SAM domain were mutated in yeast ELP3, a defect in formation of modified nucleosides was observed (Figure 5, Table 1). This observation shows that the FeS cluster and the SAM binding regions of the radical SAM domain of Elp3 are critical for the tRNA modification reaction. Substitution of glycine at position 168 to arginine, another conserved site located in the SAM binding region, reduced the wobble uridine tRNA modification to 51% of wild type (Figure 5, Table 1). In telomeric gene silencing and HU sensitivity assays, the elp3-G168R mutant displays the same phenotypes as a wild type strain suggesting that a 49% reduction in the levels of modified nucleosides do not cause phenotypes in telomeric gene silencing and DNA damage response. Two mutations in the HAT domain (Y540A and Y541A) of Elp3 did not entirely eliminate the formation of modified nucleosides at U34; 2 and 6% of mcm5s2U was detected in each mutant (Table 1). The residual level of modified nucleosides significantly improves the decoding capacity of the SUP4 encoded suppressor tRNA compared to the unmodified tRNA in the elp3 null mutant (Figure 6). This observation explains why the elp3-Y540A and elp3-Y541A mutants had increased telomeric silencing and reduced HU sensitivity compared to the elp3Δ strain (Figure 4).
Among the elp3 mutants described in Table 1, the elp3-G168R mutant, having 51% of modified nucleoside left (Figure 5 and Table 1), has the same phenotype as a wild type strain with respect to phenotypes in telomeric gene silencing and DNA damage response (Figure 4). However, this strain is resistant to killer toxin (data not shown), a phenotype tightly connected to wobble uridine tRNA modification [11]. The γ subunit of killer toxin is a tRNA endonuclease which cleaves tRNA at the anticodon region [11]. The mcm5 side chain at U34 of tRNA is important for the substrate recognition by γ toxin. In the elp3-G168R mutant, a fraction of the U34 tRNAs are missing the mcm5 side chain and the mutant is resistant to γ toxin (data not shown). However, the modified tRNAs in the elp3-G168R support the efficient expression of gene products required for telomeric gene silencing and DNA damage response. Thus, strains with tRNAs partially modified at U34 show weaker or no phenotypes compared to Elongator deficient strains.
In summary, the major function of Elongator complex in yeast is in formation of wobble uridine tRNA modifications and this function is probably conserved in eukaryotes. We suggest that when new phenotypes of Elongator mutants are discovered in yeast, an important first step is to investigate whether the phenotypes can be suppressed by over-expressing , and .
All yeast strains used in this study are listed in Table S1. Yeast transformation, media, and genetic procedures have been described previously [40]. To generate elp null mutants in different strain backgrounds, chromosomal DNA from KanMX deleted elp mutants UMY2911 (elp1::KanMX4), UMY2913 (elp2::KanMX4), UMY2915 (elp3::KanMX4), UMY2917 (elp4::KanMX4), UMY2919 (elp5::KanMX6) and UMY2921 (elp6::KanMX4) served as templates. Primers were designed to amplify DNA fragments containing the KanMX cassette and 300–500 nt flanking sequences of each ELP gene. PCR products were transformed into either W303-1A or UMY2584, and the transformants were selected by using YEPD plates containing 200 µg/ml G418. The deletion mutants were verified by PCR. To introduce asf1::KanMX4 and rtt109::KanMX4 into W303 background, chromosomal DNAs from the corresponding mutants in the deletion collection (Open biosystems) were used as templates. Primers were designed to amplify the KanMX4 cassette and 500 nt flanking sequences. PCR products were transformed into diploid strain UMY3104 and transformants were selected on G418 containing plates. The asf1::KanMX4 and rtt109::KanMX4 strains were obtained by tetrad dissection after sporulation. To construct asf1::KanMX4 elp3::KanMX4 and rtt109::KanMX4 elp3::KanMX4, the elp3::KanMX4 strain was crossed with asf1::KanMX4 or rtt109::KanMX4 to generate the diploid and double mutants were obtained by tetrad dissection. To generate elp3::KanMX4 SIR4-13Myc-KanMX6 strain, the elp3::KanMX4 strain was crossed with SIR4-13Myc-KanMX6 strain. The diploid was sporulated and the elp3::KanMX4 SIR4-13Myc-KanMX6 strain was obtained by tetrad dissection.
A two-step gene replacement procedure was used to obtain strains with different mutant alleles of ELP3. Plasmids pABY1672 (elp3-C103A), pABY1673 (elp3-C108A), pABY1676 (elp3-C118A), pABY1677 (elp3-C121A), pABY1984 (elp3-G168R) and pABY1985 (elp3-G180R G181R) were digested with EcoRI and the linearized fragments were transformed into the UMY2894. Transformants were selected on SC-Ura plates and streaked on YEPD plates. Eight independent colonies on YEPD plates were picked and streaked on 5-FOA containing plates. The strains with elp3 mutant alleles except for elp3-C103A were identified by their resistance to killer toxin and confirmed by sequencing. In order to identify the elp3-C103A mutant, DNA isolated from several candidates were sequenced.
Plasmids used in this study are listed in Table S2. The pRS306-ELP3 (pABY1554) was constructed previously [10] and used as DNA template for mutagenesis. Plasmids pABY1672 (elp3-C103A), pABY1673 (elp3-C108A), pABY1676 (elp3-C118A), pABY1677 (elp3-C121A), pABY1984 (elp3-G168R) and pABY1985 (elp3-G180R G181R) were generated by using Quickchange Lightning Multi Site-Directed mutagenesis kit according to the instruction manual (Agilent Technologies). Site-specific primers were designed by Agilent online service. To move mutant alleles of ELP3 to pRS315, pRS306-elp3 derivatives were digested using restriction enzymes BamHI and XhoI, and the excised fragments were cloned into the corresponding sites of pRS315. To generate pRS424-SIR4, SIR4 gene was amplified by PCR using W303-1A genomic DNA as template with oligos AAAA GAATTC TGTGA GTACATATAT CCGCAG and AAAA CTCGAG TTG GTATTTGATG GGTTGCTC. The PCR product was digested with EcoRI and XhoI, and cloned to the corresponding sites on pRS424.
Cells were grown at 30°C in 100 ml YEPD and harvested at OD600 = 1.5∼2. The cell pellet was resuspended in 3 ml 0.9% NaCl. The cell suspension was vortexed at room temperature for 30 minutes in the presence of 8 ml water-saturated phenol and vortexed for another 15 minutes after adding 0.4 ml chloroform. Centrifugation was carried out at 12000 g for 20 minutes. The water phase was collected and re-extracted with phenol. The final water phase was collected, mixed with 2.5 volume 99.5% ethanol and kept at −20°C for at least 3 hours. Total RNA was pelleted at 12000 g for 20 minutes. The RNA pellet was dissolved in 5 ml DE52 binding buffer (0.1 M Tris.HCl pH 7.4 and 0.1 M NaCl) and loaded onto the DE52 cellulose column. The column was washed twice with 7 ml DE52 binding buffer and the tRNA was eluted with 7 ml elution buffer (0.1 M Tris.HCl pH 7.4 and 1 M NaCl). The tRNA was precipitated with 0.7 volume of isopropanol at −20°C for at least 3 hours and pelleted by centrifugation at 12000 g for 20 minutes. The pellet was washed once with 70% ethanol and dissolved in 50 µl MQ. Purified tRNA was digested with Nuclease P1 for 16 hrs at 37°C and treated with bacterial alkaline phosphatase for 2 hours at 37°C. The hydrolysate was analyzed by high pressure liquid chromatography with a Develosil C-30 reverse-phase column as described [41].
To investigate the defect in telomeric gene silencing of Elongator mutants, 10-fold dilutions of freshly cultivated yeast cells were spotted on 5-FOA containing plates and control plates. Plates were incubated at 30°C for 2 days. To analyze the DNA damage response, 10 fold dilutions of freshly cultivated yeast cells were spotted on the plates containing 50 mM HU and control plates. The results were scored after 2 days of incubation at 30°C.
The luciferase activities were measured by GloMax 20/20 luminometer (Promega) and the dual-luciferase reporter assay system (Promega). Cells were grown to 0.5 OD600 and diluted 10 fold before use. 20 µl of diluted cell culture was mixed with 100 µl passive lysis buffer, vortexed for 12 seconds and 20 µl of cell lysate was used to determine the luciferase activity. Each culture was measured 3 times and 3 independent experiments were performed.
To determine the Sir4 protein levels, cells were grown at 30°C to OD600 = 0.5 before harvest. Cells were broken in breaking buffer (40 mM Hepes pH 7.3, 50 mM NH4Ac, 10 mM MgCl2 and 1 mM DTT) containing Complete Protease Inhibitor Cocktail Tablets (Roche Applied Science) by using FastPrep-24 homogenizer (MP biomedicals). 60 µg proteins were loaded in each lane. Mouse anti-Myc antibody (9E10) with a dilution 1∶1000 was used to detect recombinant proteins. The actin levels, used as an internal control, were detected using mouse anti-Act1 antibody (Thermo Scientific) at a 1∶2000 dilution. RNA levels were determined as previously described [42].
|
10.1371/journal.ppat.0030192 | Hemolytic C-Type Lectin CEL-III from Sea Cucumber Expressed in Transgenic Mosquitoes Impairs Malaria Parasite Development | The midgut environment of anopheline mosquitoes plays an important role in the development of the malaria parasite. Using genetic manipulation of anopheline mosquitoes to change the environment in the mosquito midgut may inhibit development of the malaria parasite, thus blocking malaria transmission. Here we generate transgenic Anopheles stephensi mosquitoes that express the C-type lectin CEL-III from the sea cucumber, Cucumaria echinata, in a midgut-specific manner. CEL-III has strong and rapid hemolytic activity toward human and rat erythrocytes in the presence of serum. Importantly, CEL-III binds to ookinetes, leading to strong inhibition of ookinete formation in vitro with an IC50 of 15 nM. Thus, CEL-III exhibits not only hemolytic activity but also cytotoxicity toward ookinetes. In these transgenic mosquitoes, sporogonic development of Plasmodium berghei is severely impaired. Moderate, but significant inhibition was found against Plasmodium falciparum. To our knowledge, this is the first demonstration of stably engineered anophelines that affect the Plasmodium transmission dynamics of human malaria. Although our laboratory-based research does not have immediate applications to block natural malaria transmission, these findings have significant implications for the generation of refractory mosquitoes to all species of human Plasmodium and elucidation of mosquito–parasite interactions.
| Malaria is arguably the most important vector-borne disease worldwide, affecting 300 million people and killing 1–2 million people every year. The lack of an effective vaccine and the emergence of the parasites' resistance to many existing anti-malarial drugs have aggravated the situation. Clearly, development of novel strategies for control of the disease is urgently needed. Mosquitoes are obligatory vectors for the disease and inhibition of parasite development in the mosquito has considerable promise as a new approach in the fight against malaria. Based on recent advances in the genetic engineering of mosquitoes, the concept of generating genetically modified (GM) mosquitoes that hinder transmission by either killing or interfering with parasite development is a potential means of controlling the disease. To generate these GM mosquitoes, the authors focused on a unique lectin isolated from the sea cucumber, which has both hemolytic and cytotoxic activities, as an anti-parasite effector molecule. A transgenic mosquito expressing the lectin effectively caused erythrocyte lysis in the midgut after ingestion of an infectious blood meal and severely impaired parasite development. This laboratory-acquired finding may provide significant implications for future malaria control using GM mosquitoes refractory to the parasites.
| Malaria, transmitted by anopheline mosquitoes, is among the worst health problems in the world, killing 1–2 million people every year, mostly African children. Lack of an effective vaccine and the emergence of Plasmodium strains resistant to many existing anti-malarial drugs have aggravated this situation. Therefore, the control of vector competence has become a more important target in malaria intervention.
Recent advances in genetic engineering of anopheline mosquitoes have raised hopes for their use as new strategies for malaria control, also the provision of powerful tools for investigating mosquito-parasite interactions. We and others have characterized tissue-specific promoters that drive robust expression of transgenes in the midgut [1,2], hemocoel [3], and salivary glands [4]. The next challenge is to identify “effector” molecules to inhibit development of malaria parasites without competitive cost to the mosquito. To date, several effector molecules have been identified (e.g., single-chain antibody fragments directed against parasite ligands [5,6], the dodecapeptide SM1 [7], PLA2 [8], a cecropin-like peptide [9], and the Vida3 peptide [10]; (see reviews [11,12]). Of these, transgenic mosquitoes expressing either SM-1 or PLA2 in a midgut-specific manner were less able to support transmission of the rodent parasite P. berghei [13,14]. However, the SM1 transgenic mosquito was not resistant to the human malaria parasite P. falciparum (M. Jacobs-Lorena, unpublished observations), and the PLA2 transgenic mosquito was significantly less fit than the wild-type [15]. In those transgenic mosquitoes generated so far, no single effector molecule has exhibited a “non-sporozoite” phenotype in the salivary glands, i.e., complete Plasmodium transmission blockade. Therefore, other effector molecules and/or mechanisms are required to generate a transgenic mosquito that is both fit and refractory to all species and strains of human Plasmodium.
Transmission of malaria parasites is absolutely dependent on availability of competent mosquito vectors. Development of Plasmodium in the mosquito begins with ingestion of an infectious blood meal containing gametocytes from a vertebrate host [16]. In the mosquito midgut lumen, female and male gametocytes mature into gametes after exposure to environmental and mosquito-specific factors. These include a drop in temperature of 5 °C and exposure to xanthurenic acid [17]. A signal transduction cascade results in the release of calcium in the cytoplasm of the activated gametocyte, initiating development and its escape from the erythrocyte [18]. After fertilization, the zygote matures into a motile ookinete. Anopheline mosquitoes rapidly concentrate the contents of the blood meal 1.5- to 2-fold, resulting in highly viscous gut content. Although little is known about the influence of these changes, we postulated that changes to the midgut environment could inhibit parasite development. We chose to express the CEL-III lectin from the sea cucumber, Cucumaria echinata. CEL-III is a Ca2+-dependent (C-type) lectin, that exhibits strong hemolytic and cell-dependent activity [19] as well as cytotoxicity toward some cultured cell lines [20] by forming ion-permeable pores in target cell membranes through oligomerization after binding to carbohydrate chains on the cell surface [21,22]. Furthermore, synthetic peptides derived from the C-terminal hydrophobic region of CEL-III exhibit strong activity against Gram-positive bacteria such as Staphylococcus aureus and Bacillus subtilis [23].
Here we show that CEL-III strongly inhibits ookinete formation in vitro, and transgenic mosquitoes expressing CEL-III in the midgut significantly inhibit oocyst formation and sporozoite production, not only of P. berghei but also P. falciparum. To our knowledge, this is the first demonstration of stably engineered anophelines in which the reduction of vectorial capacity transcends Plasmodium species.
CEL-III has strong Ca2+-dependent hemolytic activity toward human and rabbit erythrocytes, but shows only weak hemagglutination of chicken and horse erythrocytes [24]. This species-specific hemolysis is due to the binding of CEL-III to specific carbohydrate receptors on the erythrocyte surface. We examined whether CEL-III has hemolytic and hemagglutination activities toward mouse and rat erythrocytes as hosts for the rodent malaria parasite P. berghei. Figure 1A and 1B shows that the hemolytic activity of CEL-III was strong toward human and rat erythrocytes at low concentrations (IC50 = 0.3 and 0.8 μg/ml, respectively) in the presence of 5% fetal bovine serum (FBS: a source of Ca2+), whereas there was no hemolytic activity toward mouse erythrocytes. Weak hemolytic activity was observed against human and rat erythrocytes even in the absence of FBS. Similarly, CEL-III exhibited strong hemagglutination activity toward human and rat erythrocytes, but not toward mouse erythrocytes (Figure 1C). Fluorescent microscopic studies also confirmed that CEL-III bound to rat erythrocytes with numerous punctuate dots distributed throughout the cells, whereas no signals were detected in mouse erythrocytes (Figure 1D). These results suggest that carbohydrate chains on the mouse and rat erythrocyte surface may differ.
It has been reported that CEL-III is cytotoxic toward some cultured cell lines as well as toward erythrocytes [20]. Therefore, we investigated the effect of CEL-III on ookinete development in vitro. At first, CEL-III was added to cultured ookinetes in the absence of Ca2+. Figure 2A shows that bound CEL-III was observed as small punctuate dots distributed throughout the ookinete (similar the binding of CEL-III to rat erythrocytes as shown in Figure 1D), whereas no signals were detected in the ookinete without CEL-III. Next, CEL-III was incubated with gametocytes in vitro and the number of ookinetes was determined 24 h later. Figure 2B shows that CEL-III (10 μg/ml) inhibited ookinete development by approximately 95%. This inhibition was dose-dependent, with an IC50 of approximately 15 nM.
To express CEL-III in the A. stephensi midgut, we made a pAgCP-CEL-III gene cassette consisting of the promoter, 5′-UTR, and signal peptide from the A. gambiae carboxypeptidase A (AgCPA) gene [1] linked to the coding sequence of the CEL-III gene that lacked signal peptide sequence and the anopheles trypsin 1 (Antryp1) putative terminator region (Figure 3A). This gene cassette was inserted into pMinos-EGFP-RfA-F to construct pMinos-EGFP-carboxypeptidaseP-CELIII-antryp1T, then transformed into the germ line of A. stephensi embryos. A total of 876 embryos were injected and 22 fertile G0 matings were obtained. From these, one mating produced transgenic offspring expressing the egfp selectable marker. A transgenic homozygous line was obtained and propagated. A single integration event was confirmed by Southern blot analysis using genomic DNA from G4 adults (data not shown). The transgenic line has been stably maintained by blood feeding on mice or rats for over 30 generations, with no difference in reproductive fitness between transgenic and non-transgenic mosquitoes (i.e., number of eggs and hatched larvae; data not shown).
Expression profiles of the CEL-III transgene were investigated by real-time (RT)-PCR (Figure 3B). CEL-III mRNA was present in the midgut cells of sugar-fed mosquitoes and was strongly induced 6 h after blood ingestion, consistent with the pattern of expression of the endogenous A. stephensi carboxypeptidase A (AsCPA) gene, which is similar expression pattern to that of the AgCPA gene [25]. To examine whether CEL-III is secreted into the midgut lumen upon blood ingestion, transgenic mosquitoes were offered a serum meal by membrane feeding. Midgut lysates of the transgenic mosquitoes before and after the serum meal contained hemolytic activity toward human erythrocytes, but not in those of non-transgenic mosquitoes, indicating CEL-III is secreted into the midgut lumen upon feeding (Figure 3C).
Immunoblot analysis detected monomeric (48 kDa) and oligomeric (>200 kDa) forms of CEL-III in the midguts of sugar-fed transgenic mosquitoes (Figure 4). The relative mobilities of these two forms were similar to those of native CEL-III. In mosquitoes offered a blood-free ATP meal as a phagostimulant, the AgCPA promoter is activated [25,26]. Under these conditions, a slightly enhanced expression of the oligomeric form was observed 24 h after the meal. Compared to the native CEL-III, we estimate 5–10 ng of CEL-III accumulated in a single midgut after the ATP meal.
We confirmed hemolysis of human erythrocytes 24 h after a blood meal in midgut sections of a transgenic mosquito. Mosquitoes were allowed to feed on a human, then, 24 h after the blood feeding, gut sections were prepared for histology and stained with hematoxylin and eosin (HE). Compared to non-transgenic midguts which were filled with intact erythrocytes (Figure 5A and 5C), erythrocytes in the midgut of transgenic mosquito were extensively hemolyzed (Figure 5B and 5D). Lymphocytes were clearly contrasted in the midgut of transgenic mosquito (Figure 5D), but not amongst the intact erythrocytes in the midgut of non-transgenic mosquito (Figure 5C). These results are consistent with the data shown in Figure 3C, where the secretion of CEL-III into the midgut lumen caused effective hemolysis.
To investigate the effect of CEL-III expression on P. berghei development, both transgenic and non-transgenic mosquitoes were allowed to feed on the same P. berghei–infected rat and the number of oocysts formed was counted. In three experiments, the infection rate (prevalence) of transgenic mosquitoes (10.5%) was markedly reduced compared to non-transgenic mosquitoes (63.6%) (transmission blockade of prevalence, TBp; 83.5%, p < 0.01). The oocyst numbers were consistently and strongly lower in transgenic mosquitoes (transmission blockade of intensity, TBi; range 90.0 to 97.9%, mean 90.5%, p < 0.01) (Table 1). For mouse experiments, TBp in Experiment 1 and TBi in both experiments were significantly reduced in transgenic mosquitoes. Overall, the two experiments combined, both TBp and TBi were significantly reduced in transgenic mosquitoes (p < 0.01) (Table 2).
The impact of CEL-III expression on the ability of mosquitoes to transmit the parasite to uninfected animals (vectorial competence) was measured (Table 3). Vectorial competence of transgenic mosquitoes (20%) was severely impaired, compared to non-transgenic mosquitoes (100%). After a blood meal, the salivary glands of engorged mosquitoes were dissected, and numbers of sporozoites were counted. The number of sporozoites in individual salivary glands of the transgenic mosquitoes was markedly lower than that of non-transgenic mosquitoes. Importantly, sporozoite prevalence in transgenic mosquitoes (10%) was significantly reduced compared to non-transgenic mosquitoes (60%) (Table 4). These data reflect the oocyst prevalence seen in the rat experiments (Table 1).
To investigate the effect of CEL-III expression on human Plasmodium development, both transgenic and non-transgenic mosquitoes were allowed to feed on mature P. falciparum gametocyte cultures by membrane feeding, followed by determination of the number of oocysts formed (Table 5). In Experiment 1, oocyst formation in transgenic mosquitoes was significantly impaired (TBi 76.6%). In Experiment 2, TBi was 57.1%, and there was no statistically significant difference between transgenic and non-transgenic mosquitoes. Most likely, the low infection prevalence (30%) and low oocyst number (0.7 ± 1.4) in Experiment 2 affected the statistical analysis. Overall, with the two experiments combined, transgenic mosquitoes significantly impaired P. falciparum oocyst numbers (TBi 69.1%, p < 0.05), although TBp was only 7.8%.
This study demonstrates a novel “proof-of-concept” showing that transgenic mosquitoes expressing C-type lectin CEL-III significantly impairs development of both P. berghei and P. falciparum. We hypothezised that an environmental change in the midgut of anopheline mosquitoes by genetic manipulation could provide a new strategy for interrupting parasite development. CEL-III is a C-type, galactose/N-acetylgalactosamine (GalNAc)-specific lectin isolated from the body fluid of the marine invertebrate C. echinata. This lectin exhibits strong and rapid hemolytic activity and cytotoxicity through pore formation in target cell membranes. CEL-III is thought to play an important role in innate defense systems of C. echinata and therefore has the potential not only to change the environment of the mosquito midgut by rapid hemolysis of a blood meal, but also to act directly as a toxin against parasites.
For CEL-III, the N-terminal region contains two carbohydrate binding domains that have homology with the B-chains of ricin and abrin [20,27,28] and binds to the carbohydrate chains on the surface of the target cell membrane by its lectin activity. The C-terminal hydrophobic region that has antibacterial activity is believed to permeabilize the lipid bilayer of target microbes and cells [23]. CEL-III may therefore exhibit direct effector function to kill parasites. Alternatively, similar to other antimicrobial peptides or lectins, CEL-III may induce cells to undergo apoptosis [29].
In the transgenic mosquitoes, CEL-III is constitutively expressed prior to blood meal ingestion and accumulates in the midgut. Expression level of CEL-III was enhanced and reached a peak at 5–10 ng per midgut after a protein-free ATP meal. This amount is sufficient to completely hemolyze human erythrocytes in 3 μl-whole blood. As A. stephensi usually imbibes in less than 2.2 μl in a single blood meal [30], this result is consistent with the observation that complete hemolysis occurred in the midgut at 24 h after a blood meal.
Within minutes of ingestion, both male and female gametocytes escape from enveloping erythrocytes, then transform into male and female gametes. The male gametes produce eight flagellate microgametes in a process termed exflagellation, fertilizing the female gametes, giving rise first to zygotes then motile ookinetes. In the rat model, CEL-III accumulation in the midgut before a blood meal is likely to hemolyze erythrocytes infected with gametocytes immediately after a blood meal. As a result, extracellular gametocytes may be killed before differentiation. Although CEL-III cannot cause hemolysis of mouse erythrocytes, oocyst formation was also significantly reduced in the mouse model suggesting that direct parasite toxicity may be the dominant impact of the peptide. Preliminary observations suggest CEL-III reduces the efficiency of fertilization (S.Y. unpublished data). Additionally CEL-III bound to cultured ookinetes correlating with a strong killing effect on the parasites (IC50 = 15 nM) at 100- to 1,000-fold lower concentrations in vitro, when compared to other reported effector molecules, such as cecropin-like peptide [9], defensin [31], Vida3 [10], SM1 [7], and PLA2 [8]. In the rat model, the higher TBi (90.5%) may be due to additional hemolysis compared to that of the mouse model (84.8%). Although the binding specificity and mechanism by which CEL-III kills parasites in mosquitoes is unknown, findings from this study suggest that CEL-III may cause lethal damage to the female gamete and ookinete by pore formation following oligomer formation.
For P. berghei, the key property, as proposed in 1968 by Curtis [32], of vectorial competence was demonstrably and severely impaired, as measured by the relative inefficiency of transgenic mosquitoes to infect naïve mice compared to wild-type. Importantly, CEL-III transgenic mosquitoes impair sporogonic development of P. falciparum. To our knowledge, this is the first demonstration of stably engineered anophelines that affect the human Plasmodium transmission dynamics of a human malaria. Compared to the P. berghei-rat model, the TBi of P. falciparum is numerically lower (69.1%). One possible explanation for the lower TBi is that membrane feeding of in vitro cultured P. falciparum gametocytes does not contain leukocytes that may remain active in the mosquito blood meal and kill or phagocytose the liberated extracellular parasites [33,34].
In malaria endemic areas, multiple infections with Plasmodium species and strains are often observed. Those effector gene products must inhibit development of all species and strains of Plasmodium in the mosquito. As CEL-III targets erythrocytes, the “vehicles” for this parasite, as well as ookinetes, this transgenic mosquito may prove to be refractory to all species and strains of Plasmodium, including P. falciparum and P. vivax. Transgenic mosquitoes must have a minimal fitness cost, as such costs would reduce the effectiveness of the genetic drive mechanisms used to introduce transgenes into field mosquito populations. To date, there have been no single or cumulative toxic effects observed from CEL-III production in mosquitoes for fecundity (eggs laid per female). Further studies are nevertheless required to address the ability of CEL-III transgenic mosquitoes to compete with their non-transgenic siblings.
While we have demonstrated it is possible to create mosquitoes with impaired vectorial competence for more than one species of malarial parasites, we recognize there are numerous other scientific and ethical problems to be overcome before such a control strategy could be implemented.
A. stephensi mosquito strain SDA 500 was maintained at Jichi Medical University and Imperial College London. Female BALB/c mice were obtained from SEASCO (Saitama, Tokyo, Japan) and used at 7 to 8 weeks of age. Female brown Norway rats were obtained from SEASCO and used at 7 to 8 weeks of age. P. berghei strain ANKA 234 was maintained by cyclical passage through Balb/c mice and A. stephensi using standard methods [35]. P. falciparum strain 3D7 was maintained in asynchronous culture as described elsewhere [36].
CEL-III was purified from C. echinata body fluid as previously described [22]. Hemolytic activity was measured in the absence or presence of 5% FBS either by visual examination of lysis of erythrocytes or by measurement of hemoglobin release from erythrocytes using absorbance at 540 nm, as previously described [24]. Hemagglutination activity of CEL-III toward human, mouse, and rat erythrocytes was measured in the presence of 10% Dextran 4 (an osmotic protectant: SERVA, Heidelberg, Germany) as previously described [19].
P. berghei–infected mouse blood was diluted in 5 vol ookinete medium (RPMI 1640, 10% FCS, 50 μg/ml hypoxanthine, 0.024 M NaHCO3, 5 μg/ml penicillin and 5 μg/ml streptomycin, final pH 8.3) in 24-well plates with different concentrations of CEL-III and control containing the same buffer. The plate was then incubated at 19 °C on a slow moving shaker for 24 h. After 24 h, the culture was then smeared and fixed with methanol. Air-dried slides were stained with Giemsa, and then the number of ookinetes was counted in a sample of 2,000 or 5,000 RBCs.
Mouse and rat erythrocytes were prepared from whole bloods by washing five times with PBS. In vitro cultured ookinetes were purified as previously described [37]. Mouse erythrocytes, rat erythrocytes, or ookinetes were incubated with 25 μg/ml of CEL-III at room temperature for 1 h in PBS, and then washed five times with PBS. Bound CEL-III was detected by fluorescence microscopy with goat FITC-labeled anti-mouse IgG (Biosource) following mouse anti-CEL-III antiserum [19].
PCR reactions were performed with Pfu DNA polymerase (Stratagene GmbH). A gene fragment encoding amino acids 11–342 of CEL-III was amplified from plasmid pGEM-CEL-III [38] by PCR using primers pCEL-III-F1 and -R1 (Table S1). The PCR product was cloned into pENTR/D-TOPO (Invitrogen) to generate pENTR-CEL-III. A 2,311-bp DNA fragment of the putative promoter region of the AgCPA gene and its signal sequence was obtained from A. gambiae genomic DNA by PCR using primers pAgCPA-F2 and -R2 (Table S1). A 392-bp DNA fragment of the putative terminator region of Antryp1 [39] was obtained from A. gambiae genomic DNA by PCR using primers pAgAntrp1-F1 and -R1 (Table S1). The AgCPA promoter and Antryp1 terminator were assembled by overlapping PCR using primers pAgCPA-F2 and pAgAntrp1-R1, then cloned into pENTR/D-TOPO (Invitrogen) to generate pENTR-carboxypeptidaseP-antryp1T. The gene fragment encoding CEL-III was excised from pENTR-CEL-III by digestion with BglII and SphI, then cloned into the BamHI/SphI sites of pENTR-carboxypeptidaseP-antryp1T to generate plasmid pENTR-carboxypeptidaseP CEL-III-antryp1T. Transformation plasmid pMinos-EGFP-carboxypeptidaseP-CELIII-antryp1T was generated by incubation of pMinos-EGFP-RfA-F [4] and pENTR-carboxypeptidaseP-CELIII-antryp1T in the presence of LR Clonase (Invitrogen) according to the manufacturer's instructions. Primer sequence information is available in Table S1.
Embryo microinjection of the transformation and helper plasmids, screening of EGFP-expressing G0-G2 larvae, and generation of a homozygous line were performed as previously described [4].
We have cloned and sequenced a gene fragment encoding a part of the AsCPA gene from the midgut mRNA of A. stephensi by RT-PCR using primers, pAgCPA-F1 and pAgCPA-R1 (Table S1), designed for the AgCPA gene. Total RNA was isolated from mosquito midguts using an RNeasy Mini column (Qiagen). Gene-specific primers for the CEL-III, AsCPA, and ribosomal protein S7 genes were pCEL3-RT-F1/pCEL3-R2, pAsCPA-F1/pAsCPA-R1, and pAgS7-F1/pAgS7-R1, respectively (Table S1). Aliquots of cDNA representing 0.2 μg total RNA were amplified by PCR using the primer sets for detection of these genes. PCR products were separated by electrophoresis on a 2% agarose gel then visualized by ethidium bromide staining. PCR products of the S7 gene were used as controls for quality of the different mRNA preparations used in the RT-PCR analysis.
Mosquitoes were offered RPMI1640 medium containing 50% FBS through a Parafilm membrane warmed to 37 °C with a glass-watered jacket. Six h after the meal, engorged midguts from 5 mosquitoes were dissected in TBS-Ca (10 mM Tris-HCl [pH 7.5], 150 mM NaCl and 10 mM CaCl2), then homogenized in a small volume of TBS-Ca buffer, and supernatants were removed by centrifugation. The supernatants were added to human erythrocytes, and then hemolytic activity was measured by visual examination of lysis of erythrocytes as described above.
Mosquitoes were offered protein-free ATP solution (1 mM ATP, 150 mM NaCl, 10 mM NaHCO3 [pH 7.0]) through a Parafilm membrane warmed to 37 °C with a glass-watered jacket. This protein-free ATP solution was used to minimize background in subsequent western blots as previously described [25]. Engorged midguts were dissected 6 or 24 h after the meal in phosphate buffered saline (PBS), then solubilized with Laemmli buffer containing 1% 2-mercaptoethanol. The equivalent of 2 guts was separated on an 8% SDS-PAGE, electroblotted to Immobilon Transfer Membrane (Millipore), then probed with mouse anti-CEL-III polyclonal antibody. Bound antibodies were subsequently detected as previously described [5]. Native CEL-III was used for quantification of CEL-III expression per gut.
Mosquitoes were allowed to feed on a healthy Japanese volunteer. 24 h after a blood meal, engorged mosquitoes were fixed with 10% buffered formalin, then embedded in paraffin wax. Each block was cut into 4-μm sections, and then stained with HE. The Japanese volunteer gave his written consent to be included in this study after detailed explanation of the research project.
Transgenic and sibling non-transgenic mosquitoes were mixed in the same container then allowed to feed on a single infected rat or mouse. Blood-fed mosquitoes were separated after 24 h, then sorted into transgenic and non-transgenic mosquitoes using a fluorescence stereomicroscope SZX7 (Olympus) with GFP filter (excitation/emission at 480 nm/515 nm). Expression of EGFP in the abdomen of transgenic mosquitoes allowed them to be distinguished from non-transgenic mosquitoes. The two species of mosquitoes were separately housed in pots at 21 °C with 5% fructose solution. On day 15, midguts were dissected, then number of oocysts per midgut was determined. Prevalence, TBp, the mean number of oocysts in the midgut (intensity), and TBi were calculated as previously described [5,40]. Data were analyzed using the Mann-Whitney U test.
Mature gametocytes of P. falciparum (3D7) were produced in vitro as previously described [41]. Membrane feeding assays were performed to test infectivity of the P. falciparum gametocytes for mosquitoes as previously described [42]. Briefly, mature gametocyte cultures (0.3 to 0.4% final gametocytemia) were fed for 30 min at 37 °C to transgenic and non-transgenic mosquitoes through a Parafilm membrane. Engorged mosquitoes were housed in pots at 26 °C and 60%–80% relative humidity. On day 10, midguts were dissected and number of oocysts per midgut was determined. The prevalence was analyzed as above.
Transgenic and non-transgenic mosquitoes were allowed to feed on the same rat, which was infected with P. berghei. Mosquitoes that blood-fed (30 transgenic and 60 non-transgenic mosquitoes) were separated after 24 h and housed in pots at 21°C with 5% fructose solution. To measure transmission, 6 mosquitoes per group were allowed to feed on individual naïve mice 21 days after ingesting the infectious blood meal. Of 6 mosquitoes, at least 3 mosquitoes were observed to feed on each mouse. Immediately after a blood meal, engorged mosquitoes (20 of 30 transgenic and all 60 non-transgenic mosquitoes) were picked up and the salivary glands were excised, placed on a microscope slide, squashed under a cover slip, and then examined by phase-contrast microscopy (× 400). Numbers of sporozoites per salivary gland (intensity) was determined using a gland index based on Collins et al. (1977) [43]: 0; 1: 1–499; 2; 500–4,999; 3: > 5,000. The infection status of each mouse was established by examining a smear of tail blood on alternate days. Mice that had no parasites by day 30 were considered to be uninfected.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank/) accession numbers for the genes discussed in this paper are AsCPA (AB353072) and CEL-III (AB109017). |
10.1371/journal.ppat.1004699 | Direct Binding of Retromer to Human Papillomavirus Type 16 Minor Capsid Protein L2 Mediates Endosome Exit during Viral Infection | Trafficking of human papillomaviruses to the Golgi apparatus during virus entry requires retromer, an endosomal coat protein complex that mediates the vesicular transport of cellular transmembrane proteins from the endosome to the Golgi apparatus or the plasma membrane. Here we show that the HPV16 L2 minor capsid protein is a retromer cargo, even though L2 is not a transmembrane protein. We show that direct binding of retromer to a conserved sequence in the carboxy-terminus of L2 is required for exit of L2 from the early endosome and delivery to the trans-Golgi network during virus entry. This binding site is different from known retromer binding motifs and can be replaced by a sorting signal from a cellular retromer cargo. Thus, HPV16 is an unconventional particulate retromer cargo, and retromer binding initiates retrograde transport of viral components from the endosome to the trans-Golgi network during virus entry. We propose that the carboxy-terminal segment of L2 protein protrudes through the endosomal membrane and is accessed by retromer in the cytoplasm.
| The human papillomaviruses are important carcinogens, but little is known about how these non-enveloped viruses traffic to the nucleus, the site of genome replication. We use imaging, biochemical, and genetic techniques to show that a multi-subunit intracellular trafficking machine known as retromer binds directly to the L2 minor capsid protein in the virus particle to initiate its transport from the endosome to other membrane-bound organelles farther inside the cell. Most notably, knock-down of retromer expression or mutation of newly identified retromer binding sites in L2 cause the accumulation of incoming HPV16 capsids in the endosome and prevent trafficking to the Golgi. These defects can be corrected by insertion of a retromer binding site from a cellular cargo. Because all previously known retromer cargoes are cellular transmembrane proteins, the virus represents a new class of retromer cargo. In addition to elucidating the mechanism of viral endosome escape, these results suggest that retromer may play a more versatile role in cell biology than previously recognized.
| The human papillomaviruses are small non-enveloped viruses responsible for approximately 5% of human cancer deaths worldwide [1]. The cellular mechanisms involved in HPV infection are poorly understood, but they represent potential sites of anti-viral intervention. HPV consists of an ~8000 base-pair double-stranded DNA viral genome packaged in the viral capsid, which is composed of 360 molecules of the major capsid protein, L1, and up to 72 molecules of the minor capsid protein, L2, which is largely buried inside the L1 shell [2–4]. Initial binding of virus to cells is mediated by the interaction between the L1 capsid protein and heparan sulfate proteoglycans [5–8]. After cell binding, L1 and L2 undergo conformational changes, which allow cleavage of the amino-terminus of L2 at the cell surface by the protease furin [9–12]. HPV is then transferred to an as-yet-unidentified cell-surface receptor and internalized [13–17]. Disassembly of the capsid is initiated by acidification of the endosomal lumen by the vacuolar ATPase, and L1, L2, and viral DNA then traffic via retrograde pathways to the Golgi apparatus and endoplasmic reticulum [5,16–24]. Cell cycle progression and nuclear envelope breakdown appear required for HPV entry into the nucleus, where viral gene expression and DNA replication occur [25,26]. During virus trafficking, the L1 protein dissociates from the viral DNA, but the 473-amino acid L2 protein is required for efficient trafficking of the viral genome to the nucleus and remains associated with the genome during nuclear entry [20,27–34].
We performed a genome-wide siRNA screen to identify host cell genes required for HPV16 infection and discovered that infection of HeLa cervical cancer cells and immortalized cervical keratinocytes requires retromer, a cytoplasmic endosomal coat complex that mediates export of cellular transmembrane proteins from the endosome to the trans-Golgi network (TGN) or plasma membrane [22,35,36]. The cargo recognition core of the retromer consists of three subunits, Vps26, Vps29, and Vps35, all of which are required for retromer-mediated endosomal sorting and for HPV infection [22]. In cells depleted of retromer, HPV components fail to arrive at the TGN [22]. In addition, retromer is present in a stable complex with viral capsid proteins in infected cells [22].
The mechanism by which retromer supports HPV trafficking is unknown. All known retromer cargos are cellular, integral membrane proteins [35,36]. Retromer recognizes sorting signals located in the cytoplasmic domain of these transmembrane protein cargos at the endosomal membrane to effect packaging of the cargo into budding vesicles or tubules that later fuse with target membranes to deliver the cargo to its destination. Unlike known retromer cargos, the non-enveloped HPV capsid is particulate and lacks transmembrane proteins. Furthermore, when the capsid is in the endosomal lumen during the early stages of infection, it is separated from retromer in the cytoplasm by the endosomal membrane. It is possible that retromer acts indirectly by mediating trafficking of a cellular transmembrane protein that is essential for some step in HPV entry. Alternatively, HPV might have developed a strategy to access retromer even when the capsid itself is in the lumen of the endosome. Finally, virus might exit from the endosome into the cytoplasm where it can be recognized by retromer, which later mediates its entry into the Golgi. The experiments reported here demonstrate that retromer directly binds to the minor capsid protein of incoming HPV and that this interaction mediates export of virus particles from the early endosome into the retrograde vesicular pathway during the early stages of intracellular trafficking of incoming HPV.
Because retromer is required for the delivery of HPV16 to the Golgi apparatus and initiates endosome-to-Golgi transport of various cellular proteins, we hypothesized that a viral protein might bind to retromer. Inspection of the amino acid sequence of the HPV16 L2 protein revealed that its carboxy-terminal segment contains two short sequences, FYL and YYML, that resemble known retromer binding motifs, e.g., aromatic followed by any amino acid followed by leucine or methionine [ФXL/M], or phenylalanine or tryptophan followed by leucine followed by a valine or methionine [Trp/Phe-Leu-Met/Val] [37,38] (Fig. 1A). To determine if these sequences are important for HPV infection, we constructed alanine scanning mutations across this segment of L2 (Fig. 1A). Because it is difficult to introduce mutations into authentic HPV, we used pseudoviruses (PsVs) comprised of L1 and L2 encapsidating a reporter plasmid, which display the entry properties of authentic virus [39,40]. PsV containing wild-type L1, wild-type or mutant L2 with a C-terminal HA or FLAG tag, and a GFP or HcRed reporter plasmid were produced in 293TT cells [24,40]. The assembly of mutant capsids was confirmed by encapsidation of the reporter plasmid and electron microscopy, which revealed no morphologic differences from wild-type PsV (S1A Fig). In addition, when normalized by encapsidated reporter virus plasmids, wild-type and mutant pseudovirus preparations displayed a similar level of purity and contained similar levels of L1 and L2 (S1B and S1C Fig). HeLa cells were infected with wild-type and mutant PsV stocks containing the same number of encapsidated reporter plasmids, corresponding to a multiplicity of infection (MOI) of approximately 0.5 for wild-type, and successful infection was measured two days later by flow cytometry for GFP fluorescence. As shown in Fig. 1B, several mutants were competent to infect cells. Strikingly, however, the FYL/AAA mutant lacking one of the putative retromer binding sites showed a >80% reduction in infectivity, and the YYML/AAAA mutant lacking the other site showed an approximately 50% reduction. When the FYL and YYML mutations were combined to generate the double mutant (HPV16.L2DM), infectivity was essentially abolished. The double mutant showed a similar defect when the HA tag on L2 was replaced with a FLAG tag. The double mutant was also defective in HaCaT cells, a human skin keratinocyte cell line commonly used in HPV entry studies (e.g., [19,20,25]) (Fig. 1C), which also require retromer for efficient infection (S2 Fig).
The defect caused by mutations in the putative retromer binding sites in L2 suggests that retromer may interact with these sequences to promote HPV infection. To explore this possibility, we replaced FYL with the sequence tryptophan-leucine-methionine (WLM) (Fig. 1A), a retromer sorting signal from the cytoplasmic tail of the cation-independent mannose-6-phosphate receptor (CIMPR), a cellular cargo of retromer [37]. As shown in Fig. 1A and 1C, insertion of WLM into the double mutant to generate HPV16.L2WLM/DM partially restored the ability of HPV16.L2DM to infect HeLa and HaCaT cells, and a mutant containing WLM and the endogenous YYML sequence (HPV16.L2WLM) infected cells as well as wild-type PsV. Taken together, these results demonstrate that the putative retromer binding motifs in the carboxy-terminus of L2 are required for efficient infection and suggest that they act by binding to retromer. Alignment of the carboxy-terminal segment of the L2 protein from multiple HPV types showed that the FYL putative retromer binding site or a closely related sequence is highly conserved in all genera of HPV (Fig. 1D), but the YYML sequence is not. This correlates with the more dramatic defect caused by removal of the FYL sequence.
To test directly if the sequence FYL can act as a retromer sorting signal, we used an antibody internalization assay to determine if FYL is able to replace the endogenous sorting signal in a cellular retromer cargo [41]. The extracellular and transmembrane domains of the cell-surface protein CD8 were fused to the 160-amino acid cytoplasmic tail of CIMPR, which contains an endocytosis motif and the WLM retromer sorting signal identified by Seaman and colleagues [37] (Fig. 2A). This signal mediates retromer-dependent trafficking of CIMPR from the endosome to the Golgi apparatus. HeLa cells were transfected with a plasmid expressing CD8-CIMPR containing a wild-type or mutant WLM sequence, and after 24 hours, live non-permeabilized cells were incubated with a CD8 antibody for three hours at 37°C. During this incubation, the CD8-CIMPR fusion protein containing the endogenous WLM sorting signal was endocytosed and trafficked to the TGN, which was scored by co-localization with the Golgi marker GM130. As previously reported [37], substitution of the WLM sequence with AAA abolished Golgi trafficking, resulting in a punctate distribution of the fusion protein throughout the cytoplasm. Strikingly, replacement of WLM with FYL restored Golgi localization (Fig. 2B). Furthermore, siRNA-mediated knockdown of the Vps35 retromer subunit (S3 Fig) eliminated Golgi localization of the WLM and FYL CD8-CIMPR fusion proteins in the antibody internalization assay and resulted in dispersed punctate distribution of the fusion proteins (Fig. 2B). These results demonstrate that FYL can act as a retromer sorting signal in a standard trafficking assay.
The experiments described above imply that retromer sorting signals in the L2 protein are involved in HPV trafficking during virus entry. To determine the role of retromer and the putative retromer sorting motifs in HPV infection, we conducted experiments in infected cells. To first rule out the possibility that the defect in infectivity caused by the L2 mutations was due to an inability of the mutant to enter cells or undergo disassembly, we infected HeLa cells with PsV containing L2 with a carboxy-terminal FLAG epitope tag (designated HPV16.L2F). The FLAG tag is constitutively exposed on the surface of capsids and does not inhibit infectivity nor affect entry requirements, as assessed by sensitivity to several genetic and pharmacologic inhibitors of entry [24]. We stained cells at early times after infection with the anti-FLAG antibody or with the 33L1–7 antibody, which recognizes an epitope on the L1 protein that is inaccessible in intact capsids and reacts with L1 only after the capsid has entered cells and disassembly has begun [20,29,42]. Neither antibody stained uninfected cells, but cells infected with PsV containing wild-type or double mutant L2 displayed similar punctate intracellular staining with both antibodies (S5 Fig), demonstrating that the inability of the L2 double mutant to infect cells is not due to a defect in internalization or initiation of capsid disassembly.
To determine if retromer knock-down impairs exit of HPV PsV from the endosome, we used the proximity ligation assay (PLA), a specific immune-based detection system in which a fluorescent signal is generated only when two proteins of interest are nominally within 40nm [43,44]. PLA was used to determine if L2 is in close proximity to the early endosome marker EEA1 or the trans-Golgi marker TGN46 during entry. At eight and 16 hours post-infection with HPV16.L2F, we incubated fixed and permeabilized cells with anti-FLAG and the antibody recognizing the cellular component and then processed the samples for PLA. As expected, PLA did not generate signals in uninfected cells. When HeLa or HaCaT cells were infected with wild-type HPV16.L2F and stained for L2-FLAG in proximity to EEA1, a PLA signal was detected at eight hours post-infection, and by 16 hours the PLA signal associated with this compartment was reduced (Figs. 3, 4, and S6). In contrast, there was little L2/TGN46 PLA signal in the Golgi at eight hours post-infection, but there was abundant signal by 16 hours. These results indicate that virus enters the early endosome by eight hours after infection, but by 16 hours it has transited through this compartment and arrived at the Golgi. HPV trafficking was strikingly different in retromer knock-down cells. In this experiment, HeLa and HaCaT cells were transfected with siRNA targeting the retromer subunit Vps29, infected with HPV16.L2F two days later, and subjected to PLA at various times post-infection. An L2/EEA1 PLA signal was observed in both control and retromer knock-down cells eight hours after infection, confirming that retromer is not required for virus endocytosis. However, at 16 hours post-infection, the L2/TGN46 PLA signal was essentially undetectable, whereas there was a striking increase in the L2/EEA1 PLA signal (Figs. 3, 4, and S6). This is in marked contrast to infected parental cells, which contained little endosomal L2 at 16 hours after infection. These experiments demonstrate that retromer knock-down in both HeLa and HaCaT cells causes the accumulation of L2 in the early endosome and prevents the arrival of L2 into the Golgi.
If the non-infectious phenotype of HPV16.L2DM is due to interference with retromer recognition, this mutant should fail to exit from the early endosome in cells with intact retromer function. To test this, we infected HeLa and HaCaT cells with FLAG-tagged HPV16.L2DM and used PLA to assess localization of incoming virus. As shown in Figs. 3, 4, and S6, an L2/EEA1 PLA signal was observed in cells infected with HPV16.L2DM at eight hours after infection. At 16 hours post-infection, an L2/TGN46 PLA signal was not observed, while the L2/EEA1 PLA signal markedly increased. Thus, mutation of the putative retromer binding motifs in the carboxy-terminal segment of L2 is functionally equivalent to knocking down retromer function. Importantly, when WLM was inserted into HPV16.L2DM at the original position of FYL (to generate HPV16.L2WLM/DM), a substantial restoration of the L2/TGN46 PLA signal was observed at 16 hours after infection in both cell types, together with a reduction in the EEA1 signal at this time, confirming that the WLM retromer motif restores exit of L2 from the early endosome and trafficking to the Golgi. Similar results were obtained with multiple independent stocks of HPV16.L2DM and HPV16.L2WLM/DM. These results strongly suggest that the trafficking defect displayed by the double L2 mutant is due to impaired retromer binding.
Recognition of cargo is a major factor underlying retromer recruitment to membranes [45]. To test whether the trafficking defect displayed by the L2 double mutant correlated with impaired association between L2 and endogenous retromer in infected cells, we performed PLA with anti-FLAG and anti-Vps35. HeLa cells were infected with FLAG-tagged HPV16 PsV containing wild-type L2, L2DM, or L2WLM/DM, and PLA was performed at eight and 16 hours post-infection. As shown in Fig. 5, a PLA signal was observed for wild-type L2 and retromer at eight hours. This signal decreased substantially by 16 hours, consistent with a transient association between L2 and retromer as HPV exits the early endosome. In contrast, at eight hours after infection with HPV16.L2DM, the PLA signal for L2 and Vps35 was decreased by 70% compared to cells infected with the wild-type PsV. The signal persisted at this level at 16 hours, despite the markedly increased amount of the mutant L2 protein in the early endosome at this time. Replacing FYL in the double mutant with WLM restored transient association of the L2 protein with retromer. Taken together, these results indicate that mutations that impair the association of L2 with retromer in infected cells also impair L2 export from the early endosome and demonstrate that the WLM sequence restores association between L2 and retromer, as well as restoring endosome exit.
To determine if the carboxy-terminal segment of L2 is sufficient to bind to retromer, we conducted in vitro pull-down experiments. First, we employed biotinylated peptides, one from the amino-terminal portion of the L2 protein, one from the middle of the L2 protein, and one from the carboxy-terminal portion, including the two putative retromer binding sites (Fig. 6A). These peptides were incubated with detergent lysates of uninfected HeLa and HaCaT cells, and cellular proteins that bound to the peptides were collected on streptavidin beads, subjected to SDS-polyacrylamide gel electrophoresis and immunoblotted for retromer subunits. The carboxy-terminal peptide containing the retromer motifs precipitated endogenous Vps29 and Vps35, whereas the other two peptides were devoid of retromer binding activity (Figs. 6B and 6C, left panels, and S7).
We also conducted pull-down experiments with carboxy-terminal peptides containing mutations in the retromer binding motifs. As shown in Fig. 6B and 6C, mutations in either retromer motif eliminated retromer binding in HeLa and HaCaT cell lysates, as did mutation of both sites. In some experiments, slight binding to the YYML/AAAA mutant was observed (S8 Fig), but binding to the FYL/AAA mutant was never detected, suggesting that the FYL mutation causes a more severe defect in retromer binding, consistent with the more dramatic defect in infection caused by the FYL mutation. Importantly, replacement of FYL with the WLM retromer motif restored a significant level of retromer binding in extracts of either cell type (Fig. 6). Taken together, these results show that the retromer sorting motifs in the C-terminus of L2 bind to endogenous retromer in vitro, and that mutations that inhibit infectivity and endosome exit interfere with retromer binding.
To determine if retromer directly recognizes L2, we tested whether the carboxy-terminus of L2 was able to bind to active human retromer assembled from individual Vps26, Vps29, and Vps35 subunits purified from E. coli and immobilized on glutathione resin. We previously showed that retromer assembled in this way bound to the cellular retromer cargo, DMT1-II [45]. A 24-amino acid wild-type or mutant segment of L2 containing the retromer binding sites was fused to poly-histidine-tagged maltose binding protein (MBP), which was also expressed and purified from E. coli (Fig. 7A). The L2 fusion protein was incubated with immobilized retromer, and the L2 fusion protein bound to retromer was eluted and detected following SDS-PAGE. As shown in Figs. 7B and 7C, retromer captured the L2 fusion protein containing the carboxy-terminal segment of the wild-type L2 protein, indicating that retromer and this segment of L2 interact directly. In contrast, the FYL and YYML alanine substitutions, alone or in combination, drastically decreased retromer binding. Thus, the carboxy-terminal segment of the L2 protein binds directly to retromer via sites required for exit from the early endosome.
Viruses utilize cellular machinery to reach the site of viral genome replication. Therefore, studies of virus entry not only reveal important features of the virus life cycle, but also elucidate the mechanisms cells use to ensure that cellular components are present in their proper intracellular locations. We previously identified retromer as a factor required for trafficking of HPV16 to the Golgi apparatus during infection [22], but our published experiments did not determine if retromer plays a direct or indirect role in HPV infection. HPV is a non-enveloped virus that lacks transmembrane proteins and is present in the endosomal lumen early during entry. Thus retromer, which is present in the cytoplasm and transports transmembrane proteins, could act indirectly on a cellular cargo to support HPV entry, or it could recognize the HPV capsid in an unconventional manner. The experiments reported here reveal that the papillomavirus capsid is a new class of retromer cargo and that a direct interaction between retromer and the L2 minor capsid protein is required for L2 to exit the endosome and traffic to the Golgi. Because L2 is closely associated with the viral genome throughout the entry process and manipulations that interfere with L2 trafficking also inhibit infectivity, we conclude that the observed behavior of L2 reflects the behavior of the viral components required for infectivity.
Several lines of evidence demonstrate that L2 is a retromer cargo. Retromer knock-down causes HPV L2 to accumulate in the early endosome in HeLa and HaCaT cells. Furthermore, the carboxy-terminal segment of the L2 protein contains short sequences that resemble known retromer binding motifs, and mutations in these motifs interfere with the ability of L2 to associate with retromer in infected cells and inhibit the export of PsV from the early endosome and its delivery to the Golgi apparatus in HeLa and HaCaT cells. Importantly, these defects are rescued by replacement of the major retromer binding site in L2 with a retromer sorting signal from a cellular protein. In addition, this L2 sequence can function in a standard retromer sorting assay. Finally, in vitro binding studies showed that these sites in the L2 protein bind directly to retromer. Taken together, these results demonstrate that a direct interaction between the carboxy terminus of the L2 protein and retromer is required for exit of L2 from the early endosome and subsequent entry into the Golgi during HPV16 entry. These findings strongly suggest that retromer sorts HPV into an endosome-derived transport vesicle that ferries HPV or a subviral structure containing L2 and viral DNA to the TGN. Elsewhere, we showed that inhibition of γ-secretase blocks HPV trafficking after export from the early endosome but before delivery to the Golgi and the ER [24], indicating that γ-secretase is required in HPV entry after retromer action.
The peptide and fusion protein pull-down experiments suggest that both the FYL and YYML motifs are required for efficient retromer binding in vitro, although YYML appears less important (S8 Fig). However, the more dramatic infectivity defect caused by the FYL mutation and the lack of YYML conservation indicates that the FYL sequence is more important in cells and during natural infection.
FYL or a closely related sequence is present in the same position in all sequenced HPV L2 proteins examined and in the great majority of animal papillomaviruses, implying that the ability of retromer to recognize the L2 protein and export it from the early endosome arose early during papillomavirus evolution and remains an important feature of the virus life cycle. Some positions flanking FYL are also highly conserved, but they are not required for HPV16 entry in the assays used here. We also note that FYL and YYML do not match the canonical CIMPR retromer binding motif, Trp/Phe-Leu-Met/Val. FYL does appear similar to the ФXL/M motif in the mammalian iron transporter, DMT1-II [38]. However, comparison of this sequence in numerous HPV types reveals important differences. Although the ФXL/M motif can accommodate tryptophan in the first position, in 45 HPV L2 proteins examined, ~70% contain phenylalanine at this position, ~30% contain tyrosine, and none contain tryptophan. At the second position, ~75% HPV contain tyrosine and only one contains leucine, while wild-type DMT1-II contains leucine. Thus, although the HPV L2 and DMT1-II motifs are related, the examination of numerous HPV types suggests that evolution has selected specific residues at these positions in the viral protein that differ from the sequence in DMT1-II. Similar non-canonical retromer binding motifs may exist in additional retromer cargos from cells or other viruses.
The L2 protein may be analogous to capsid proteins of other non-enveloped viruses that undergo conformation changes that expose hydrophobic peptides that insert into cell membranes and disrupt them to allow capsid entry into the cytoplasm [46,47]. We propose that the carboxy-terminal segment of L2 causes a more subtle perturbation of membrane structure, allowing it to protrude through the endosomal membrane where it is recognized by retromer on the cytoplasmic face of the membrane. Cells may contain other similar, as-yet-unrecognized, non-conventional retromer cargos, and other viruses may use a similar entry mechanism.
A peptide derived from the carboxy-terminal segment of L2 including YYML and an adjacent, conserved highly basic sequence exhibits membrane bilayer-disrupting activity in vitro and can mediate integration of a reporter protein into cell membranes [31]. This peptide disrupts membranes only at low pH [31]. Therefore, endosome acidification may trigger its penetration through the endosomal membrane during virus entry. Such behavior may explain at least in part the requirement for endosome acidification during HPV entry [17,18,23].
It is possible that the carboxy-terminal segment of L2 cooperates with another segment of the L2 protein exposed on the surface of the capsid to mediate membrane penetration and/or retromer recognition in intact cells. The L2 amino-terminus contains a transmembrane-like domain required for infectivity [34]. This L2 segment may stabilize the association of L2 with the endosomal membrane or retromer or facilitate the passage of the carboxy-terminal segment containing the retromer binding sites through the membrane. In addition, a central segment of L2 binds SNX17, another cytoplasmic protein required for efficient infection [30], implying that this L2 segment is also accessible to the cytoplasm in some situations. Interestingly, SNX17 is reported to cooperate with retromer in mediating recycling of the Notch ligand, Jag1a [48]. Finally, the basic sequence adjacent to YYML resembles motifs present in cell-penetrating peptides, which can carry protein cargo across membranes into cells [49]. Further analysis will determine if any of these sequences play a role in retromer action during HPV infection.
In summary, these studies elucidate the role of retromer in HPV entry, demonstrating that it binds directly to the HPV capsid and mediates export of this unconventional cargo from the early endosome on its journey into the cell. We propose that retromer recognition in this system involves a novel mechanism whereby a segment of a minor capsid protein of a non-enveloped virus protrudes through a cellular membrane into the cytoplasm. Further analysis of the role of retromer in HPV16 infection will provide new insights into virus entry, retromer function, and intracellular trafficking.
HaCaT cells are a spontaneously immortalized line of human skin keratinocytes obtained from G. Paolo Dotto (Massachusetts General Hospital) [50]. 293TT cells were obtained from Dr. Christopher Buck (NIH). HeLa-Sen2 cells (designated here HeLa cells) are a previously described cloned strain of HeLa cells that infects efficiently with SV40 and HPV16 pseudovirus [51]. HeLa-M cells, a strain of HeLa-S3 cells that transfects efficiently, were obtained from Walther Mothes (Yale University). All cells were cultured in Dulbecco’s MEM (DMEM) with 10% fetal bovine serum (FBS), 10mM L-glutamine, 10mM HEPES pH 7.2 and standard antibiotics (Pen/Strep).
HPV16-GFP pseudovirus containing an HA tag at the C-terminus of L2 (designated here HPV16.L2HA) was generated by using a plasmid obtained from Patricia Day. We also used PsV designated HPV16.L2F, which contains an L2 protein with a carboxy-terminal 3xFLAG-tag constitutively exposed on the surface of the capsid [24]. The HPV16 L2 C-terminal mutants were produced in either the HA- or the FLAG-tagged p16sheLL expression plasmid using the QuickChange site-directed mutagenesis system and the primers listed in S1 Table. The L1 and L2 genes in each mutant were sequenced in their entirety. Mutations in the retromer sorting motif in the CD8-CIMPR fusion protein were inserted into a plasmid expressing the wild-type fusion protein (a gift from Matthew Seaman, Cambridge Institute for Medical Research). siRNA targeting Vps26, Vps29, and Vps35 and the control scrambled siRNA were purchased from Dharmacon (Lafayette, CO). Sequences of all oligonucleotides used in this study are listed in S1 Table.
pCINeo-GFP plasmid (obtained from Christopher Buck (NIH) and pCAG-HcRed plasmid (purchased from Addgene, Plasmid 11152) were used as reporter plasmids. Pseudoviruses were produced by co-transfecting 293TT cells with a p16sheLL plasmid expressing L1 and wild-type or mutant L2 and a reporter plasmid, and purified by density gradient centrifugation in OptiPrep (Sigma-Aldrich, #D1556) as previously described [22,40]. Encapsidated GFP or far-red genomes were quantified by qPCR as described [52,53]. Briefly, 5 μl of each pseudovirus preparation was treated with 4 μl of RQ1 DNAase (Promega, M6101) in 100 μl DNAase buffer (50mM Tris HCl pH 7.6, 10mM MgCl2) for one hour at 37°C. The DNAase was inactivated by incubation at 75°C for 30 min, followed by the addition of 50 μg of proteinase K (PK, Roche) for one hour at 37°C, in PK buffer (10mM Tris HCl pH 8.0, 10mM EDTA, 0.25% SDS) [52]. DNA was isolated using a PCR purification kit, and the number of encapsidated genomes was determined by qPCR using primers for the GFP or far-red gene, using a 10-fold serial dilution of pCINeo-GFP or pCAG-HcRed plasmid (109 to 103 genomes/μl) analyzed on the same plate as a standard. Encapsidated genomes for all of the PsV stocks used in any one experiment were quantified in parallel. To examine the purity and content of L1 and L2 in PsVs, Optiprep-purified PsV preparations containing FLAG-tagged L2 were denatured in SDS-Laemmli sample buffer (108 packaged reporter plasmid genomes/lane) and electrophoresed on a SDS-10% polyacrylamide gel for 1.5h at 150V. Proteins were subjected to silver staining or transferred from the gel to PVDF membrane (Millipore Immobilon, 0.2 μm, ISEQ15150), which was probed with 0.5 μg/mL primary anti-L1 (BD Pharmingen, 554171) or anti-FLAG (Sigma, F3165) antibody. Following incubation with 1:10,000 dilution of horseradish peroxidase-coupled secondary antibodies, bands were visualized by luminescence (SuperSignal West Pico, Thermo Scientific, 34080).
Freshly glow-discharged 200 mesh Formvar/carbon-coated copper grids (Electron Microscopy Services, CG200-Cu) were inverted on drops of gradient-purified PsV diluted 1:9 in phosphate-buffered saline (PBS), and virus allowed to adsorb for five minutes. The grids were washed twice in deionized water and stained by two one-minute incubations on drops of Nano-W (Nanoprobes, Nephank, NY) before removing excess stain by gentle blotting with Whatman #1 filter paper. The grids were air-dried before viewing on a FEI Tecnai Biotwin transmission electron microscope at 80Kv. Images were taken using a Morada CCD camera and iTEM (Olympus) software, and ImageJ was used to provide sizing information based on a scale bar embedded in the images.
5×104 HeLa or HaCaT cells were plated in 12-well plates. Cells were infected with wild-type HPV16 PsV at MOI ~0.5 GFP-transducing units per cell (i.e., enough virus to result in GFP expression in one-half of the cells as assessed by flow cytometry on a BD Biosciences FACSCalibur flow cytometer 48 hours post-infection). The number of packaged wild-type reporter plasmids required to achieve this MOI in unmanipulated cells was quantified by qPCR, and an equivalent number of mutant genomes were used to infect cells. Depending on the experiment, 150–300 reporter plasmid genomes per cell resulted in an MOI of ~0.5 for wild-type pseudovirus in HeLa cells; approximately five-fold more virus was required to attain this MOI in HaCaT cells. In some experiments, cells were transfected with siRNA prior to infection. To confirm retromer knock-down, 5×105 cells were plated in a 6-well plate and reverse-transfected with 40nM siRNA targeting Vps26, Vps29, or Vps35. Forty-eight hours later, cells were lysed in sample buffer, electrophoresed, and analyzed by immunoblotting for Vps35.
3 to 5×104 HeLa cells were plated in eight-well chambered glass slides and infected the next day with wild-type PsV at MOI 20 or mutant PsV containing the same number of reporter plasmid genomes. (The lower sensitivity of immunofluorescence or PLA [see below] compared to reporter gene expression necessitated an MOI of 20 or 50 to visualize virus components.) Eight hours post-infection, the cells were fixed for 15 min at room temperature with 4% paraformaldehyde (Electron Microscopy Sciences, #15710), washed with PBS and then permeabilized with 0.5% Triton X-100 for 20 min at room temperature in PBS. The cells were blocked for one hour in 1% bovine serum albumin and 3% goat serum, and immunostained with 1:200 33L1–7 (obtained from Martin Sapp (LSU)). After extensive washes, AlexaFluor 594-conjugated goat anti-mouse secondary antibodies were added at 1:300 dilution for 40 minutes at room temperature. Nuclei were stained with 1:100 dilution of 4′,6-diamidino-2-phenylindole (DAPI), cells were washed extensively, and slides were mounted in Prolong gold anti-fade (Molecular Probes). Images were recorded on a ZEISS Axiovert 200 inverted fluorescent microscope using appropriate filters processed with ImageJ.
3×104 HeLa-M cells in 8-chambered glass slides were transfected with 10nM Vps35 or RISC-free siRNA. Twenty-four hours later, the Trans-IT HeLaMONSTER reagent (Mirus Bio) was used to transfect cells with 1 μg of a plasmid expressing a CD8-CIMPR fusion protein containing WLM (wild-type), AAA, or FYL. Twenty-four hours later, live non-permeabilized cells were incubated at 37°C with a 1:400 dilution of an antibody that recognizes the extracellular domain of CD8 (Ancell, 153–020). After three hours, the cells were fixed for 15 min at room temperature with 4% Formalde-Fresh, permeabilized with 0.5% Triton X-100 for 20 min, and then blocked with 5% donkey and goat serum for 30 minutes at room temperature. The cells were then stained with anti-GM130 (Abcam, ab52649 [1:200]) overnight at 4°C, washed five times with the blocking solution, and incubated with conjugated secondary antibody (Life Technologies [1:500]) for 30 minutes at room temperature. Cells were mounted with Duolink in situ Mounting Medium with DAPI, imaged on a ZEISS Axiovert 200 inverted fluorescent microscope and processed with Image J.
5×104 HeLa cells grown overnight on glass coverslips in 24-well plate were transfected with siRNA targeting retromer subunit Vps29 or control scrambled siRNA with Lipofectamine RNAi Max reagent (Life Technologies, Carlsbad, CA) 48 hours prior to infection. The cells were then infected with wild-type or mutant HPV16.L2F at MOI of 50 (according to genome normalization), fixed with 4% Formalde-Fresh at eight or sixteen hours post-infection, and permeabilized with 1% saponin for one hour at room temperature. The cells were incubated with anti-FLAG mouse antibody (Sigma, F3165 [1:1000]) to label L2 and an antibody recognizing EEA1 (Cell Signaling, C45B10 [1:100]) or TGN46 (Abcam, ab50595 [1:200]). Alternatively, cells were incubated with anti-FLAG rabbit antibody (Cell Signaling, 2368 [1:500]) and anti-Vps35 antibody (Abcam, 57632 [1:500]). PLA was performed with Duolink reagents from Olink Biosciences (Uppsala, Sweden) as described [22,54]. Briefly, samples were incubated with a pair of suitable PLA probes at 1:5 in a humidified chamber for one hour and processed for ligation for 30 min at 37°C. DNA was then amplified with fluorescent substrates for 100 min at 37°C. The nuclei were stained by incubation with 5μg/ml DAPI for 10 min and images were acquired as described above. Approximately 100 nuclei were imaged in each sample. The images were processed with ImageJ and quantitatively analyzed with BlobFinder software to measure total fluorescence intensity in each sample. The average fluorescence intensity per cell in each sample was normalized to the control sample as indicated in each experiment. All the experiments were done independently three times with similar results, and one representative experiment is shown.
Peptides shown in Fig. 6 were purchased from NeoBioLab (Cambridge, MA) at >95% purity. L2-N was biotinylated at its C-terminus with the N-terminus unmodified, while all the other peptides were biotinylated at their N-terminus and amidated at their C-terminus. L2-N was dissolved in sterile deionized water containing 0.01% sodium azide, L2-M was initially solubilized in a small amount of DMSO (~ 70–80 μl) and then dissolved in sterile deionized water with 0.01% sodium azide. L2-C was initially resuspended in 30% acetic acid and DMSO (~ 70 μl each), and then dissolved in sterile deionized water with 0.01% sodium azide. The FYL, YYML, DM and WLM peptides were initially solubilized in a small amount of 30% acetic acid (~80 μl), and then dissolved in sterile deionized water with 0.01% sodium azide. Peptide stocks (3.5–5.6 mg/ml) were aliquoted and stored at-20°C.
HeLa or HaCaT cells plated in six-well plates were lysed at ~80% confluency with 500 μl RIPA-MOPS buffer (20mM morpholinepropanesulfonic acid [pH 7.0], 150mM NaCl, 1% Nonidet P-40, 1mM EDTA, 1% deoxycholic acid, 0.1% sodium dodecyl sulfate [SDS]) supplemented with protease inhibitors (1X HALT protease and phosphatase inhibitor cocktail [Thermo Scientific]) or with 500 μl HEPES buffer (20mM Hepes pH8, 50mM NaCl, 5mM MgCl2, 1mM dithiothreitol (DTT), and 1.0% triton X-100) containing 1:100 Halt TM Protease & Phosphatase Inhibitors (Thermo Scientific, Prod # 78443). The lysate was centrifuged at 14,000 rpm for 20 min, and the supernatant was incubated with 10 μg of a biotinylated peptide for two hours at 4°C. 40 μl of streptavidin agarose beads slurry (Pierce, cat# 20349) was added, and the mixture was gently rocked for 45 min at 4°C. Beads were recovered by centrifugation and washed four times with RIPA-MOPS buffer supplemented with NaCl to a total of 0.4M or with HEPES buffer. Samples were analyzed by SDS-PAGE and immunoblotting with Vps35 (Abcam, ab57632) or Vps29 (Santa Cruz, sc-98611) antibody.
Individual human Vps26, Vps29, and GST-tagged Vps35 subunits were expressed individually in E. coli, and the assembled trimeric retromer complex was immobilized on GSH resin via the GST-tag on Vps35 as described [45]. Maltose binding protein (MBP)-L2–6His fusion proteins containing a C-terminal segment from wild-type or mutant HPV16 L2 (amino acids 434–457) were expressed in bacteria and purified using the AKTA-Prime plus FPLC system equipped with a His-trap column. The sequence appended to the C-terminus of MBP in the pMal-C2 expression vector (New England Biolabs) was GSASPQYTIIADAGDFYLHPSYYMLRKHHHHHHC (L2 sequence underlined). Purified proteins were exchanged into 20mM Hepes pH 8, 50mM NaCl and quantified by bicinchoninic acid assay. Five or 10 μM of each fusion protein was incubated with assembled retromer trimer immobilized on GSH resin for two hours at 4°C in 20mM HEPES pH 8.0, 50mM NaCl, 5mM MgCl2, 1mM DTT, and 0.1% Triton X-100. Beads were washed twice in HEPES buffer, suspended in SDS loading buffer, boiled, and subjected to SDS-PAGE and anti-His immunoblotting. Bands corresponding to the MBP-L2-His constructs were quantified by Image Lab.
|
10.1371/journal.ppat.1006434 | Genome-wide transposon mutagenesis of Proteus mirabilis: Essential genes, fitness factors for catheter-associated urinary tract infection, and the impact of polymicrobial infection on fitness requirements | The Gram-negative bacterium Proteus mirabilis is a leading cause of catheter-associated urinary tract infections (CAUTIs), which are often polymicrobial. Numerous prior studies have uncovered virulence factors for P. mirabilis pathogenicity in a murine model of ascending UTI, but little is known concerning pathogenesis during CAUTI or polymicrobial infection. In this study, we utilized five pools of 10,000 transposon mutants each and transposon insertion-site sequencing (Tn-Seq) to identify the full arsenal of P. mirabilis HI4320 fitness factors for single-species versus polymicrobial CAUTI with Providencia stuartii BE2467. 436 genes in the input pools lacked transposon insertions and were therefore concluded to be essential for P. mirabilis growth in rich medium. 629 genes were identified as P. mirabilis fitness factors during single-species CAUTI. Tn-Seq from coinfection with P. stuartii revealed 217/629 (35%) of the same genes as identified by single-species Tn-Seq, and 1353 additional factors that specifically contribute to colonization during coinfection. Mutants were constructed in eight genes of interest to validate the initial screen: 7/8 (88%) mutants exhibited the expected phenotypes for single-species CAUTI, and 3/3 (100%) validated the expected phenotypes for polymicrobial CAUTI. This approach provided validation of numerous previously described P. mirabilis fitness determinants from an ascending model of UTI, the discovery of novel fitness determinants specifically for CAUTI, and a stringent assessment of how polymicrobial infection influences fitness requirements. For instance, we describe a requirement for branched-chain amino acid biosynthesis by P. mirabilis during coinfection due to high-affinity import of leucine by P. stuartii. Further investigation of genes and pathways that provide a competitive advantage during both single-species and polymicrobial CAUTI will likely provide robust targets for therapeutic intervention to reduce P. mirabilis CAUTI incidence and severity.
| Proteus mirabilis is a common cause of single-species and polymicrobial catheter-associated urinary tract infections (CAUTIs). Prior studies have uncovered P. mirabilis virulence factors for single-species ascending UTI, but little is known concerning pathogenesis during CAUTI or polymicrobial infection. Using transposon insertion-site sequencing (Tn-Seq), we performed a global assessment of P. mirabilis fitness factors for CAUTI while simultaneously determining how coinfection with another CAUTI pathogen, Providencia stuartii, alters P. mirabilis fitness requirements. This approach provides six important contributions to the field: 1) the first global estimation of P. mirabilis genes essential for growth, 2) validation of a role for known P. mirabilis fitness factors during CAUTI, 3) identification of novel fitness factors, 4) identification of core fitness factors for both single-species and polymicrobial CAUTI, 5) identification of single-species fitness factors that are complemented during polymicrobial infection, and 6) identification of factors that only provide a competitive advantage during polymicrobial infection. We further demonstrate that the CAUTI model can be used to examine the interplay between fitness requirements of both species during coinfection. Investigation of fitness requirements for other pathogens during single-species and polymicrobial CAUTI will elucidate complex interactions that contribute to disease severity and uncover conserved targets for therapeutic intervention.
| The Gram-negative bacterium Proteus mirabilis thrives in a wide variety of environments, including soil, water sources, sewage, and as a commensal in the intestinal tract of humans and animals [1, 2]. P. mirabilis has also been identified as the causative agent of numerous human illnesses including cystitis, pyelonephritis, prostatitis, as well as intra-abdominal, wound, eye, and burn infections [2]. While it is capable of causing uncomplicated urinary tract infections (UTIs), this organism is a much more common cause of catheter-associated UTI (CAUTI) [3–5]. Indeed, we recently identified P. mirabilis as the most common cause of CAUTIs in twelve Michigan nursing homes [6]. CAUTIs are frequently polymicrobial [7, 8], and P. mirabilis is one of the most common organisms present during polymicrobial urine colonization and infection [3, 6]. UTIs and CAUTIs involving P. mirabilis are typically complicated by the formation of bladder and kidney stones (urolithiasis) and permanent renal damage [9–11], and may progress to bacteremia [12, 13]. Despite these potentially severe complications of P. mirabilis infection, there are no currently licensed vaccines available for this organism and multidrug-resistant isolates are increasingly common [14, 15].
No previous studies have explored P. mirabilis genes essential for growth in vitro, but numerous fitness and virulence factors have been examined in a murine model of ascending UTI, including but not limited to urease, fimbriae, and hemolysin (see [2, 16] for review). Fitness factors for colonization of the urinary tract have also been explored in three non-saturating signature-tagged mutagenesis (STM) studies in the ascending UTI model using P. mirabilis strain HI4320, which achieved a combined 70% theoretical coverage of the genome and revealed traditional virulence factors, metabolic pathways important for infection, and fitness factors with no prior links to P. mirabilis pathogenicity [17–19]. For P. mirabilis infection studies, the murine model of ascending UTI is generally considered to represent complicated UTI due to the occurrence of urolithiasis in this infection model [20]. However, we recently adapted a murine model of CAUTI for investigation of P. mirabilis pathogenicity, and verified that maintenance of a catheter segment within the bladder dramatically increases inflammation and infection severity [21]. Thus, different genes and pathways are likely required for fitness in the catheterized bladder environment than those identified in the ascending UTI model.
An additional consideration in identifying fitness and virulence factors of P. mirabilis is the frequent involvement of this organism in polymicrobial infections. We recently determined that other uropathogens, such as Escherichia coli, Enterococcus species, and Providencia stuartii, are capable of influencing P. mirabilis urease activity [21]. Co-culture with P. stuartii also enhanced P. mirabilis cytotoxicity independent of urease activity, altered the immune response to infection, and resulted in greater tissue damage, indicating that other virulence factors are affected by polymicrobial interactions [21]. All prior studies of P. mirabilis fitness factors have been conducted with pure cultures of P. mirabilis in isolation, yet the potential clearly exists for other organisms to influence expression of P. mirabilis virulence factors, metabolic requirements, and factors required for adaptation to changes in the bladder environment in response to the presence of a second pathogen. In this study, we generated a genome-saturating library of transposon mutants and utilized transposon insertion-site sequencing (Tn-Seq) to identify P. mirabilis genes essential for growth in rich medium and the full arsenal of fitness factors for single-species infection in a CAUTI model, while concurrently determining how fitness requirements changed during polymicrobial infection with P. stuartii, a common co-colonizing partner of P. mirabilis [3, 4, 22–25].
Based on the P. mirabilis HI4320 genome size (4.063 Mbp, with approximately 3747 genes [26]), it is estimated that 34,249 transposon mutants are required for 99.99% probability of full genome coverage [27]. A similar strategy as detailed by Crimmins et al. [28] for maximum colonization density was used to determine the appropriate transposon library pool density. From our recent investigation of experimental CAUTI in CBA/J mice, the minimum colonization density achieved by P. mirabilis in the urine, bladder, and kidneys of mice 4-days post inoculation is expected to be ~1x104 CFU [21]. Preliminary experiments confirmed a minimum bladder colonization of 1x104 CFU/gram of tissue at 4-h and 4-days post-inoculation, and indicated lack of a significant bottleneck in the CAUTI model (S1 Fig). We therefore concluded that generation of transposon pools containing 1x104 transposon mutants would be ideal. Approximately 50,000 transposon mutants from three independent matings were collected and pooled in groups of 1x104 mutants to generate five transposon mutant libraries. Randomness of insertions was verified by Southern blot, and the majority of mutants harbored only one transposon insertion as expected (S2 Fig).
Raw Illumina reads were filtered and transposon insertion-sites were uniquely mapped to genomic coordinates using a method adapted from the previously published work by Goodman et al. [29]. Saturation was achieved for the P. mirabilis HI4320 chromosome and 36 Kb plasmid (pHI4320), and the majority of insertion-sites were represented in three or more input pools (Fig 1). However, there were a few noticeable gaps in both the chromosome and plasmid maps. To determine if these gaps represented genes essential for growth of P. mirabilis in LB medium, a Bayesian mixture model was used to estimate essentiality based on absence or underrepresentation of transposon insertions in these genes within the input pools (see Materials and methods). The model identified 436 genes (11.6% of the 3747 genes encoded by P. mirabilis HI4320) as potentially essential for P. mirabilis growth (S1 Table), 279 (64.0%) of which were present in the bacterial section of the Database of Essential Genes (DEG, http://www.essentialgene.org/). Three hundred thirty-three of the estimated essential genes had a Cluster of Orthologous Groups of proteins (COG) assignment (Fig 2).
Among the list of P. mirabilis estimated essential genes are numerous genes listed as essential in the majority of bacterial species present in DEG, including genes pertaining to cell cycle control and division (such as the fts cell division proteins), cell wall biogenesis, replication (including DNA polymerase I, DNA polymerase III, and DNA gyrase), RNA polymerase σ70 (rpoD), σ32 (rpoH), and σ24 (rpoE), nucleoid protein H-NS (hns), numerous ribosomal proteins and tRNA synthetases, and ATP synthase. There are five ATP-dependent proteases in bacteria that comprise the majority of energy-dependent protein degradation: HflB (FtsH), the Clp proteases (ClpAP, ClpXP, and ClpYQ), and the Lon protease [30]. Insertions in three of these proteases (HflB, ClpAP, and ClpXP) were absent from the input pools and therefore identified as essential for growth of P. mirabilis in rich medium, underscoring the importance of degrading misfolded and unstable proteins for P. mirabilis growth in rich medium. The replication initiation protein PMIP01 on the P. mirabilis plasmid (pHI4320) was also identified as essential.
Infection studies to screen for P. mirabilis fitness factors during single-species and polymicrobial CAUTI were conducted in parallel to utilize the same inocula as the input samples for each infection type. A conceptual model of the infection and sequencing scheme is shown in Fig 3. For each of the five transposon library pools, 5–10 CBA/J mice were transurethrally inoculated with 1x105 CFU (10X coverage of each mutant within the pool), and 5–10 CBA/J mice were inoculated with 1x105 CFU of a 1:1 mixture of the transposon library and wild-type P. stuartii BE2467 (5x104 CFU of the P. mirabilis transposon mutant library for 5X coverage of each mutant within the pool). In all cases, a 4 mm segment of sterile silicone catheter tubing was carefully advanced into the bladder during inoculation and retained for the duration of the study as described previously [21]. Due to the level of encrustation in mice infected with urease-positive organisms, the catheter segments were generally embedded in the bladder tissue and therefore were not removed prior to homogenization of bladder samples. Thus, all CFUs recovered from bladder samples actually represent colonization of the catheter segment as well as the bladder.
Single-species infections and coinfections with the P. mirabilis transposon mutant pools resulted in comparable bladder and kidney colonization in all mice (Fig 4A and 4B). Differential plating revealed that the majority of coinfected mice were highly colonized by both P. mirabilis and P. stuartii, as expected (Fig 4C). In cases where an individual mouse exhibited low colonization by P. mirabilis in either the bladder or kidneys (<1x104 CFU), the samples from that mouse were excluded from further study. Bladder and kidney output samples for sequencing were chosen from the most highly colonized mice as follows: four mice from the single-species infection group and four from the coinfection group, per transposon pool, resulting in 20 bladder samples and 20 kidney samples from each infection type. A fitness index was calculated for each gene by taking into account the number of unique insertion-sites within that gene and the depth of reads at each insertion-site for the combined output samples recovered from the organs of infected mice compared to the combined input pools (see Materials and methods).
Tn-Seq from single-species CAUTI identified 629 genes (16.8% of the total genes encoded by P. mirabilis HI4320) as candidate fitness factors. Due to the stringent cutoffs that were used for estimation of fitness factors (see Materials and methods section), the range of fitness defects was compressed, spanning a 2- to 40-fold reduction in recovery from output samples compared to the input samples. The top candidate fitness factors for colonization of the catheterized bladder (excluding tRNAs, pseudogenes, and genes not present in COG) are shown in Table 1. Four of these genes were previously identified by signature-tagged mutagenesis (STM) in the ascending UTI model [18, 19], and eight were previously shown to be upregulated by P. mirabilis in the ascending UTI model [31]. Importantly, many of the top hits for bladder colonization represent multiple genes in an operon, including the phosphate ABC transporter (pstSCAB) and the cytochrome bo3 quinol oxidase (cyoABCDE). Top candidate fitness factors for kidney colonization (again excluding tRNAs, pseudogenes, and genes not present in COG) are shown in Table 2. These include the same four genes identified by STM that were important for bladder colonization, and fourteen genes that were upregulated by P. mirabilis in the ascending UTI model [31].
Upon further analysis, fitness factors for single-species CAUTI fell into three categories: 93 genes for colonization of the catheterized bladder that were not significant for kidney colonization (S2 Table), 209 genes for kidney colonization that were not significant for bladder colonization (S3 Table), and 286 genes that were important for colonization of both organs (S4 Table). Fitness factors for colonization of each organ were randomly distributed across the chromosome and plasmid (S3 Fig). Of the 286 genes important for both bladder and kidney colonization, 203 were present in COG: 24 pertained to transcription (10.9% of functional category assignments); 21 (9.5%) translation, ribosomal structure, and biogenesis; 19 (8.6%) amino acid transport and metabolism; 19 (8.6%) post-translational modification, protein turnover, and chaperones; 18 (8.2%) energy production and conversion; and 16 (7.3%) cell wall and envelope biogenesis (Fig 5A). Regarding fitness factors for bladder colonization alone, 64 of the 93 candidate factors were present in COG and revealed a greater proportion of genes pertaining to amino acid transport, carbohydrate transport and metabolism, and inorganic ion transport and metabolism than the genes important for colonization of both organs, and fewer candidate genes pertaining to translation and cell wall biogenesis (Fig 5B). Interestingly, the candidate fitness factors for kidney colonization alone followed a similar distribution as the genes important for colonization of both the bladder and kidneys with a few notable exceptions, including fewer genes pertaining to translation or carbohydrate transport and metabolism, and more pertaining to replication, recombination, and repair and extracellular structures. (Fig 5C). Thus, functional categories important for kidney colonization are likely to be involved in fitness during bladder colonization as well.
Overall, of the fifty-four P. mirabilis fitness factors previously identified by signature-tagged mutagenesis (STM) in the ascending UTI model [17–19], 31 (57%) were significant for CAUTI (denoted with a “*” in the supplemental tables). These include the phosphate ABC transporter (pstSCAB), pyruvate dehydrogenase (aceE and aceF), an AsnC transcriptional regulator (PMI1431), a d-methionine ABC transporter protein (metN), inosine-5’-monophosphate dehydrogenase (guaB), a bifunctional polymyxin resistance protein (arnA), a polysaccharide deacetylase next to the Arn operon (PMI1046), a dihydrouridine synthase (dusB), and a hypothetical protein (PMI3700). Urease is another well-known fitness factor for P. mirabilis UTI and prior STM studies that we previously verified as important for colonization in the CAUTI model [21], and a urease accessory protein (ureG) required for the catalytically-active enzyme was also identified in the present CAUTI Tn-Seq.
Tn-Seq from coinfection with P. stuartii revealed 1570 candidate P. mirabilis fitness factors, ranging from 2- to 30-fold reduction of transposon insertions in these genes from the combined output samples compared to the input samples. Candidate fitness factors represented 217/629 (34.5%) of the genes identified above for single-species CAUTI, indicating substantial overlap between the two infection models, as well as 1353 additional factors that may specifically contribute to colonization during coinfection. Sixty-two genes were candidate fitness factors for bladder colonization alone (S5 Table), 10 of which were also important for bladder colonization during single-species infection. These bladder-specific fitness factors, regardless of infection type, include two members of a carbohydrate ABC transporter (ugpC and ugpE), a nucleoid-associated protein (ndpA), the anti σE protein (rseA), an HxlR-family transcriptional regulator, and a siderophore receptor (ireA).
213 genes were candidate fitness factors for kidney colonization alone (S6 Table), 61 of which were also important for kidney colonization during single-species infection. The majority of the kidney-specific fitness factors, regardless of infection type, pertained to amino acid transport and metabolism (including the d-serine ammonia-lyase dsdA and the l-serine ammonia lyase sdaA), cell wall and envelope biogenesis, energy production and conversion (including succinate dehydrogenase and formate acetyltransferase), posttranslational modification and protein turnover (sufD, an iron-regulated ABC transporter permease protein, and surA, a periplasmic chaperone for outer membrane proteins), transcriptional regulation (including the phoP two-component response regulator, and an RpiR-family transcriptional regulator encoded by PMI2974), and four plasmid-encoded genes (PMIP09, which encodes the pilX4 conjugal transfer protein, PMIP10, PMIP28, and PMIP30). The Sec-independent protein translocase TatC (twin arginine transporter) was also identified as important for kidney colonization but not bladder colonization in both infection types.
909 genes were important for both bladder and kidney colonization during coinfection (S7 Table), 717 of which were present in COG and primarily pertained to transport and metabolism of amino acids, inorganic ions, carbohydrates, as well as energy production and conversion (Fig 10A). The COG functional categories represented by P. mirabilis fitness factors during coinfection followed a similar trend as single-species infection: fitness factors for bladder colonization alone revealed a greater proportion of genes pertaining to translation and transcription compared to the genes important for colonization of both organs (Fig 10B), and fitness factors for kidney colonization alone followed a similar distribution as genes important for colonization of the bladder and kidneys, albeit with an increase in factors with unknown function and a decrease in factors pertaining to cell motility (Fig 10C).
Fifteen of the genes important for colonization of both the bladder and kidneys during coinfection were also important for colonization of both organs during single-species infection. Seven of these genes have unknown functions or are not in COG; the other genes pertain to translation, ribosomal structure, and biogenesis (dusC, PMI3283, and rimI, an alanine acetyltransferase), cell wall/envelop biogenesis and defense (mdtA, a multidrug resistance efflux transporter), motility (PMI2642), and inorganic ion transport and metabolism (ppaA). The factors encoded by these genes appear to represent a putative core set of fitness requirements for P. mirabilis colonization of the murine urinary tract.
Intriguingly, there were also 109 genes for which transposon insertions resulted in a fitness defect during single-species infection but provided a competitive advantage during polymicrobial CAUTI (S8 Table). Seventy-nine of these genes were present in COG, and included factors involved in energy production and conversion, transcription, carbohydrate transport and metabolism, inorganic ion transport (particularly iron and phosphate), amino acid transport and metabolism, and nucleotide transport and metabolism.
Proteus mirabilis HI4320 is a model organism that has been used for decades to explore virulence determinants of this unusual bacterial species. With the availability of the complete genome sequence in 2008 [26], prior signature-tagged mutagenesis studies [17–19], and in vivo transcriptome assessment [31], much has been learned concerning how this organism differentiates into swarm cells, how it survives in the environment and during infection, the metabolic processes that allow for adaptation to the bladder and kidney environments, and virulence factors that contribute to ascending UTI (see [2, 16] for review). Despite this wealth of information, numerous questions remained regarding global P. mirabilis fitness and virulence factors, particularly concerning the relevance of known fitness factors to catheter-associated infection.
The present study details the first use of Tn-Seq for identification of fitness factors in a murine model of catheter-associated urinary tract infection (CAUTI), as well as the first use of Tn-Seq during polymicrobial infection. Strengths of this study include: 1) the use of multiple transposon library pools and four “replicate” mice per infection pool, 2) stringent cutoffs for analysis of insertion-site reads, and 3) generation and testing of mutants in each of the bacterial species under investigation for mechanistic insight into altered fitness requirements during coinfection. It is important to note that our approach is restricted by a few caveats common to all Tn-Seq studies, including an inability to assess the fitness contribution of genes for which insertions were not present in the input pools and difficulty assessing the fitness contribution of secreted factors due to complementation by other mutants within the input pools. Another limitation of the study comes from the use of two urease-positive organisms for our CAUTI coinfection model: the catheter segments were generally embedded in the bladder tissue, so we were unable to remove them for a separate assessment of fitness factors for catheter colonization versus bladder colonization. Thus, the fitness factors for bladder colonization likely represent a combination of factors for colonization of the catheter segment as well as the bladder epithelium.
Transposon insertions may also have polar effects on genes that are within an operon, a consideration that needs to be explored for follow-up studies on fitness factors of interest. For the eight mutants used for validation of the primary screen, four are within operons. The transcription unit for hslU includes hslV, the other subunit of the ATP-dependent protease that was similarly found to be a fitness factor. The transcription unit for PMI1518 includes PMI1519, encoding a hypothetical substrate binding protein. PMI1519 is upstream of PMI1518 in the transcription unit, and this gene contained transposon insertions but was not identified as a fitness factor, indicating the central importance of the efflux protein encoded by PMI1518. For ilvD, the transcription unit is organized as ilvGMEDA, and there were differences in the fitness defects for each of these genes as show in S9 Table. ArnA also resides in the middle of a transcription unit, but the defects for this operon are not likely due to polar effects as the first four genes were fitness factors for single-species CAUTI but not during coinfection, and the remaining three genes in the operon were identified as essential for growth in rich medium.
Despite these limitations, our approach uncovered numerous previously unrecognized P. mirabilis fitness determinants and explored the impact of polymicrobial infection on fitness requirements. Indeed, fitness defects were verified for 7/8 (88%) candidate fitness factors tested by direct co-challenge with wild-type P. mirabilis and 3/3 mutants tested during coinfection. Results of interest from the essential gene analysis, single-species CAUTI, and polymicrobial CAUTI are detailed below.
The majority of the large regions of the chromosome that lacked transposon insertions and were visible as gaps in Fig 1 did indeed correspond to genes identified as essential. For instance, the largest region that lacked transposon insertions was 12,423 bp encompassing 18 genes involved in cell division and cell wall biogenesis. The next largest regions that lacked transposon insertions were a 7,543 bp region encompassing genes necessary for translation, and a 6,650 bp region encompassing an intergenic region and murI, a glutamate racemase that is essential for cell wall biosynthesis. However, this was not always the case for gaps in transposon insertions on plasmid pHI4320. The largest region lacking insertions on the plasmid was 1,252 bp encompassing an intergenic region and a portion of PMIP30, which encodes a putative colicin. This gene was not identified as essential as there were 81 insertion sites after the gap which were well-represented in the input pools. Another gap on the plasmid of 826 bp corresponds to a portion of PMIP42, a putative RelB antitoxin. Similar to PMIP30, PMIP42 contained two insertion sites after the gap that were highly represented in the input pools, which likely caused it to fall below the threshold for being considered essential. Thus, the model used to identify putative essential genes provided very conservative estimates, and is likely an underrepresentation of the genes required for optimal growth of P. mirabilis in LB. It is therefore notable that the other ~500 bp gaps in transposon insertions on the plasmid were within a conjugal transfer protein (PMIP09), a colicin immunity protein (PMIP31), a plasmid stability/partitioning protein (PMIP34), a type IA DNA topoisomerase (PMIP25), and HN-S family protein (PMIP26). These findings are in agreement with the replication initiation protein (PMIP01) being essential, and underscore the importance of plasmid maintenance and replication to P. mirabilis growth.
In addition to the numerous genes previously identified as essential for growth in other bacterial species, the list of P. mirabilis estimated essential genes contained a few unusual items. For instance, all but one of the genes in the non-oxidative pentose phosphate pathway were identified as putative essential genes (rpiA, rpe, tktA, and talB), indicating a central role for this pathway during growth of P. mirabilis in rich medium. Phosphoglycerate kinase (pgk), which is involved in glycolysis, gluconeogenesis, and glycerol degradation, was also identified as a possible essential gene. Factors involved in inorganic ion transport and metabolism were a surprising find, as these genes are not commonly identified as essential. These included genes involved in potassium uptake (trkA and trkH), intracellular sulfur oxidation (PMI2797 and PMI2798), tellurite resistance (terC), a sulfite reductase (PMI0794), and magnesium and cobalt efflux (corC). TrkA and TrkH comprise a high-rate low-affinity potassium-translocating system that requires ATP via the Sap system [37], and all members of the Sap dipeptide transport system transcriptional unit (sapBCDF) were also identified as putative essential genes.
Six fimbrial genes were also identified as potential essentials, including three homologs of the mrpJ fimbrial operon regulator [38]. Little is known concerning the identified fimbrial genes (pmpB, fim3J, fim5G, fim7D, fim8J, fim10D, and fim10J), or why disruption of these genes would impact growth in rich medium. Three of these fimbrial genes encode homologs of mrpJ, a transcriptional regulator that represses motility but also influences expression of other adhesins, virulence factors, and metabolic pathways [39]. Homologs of mrpJ are also capable of regulating motility and adherence [38], so it is possible that these mrpJ homologs similarly play a role in a broad regulatory network that could contribute to growth, or that tight regulation of the expression of these operons is important during growth in rich medium. Two homologs of the type VI secretion system (T6SS) secreted protein Hcp (PMI0750 and PMI1117) were also identified as putative essential genes, as well as 17 transposases and 1 phage repressor protein. In each of these cases, the identified genes contained >15 TA sites for transposon insertion, but <3 total insertions were recovered from the five combined input pools. While these genes may be important for growth in rich medium, it is also possible that the mariner transposon wasn’t able to efficiently target these chromosomal locations. Thus, the importance of the unique genes estimated to be essential for P. mirabilis should be interpreted with caution.
As mentioned above, the genes involved in putrescine uptake and biosynthesis were only identified as fitness requirements for P. mirabilis single-species CAUTI and not polymicrobial CAUTI, indicating that the presence of P. stuartii alleviates the polyamine requirement of P. mirabilis. The same appears to be true of the high-affinity zinc transport system encoded by znuABC. Stress responses also appear to be more important during single-species CAUTI than coinfection, as the majority of the stress-related genes identified as P. mirabilis fitness factors for single-species CAUTI were not significant during coinfection. Similarly, glutamine synthetase (glnA) was important for single-species fitness but not coinfection, further underscoring the differences in metabolic pathways favored by P. mirabilis during these two infection types. Further research is needed to determine which of these shifts in fitness requirements are specifically due to P. stuartii and which are due to the altered host environment and response to coinfection compared to P. mirabilis single-species infection.
The combination of genome-saturating transposon mutant libraries and Tn-Seq has allowed for the first global estimation of P. mirabilis essential genes, validation of numerous P. mirabilis virulence factors and fitness determinants from decades of studies using the ascending model of UTI, the discovery of novel fitness determinants specifically for CAUTI, and a stringent assessment of how polymicrobial infection influences fitness requirements. For instance, proteobactin, a yersiniabactin-related siderophore (nrp), and heme receptors were promising targets for perturbing P. mirabilis ascending UTI [56], but these factors do not appear to be important for fitness during single-species infection in the CAUTI model. BCAA biosynthesis is an intriguing target for reducing bacterial colonization, but our results indicate that this pathway is only important for P. mirabilis during polymicrobial infection and therefore would not be a suitable target for single-species infection by P. mirabilis. In contrast, our data indicate that polyamine uptake and biosynthesis may be promising targets for perturbing P. mirabilis single-species CAUTI, but these pathways were not important during coinfection, which may limit their potential as therapeutic targets given the high frequency of P. mirabilis polymicrobial colonization and infection.
Further research is needed concerning the numerous genes and pathways that provided a competitive advantage to P. mirabilis during both single-species and polymicrobial CAUTI, as the underlying mechanisms of these fitness requirements are the most likely to provide conserved targets for therapeutic intervention aimed at reducing P. mirabilis colonization or minimizing risk of progression to severe infection and urosepsis. We have clearly demonstrated that the CAUTI coinfection model can be used to examine the interplay between fitness requirements for both species during coinfection. It is therefore likely that investigation of the fitness requirements of other common CAUTI pathogens for single-species and polymicrobial CAUTI will further elucidate complex bacterial interactions that contribute to disease severity, and may even uncover conserved bacterial targets for therapeutic intervention.
|
10.1371/journal.pbio.3000201 | Up-regulation of FOXD1 by YAP alleviates senescence and osteoarthritis | Cellular senescence is a driver of various aging-associated disorders, including osteoarthritis. Here, we identified a critical role for Yes-associated protein (YAP), a major effector of Hippo signaling, in maintaining a younger state of human mesenchymal stem cells (hMSCs) and ameliorating osteoarthritis in mice. Clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR associated protein 9 nuclease (Cas9)-mediated knockout (KO) of YAP in hMSCs resulted in premature cellular senescence. Mechanistically, YAP cooperated with TEA domain transcriptional factor (TEAD) to activate the expression of forkhead box D1 (FOXD1), a geroprotective protein. YAP deficiency led to the down-regulation of FOXD1. In turn, overexpression of YAP or FOXD1 rejuvenated aged hMSCs. Moreover, intra-articular administration of lentiviral vector encoding YAP or FOXD1 attenuated the development of osteoarthritis in mice. Collectively, our findings reveal YAP–FOXD1, a novel aging-associated regulatory axis, as a potential target for gene therapy to alleviate osteoarthritis.
| Stem cell aging contributes to aging-associated degenerative diseases. Studies aiming to characterize the mechanisms of stem cell aging are critical for obtaining a comprehensive understanding of the aging process and developing novel strategies to treat aging-related diseases. As a prevalent aging-associated chronic joint disorder, osteoarthritis is a leading cause of disability. Senescent mesenchymal stem cells (MSCs) residing in the joint may be a critical target for the prevention of osteoarthritis; however, the key regulators of MSC senescence are little known, and targeting aging regulatory genes for the treatment of osteoarthritis has not yet been reported. Here, we show that Yes-associated protein (YAP), a major effector of Hippo signaling, represses human mesenchymal stem cell (hMSC) senescence through transcriptional up-regulation of forkhead box D1 (FOXD1). Lentiviral gene transfer of YAP or FOXD1 can rejuvenate aged hMSCs and ameliorate osteoarthritis symptoms in mouse models. We propose that the YAP–FOXD1 axis is a novel target for combating aging-associated diseases.
| Mesenchymal stem cells (MSCs) are widely distributed in adult tissues and have the capacities of self-renewal and differentiation into multiple cell lineages, such as chondrocytes, osteoblasts, and adipocytes [1]. MSCs are involved in tissue repair and homeostatic maintenance [2,3]. Over time, MSCs exhibit an age-associated decline in their number and function [4–6], namely, MSC senescence, which may be implicated in the loss of tissue homeostatic maintenance and leads to organ failure and degenerative diseases [7–10]. Therefore, an understanding of the mechanisms underlying MSC senescence will likely reveal novel therapeutic targets for ameliorating degenerative diseases.
Osteoarthritis is a prevalent aging-associated disorder that is characterized by the progressive deterioration of articular cartilage [11,12]. In osteoarthritis joints, degenerative changes start with cellular disorganization, gradual stiffening, and irregular surface of superficial zone followed by loss of matrix, clefts, and osteophyte formation in the deep articular cartilage [13,14]. Accordingly, disruption of the superficial zone of cartilage is an onset of osteoarthritis. Previous reports have demonstrated that cells isolated from the superficial zone of mouse and human articular cartilage express MSC markers, including cluster of differentiation (CD) 105, CD166, CD29, and exhibit MSC characteristics [15–20]. Cell death induced by oxidative stress or wound occurs primarily at the surface zone of cartilage [21,22]. When such cell death is inhibited by chemicals, cartilage disorganization and matrix loss are greatly reduced [23]. Therefore, MSCs or chondrocyte progenitor cells residing in the superficial zone of cartilage may be a critical target for the prevention of osteoarthritis. Although the transplantation of ex vivo cultures of MSCs into the osteoarthritic joint has been shown to improve the symptoms [24–26], the rejuvenation of endogenous senescent MSCs may also be a therapeutic option for osteoarthritis. The localized nature of osteoarthritis, which has no major extra-articular or systemic manifestations, makes it an ideal candidate for local, intra-articular gene therapy [27,28]. However, gene therapy strategies aiming at alleviating senescence, particularly MSC senescence, for treating osteoarthritis have not yet been reported.
Yes-associated protein (YAP) and transcriptional coactivator with PDZ-binding motif (TAZ) are primary targets of the Hippo signaling pathway, which plays important roles in the regulation of development, homeostasis, regeneration, and so forth [29–31]. The Hippo kinase cascade phosphorylates YAP and TAZ, resulting in their cytoplasmic retention and proteolytic degradation. When the Hippo pathway is inactive, YAP and TAZ translocate into the nucleus and interact with transcription factors to regulate the expression of target genes [32]. YAP and TAZ, as paralogs, have been demonstrated as key regulators in organ size control [33] and essential transducers of mechanical signals [34]. Here, we identified a critical role for YAP, but not TAZ, in regulating human MSC (hMSC) senescence. YAP exerted a geroprotective effect on hMSCs through the transcriptional activation of forkhead box D1 (FOXD1) in a TEA domain transcription factor (TEAD)-dependent manner. Gene therapy with lentiviral vectors encoding YAP or FOXD1 prevented cellular aging and attenuated osteoarthritis in mice. Our data suggest that YAP and its downstream target FOXD1 are novel suppressors of hMSC senescence and that the YAP–FOXD1 regulatory axis represents a potential therapeutic target for osteoarthritis.
We first used Clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated protein 9 nuclease (Cas9)-mediated gene editing [35] to generate isogenic human embryonic stem cells (hESCs) lacking YAP or TAZ to study the functions of YAP and TAZ in regulating human stem cell homeostasis (S1A, S1B and S1G Fig). Successful gene targeting at the YAP or TAZ locus were verified by genomic polymerase chain reaction (PCR) and reverse transcription quantitative PCR (RT-qPCR) (S1C and S1D Fig). Western blot and immunofluorescence further confirmed the complete ablation of YAP and TAZ protein in YAP−/− and TAZ−/− hESCs, respectively, with no detectable compensations between YAP and TAZ (Fig 1A, 1B and 1C). Both YAP−/− and TAZ−/− hESCs maintained normal pluripotency (Fig 1D) and cell cycle kinetics (Fig 1E) and were able to differentiate into tissues composed of all 3 germ layers in vivo (S1E Fig). Karyotype and genome-wide copy number variation (CNV) analyses demonstrated that genomic integrity was maintained in YAP−/− and TAZ−/− hESCs after more than 30 passages (Figs 1F and S1F). Moreover, YAP−/− and TAZ−/− hESCs displayed transcriptional profiles that were highly similar to wild type (WT) hESCs (Fig 1G).
We next differentiated WT, YAP−/−, and TAZ−/− hESCs into hMSCs (Fig 2A) [35–38]. The derived hMSCs expressed a series of hMSC markers including CD105, CD166, CD29, CD90, CD73, CD44, CD13, and human leukocyte antigens, A, B and C (HLA-ABC) and were negative for hematopoietic or skeletal lineage markers CD34, CD43, CD45, CD14, CD19, podoplanin (PDPN), and CD164, resembling the resident CD105+, CD166+, and CD29+ MSCs in the superficial zone of articular cartilage (Figs 2A and S2A) [39–41]. Whereas WT hMSCs were able to differentiate into chondrocytes, osteoblasts, and adipocytes, YAP−/− and TAZ−/− hMSCs exhibited compromised differentiation abilities into osteoblasts and chondrocytes (S2B–S2E Fig). Additionally, the growth rates of WT, YAP−/−, and TAZ−/− hMSCs were analyzed through in vitro serial passaging. Compared with WT and TAZ−/− hMSCs, YAP-deficient hMSCs exhibited early-onset aging characteristics and arrested at passage 6 (Fig 2B). Increased levels of senescence-associated–β-galactosidase (SA-β-gal) activity (Fig 2C) and increased expression of P16, P53, and P21 (Fig 2D) were detected at as early as passage 4. Concomitantly, YAP−/− hMSCs showed a series of premature phenotypes, including (1) a decreased percentage of cells in synthesis (S) phase and increased percentages of cells in gap phases (G0, G1, and G2) and mitosis (M) phase (Fig 2E), (2) a lower percentage of Ki67-positive cells (S2F and S2G Fig), (3) decreased levels of the nuclear lamina-associated protein 2 (LAP2; S2F and S2G Fig), (4) reduced levels of heterochromatin protein 1 alpha (HP1α) and heterochromatin protein 1 gamma (HP1γ; S2F and S2G Fig), and (5) higher levels of reactive oxygen species (ROS; S3A Fig), compared with WT and TAZ−/− hMSCs. We subsequently examined whether the YAP deficiency resulted in stem cell attrition in vivo. WT, YAP−/−, and TAZ−/− hMSCs were transduced with a lentiviral vector expressing luciferase (Luc) and injected into the tibialis anterior (TA) muscles of immunodeficient mice. Consistent with the in vitro observations, YAP−/− hMSCs, but not TAZ−/− or WT hMSCs, exhibited an accelerated functional decay after transplantation in vivo (Figs 2F and 2G and S3B and S3C).
To further evaluate the effect of YAP in hMSCs, WT hMSCs were transduced with lentiviruses encoding a single guide RNA (sgRNA) targeting YAP or a non-targeting control (NTC) sgRNA, as well as CRISPR/Cas9 [42,43]. Phenotypic characterizations revealed that the down-regulation of YAP in hMSCs also resulted in a similar premature aging phenotype (S3D and S3E Fig). By contrast, ectopic expression of YAP rescued the premature senescence observed in YAP−/− hMSCs, as evidenced by the reduced percentage of SA-β-gal–positive cells (Fig 2H), enhanced growth rate and clonal expansion ability (S3F and S3G Fig), decreased expression of P16 and P21 (S3H Fig), lower levels of ROS (S3I Fig), and slower in vivo decay after engraftment (Figs 2I and S3J). Taken together, these data suggest that YAP, but not TAZ, plays an essential role in protecting hMSCs from premature senescence.
Given that YAP and TAZ displayed distinct functions in regulating hMSC senescence, we next examined whether there were differences in the subcellular localizations of YAP and TAZ. In hMSCs, YAP was predominantly located in the nucleus, whereas TAZ was in the cytoplasm (Fig 3A). It has been shown that nuclear YAP binds transcription factors, including TEAD family of transcription factors (TEAD1, 2, 3, and 4), as a transcriptional coactivator to induce target gene expression and thus regulate a series of cellular processes [44]. To test whether the nuclear YAP acted in conjunction with TEAD to regulate hMSC senescence, we blocked the activities of all the members of TEAD family in hMSCs as confirmed by immunoblotting analysis (referred to as TEADs knockdown [KD] and KO hMSCs, TEADs KD/KO hMSCs; Fig 3B). Similar to YAP-deficient hMSCs, TEADs KD/KO hMSCs also showed major phenotypes of premature senescence, such as an increased number of SA-β-gal–positive cells (Fig 3C), compromised clonal expansion abilities (S4A Fig), and up-regulation of P16 and P21 (S4B Fig). These observations suggest that YAP safeguards hMSCs from premature senescence in a TEAD-dependent manner.
To elucidate the molecular mechanism underlying YAP–TEAD regulation of hMSC senescence, RNA sequencing (RNA-seq) analyses of WT, YAP−/−, and TAZ−/− hMSCs were performed (S4C Fig). TAZ−/− hMSCs displayed comparable transcriptional features compared to those of WT hMSCs, whereas YAP−/− hMSCs exhibited a substantial number of differentially expressed genes (Figs 3D and S4E and S2 and S3 Datas). We observed few overlaps between differentially expressed genes in YAP KO and TAZ KO hMSCs compared to WT cells (S4D Fig), consistent with differential subcellular localization patterns of YAP and TAZ in hMSCs. Searching for TEAD binding motifs in the genome [45,46] identified 476 (55%) of the 862 down-regulated genes in YAP−/− hMSCs as potential TEAD targets (P < 1.0 × 10−4; Fig 3E). Among them, FOXD1 was the most significantly down-regulated gene in YAP−/− hMSCs (S4 Data).
Western blotting verified the down-regulation of FOXD1 expression in YAP−/− hMSCs (Fig 3F) as well as its up-regulation upon the reintroduction of YAP (Fig 3G), suggesting that FOXD1 was transcriptionally controlled by YAP. We examined the FOXD1 promoter region, including 1,500 bp upstream of the transcriptional start site (TSS) and identified 4 putative TEAD binding sites between −1,500 and −1,000 bp and 1 between −1,000 bp and the TSS (Fig 3H). Accordingly, we detected these 2 regions followed by chromatin immunoprecipitation (ChIP) using YAP and TEAD4 antibodies, revealing that YAP and TEAD4 bound predominantly within 1,000 bp upstream of the FOXD1 TSS, where there was a putative TEAD binding site (Fig 3I and 3J). Next, we cloned this promoter region (−1,000 bp to the TSS) as a transcriptional element upstream of a basic Luc reporter. Reporter activity was lower in YAP−/− hMSCs than in WT cells (Fig 3K) and was increased upon YAP or TEAD4 overexpression. Luc activity was even higher upon the expression of a constitutively activated YAP mutant (YAP-S127A) and was further enhanced by coexpression of YAP and TEAD4 (Fig 3L). The high levels of Luc activity were significantly abolished when we mutated the predicted TEAD binding site (Fig 3M). By contrast, ChIP assay demonstrated that TAZ did not bind to the FOXD1 promoter (S4F Fig), and the Luc activity was insensitive to cellular TAZ levels (Figs 3K and S4G). Therefore, the YAP–TEAD pathway, but not TAZ, transcriptionally activates FOXD1 expression.
FOXD1 was initially implicated in renal development [47], but there was a lack of evidence for a link between FOXD1 and cellular senescence. To investigate whether FOXD1 participated in YAP deficiency-induced accelerated senescence of hMSCs, we knocked out FOXD1 in hMSCs using a lentiviral vector-dependent CRISPR/Cas9 system [42,43] (Figs 4A and S5A). FOXD1 depletion in hMSCs increased the percentage of SA-β-gal–positive cells (Fig 4B), inhibited clonal expansion (S5B Fig), and up-regulated P16 and P21 expression (S5C Fig), recapitulating the major phenotypes implicated in premature senescence caused by the YAP deficiency. The overexpression of FOXD1 in YAP−/− hMSCs effectively alleviated the accelerated senescence (Fig 4C). In addition, we also examined the gene expression profile of FOXD1 KO hMSCs using RNA-seq (S5 Data). FOXD1 KO decreased the expression of genes that were mainly associated with cell division and DNA replication, which ultimately contributed to the senescence phenotypes (Fig 4D and S6 Data). Combined analyses with WT and YAP−/− hMSCs showed that FOXD1-deficient and YAP−/− hMSCs were similar to each other at the transcriptomic level (S5D and S5E Fig). Many differentially expressed genes were overlapped between YAP−/− (compared to WT) and FOXD1 KO (compared to NTC-transduced) hMSCs, including 116 up-regulated genes accounting for 20% of the total up-regulated genes in YAP−/− hMSCs and 276 down-regulated genes accounting for 32% of the total down-regulated genes in YAP−/− hMSCs (Fig 4E and 4F), implying an important role for FOXD1 in mediating YAP deficiency-induced premature cellular aging. Of note, many of those commonly down-regulated genes were elevated upon ectopic expression of FOXD1 in YAP−/− hMSCs (Fig 4G). Conversely, ectopic expression of YAP in FOXD1 KO or TEADs KD/KO hMSCs did not exert obvious rescue effect on the senescence phenotypes (S5F and S5G Fig). Taken together, these data indicate that down-regulation of FOXD1, an effector of YAP–TEAD signaling, contributes to the premature senescence induced by YAP deficiency.
To further elucidate the relationship between the YAP–FOXD1 axis and human stem cell aging, we examined the expression levels of YAP, pan-TEAD, and FOXD1 in both replicative-senescent (RS) hMSCs and Werner syndrome (WS) hMSCs, a human stem cell model for premature aging disorder WS [37,48]. Western blotting revealed decreased levels of YAP, pan-TEAD, and FOXD1 in both types of senescent hMSCs (Fig 5A and 5D). Moreover, the activity of 8 × GTIIC-Luc, a YAP/TAZ-responsive reporter, decreased in both RS hMSCs and WS hMSCs (Fig 5B and 5E). Lentiviral overexpression of YAP or FOXD1 effectively attenuated the senescent features of RS hMSCs (Figs 5C and S6A–S6C) and WS hMSCs (Figs 5F and S6D–S6G). We also observed diminished protein levels of YAP and FOXD1 in RS-primary hMSCs isolated from human bone marrow (BM-hMSCs) (Fig 5G). In BM-hMSCs, KO of YAP or FOXD1 with a CRISPR/Cas9 system promoted cellular senescence (Fig 5H–5K), whereas the overexpression of YAP or FOXD1 delayed BM-hMSC senescence (Fig 5L and 5M). Collectively, these observations establish a geroprotective role for the YAP–FOXD1 axis in alleviating hMSC aging.
Mesodermal cellular aging has emerged as a fundamental hallmark of aging-related disorders, including osteoarthritis, one of the most common degenerative diseases, the incidence of which increases significantly with age. Dysfunction of MSCs residing in the superficial zone of cartilage precedes osteoarthritis [15–19,21,22] that is characterized by articular cartilage degradation [49–51]. To validate a role of hMSC senescence in driving osteoarthritis, we injected young hMSCs, RS hMSCs, and RS hMSCs overexpressing YAP or FOXD1, respectively, into the joints of immunodeficient mice and performed histological assessment of the joints 1 month later (S7A Fig). In line with a previous report [52], Safranin O staining revealed the delamination of the articular surface and erosion of articular cartilage in the RS hMSC–administrated joints (S7B and S7C Fig). However, no osteoarthritis-related features manifested in the joints transplanted with young hMSCs or RS hMSCs overexpressing YAP or FOXD1. RT-qPCR further demonstrated that RS hMSCs, rather than young hMSCs, and RS hMSCs overexpressing YAP or FOXD1 induced aging markers in mouse joints (S7D Fig). These results suggest that accumulation of senescent MSCs in joints contributes to the development of osteoarthritis, which can be eliminated by YAP or FOXD1 overexpression.
The elimination of local senescent cells using pharmacological or genetic approaches have been proven effective in attenuating age-associated bone loss and development of post-traumatic osteoarthritis in rodents [51,53]. Given the ability of YAP or FOXD1 to rejuvenate senescent MSCs, we hypothesized that intra-articular injection of lentiviral vectors expressing YAP or FOXD1 might exert a therapeutic effect on osteoarthritis. To test this, we performed an anterior cruciate ligament transection (ACLT) surgery widely used to trigger osteoarthritis in mice and then administrated the lentiviruses expressing flag-tagged Luc, YAP, or FOXD1 intra-articularly (Fig 6A). The lentiviral vectors steadily expressed exogenous proteins in and around the joints receiving virus injection for at least 7 weeks (S8A Fig). High expression levels of YAP and FOXD1 were detectable by RT-qPCR (Fig 6B and 6C); immunohistochemical analysis of the flag-tagged Luc, YAP, and FOXD1 further verified the persistent infection of the lentiviruses and expression of indicated proteins primarily in the superficial zone of articular cartilage (S8B Fig). As expected, ACLT induced the accumulation of P16-positive senescent cells in the articular cartilage, particularly in the superficial zone of cartilage, of the osteoarthritis mice (Fig 6D), which was accompanied by decreased levels of YAP and FOXD1 (S8C and S8D Fig). YAP or FOXD1 gene therapy reduced the number of senescent cells and alleviated ACLT-induced articular cartilage erosion and clefts (Fig 6D and 6E). Consistently, a substantial proportion of gene expression changes in the joints induced by ACLT were reversed by YAP or FOXD1 gene therapy (S8C–S8F Fig and S7 Data). For instance, increased expression of genes associated with inflammation (Mmp13, Il6, etc.), cellular senescence (P21, Serpine1, etc.), and cell apoptosis (Dapk1, Casp4, etc.) were observed in ACLT-induced osteoarthritis joints, and the expression levels of most of these genes were diminished upon YAP or FOXD1 treatment. Moreover, the YAP or FOXD1 treatment enhanced the expression of proliferation markers (Ki67, Aspm, etc.) and chondrocyte differentiation-related genes (Col2a1, Acan, etc.) (Fig 6F). Taken together, these data suggest that the YAP- or FOXD1-mediated alleviation of cellular senescence in local bone joints helps create a prochondrogenic environment and alleviates disease symptoms.
Cellular senescence and stem cell exhaustion are hallmarks of aging [54]. Accelerated attrition of the MSC pool has been observed in human stem cell and mouse models of premature aging disorders, including WS and Hutchinson Gilford progeria syndrome (HGPS) [37,55]. Transplantation of mesoderm-derived stem cells from young animals increases the lifespan of progeroid mice [56]. Quercetin has been shown to alleviate MSC senescence [57], improve physical function, and increase lifespan in aged mice [58]. From this perspective, senescent MSCs could be good therapeutic targets for aging-associated degenerative disorders. In this study, we presented several lines of evidence supporting a geroprotective role of YAP and FOXD1 in rejuvenating hMSCs: (1) YAP is required for preventing premature senescence of hMSCs; (2) YAP transcriptionally activates FOXD1 expression whereas YAP deficiency results in down-regulation of FOXD1, which contributes to the early-onset of cellular aging; and (3) lentiviral gene transfer of YAP or FOXD1 alleviates cellular senescence and osteoarthritis. Our findings define a critical role of the YAP–FOXD1 axis in regulating hMSC aging, which highlights new avenues for translation into geriatric and regenerative medicine.
The Hippo-YAP/TAZ signaling pathway is an evolutionarily conserved pathway that regulates cell proliferation and apoptosis. Here, we focused on the function of YAP and/or TAZ in regulating hMSC senescence. We generated isogenic YAP- or TAZ-deficient hMSCs. Compared with WT cells, TAZ−/− hMSCs showed minimal effect in cell growth, whereas YAP−/− hMSCs exhibited accelerated senescence. The cytoplasmic localization of TAZ underlays its inactivation in hMSCs, whereas nuclear YAP was essential for counteracting hMSC senescence. Consistent with our observations, emerging studies have revealed the differences between YAP and TAZ. For example, TAZ promotes the myogenic differentiation of myoblasts at late stages of myogenesis, whereas YAP inhibits this process in mice [59]. However, in-depth insights into the molecular mechanisms underlying these functional differences require further investigations.
FOXD1 is a new downstream target of YAP, loss of which mediates the senescent phenotype of YAP-deficient hMSCs. As a member of the forkhead box family of transcription factors, FOXD1 is known to regulate kidney development during organogenesis [60,61]. Recently, FOXD1 has been shown to promote cell proliferation by targeting the sonic hedgehog pathway and cyclin-dependent kinase inhibitors [62,63]. FOXD1 also facilitates the reprogramming of mouse embryonic fibroblasts (MEFs) into induced pluripotent stem cells (iPSCs) [64]. Here, we identified a geroprotective role for FOXD1 as a transcriptional target of YAP in rejuvenating hMSCs. Overexpression of YAP or FOXD1 delayed replicative and pathological senescence, implying a therapeutic potential of targeting the YAP–FOXD1 axis to relieve aging-associated degenerative diseases.
In a therapeutic context, we provided a proof-of-concept evidence that intra-articular lentiviral transduction of a single protein exerted therapeutic effects on ACLT-induced osteoarthritis, an age-related disorder. Because ACLT-induced osteoarthritis is accompanied by the accumulation of senescent cells [65], efforts have been made on chemical-induced elimination of senescent cells to alleviate osteoarthritis in mouse models [50,51]. With gene therapy offering novel therapeutic options for osteoarthritis [66], intra-articular injection presents a minimally invasive procedure that avoids conventional barriers to joint entry, increases bioavailability, and lowers systemic toxicity [67]. For the first time, our study shows that the intra-articular injection of lentiviruses expressing YAP or FOXD1 reduces the number of senescent cells, inhibits articular inflammation and cartilage erosion, and ameliorates the pathological symptoms. Therefore, gene therapy via the introduction of geroprotective factors aiming at rejuvenating senescent cells may represent a new avenue to treating osteoarthritis in the future.
All animal experiments were conducted in compliance with animal protocols approved by the Chinese Academy of Science Institutional Animal Care and Use Committee, licensed by the Science and Technology Commission of Beijing Municipality (SYXK-2016-0026). All mice were housed under a 12-hour light/dark cycle at constant temperature (22°C). Food and water were available ad libitum. Mice were anaesthetized using isoflurane and euthanized with CO2 followed by cervical dislocation.
Human H9 ESCs as well as derived YAP−/− and TAZ−/− hESCs were maintained on feeder layers of mitomycin C–inactivated MEFs in hESC medium [68] (DMEM/F12 [Thermo Fisher Scientific, Waltham, MA], 20% Knockout Serum Replacement [Thermo Fisher Scientific], 0.1 mM nonessential amino acids [NEAAs; Thermo Fisher Scientific], 2 mM GlutaMAX [Thermo Fisher Scientific], 1% penicillin/streptomycin [Thermo Fisher Scientific], 55 μM β-mercaptoethanol [Thermo Fisher Scientific], and 10 ng/ml bFGF [Joint Protein Central, Incheon, Korea]) or on Matrigel (BD Biosciences, San Jose, CA, USA) in mTeSR medium (STEMCELL Technologies, Vancouver, Canada). hESCs derived hMSCs and BM-hMSCs (purchased from Lonza, Basel, Switzerland) were cultured in hMSC medium (αMEM + GlutaMAX [Thermo Fisher Scientific], 10% fetal bovine serum [Gibco, Cat: 10099–141, Lot: 1616964], 1% penicillin/streptomycin [Thermo Fisher Scientific], and 1 ng/ml bFGF [Joint Protein Central]). No mycoplasma contamination was observed during cell culture.
Antibodies used for western blotting, immunostaining and flow cytometry included anti-YAP (15407; Santa Cruz Biotechnology, Santa Cruz, CA), anti-YAP (52771; Abcam, Cambridge, MA), anti-TAZ (4883; Cell Signaling Technology, Danvers, MA), anti-GAPDH (25778; Santa Cruz Biotechnology), anti-P16 (550834; BD), anti-P21 (2947s; Cell Signaling Technology), anti-P53 (1101; Abcam), anti-β-tubulin (5274; Santa Cruz Biotechnology), anti-β-actin (69879; Santa Cruz Biotechnology), anti-Pan TEAD (13295; Cell Signaling Technology), anti-FOXD1 (PA5-27142; Thermo Fisher Scientific), anti-NANOG (21624; Abcam), anti-OCT3/4 (5279; Santa Cruz Biotechnology), anti-SOX2 (17320; Santa Cruz Biotechnology), anti-TUJ1 (T2200; Sigma, St. Louis, MO), anti-FOXA2 (8186; Cell Signaling Technology), anti-SMA (A5228; Sigma), anti-Ki67 (VP-RM04; Vector Labs, Burlingame, CA), anti-LAP2 (611000; BD), anti-HP1α (2616S; Cell Signaling Technology), anti-HP1γ (2619; Cell Signaling Technology), anti-Aggrecan (AF1220; R&D, Minneapolis, MN), anti-Osteocalcin (MAB1419; R&D), anti-FABP4 (AF3150; R&D), anti-CD73 (550741; BD Bioscience), anti-CD90 (555595; BD Bioscience), anti-CD105 (17–1057; eBioscience, San Diego, CA), anti-CD29 (303004; Biolegend, San Diego, CA), anti-CD44 (550989; BD Bioscience), anti-CD13 (301705; Biolegend), anti-CD166 (343903; Biolegend), anti-HLA-ABC (560168; BD Bioscience), anti-CD34 (555822; BD Biosciences), anti-CD43 (580198; BD Biosciences), anti-CD45 (555482; BD Biosciences), anti-CD14 (555398; BD Biosciences), anti-CD19 (555415; BD Biosciences), anti-PDPN (17-9381-41; eBioscience), and anti-CD164 (324805; Biolegend).
CRISPR/Cas9-mediated gene targeting was performed using previously described methods, with some modifications [69]. The YAP or TAZ gRNA was cloned into the gRNA vector (Addgene #41824). The donor plasmid for homologous recombination containing homology arms and a neo cassette was described previously [70]. Briefly, 5 × 106 H9 ESCs were mixed with the plasmid cocktail and electroporated. After electroporation, cells were plated on a G418-resistant MEF feeder layer. Two days later, cells were treated with 100 μg/ml G418 (Gibco, 10131027) for screening. After 2 weeks of selection, G418-resistant clones were manually picked, transferred to 96-well plates, and expanded for genotyping. Gene-targeted clones were identified using genomic PCR. gRNA sequences and primers are listed in S1 Data.
hMSCs were differentiated from hESCs as previously described [70–72]. Briefly, hESCs were dissociated into embryoid bodies and then plated on Matrigel-coated plates in differentiation medium (α-MEM + GlutaMAX [Thermo Fisher Scientific], 10% FBS [Gibco, Cat: 10099–141, Lot: 1616964], 1% penicillin/streptomycin [Thermo Fisher Scientific], 10 ng/ml bFGF, and 5 ng/ml TGFβ [HumanZyme, Chicago, IL]). After 10 days, the confluent MSC-like cells were passaged to gelatin-coated plates and sorted by FACS to purify CD73/CD90/CD105 triple-positive hMSCs, which were further characterized by flow cytometry analysis of the surface antigens, including CD166, CD29, CD44, CD13, HLA-ABC, CD34, CD43, CD45, CD14, CD19, PDPN, and CD164. The functionality of hMSCs was verified by differentiation to osteoblasts, chondrocytes, and adipocytes.
Lentiviral CRISPR/Cas9-mediated gene editing was performed as previously described [42]. Briefly, the sgRNA targeting YAP or FOXD1 was cloned into lentiCRISPRv2 vector (Addgene #52961), which contains 2 expression cassettes, hSpCas9 and the chimeric sgRNA. Then, the plasmids were packaged into lentiviruses and transduced into hMSCs; 72 hours later, transduced cells were treated with 1 μg/ml puromycin (Gibco, A1113803) for enriching transduced cells. The sgRNA sequences are listed in S1 Data.
We generated 2 lentiviral constructs to silence the expression of TEAD1, 2, 3, and 4: one containing an shRNA targeting TEAD1, TEAD3, and TEAD4 and the other containing an sgRNA targeting TEAD2 [73]. The shRNA was cloned into the PLVTHM vector (Addgene #12247), and the sgRNA was cloned into lentiCRISPRv2. We then cotransduced these lentiviral constructs into hMSCs. Seventy-two hours later, transduced cells were enriched by treatment with 1 μg/ml puromycin (Gibco, A1113803). The targeting sequences are listed in S1 Data.
For packaging of the lentivirus, HEK293T cells were cotransfected with lentiviral vectors, psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). Viral particles were collected by ultracentrifugation at 19,400 g for 2.5 hours.
For the cell cycle analysis, the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (C-10419; Molecular Probes, Eugene, OR) was used according to the manufacturer’s instructions. For ROS measurements, living cells were incubated with ROS indicators (1 μM CM-H2DCFDA, C6827; Molecular Probes). All experiments were measured with an LSRFortessa cell analyzer (BD), and data were analyzed using FlowJo software (TreeStar, Ashland, OR).
Cells were fixed with 4% paraformaldehyde for 30 minutes, washed with PBS, permeabilized with 0.4% Trion X-100 in PBS, and then blocked with 10% donkey serum (Jackson ImmunoResearch Labs, West Grove, PA). Afterwards, cells were incubated with primary antibodies in blocking solution at 4°C overnight, followed by an incubation with the corresponding secondary antibodies and Hoechst 33342 for 1 hour at room temperature.
The SA-β-gal staining of hMSCs was conducted using a previously described method [74].
For western blotting, cells were lysed in RIPA buffer containing a protease inhibitor cocktail (Roche) and quantified with a BCA kit. Generally, 20 μg of cell lysate was subjected to SDS-PAGE and electrotransferred to a PVDF membrane (Millipore, Billerica, MA). Then, the membrane was incubated with primary and HRP-conjugated secondary antibodies. Western blot data were quantified with Image Lab software for the ChemiDoc XRS system (Bio-Rad, Hercules, CA). For RT-qPCR, cellular total RNA was extracted using TRIzol (Thermo Fisher Scientific), and genomic DNA was removed with a DNA-free Kit (Ambion, Austin, TX), followed by cDNA synthesis with the GoScript Reverse Transcription System (Promega, Madison, WI). RT-qPCR was performed with qPCR Mix (TOYOBO, Tokyo, Japan) in a CFX384 Real-Time system (Bio-Rad). For genomic PCR, genomic DNA was extracted with a DNA extraction kit (TIANGEN, Beijing, China), and PCR was conducted using PrimeSTAR (TAKARA, Tokyo, Japan).
Two thousand cells were seeded in each well of a 12-well plate and then cultured until clear cell colonies formed to determine the clonal expansion abilities of hMSCs. The relative colony area was then determined by performing crystal violet staining and measured using ImageJ software.
The indicated fragments of the FOXD1 promoter were amplified by PCR and cloned into the pGL3-Basic vector (Promega). The mutant of pGL3-FOXD1 promoter 2-Luc was constructed with the Fast Mutagenesis System kit (FM111; Transgen Biotech, Beijing, China). PGL3-FOXD1 promoter 2-Luc or PGL3-FOXD1 promoter 2(mut)-Luc was transfected into hMSCs together with vectors expressing the proteins of interest and Renilla-Luc, which was used to normalize the transfection efficiency. For detection of the 8 × GTIIC-Luc activity, the 8 × GTIIC reporter (Addgene #34615) and Renilla-Luc plasmids were cotransfected into hMSCs. Cells were harvested 72 hours later using the Dual-Luciferase Reporter Assay System (Vigorous Biotechnology, Beijing, China) and assayed according to the manufacturer’s instructions.
ChIP was performed using a previously reported protocol with minor modifications [75]. Briefly, cells were cross-linked with 1% (v/v) formaldehyde for 15 minutes at room temperature, and the reaction was terminated by the addition of 125 mM glycine and an incubation for 5 minutes at room temperature. Then, cells were scraped and lysed in lysis buffer. After sonication, protein-DNA complexes were incubated with antibody-coupled Protein A beads at 4°C overnight. After elution and reverse cross-linking at 68°C, DNA was purified by phenol/chloroform extraction and ethanol precipitation and then subjected to qPCR analysis. Antibodies for ChIP included anti-YAP (14074; Cell Signaling Technology), anti-TAZ (4883; Cell Signaling Technology), anti-TEAD4 (101184; Santa Cruz Biotechnology), and normal rabbit IgG (2027; Santa Cruz Biotechnology) as a negative control.
For the teratoma analysis, 5 × 106 hESCs were subcutaneously injected into NOD-SCID mice (6 to 8 weeks, male). After 8 to 12 weeks, the tumors were excised, fixed, dehydrated, embedded in O.C.T. compound, sectioned while frozen, and analyzed by immunostaining.
For hMSC transplantation assays, 1 × 106 hMSCs transduced with a lentivirus expressing Luc were injected into the midportion of the TA muscle of nude mice (6 to 8 weeks, male). Then, 0, 1, 3, 5, and 7 days after transplantation, mice were treated with D-luciferin (GoldBio, St. Louis, MO) and imaged with an IVIS spectrum imaging system (XENOGEN, Caliper, Waltham, MA). Bioluminescence images were acquired in “auto” mode.
For RS-hMSC–induced osteoarthritis, we transplanted PBS, young hMSCs, RS hMSCs, and RS hMSCs overexpressing YAP or FOXD1 intra-articularly into the joints of NOD-SCID mice (6 to 8 weeks, male). Firstly, the mice were anaesthetized using isoflurane, and skin around the joints were shaved. For each injection, the needle was inserted beneath the middle patellar ligament, and a volume of 10 μl containing either PBS or 3 × 106 cells was injected intra-articularly. One month later, the mice were euthanized, and the joints were collected for mRNA quantification and histological assessments.
For surgically induced osteoarthritis, we performed ACLT surgery on 8-week-old male C57BL/6 mice. Animals were anaesthetized, and their hindlimbs were shaved. After the opening of the joint capsule, the anterior cruciate ligament was transected with microscissors under a surgical microscope. After irrigation with saline to remove tissue debris, the skin incision was closed. Then, 7 days later, a total volume of 10 μl of the indicated lentivirus was injected intra-articularly. At week 8, the mice were euthanized, and the joints were collected for mRNA quantification and histological assessments.
Mouse joints were fixed with 4% paraformaldehyde overnight, decalcified with 5% methanoic acid for 7 days, and embedded in paraffin. Sections (5 μm) were cut from the paraffin blocks and stained with Fast Green FCF (0.02%) and Safranin O (0.1%). Joint pathology was quantified using the OARSI scoring system [13].
For immunohistochemical staining, paraffin-embedded tissue sections were subjected to a heat-mediated antigen retrieval procedure, and then endogenous peroxidases were blocked with hydrogen peroxide. Next, tissue sections were incubated with a primary antibody overnight. Finally, the appropriate secondary antibody (ZSGB-BIO, Beijing, China) was added to the sections, which were then incubated for 30 minutes. Antigen-positive cells were visualized using the DAB Substrate kit (ZSGB-BIO). Anti-P16 antibody (54210; Abcam) and anti-flag (166355; Santa Cruz Biotechnology) were used as the primary antibodies.
First, genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen, Duesseldorf, Germany) according to the manufacturer’s instructions. DNA was sheared into fragments of approximately 300 bp using Covaris, and then the library of the fragmented DNA was constructed using the NEBNext ultra DNA Library Prep Kit for Illumina (NEB, Beverly, MA), according to the manufacturer’s protocol. The libraries were sequenced on an Illumina HiSeq 4000 platform. For CNV identification, we used the published R package HMMcopy [76]. Briefly, the genome was binned into consecutive 1 Mb windows with read Counter, and then we calculated the absolute number of reads detected in each window. We estimated the copy number with GC and mappability corrections with HMMcopy.
Total RNA was extracted from cultured human cells or mouse joints using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol. For cells, 1 × 106 cells were analyzed in biological triplicate. For mouse joints, we mixed the RNA extracted from the sample group, and then divided the sample into 3 technical replicates. One to two micrograms of total RNA was used to construct sequencing libraries using the NEBNext Ultra RNA Library Prep Kit for Illumina (NEB). The libraries were sequenced on an Illumina HiSeq 4000 platform. RNA-seq reads were aligned to the hg19 or mm10 reference genome using TopHat2 software [77]. The analysis of differentially expressed genes was performed using DESeq2 [78] based on read counts.
Promoter for TEAD-binding sites analysis was defined as 3 kb upstream and 500 bp downstream of TSS. TEAD-binding sites with P < 1.0 × 10−4 among the promoter regions were found by FIMO (http://meme-suite.org/doc/fimo.html) using the TEAD motif downloaded from JASPAR database (http://jaspardev.genereg.net/).
Results are presented as the mean ± SD. Two-tailed Student t tests were used to compare differences between treatments. P < 0.05, P < 0.01, and P < 0.001 were considered statistically significant ("*", "**", and "***", respectively).
|
10.1371/journal.pntd.0001921 | Salmonella Typhi in the Democratic Republic of the Congo: Fluoroquinolone Decreased Susceptibility on the Rise | Drug resistance of Salmonella enterica serovar Typhi (Salmonella Typhi) to first-line antibiotics is emerging in Central Africa. Although increased use of fluoroquinolones is associated with spread of resistance, Salmonella Typhi with decreased ciprofloxacin susceptibility (DCS) has rarely been reported in Central Africa.
As part of a microbiological surveillance study in the Democratic Republic of the Congo (DR Congo), Salmonella Typhi isolates from bloodstream infections were collected prospectively between 2007 and 2011. The genetic relationship of the Salmonella Typhi isolates was assessed by pulsed-field gel electrophoresis (PFGE). The antimicrobial resistance profile of the isolates was determined and mutations associated with DCS were studied. In total, 201 Salmonella Typhi isolates were collected. More than half of the Salmonella Typhi isolates originated from children and young adults aged 5–19. Thirty different PFGE profiles were identified, with 72% of the isolates showing a single profile. Multidrug resistance, DCS and azithromycin resistance were 30.3%, 15.4% and 1.0%, respectively. DCS was associated with point mutations in the gyrA gene at codons 83 and 87.
Our study describes the first report of widespread multidrug resistance and DCS among Salmonella Typhi isolates from DR Congo. Our findings highlight the need for increased microbiological diagnosis and surveillance in DR Congo, being a prerequisite for rational use of antimicrobials and the development of standard treatment guidelines.
| Typhoid fever, caused by infection with Salmonella enterica serovar Typhi (Salmonella Typhi), is an important health problem in sub-Saharan Africa. Multidrug resistance of Salmonella Typhi to the first line antibiotics is spreading and treatment of typhoid fever increasingly relies on fluoroquinolone antibiotics such as ciprofloxacin. Increased use of fluoroquinolones is however associated with spread of resistance as well. In sub-Saharan Africa, microbiological cultures to detect invasive bacterial diseases are frequently absent and the extent of the problem is poorly known. In the present study, 201 Salmonella Typhi isolates were collected between 2007 and 2011 in DR Congo, mainly from children and young adults. For the first time, widespread Salmonella Typhi multidrug resistance (30.3%) and decreased ciprofloxacin susceptibility (15.4%) is described in Central Africa. Decreased ciprofloxacin susceptibility was associated with point mutations in the quinolone resistance determining region of the gyrA gene. Resistance to azithromycin, an alternative for treatment of uncomplicated typhoid fever in the case of decreased ciprofloxacin susceptibility, was still rare (1.0%). Our findings demonstrate emergence of multidrug resistance and fluoroquinolone decreased susceptibility in DR Congo, and highlight the need for increased microbiological diagnosis and surveillance, being a prerequisite for rational use of antimicrobials and the development of standard treatment guidelines.
| Typhoid fever is endemic in the Democratic Republic of the Congo (DR Congo). Although antimicrobial resistance data are sparse for Central Africa, drug resistance to first-line antibiotics is clearly emerging [1]. Also in DR Congo, multidrug resistance (MDR) [defined as co-resistance to first-line antibiotics ampicillin, chloramphenicol and trimethoprim/sulphamethoxazole (TMP-SMX)] in Salmonella enterica serotype Typhi (Salmonella Typhi) has been observed [2], [3]. Facing widespread MDR, fluoroquinolones have become the drugs of choice for treating typhoid fever, but their increased use has been associated with a spread in low-level fluoroquinolone resistance - further referred to as decreased ciprofloxacin susceptibility (DCS) [4]. Salmonella Typhi with DCS is common in Asia and has been reported in East Africa and South Africa [5]–[7], but apart from a single case reported from Cameroon [8], and two cases in DR Congo, Salmonella Typhi with DCS has not yet been observed in the central African region [2], [3], [9].
The present study describes the antimicrobial resistance profile of a prospective collection of Salmonella Typhi isolates recovered as part of a microbiological surveillance study from blood cultures obtained from patients in DR Congo over the years 2007–2011. Mutations associated with DCS were studied and the genetic relationship of the Salmonella Typhi isolates was assessed.
Ethical approval was granted by the Ethical Committee of the University of Antwerp, Belgium and from the Ministry of Health in DR Congo. The present study complies with the World Health Organization and international guidelines (European Society of Clinical Microbiology and Infectious Diseases Study Group for Antimicrobial Resistance Surveillance and Clinical Laboratory Standards Institute) on antibiotic surveillance for which no recommendation for an informed consent has been issued. The diagnostic procedure – blood cultures – is part of the standard diagnostic work-up of patients with a suspicion of bacteremia. Clinical information -as presented- and information about use of antibiotics was the standard information present on the laboratory request form. Data have been reviewed and analyzed anonymously.
Between April 2007 and January 2011, blood cultures were performed on 9,634 patients suspected of typhoid fever or other systemic infections. Patients were seen at health care facilities in seven out of 11 provinces in DR Congo: Kinshasa, Bas-Congo, Bandundu, Equateur, Kasai Occidental, Kasai Oriental and Oriental province (Figure 1). In Kinshasa, health care facilities involved in the detection and study of the epidemic increase of typhoid fever-associated peritonitis of 2004 were selected [3]. Health care facilities in other provinces were recruited based on the past or actual existence of microbiological laboratories, professional contacts and the accessibility to reliable shipment facilities. Criteria for blood culture sampling were clinical suspicion of bacteremia associated with a local (pneumonia, urinary tract infection, meningitis or other) or systemic (typhoid fever, endocarditis) infection diagnosed at consultation or admission. Typhoid fever was defined according to the case definitions of the Ministry of Health surveillance of communicable diseases [10]. At the start of the surveillance project, teams of clinicians and laboratory technicians were trained in indications and sampling of blood cultures. For children <14 years, 1–4 ml of blood was inoculated into a pediatric blood culture vial (Bact/ALERT FP; bioMérieux; Marcy L'Etoile; France). For adults, 2×10 ml of blood was inoculated into aerobic blood culture vials (Bact/ALERT FA; bioMérieux; Marcy L'Etoile; France). Age, gender, geographic origin, use of antibiotics prior to blood culture sampling and presumptive diagnosis (focus of bacteremia including suspicion of typhoid fever) were recorded. Inoculated vials were shipped to the Institut National de Recherche Biomédicale (INRB) in Kinshasa, incubated at 35°C and daily checked for growth by visual inspection of the chromatographic growth indicator. Grown cultures were Gram stained, subcultured and identified to the species level. Skin or environmental bacteria (coagulase negative staphylococci, Corynebacterium spp., Propionibacterium acnes and Bacillus spp.) were categorized as contaminants, the other bacteria were considered as clinically significant organisms [11]. Suspected colonies of Salmonella were identified as Salmonella Typhi using standard biochemical methods (characteristic aspect on Kligler Iron Agar (acid from glucose, no gas, trace of H2S), negative tests for urease, oxidase, β-galactosidase and indole production tests, positive tests for lysine decarboxylase) and the serotype of Salmonella Typhi (O:9;H:d;Vi+) was confirmed with commercial antisera (Remel, Lenexa, Kansas). Identity of Salmonella species isolates was confirmed using the Vitek II system (Card GN21 341, bioMérieux). At the National Reference Laboratory for Salmonella and Shigella (Institute of Public Health, Brussels), the serotype of the Salmonella isolates was re-confirmed by slide agglutination with commercial monospecific antisera (Sifin, Berlin, Germany), following the Kauffmann-White scheme [12]. For analysis in the present study, only the first isolate per patient was considered.
Susceptibility tests for ampicillin, cefotaxime and TMP-SMX were performed using the Vitek II system (Card AB AST-N156, bioMérieux). For nalidixic acid, ciprofloxacin, chloramphenicol and azithromycin, minimal inhibitory concentration (MIC) values were determined using the E-test macromethod (bioMérieux). Breakpoints for resistance were respectively ≥32 mg/l for ampicillin, ≥4 mg/l for cefotaxime, ≥32 mg/l for naladixic acid, ≥16 mg/l chloramphenicol (considering intermediate susceptible isolates resistant) and ≥4/76 mg/l for TMP-SMX using Clinical and Laboratory Standards Institute definitions [13]. DCS was defined according to European Committee on Antimicrobial Susceptibility testing (EUCAST) V 2.0. guidelines, i.e. a MIC-value for ciprofloxacin >0.064 mg/l [14]. For azithromycin, EUCAST V.2.0. suggests MICs of >16 mg/l as defining resistance [14]. Multidrug resistance was defined as co-resistance to ampicillin, chloramphenicol and TMP-SMX.
Screening for mutations causing DCS was performed by amplification and sequencing of the quinolone resistance-determining regions (QRDRs) of the gyrA, gyrB, and parC genes. The presence of the plasmid-mediated quinolone resistance qnr genes (qnrA, qnrB, and qnrS) was determined using PCR [15].
Pulsed-field gel electrophoresis (PFGE) was performed according to the PulseNet protocol for molecular subtyping of Salmonella [16], using XbaI as restriction enzyme (New England Biolabs, Leusden, The Netherlands). For cluster analysis, Bionumerics 5.1 was used (Applied Maths NV, Sint-Martens-Latem, Belgium), with as comparison settings the Dice similarity coefficient and UPMGA dendrogram type (optimization 0.50%, position tolerance 1.50%). PFGE profiles obtained were compared to PFGE of Salmonella Typhi profiles stored: In 1° the Institute of Public Health (Brussels) database, originating from Belgium (n = 27), Morocco (n = 8), Egypt (n = 1), Burkina Faso (n = 1), Niger (n = 1), Cambodia (n = 20), India (n = 11), Pakistan (n = 7), Bangladesh (n = 3), Indonesia (n = 1), Sri Lanka (n = 1), and Thailand (n = 1); 2° the PFGE database of the Centre for Enteric Diseases (CED) of the National Institute for Communicable Diseases in South Africa, containing patterns for 730 Salmonella Typhi isolates.
All data were entered in an Excel database (Microsoft Corporation, Redmond, Washington, USA). Proportions were assessed for statistical significance using the Chi square test, considering p<0.05 as significant (Stata 10, StatCorp, Texas, USA).
From the 9634 blood cultures performed, 989 (10.3%) clinically significant organisms were grown, including 201 isolates of Salmonella Typhi, representing respectively 2.1% of all blood cultures and 20.3% of all clinically relevant organisms. There were no annual or seasonal differences in isolation rates. The geographic site of cases, positive on blood cultures for Salmonella Typhi in DR Congo, is shown in figure 1. Among the 201 Salmonella Typhi blood cultures isolates, 110 (54.7%) were recovered from Bas-Congo and 67 (33.3%) from Kinshasa. In Kinshasa, the isolation rate of Salmonella Typhi (67/5465, 1.2%) was lower compared to all other provinces in DR Congo (134/4064, 3.4%, p<0.001). The median age of patients infected with Salmonella Typhi was 15 years (interquartile range 8–25), but infection of young children was also common (Table 1). Over half of blood cultures were in children <10 years, yet 32.8% of the Salmonella Typhi isolates recovered were from children in this age group. The most affected age group were persons aged 10–19 years, in whom nearly 60% of the organisms isolated were Salmonella Typhi. In addition, this age group contributed to only approximately 10% of the blood culture samples, yet accounted for 30% of all Salmonella Typhi isolates recovered.
In patients with clinically significant organisms, presumptive diagnosis of typhoid fever was made in 53.0% (524/989) of the cases. In the 201 patients from whom Salmonella Typhi was cultured, presumptive diagnosis of typhoid fever at the moment of sampling was made in 80.6% (162/201) of the cases. A total of 21.0% (34/162) of these patients suffered from abdominal distension and/or gastro-intestinal bleeding and were classified as complicated typhoid fever. Other presumptive diagnoses (for several patients more than one presumptive diagnosis was mentioned) included complicated urinary tract infection (14.4%), pneumonia (7.0%), meningitis (2.0%), malaria (5%) and other non-specified causes of bacteremia (16.4%); for three patients (1.5%), no data were available. Nearly half (93/201, 46.3%) of the patients had received antibiotics within 48 hours prior to sampling of blood cultures, mostly first-line antibiotics.
Resistance against ampicillin, chloramphenicol or TMP-SMX were observed in 64.7% (130/201), 41.3% (83/201) and 57.7% (116/201) of isolates, respectively. MDR and DCS were observed in 30.3% (61/201) and 15.4% (31/201) of isolates respectively; combined MDR and DCS occurred in 7.5% (15/201) of isolates. Isolates with DCS corresponded to nalidixic acid resistant isolates and vice versa. Only two (1.0%) isolates had azithromycin MIC values exceeding 16 mg/l (i.e. 24 mg/l); both isolates were also resistant against ampicillin and TMP-SMX (one also combined with DCS and nalidixic acid resistance). No cefotaxime resistance was observed. Fifty-one isolates (25.4%) were fully susceptible to all three first-line antibiotics and 50 (24.9%) were susceptible to all seven antibiotics tested. There was no apparent relationship between antimicrobial resistance and patient age, year of isolation, province of isolation or administration of antibiotics prior to isolation.
All 31 isolates with DCS were analysed for mutations in the QRDRs of the gyrA, gyrB, and parC genes. No mutations were detected in gyrB or parC genes. No qnrA or qnrB genes were detected. For one isolate, the qnrS gene was detected. Apart from a mutation at codon 133 in the gyrA gene (conferring a Glu to Gly change in the GyrA protein), which was also present in nalidixic acid susceptible strains, all 31 isolates had DCS-associated mutations in the gyrA gene, conferring the following amino acid mutations in the GyrA protein:(i) Ser83 changed into Phe or Tyr (n = 22), or (ii) Asp87 changed into Gly, Tyr or Asn (n = 9).
Among 185 isolates tested for PFGE, 30 different profiles were observed (Supplemental file). In 132/185 isolates (71.4%), an indistinguishable PFGE profile occurred. This profile was the main or single PFGE profile over time and geography, although its proportion was lower (p = 0.02) in Kinshasa province (39/64 isolates, 60.9%) compared to the other provinces in DR Congo (93/121, 76.9%). In Kinshasa, the highest variation of profiles - in total 22 - was noted. Of the isolates showing the main/predominant profile, 31.1% (41/132) were MDR and 17.4% (23/132) were DCS, while 11 (8.3%) showed a combined MDR and DCS. Comparison of the PFGE profiles of Salmonella Typhi from DR Congo with isolates from other geographic origins revealed one PFGE profile from a Salmonella Typhi isolate from DR Congo that was indistinguishable from a PFGE profile from an isolate recovered in Belgium, and 5 Congolese PFGE profiles that were indistinguishable from PFGE profiles of 9 Salmonella Typhi isolates recovered in South Africa.
The present study demonstrated widespread MDR and DCS among Salmonella Typhi isolates from DR Congo. DCS resistance was associated with point mutations in the gyrA gene. Azithromycin resistance was rare. Nearly 75% of isolates were associated with a single PFGE profile. Our data indicate that infection with Salmonella Typhi is highly clinically relevant for school-aged children and young adults.
Our study has some limitations. Despite our attempt to survey large parts of DR Congo, the geographical origin of the isolates could have been biased by logistic difficulties. Connections and communications by road are very limited in DR Congo and except from the Bas-Congo and Kinshasa provinces, samples had to be shipped and transported by air plane. Apart from shipment delays, antibiotic use prior to sampling may have affected culture yields and biased the resistance profiling towards resistant organisms. Further, our study did not allow calculation of incidence rates because referral patterns and catchment populations among participating centers differed; for instance financial constraints prevent many patients from consulting. Medical doctors are not familiar with microbiological culture tools; they rely on the clinical picture or the Widal test for diagnosis of typhoid fever [10]. There were undoubtedly differences in application and interpretation of the clinical criteria as well variations in sampling intensity. Drug efflux was not examined in DCS strains, so we cannot exclude the possibility that it played a role in the observed fluoroquinolone resistance. Finally, PFGE as a genotyping method has several limitations, including insensitivity when dealing with likely clonally related strains. As a consequence, single nucleotide polymorphism differences that may be important in identifying subtypes among closely related Salmonella Typhi strains may have been missed [17]. On the other hand, despite these limitations, the yield of clinically significant organisms was in line with international standards and similar for all settings and over time. In addition, we were able to document antimicrobial resistance patterns at a large scale in the central African region where the typhoid fever situation is notoriously poorly characterized [4].
Salmonella Typhi infection appears to prevail among school-aged children and young adults. Of note, in the present study, Salmonella Typhi was also isolated from children under 10 years of age at a positivity rate of 32.8%. This value is higher than previously reported from the Kivu province (not included in our current study) in Eastern DR Congo where a previous study reported a positivity rate of 7.1% amongst children of this age group, amongst a total of 409 Salmonella Typhi cultures sampled during the years 2002–2006) [2]. Patients' age distribution in the present study compares with that observed in countries with medium to high incidence rates of typhoid fever [18], and was similar to a Ugandan report [9] although different from data obtained in South Africa and Malawi, where young children seem relatively less affected [5], [19]. This has been highlighted by Bhutta and contrasted with south Asian countries [20], although some geographic differences may exist [19]. In urban Kinshasa, the isolation rate of Salmonella Typhi appeared to be lower compared to the other (more rural) provinces in DR Congo, which seems in contrast to observations in Kenya [21]. Although this seems to suggest that incidence might be lower in urban than in rural areas, one should be careful interpreting these results since referral patterns and catchment populations may have differed, as well as the type of sampling sites. In the provinces, sampling sites were uniquely referral hospitals, whereas in Kinshasa, private clinics were also included. Suspicion of meningitis in a minority of patients, suggests that neurological symptoms might have been present [5].
Although commonly described in our present study, MDR to first line antibiotics, was less prevalent in the current isolate collection compared to previous reports of MDR in Salmonella Typhi from DR Congo [2], [3], for Nigeria [22], Kenya [6], [21], and Malawi and Mozambique [5]. Although discriminative power of PFGE using one restriction enzyme might have been insufficient [23], PFGE suggests limited genotypic diversity with a single Salmonella Typhi clone prevailing in DR Congo, while in Kinshasa a more diverse panel of profiles was observed. As observed in Kenya, MDR and sensitive clones seem to share similar PFGE patterns [6]. The Salmonella Typhi PFGE patterns from DR Congo have been submitted to the Global PulseNet Salmonella Typhi Database (www.pulsenetinternational.org), for consideration to be included in to the database in order to allow for international comparison of the Salmonella Typhi isolates. A limited comparison of the Congolese PFGE patterns with patterns of Salmonella Typhi from other origins revealed six patterns that were indistinguishable from Belgian and South African PFGE patterns. Exchange of Salmonella isolates between Belgium and DR Congo seems plausible in view of the historical links and the frequent travelling between the two countries. It was also not unexpected to find DR Congo pulsotypes indistinguishable from South African pulsotypes, as there is much migration of DR Congo nationals into South Africa.
For the first time in Central-Africa, widespread DCS related to the presence of gyrA mutations was reported. DCS may lead to treatment failure [24] and points to emerging fluoroquinolone resistance. The DCS rate was similar to the rate described in Nigeria, with comparable MIC50 and MIC90 values [22]. It appeared higher than in Kenya [6], [21], Malawi and Mozambique [5], South-Africa [7] and Uganda [9] although these studies might have applied older CLSI guidelines. Fluoroquinolone resistant isolates showed mutations in gyrA which conferred amino acid mutation at codons 83 and 87 in the GyrA protein; these mutations were similar to previous observations in South Africa [7]. The fact that DCS was invariably predicted by nalidixic acid resistance should be further explored; in particular future studies should be carried out to examine the predictive value of screening for DCS by simple nalidixic acid disk diffusion test [25]. As there is limited human migration between provinces (some of the presently studied provinces are only accessible by air or river), the similar rates of DCS over all provinces may suggest spontaneous and independent mutations induced by selection pressure rather than a single resistant clone spreading throughout the country. Indeed, fluoroquinolone antibiotics have been increasingly used in DR Congo since the MDR Salmonella Typhi outbreaks in 2004 [3]. Azithromycin may be a valuable alternative for treatment of uncomplicated typhoid fever in the case of DCS [4], and there were low levels of resistance to this antibiotic. However, as its patent has expired recently, it can be expected that the market in resource limited settings will be flooded by newer cheaper generics, with the danger of indiscriminate use resulting in the emergence and spread of azithromycin resistance. Indeed, from site visits we observed that azithromycin is now increasingly been introduced and promoted as an oral antibiotic for many indications.
The present results provide an early warning sign for the emerging resistance of a bacterial key-pathogen to affordable antibiotics in DR Congo. Together with recent data we described about the occurrence of MDR bacteria in drinking water in Kinshasa city [26], they provide evidence of a serious problem of antibiotic resistance in the community setting of DR Congo. The need for microbiological diagnosis and surveillance is highlighted. Surveillance not only timely detects outbreaks but is also a prerequisite for rational use of antimicrobials and the development of standard treatment guidelines [4], [27] which in turn are needed to contain antibiotic resistance.
This study was funded by Directorate General of Development Cooperation of the Belgian Government through Institutional Collaboration INRB-ITM (Network Program on Laboratory Quality Management; Project 3.21). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The authors declare that they have no conflicting interests in relation to this work.
|
10.1371/journal.ppat.1005451 | Dengue Virus Nonstructural Protein 5 (NS5) Assembles into a Dimer with a Unique Methyltransferase and Polymerase Interface | Flavivirus nonstructural protein 5 (NS5) consists of methyltransferase (MTase) and RNA-dependent RNA polymerase (RdRp) domains, which catalyze 5’-RNA capping/methylation and RNA synthesis, respectively, during viral genome replication. Although the crystal structure of flavivirus NS5 is known, no data about the quaternary organization of the functional enzyme are available. We report the crystal structure of dengue virus full-length NS5, where eight molecules of NS5 are arranged as four independent dimers in the crystallographic asymmetric unit. The relative orientation of each monomer within the dimer, as well as the orientations of the MTase and RdRp domains within each monomer, is conserved, suggesting that these structural arrangements represent the biologically relevant conformation and assembly of this multi-functional enzyme. Essential interactions between MTase and RdRp domains are maintained in the NS5 dimer via inter-molecular interactions, providing evidence that flavivirus NS5 can adopt multiple conformations while preserving necessary interactions between the MTase and RdRp domains. Furthermore, many NS5 residues that reduce viral replication are located at either the inter-domain interface within a monomer or at the inter-molecular interface within the dimer. Hence the X-ray structure of NS5 presented here suggests that MTase and RdRp activities could be coordinated as a dimer during viral genome replication.
| Many plus-strand RNA viruses encode a viral RNA polymerase and capping enzymes to synthesize a 5’-capped RNA genome. However, how these two activities are coordinated during viral replication is not understood. In flaviviruses, polymerase and capping enzymes are encoded in a single multifunctional protein, where separate domains within the polypeptide are responsible for these activities; flavivirus NS5, composed of the polymerase and methyltransferase domains, carries out viral RNA synthesis, 5’-RNA capping, and RNA cap methylations. Previous NS5 monomer structures were unable to provide mechanistic insight into how the two domains communicate or the quaternary organization of the functional enzyme. We have determined the crystal structure of dengue virus NS5 and show that the NS5 dimer is likely the biological assembly of NS5, and RNA synthesis and RNA capping may be coordinated by the dimer. We found that essential interactions between the two NS5 domains can be maintained either within a monomer or via inter-molecular interactions within a dimer, and thus NS5 can adopt multiple conformations while preserving necessary interactions between the methyltransferase and polymerase domains. Using dengue virus, we additionally determined that such specific interaction between the two NS5 domains is the major determinant of viral replication.
| Flaviviridae includes at least 70 mosquito- and tick-borne viral species, some of which act as the etiological agents of human diseases such as dengue virus (DENV), West Nile virus (WNV), Japanese encephalitis virus (JEV), and yellow fever virus (YFV). Dengue is the most prevalent mosquito-borne virus, with nearly 400 million annual cases worldwide [1]. Infection with one of four dengue virus serotypes (DENV1-4) can lead to febrile illness and flu-like symptoms, or can progress to the more severe dengue hemorrhagic fever or dengue shock syndrome. Development of the more severe hemorrhagic fever is more likely if recovery from infection by one dengue serotype is followed by subsequent infection by a second serotype [2]. There are currently no effective antiviral drugs for the treatment of flavivirus infections, and no vaccine is available for protection against severe diseases caused by any of the DENV serotypes or WNV [3].
The 10–11 kb positive (+) sense flaviviral genome consists of a single open reading frame (ORF) flanked by 5’ and 3’ untranslated regions (UTR). The 5’ terminal nucleotide A is modified by the addition of a type 1 cap (N7MeG5’-ppp-5’A2’OMe). The viral ORF is translated into a single polyprotein that is co- and post-translationally cleaved by host and viral proteases to produce ten proteins, including three structural proteins (capsid, pre-membrane, and envelope) and seven nonstructural (NS) proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) [4]. All NS proteins, along with the viral RNA and cellular factors, form a viral replication complex embedded within virus-induced membrane vesicles located at the rough endoplasmic reticulum of infected cells [5–8]. NS3 and NS5 are the key enzymes in the replication complex, as together they account for all catalytic activities required for genome replication and RNA capping at the 5’ UTR. NS3 consists of an N-terminal serine protease domain, which requires NS2B as a cofactor, and a C-terminal helicase domain possessing three distinct activities: RNA helicase, nucleoside triphosphatase, and 5’ RNA triphosphatase [4]. NS5, the largest NS protein at 103 kDa, consists of an N-terminal methyltransferase (MTase) domain possessing three activities necessary for cap synthesis (guanylyltransferase, guanine-N7-methyltransferase, and nucleoside-2’O-methyltransferase) and a C-terminal RNA-dependent RNA polymerase (RdRp) domain that carries out de novo RNA synthesis [9–14]. Replication of the (+) strand RNA genome is an asymmetric and semi-conservative process wherein the antigenome is present only as a double-stranded (ds) RNA replication intermediate [15]. After (+) strand RNA synthesis by the NS5 RdRp, the cap structure is added to the 5’ end of the genome by four enzymatic reactions catalyzed by NS3 and NS5 [16]. First, the 5’ γ-phosphate of the (+) RNA is cleaved by NS3 5’-RNA triphosphatase activity to produce a diphosphate-terminated RNA (pp-5’A). Next, the NS5 MTase catalyzes the transfer of a GMP moiety from GTP to the RNA (G5’-ppp-5’A). Finally, NS5 MTase catalyzes sequential guanine-N7- and nucleoside-2’O-methylations using S-adenosyl-L-methionine (SAM) as a methyl donor and produces a type 1 cap structure (N7MeG5’-ppp-5’A2’OMe) [16,17].
The physical linkage of two distinct catalytic domains within NS5 suggests that RNA synthesis and genome capping activities may be coupled during viral replication [18]. Communication between the MTase and RdRp domains has indeed been observed in several flaviviruses via reverse genetic, biochemical, and structural experiments [19–24]. Here we present the crystal structure of full-length DENV3 NS5. A total of eight copies of full-length NS5 are arranged as four dimers within the crystallographic asymmetric unit (ASU). All eight molecules have essentially the same relative orientation of the MTase and RdRp domains, and this arrangement differs from that seen in JEV NS5 [25]. In the structure presented here, the linker region between the MTase and RdRp domains is fully resolved, and comparison with the JEV NS5 structure suggests that this region acts as a hinge that allows NS5 to adopt multiple conformations. Furthermore, many residues that reduce viral replication are located at the inter-domain and inter-molecular interfaces, and thus the domain-domain interactions within a monomer as well as monomer-monomer interactions within a dimer likely play a role during viral genome replication. The DENV NS5 structure will aid the development of structure-based inhibitors that would interfere with inter-domain or inter-molecular interactions.
The crystallographic ASU contains eight copies of full-length DENV3 NS5, in which four sets of dimers (AB, CD, EF, and GH) are arranged in a saddle-like shape (Fig 1). The AB and CD dimers are related to the GH and EF dimers, respectively, by a 2-fold non-crystallographic symmetry (NCS) axis. Superimposition of the eight NS5 monomers and four NS5 dimers within the ASU yields rmsd values ranging from 0.4 to 1.9 Å, and from 0.8 to 2.3 Å, respectively, indicating that all copies of NS5 are nearly superimposable with one another. Thus the relative arrangement of MTase and RdRp domains within a monomer and the arrangement of monomers within a dimer are conserved in all eight molecules. These eight molecules within the ASU are subject to entirely different packing forces, hence the observed arrangement is not an artifact of crystallization (S1 Fig).
Individual MTase and RdRp domains in full-length NS5 have conserved protein folds. The MTase (residues 1–262) adopts the SAM-dependent methyltransferase fold composed of four helices surrounding a central 7-stranded β-sheet (Fig 2A), similar to previously determined flavivirus MTase structures [13]. The active site, containing a catalytic K61-D146-K180-E216 (KDKE) motif, is positioned in the center of the β-sheet. The MTase core fold is surrounded by N- and C-terminal extensions that interact with each other. Although no SAM or S-adenosyl-L-homocysteine (SAH, a byproduct of the methyltransferase reaction) were added during purification or crystallization, SAH is clearly visible in the electron density in the active site (S2 Fig). The C-terminal RdRp (residues 273–900) adopts the canonical right hand polymerase fold, consisting of fingers, palm, and thumb subdomains (Fig 2A). The palm subdomain contains a catalytic G662-D663-D664 (GDD) metal-binding motif. Flavivirus RdRps initiate RNA synthesis de novo, without the need of a primer. The priming loop in the thumb subdomain is proposed to stabilize the de novo initiation complex and occlude access to the template-binding channel [26]. During elongation, however, the priming loop is proposed to open to allow exit of dsRNA products, requiring significant movement. The NS5 construct used to solve the crystal structure (NS5-Δ6) contained a six-residue deletion (795WSIHAH800) in this priming loop, which was rationally designed to eliminate the need for this movement. This protein has higher polymerase activity than the wild-type (WT) protein when a subgenomic RNA is used as a template [27] (Fig 2B). Thus the structure represents a biologically relevant form of the polymerase. The shortened priming loop of the protein follows continuous density in the electron density map, suggesting that the deletion did not significantly affect the protein fold.
Three areas of the DENV RdRp had been unresolved in the previous crystal structures [28–30]. While this manuscript was in preparation, a DENV NS5 structure was published [31], and thus was included in comparison. First, the region containing motif G (residues 408–417) in the fingers subdomain is disordered. Motif G is proposed to form a part of the template-binding channel and regulate RNA template binding and translocation [25]. Three monomers (B, C, and E) out of the eight copies in the DENV NS5 ASU have connected density for a long loop near the entrance to the template-binding channel (Fig 2C). This is the first time that motif G has been fully resolved in the DENV RdRp structures. The conformation of the loop is similar to that observed in JEV NS5 [25] (Fig 2C). Second, the fingertip region formed by residues 454 to 469 (motif F) is disordered in all eight copies. Motif F is well ordered in JEV NS5 and forms part of the template-binding channel near the MTase-RdRp interface (Fig 2C). Unexpectedly, when JEV and DENV RdRp domains are overlaid, motif F in JEV NS5 sterically clashes with the MTase in the DENV NS5 structure due to differences in the MTase and RdRp interface (see below). This suggests that motif F adopts several conformations depending on the interactions between MTase and RdRp domains. Finally, the C-termini (residues 884–899) of four monomers (A, D, F and G) were clearly resolved for the first time. The C-terminus forms an α-helix and interacts with the MTase of a neighboring NS5 (Figs 2C and 3B). Involvement of the C-terminus is further discussed in the intermolecular interactions below.
When the eight RdRp domains are overlaid, the biggest difference is the orientation of the thumb subdomain relative to the palm and fingers subdomains (Fig 2E). For example, the thumb subdomain in monomer B is rotated ~5° around a hinge point at residues G699 and W700 relative to the other monomers. This results in a 6 Å movement of an α-helix in the thumb, as calculated by the program DynDom [32]. When the RdRp domains were further compared to other flavivirus RdRp structures, movement of the thumb subdomain relative to the palm and fingers subdomains ranges up to 13 Å (Fig 2E). Such flexibility of the thumb subdomain could control the size of the template-binding channel to accommodate either an ssRNA template or a dsRNA product during RNA synthesis.
The unusually large number of molecules in the crystallographic ASU allows us to examine interactions of NS5 domains without the influence of crystal contacts. Although the individual NS5 molecules are not restricted by crystal lattice contacts, all eight copies of NS5 have the same relative orientation of the MTase and RdRp domains. Looking at the canonical right hand orientation of the RdRp, the MTase is located behind the RdRp near the template entry channel, opposite the priming loop (Fig 2A). In this arrangement of two domains, the active site of the MTase and the dsRNA exit of the RdRp face in opposite directions, exposed to solvent. The MTase interacts mainly with the fingers subdomain of the RdRp, and buries ~1000 Å2 of surface area as calculated by the program PISA [33]. Two areas of NS5 contribute to the inter-domain interface—the linker between the MTase and RdRp domains, and the interface involving two helices on the periphery of the MTase core and the backside of the fingers subdomain (Figs 2A and S3).
The MTase and RdRp domains are connected by a 10-residue linker (residues 263–272). The linker has very low sequence identity among flavivirus NS5 proteins (one identical residue out of 10 residues), and was structurally unresolved in both the crystal structures of individual MTase and RdRp domains, as well as in the full-length JEV NS5 structure [25,30]. Hence the linker was predicted to be flexible. In the current structure, the linker is well ordered, and has complete density connecting the MTase and RdRp domains (Fig 2D). The linker is in an extended conformation and makes extensive interactions with each domain. R262 and E267 form hydrogen bonds with the MTase core (A92, L94, K95, V97, N112, and Y119) (Fig 2D). Residues P268, E269, and N272 interact primarily with the fingers subdomain of the RdRp (R361, K595, and N574) (Fig 2D). The average temperature factor for the linker ranges from 80 to 127 Å2 amongst the eight monomers, and is similar to the temperature factor of the neighboring residues in each monomer. The interactions between the linker and NS5 domains support previous observations that incorporation of the linker (residues 263–272) into the DENV1-4 RdRp domains improves thermal stability compared to the RdRp domain alone [30].
Adjacent to the linker region, there are additional molecular contacts at the MTase and RdRp interface. These include a cation-pi interaction between R68 and F348, and salt bridges between E67 and R352, and between E252 and R352 (S3 Fig). Hydrogen bonding interactions are also observed between Q63 and R352, G93 and K300, and D256 and E356. Most of the residues involved in the domain-domain interactions are conserved, and thus the MTase and RdRp interactions are likely preserved in DENV NS5 from all four serotypes. Several residues located in the MTase and RdRp domain interface have previously been mutated and shown to reduce viral replication. For example, virus containing the double mutation K356A/E357A in DENV4 NS5 (corresponding to K355/E356 in DENV3 NS5) reduced viral replication 100-fold relative to the WT virus [22,34]. Similarly, no virus was detected after transfection of DENV2 RNA transcripts containing R353A/K358A (R352/K357 in DENV3) or E357A/K358A/D360A (E356/K357/D359 in DENV3) in NS5 [34]. A F349D substitution in DENV2 NS5 (F348 in DENV3 NS5) also severely impaired viral replication [22]. Thus the MTase and RdRp interactions shown in the crystal structure are likely important for viral replication.
Flavivirus NS5 is proposed to adopt multiple conformations from compact to more extended forms based on small-angle X-ray scattering (SAXS) [35]. Crystal structures of JEV and recently solved DENV NS5 (PDB entry 4V0Q) indeed show different arrangements of MTase and RdRp domains along with different linker conformations [25,31]. It was thus interesting that the arrangement of MTase and RdRp domains in our DENV NS5 structure was similar to those in the 4V0Q structure. Despite the different crystallization conditions (1.0 M succinic acid and 1% monomethylether 2000 in our condition vs. 0.2 M calcium acetate and 10–20% PEG 8000 in the 4V0Q structure) and symmetry of crystals (P3221 with 8 molecules in ASU vs. P21212 with one molecule in ASU), the two DENV NS5 structures can be superimposed with rmsd values of 1.0–2.5 Å for each of the eight NS5 monomers. The 10-residue linkers, including a short 310 helix, are also similar and they can be superimposed with rmsd values of 0.4–0.6 Å (Fig 2C). We next compared the NS5 crystal structure to the SAXS profile, which showed an average radius of gyration (RG) and maximum dimension (Dmax) of 34 and 125 Å, respectively. The RG and Dmax of both DENV NS5 structures are calculated to be 31 and 98 Å, respectively, using the program CRYSOL [36]. Thus, the crystal structure would correspond to a compact form of NS5 [35]. Since both structures have the same domain arrangements within the monomer despite the protein’s ability to adopt multiple forms in solution, the compact form of NS5 may be a preferred structure at high protein concentration.
The eight molecules of NS5 in the crystallographic ASU consist of four dimers (AB, CD, EF, and GH in Fig 1), in which the relative arrangement of both monomers is fixed (type I dimer). The four dimers can be superimposed with rmsd values ranging from 0.8 Å (between AB and GH dimers) to 2.3 Å (between EF and GH dimers). The periphery of the core MTase domain in one NS5 molecule interacts with the base of the palm subdomain in the second NS5 molecule, and buries 1100–1200 Å2 of surface area between monomers in each pair (Fig 3A). This extensive intermolecular dimer interaction even includes two hydrogen bonds between main chain atoms of S128 in the MTase of one monomer and G526 in the RdRp of the neighboring monomer as in a short β-sheet (Fig 3A). MTase residues E112, S117, and V124 in one monomer interact with RdRp residues H701, N676, and R673 in the neighboring monomer, respectively. P113 forms an additional hydrogen bond with H701. The type I dimer is also observed in the recently solved DENV NS5 structure (PDB code 4V0Q) [31]. Although the NS5 structure was reported to be a monomer, i.e., one molecule in the ASU, the structure shows a crystallographic dimer that is nearly identical to our type I dimer. The crystallographic dimer can be superimposed with the type I dimer with an rmsd of 2.0 to 3.1 Å (for 1599–1671 Cα carbons). As mentioned previously, crystallization conditions and the symmetry of the crystals are significantly different between the two crystal forms. Crystal contact areas of the five dimers (four dimers in the current structure and the crystallographic 4V0Q dimer) were determined using the program CONTACT with a 4 Å cutoff. The crystal contacts were also quite different for each of the five dimers (S1 Fig). Thus, assembly of DENV NS5 into the type I dimer is not an artifact of crystallization.
Because of the arrangement of dimers within the 8 copies of NS5 in the ASU, there is a second type of dimer interaction (type II dimer) only observed between monomers A and F, and between D and G (Fig 3B). In the type II dimer, the tip of the thumb subdomain of one NS5 molecule (where the fingertips and the thumb domain meet) interacts with the same area in the neighboring molecule. This interface includes mostly hydrophobic interactions and buries ~1600 Å2 of surface area. Interestingly, only the four NS5 monomers in type II dimers (A, D, F, and G) have the extended density for the C-terminus (residues 884–899), contributing to the highly buried surface area. The C-terminus of each of the four monomers forms an α-helix and is located near the C-terminal α-helix (residues 228–243) in the neighboring MTase (Figs 2C and 3B). The C-terminus of flavivirus NS5 is thought to be flexible because the amino acid sequences are less conserved (< 25% identity), and it has not been observed in any crystal structures. The current structure indicates that the C-terminus of NS5 can be ordered and mediate interactions with other NS5 molecules.
The two types of NS5 dimer interactions with significant interface areas (>1,000 Å2) suggest that NS5 may function as a dimer in the viral replication complex. Although recombinant NS5 is a monomer in solution [35,37], cellular interactions between NS5 molecules have been shown using pull-down and fluorescence resonance energy transfer (FRET) assays [19,38]. For instance, the WNV RdRp domain pulls down the MTase domain and full-length NS5, indicating that inter-domain and inter-molecular interactions exist for NS5 in cells [19]. FRET assays also show that NS5 homo-oligomerizes via its RdRp domain [38]. Thus, if multiple copies of NS5 are present in the replication complex, a specific dimer of NS5, such as one shown in the crystal structure, may be formed preferentially. We thus investigated whether any of the NS5 residues located at one of the dimer interfaces have been shown to be important for viral replication. Indeed, several mutations in or near the type I dimer interface severely reduced viral replication. Individual mutations of P113D and W121D in DENV2 NS5 almost completely disrupted viral replication [22,23]. Both P113 and W121 are conserved in all DENV NS5 sequences, and interact with H701 and G260 at the type I dimer interface, respectively (Fig 3A). Several paired charge-to-alanine mutations in DENV4 NS5 led to viral attenuation in mice [34]. A double mutant K524A/K525A or K525A/D526A in DENV4 NS5 (K523/I524/P525 in DENV3) reduced viral replication by ~100 fold [34]. Other pairs such as E642A/R643A and D654A/R655A in DENV4, corresponding to K641/K642 and E653/R654 located near the dimer interface in DENV3 NS5, reduced viral replication by ~10,000 fold (Fig 3A). Thus the intermolecular interactions observed in the NS5 type I dimer seem to be important for viral replication. None of the residues in the type II dimer interface have been included in published mutagenesis studies, and thus it is not clear whether the type II dimer also plays a role in viral replication.
The structure of DENV NS5 was compared to the JEV NS5 structure [25]. As in the DENV NS5 structure, the JEV MTase is positioned behind the RdRp when looking at the canonical right hand orientation of RdRp, and interacts with the fingers subdomain (Fig 4). However, the relative orientations of MTase and RdRp in DENV and JEV NS5 differ significantly. With the MTase of DENV and JEV NS5 superimposed (rmsd = 0.99 Å for 253 Cα atoms), the RdRp domains are related by a rotation of 102° and a ~5Å translation as calculated by DynDom [32] (Fig 4A). This twisting motion would require significant bending of residues 262–272 (the linker), using residues 260–262 (260GTR262) as the pivot (Fig 4B). The ‘GTR’ sequence is conserved in flavivirus NS5, and individual substitutions of G261A, T262V, and R263L in DENV2 and JEV NS5 greatly impair viral replication and virus production [22]. Thus, flavivirus NS5 likely uses the GTR pivot to sample different conformations, repositioning the MTase and RdRp domains during viral replication.
Due to the different arrangements of MTase and RdRp domains, DENV and JEV NS5 have different domain interfaces (Fig 4C). The domain interface in JEV NS5 buries ~860 Å2 surface area and centers around a hydrophobic core comprised of six residues: P113, L115, and W121 in the MTase domain, and F351, F467, and P585 in the RdRp domain. When individual mutations of the six residues were introduced into infectious JEV or replicon systems, virus replication was significantly impaired, while their polymerase activities were largely unaffected [22,23]. We mapped the corresponding residues on the DENV NS5 structure to determine whether they are involved in domain-domain interactions. The conserved RdRp residues F348 and P584 are found in the MTase and RdRp domain interface in the DENV NS5 monomer, similar to those involved in the intra-molecular interactions in the JEV NS5 monomer. F464 is disordered in the DENV NS5 structure. However, MTase residues P113, P115 (corresponding to L115 in JEV NS5) and W121 are not involved in domain-domain interactions in the DENV NS5 monomer (Fig 4C). Instead, the loop containing these MTase residues is part of the monomer-monomer interface in the NS5 type I dimer (Figs 3A and 4C). Thus, the same MTase region mediates intra- and inter-molecular interactions with the RdRp domain in JEV and DENV NS5, respectively. This surprising discovery again suggests that the inter-molecular interactions in the DENV NS5 dimer (type I) are likely important for viral replication.
NS5 is the most conserved protein in flavivirus, and sequence identities among DENV1-4 NS5 range from 74 (DENV2 vs. DENV4) to 84% (DENV1 vs. DENV3). Despite the high sequence identity, chimeric NS5, wherein the MTase and RdRp from different DENV serotypes or from DENV and WNV (66% sequence identity) are combined in a single polypeptide, is not capable of carrying out replication [39,40]. We have previously shown that a DENV2 infectious RNA containing an NS5 chimera with DENV2 MTase (residues 1–270) replaced by the DENV4 MTase was severely impaired and unable to accumulate viral RNA and virus particles [39]. Repeated passages of the chimeric RNA-transfected cells yielded viruses that contain a mutation in either NS5 MTase (K74I) or NS3 helicase (D290N). Thus, serotype-specific interactions either between the MTase and RdRp domains and/or between NS5 and other components of the replication complex are required for efficient viral RNA synthesis. To determine whether inter-domain interaction within NS5 plays a significant role in viral replication, we replaced the entire NS5 with DENV4 NS5 in the full-length DENV2 infectious RNA. Surprisingly, replication of the DENV2 RNA containing the DENV4 NS5 was delayed only slightly compared to the wild-type DENV2 RNA, significantly faster than replication of the chimeric RNA containing the DENV4 MTase (Fig 5A). This suggests that the serotype-specific interactions between MTase and RdRp domains are the major determinant of efficient viral replication, rather than interactions between individual NS5 domains and RNA or other proteins.
In light of the DENV NS5 structure presented here, we analyzed whether virus serotype-specific interactions exist between the MTase and RdRp domains that could explain the low viral replication of the NS5 chimera, and whether serotype-specific residues accumulate at a particular protein surface that would facilitate serotype-specific protein interactions. Multiple amino acid sequences of four DENV serotypes were analyzed to identify individual residues that are specific to each serotype. Serotype-specific NS5 residues were identified such that the residues are conserved within one DENV serotype, but not conserved among different serotypes. A total of 119 residues were identified as serotype-specific, and 38 positions out of 119 were identified for more than two serotypes (Fig 5B). These 38 residues were then mapped on the DENV3 NS5 structure (Fig 5C). Most of the residues were located on the protein surface, with several residues in the domain-domain and monomer-monomer interfaces. For example, K96, E296, and linker residues H263 and N265 are found in the domain-domain interface (Fig 5C). The more surprising discovery, however, was that many serotype-specific residues clustered on one face of the dimer, while the opposite face of the dimer, where the active sites of NS5 are exposed to solvent, contained few serotype-specific residues (Fig 5C). The best-documented example of serotype-specific differences in DENV NS5 is their nuclear localization. DENV2 and DENV3 NS5 have a functional nuclear localization signal (NLS, residues 369–406) that is recognized by importin-α/β, and predominantly localize to the nucleus [41,42]. By contrast, DENV4 NS5 does not have a functional NLS, and localizes to the cytoplasm [43]. We mapped the NLS sequence on the NS5 structure to determine whether it is located on the serotype-specific side of the dimer (Fig 5C). The NLS indeed mapped to the serotype-specific surface, and four residues within the NLS—residues 375, 383, 385, and 404—were identified as serotype-specific (Fig 5B). Thus, serotype-specific residues accumulate on a particular NS5 dimer surface that is likely mediate NS5-protein interactions that are specific for each serotype.
Genome replication in flavivirus is carried out by a membrane-bound viral replication complex that consists of viral NS proteins, viral RNA, and unknown host proteins [7]. Neither the exact composition of the replication complex nor the stoichiometry of viral NS proteins within the replication complex is currently known. The DENV NS5 structure reported here shows two dimer types with significant interface areas between the monomers (> 1,000 Å2). In particular, the type I dimer is formed by all eight NS5 molecules in the crystallographic ASU, even though each of these eight molecules is in a different chemical environment in the crystal (S1 Fig). The type I dimer was also observed in recently solved DENV NS5 structure [31], and thus the DENV NS5 clearly assembles into the dimer. Many NS5 residues involved in the monomer-monomer interface in the type I dimer are important for viral replication, suggesting that the NS5 dimer may be the biological unit in the replication complex. Recombinant NS5 has been shown to be a monomer in solution [35,37]. However, these experiments were performed in membrane-free, high salt (400 mM NaCl) buffers that do not necessarily reflect the cellular environment. In the membrane-bound replication complex, where free diffusion is limited to a 2-D surface, NS5 may form a higher order oligomer. Cellular interaction and pull-down assays indeed demonstrate that NS5 interacts with itself [19,38]. Furthermore, NS5 is anchored to the membrane by its interaction with NS3 in the replication complex [38]. NS3 itself does not contain any membrane-associated region, but the N-terminal protease domain requires the cofactor NS2B that forms a dimer on the membrane [38]. Thus, the membrane anchored NS5-NS3-NS2B complex could contain a dimeric form of NS5. Other viral polymerases such as poliovirus 3D, which is also a monomer in solution, oligomerize in order to facilitate efficient binding of RNA [44]. Disruption of the protein-protein interface of the poliovirus 3D oligomers led to low viral replication in infected cells, suggesting the biological significance of oligomerization [44].
The physical linkage of the MTase and RdRp domains in a single polypeptide of NS5 suggests that RNA synthesis and capping may be coupled during viral genome replication. In light of the dimer structure, we compared the NS5 monomer and dimer in terms of the relative orientation of the RdRp dsRNA exit and MTase active site, which could be important for the coordination of RdRp and MTase activities (Fig 6). The location of the ssRNA-binding site in the MTase domain of the full-length NS5 was identified based on superposition of the MTase domains in our structure with the structure of the isolated MTase domain-RNA complex [45] (PDB code 2XBM).
In the NS5 monomer, the dsRNA exit site of the RdRp and the active site of the MTase face opposite directions in both the DENV and JEV NS5 structures (Fig 4). Thus, the NS5 monomer model would require a large conformational change to pass the RNA from the RdRp directly to the MTase. Consequently, monomer models with multiple large conformational changes have been proposed [25,46]. Using small-angle X-ray scattering, monomeric NS5 has been shown to adopt multiple conformations in solution ranging from compact to more extended forms [35]. The DENV and JEV NS5 structures clearly indicate that NS5 adopts different arrangements of MTase and RdRp domains related by rotation along the linker (Fig 4). However, multiple large conformational changes could be limited in the replication complex, where NS5 is bound by viral and host proteins on the membrane. The NS5 dimer could thus allow coordination between MTase and RdRp domains within and across the NS5 molecules without requiring large conformational changes inherent in the monomer model. In the DENV NS5 dimer structure, the dsRNA exit site of the RdRp in one monomer and the MTase active site of its partner in the NS5 dimer face the same direction (Fig 6). The distance from the dsRNA exit of one monomer to the entrance to the MTase active site of its partner is also considerably closer than the distance to its own MTase. Thus the dsRNA product of the RdRp could easily access the MTase active site of its neighbor in the NS5 dimer (Fig 6). The dimeric form could still allow small rearrangements of MTase and RdRp domains.
How flavivirus coordinates RNA synthesis and 5’-RNA capping is not well understood. It is currently not clear how NS5 modulates multiple interactions with different RNA forms (plus, minus, double stranded, and capped RNA). In addition, following (+) strand RNA synthesis, the 5’ triphosphate end of the RNA must be dephosphorylated by NS3 5’ RNA triphosphatase activity prior to the cap addition, so NS5 is also required to interact with NS3 for viral replication. Thus, NS5 may need the flexibility to form a monomer and dimer to carry out multiple functions in the replication complex. Future effort will be directed toward validating these NS5 models. The current structure provides a framework to test how NS5 coordinates multiple reactions within a single polypeptide, and to design NS5 inhibitors of dengue virus replication.
DENV3 NS5 constructs Δ2 and Δ6 with the priming loop shortened by either two (797IH798) or six amino acids (795WSIHAH800), respectively, were designed to make NS5 more amenable to crystallization. The plasmid encoding DENV3 NS5-Δ2 was generated from the wild-type NS5 clone (900 residues) using the QuikChange II site-directed mutagenesis kit (Agilent Technologies) with the oligonucleotide primers 5’ CCTACAAGCAGAACGACATGGTCTGCTCACCATCAGTGGATGACTAC 3’ (forward) and 5’ GTAGTCATCCACTGATGGTGAGCAGACCATGTCGTTCTGCTTGTAGG 3’ (reverse). The plasmid containing NS5-Δ2 was then used to produce NS5-Δ6 by another round of site-directed mutagenesis using the oligonucleotide primers 5’ CCCACGAGCAGAACGACACATCAGTGGATGACTACAG 3’ (forward) and 5’ CTGTAGTCATCCACTGATGTGTCGTTCTGCTCGTGGG 3’ (reverse). After DNA sequence confirmation of the NS5-Δ2 and Δ6 plasmids, the plasmids were transformed into BL21-CodonPlus-RIL Escherichia coli cells (Stratagene). His-tagged NS5 proteins were purified as previously reported with minor differences [35]. Briefly, 1 L of Luria Broth medium supplemented with 25 μg/ml chloramphenicol and 50 μg/ml kanamycin was inoculated with 5 mL of start culture. After reaching an OD600 of 0.8, the cells were induced by the addition of 1 mM isopropyl 1-thio-β-D-galactopyranoside and grown at 18°C overnight. Pelleted cells were lysed by sonication in lysis buffer [50 mM sodium phosphate pH 8.0 and 1 M NaCl, supplemented with 50 μg/mL ribonuclease A, 100 μg/mL deoxyribonuclease A, and one tablet of protease inhibitor cocktail (Roche Applied Science)]. NS5 was purified first by affinity chromatography using TALON cobalt affinity resin (Clontech) and an imidazole gradient of 5 to 150 mM in elution buffer (25 mM sodium phosphate, pH 7.0 and 500 mM NaCl). Fractions containing NS5 were concentrated to ~1 mL and further purified by size exclusion chromatography using a HiLoad 16/60 Superdex 200 preparative grade column (GE Healthcare) in 20 mM Tris-HCl pH 7.0, 300 mM NaCl, and 1 mM DTT. NS5 containing fractions were pooled. The protein concentration was determined using a Nanodrop 1000 spectrophotometer (Thermo Scientific) with a theoretical molar extinction coefficient of 217,000 M-1cm-1 and a molecular weight of 104,000 Da.
Polymerase activity assays for wild-type, Δ2, and Δ6 NS5 were performed using the subgenomic RNA template as previously described [27]. Briefly, the assays were conducted in a standard reaction mixture (50 μL) containing 50 mM Tris-HCl (pH 8.0), 50 mM NaCl, 5 mM MgCl2, template RNA (0.2 μg; 0.2 pmol), 500 μM (each) ATP, CTP, and UTP, 10 μM unlabeled GTP, 10 μCi of [α-32P]GTP, and 100 μM DTT. The subgenomic RNA template (719 nt) contains both 5’- and 3’-UTR regions of the DENV2 genome. The reaction was started by adding 10 μg (100 pmol) of purified NS5 before incubation at 37°C for 1 h. The reaction was terminated by acid phenol-chloroform extraction, followed by purification on a Bio-Rad P-30 column to remove the unincorporated nucleotides. Radioactive RNA products were analyzed by formaldehyde-agarose gel electrophoresis and visualized by autoradiography. Band intensities were measured with a PhosphoImager (Molecular Dynamics).
NS5-Δ2 and Δ6 proteins were concentrated to ~10 mg/mL and screened using the sitting drop vapor diffusion method in 96-well plates using a Phoenix RE liquid handling robot (Rigaku). Several conditions produced small crystals within 2 weeks for Δ2 and Δ6, which were further optimized. The best-diffracting NS5-Δ6 crystals were lens-shaped, and grew to full size within 3–7 weeks in 1.0 M succinic acid (pH 7.0), 0.1 M HEPES pH 7.0, and 1% PEG monomethylether 2000. Crystals were harvested by cryo-cooling in liquid nitrogen after soaking for ~20 seconds in well solution supplemented with 25% ethylene glycol.
X-ray diffraction data were collected at 100 K at the Advanced Photon Source beamline 21 (Argonne National Laboratory, Chicago). Four datasets were collected from three crystals. Reflections were indexed and integrated using HKL2000, and four datasets were scaled and merged together using SCALEPACK [47]. A 3.6 Å resolution cutoff was applied using CC1/2 and CC* values of 0.157 and 0.520, respectively, in the highest resolution shell (3.66–3.60 Å) [48]. The crystals belonged to space group P3221 with a = b = 215.3 Å, c = 480.7 Å, and contained eight NS5 monomer in the ASU with a corresponding solvent content of 68%. The initial structure solutions were obtained by molecular replacement with DENV3 MTase (residues 7–262, PDB code 3P97) and RdRp (residues 272–883, PDB code 4HHJ) as search models using the program PHASER in the Phenix suite [49]. Seven MTase and seven RdRp solutions were found, and the eighth MTase was placed manually. After refinement with seven RdRp solutions and eight MTase, the eighth RdRp solution was found using Phenix. The eighth RdRp domain (monomer H in Fig 1) has the weakest density and is missing 133 residues. Upon initial structure solution, continuous electron densities in the 2Fo-Fc map were clearly visible between the C-terminus of MTase and the N-terminus of RdRp, indicating which copies of the MTase and RdRp were part of each of eight intact polypeptide chains. The distance between the terminal ends of each corresponding MTase/RdRp pair was ~24 Å (distance between Cα atoms of R262 and N272), excluding the possibility of any other domain arrangement and making the assignment of continuous protein chains unambiguous. The 2Fo-Fc map also indicated that either SAM or SAH is bound to the MTase active site. We modeled SAH in the density, because SAH was copurified in several high-resolution MTase structures [13,50], and the density for the additional methyl group of SAM was missing (S2 Fig). Additionally, the RdRp contains two metal ions coordinated by tetrahedral geometry with one metal-binding site consisting of H712, H714, C728, and C853 in the thumb subdomain, and the second site consisting of E437, H441, C446, and C449 in the fingers subdomain. Since no metal was added during crystallization, the metal ions must have come from the growth medium and been copurified with the protein. Based on the tetrahedral geometry, coordinating residues, and the positions where zinc atoms have previously been identified in the structures of other flavivirus RdRps [20,29], zinc atoms were modeled into the electron density.
Manual model building was carried out using Coot [51], and iterative cycles of refinement were carried out using phenix.refine. Initially, a global NCS was used during refinement owing to the presence of eight copies of the NS5 monomer in the ASU, which was relaxed to a torsional NCS during subsequent rounds of refinement. TLS (translation, liberation, and screw motion) refinement was also used with TLS groups automatically defined by the TLSMD server [52]. The final model contains eight full-length NS5, eight SAH, and sixteeen zinc atoms with the R and Rfree factors of 23.8 and 27.4%, respectively (Table 1). A Ramachandran plot shows 95.4 and 4.5% of residues in allowed or generally allowed regions, and 0.1% outliers. The final model and structure factors were deposited to the Protein Data Bank with accession code 5CCV.
The buried surface areas and interface residues between MTase and RdRp domains and between NS5 monomers were determined using the program PISA and Contact, respectively, in the CCP4 suite [33,53]. Separate domains were delineated by the following residue ranges: MTase, residues 7–271 in DENV3 and residues 5–270 in JEV; RdRp, residues 272–892 in DENV3 and residues 274–895 in JEV. The rmsds of two structures were calculated using Pymol [54]. The conformational differences between the DENV and JEV NS5 structures were analyzed using the DynDom protein domain motion analysis program [32]. The maximum dimension of the NS5 monomer was calculated by CRYSOL [36].
Construction of a full-length cDNA of DENV2 (New Guinea C strain) and its NS5 chimera containing the DENV4 MTase in yeast/Escherichia coli shuttle vector were previously described [39,55]. The replacement region was amplified by PCR using primers, 5’ TCCATCATGAAGAAC ACAACCAACACGAGAAGGGGAACTGGGACCACAGGAGAG 3’ (forward) and 5’ GACCTG ACTTCTAGCCTTGTTTCATGTTAGTTTTGCCTTTTACAGAACTCCCTCACTCT 3’ (reverse), and pRS424 DENV4 cDNA (GenBank accession number M14931.2). The amplified DNA fragment was mixed with the StuI and AatII-double digested pRS424-FLDV2 cDNA encoding full-length DENV2 RNA. Yeast recombination method was used [55] to create a chimera virus cDNA having DENV4 full-length NS5 in DENV2 backbone. The chimera plasmid was linearized using the BcgI enzyme at the 3′-end of the viral sequence and was used as the template for in vitro transcription catalyzed by SP6 RNA polymerase (Epicenter Biotechnologies) in the presence of the 7-MeGpppG cap analog.
The DENV2 RNA (∼3 μg) containing either the wild type (DENV2) NS5, a NS5 chimera (DENV4 MTase and DENV2 RdRp), or DENV4 NS5 were transfected by electroporation (Amaxa Nucleofector II system, Amaxa Biosystems, Cologne, Germany) into BHK-21 cells (American Type Culture Collection, Manassas, VA), as previously described [39]. Briefly, ∼1 × 106 cells were resuspended in 100 μl of Ingenio solution (Mirus Bio, Madison, WI). After pulsing, cells were carefully transferred into prewarmed complete medium (Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% fetal bovine serum and 1× streptomycin/penicillin), and allowed to recover for 5 min at 37°C in an incubator. Cells were resuspended in 10 ml of complete DMEM and incubated in a T-12.5 flask. On days 2 and 9, cells were trypsinized and transferred into a T-25 and T-75 flask, respectively. This procedure was repeated using one-third of the trypsinized cells from a T-75 flask every 5–7 days. For immunofluorescence assay, RNA-transfected cells at the end of indicated time periods were seeded into a slide (LabTek), and fixed by treatment with acetone. Cells were incubated with a 1:200 dilution of 7E11, a monoclonal antibody against DENV2 NS1. Fluorescein isothiocyante (FITC)-labeled, goat anti-mouse immunoglobulin G conjugate (Kirkegaard & Perry Laboratories) was used as a secondary antibody at a 1:100 dilution. Immunofluorescence photomicrographs (×200 magnification) were acquired using a Leitz Diaplan microscope coupled to the Leica/Wild MPS48 automated photographic system. The numbers and intensities of positive cells were compared utilizing the ImageJ program (National Institutes of Health), as previously described.
Serotype-specific residues were identified in two steps. First, seven to ten NS5 sequences from each DENV serotype were randomly selected for multiple sequence alignment using Clustal W [56]. DENV1 NS5 sequences include GenBank codes AHI43715.1, ACW82945.1, ACJ04223.1, AGN94878.1, and AAK29447.1, and UniProt codes P33478.1 and P17763.2. DENV2 NS5 sequences include UniProt codes P07564.2, Q9WDA6.1, P14337.2, P12823.1, P29991.1, P14340.2, and P29990.1, and GenBank codes ABY65725.1, AII99332.1 and AHB63929.1. DENV3 NS5 sequences include GenBank codes ABV54900.1, YP_001621843.1, AHG23213.1, Q99D35.1, ACY70817.1, AAS49486.2, ABV54900.1, ABV03585.1, YP_001621843.1, and ABV54900.1. DENV4 NS5 sequences include GenBank codes GNWVDF, AHG23274.1, ACW83012.1, ACQ44391.1, ABO45246.1, AEX91754.1, and AGI95993.1. Serotype-specific residues that are conserved in each serotype, but different among serotypes were selected. Next, the virus variation resource at NCBI (http://www.ncbi.nlm.nih.gov/genome/viruses/variation/dengue/) was used to remove residues that have some variations in each serotype [57].
|
10.1371/journal.pgen.1003522 | ttm-1 Encodes CDF Transporters That Excrete Zinc from Intestinal Cells of C. elegans and Act in a Parallel Negative Feedback Circuit That Promotes Homeostasis | Zinc is an essential metal involved in a wide range of biological processes, and aberrant zinc metabolism is implicated in human diseases. The gastrointestinal tract of animals is a critical site of zinc metabolism that is responsible for dietary zinc uptake and distribution to the body. However, the role of the gastrointestinal tract in zinc excretion remains unclear. Zinc transporters are key regulators of zinc metabolism that mediate the movement of zinc ions across membranes. Here, we identified a comprehensive list of 14 predicted Cation Diffusion Facilitator (CDF) family zinc transporters in Caenorhabditis elegans and demonstrated that zinc is excreted from intestinal cells by one of these CDF proteins, TTM-1B. The ttm-1 locus encodes two transcripts, ttm-1a and ttm-1b, that use different transcription start sites. ttm-1b expression was induced by high levels of zinc specifically in intestinal cells, whereas ttm-1a was not induced by zinc. TTM-1B was localized to the apical plasma membrane of intestinal cells, and analyses of loss-of-function mutant animals indicated that TTM-1B promotes zinc excretion into the intestinal lumen. Zinc excretion mediated by TTM-1B contributes to zinc detoxification. These observations indicate that ttm-1 is a component of a negative feedback circuit, since high levels of cytoplasmic zinc increase ttm-1b transcript levels and TTM-1B protein functions to reduce the level of cytoplasmic zinc. We showed that TTM-1 isoforms function in tandem with CDF-2, which is also induced by high levels of cytoplasmic zinc and reduces cytoplasmic zinc levels by sequestering zinc in lysosome-related organelles. These findings define a parallel negative feedback circuit that promotes zinc homeostasis and advance the understanding of the physiological roles of the gastrointestinal tract in zinc metabolism in animals.
| Zinc is an essential mineral nutrient involved in many physiological processes, and it plays a critical role in human health. Insufficient dietary zinc causes a wide range of health problems, and excess dietary zinc causes toxicity. Furthermore, genetic mutations affecting zinc metabolism have been implicated in a variety of human diseases. Therefore, animals require homeostatic mechanisms that effectively regulate zinc metabolism in response to dietary fluctuations. The gastrointestinal tract is a major tissue that orchestrates zinc metabolism in animals, and zinc transporters are key molecular regulators involved in this process. To understand these regulatory mechanisms, we used bioinformatic techniques to identify 14 genes that encode predicted Cation Diffusion Facilitator (CDF) family zinc transporters in the C. elegans genome. We demonstrated that one of these, ttm-1, functions in intestinal cells to promote zinc excretion, and this activity protects animals from zinc toxicity. Genetic analysis revealed that zinc excretion mediated by TTM-1B is coordinated with zinc storage mediated by CDF-2, and these transporters function in a parallel negative feedback circuit to maintain zinc homeostasis in intestinal cells. These findings provide molecular and physiological insight into the regulatory mechanisms of zinc metabolism in animals.
| The trace element zinc is essential for all biological systems. Zinc is a well established structural and enzymatic cofactor required for the function of numerous proteins [1], and emerging evidence indicates that zinc functions as a signaling molecule in several biological processes, including development, immune response and synaptic transmission [2]–[4]. Organisms have evolved mechanisms that control zinc homeostasis and metabolism in individual cells and the entire body, and these mechanisms are critical for human health. Diets that contain deficient or excess zinc and result in impaired zinc homeostasis cause a wide range of defects in human health [5], [6]. Furthermore, genetic mutations that cause aberrant zinc metabolism are implicated in a variety of human diseases, such as cancer, diabetes and neurodegenerative diseases [7]–[13]. Thus, understanding mechanisms of zinc homeostasis has important implications for human health.
Zinc metabolism is regulated at the cellular and organismal levels. In eukaryotic cells, the movement of zinc ions across membranes is mediated by transmembrane zinc transporter proteins from two major families, Cation Diffusion Facilitator (CDF/ZnT/SLC30) and Zrt-, Irt-like protein (ZIP/SLC39) [14]. CDF proteins have six conserved transmembrane motifs (TMs) and transport cytoplasmic zinc out of the cell or into the lumen of intracellular organelles, thereby decreasing the level of cytoplasmic zinc. By contrast, ZIP proteins have eight conserved TMs and transport zinc in the opposite directions, thereby increasing the level of cytoplasmic zinc. It appears that almost all zinc ions in the cytoplasm are bound to small molecules or soluble proteins such as metallothioneins [15]. Metallothioneins are small, cysteine-rich proteins that bind up to seven zinc atoms and may function in zinc detoxification and storage. A critical question is how the activity of zinc transporters and metallothioneins are coordinately regulated to maintain zinc homeostasis. Some evidence indicates that, in response to high zinc conditions, the expression of specific CDF proteins and methallothioneins are induced to promote zinc excretion and sequestration in intracellular organelles and proteins, thereby protecting from zinc toxicity [16], [17].
At the organismal level, the gastrointestinal tract is the major site of zinc metabolism that mediates dietary zinc uptake and distribution to the body [18], [19]. In humans, dietary zinc is absorbed into enterocytes across the apical plasma membrane by the action of ZIP4 [13]. Loss-of-function mutations of ZIP4 cause the recessive genetic disease acrodermatitis enteropathica, which is characterized by symptoms of zinc deficiency [8]. ZIP4 expression is highly responsive to dietary zinc levels, which regulates zinc influx into enterocytes [20]. Zinc absorbed in enterocytes is transported by ZnT1 across the basolateral plasma membrane to allow distribution to other tissues [21]. ZnT1 expression is also responsive to dietary zinc levels and induced in high zinc conditions [21], [22]. ZnT1 is essential for viability, since ZnT1-deficient mice display embryonic lethality [23]. Zinc excretion is not as well characterized as zinc absorption. Zinc is excreted in the feces, and a major source of fecal zinc is pancreatic and biliary secretion of zinc containing enzymes [18]. However, the primary role of these enzymes is likely to be digestion rather than zinc homeostasis. The role of enterocytes in zinc excretion via zinc transporters is only beginning to emerge. The ZnT5 variant B is localized to the apical membrane of enterocytes and is reported to function in both zinc efflux and influx [24], [25]. Additional zinc transporters are localized to intracellular compartments in enterocytes. For example, ZnT5A, ZnT6, ZnT7 and ZIP7 are localized to the Golgi complex, and ZnT2 and ZnT4 are localized to endosomal or lysosomal vesicles [19]. These proteins might be involved in zinc excretion through the secretory pathway, although in vivo evidence for this function has not been well established.
The nematode C. elegans is a useful model organism to study zinc metabolism. Facilitated by powerful genetic techniques, studies of C. elegans have contributed to the discovery of novel functions of zinc ions in signal transduction during development and to understanding mechanisms that regulate zinc metabolism and homeostasis [26]–[31]. The C. elegans genome encodes many evolutionarily conserved proteins involved in zinc metabolism, including members of the CDF, ZIP and methallothionein families [14], [32], indicating that it will be relevant to understanding mechanisms of zinc metabolism in higher animals. Similar to higher animals, intestinal cells of C. elegans are the point of zinc entry and the critical tissue for zinc homeostasis. CDF-1 and SUR-7 are expressed in intestinal cells and localized to the plasma membrane and the ER/Golgi complex, respectively [26], [31]. Lysosome-related organelles in intestinal cells, known as gut granules, function as a physiological zinc storage site that is important for zinc detoxification and mobilization in response to fluctuating dietary zinc levels, and CDF-2 plays an essential role in this process [30]. Furthermore, the metallothionein genes mtl-1 and mtl-2 are highly expressed in intestinal cells [32]. Therefore, the intestine of C. elegans appears to function as a major organ that controls zinc metabolism and may be a relevant model for the gastrointestinal tract in mammals.
Here we identified 14 predicted CDF family members encoded by the C. elegans genome using iterative sequence based homology search methods. We focused on ttm-1 (toxin-regulated target of p38MAPK), because the predicted TTM-1 proteins are highly related to vertebrate CDF proteins and the role of this gene in zinc metabolism had not been characterized. ttm-1 was initially identified as a downstream target of p38 MAP kinase that is induced in response to pore-forming bacterial toxins [33]. In addition, ttm-1 was reported to be induced by cadmium exposure [34]. To elucidate the function of ttm-1 in zinc metabolism, we characterized the ttm-1 gene structure, expression and function. ttm-1 encodes two protein isoforms: TTM-1A was not zinc regulated, whereas TTM-1B was induced by high levels of dietary zinc. Each isoform displayed unique cellular and subcellular expression patterns, and TTM-1B displayed a striking subcellular localization at the apical plasma membrane of intestinal cells. Using a zinc-specific fluorescent dye to visualize zinc and quantitative zinc level analyses, we showed that ttm-1 functions in zinc excretion from intestinal cells into the lumen of the gut. Zinc excretion mediated by TTM-1B was coordinated with zinc storage in gut granules mediated by CDF-2, and these two genes function in a parallel negative feedback circuit to promote zinc detoxification and homeostasis. These studies rigorously document zinc excretion from intestinal cells mediated by a CDF protein and the coordinated action of multiple CDF proteins in zinc detoxification, and may be relevant to understanding zinc metabolism in higher animals.
To identify C. elegans CDF proteins, we employed iterative sequence-based homology search methods using the Position-Specific Iterated BLAST (PSI-BLAST) [35]. Beginning with the protein sequence of CDF-1, a total of 14 C. elegans proteins were identified, including SUR-7 and CDF-2 that were previously characterized and eleven additional CDF proteins (Figure 1). To analyze evolutionary conservation of CDF proteins, we compared the 14 C. elegans proteins with CDF proteins of yeast, plants and humans using the reciprocal PSI-BLAST. This analysis generated a phylogenetic tree of CDF proteins and suggested that CDF proteins have diverged into several subfamilies (Figure 1). Human ZnT proteins were present in four clusters, and all of these clusters contained C. elegans proteins, suggesting that C. elegans contains CDF proteins that may be relevant to the function of most or all human CDF proteins. There were also clusters that contained C. elegans proteins but no human proteins and clusters that contained only plant and yeast proteins, suggesting that some CDF proteins in yeast, plants and worms have functions not represented in humans.
We focused on ttm-1 because it is highly related to human ZnT2, 3, 4 and 8 (Figure 1). ttm-1 expression was previously analyzed, since it is induced by pore-forming bacterial toxins and cadmium exposure [33], [34], but the function of ttm-1 has not been reported. The computational algorithm Gene Finder and the analysis of cDNAs indicated that ttm-1 consists of five exons and encodes two transcripts; ttm-1a is generated by splicing exons 1, 2, 4, and 5, whereas ttm-1b is generated by splicing exons 3, 4, and 5 (Figure 2A). To determine if the two transcripts result from alternative splicing of a single primary transcript or alternative transcription initiation sites, we used 5′ rapid amplification of cDNA ends (5′ RACE). If the two transcripts were generated by alternative processing of the same precursor, then ttm-1a and ttm-1b are predicted to contain SL1 and SL2 trans-spliced leader sequences at the 5′ends, respectively. By contrast, if ttm-1a and ttm-1b are independently transcribed, then both are predicted to contain the SL1 trans-spliced leader sequence [36]. The SL1 trans-spliced leader sequence was detected at the 5′ end of both ttm-1a and ttm-1b transcripts (Figure S1). These results indicate that the ttm-1 locus has two transcription initiation sites that generate two primary transcripts that are spliced to form ttm-1a and ttm-1b (Figure 2A).
The predicted TTM-1A and TTM-1B proteins have the signature conserved motifs of the CDF family: six transmembrane motifs (TMs) and a conserved histidine-rich motif, (HX)n, in the loop between TM IV and V (Figure 2B). The (HX)n motif is implicated in zinc binding and regulation of transport activity [37]. TTM-1A and TTM-1B showed the highest sequence similarity with CDF-2 from C. elegans and ZnT2 from humans (Figure 2B). As a result of different transcription start sites, TTM-1A and TTM-1B have distinct amino acid sequences at N-terminus. TTM-1B has an additional histidine-rich motif in the N-terminus, which is similar to the (HX)n motif in the loop between TM IV and V (Figure 2B), suggesting that TTM-1B may have an N-terminal zinc-binding domain that influences transporter function or regulation by zinc.
To determine cell types that express ttm-1, we generated plasmids that express green fluorescent protein (GFP) under the control of the predicted ttm-1a and ttm-1b promoters, introduced these constructs into transgenic animals, and monitored fluorescence. For ttm-1a, the promoter region consisted of ∼1.2 kb of genomic DNA that extends from the ttm-1a start codon upstream to the 3′ end of the adjacent gene (Figure 2A). Pttm-1a::GFP transgenic animals displayed GFP expression in the hypodermis and the intestine (Figure 3A). For ttm-1b, the promoter region consisted of ∼6 kb of genomic DNA that extends from the ttm-1b start codon in exon 3 upstream into intron 2 (Figure 2A). Pttm-1b::GFP transgenic animals displayed GFP expression in multiple tissues including the intestine, head neurons, seam cells, hypodermis and the vulva (Figure 3B). In both cases, similar patterns were displayed by multiple, independently derived transgenic strains, indicating these results are reproducible. These results indicate that both ttm-1 transcripts are expressed in intestinal cells, and each has a distinct expression pattern in non-intestinal cells.
To determine whether the expression of ttm-1 is regulated by dietary zinc levels, we cultured wild-type animals on noble agar minimal media (NAMM) containing 0 µM or 200 µM supplemental zinc [27] and analyzed mRNA levels using quantitative real time PCR (RT-PCR). The level of ttm-1a mRNA was not affected by dietary zinc, similar to the control gene ama-1 which encodes the large subunit of RNA polymerase II and is not regulated by dietary zinc [28]. By contrast, the level of ttm-1b mRNA was significantly elevated by supplemental dietary zinc, similar to the zinc-inducible gene cdf-2 [28], [30] (Figure 4A). These results indicate that dietary zinc increases the synthesis and/or decreases the degradation of ttm-1b mRNA.
To determine how ttm-1b expression is regulated in different tissues in response to high levels of dietary zinc, we analyzed Pttm-1b::GFP transgenic animals. Pttm-1b::GFP transgenic animals cultured with 200 µM supplemental zinc displayed increased GFP expression in intestinal cells (Figure 4B, top). However, GFP expression was not increased in the other tissues, such as head neurons (Figure 4B, bottom). Thus, the 6 kb ttm-1b promoter fragment contains regulatory elements that are sufficient to mediate zinc-responsive transcriptional induction in intestinal cells, suggesting regulation occurs at the level of transcriptional initiation. Furthermore, ttm-1b expression appears to be differentially regulated by dietary zinc in different cell types.
To analyze the intracellular localizations of TTM-1A and TTM-1B proteins, we generated plasmids that encode full length TTM-1A or TTM-1B protein fused to GFP under the control of the corresponding promoter (Figure 2A) and introduced these plasmids into transgenic animals. TTM-1A::GFP was expressed in the intestine and hypodermis (Figure 5A–5F); the same tissue distribution pattern was displayed by transgenic animals expressing GFP alone under the control of the ttm-1a promoter, indicating that the coding sequences do not substantially influence the tissue expression pattern. In intestinal and hypodermal cells, TTM-1A::GFP displayed a punctuate pattern (Figure 5C–5F). TTM-1A::GFP did not colocalize with the autofluorescence of gut granules in intestinal cells or with MitoTracker in hypodermal cells (data not shown), suggesting that TTM-1A does not localize to lysosome-related organelles or mitochondria.
TTM-1B::GFP was expressed in multiple tissues including intestine, head neurons and seam cells (Figure 5G–5L), consistent with the tissue distribution pattern displayed by Pttm-1b::GFP transgenic animals. TTM-1B::GFP was localized to intracellular compartments in several tissues such as seam cells (Figure 5I, 5J) and head neurons (data not shown). However, in intestinal cells, TTM-1B::GFP was localized to the plasma membrane and restricted to the apical side of the cell which forms the surface of the intestinal lumen (Figure 5K, 5L). Thus, TTM-1B displayed distinct intracellular localization in different cell types.
CDF-1 is localized to the plasma membrane of intestinal cells [26]. To compare the localization of CDF-1 with TTM-1B, we immunostained transgenic animals expressing CDF-1::GFP with an anti-GFP antibody to amplify GFP signal in intestinal cells and eliminate autofluorescence from gut granules. CDF-1::GFP was detected on the plasma membrane where it was restricted to the basolaternal surface of intestinal cells that face the body cavity and absent from the apical surface (Figure 5M, 5N). These results suggest that TTM-1B and CDF-1 are localized in a non-overlapping pattern on the apical and basolateral plasma membranes of intestinal cells, respectively.
To characterize the function of ttm-1, we obtained the mutant allele ttm-1(ok3503) that was generated by the C. elegans Gene Knockout Project at Oklahoma Medical Research Foundation (OMRF). By analyzing the genomic DNA sequence of this allele, we determined that the ok3503 mutation is a deletion of 877 bp beginning in exon 4 and extending to intron 4 (Figure 2A and Figure S2). This deletion removes the coding sequence for part of TM II, all of TM III, IV, V and VI, and the (HX)n motif (Figure 2B). Because these motifs are highly conserved in the CDF family, the ok3503 mutation is likely to severely disrupt the activity of TTM-1 proteins. The deleted region affects both TTM-1A and TTM-1B (Figure 2A), suggesting that ttm-1(ok3503) is a strong loss-of-function mutation that reduces the activity of both isoforms. To determine the effect of the ok3503 mutation on transcription of ttm-1, we analyzed ttm-1a and ttm-1b mRNA levels by RT-PCR. Both ttm-1 transcripts were detected in the ttm-1(ok3503) strain, and ttm-1b expression was elevated in response to high levels of dietary zinc, similar to wild-type animals (data not shown). Thus, the ttm-1(ok3503) allele does not appear to affect transcription initiation.
To determine the role of ttm-1 in zinc metabolism, we measured total zinc content of worms using inductively coupled plasma-mass spectrometry (ICP-MS). With 0 µM supplemental zinc, wild-type and ttm-1(ok3503) mutant animals displayed similar total zinc content. In the presence of 200 µM supplemental zinc, ttm-1 mutant animals displayed approximately 40% higher total zinc content than wild-type animals (Figure 6A). Thus, ttm-1 may promote zinc excretion in high levels of dietary zinc. Expression levels of metallothionein genes can be regulated by a variety of stressors, such as heavy metals and oxidative damage, and metallothionein gene expression is upregulated by high levels of environmental zinc [38]. To indirectly measure zinc levels, we analyzed mtl-1 and mtl-2 mRNA levels using RT-PCR. Both wild type and ttm-1 mutant animals displayed a substantial increase in the levels of mtl-1 and mtl-2 transcripts when exposed to 100 µM supplemental zinc (Figure 6B, 6C), demonstrating that C. elegans metallothionein genes can be induced by dietary zinc. Furthermore, in the presence of 0 µM and 100 µM supplemental zinc, ttm-1 mutant animals displayed approximately 4-fold and 2-fold higher levels of mtl-1 mRNA, respectively, compared to wild-type animals (Figure 6B, 6C). Similar results were observed for mtl-2 mRNA expression (Figure 6B, 6C). Together, these results indicate that zinc hyperaccumulates in ttm-1 mutant animals, suggesting that TTM-1 isoforms function in the excretion of zinc out of the body.
Zinc-responsive fluorescent dyes have been used to visualize the distribution of zinc in various model systems, and we previously reported that FluoZin-3 specifically detects labile zinc stored in gut granules in C. elegans [30]. To visualize zinc in other locations in C. elegans, we analyzed additional zinc-responsive fluorescent dyes and discovered that wild-type animals incubated with the dye Zinpyr-1 displayed fluorescence primarily in the pseudocoelom (Figure 6E). Animals cultured with 100 µM supplemental zinc displayed elevated Zinpyr-1 fluorescence in the pseudocoelom compared to 0 µM supplemental zinc (Figure 6D, 6E), suggesting that Zinpyr-1 fluorescence is a measure of labile zinc and the level of pseudocoelomic zinc is responsive to dietary zinc levels. The pseudocoelom is a fluid-filled space that interfaces with the basolateral surface of intestinal cells and epithelial cells and surrounds internal organs, such as the gonad, muscles and neurons. The fluid in the pseudocoelom is likely to transport nutrients such as zinc from the intestinal cells to other tissues, and Zinpyr-1 appears to visualize this pool of labile zinc.
To analyze the regulation of pseudocoelomic zinc, we examined cdf-1 mutant animals. CDF-1 was localized to the basolateral surface of the plasma membrane of intestinal cells, the interface between the intestine and pseudocoelom (Figure 5M, 5N), and thus we predicted that reducing CDF-1 activity might decrease zinc levels in the pseudocoelom. cdf-1(n2527) is a strong loss-of-function mutation that causes a nonsense change [26]. cdf-1 mutant animals displayed decreased Zinpyr-1 fluorescence in the pseudocoelom compared to wild-type animals, and the difference was striking in 100 µM supplemental zinc (Figure 6D, 6E). These results suggest that Zinpyr-1 specifically detects zinc in the pseudocoelom and CDF-1 plays a key role in transporting zinc from intestinal cells to the pseudocoelom.
ttm-1 mutant animals displayed elevated Zinpyr-1 fluorescence in the pseudocoelom compared to wild-type animals (Figure 6D, 6E). Thus, the pseudocoelom is one compartment that hyperaccumulates zinc in ttm-1 mutant animals. To investigate the relationship between TTM-1 isoforms and CDF-1, we analyzed ttm-1;cdf-1 double mutant animals. Double mutant animals displayed a low level of Zinpyr-1 fluorescence in the pseudocoelom, similar to cdf-1 single mutant animals (Figure 6D, 6E). These results suggest that the accumulation of zinc in the pseudocoelom displayed by ttm-1 mutant animals requires the activity of CDF-1. We propose that TTM-1 isoforms function upstream of CDF-1 to reduce zinc levels in the pseudocoelom by localizing to the apical surface of the plasma membrane of intestinal cells and functioning to excrete zinc. In ttm-1 mutant animals, excess zinc in the cytoplasm of intestinal cells is transported into the pseudocoelom by CDF-1.
To determine if zinc excretion by TTM-1 isoforms contributes to zinc detoxification, we examined zinc sensitivity by measuring the growth rate of animals in the presence of different concentrations of supplemental zinc. In all zinc conditions, ttm-1 mutant animals displayed similar or marginally reduced growth rates compared to wild-type animals (Figure 7A). To address the possibility that other CDF proteins may function redundantly with TTM-1 isoforms, we examined double mutant animals with cdf-1(n2527), cdf-2(tm788) and sur-7(ku119). cdf-2(tm788) is a strong loss-of-function mutation caused by a deletion/insertion that disrupts transcription [28], and sur-7(ku119) is a partial loss-of-function mutation caused by a base pair change in a splice site [31]. cdf-1 and sur-7 mutant animals displayed reduced growth rates in the presence of supplemental zinc compared to wild-type animals (Figure 7A and 7B), consistent with the previous reports that cdf-1 and sur-7 mutant animals are hypersensitive to zinc [26], [31]. The zinc sensitivities of ttm-1;cdf-1 and ttm-1;sur-7 double mutant animals were similar to cdf-1 and sur-7 single mutant animals, respectively (Figure 7A and 7B). By contrast, ttm-1;cdf-2 double mutant animals displayed strikingly enhanced zinc sensitivity compared to cdf-2 single mutant animals (Figure 7C), indicating that TTM-1 isoforms have an important function in zinc detoxification in animals that lack CDF-2. To investigate whether CDF-2 activity is altered to compensate for the loss of TTM-1 isoforms, we compared cdf-2 mRNA levels between wild type and ttm-1 mutant animals. cdf-2 mRNA levels were higher in ttm-1 mutant animals (Figure 7D), suggesting that increased CDF-2 expression is a mechanism that compensates for the loss of ttm-1 function in zinc detoxification.
CDF-2 mediates zinc storage in gut granules. Glo mutant animals are defective in the biogenesis of gut granules [39], [40] and, therefore, are defective in zinc detoxification [30]. pgp-2 mutant animals have a moderately reduced number of gut granules compared to wild-type animals [40], and glo-1 mutant animals have a severely reduced number of gut granules [39]. ttm-1;pgp-2 double mutant animals displayed enhanced zinc sensitivity compared to pgp-2 single mutant animals (Figure 7E). ttm-1;glo-1 double mutant animals displayed substantially enhanced zinc sensitivity compared to glo-1 single mutant animals (Figure 7F). These results are consistent with the analysis of CDF-2 and suggest that TTM-1 isoforms function in zinc detoxification in cooperation with zinc storage in gut granules.
To determine the role of TTM-1 isoforms in zinc detoxification, we generated transgenic animals that express TTM-1A::GFP or TTM-1B::GFP. Transgenic ttm-1;cdf-2 mutant animals expressing either TTM-1A::GFP or TTM-1B::GFP displayed increased growth rates compared to ttm-1;cdf-2 mutant animals (Figure 7G), suggesting that both TTM-1A::GFP and TTM-1B::GFP isoforms are functional zinc transporters and contribute to zinc detoxification. To further analyze the site of action, we generated transgenic animals expressing TTM-1B::GFP under the control of cdf-2 promoter (Figure S3). In these animals, TTM-1B::GFP is expressed specifically in intestinal cells and is zinc inducible. Intestine-specific expression of TTM-1B::GFP in ttm-1;cdf-2 double mutant animals rescued zinc sensitivity to the level of cdf-2 single mutant animals (Figure 7H). Thus, TTM-1B expression in intestinal cells was sufficient to rescue the zinc hypersensitivity phenotype of ttm-1;cdf-2 mutant animals. These results suggest that TTM-1B localized to the apical surface of the plasma membrane of intestinal cells plays a critical role in zinc excretion and detoxification.
To further characterize the relationship between ttm-1 and cdf-2, we analyzed ttm-1;cdf-2 double mutant animals for total zinc content using ICP-MS. In the presence of 0 µM and 200 µM supplemental zinc, ttm-1;cdf-2 mutant animals displayed lower total zinc content than wild-type animals, similar to cdf-2 mutant animals (Figure 8A). These results indicate that although the ttm-1 mutation causes a defect in zinc excretion, the cdf-2 mutation prevents this excess zinc from being stored in gut granules, resulting in an overall decrease in organismal zinc. Thus, CDF-2 is required for the hyperaccumulation of zinc in ttm-1 mutant animals, and TTM-1 isoforms and CDF-2 act antagonistically with respect to total organismal zinc content.
To indirectly measure the level of cytoplamic zinc in intestinal cells, we determined mRNA levels of mtl-1 and mtl-2 using RT-PCR. In the absence of supplemental zinc, ttm-1;cdf-2 double mutant animals displayed strikingly elevated levels of mtl-1 and mtl-2 expression compared to wild-type animals, approximately 35-fold and 15-fold increases, respectively (Figure 8B). With 200 µM supplemental zinc, the expression of mtl-1 and mtl-2 in ttm-1;cdf-2 double mutant animals was further elevated (Figure S4A and S4B). These results indicate that both ttm-1 and cdf-2 function to inhibit mtl-1/2 expression, presumably by reducing the level of cytoplasmic zinc in intestinal cells. Furthermore, the synergistic effect of combining the two mutations suggests that ttm-1 and cdf-2 function redundantly to decrease cytoplasmic levels of zinc. Together, the findings that total zinc content decreased whereas mtl-1/2 expression dramatically increased suggests that mtl-1/2 gene expression responds to zinc levels in the cytoplasm but does not respond to zinc levels in the storage gut granules. Thus, impaired zinc storage in ttm-1;cdf-2 double mutant animals causes a reduction in gut granule zinc content and an overall decrease in total zinc content. By contrast, the combination of impaired zinc storage and excretion causes an increase in the levels of cytoplasmic zinc that induces expression of mtl-1/2. These highly elevated levels of mtl-1 and mtl-2 expression were not observed in ttm-1;cdf-1 or ttm-1;sur-7 mutant animals (Figure S4C–S4F), indicating the specificity of the genetic interaction between ttm-1 and cdf-2. These results indicate that zinc homeostasis is severely impaired in ttm-1;cdf-2 mutant animals and suggest that TTM-1 isoforms and CDF-2 are key regulators of zinc homeostasis that act in a coordinated manner to control the levels of cytoplasmic zinc.
By exhaustively searching the C. elegans genome for CDF proteins, we identified 14 predicted family members. C. elegans contains one or more proteins that are highly similar to all ten human CDF proteins, indicating that mechanisms of zinc metabolism are conserved in nematodes and mammals. We previously demonstrated that C. elegans CDF-2 plays a critical role in detoxification by transporting zinc into lysosome-related organelles in intestinal cells [30]. Here we focused on the cdf gene ttm-1, because TTM-1 isoforms are highly related to CDF-2, and these C. elegans proteins are similar to human ZnT2, ZnT3, ZnT4 and ZnT8.
The analysis of ttm-1 demonstrated that the locus generates two transcripts, ttm-1a and ttm-1b, that use alternative transcription start sites. The ttm-1a and ttm-1b promoters appear to function independently based on the analysis of reporter constructs in transgenic animals. Both ttm-1a and ttm-1b were expressed in intestinal cells, but ttm-1a was also expressed in hypodermal cells whereas ttm-1b was also expressed in neurons and seam cells. Although endogenous TTM-1 proteins were not visualized, these expression patterns are likely to reflect the endogenous expression patterns, since constructs containing these promoters expressing TTM-1 isoforms fused to GFP were capable of rescuing the zinc hypersensitivity phenotype of ttm-1 mutant animals. A second difference between the promoters was revealed by the response to high zinc conditions. ttm-1b mRNA levels were increased in response to high levels of dietary zinc, whereas ttm-1a mRNA levels were not changed. The ttm-1b promoter was sufficient to confer zinc-inducibility when fused to the reporter GFP, indicating that regulation occurs at the level of transcriptional initiation. The cis-acting DNA elements and trans-acting proteins that mediate transcriptional induction have not been defined.
The ttm-1a and ttm-1b transcripts encode proteins that have unique N-termini and share a common C-terminus. Both TTM-1A and TTM-1B contain six predicted membrane spanning domains and a histidine-rich motif, suggesting both protein isoforms are functional zinc transporters. The TTM-1A and TTM-1B isoforms displayed distinct subcellular localizations. In intestinal cells, TTM-1B was localized to the apical surface of the plasma membrane, whereas TTM-1A was localized to intracellular vesicles. TTM-1A-containing vesicles did not colocalize with mitochondria or gut granules, which are lysosome-related organelles. Although the identity of the TTM-1A-containing vesicles was not further characterized, the homology of TTM-1A with CDF-2 and ZnT2 raises the possibility that TTM-1A may be localized to a distinct lysosomal compartment. Because the two TTM-1 isoforms have unique N-terminal sequences, we infer that the N-terminal regions contain amino acid motifs that target the isoforms to distinct subcellular localizations. Furthermore, the TTM-1B isoform displayed a different subcellular localization in different cell types; TTM-1B was localized to intracellular vesicles in neurons and seam cells. Thus, TTM-1B may respond to cell type-specific mechanisms for subcellular localization and play distinct roles in zinc metabolism in different cell types. These results illustrate how a single zinc transporter locus can make complicated contributions to zinc metabolism by generating two protein isoforms that are expressed in overlapping and unique tissues, have different subcellular localizations and respond differently to dietary zinc.
To determine the function of ttm-1 in zinc metabolism, we analyzed a deletion mutation that lacks the coding region for highly conserved domains and is likely to severely reduce the activity of both the TTM-1A and TTM-1B isoforms. Zinc levels in ttm-1 mutant animals were evaluated using three independent methods; the measurement of total zinc content by ICP-MS, the analysis of mtl-1/2 transcript levels by RT-PCR, and the visualization of labile zinc using a zinc-responsive fluorescent dye, Zinpyr-1. ttm-1 mutant animals displayed significantly elevated total zinc content, demonstrating excess zinc is present somewhere in the animal. To define the anatomical location(s) of the excess zinc, we showed that ttm-1 mutant animals displayed elevated mtl-1 and mtl-2 mRNA levels. Because these genes are transcribed specifically in intestinal cells and are likely to respond to the level of cytoplasmic zinc, these results suggest that ttm-1 mutant animals have elevated levels of cytoplasmic zinc in intestinal cells. Because TTM-1B is localized to the apical surface of the plasma membrane in intestinal cells, these results suggest that TTM-1B contributes to zinc homeostasis by promoting zinc excretion into the intestinal lumen (Figure 9A).
Using the newly developed technique of Zinpyr-1 staining, we demonstrated that there is a pool of labile zinc in the pseudocoelom. We hypothesized that the source of pseudocoelomic zinc is excretion from intestinal cells mediated by CDF-1, the homolog of vertebrate ZnT1 [26]. Consistent with this hypothesis, we showed that CDF-1 was localized specifically on the basolateral surface of the plasma membrane of intestinal cells. Importantly, cdf-1 mutant animals displayed reduced Zinpyr-1 staining in the pseudocoelom. We previously reported that total zinc content is elevated in cdf-1 mutant animals [28], suggesting that cdf-1 mutant animals accumulate zinc in intestinal cells as a result of defective zinc transport into the pseudocoelom. ttm-1 mutant animals displayed higher Zinpyr-1 fluorescence in the pseudocoelom than wild-type animals, indicating that ttm-1 is necessary to limit the level of zinc in the pseudocoelom. The high Zinpyr-1 fluorescence in ttm-1 mutant animals was decreased in ttm-1;cdf-1 double mutant animals. Based on these results, we propose that ttm-1 functions in intestinal cells to promote zinc excretion into the intestinal lumen. In ttm-1 mutant animals, zinc accumulates in the cytoplasm of intestinal cells, and some of this excess zinc is transported into the pseudocoelom by CDF-1 (Figure 9A).
TTM-1A and TTM-1B in intestinal cells were localized to intracellular vesicles and the apical surface of the plasma membrane, respectively. The subcellular localization of TTM-1B indicates a direct role in excretion by transport of zinc across the apical surface of the plasma membrane. The subcellular localization of TTM-1A is consistent with an indirect role in excretion, if TTM-1A-containing vesicles fuse with the apical surface of the plasma membrane, or a role in sequestration. To analyze the function of each isoform separately, we generated transgenic animals that expressed only TTM-1A or TTM-1B in the background of ttm-1(ok3503). Both isoforms displayed rescue activity, indicating that both isoforms can be functionally important. Two additional experiments highlight the importance of TTM-1B. First, TTM-1B but not TTM-1A was induced by high levels of dietary zinc specifically in intestinal cells, suggesting TTM-1B plays a more prominent role during zinc excess. Second, the intestine-specific expression of TTM-1B alone was sufficient to rescue the zinc hypersensitivity phenotype of ttm-1;cdf-2 mutant animals. These results suggest that the direct transport of zinc across the apical plasma membrane of intestinal cells by TTM-1B plays an important and physiological role.
The yeast Saccharomyces cerevisiae does not appear to excrete zinc, since CDF proteins are not localized to the plasma membrane [41]. To detoxify excess zinc, yeast cells sequester zinc in an intracellular vesicle, the vacuole. In mammals, pancreatic enzymes and bile that contain zinc are a source of zinc excretion in feces [18]. However, the primary purpose of pancreatic enzymes and bile is nutrient absorption, not zinc homeostasis, and the role of enterocytes in zinc excretion has not been well defined. Our discovery of zinc excretion from the apical plasma membrane of C. elegans intestinal cells mediated by a conserved CDF protein raises the possibility that this mechanism may be conserved in mammals.
Several mammalian CDF proteins are expressed in enterocytes, including ZnT1, ZnT2, ZnT4, ZnT5, ZnT6 and ZnT7 [19]. The protein sequence of ZnT2 is highly similar to TTM-1 isoforms, and ZnT2 has intriguing functional similarities. ZnT2 expression is induced by high dietary zinc, and two isoforms of ZnT2 result from alternative splicing; one isoform localizes to the plasma membrane, whereas and the other isoform localizes to intracellular endosome/lysosome-like vesicles [22], [42]. ZnT2 promotes zinc excretion from mammary epithelial cells [7]. Similarly, ZnT4 has been localized to the plasma membrane and intracellular vesicles, its subcellular localization changes in response high zinc conditions, and it functions in zinc secretion from mammary epithelial cells [19]. The ZnT5 gene encodes two isoforms, ZnT5A and ZnT5B, and each isoform exhibits a distinct subcellular localization; ZnT5A is localized to the Golgi complex, whereas ZnT5B is localized to the apical membrane of enterocytes in the small intestine [24], [43]. Functional studies of ZnT5B using cell lines and Xenopus oocytes suggest that ZnT5B may be a bidirectional zinc transporter that mediates both zinc influx and efflux [25]. These results suggest that ZnT2, ZnT4 and/or ZnT5B may function in the excretion of zinc from enterocytes into the intestinal lumen, but these functions have not been demonstrated in vivo.
To analyze how the activity of multiple zinc transporters is coordinated to promote homeostasis, we analyzed genetic interactions between ttm-1 and previously characterized cdf genes. ttm-1;cdf-2 double mutant animals displayed a dramatic induction of mtl-1/2 mRNA levels, indicating that the level of cytoplasmic zinc is highly elevated in intestinal cells. Thus, the coordinated actions of TTM-1B in zinc excretion and CDF-2 in zinc storage maintain cytoplasmic levels of zinc in intestinal cells. Consistent with this model, ttm-1;cdf-2 double mutant animals displayed extreme sensitivity to dietary zinc, indicating these genes function coordinately to promote zinc detoxification. Similar defects in zinc homeostasis were observed when the ttm-1 mutation was combined with mutations in glo genes that cause defects in gut granule biogenesis and zinc storage. These results confirm that the genetic interaction of ttm-1 with cdf-2 reflects the role of CDF-2 in zinc storage in gut granules of intestinal cells (Figure 9A). The genetic interaction between ttm-1 and cdf-2 was specific, since reducing the activity of ttm-1 did not strongly affect the zinc sensitivity of cdf-1 and sur-7 mutant animals.
Our results indicate that coordination between ttm-1 and cdf-2 occurs at two levels; one level is transcriptional control, and a second level is protein function. Here we demonstrated that reducing the activity of ttm-1 resulted in increased expression of cdf-2 mRNA. We previously showed that reducing the activity of cdf-2 resulted in increased expression of ttm-1 mRNA [30]. These results suggest that ttm-1 mutant animals and cdf-2 mutant animals have elevated cytoplasmic zinc in intestinal cells, which induces expression of zinc-regulated transcripts, including ttm-1 and cdf-2. At the level of protein function, both CDF-2 and TTM-1B function to reduce the level of cytoplasmic zinc, thereby enhancing zinc detoxification and reducing expression of zinc-regulated transcripts. This is a negative feedback circuit because rising levels of cytoplasmic zinc induce CDF-2 and TTM-1B protein expression, which in turn reduce the levels of cytoplasmic zinc. This is a parallel circuit since CDF-2 and TTM-1B proteins reduce cytoplasmic zinc by independent mechanisms: storage in lysosome-related organelles and excretion from intestinal cells, respectively. Because ttm-1b and cdf-2 are both zinc inducible, and both proteins promote low levels of cytoplasmic zinc, we propose that these genes constitute a parallel negative feedback circuit that maintains zinc homeostasis (Figure 9B).
Whereas cdf-2 single mutant animals displayed measurable zinc hypersensitivity, ttm-1 single mutant animals appeared to be similar to wild type in zinc sensitivity. One interpretation of these results is that zinc storage mediated by CDF-2 is the primary response to high zinc conditions and zinc excretion mediated by TTM-1B is the secondary response. The primacy of zinc storage relative to excretion may be a strategy for optimal handling of excess zinc. Zinc is an essential nutrient, and the availability of zinc may fluctuate in natural environments. When exposed to zinc abundance, it may be strategic for animals to first store zinc and later excrete excess zinc, thereby optimizing preparation for future exposure to zinc-deficient conditions. Therefore, the coordinated activity of CDF-2 mediated zinc storage and TTM-1B mediated zinc excretion may maintain zinc homeostasis and also optimize storage to promote survival in environments with fluctuating zinc levels.
C. elegans strains were cultured at 20°C on nematode growth medium (NGM) with a lawn of E. coli OP50 for food [44]. The wild-type strain and parent of all mutant strains was Bristol N2. The following mutations were used: pgp-2(kx48) I [40], glo-1(zu391) X [39], cdf-1(n2527) X [26], cdf-2(tm788) X [28] and sur-7(ku119) X [31]. ttm-1(ok3503) III was generated by the C. elegans Gene Knockout Project at OMRF, which is part of the International C. elegans Gene Knockout Consortium. We backcrossed ttm-1(ok3503) five times with N2 before analysis. The molecular lesion of ttm-1(ok3503) was defined by determining the DNA sequence of the ttm-1 locus that was PCR-amplified using the following primers: cccgccaaaaattattcaga and accgtaatgggacagacagc. Double mutant animals were generated by standard methods, and genotypes were confirmed by PCR or DNA sequencing.
To identify C. elegans CDF proteins, we conducted an iterative sequence-based homology search using PSI-BLAST [35]. Briefly, we searched open reading frames (ORFs) of C. elegans with the CDF-1 protein sequence and identified proteins with an E-value of 10−3 or less. These protein sequences were used to search for additional C. elegans ORFs with high similarity. To analyze evolutionary conservation, we used C. elegans CDF proteins to search ORFs of S. cerevisiae, A. thaliana and humans using PSI-BLAST. The identified orthologs were analyzed by reciprocal PSI-BLAST; each was used to search C. elegans ORFs to determine the most similar proteins. Multiple sequence alignment of CDF proteins was carried out using ClustalW, and the resulting alignment was used to generate a phylogenetic tree using MEGA [45].
RNA was isolated from synchronized wild-type animals at the adult stage using TRIzol (Invitrogen), and 5′RACE was performed using the 5′ RACE System for Rapid Amplification of cDNA Ends Version 2.0 according to the manufacturer's protocol (Invitrogen). Briefly, RNA was reverse-transcribed into cDNA using a gene specific primer (GSP1) which hybridized to exon 4 of ttm-1. cDNA was tailed with oligo(dC), PCR-amplified using Abridge Anchor Primer and another gene specific primer (GSP2) that is positioned approximately 30 bp upstream of GSP1, and fractionated by agarose gel electrophoresis. Two PCR products of different sizes were observed, and the DNA sequence of each was determined. GSP1 is gtaaccgaatgaaagacgct, and GSP2 is gagaattcaagacgtgcacaacgaatcg.
Eggs were isolated from gravid adult hermaphrodites by bleaching, allowed to hatch in M9 buffer overnight, and the worms were cultured on NGM dishes for approximately 2.5 days. Synchronized animals at L4/young adult stage were washed and then cultured on noble agar minimum medium (NAMM) dishes supplemented with zinc sulfate (ZnSO4) and seeded with concentrated OP50. After 16–24 hr, animals were washed and collected for RNA isolation. RNA analysis was performed as previously described with modifications [28]. Briefly, RNA was isolated using TRIzol (Invitrogen) and treated with DNase I, and cDNA was synthesized using the High-Capacity cDNA Reverse Transcription kit according to the manufacturer's protocol (Applied Biosystems). PCR was performed using an Applied Biosystems 7900 Fast Real-Time PCR System thermocycler and SYBR Green PCR Master Mix (Applied Biosystems). Fold change was determined by comparing target gene expression with the reference gene expression (ama-1 and rps-23) under the same conditions. The primers used for PCR were: ama-1, atcggagcagccaggaactt and gactgtatgatggtgaagctgg; rps-23, aaggctcacattggaactcg and aggctgcttagcttcgacac; ttm-1a, aacgttttcgacggaggagg and ctctctcgactctggcaacc; ttm-1b, catgggcactcacacacacac and ctcggcgacccttttgatatttc; cdf-2, atagcaatcggagagcaacg and tgtgacaattgcgagtgagc; mtl-1, ggcttgcaagtgtgactgc and cctcacagcagtacttctcac; mtl-2, ggtctgcaagtgtgactgc and gcagcagtattgctcacagc.
For the ttm-1a promoter fusion construct, the region between the ttm-1a start codon and the 3′ end of the adjacent upstream gene (∼1.2 kb) was PCR amplified using genomic DNA from wild-type animals. For the ttm-1b promoter, the region extending ∼6 kb upstream of the ttm-1b start codon was generated by PCR amplifying two fragments of DNA using genomic DNA from wild-type animals. We inserted these promoter regions, GFP coding sequence, and the unc-54 3′UTR into pBluescript SK+ (Stratagene) to generate pHR17 [Pttm-1a::GFP] and pHR6 [Pttm-1b::GFP]. For TTM-1 translational fusion constructs, ttm-1a cDNA was PCR amplified using EST clone yk1572h06 that was obtained from the National Institutes of Genetics in Japan. ttm-1b cDNA was generated by combining exon 3 and exon 4–5 fragments that were PCR amplified from EST clone yk1572h06 and wild-type genomic DNA, respectively. The appropriate cDNA (without the stop codon) was inserted into pHR17 and pHR6 to generate pHR4 [Pttm-1a::TTM-1A::GFP] and pHR7 [Pttm-1b::TTM-1B::GFP]. ttm-1b cDNA was also inserted into pSC7 [Pcdf-2::GFP] to generate pHR8 [Pcdf-2::TTM-1B::GFP]. For the cdf-1 translational fusion construct, the genomic DNA fragment containing the cdf-1 promoter and coding sequence were inserted with GFP and the unc-54 3′UTR into MM016 to generate pDP13 [Pcdf-1::CDF-1::GFP;unc-119(+)].
We generated transgenic animals containing extrachromosomal arrays by injecting hermaphrodites with pHR4, pHR6, pHR7, pHR8, or pHR17 and the coinjection marker pCJF104 [Pmyo-3::mCherry] and selecting F1 progeny expressing mCherry in body wall muscles. We generated transgenic animals containing integrated arrays by bombardment transformation with pDP13 into unc-119(ed3) animals [46]. We selected progeny that were non-Unc and segregated only non-Unc progeny.
For live fluorescence microscopy, animals were paralyzed in a drop of 10 mM levamisole in M9 buffer on 2% agarose pads on microscope slides. Immunostaining of transgenic animals was performed as described by Davis et al [28]. Briefly, animals were fixed in methanol and acetone, rehydrated, and stained with a rabbit anti-GFP antibody (Invitrogen) and an Alexa Fluor 488 goat anti-rabbit secondary antibody (Invitrogen). Fluorescence was visualized using a Zeiss Axioplan 2 microscope equipped with a Zeiss AxioCam MRm digital camera.
Zinpyr-1 (USBiological, Z0530) was reconstituted in dimethylsulfoxide (DMSO) to generate a 5 mM stock solution and diluted in M9 buffer to a final concentration of 20 µM. Hermaphrodites at the L4/young adult stage were cultured for 12–16 h on NAMM dishes supplemented with ZnSO4 and seeded with concentrated OP50. Animals were washed, transferred into 20 µM Zinpyr-1 in M9, and incubated for 3–4 h in the dark. Animals were washed three times with M9, paralyzed in 10 mM levamisole in M9, mounted on 2% agarose pads on microscope slides, and visualized as described above. For comparison, images were captured using identical settings and exposure times. To quantify Zinpyr-1 staining, we categorized individual worms into four groups based on visual inspection according to the fluorescence intensity and pattern; Low was defined as no detectable or weak fluorescence, Medium was defined as low level fluorescence in a wide area, High was defined as moderate fluorescence in a wide area and strong fluorescence in a small area, and Highest was defined as strong fluorescence in a wide area of the body.
Metal content analysis was performed as previously described [30], with modifications. For sample preparation, large populations of animals were generated by culturing on multiple 100 mm NGM dishes. Animals were washed and cultured on multiple 100 mm NAMM dishes supplemented with ZnSO4 and seeded with concentrated OP50. After ∼24 h, animals were washed three times in magnesium-free (Mg-free) M9 containing 0.01% Tween-20, incubated in 1 mM serotonin in Mg-free M9 for 30 min to remove bacteria from the intestinal lumen, washed twice in Mg-free M9, transferred to preweighed tubes and frozen at −80°C. For ICP-MS, samples were freeze-dried, reweighed to obtain the dry pellet weight, and digested by incubation in a hot block digester with concentrated nitric acid and hydrogen peroxide solution. The solution was diluted with water, and internal standards were added to correct for matrix effects. Instrument calibration standards were prepared from multi-element stock solutions (High-Purity Standards) to generate a linear calibration curve, and samples were analyzed using a Perkin Elmer NexION ICP-MS. The zinc content was determined as a value in parts-per-million (ppm) by dividing zinc weight by dry worm pellet weight (µg/g).
Zinc sensitivity assays were conducted as previously described [30]. Briefly, eggs were isolated from gravid adult hermaphrodites by treating with NaOH and bleach and allowed to hatch in M9 overnight. Synchronized L1 animals were then cultured on NAMM dishes supplemented with ZnSO4 and seeded with concentrated OP50. After ∼3 days, animals were washed, paralyzed with 10 mM sodium azide (NaN3) in M9, mounted on a 2% agarose pad on a microscope slide, and visualized as described above. The length of animals was measured using ImageJ software (NIH) by drawing a line from the nose to the tail tip.
|
10.1371/journal.pgen.1003265 | Massive Mitochondrial Gene Transfer in a Parasitic Flowering Plant Clade | Recent studies have suggested that plant genomes have undergone potentially rampant horizontal gene transfer (HGT), especially in the mitochondrial genome. Parasitic plants have provided the strongest evidence of HGT, which appears to be facilitated by the intimate physical association between the parasites and their hosts. A recent phylogenomic study demonstrated that in the holoparasite Rafflesia cantleyi (Rafflesiaceae), whose close relatives possess the world's largest flowers, about 2.1% of nuclear gene transcripts were likely acquired from its obligate host. Here, we used next-generation sequencing to obtain the 38 protein-coding and ribosomal RNA genes common to the mitochondrial genomes of angiosperms from R. cantleyi and five additional species, including two of its closest relatives and two host species. Strikingly, our phylogenetic analyses conservatively indicate that 24%–41% of these gene sequences show evidence of HGT in Rafflesiaceae, depending on the species. Most of these transgenic sequences possess intact reading frames and are actively transcribed, indicating that they are potentially functional. Additionally, some of these transgenes maintain synteny with their donor and recipient lineages, suggesting that native genes have likely been displaced via homologous recombination. Our study is the first to comprehensively assess the magnitude of HGT in plants involving a genome (i.e., mitochondria) and a species interaction (i.e., parasitism) where it has been hypothesized to be potentially rampant. Our results establish for the first time that, although the magnitude of HGT involving nuclear genes is appreciable in these parasitic plants, HGT involving mitochondrial genes is substantially higher. This may represent a more general pattern for other parasitic plant clades and perhaps more broadly for angiosperms.
| Recent studies have suggested that plant genomes have undergone potentially rampant horizontal gene transfer (HGT), especially in the mitochondrial genome. Here, using phylogenomic approaches, we demonstrate that as much as ∼40% of the mitochondrial genes in the parasitic plant species Rafflesiaceae are acquired from their hosts via HGT. These transgenes are likely functional in their recipient species and in some cases appear to have displaced native copies in the same genomic location. These results establish for the first time that, although the magnitude of HGT involving nuclear genes is appreciable in parasitic plants, HGT involving mitochondrial genes is substantially higher.
| Recent studies have suggested that plant genomes have undergone potentially rampant horizontal gene transfer (HGT) [1], [2], especially in the mitochondrial genome [3]–[7]. Parasitic plants have provided the strongest evidence of HGT [8]–[12], which appears to be facilitated by the intimate physical association between the parasites and their hosts [8], [10], [12]–[15]. One parasitic plant clade that appears to be prone to HGT is Rafflesiaceae sensu stricto, which belong to the order Malpighiales [8], [16]–[18] and whose members possess the largest flowers in the world. Rafflesiaceae are endophytic holoparasites, which lack leaves and stems. This family includes the genera Rafflesia (∼28 species), Rhizanthes (four species), and Sapria (three species), and provides one of the best opportunities to investigate HGT in plants because (i) the parasites have a very narrow host specialization range on members of the grapevine family (Tetrastigma spp., Vitaceae), (ii) complete genome sequences, including fully annotated mitochondrial and plastid genomes, are available for close relatives of the parasites (Ricinus communis, Euphorbiaceae) [19], [20] and their hosts (Vitis vinifera, Vitaceae) [21]–[23], and (iii) the hosts and parasites are separated by at least 115 million years of evolution (Figure 1A) [24]–[27]. These factors make it easier to distinguish transgenes from native genes in Rafflesiaceae using phylogenomic tools.
A recent phylogenomic study demonstrated that in Rafflesia cantleyi, about 2.1% of nuclear gene transcripts were likely acquired from its obligate host [28]. This study, however, did not include a thorough investigation of the mitochondrial genome. Here, we comprehensively sequenced 38 mitochondrial genes from R. cantleyi and five additional species, including two of its closest relatives and two host species. Our results reveal an extraordinarily high degree of HGT in the mitochondrial genome of Rafflesiaceae involving genes that were likely acquired from its host at various time intervals. Most of these transgenic sequences possess intact reading frames and are actively transcribed indicating that they are potentially functional. Additionally, some of these transgenes maintain synteny with their donor and recipient lineages suggesting that native genes in Rafflesiaceae have likely been displaced via homologous recombination. These results establish for the first time that although the magnitude of HGT involving nuclear genes is appreciable in these parasitic plants, HGT involving mitochondrial genes is substantially higher.
We used next-generation sequencing to comprehensively sequence the mitochondrial genomes of three species that span the crown node of Rafflesiaceae: Rafflesia cantleyi, Rafflesia tuan-mudae, and Sapria himalayana (Figure 1A, see also Table S1). We then extracted the 38 mitochondrial genes from our de novo assembled contigs that ranged in size from 2 to 54 kilobases (kb). These 38 protein-coding and ribosomal RNA genes are present in the mitochondrial genomes of both Ricinus and Vitis, and are also common to most angiosperms [29]. We included 35, 33, and 59 gene sequences from R. cantleyi, R. tuan-mudae and S. himalayana, respectively, for further analyses. While repetitive sequences made assembly of the entire chromosome impractical, high sequence coverage (Table S1) ensured that we have sequenced all coding regions in these mitochondrial genomes.
Several lines of evidence suggest that all gene sequences we assembled here are localized to the mitochondrial genome of Rafflesiaceae. First, the genome libraries for two of our three Rafflesiaceae species (i.e., R. cantleyi and S. himalayana) were prepared from fresh tissue using sucrose gradient centrifugation, which are enriched for plant organelles [30]. Since the plastid genome has apparently been lost in Rafflesiaceae [31], our libraries are heavily enriched for mitochondria. Second, plastid, mitochondrial and nuclear genes in plant cells differ widely in copy number: plastid genes are generally present in hundreds to thousands of copies per cell, mitochondrial genes in tens to hundreds of copies per cell, while nuclear genes are usually present in only two copies per cell [7], [32]. To investigate if gene sequences assembled here have copy numbers that correspond with a mitochondrial localization, we compared gene copy number here to 1,305 genes previously determined to be localized to the nuclear genome of R. cantleyi [28] and R. tuan-mudae [18]. Our results demonstrate that copy numbers for all putative mitochondrial gene sequences in Rafflesiaceae are one to two orders of magnitude greater than for nuclear genes (Table S2), with means of 155-, 68-, and 160-fold greater for R. cantleyi, R. tuan-mudae, and S. himalayana, respectively (p-value<2.2×10−16, Welch's t test). These copy numbers are consistent with a mitochondrial localization, but not high enough to suggest localization in, or the existence of, a plastid genome in Rafflesiaceae. Third, when comparing assembled gene sequences from R. cantleyi with our previously published complementary DNA (cDNA) library [28], we identified cytosine-to-uracil (C-to-U) RNA editing in seven genes (i.e., atp1, atp4, atp6, cox2, nad1, rps4, and rps12), which is a common characteristic of mitochondrial genes [33]. These results collectively indicate that these gene sequences are most likely localized to the mitochondrial genome of Rafflesiaceae, although complete assembly of these mitochondrial genomes will be required to definitively confirm our results.
To estimate the magnitude of HGT in the mitochondrial genome of Rafflesiaceae, we also sequenced the same 38 mitochondrial genes from three species of Vitaceae: Tetrastigma cruciatum, which is the host of S. himalayana [34], Tetrastigma rafflesiae, which is the host of R. cantleyi and R. tuan-mudae [34], [35], and Leea guineensis (Figure 1A, see also Table S1). The latter represents the earliest diverging lineage of Vitaceae [36], [37], which allows us to determine if putative transgenic sequences from Rafflesiaceae are phylogenetically nested within the host clade Vitaceae.
Our newly sequenced mitochondrial gene sequences from the six species were then analyzed using maximum likelihood (ML) with homologous sequences from 27 other seed plants whose mitochondrial genomes have been sequenced and fully annotated (Figure 1A, see also Table S3; Arabidopsis thaliana [Brassicaceae], Beta vulgaris [Amaranthaceae], Boea hygrometrica [Gesneriaceae], Brassica napus [Brassicaceae], Carica papaya [Caricaceae], Citrullus lanatus [Cucurbitaceae], Cucumis sativus [Cucurbitaceae], Cucurbita pepo [Cucurbitaceae], Cycas taitungensis [Cycadaceae], Daucus carota [Apiaceae], Lotus japonicus [Fabaceae], Malus domestica [Rosaceae], Millettia pinnata [Fabaceae], Mimulus guttatus [Phrymaceae], Nicotiana tabacum [Solanaceae], Oryza sativa [Poaceae], Phoenix dactylifera [Arecaceae], Raphanus sativus [Brassicaceae], Ricinus communis, Silene latifolia [Caryophyllaceae], Sorghum bicolor [Poaceae], Spirodela polyrrhiza [Araceae], Tripsacum dactyloides [Poaceae], Triticum aestivum [Poaceae], Vigna radiata [Fabaceae], Vitis vinifera, and Zea mays [Poaceae]). These reference species represent a broad sampling of most major flowering plant clades [38]. Each Rafflesiaceae gene sequence was placed into one of three categories–i.e., VGT, HGT, or unassigned–on the basis of its phylogenetic position and ML bootstrap percentage (BP) support following Xi et al. [28]. We applied two BP thresholds to categorize each gene sequence. Our more conservative estimate applied a 70 BP threshold; this BP threshold has been shown to correspond to a very high probability that the clade is real [39]. Here, gene sequences whose placements were consistent with accepted species' relationships (i.e., Rafflesiaceae gene sequences were sister to their closest relative Ricinus with ≥70 BP; [17], [18]) were scored as VGT; HGT was inferred when gene sequences were placed elsewhere with ≥70 BP; and gene sequences with <70 BP were left unassigned. To explore if our estimates of HGT were sensitive to our thresholds, we also categorized these gene sequences by applying a less conservative threshold using ≥50 BP.
Our phylogenetic analyses of the 38 mitochondrial genes indicated that, for the 30 autotrophic species included here (i.e., 27 reference species, two Tetrastigma species, and Leea; Figure 1A), phylogenetic placements largely agreed with accepted relationships between families [38], [40] using both the 70 and 50 BP thresholds (Figures S1A and S2). The only three exceptions were for atp1 where Brassicaceae (i.e., Arabidopsis+Brassica+Raphanus) was placed sister to the asterids (i.e., Boea+Mimulus+Nicotiana) with 81 BP (a similar topology was also identified by Nickrent et al. [10]); for atp4 where Brassicaceae was placed sister to Fabaceae (i.e., Lotus+Millettia+Vigna) with 93 BP; and for cox1 where Brassicaceae was placed sister to Caryophyllales (i.e., Beta+Silene) with 87 BP (Figure S1A). These results indicate that applying both the 70 and 50 BP thresholds yields very low false positive estimates of HGT in these autotrophic species, for which we expect little or no HGT to occur.
In contrast, in the three holoparasitic Rafflesiaceae species, 11 mitochondrial genes demonstrated evidence for one or more cases of HGT using our more conservative 70 BP threshold: of the 21 gene sequences with ≥70 BP in R. cantleyi, five gene sequences (24%) showed evidence of HGT, 5 of 19 (26%) in R. tuan-mudae, and 11 of 27 (41%) in S. himalayana. Furthermore, vertical placements of these putative transgenic sequences were rejected in 18 of 21 cases using the approximately unbiased (AU) test (Table 1). For the less conservative 50 BP threshold, the number of mitochondrial genes that showed evidence of HGT increased to 16; however, the relative frequencies of HGT are nearly identical with those above: 29% in R. cantleyi (7 of 24), 32% in R. tuan-mudae (7 of 22), and 47% in S. himalayana (16 of 34). This indicates that our less conservative threshold does not increase false positive rates. Thus, given the consistency of our estimates of HGT using both thresholds, we treat these transgenes collectively in the discussion below unless otherwise indicated. Two additional findings support the reliability of our HGT inferences: first, the phylogenetic placements of these transgenic sequences were not obviously biased by C-to-U RNA editing (see Figure S1B for phylograms with RNA editing sites excluded from our alignments); second, seven of our large assembled contigs contained both transgenes and native genes (Figure 2), indicating that these transgenes were clearly integrated into the mitochondrial genome of Rafflesiaceae. Therefore, rates of mitochondrial HGT in Rafflesiaceae appear to be extraordinarily high, and well above the false positive rates established from the 30 autotrophic species included here.
Of the 11 mitochondrial genes that showed evidence of HGT using our more conservative threshold, four (i.e., cob, cox3, rpl5, and rps4) maintained both horizontally and vertically transferred homologs, and seven included only transgenic sequences (i.e., atp4, cox1, cox2, rps1, rps7, rps13, and sdh3) (Table 1). An additional five mitochondrial genes showed evidence of HGT using our less conservative threshold, one of them (atp9) maintained both horizontally and vertically transferred homologs, and four included only transgenic sequences (i.e., atp1, rpl2, rps14, and sdh4) (Table 1). Of those genes that included only transgenic copies in Rafflesiaceae all had homologs present in the mitochondrial genome of Ricinus, which suggests that they were likely present as native copies ancestrally in Rafflesiaceae and were subsequently displaced by transgenic homologs. One example is illustrated by our assembled contig containing the genes nad5 exons A and B and rps7 (Figure 2A). In Rafflesiaceae, nad5 exon B was identified as a native sequence (84 BP) in our phylogenetic analyses, while rps7 was identified as a transgene (96 BP). The phylogenetic placement of nad5 exon A within Vitaceae is also consistent with HGT, but support for this placement is <50 BP (Figure 2A, see also Figure S1C). However, the synteny of nad5 exons A and B is conserved among Ricinus, Vitis, and all three Rafflesiaceae species, suggesting that the native copy of nad5 exon A in Rafflesiaceae may have been displaced by a horizontally transferred DNA fragment via homologous recombination [41]. This hypothesis is further supported by the fact that nad5 exon A is immediately adjacent to the well-placed transgene rps7 in Rafflesiaceae, which exactly matches the synteny of Vitis but not Ricinus. To better locate the recombination breakpoint, we analyzed the intron region between nad5 exons A and B. This ∼1-kb region is highly conserved across angiosperms and can be easily aligned for phylogenetic analysis. We found that nad5 intron A/B was clearly identified as a native sequence (82 BP; Figure 2A, see also Figure S1C), therefore, the breakpoint is likely very close to the junction of nad5 exon A and intron A/B. Although the integration of foreign DNA via homologous recombination is common in bacteria [41], reports of this phenomenon are rare for plants (e.g., atp1 gene [42] and rps11 gene [1], [4]). Such direct homologous recombination, which is likely facilitated by the intimate physical association between Rafflesiaceae and their hosts [14], [15] combined with the frequent fusion of plant mitochondria [1], [43]–[45], may obviate the need to invoke a transposable element, bacterium, or virus for catalyzing the insertion of a DNA fragment from donor to recipient in plants.
Additionally, in the mitochondrial genome of S. himalayana, we found evidence of HGT involving 14 genes that were potentially of plastid origin using our more conservative threshold, only one of which was also identified in R. tuan-mudae (Table 1, see also Figure S1D). Thirteen of these genes support the conclusion that they were acquired via host-to-parasite HGT, because in each case S. himalayana is placed sister to, or nested within, Vitaceae with ≥70 BP. Only atpA from S. himalayana and R. tuan-mudae were placed elsewhere phylogenetically, sister to Daucus with 94 BP. Furthermore, for six of these 14 genes (i.e., atpB, atpI, psaB, psbA, psbC, and psbD), the transgenic sequences from S. himalayana were sister to the mitochondrial and not the plastid homologs from Vitaceae (Figure S1D). These six genes, plus three additional plastid genes (i.e., atpA, ndhB, and rbcL), have been shown to be incorporated into the mitochondrial genome of Vitis [23]. Together, these results suggest that the majority of these plastid genes were likely acquired via HGT from the host mitochondrial genome, instead of from its plastid genome.
Finally, seven of our assembled contigs demonstrated that synteny was maintained between transgenes from Rafflesiaceae and genes from the close relative of their hosts, Vitis, whose mitochondrial and plastid genomes were both fully annotated (i.e., rps7+nad5 exon A, psbC+psbD, sdh3+rpl5+rps14+cob+cox1, rpoC1+rpoC2, and sdh4+cox3; Figure 2 and Figure S3). This, combined with our finding that two transgenes bear introns (i.e., cox1 and rpl2), firmly supports our previous suggestion that transgenes in Rafflesiaceae are likely transferred as larger DNA fragments versus shorter mRNAs [28].
Most previously reported mitochondrial transgenes in plants appear to be non-functional, i.e., they have been shown to be either introns (e.g., [3], [6], [8], [46], [47]) or pseudogenes (e.g., [7], [9]). However, among all mitochondrial transgenes identified here, six of seven sequences in R. cantleyi and R. tuan-mudae, and 13 of 16 sequences in S. himalayana maintain their reading frames (Table 1). To further understand if these transgenes are expressed, we re-examined the recently published transcriptome of R. cantleyi [28] to quantify gene expression levels of these mitochondrial genes. Our results indicate that all transgenes in R. cantleyi show evidence of expression (Table 1). Furthermore, although native genes in R. cantleyi show higher overall levels of gene expression than transgenes (Figure S4), this difference is not significant (p-value = 0.19, Welch's t test). Thus, transgenes are actively transcribed in this species, suggesting that they have functional promoters and likely play a role in cellular function.
Our broad phylogenomic assessment of mitochondrial genome provides a unique opportunity to determine if HGT we identified in Rafflesiaceae is relatively ancient or more recent. For five genes that show evidence of HGT (i.e., atp1, atp4, cox1, rps7, and atpA; Figure S1A and S1D), it is most parsimonious to infer that they each result from an ancient HGT event. The more ancient origin is supported by the fact that transgenic sequences from Rafflesia and Sapria form a clade. Furthermore, we found that some of these transgenes maintained synteny between Rafflesia and Sapria (e.g., atp1 and rps7; Figure 2A and 2B). Therefore, these gene transfers appear to have been relatively ancient and likely occurred after the origin of stem group Rafflesiaceae (95% highest posterior density [HPD] interval of 83.1–109.5 Ma; [17], [48]) and before the origin of crown group Rafflesiaceae (69.5–95.9 Ma; [17], [48]). Both of these estimated clade ages, accounting for 95% HPD intervals, are outside the age of stem group Tetrastigma (36.4–65.3 Ma; [49]), and well outside the age of crown group Tetrastigma (25.7–49.3 Ma; [49]) (Figure 1B).
This raises the distinct possibility that Rafflesiaceae has had former host associations with other plant lineages (perhaps within Vitaceae, but also outside of the family), which may have served as past donors of transgenes. We have previously referred to this as the ghost of HGT's past [50]. In support of this possibility, none of the more ancient transgenic sequences we identified grouped with their current hosts Tetrastigma, as would be expected if these species served as hosts. Two genes (i.e., atp1 and rps7) involved in these more ancient HGT events are sister to Vitaceae, suggesting that close relatives of Tetrastigma may have served as past transgenic donors. In three other cases (i.e., atp4, cox1, and atpA), however, transgenic sequences do not group closely to Vitaceae (e.g., Cucurbitaceae and Daucus; Table 1), indicating different transgenic donors (no evidence of gene conversion, which would confound phylogenetic placements, was detected in these genes using the OrgConv package [51] with p-value<0.001). To our knowledge, this is the first evidence that Rafflesiaceae may have previously parasitized different host species, which served as transgenic donors in the past. Further taxon sampling of these genes by co-authors Z.X. and Y.W. is underway and should allow us to determine those previous host donors more precisely.
For the remaining 28 instances of HGT, it is most likely that these were the result of more recent gene transfers. Transgenic sequences in these cases are found exclusively in either Rafflesia (i.e., rps4) or Sapria (e.g., atp9 and cob), or if identified in both Rafflesia and Sapria they do not form a clade (i.e., rpl2 and rps1) (Figure S1A and S1D). Evidence of such recent HGT is especially prevalent in S. himalayana: 17 of its transgenic sequences are sister to, or nested within, Tetrastigma (Table 1). Moreover, our phylogenetic analyses indicate that some of these sequences may have resulted from multiple independent gene transfers involving the same gene because transgenic sequences from Rafflesiaceae do not form a clade. In some cases, these gene transfers appear to involve multiple transgenic sequences within a single species for the same gene (i.e., cox1 in Sapria, which possesses two distinct transgenic sequences that appear to have been transferred independently). In other cases, gene transfers involve multiple transgenic sequences in different species for the same gene (i.e., rpl2 and rps1, which show independent transfer events for Rafflesia and Sapria). These more recent HGT events are further supported by synteny: transgenic sequences involving the same gene from Rafflesia and Sapria are located at different positions in the mitochondrial genome (e.g., rpl2; Figure 2A and 2C). Why some genes exhibit repeated HGT is fertile ground for future investigation.
Our study is the first to comprehensively assess the magnitude of HGT in plants involving a genome (i.e., mitochondria) and a species interaction (i.e., parasitism) where it has been hypothesized to be potentially rampant. These results reveal a high degree of HGT in the mitochondrial genome of Rafflesiaceae involving genes that were likely acquired from its host at various time intervals. We previously established that in R. cantleyi, about 2.1% of nuclear gene transcripts have likely been acquired from its host via HGT [28]. In contrast, our study conservatively indicates that 24–41% of the mitochondrial gene sequences show evidence of HGT in Rafflesiaceae, depending on the species. These results establish for the first time that although the magnitude of HGT involving nuclear genes is appreciable, HGT involving mitochondrial genes in these parasitic plants is an order of magnitude higher. This elevated rate of HGT involving the mitochondrial genome may represent a more general pattern for other parasitic plant clades, and perhaps more broadly for angiosperms.
For R. cantleyi and S. himalayana, mitochondria were isolated from ∼30 grams of fresh material from flower buds using the sucrose gradient centrifugation protocols of Jansen et al. [30]. DNA extracted from purified mitochondria was amplified with the REPLI-g Midi Kit (Qiagen, Inc.). When we were unable to acquire fresh material, total genomic DNA (gDNA) was extracted from silica-dried material using the DNeasy Plant Mini kit (Qiagen, Inc.), and treated with RNase A at 60°C for 1.5 hours to remove any residual RNA contamination. For each species, an Illumina library with the insert size of 350±50 bp was prepared from five micrograms of DNA following the protocols of Bentley et al. [52]. All libraries were sequenced on the Genome Analyzer II (Illumina, Inc.) with 100 bp paired-end runs at the FAS Center for Systems Biology at Harvard University (Table S1).
Illumina reads were assembled de novo in ABySS v1.2.1 [53] using default parameters (Table S1). The assembled contigs were annotated against published mitochondrial and plastid genomes from 27 seed plants (Table S3) with BLASTN v2.2.23 [54] using an e-value ≤10−5. Gene sequences from all species were then queried against themselves using BLASTN v2.2.23. BLASTN hits with an e-value ≤10−10 were passed to MCL v08-312 [55] for Markov clustering. Only those gene clusters that included at least Cycas/Spirodela (for outgroup rooting), Rafflesia/Sapria (sequences under investigation), Ricinus (close relative of Rafflesiaceae), and Vitis (close relative of Tetrastigma) were retained. The nucleotide sequences of each gene were first aligned using MAFFT v6.624 [56], and then manually inspected and realigned if necessary.
To assess gene copy number and corresponding genomic compartment localization of our assembled gene sequences, we mapped the Illumina gDNA reads from R. cantleyi [28], R. tuan-mudae, and S. himalayana to gene sequences identified here and to the 1305 nuclear genes identified from R. cantleyi [28] and R. tuan-mudae [18] using Bowtie v0.12.7 [57] (Table S2). To avoid complications with intron regions, we first divided each Illumina read into multiple 25 bp fragments following Kim and Salzberg [58], and then mapped each 25-mer with zero mismatches and unique mapping.
Our ML analyses were conducted for all genes using RAxML v7.2.8 [59] with the GTR+Γ nucleotide substitution model. The best-scoring ML tree and BP for each gene were obtained using the rapid bootstrap algorithm [60] with 500 replicates (Figures S1 and S2). For nad5, we also performed ML analyses on three gene regions separately (i.e., nad5 exon A, intron A/B, and exon B; Figure 2A, see also Figure S1C), which allowed us to determine the location of homologous recombination more accurately (see above).
Alternative topology tests were performed in an ML framework using the approximately unbiased (AU) test [61] as implemented in scaleboot v0.3-3 [62] (Table 1). To generate constrained ML trees for genes that show evidence of HGT, we enforced all transgenic and native (when present) sequences from Rafflesiaceae to be monophyletic with Ricinus, and then conducted ML searches using these constraints.
To estimate the gene expression level in R. cantleyi, the Illumina cDNA reads from Xi et al. [28] were mapped onto the assembled R. cantleyi mitochondrial gene sequences using Bowtie v0.12.7 [57] as described above. cDNA reads that mapped onto each gene sequence were then summed and further normalized to reads per kilobase per million reads (RPKM [63]; Table 1, see also Figure S4).
Tremendous care was taken to avoid and/or detect host or lab contamination during our sample preparation and data analyses. First, our DNA sample preparation and genome library sequencing of Rafflesiaceae were performed separate from any work involving Tetrastigma; thus, laboratory contamination of our Rafflesiaceae DNAs with Tetrastigma is unlikely. Second, the plastid genome has apparently been lost in Rafflesiaceae [31]. If there were any host contamination, the host's plastid gene sequences should be easily detected in our sequence data. This was not the case. Third, the mitochondrial genome sequences of R. cantleyi and R. tuan-mudae were generated from two different sources, i.e., a fresh flower bud using sucrose gradient centrifugation and silica-dried perigone lobes using total gDNA extraction, respectively (Table S1). If one of these samples, or genome libraries, were contaminated, we would not expect to have identified the identical set of transgenes from these samples. Similarly, for S. himalayana, all transgenes identified from the genome library prepared using sucrose gradient centrifugation were verified in our second library of this species that was prepared from total gDNA (Table S2). Fourth, most transgenic sequences identified here possess some amount of sequence divergence when directly compared with homologs from their current host. For example, all 15 transgenic sequences from Rafflesia show some degree of sequence divergence when directly compared with homologs from their host species, T. rafflesiae (mean DNA sequence distance = 0.042189). Similarly, 26 of 30 transgenic sequences from Sapria show some degree of sequence divergence when directly compared with homologs from their host species, T. cruciatum (mean DNA sequence distance = 0.020265) (Figure S1A and S1D). These sequence distances are significantly greater (p-value<0.01, Welch's t test) than those between the two included Tetrastigma species (mean DNA sequence distance = 0.001984). This is despite the fact that these two Tetrastigma species have diverged from each other at least 10 Ma [49]. Furthermore, three transgenic sequences from Rafflesia and 13 transgenic sequences from Sapria contain nonsense mutations (Table 1). These results strongly indicate that some period of evolution has elapsed since the time of HGT. Fifth, all seven transgenes from R. cantleyi show evidence of gene expression based on its transcriptome (Table 1), and levels of expression are not significantly different between transgenes and native genes (p-value = 0.19, Welch's t test; Figure S4). Lastly, and perhaps most importantly, seven of our assembled contigs contain both transgenes and native genes (Figure 2) indicating that these transgenes are clearly integrated into the mitochondrial genome of Rafflesiaceae.
|
10.1371/journal.pcbi.1003876 | Canonical Correlation Analysis for Gene-Based Pleiotropy Discovery | Genome-wide association studies have identified a wealth of genetic variants involved in complex traits and multifactorial diseases. There is now considerable interest in testing variants for association with multiple phenotypes (pleiotropy) and for testing multiple variants for association with a single phenotype (gene-based association tests). Such approaches can increase statistical power by combining evidence for association over multiple phenotypes or genetic variants respectively. Canonical Correlation Analysis (CCA) measures the correlation between two sets of multidimensional variables, and thus offers the potential to combine these two approaches. To apply CCA, we must restrict the number of attributes relative to the number of samples. Hence we consider modules of genetic variation that can comprise a gene, a pathway or another biologically relevant grouping, and/or a set of phenotypes. In order to do this, we use an attribute selection strategy based on a binary genetic algorithm. Applied to a UK-based prospective cohort study of 4286 women (the British Women's Heart and Health Study), we find improved statistical power in the detection of previously reported genetic associations, and identify a number of novel pleiotropic associations between genetic variants and phenotypes. New discoveries include gene-based association of NSF with triglyceride levels and several genes (ACSM3, ERI2, IL18RAP, IL23RAP and NRG1) with left ventricular hypertrophy phenotypes. In multiple-phenotype analyses we find association of NRG1 with left ventricular hypertrophy phenotypes, fibrinogen and urea and pleiotropic relationships of F7 and F10 with Factor VII, Factor IX and cholesterol levels.
| Pleiotropy appears when a variation in one gene affects to several non-related phenotypes. The study of this phenomenon can be useful in gene function discovery, but also in the study of the evolution of a gene. In this paper, we present a methodology, based on Canonical Correlation Analysis, which studies gene-centered multiple association of the variation of SNPs in one or a set of genes with one or a set of phenotypes. The resulting methodology can be applied in gene-centered association analysis, multiple association analysis or pleiotropic pattern discovery. We apply this methodology with a genotype dataset and a set of cardiovascular related phenotypes, and discover new gene association between gene NRG1 and phenotypes related with left ventricular hypertrophy, and pleiotropic effects of this gene with other phenotypes as coagulation factors and urea or pleiotropic effects between coagulation related genes F7 and F10 with coagulation factors and cholesterol levels. This methodology could be also used to find multiple associations in other omics datasets.
| Pleiotropy refers to a phenomenon in which a single locus affects two or more apparently unrelated phenotypic traits. It is often identified as a single mutation that affects these two or more wild-type traits [1]. The study of pleiotropic genes usually involves the mapping of phenotypic traits to a single mutant locus. When two or more traits consistently segregate with a particular mutation, this mutation is then classified as pleiotropic. In the case of S. cerevisiae (yeast), it has been argued that the pleiotropic effects of a gene are not usually conferred by multiple molecular functions of the gene, but by multiple consequences (biological processes) of a single molecular function [2]. Tyler et al. [3] defined the concept of vertical and horizontal pleiotropy, extending the definition of relational and mosaic pleiotropy proposed by Hadorn and Mittwoch [4]. Vertical or relational pleiotropy appears when a mutation in one gene produces a modification of one particular phenotype, which leads to modification in one or several related phenotypes. By contrast, horizontal or mosaic pleiotropy appears when one mutation in one gene with a causal implication in several biological mechanisms, causes a disruption in these mechanisms. This causes alteration in very different phenotypes, which are observable at the same physiological level. Some papers [5] [6] [7] have established a high level of pleiotropy for certain genes, particularly genes associated with disease [8].
To discover such associations we could use a range of multivariate techniques which highlight the dependence of a single variable on a set of independent variables. Some proposals are based on combining univariate association measures for different phenotypes in order to find pleiotropic effects, such as PRIMe [9] or Yang et al's approach [10], based in O'Brien's method [11]. An approach taken by Li [5], uses Fisher's combined p-value approach [12], adjusting Fisher's combined measure using a Satterwhite approximation method. Other approaches use a Bayesian network approach [13] or multiple regression analysis [14]. However, for the purposes of pleiotropy analysis, we are most interested in finding dependencies between two multivariate sets of variables, rather than a relation of one set with one dependent variable. Various techniques have also been introduced to deal with such multivariate problems [15]. An example of this multiple SNP/multiple phenotype analysis is GUESS [16], which is an implementation of a Bayesian variable selection algorithm for multiple regression using evolutionary Monte Carlo techniques: the algorithm selects relevant SNPs and identifies the contribution of each SNP to single or multiple traits.
In this paper, we will focus on Canonical Correlation Analysis (CCA) [17], which uses linear combinations of variables derived from two sets of data objects and finds those combinations which are maximally correlated with each other. The variables found in the first iteration of the method give the first set of canonical variables. In subsequent iterations we seek variables which maximize the same correlation function, subject to the constraint that they are uncorrelated with previous sets of canonical variables.
CCA has been used as an efficient and powerful approach for both univariate and multivariate gene-based association tests. For genomic multivariate data analysis, such an approach would involve finding linear combinations over very large blocks of features, typically involving tens of thousands of features. However, to use CCA, the number of samples should be more than the number of features. To handle this issue, some solutions have been proposed for genomic data integration, such as sparse CCA [18]. With this approach, sparsity is intrinsically achieved by the algorithm so that the number of features used is less than the sample size. This method maximizes the correlations between these selected subsets using a regularization procedure similar to LASSO. Adaptive SCCA [19] selects fewer features which are more correlated and Waaijenborg et al. [20], propose a method called penalized CCA to find associations between gene expression and copy number variation data. Other variants on CCA which are applicable include non-linear extensions of CCA, such as kernel CCA [21], [22], Bayesian approaches to CCA [23], [24] and sparse CCA models for handling more than two types of data [25].
CCA for association analysis was proposed initially by Ferreira and Purcel [26] and subsequently extended [27]. Both these papers apply CCA to multiple trait/single genotype analysis (pleiotropy analysis), while the latter also considers the case of several markers (gene centered pleiotropy analysis) and several traits, or several markers and one trait (epistasis analysis). Since the original publication [26], CCA has been used for multiple association analysis elsewhere, including a single SNP, multiple phenotype association approach [28] to analyze blood phenotypes related with metabolic syndrome in mice, and use of a sparse version of CCA to discover associations between single locus and multiple neuroimage phenotypes [29]. Further applications of CCA include a study [26] of pleiotropy in white cell related traits using a single locus/multiple trait approach, and use of CCA for single SNP/multiple trait analysis to find different child behavior profiles [30].
In this paper we propose an alternative approach for using CCA in which we select feature sets via biological insight, based on association with a gene, a pathway or another biologically relevant grouping. As detailed below, to maximize the association between genetic data and different phenotypes, we combine the CCA approach proposed by Ferreira and Purcell [26] with an optimization technique, drawn from integer programming. We will refer to any discovered significant associations between subsets of the genetic and phenotype data as putative association rules.
Our results are divided into (a) single gene/single phenotype; (b) single gene/multiple phenotype (in which the algorithm identifies the set(s) of phenotypes associated with a single gene); (c) multiple gene/single phenotype (in which the algorithm identifies the set(s) of genes associated with a single phenotype); (d) multiple gene/multiple phenotype (in which the algorithm selects sets of both genes and phenotypes that correlate).
This approach consists of a gene centered association analysis with each single phenotype using simple CCA without any search heuristic. It is exactly the same approach used previously by Tang and Ferreira [27], consisting of a multiple association of all the SNPs close to a gene (see Methods for more detail) with a particular phenotype. In order to correct for multiple testing, we use a Bonferroni correction for 3648 genes and 82 phenotypes, giving a “threshold” p-value of 1.67×10−06 corresponding to p = 0.05 for a single test. We found 62 genes with significant association (p<1.67×10−06). Most of the time this association reflects the most associated SNP in a gene. The most important associations are presented in the hive plot in Figure 1.
In Table 1 we show some of the associations found and publications that supports these findings. All such associations can be found in Table S1, where we compare the association values between this approach and conventional single SNP association tests.
Although most genes have more than one associated SNP reading, we found a non-reported association (p = 4.93×10−10) between the single SNP rs11264341, located in the intronic region of gene TRIM46 (ENTREZ GENE # 80128), and the serum magnesium phenotype. This SNP is in LD with SNP rs4072037 (r2 = 0.54) in MUC1, which has been previously related with serum magnesium. Close to this SNP (>8 kb), and not in LD, we found an association in gene MUC1 (ENTREZ GENE # 4582) with serum magnesium (p = 1.37×10−14) which has been previously reported by Meyer et al. [31].
We found a previously reported association of gene SURF4 (ENTREZ GENE # 6836) and Von Willebrand Factor (vWF), but also a non-previously-reported association with Factor VIII (1.57×10−24) and alkaline phosphatase (ALP) (3.1×10−9). SNPs in SURF4 are in some LD (r2 = 0.696) with SNP in C9orf96, which has been related with vWF [32], also with SNPs in ABO(ENTREZ GENE # 28) (r2 = 0.502) which has been related with vWF [32] and ALP [33] [34]. We could expect the association between Factor VIII and vWF, because there is a high correlation between its serum concentrations (0.70), and vWF acts as a carrier protein of Factor VIII. However, the correlation between ALP and vWF are 0.14. This is a clear example of vertical pleiotropy, where variants in SURF4 are causal of vWF, and vWF glycoprotein is the carrier for Factor VIII glycoprotein in blood. Another important association (p = 2.18×10−8), which has not been reported, is between gene NSF (ENTREZ GENE # 4905) and triglycerides. NSF is related with genes KIAA1377 (ENTREZ GENE # 57562) and LUC7L2 (ENTREZ GENE # 51631), through the PPI network, which are also related with the LPL gene (ENTREZ GENE # 4023). Finally the MYBPHL gene is associated (p = 3.19×10−8) with low density lipoprotein (LDL) cholesterol, which has not been previously reported. However, SNPs in this gene are in LD with SNPs in CELSR2 (ENTREZ GENE # 1952) (r2 = 0.546), PSRC1 (ENTREZ GENE # 84722) (r2 = 1) and SORT1 (ENTREZ GENE # 6272) (r2 1) rs12740374, which is associated [35], [36], [37], [38], [39] with LDL cholesterol.
Left ventricular hypertrophy can be detected through ECG parameters such as Cornell product [40] or QRS product [40], [41]. Using the CCA gene-centered association approach we have identified a number of genes associated with these two clinical parameters, which are also positively associated with cardiovascular diseases such as stroke [42]. We found association between ACSM3 (ENTREZ GENE # 6296) and Cornell product (p = 2.38×10−8). This gene was previously reported to associate with hypertension in rats [43] and in humans [44] and also with obesity hypertension in humans [45], but there is some controversy [46]. Other studies relate it with ventricular deformations such as left ventricular mass index and mean wall thickness [47]. The ERI2 gene was also associated with Cornell product (p = 7.87×10−9). This gene overlaps ACSM3 (ERI2 SNPs is a subset of ACSM3). No association with left ventricular hypertrophy or hypertension has been reported previously. IL18RAP (ENTREZ GENE # 8807) was associated with Cornell product (p-value 1.07×10−8) and QRS voltage product (p-value 1.72×10−10). SNPs in this gene have been associated [48] with echocardiography left ventricular obtained measures. In Grisoni et al. [49], using different SNPs in the same gene, the authors did not find any association between IL18RAP and any cardiovascular diseases (CVD) risks. However, Tiret et al. [50] found a significant association between IL18 family gene SNPs and mortality. We found association between IL23R (ENTREZ GENE # 149233) and Cornell product (3.3×10−12). This gene has been associated with left ventricular hypertrophy [51] and idiopathic dilated cardiomyopathy in Chen et al. [52]. It is interesting to note the importance of autoimmune related genes (IL18RAP and IL13R) in left ventricular hypertrophy or idiopathic dilated cardiomyopathy. A relation between autoimmune response and idiopathic dilated cardiomyopathy has been suggested in San Martin et al. [53] and Lappe et al. [54].
Finally, gene NRG1 (ENTREZ GENE # 3084) presents an association with phenotypes Cornell product (2.71×10−14) and QRS voltage product (2.97×10−11). This gene has been associated to cardiovascular development in mouse [55], through the NRG1/ErbB signaling pathway [56], [57], that is involved in angiogenesis, blood pressure and skeletal muscle response to exercise. In humans, serum NRG-beta has been found elevated in patients with severe systolic heart failure [58]. In McBride et al. [59], no association was found between SNPs in NRG1 and a group of congenital heart malformations (left ventricular outflow tract, defects of aortic valve stenosis, coarctation of the aorta and hypoplastic left syndrome). One of the drawbacks of CCA analysis, which could affect our understanding of the role of NRG1, is that this method lacks power when a gene is larger than 100 Kb [27], and NRG1 has a length of 1.1 Mb.
In order to analyze the association of all the SNPs in one gene and multiple phenotypes, we use CCA and a genetic algorithm as an optimization method, to select the most important phenotypes, as described in the Methods section.
In Table 2 we show some of the most important pleiotropic genotype/multiple phenotype associations, including the p-value of CCA association and the phenotypes with which they are associated. We also show Fisher's combined association value and, in parentheses, the association value of the genes and the single phenotype. In Table S2 we show all the results for associations between one gene/multiple phenotypes. In order to correct for multiple associations, we use a Bonferroni correction for 3648 genes and combinations of 82 phenotypes in subsets of 24 to 2 groups. We chose 24 because it is the maximum number of different phenotypes in one association rule (an association rule is a combination of a number of phenotypes associated with a number of genes) selected by the genetic algorithm (see the multiple test association correction paragraph in Methods). This combination gives 5.36×1020 different phenotypic rules, giving a threshold p-value of 2.55×10−25 equivalent to p = 0.05 for a single test. In Figure 2, we use a heatmap plot to represent the most important (higher association) pleiotropic relations between phenotypes and genotypes. Also, we use a hive plot (interactive plot available online) in Figure 3. In this diagram, vertical axis represents the association between the phenotype (left axis) and genotype (right axis). Association rules are ordered in the diagram following the association value (the higher association, the higher in the plot).
Gene ABO which has an indicated association (p-value 2.47×10−147) with coagulation (tissue plasminogen activation, Factor VIII and Von Willebrand factor levels), but also with serum levels of ALP (previously reported in Yuan et al. [34] and creatinine. Gene SURF4 (ENTREZ GENE # 6836), which has been previously associated with Von Willebrand Factor, Factor VIII and ALP, is also associated with ECG measures, MMP-9 (inflammatory marker) and mean cell volume (average red blood cell volume) among others. Gene HRG presents a weak association (p-value 1.88×10−24, corrected threshold 2.55×10−25) with some factors related with coagulation, such as activated partial thromboplastin time (APTT), ratio activated protein C (APC)/APTT, volume, total protein and Factor IX. Finally gene CETP (ENTREZ GENE # 1071) shows weak association (p-value 3.65×10−20) with cholesterol as expected, but also with coagulation factors (Von Willebrand Factor, Factor VII and sCD40 ligand).
Gene F10 (ENTREZ GENE # 2159), presents association with coagulation factors (Factor VII and Factor IX), but also with cholesterol, similar to gene F7, which also presents association with diastolic blood pressure. For the gene NRG1, we found association with ECG measures of ventricular hypertrophy, but also with urea and fibrinogen. Gene IL18RAP is weakly associated (p-value 5.07×10−21) with white cell counts (white cells, neutrophils, lymphocytes), with alanine transaminase (ALT) and glucose, but also with ECG measures of ventricular hypertrophy. Gene IL23R is weakly associated (p- value 2.19×10−18) with levels of interleukin 18 but also with adiponectin and ECG measures of ventricular hypertrophy). Gene ALOX5AP (ENTREZ GENE # 241) has been related with myocardial infarction and stroke [60], and also with inflammatory activity and atherosclerosis [61]. In our results it presents some association with some phenotypes related with immune response (white blood count, neutrophils, CD40 or total protein) but also with some markers of ECG related with hypertension. And gene GPR98 (ENTREZ GENE # 84059) is related in our analysis with immune response phenotypes and insulin related phenotypes (insulin, HOMA score) and in some cases with an association with CVD. No relation between this gene and these phenotypes has been reported, but some association was reported with carotid diseases and body weight [62], [63].
In this case, instead of selecting the most associated phenotypes for each gene, the GA selects the most associated genes for each phenotype. This operation is more computationally expensive than the previous one, because of the high number of genes (3648) involved. In order to correct for multiple testing, we use a Bonferroni correction for 82 phenotypes and a combination of 3648 genes in subsets of 29, 28, 27…1 groups. We choose 29 because this is the maximum number of different genes in one rule. This combination gives 2.03×1072 different genotypic rules, giving a threshold p-value of 2.99×10−75 equivalent to p = 0.05 for a single test. Ferreira [27] comments there may be a lack of power related with gene size for CCA for the case of multiple gene analysis. However, we consider that this analysis could contribute if the involved genes are small. Some of the most interesting rules are shown in Table 3. Other significant and non-significant enrichment analyses of the genes in the rules are listed in supplementary Table S3 and S4.
The Von Willebrand factor association (p-value 1.69×10−117–2.49×10−119) is led by individual association with gene ABO (p-value 9.43×10−112), and two of three significant pathways present more CCA association that Fisher multiple association.
The bilirubin association (p-value 6.76×10−115–3.79×10−118) is most influenced by genes in the UGT1 family, and all pathways present more CCA association than Fisher multiple association. The FVII association is led by genes F7 and EDEM2 (ENTREZ GENE # 55741) or PROCR (ENTREZ GENE # 10544). Finally FVIII association is led by ABO gene.
Regarding the enrichment analysis (Table S4), some interesting enrichments has been found, such as Factor VII and Human Phenotype Pathway “Abnormality of the coagulation cascade”, KEGG pathway “Complement and coagulation cascades” and Reactome pathway “Formation of Fibrin Clot (Clotting Cascade)”, or Factor VIII and KEGG pathways “ECM-receptor interaction” (pathway related with hemophilia, directly related with factor VIII). From non significant rules, APTT related genes are annotated with GO Terms “negative regulation of blood coagulation”, “blood coagulation fibrin clot formation”, “blood coagulation intrinsic pathway” and Reactome pathway “formation of fibrin clot”. Finally, LDL cholesterol is annotated with LDL gene related annotations
Finally, we use a CCA - two population genetic algorithm approach for multiple gene/multiple phenotype rule extraction. As a result, a set of 56 rules that relate the most associated set of genes with phenotypes was obtained. Following our previous multiple association corrections, the maximum number of genes in the obtained rules is 22 and the maximum size of the phenotypes is 9, so there is a possible population of 1.94×1057 gene rules and 3.3×1011 phenotypes, that determine a threshold p-value of 7.71×10−70 (equivalent to p = 0.05 for a single test).
Table 4 shows some of these association rules, and a complete list of 56 rules can be found in Table S5. An enrichment analysis can also be found in Table S6.
The bigger association obtained rule, genes F7, ABO, MRPS28, UGT1A3 and SURF4 with phenotypes bilirubin FVII and vWF, presents an association probability under 2.22×10−308, which was below our machine precision and therefore recorded as zero. We have identified some patterns in the multiple genes/multiple phenotype pleiotropic rules. ABO and SURF4 has similar relations with ALP, FVIII and vWF, F7 and F5 with FVII, F5 and HRG with APTT and ratio APC/APTT, F12 with APTT and NRG with Cornell product and QRS voltage product. Most of the rules obtained here are combinations of these.
The enrichments analysis of multiple phenotypes reveals interesting results, such as a rule formed by phenotypes bilirubin, alp, APTT, ratio APC/APTT and Von Willebrand Factor which were enriched for HP pathways “Prolonged partial thromboplastin time” and “Prolonged whole-blood clotting time”, KEGG pathway “Complement and coagulation cascades” and Reactome pathway “Formation of Fibrin Clot (Clotting Cascade)”. This rule is not a clear example of pleiotropy, because all genes and phenotypes are related with clotting, but it is clear that the inclusion of all genes and phenotypes in the same rules increases the association. Rules including phenotypes QRS duration, Cornell Index and Cornell Product, are annotated with hypertension GO terms and linked with genes that support these annotations. Also rules including phenotypes related with left ventricular hypertrophy are enriched with the GO term “epithelium development” and linked with genes related with cardiovascular development.
In the case of single gene/single phenotype analysis, we are not looking for pleiotropic effects, but for a combined gene-based association effect, and some interesting results were found. The complete list of gene-based significant and previously reported associations can be found in Table 1. One of the drawbacks of CCA analysis, which could affect our understanding of the role of NRG1, is that this method lacks power when a gene is larger than 100 Kb [27], and NRG1 has a length of 1.1 Mb.
In the case of single gene and multiple phenotype association, our results show that the p-values (both CCA and Fisher) increase when more related phenotypes are included in the phenotype set. As expected, most of these phenotypes are correlated/associated. However, not all phenotypic sets are correlated. An example can be observed in gene F7 (ENTREZ GENE # 2155), which is associated with phenotypes total cholesterol, Factor VII and Factor IX. Correlation exists between total cholesterol and Factor VII (0.28), Factor VII and Factor IX (0.39), but not between cholesterol and Factor IX (0.09). In some cases, Fisher's combined p-value approach shows equal or bigger association than CCA, which could mean that CCA association shows the cumulative effects of individual associations. In contrast, when Fisher's multiple association p-value is smaller than CCA association, this could suggest that CCA association analysis has found pleiotropic effect between a gene and these phenotypes. Some examples of the first group are association of gene ABO with coagulation phenotypes. In contrast, examples of pleiotropic effects appear in genes F10 (ENTREZ GENE # 2159) or F7, which presents an association with coagulation factors (Factor VII and Factor IX), but also with cholesterol.
In the case of multiple gene/single phenotype, using this robust association threshold, we have identified a set of pathways that are associated significantly with phenotypes Von Willebrand factor, bilirubin, FVII and FVIII. The whole list of pathways is listed in Table S3. It's interesting to see that there is no significant difference between the CCA and Fisher's association, in contrast with the differences shown in the previous Section, which supports the fact that CCA could detect pleiotropy patterns.
In conclusion, in this paper we have applied a canonical correlation analysis approach for association in multivariate datasets, finding correlations between gene-centered genetic variants and phenotypes. This multivariate approach allows us to mine pleiotropic relations between one or a set of genes and a set of phenotypes. In term of single gene/single phenotype association, we have found non-reported associations of gene NSF and triglycerides and genes ACSM3. ERI2, IL18RAP, IL23RAP and NRG1 with phenotypes related with left ventricular hypertrophy. We use a genetic algorithm as feature selection algorithm in order to find pleiotropy patterns in phenotypes. Using this approach we found pleiotropy patterns in genes F7 and F10 with phenotypes Factor VII, Factor IX and cholesterol; NRG1, with left ventricular hypertrophy related phenotypes, but also with fibrinogen and urea or IL18RAP or IL23RAP, related with immune response related phenotypes, but also with ECG measures.
Despite the possible drawbacks of CCA, related to power when the length of a gene is greater than 100 Kb, or increases of type I error when features are not normally distributed, we found that CCA can be used as a powerful tool to find gene-centered association, multivariate association and pleiotropic patterns. Also, this tool can be extended to find non-linear canonical correlation relations using kernel based approaches such as KCCA. Future research directions include improving the search method, using other meta-heuristics such as Tabu Search, Simulated Annealing or Particle Swarm Optimization, or sparse regularization methods.
The British Women's Heart and Health Study (BWHHS) is a UK-based prospective cohort study of 4286 healthy women aged 60–79 years at baseline (1999–2001). Participants were selected at random from general practice registers in 23 UK towns [64]. A range of baseline data sources (blood samples, anthropometry, health/medical history, echocardiography measures, etc.) was collected between 1999 and 2001, and DNA extracted from 3884 participants. Although the cohort has been followed-up in subsequent phases, all data presented here is based on the recruitment (baseline) phase.
Multi-centre (London Multi-centre Regional Ethics Committee) and local research ethics committees provided approval for the BWHHS study and informed consent was obtained from the women to complete the data used in this study.
Genotyping was performed using the Illumina HumanCVD BeadArray (Illumina Inc, San Diego, USA), which comprises nearly 48,742 SNPs in over 2,100 genes selected on the basis of cardiovascular candidacy by an international consortium of experts [65]. Genotypes were called using a Illumina BeadStudio (v3) Genotyping Module. Samples with a genotype call rate <90%, Hardy Weinberg disequilibrium <10−7 and minor allele frequency <1% were excluded from the analysis, following insight from previous work on this array and patient cohort [66]. Non-European samples were also excluded from analysis. Principal components analysis identified no evidence of population stratification (consistent with self-reported ancestry).
The different phenotypes used in this study consisted of 11 directed and derived electrocardiogram (ECG) measures, obtained as described in Gaunt et al. [67], 64 blood measures, 2 blood pressure readings, 3 anthropometric measures, HOMA score (derived from glucose and insulin values) and an indicator of whether a patient has suffered cardiovascular disease. These data were measured as described in Lawlor et al. [64].
All data were analyzed using R (The R project for statistical computing, http://www.r-project.org/). Due to the high number of missing values present in the phenotypic data (7575 of 312984 values, median of 55 (1.42%) missing values per phenotype, max 509 (13.17%) and min 19 (0.49%)), we followed a strategy of phenotypic data imputation based on a k-nearest neighbor approach, implemented in the R package “Imputation” [68] (http://cran.r-project.org/web/packages/imputation/index.html), with a k of 5. In order to test how these imputed values affected the association profile, we compared the single association values of imputed data versus data with missing values removed. The results show that the associations are the same or lower in the imputed values, so imputation does not create false associations. All phenotypic data was normalized to mean zero and standard deviation one.
All the approaches for analysis in this work were based on a “gene-centered” perspective. Genotype data, both intronic and exonic, was assigned to the genomically closest gene using the function “ClosestBED” from the suite “BEDTools” [69] (http://bedtools.readthedocs.org/en/latest/). In order to avoid multicollinearity in genotype data, we applied two-stage linkage disequilibrium (LD) pruning as described in Tang and Ferreira [27]. We removed SNPs with a high LD (r2>0.64) with other markers and also a high correlation between linear combinations of SNPs using Variance Inflation Factor (VIF) [70] in order to exclude SNPs with a VIF>2 with other markers. In order to select the most appropriate value for r2, we developed several experiments to test the CCA single gene/single phenotype association using a range of r2 (0.5–0.99), and best results was obtained pruning SNPs with r2>0.64. The value of VIF>2 was selected based in the recommendation of the original CCA paper [27].
As mentioned above, unlike other approaches to pleiotropy analysis, in this study we used a gene-centered approach. This perspective allowed us to capture all the pleiotropic effects in one gene, instead of the pleiotropic effects caused by just one variation. But we are also interested in studying the pleiotropic effects of a set of genes in several phenotypes. In order to do this, we divided the study into four stages. Firstly, we studied the individual association between each gene (which may consist of one or more SNPs) and a single phenotype to establish a gene-centered association baseline. This approach did not reveal any pleiotropy, of course, but it is worth pursuing for two reasons. Firstly, it was interesting to find if inclusion of several SNPs increases the association value over a single SNP approach. Secondly, we got a baseline gene association value that we used as a comparator for the CCA association analyses in our subsequent analysis.
For our second stage we studied the association between one single gene and a set of phenotypes. The aim of this analysis was to reveal possible gene-based pleiotropic effects. Our next stage was to study association effects between multiple genes and a single phenotype (gene-based epistasis analysis). The aim of this analysis was to discover pathway based baseline association between a set of genes and a single phenotype. Finally, our last stage consisted in studying the association between a set of multiple genes and different phenotypes. Here we expected to find the pleiotropic effects of a set of genes in multiple phenotypes, with increased statistical significance for the indicated association rules.
Canonical Correlation Analysis (CCA) allows us to find linear combinations of two sets of variables with the highest correlations. The aim of this work was to find correlation between a set of genotype data and a set of phenotype data. The CCA algorithm was based on a method proposed by Tang and Ferreira [27]. In order to test the significance of all canonical correlations, Wilk's Lambda and Rao's F approximation were calculated. Let q be the number of SNPs in the genotype, p the number of phenotypes evaluated, n the number of samples and cj the number of canonical components calculated. Wilk's Lambda is calculated as follow:And Rao's F approximation:WhereThe methods for CCA analysis analyzed in the previous Section could be computationally expensive with a large number of features and samples. In order to use standard CCA we can also divide the feature set into small subsets using biological insight (eg the set of SNPs in the region of a specific gene). In this paper we will simply use feature sets in which the SNPs are linked to a single gene, defining the link to a gene by genomic proximity. In this case, feature selection is not necessary because the number of samples is larger than the number of features.
To find those sets which have a high correlation, according to CCA, we need to use an optimization method, with the association value as the fitness function for this optimization procedure. We have formulated this optimization step as an integer programming problem which can therefore be addressed using a metaheuristic procedure to find an approximately good solution in a computationally tractable time. We have decided not to use methods such as hill climbing or similar local methods, because they are prone to capture by local minima. In this particular problem, any big single association could be assumed to be a local minimum and a hill climbing approach could not exit easily. Instead of this, we have decided to use global methods, such as Tabu Search [71], Particle Swarm Optimization [72] or Genetic Algorithm (GA) used here, as a well known and well used approach to this type of problem and with an effective means for evading local minima. A GA is a metaheuristic, initially proposed by Holland [73] and Goldberg [74]. This procedure is based on the principles of evolution and natural selection, with steps analogous to inheritance, mutation and crossover. It is initialized with a set of solutions, each representing one possible solution to the problem. The performance of each proposed solution is estimated using the fitness function, which measures how well an individual solution is adapted to the proposed problem. The method then iteratively evolves a high-fitness solution. The “genalg” R package (http://cran.r-project.org/web/packages/genalg/index.html) was used as a binary implementation of a GA. However, because of the requisites of the multiple gene/multiple phenotype analysis, this code was modified in order to include two population searches (Modified source of genealg package is available in http://github.com/jseoane/gaCCA). One of those populations represents different solutions for gene selection, and the other represents different solutions for phenotype selection. The search strategy is applied in parallel over the two populations and the fitness function is evaluated simultaneously over the selected set of genes and selected set of phenotypes, when calculating their CCA association value.
The encoding of the genetic algorithm is a binary encoding, widely used in feature selection approaches, where if the feature is set to 1, it is included in the analysis and is not included otherwise. The fitness function in the three versions is defined by the CCA association value. Regarding the parameterization, the population size is 100 for single gene/multiple phenotype, 600 for multiple gene/single phenotype and 1000 for multiple gene/multiple phenotype. The mutation depends on the size of the GA chromosome (1/82 in the case of multiple phenotype, 1/3248 for multiple gene). The elitism (how many samples of the population are conserved between generations) is 20/35/100, respectively for each of the versions. Finally, the “zero to one ratio”, which controls the number of features in the chromosome is set to 50 in the case of multiple phenotype and 700 in the case of multiple gene.
In order to avoid multiple testing associations which arise by chance, we applied a Bonferroni correction. In this case, the Bonferroni correction should be applied to both sides of the association. In this case the association is calculated over p phenotypes and g genes, so a 0.05 of confidence should need 0.05/(p*g). But when a GA is used, millions of associations are considered, so we approximate the Bonferroni correction over the search space of the algorithm (i.e. if we expect rules of p′ phenotypes and g′ genes, the search space over genes are combinations of g different genes over g′, g′-1, g′-2,‥,2, and the search space over phenotypes are combinations of p different phenotypes over p′, p′-1, p′-2,‥,2). The highly conservative final association threshold proposed is 0.05/(length of search space in genes * length of search space in phenotypes), though we ranked and considered all results by p-value in our analysis. In order to compare the CCA combined association measure with other measures, we have chosen a statistical measure based on Fisher's combined p-value approach proposed in Li et al. [5].
During phase two and phase four of the analysis, a set of genes related with one or several phenotypes is obtained. In order to functionally annotate these sets of genes, we perform an enrichment analysis, detecting GO ontology terms, KEGG, Reactome or Phenotype annotations that are significantly present in our pathways. We use the enrichment analysis tool g:Profiler [75] (http://biit.cs.ut.ee/gprofiler/), through R package “gProfiler” (http://cran.r-project.org/web/packages/gProfileR/index.html). In order to calculate the p-values for each enrichment, the method first simulate 10 millions of queries (sets of genes) randomly to see how was the p-values distribution according the query size. Then analytically derived the p-value threshold for each query size (for more details consult g:SCS threshold section in the Reimand paper).
Because some association values were close to zero, note that all calculations were performed in a 64-bit Linux R environment where the lowest positive value is 2.22×10−308, which means that values below this threshold were treated as zero.
|
10.1371/journal.ppat.1007884 | Hidden regulation of herpes simplex virus 1 pre-mRNA splicing and polyadenylation by virally encoded immediate early gene ICP27 | In contrast to human cells, very few HSV-1 genes are known to be spliced, although the same pre-mRNA processing machinery is shared. Here, through global analysis of splice junctions in cells infected with HSV-1 and an HSV-1 mutant virus with deletion of infectious cell culture protein 27 (ICP27), one of two viral immediate early (IE) genes essential for viral replication, we identify hundreds of novel alternative splice junctions mapping to both previously known HSV-1 spliced genes and previously unknown spliced genes, the majority of which alter the coding potential of viral genes. Quantitative and qualitative splicing efficiency analysis of these novel alternatively spliced genes based on RNA-Seq and RT-PCR reveals that splicing at these novel splice sites is efficient only when ICP27 is absent; while in wildtype HSV-1 infected cells, the splicing of these novel splice junctions is largely silenced in a gene/sequence specific manner, suggesting that ICP27 not only promotes accumulation of ICP27 targeted transcripts but also ensures correctness of the functional coding sequences through inhibition of alternative splicing. Furthermore, ICP27 toggles expression of ICP34.5, the major viral neurovirulence factor, through inhibition of splicing and activation of a proximal polyadenylation signal (PAS) in the newly identified intron, revealing a novel regulatory mechanism for expression of a viral gene. Thus, through the viral IE protein ICP27, HSV-1 co-opts both splicing and polyadenylation machinery to achieve optimal viral gene expression during lytic infection. On the other hand, during latent infection when ICP27 is absent, HSV-1 likely takes advantages of host splicing machinery to restrict expression of randomly activated antigenic viral genes to achieve immune evasion.
| Little is known regarding to how HSV, a large DNA virus and known to contain very few spliced genes, escapes host pre-mRNA splicing machinery. Here, by establishing a high throughput splice junction identification platform and quantitative analysis method to assess splicing efficiency based on high throughput data, we find that HSV-1 encodes hundreds of previously unknown alternative splice junctions; however, splicing of these novel spliced genes is largely silenced in wild-type HSV-1 infected cells, explaining why only very few spliced genes have been previously identified in HSV-1. Moreover, ICP27 is required for splicing inhibition and 3’ end formation of ICP34.5, the major viral neurovirulence factor and also the major target of latently expressed viral miRNAs. These findings not only fundamentally change the view of HSV gene structure, but also reveal a mechanism by which HSV employs host splicing and polyadenylation machineries to achieve optimal gene expression during acute infection and may also contribute to immune evasion during latency when ICP27 is not expressed.
| HSV-1 and HSV-2, two closely related human herpes viruses, establish lifelong incurable latency in and reactivate preferentially from trigeminal ganglia and dorsal root ganglia to cause orofacial and genital herpes, respectively. Although infections are usually mild, these viruses can cause severe disease including encephalitis and neonatal herpes. During latency in terminally differentiated neurons, expression of viral genes is repressed, except for the latency-associated transcript (LAT) and latency-associated miRNAs [1–3]. During acute infection, herpesvirus genes are expressed in a coordinated temporal cascade characterized by three kinetic classes, immediate-early (IE or α), early (β), and late. Late genes are further divided into two subclasses: leaky-late (γ1) genes that are expressed at very low levels at early times after infection and are dramatically upregulated at late times as a result of the increased number of genomes present after DNA replication, and true late genes (γ2) that are expressed exclusively after and are dependent upon viral DNA replication. HSV infected cell culture polypeptide 27 (ICP27), along with ICP4, are the only two IE genes essential for virus replication [1, 4]. ICP27, highly conserved between HSV-1 and HSV-2, is also the only one of the five HSV-1 IE genes that has clear homologs in all characterized mammalian herpesviruses (8). ICP27 is known to be required for efficient expression of some viral DNA replication-related early genes and late viral genes as well as for virus growth [5, 6]. ICP27 plays a role in transcriptional regulation through association with the C-terminal domain of RNA polymerase II [7, 8] and interacts with viral transactivating proteins encoded by immediate early genes including ICP4 and ICP0 [9–11]. ICP27 forms homo-dimers [12, 13], interacts with U1 snRNP through its C-terminal domain, and colocalizes with U1 and U2 snRNPs [14, 15]. ICP27 also interacts with splicing factors such as SRSF1, SRSF2, SRSF3, and SRSF7 through its C-terminal domain, and SR protein kinase 1 (SRPK1) through its N-terminal RGG RNA-binding domain [16–19]. Recently, ICP27 was shown to inhibit splicing of certain introns and promote use of alternative 5′splice sites (ss) in a small percentage of cellular genes in a sequence specific manner [20]. ICP27 also promotes co-transcriptional cellular pre-mRNA 3’ end formation using cryptic polyadenylation signals (PAS) in proximal introns, generating hundreds of novel, intronless GC-rich cellular transcripts that resemble HSV genes [20].
Although HSV-1 pre-mRNAs are transcribed in the nucleus by host transcription and RNA processing machineries, only 6 genes out of at least 84 genes, including 3 out of the 5 immediate early genes (ICP0, ICP22 and ICP47), a latently expressed gene (the latency associate transcript, or LAT) and two late genes (UL15 and gC), have until now been identified as spliced genes [1, 21]. Recently, a few novel splice isoforms including two antisense transcripts, UL41-42C (transcript initiated antisense to UL41) and AST-2 (transcript antisense to UL36), as well as UL49sp (splice site flanked by a unusual GC-AG intron) were identified using high throughput long-read sequencing in HSV-1 infected Vero cells [22].
Since ICP27-targeted host genes contain high GC content and cytosine-rich sequences, resembling HSV genes [20], we hypothesized that ICP27 likely co-evolved with the GC-rich viral genome and may have additional unknown viral targets. In this report, we further investigate the role of ICP27 in regulating pre-mRNA processing of viral genes. In addition to discovery of novel alternative splice sites for known viral spliced genes, we identify 22 novel viral spliced genes, most of which are tightly controlled by ICP27. Furthermore, we find that ICP27 tightly regulates expression of monocistronic ICP34.5 mRNA by inhibiting splicing and activating a PAS in the newly identified proximal intron, which represents a novel regulation mechanism for viral gene expression.
HSV infects most cell types in vitro including human kidney HEK-293 cells, which have been widely used in previous pre-mRNA splicing and polyadenylation related studies. To further understand ICP27’s role in viral pre-mRNA processing in a way aligning directly to previous findings [20], we performed RNA-Seq using poly(A)-enriched RNA purified from HEK-293 cells infected with wild-type HSV-1 strain KOS, or an ICP27 deletion mutant (d27-1) in the presence or absence of the viral polymerase inhibitor phosphonoacetic acid (PAA) at 4 hours post infection (hpi) or 7 hpi. The RNA-Seq data were analyzed using CLC genomic Workbench and the HSV-1 consensus sequence without the terminal repeat sequences was used as the reference (Fig 1). In KOS infected HEK-293 cells, reads mapping to the HSV-1 consensus sequence increased from approximately 43.2% at 4 hpi to 73.6% at 7 hpi, a result similar to that previously reported in infected MRC-5 human fibroblast cells [23]. However, deletion of ICP27 reduced reads mapping to the HSV-1 genome to approximately 10.5% at 4 hpi and 15.4% at 7 hpi, a reduction much greater than that induced by the viral polymerase inhibitor phosphonoacetic acid (PAA), which markedly reduced transcription only of γ2 genes (the subset of viral DNA replication-dependent γ genes). Reads mapping to the IE gene UL54 (ICP27) are not detectable in d27-1 infected cells since the coding region of ICP27 was deleted in d27-1. Deletion of ICP27 does not appreciably affect other α viral gene expression, but reduced non-α (i.e., β and γ) viral gene expression, confirming ICP27’s essential role in promoting β and γ viral gene expression.
The RNA-Seq data were also analyzed using CLC Genomics Workbench and the human genome consensus sequence (HG19) as the reference. Aberrant mRNA processing of ICP27-targeted cellular genes, previously identified as a result of ectopic HSV-2 ICP27 expression [20], was also observed in HSV-1 infected cells (with or without PAA), but not in d27-1 infected cells, confirming HSV-1 and ICP27’s role in mediating aberrant pre-mRNA processing in infected cells. The three previously described types of ICP27 mediated aberrant pre-mRNA processing (aberrant polyadenylation, aberrant use of 5’ss and intron retention) are apparent in three representative genes, PPTC7, ZER1 and POLR2A, respectively (Fig 2). The RNA-Seq results for these three representative genes in KOS and d27-1 infected cells are consistent with RT-PCR and Northern blot results [20]. HSV-1 thus mediates aberrant pre-mRNA processing in a manner similar to ectopic expression of ICP27 alone. Expression of ICP27 in the context of viral infection appears to induce additional intron retention in POLR2A that was not observed with ectopic expression of HSV-2 ICP27 in transfected cells, suggesting either subtle differences between HSV-1 and HSV-2 ICP27 or that virus-produced ICP27 more efficiently inhibits splicing than does ectopic expression in transfected cells.
We used CLC Genomics Workbench to further analyze the RNA-Seq data to view the splicing pattern of known viral spliced genes (as presented in Fig 1) at a resolution of single genes. Expression of ICP27 does not appear to influence pre-mRNA splicing of ICP47, ICP22, or ICP0 intron 2 (Fig 3A–3D). Slight retention of ICP0 intron 1 was observed in HSV-1 strain KOS-infected cells; however, retention of ICP0 intron 1 was accompanied by increased reads mapping upstream of the ICP0 transcription initiation site (Fig 3C). Because this approach may not detect splicing patterns of other known spliced genes due to much lower expression levels or to complex transcription patterns, the high throughput data were also mapped to a 44-bp reference unspliced sequence containing 22 bp of sequence from each exon and 22 bp sequence from the adjacent intron, and a 44-bp reference spliced sequence containing 22 bp sequence from each of the two exons expected to be joined following splicing. Each result was manually examined to confirm the results. The 44-bp length of these reference sequences detected splices in known spliced genes including ICP47, ICP0, UL15 and gC. The percentage of intron removal was thus calculated for individual genes in each of the five RNA-Seq data sets based on the reads mapping to exon-exon junction sequence relative to total reads mapped to both exon-exon junction and exon-intron junction sequences. The quantitative data support the graphical results in Fig 3A, 3B and 3C. ICP27 does not significantly influence the splicing of IE genes, except for intron 1 of ICP0 (Fig 3D). The splicing efficiency for ICP0 intron 1 appears to increase from approximately 85% in KOS-infected cells to over 99% in d27-1 infected cells. Increase of ICP0 intron 1 retention is coupled with an increase in reads mapping to sequences upstream of the ICP0 transcription initiation site. Co-transcriptional splicing of the first intron in a gene can be greatly enhanced by the RNA capping machinery and a large distance from a cap-proximal 5′ss to the RNA 5′ cap may reduce the chance of this splice site being recognized by cellular splicing machinery [24–26]. Taken together with the mapping results in Fig 3C, this implies that the observed retention of ICP0 intron 1 in KOS infected cells is likely in readthrough transcripts from upstream alternative promoters, for which splicing of ICP0 intron 1 is likely less efficient due to the increased distance between the 5’ cap and the 5’ss of ICP0 intron 1. The actual impact of ICP27 on the splicing of monocistronic ICP0 intron 1 is thus minimal, consistent with previous observations [27].
To confirm ICP27’s role in viral gene expression and obtain more precise quantitative information, we also performed RNA-Seq using Vero cells (monkey kidney cells that have been widely used in HSV studies) infected with KOS or d27-1 (S1 Fig). The infection was performed in triplicate in 6 well plates and the poly(A) selected RNA samples were prepared at 7 hpi. In this RNA-Seq data, the impact of ICP27 on overall viral gene expression was similar to the results obtained in infected HEK-293 cells as shown in Fig 1. The mean relative splicing efficiency and standard deviation were calculated based on the mapping results as described above. Retention of ICP0 intron 1, although less severe than in infected HEK-293 cells (Fig 3C), was also coupled with an increase of reads mapping to the sequences upstream of the ICP0 transcription initiation site (Fig 3E and 3F). Thus, ICP27 does not appear to inhibit splicing of IE genes.
To understand ICP27’s role in global viral pre-mRNA processing, we further mapped potential viral splice junctions using MapSplice 2, software that identifies potential splice junctions relative to a reference genome without relying on sequence annotations (31), with the default parameter settings and the RNA-Seq data for d27-1 and KOS infected HEK-293 cells (7 hpi) presented in Figs 1, 2 and 3. Possible splice junctions were detected relative to the HSV-1 reference sequence (raw data are presented in S1 and S2 Tables). Most newly identified introns possess canonical splice junctions flanked by GT(GU) and AG. Although the total viral read counts in KOS infected cells are nearly 5-fold more than d27-1 infected cells (Fig 1), the total read counts of the splice junctions mapping to the viral genome were similar between KOS and d27-1 infected cells (148,652 for KOS infected cells, and 143,990 for d27-1 infected cells). A total of 1940 and 450 splice junctions mapping to the HSV-1 genome were identified in KOS infected cells and d27-1 infected cells, respectively. KOS infected cells contained significantly more rare splice junctions (reads ≤ 2) than did d27-1 infected cells (85% vs 44%), suggesting that splicing of these predicted and known viral genes may be generally inhibited in wild-type HSV-1 infected cells as compared with ICP27 deletion mutant virus infected cells.
We next mapped the predicted splice junctions listed in S1 and S2 Tables to the viral genome. Ten (10) out of eleven (11) previously identified splice junctions in six known HSV-1 spliced genes, including three IE genes (ICP0, ICP47 and ICP22), two late genes (UL15 and gC) and one latent gene (LAT), and two recently identified spliced transcripts (UL41-42C and AST-2) [28], are also identified by MapSplice 2 (Tables 1 and S1 and S2). Only very few reads (<10) were identified for splicing of AST-2 intron 2 and no read was obtained for splicing of AST-2 intron 1 or UL41-42C intron 3. No read was obtained for splicing of UL49sp, a newly identified spliced transcript with a 74 bp intron flanked by GC-AG [22]. Total reads for these known splice junctions identified in KOS and d27-1 infected cells accounted for approximately 92.3% and 89.5% of total viral splice junctions detected, respectively.
In addition to the known spliced junctions identified, a total of 13 novel alternative splice junctions for transcripts of 7 of the known spliced HSV-1 transcripts, including LAT, ICP0, UL15, ICP22, ICP47, gC and UL41-42C were identified and the reads for these novel alternative splice junctions accounted for approximately 2.2% of the total junctions identified (Tables 1 and S1). We confirmed the novel splice junctions mapping to LAT, ICP0 and UL15 by RT-PCR and sequencing.
During latency of HSV-1, the most abundant viral transcript is the latency-associated transcript (LAT), a noncoding RNA. Primary LAT is a low-abundance transcript of 8.5 kb in latently infected neurons. LAT is spliced, leading to accumulation of abundant 2.0 kb and 1.45 kb highly stable introns in the nucleus [29, 30] and LAT-encoded miRNAs [2, 3]. The 2.0 kb intron appears to be the major species in the latently infected neuron and the 1.45 kb intron flanked by “GC-AG” is produced via secondary pre-mRNA splicing using splice sites within the 2.0 kb intron [30]. The splice junction of the 2.0 kb LAT intron was identified by MapSplice 2 (Table 1), as was a novel splice junction flanking a 1.68 kb GC-AG intron (Fig 4A). The 5’ss of the 1.68 kb intron is the same as previously described second 5’ss of the 1.45 kb intron, which was previously shown to be an intron within the 2.0 kb intron (illustrated in Fig 4A). The alternative splice junctions for the 1.68 kb LAT was confirmed by RT-PCR and subsequent sequencing of the PCR fragments (Fig 4A). However, neither the 1.68 kb LAT intron nor the previously reported 1.45 kb LAT intron was readily detectable in KOS or d27-1 infected Vero cells by Northern hybridization (Fig 4A). Splicing of the 1.68 kb intron flanked by “GC-AG” was much less efficient as compared to the 2.0 kb intron (Fig 4B). Reads mapping to the splice junction for the 1.45 kb intron were not found in the high throughput data in either KOS or d27-1 infected cells (Fig 4B).
ICP0, one of the five IE genes, is an E3 ubiquitin ligase that promotes viral gene expression and inhibits host cell response. ICP0 is non-essential at high multiplicities of infection [1]. Use of an alternative 3’ss located within intron 2 of the most common ICP0 mRNA isoform generates a one amino acid polymorphism and was confirmed by RT-PCR and sequencing (Fig 4C). The frequency of using the alternative 3’ss for intron 2 is unaffected by the presence or absence of ICP27 (Fig 4D). An alternative minor 5’ splice site for ICP0 intron 1 was also confirmed by RT-PCR (Fig 4C).
UL15, an essential γ gene, is required for viral DNA cleavage and packaging [1]. Four alternative 5’ss and one 3’ss were confirmed by RT-PCR and sequencing (Fig 4G). Each of the 5 alternative splices in UL15 destroys or truncates the open-reading frame (ORF) of UL15 protein (Fig 4E). Although these splice junctions map to UL15, some splice junctions such as the junctions using 5’ss nt27264 and 28566 may represent readthrough transcripts transcribed from upstream promoters. The relative splicing efficiency of the known splice junction (2990^33581) of UL15 (γ) was modestly increased from approximately 80% in KOS-infected cells to over 98% in d27-1 infected cells. Alternative splicing of UL15 was very inefficient in KOS infected cells but was significantly increased in d27-1 infected cells (Fig 4F and 4G).
We next analyzed splice junctions listed in S1 and S2 Tables mapping to previously unknown spliced genes in both KOS and d27-1 infected cells. Since low abundance splice junction reads likely represent very low splicing efficiency, only splice junctions greater than 65 nt with more than 30 reads from either KOS or d27-1 infected cells were selected for further verification. Viral transcripts with at least one splice junction verified experimentally by RT-PCR and sequencing are summarized in Table 2. Approximately 56 novel splice junctions were mapped to 20 viral genes, including in 2 previously uncharacterized viral transcripts mapping complementary to UL4 and UL22, named UL4-5C and UL22C. These novel splice junctions accounted for approximately 3.3% and 1.1% of total viral splice junctions for d27-1 and KOS infected cell (7 hpi), respectively. Since the cut-off for further analysis of putative splice junctions was set to 30 reads, the actual number of novel viral spliced transcripts almost certainly exceeds 22. These novel spliced genes include viral DNA replication-related early genes encoding UL5 protein (helicase/primase), UL52 protein (helicase/primase), UL12 protein (exonuclease), TK protein (thymidine kinase) and ICP8 protein (single-strand DNA-binding protein), as well as late genes encoding multiple glycoproteins (gH, gL, gE, gD and gB), and virus-host interaction factors including ICP34.5, US3, UL37, UL24, US11 and the virus-host shut-off protein (VHS) that are key viral virulence factors (Table 2). Splicing of these pre-mRNA transcripts destroys, truncates or internally deletes the open-reading frame (ORFs) for the protein encoded by each of these genes, indicating that ICP27-mediated aberrant pre-mRNA processing contributes to efficient expression of full-length viral proteins encoded by these genes with hidden splices.
All newly confirmed introns possess canonical splice junctions flanked by GT(GU) and AG except for three transcripts, including the LAT 1.68 kb intron and UL26, for which splice junctions are flanked by GC and AG. Most of the splice junctions flanked by GC and AG including that of UL49sp, a recently identified spliced gene, could not be confirmed by RT-PCR in infected cells, suggesting that the majority of the predicted splice junctions flanked by GC-AG may represent sequencing error due to the high GC content of the HSV-1 genome. A GC-AG type intron was predicted for UL42; however, sequencing of the RT-PCR bands indicates that the “GT” located 4-bp downstream of the predicted “GC” is included in the intron. In contrast, the vast majority of splice junctions flanked by GT(GU) and AG could be verified by RT-PCR. Two novel splice junctions (UL4-5C and UL22C) were mapped antisense to the coding region of UL4-5 and UL22, potentially representing read-through transcripts.
While HSV-2 ICP34.5 contains a 154-bp intron within the coding region and splicing inhibition of the HSV-2 ICP34.5 pre-mRNA by ICP27 results in a truncated form of ICP34.5 [31, 32], HSV-1 ICP34.5, encoding the major viral neurovirulence factor, was not previously known to be a spliced gene. Splices from a 5’ splice junction (nt125650) in the coding region of ICP34.5 to 3’ splice junctions (predominantly) at nt124046 in the 5’ UTR region of ICP0 and (less frequently) at nt123186, which is also the acceptor splice site of ICP0 intron 1 were identified (Table 2 and Fig 5A). The predominant spliced ICP34.5 transcript isoform (125650^124046) was readily detectable in d27-1 infected cells by RT-PCR but was barely detectable in KOS infected cells (Fig 5B). Two minor splice junctions (124467^124046 and 124467^123186) were also confirmed by RT-PCR (Fig 5B). Expression of HSV-1 ICP34.5 protein was abolished in d27-1 infected cells (Fig 5C). Further analysis of the splicing efficiency of these novel splice sites in the ICP34.5 region indicates that splicing of the ICP34.5 pre-mRNA is only efficient when ICP27 is absent (Fig 5D). When ICP27 is absent, 12560^124046 is the major splice with approximately 62% intron removal efficiency, followed by 123402^123186, 12560^123186 and 12560^124047. Splicing of 124467^124046 and 124467^123186 is very inefficient (below 0.5%) in the presence of ICP27 or not. Splicing of these novel introns is more efficient when ICP27 is absent. As discussed in Fig 3, splicing of ICP0 intron 1 and intron 2 is efficient with slight intron retention in ICP0 intron 1 in KOS infected cells. To be more rigorous, we also performed quantitative analysis of the novel alternative 5’ss (nt123402) of ICP0 intron 1. Splicing of this alternative ICP0 intron 1 appears to be enabled in the presence of ICP27, similar to the case with the alternative 3’ss (nt 122380 and 122377) of ICP0 intron 2 (Fig 4D). We next performed Northern blot of viral RNAs containing ICP34.5 exon 1 in HEK-293 cells infected with wild-type HSV-1 or ICP27 mutants using a probe corresponding to ICP34.5 exon 1 (Fig 5E). m15 has a two-amino acid mutation on the C-terminal domain and d4-5 contains deletion of the N-terminal RGG RNA binding domain /SRPK-1 binding domain. A ~1.3 kb band corresponding to the monocistronic ICP34.5 mRNA (Isoform I in Fig 5A), for which the ICP34.5 polyadenylation signal (PAS) is within the newly identified intron, was abolished in d27-1 and m15 infected cells and significantly reduced in d4-5 infected cells, while a ~3 kb band corresponding to the novel ICP34.5-ICP0 splice isoforms (Isoforms IV, III and II) was only present in d27-1 infected cells, consistent with the RT-PCR and Western blot findings (Fig 5C). A ~4.5 kb band corresponding to readthrough ICP34.5-ICP0 splice isoforms (Isoforms X/XI/XII) is present in cells infected with both wild-type and ICP27 mutants. Other ICP34.5 splicing isoforms at ~5 kb (Isoforms XIV/XIII/XV) predicted to exist based on these splice sites were also detected. Based on the relative splicing efficiency of these introns, the major products corresponding to the 1.3 kb, 3 kb, 4.5 kb and 5 kb Northern blot bands appear to be Isoform I, IV, X and XIV, respectively. Thus, the Northern blot results not only confirm ICP27-dependent splicing inhibition of ICP34.5 pre-mRNAs, but also demonstrate that cleavage and polyadenylation of the monocistronic ICP34.5 mRNA is dependent on ICP27 and that the ICP27-dependent effects are dependent on ICP27 C-terminal sequences, as is the case for ICP27-mediated premature termination of cellular mRNAs in HSV-2 ICP27 transfected cells [20]. A similar result was also obtained in infected Vero cells (Fig 5F), confirming that expression of the 1.3 kb monocistronic ICP34.5 mRNA as well as splicing involving the ICP34.5 5’ss is significantly increased when ICP27 is absent. We further analyzed the relative splicing efficiencies of the novel splice sites mapping to the ICP34.5-ICP0 region using the RNA-Seq data obtained from infected Vero cells (Fig 5G). The quantitative results from Vero cells infected and sequenced in triplicate are comparable to those obtained from HEK-293 cells (Fig 5D). The splicing at these novel splice sites in the ICP34.5 region indicates that splicing of the ICP34.5 pre-mRNA is only efficient when ICP27 is absent (Fig 5D), also confirming that the 12560^124046 is the major splice among these novel ICP34.5 isoforms. Thus, the RNA that encodes HSV-1 ICP34.5 protein is efficiently expressed only when ICP27 is present to inhibit its splicing and promote 3’ formation using its own PAS. Quantitative analysis also confirmed that alternative splicing of ICP0 intron 1 does not appear to depend on absence of ICP27. These results in Vero cells demonstrate that ICP27-mediated aberrant viral pre-mRNA processing is not host cell type dependent.
The latently and acutely expressed HSV-1 miR-H6 maps antisense to the LAT promoter regions and was reported to play a role in regulating expression of ICP4, the other IE genes (along with ICP27) required for viral replication [3]. While the primary miR-H6 (pri-miR-H6) transcript remains unknown, transcripts in this region including AL and TAL antisense to LAT exon 1 were reported previously [33, 34]. Data obtained from HSV-2 suggest that transcription of HSV-2 pri-miR-H6 is likely initiated just upstream (relative to pri-miR-H6) of the LAT TATA box [35]. Here, we show that miR-H6 maps to newly identified introns that share a common 3’ss mapping to the C-terminus of UL1 encoding glycoprotein L (gL) (Fig 6A), implying that pri-miR-H6 is also a spliced gene and miR-H6 is an intronic miRNA. This splice destroys the coding region of gL but the entire coding region of UL2, which encodes uracil-DNA glycosylase (UDG) is maintained. Thus, these two previously unidentified splice isoforms are named (pri-miR-H6/UDG). The gene structure of pri-miR-H6/UDG (7618^9770) resembles the gene structure of ICP34.5-ICP0, since both involve an intronic PAS located within 1 kb of the 5’ss. Both spliced isoforms (pri-miR-H6/UDG and the gL spliced isoforms) were confirmed by RT-PCR in KOS infected cells but were more abundant in d27-1 infected cells (Fig 6B). Quantitative splicing efficiency using the RNA-Seq data obtained from infected HEK-293 cells indicates that splicing of the pri-miR-H6/UDG transcript is efficient only when ICP27 is absent (Fig 6C). Quantitative splicing analysis using RNA-Seq of infected Vero cells (in triplicate) confirmed the results in HEK-293 cells (Fig 6D). UDG, which removes uracil that was mis-incorporated or arose by deamination in viral DNA in terminally differentiated neurons where endogenous cellular UDG activity is diminished, is required for DNA replication [36, 37]. This splicing that bridges the pri-miR-H6 promoter and the UDG coding sequences may potentially provide a mechanism for expression of UDG from the miR-H6 primary transcript during latency or early reactivation when ICP27 is absent.
A splice variant using the 5’ss (nt3721) of ICP0 exon 2 and the 3’ss (9770) of UL1 of ICP0-UL1 (3721^9770) was very recently reported [38]. In d27-1 infected cells, there is indeed one read mapping to the splice junction of ICP0-UL1 (3721^9770); however, there are no corresponding reads identified in KOS infected cells (S1 and S2 Tables). Splicing of ICP0-UL1 (3721^9770) is below the detection limit and estimated to be less than 0.03% of the ICP0 exon 2 and exon 3 splice junctions in both KOS and d27-1 infected HEK-293 cells (Fig 6C). ICP0-UL1 (3721^9770) is also under the detection limit in both KOS or d27-1 infected Vero cells (Fig 6D), suggesting splicing of ICP0-UL1 (3721^9770) is much less efficient compared to other splice variants involving either the ICP0 5’ss (nt 3721) or UL1 3’ss (nt 9770).
UL5 and UL52 are essential early genes encoding subunits of the viral helicase and primase complex, which is only weakly expressed in infected cells [39]. Splicing of UL5 and UL52 were confirmed by RT-PCR and sequencing (Fig 7A). Splicing efficiencies of UL5 and UL52 are approximately 20%-30% in HEK-293 cells and approximately 14–16% in Vero cells when ICP27 is not present; however, in the presence of ICP27, splicing of these two genes is efficiently inhibited (Fig 7B and 7C). Expression of UL5 protein is much reduced in d27-1 infected Vero cells or L2-5 cells that stably express UL5 under the ribonucleotide reductase (ICP6) promoter (Fig 7D), consistent with previous findings on the role of ICP27 on UL5 expression [5]. The predicted protein corresponding to spliced UL5 was not detected by Western blot, suggesting that it may be unstable. In addition to UL5 and UL52, DNA replication-related early genes including ICP8, TK, and UL12 are also spliced genes and ICP27 appears to inhibit splicing of these transcripts (Table 1 and Fig 8). While not all of these genes appear to be efficiently spliced in the absence of ICP27, the combined effect of splicing in several replication-related genes likely exceeds that in any single gene. The finding that in addition to increasing expression of specific viral DNA replication related early genes as reported previously [5] and as shown in Fig 2, ICP27 also contributes to maintain the functional full-length ORFs of targeted early genes by preventing co-transcriptional pre-mRNA splicing reveals a new regulatory mechanism at the pre-mRNA splicing level by which ICP27 controls viral DNA replication, and thus also expression of genes from newly synthesized DNA.
At least one of the splice junctions for each of the novel spliced genes listed in Table 2 was confirmed by RT-PCR and subsequent sequencing (Fig 8A and 8B). Furthermore, splicing is also much more efficient in d27-1 infected cells for most of the novel spliced genes, consistent with our observations for ICP34.5, UL5, UL52 and UL15 (Figs 4–6). The RT-PCR splicing patterns of a few inefficiently spliced transcripts (with relative splicing efficiency <5%) including UL42, UL26, UL22C and UL4-5C were not obviously different in the presence vs. absence of ICP27, and splicing of UL26C and UL4-5C appeared to increase in the presence of ICP27 (Fig 8B). The relative splicing efficiency of these novel spliced transcripts were also quantified using the RNA-Seq data. Splicing of gC, AST-2 and UL41-42C was much less efficient in KOS infected cells and was significantly increased in d27-1 infected cells with only one major exception (Fig 8C and 8D). Splicing of UL41-42C appears to be reduced in d27-1 infected cells (Fig 8C); however, transfection of a UL41-42C minigene with or without an ICP27 expression plasmid does not obviously increase splicing efficiency of intron 1 and 2, suggesting that the discrepancy is likely due to extremely low abundance of UL41-42C transcripts in d27-1 infected cells. The splicing efficiency of gC, a previously identified spliced gene (γ2), increased dramatically from approximately less than 0.5% in KOS-infected cells to more than 50% in d27-1 infected cells, which is consistent with previous observations [31]. We made a similar observation for AST-2, a spliced transcript identified recently by long-read high throughput sequencing in wild-type HSV-1 infected cells using the PacBio sequencing system [22].
MRC-5 diploid human fibroblast cells and Vero African green monkey kidney epithelial cells were infected with KOS and d27-1. RT-PCR was performed using primers targeting representative novel spliced genes including ICP34.5, UL5, UL52, gH, US3 and gE (S2A Fig). Similar splicing patterns (by RT-PCR) were observed in Vero and MRC-5 cells, confirming that ICP27-mediated effects on pre-mRNA splicing is not cell type dependent, consistent with the Northern blot results for ICP34.5 mRNAs. Quantitative splicing efficiency analysis for the novel spliced genes (described in Table 2 and not described in the previous figures) using the RNA-Seq data obtained from infected Vero cells (in triplicate) was determined (S2B and S2C Fig). The gB splice junction (54946^54453) was under the detection limit of the quantitative analysis; however, all of the other novel splice junctions (presented in Table 2) were confirmed in the Vero RNA-Seq data, with splicing of most of these transcripts inhibited by ICP27 as in infected HEK-293 cells (Fig 8C and 8D). The transcripts in Vero and HEK-293 cells for which splicing was not significantly influenced by ICP27 include UL41-42C, UL4-5C and UL22C, which are transcribed complementary to known viral genes and have unknown significance in viral pathogenesis. The overall splicing efficiency in Vero cells (when ICP27 is absent) appears to be lower than that in HEK-293 cells, suggesting that cell type specific splice factors may influence the alternative splicing process of ICP27 targeted spliced genes.
Most viral transcripts except for IE transcripts are reduced in d27-1 infected cells compared to KOS infected cells (Fig 1). To further quantify the effect of ICP27 on the accumulation of ICP27-targeted genes, especially the functional open-reading frames, we further analyzed relative expression levels of ICP27 targeted novel spliced genes (listed in Tables 1 and 2) using RNA-Seq data. In this analysis, the RNA-Seq data were mapped to reference sequences selected to represent the total expression level of different splice variants as well as the ORFs of these genes. ICP27 does not appear to affect the accumulation of IE gene transcripts including ICP0, ICP22, ICP4, or ICP47 (as indicated by the shared exon 2 sequence from US11), which serve as an infection control (S3A and S3B Fig). There is an approximately 8–10 fold reduction in levels of two γ2 late genes including gC and LAT in the presence of PAA, the DNA synthesis inhibitor, consistent with the previous finding that expression of gC and LAT in infected cell cultures depends on viral DNA replication [1, 40]. Accumulation of γ1 late and β genes appears to be relatively unaffected by DNA replication inhibition with PAA (< 2-fold changes); however, expression of many of these genes are significantly reduced when ICP27 is absent, indicating that ICP27’s role in regulating expression of these genes goes beyond just its influence on DNA replication. Accumulation of all other ICP27-targeted genes (ORFs) was positively correlated with ICP27. In d27-1 infected cells, reduced accumulation of ICP27 targeted transcripts ranged from approximately 470-fold for gC (UL44) to 1.7-fold for UL15 (S3B Fig) relative to cells infected with strain KOS. The median fold-reduction of ICP27-targeted gene levels in d27-1infected cells was approximately 18-fold for infection at 4 hpi and 8-fold at 7 hpi. Substantial reduction of accumulation of the gC transcript in d27-1 infected cells is consistent with previous findings that ICP27 regulates gC mRNA accumulation through a responsive element (also a C-rich sequence) on the gC mRNA [41], in addition to its role in splicing inhibition of the gC transcript [21]. Following gC, the other most responsive genes including UL26, US3, UL52, UL42, and LAT were more than 60-fold reduced in d27-1 infected cells. Genes including gH (UL22), UL37, VHS (UL41) and gE (US8) were approximately 20-fold reduced at 5 hpi in d27-1 infected cells (S3B Fig). Expression of ICP34.5 (as a total of transcript variants) was reduced by approximately 14-fold at 4 hpi and 9-fold at 7 hpi in d27-1 infected cells, consistent with the observation by Northern blot (Fig 5E and 5F). Consistently with previous findings [5, 42], replication-associated spliced early genes including UL5, UL52, UL42 and TK were also reduced significantly in d27-1 infected cells (6- to 73 -fold), while expression of ICP8 (β) was only modestly reduced (approximately 2.5-fold) in d27-1 infected cells. Due to the complexity of the transcription patterns, including sharing of PAS by different viral genes, read-through transcripts and viral transcription in both directions in certain locations, the RNA-Seq data should be interpreted cautiously in the context of nearby viral gene structures. Thus, splice junctions mapping in unknown transcripts antisense to known ORFs including UL22C, UL4-5C, AST-2 and UL41-42C were not included in this analysis. Nevertheless, these data are consistent with previously described results describing ICP27’s impact on virus gene accumulation [1], and illustrates an additional mechanism by which ICP27 regulates expression of its targeted viral genes.
By transcriptome analysis in cells infected with an HSV-1 ICP27-deletion mutant, we identified hundreds of novel splice sites mapping to the HSV genome and experimentally confirm at least 22 novel viral alternatively spliced genes, many of which are essential for efficient viral replication. We find that ICP27 inhibits splicing and promotes efficient polyadenylation using a proximal intronic PAS to facilitate expression of ICP34.5 and pri-miR-H6/UDG, both of which are novel spliced genes with intact ORFs. These findings not only fundamentally change our understanding of HSV-1 gene structure by quantitively mapping alternative splicing to more than one third of the known viral genes, but also reveal a novel mechanism by which ICP27 hijacks host splicing and polyadenylation for optimal viral gene expression. These findings also imply that during latent infection when ICP27 is absent, HSV-1 likely takes advantage of host splicing machinery to restrict expression of randomly activated antigenic viral genes to achieve immune evasion.
Analysis of high throughput RNA-Seq data for splice variants remains challenging. In this study, we used MapSplice 2 software to map pair-ended high throughput RNA-Seq data to an HSV-1 reference genome for novel splice junction discovery. This powerful approach was able to identify nearly all previously known splice junctions, as well as hundreds of novel splice junctions. We did not attempt to verify every single one of the hundreds of splice junctions identified (S1 and S2 Tables) but chose novel splice junctions with ≥ 30 reads for further experimental verification. We were able to confirm at least 22 previously unidentified spliced genes, recognizing that splice junctions with lower read counts could also likely be verified experimentally. For example, the very recently reported ICP0-UL1 splice junction [38] was detected with 1 read in d27-1 infected HEK-293 cells (S1 Table). However, no ICP0-UL1 splice junction read was identified using the RNA-Seq obtained in KOS infected cells or Vero cells infected with either KOS or d27-1 (S2 Table and Fig 6E), suggesting splice junctions even with extremely low read counts (e.g., one read count as listed in S1 Table) may be experimentally verifiable and thus that splicing is a widespread event in HSV-1 infected cells.
Due to the high GC content in HSV, we found that determination of splicing patterns by RT-PCR favors smaller PCR fragments representing spliced products, which is prone to overestimating splicing efficiency, especially for larger introns. Thus, we established a relative quantitative method by calculating the percentage of the exon-exon junction reads among the total exon-intron junctions and exon-exon junctions for each splice based on RNA-Seq data. This quantitative analysis is in agreement with the RT-PCR results for most of the spliced genes and is also consistent with Northern blot results in this this and previous studies, providing a powerful high throughput tool for better understanding the nature of a splice junction.
Little is known regarding to how HSV, a large DNA virus and known to contain very few spliced genes, escapes host pre-mRNA splicing machinery. Pre-mRNA splicing is a “default” process initiated by binding of U1 snRNP to the consensus 5’ss and of U2 snRNP to the consensus 3’ss. Splicing factors can regulate gene expression by influencing the inclusion or exclusion of particular exons in a gene’s mRNA [43]. We identified much more alternative splicing in a large group of novel viral spliced genes in infected cells when ICP27 is absent. When ICP27 is present, most of the splices in these novel viral spliced genes are largely silenced, suggesting the viral IE protein functions in a way analogous to a splicing factor to inhibit splicing both of its own transcripts and of a small percentage of host transcripts that resemble HSV genes in a gene/sequences specific manner [20]. Thus, HSV ICP27 likely coevolved with GC-rich viral genes that contain C-rich sequences and co-opts host splicing machinery to ensure the correctness of viral ORFs.
Interaction of U1 snRNP with 5’ss and PAS near the transcription start site controls the length of cellular mRNAs and promoter directionality [44–47]. We showed previously that ICP27 counteracts U1 snRNA’s function and promotes expression of hundreds of cellular short intronless transcripts resembling HSV genes [20]. Here, we show that ICP27 also toggles expression of the HSV-1 monocistronic ICP34.5 mRNA that encodes the major viral neurovirulence factor by activating the proximal PAS located in the newly identified intron and inhibiting splicing of the newly identified intron (Fig 5). In ICP27 deletion mutant infected HEK-293 and Vero cells, approximately 64% to 75% of ICP34.5 transcripts are spliced, destroying the coding sequence. Interestingly, ICP34.5 is also negatively regulated by the two most abundant latently expressed miRNAs mapping antisense to its 5’ UTR and exon 1 [2, 3, 48, 49], further suggesting the importance to viral pathogenesis of tight regulation of HSV-1’s major neurovirulence factor. The distance between the newly identified 5’ss of ICP34.5 and its PAS, which is located in the newly identified intron, is within the range (≤ 1 kb) typical for both ICP27 or U1 snRNP inhibitor mediated activation of intronic PAS of cellular genes [20, 44–46]. Because inhibition of U1 snRNP’s binding to a 5’ss also typically relieves its inhibition of polyadenylation at a downstream PAS (typically within 1 kb of the 5’ss) [44–46], and ICP27 is known to interact with U1 snRNP through its C-terminal domain, colocalizing with U1 and U2 snRNPs [14, 15], this suggests that U1 snRNP is likely to be involved in the mechanism of ICP27-mediated splicing inhibition and activation of intronic PAS.
The primary transcript of miR-H6, a latently and acutely expressed viral miRNA that was reported to target the key viral transactivator ICP4 [3], has not been identified although it is certainly transcribed antisense to the LAT. There are reports of transcripts antisense to LAT such as AL RNA and TAL RNA antisense to LAT exon 1; however, the 5’ start sites and 3’ ends of these RNAs has not been determined [33, 34]. Identification of novel splice junctions for pri-miR-H6 suggests that during latency when ICP27 is absent, pri-miR-H6 is likely a spliced transcript. The gene structure around the splice 7616^9770 appears similar to that of ICP34.5, with a PAS mapping to approximately 700 bp downstream of the novel splice site (nt7618) and with the splicing inhibited by ICP27 (Fig 6). Splicing of the pri-miR-H6-UDG transcript during latency when ICP27 is absent may potentially lead to expression of UDG, a critical viral DNA repair enzyme. It is known that the endogenous enzymatic UDG activity is absent in terminally differentiated neurons.
In contrast to its role in virus-host shutoff by inhibiting expression of selected cellular transcripts [20], ICP27-mediated aberrant pre-mRNA processing is required to efficiently express full-length viral ORFs of ICP27 targeted viral genes during the coordinated temporal cascade of gene expression to promote efficient viral replication. Thus, ICP27 ensures the quality of its targeted viral transcripts. For example, inhibition of splicing of the low-abundance UL5 and UL52, both essential early genes that encode the primase/helicase complex, may also contribute to a “switch” effect by which ICP27 regulates viral DNA replication and the many viral genes that rely on viral DNA replication. Splicing control of critical virulence factors, such as ICP34.5, as well as essential glycoproteins, such as gH and gC, likely also collectively contributes to the avirulent phenotype of ICP27 deletion mutants.
High GC content (approximately 68% to 70% in HSV) can contribute to intron retention or reduced splicing efficiency of mammalian genes [50, 51] and likely contributes to the fairly low observed baseline splicing efficiency of many of the novel spliced genes identified in this study. Even low splicing efficiency indicates that, these genes contain authentic U1 snRNP binding sites at 5’ss. U1 snRNP binding to 5’ss inhibits 3’ end formation at proximal PAS (typically <~1 kb from the transcription start site), a mechanism used by the cell to define the length and direction of its transcripts. Persistent U1 snRNP binding to cryptic 5’ss in proximity to a PAS can prevent accumulation of certain adenovirus, polyomavirus and bovine papillomavirus (BPV) mRNAs [52–54]. We thus hypothesize that some of the HSV genes, such as intronless short transcripts with cryptic 5’ss near a proximal PAS are also subject to U1 snRNP-mediated restriction and that ICP27 is required to remove U1 snRNP-mediated suppression of polyadenylation, analogously to ICP27’s role in promoting expression of cellular intronless transcripts polyadenylated from proximal intronic PAS as well as the ICP34.5 monocistronic RNA [20] (Fig 5). Thus, the overall consequence of ICP27 mediated splicing inhibition during lytic infection likely ensures not only the quality (correct ORF and stabilized monocistronic mRNAs) but also the quantity (abundance) of certain ICP27-targeted viral genes, in addition to ICP27’s known role in RNA exporting and transcription [1, 55, 56].
While the virus might have been able to achieve expression of full-length genes via mutation of its splice sites, conserved splice site sequences among different HSV-1 strains suggests that ICP27-regulated aberrant posttranscriptional pre-mRNA processing likely has additional important functions. For example, this mechanism may also help reduce accidental expression of full-length viral antigens targeted by ICP27 during latency when ICP27 is absent. Indeed, recent studies revealed that HSV latency is not entirely quiescent and frequent switching on of certain antigenic lytic genes has been reported in immunological and molecular studies [57–59]. Thus, posttranscriptional regulation including splicing and the LAT-encoded miRNAs that disrupt major viral antigens and genes required for viral replication during latency when ICP27 is absent may contribute to immune-evasion and maintenance of viral latency. We also hypothesize that binding of U1 snRNP may play an important role in suppression of polyadenylation of certain ICP27 targeted viral genes during latency in the absence of ICP27, further contributing to immune-evasion and maintenance of viral latency.
Other viruses, such as papillomavirus, polyomavirus, adenovirus, retrovirus, and influenza virus, for which viral mRNA is transcribed in the nucleus, take advantage of the host pre-mRNA splicing and polyadenylation machinery to encode more viral products using limited viral DNA sequences through alternative splicing and polyadenylation to suit their viral life cycles [60]. Many of these viruses encode viral proteins to co-opt the cellular pre-mRNA processing machinery. For example, influenza NS-1 interacts with SRSR2, U6 snRNA and NS1-BP and altering host and viral mRNA splicing [60, 61]. NS-1 also interacts with CPSF30, polyadenylation factor required for the 3’ end processing of cellular pre-mRNA, resulted in reduced expression of cellular antiviral genes but not viral genes [61].
In contrast to many other viruses, the HSV-1 life cycle contains both a lytic infection phase, characterized by a coordinated temporal cascade, and a latent infection phase, characterized by the absence of significant viral antigen expression and viral DNA replication. During lytic infection, through its IE protein ICP27, HSV-1 activates PAS contained within the proximal intron and near the transcription start site of its genes, while inhibiting splicing of viral and cellular genes in a gene/sequence specific manner to achieve optimal viral gene expression. Many ICP27-targeted cellular genes are related to host immune response [20]. During latent infection in the absence of ICP27, HSV-1 likely uses host RNAi and splicing machinery to restrict expression of randomly activated viral antigens to achieve optimal immune evasion. Further investigation of the details of ICP27 mediated aberrant pre-mRNA processing will likely yield insight both into mechanisms of viral pathogenesis, potentially leading to identification of new targets for antiviral strategies, and into the mechanisms by which the cell itself controls alternative polyadenylation and splicing of selected genes.
HEK- HEK-293, MRC-5 and Vero cells were obtained from ATCC. L2-5 cells, a UL5 mutant complementary cell line established from Vero cells that stably expresses HSV-1 UL5, were obtained from Dr. Sandra K. Weller [62]. HSV-1 strain KOS, HSV-1 ICP27 mutant viruses (as shown in Fig 5E), and the V27 ICP27-complementing Vero cell line used to grow ICP27 mutant viruses were obtained from Dr. Stephen Rice (University of Minnesota) [63, 64]. Anti-HSV ICP4 antibody (Santa Cruz) and anti-Flag antibody (Sigma) were sourced commercially. Anti-HSV-1 ICP34.5 antibody was obtained from Dr. Ian Mohr [65]. Anti-HSV-1 UL5 antibody was prepared from rabbits using peptides (AGGERQLDGQKPGPP and LTSNPASLEDLQRR).
HEK-293 cells were infected with HSV-1 KOS or an ICP27 deletion mutant, d27-1 at a MOI of 5 in the presence or absence of the viral DNA replication inhibitor, phosphonoacetic acid (PAA), at 300 mg/mL. At four or seven hours post-infection (hpi), total RNAs were purified with the All-Prep DNA/RNA Kit (Qiagen). cDNA libraries were prepared from polyadenylated RNA using the TruSeq RNA sample Kit V2 (Illumina) and were sequenced on the NextSeq 500 according to the manufacturer’s instructions (Illumina). The six samples shared a single sequencer lane. Vero cells (in 6-well plates) were infected with HSV-1 KOS or d27-1 at a MOI of 5 in triplicate. At 7 hpi, total RNAs were purified with the All-Prep DNA/RNA Kit (Qiagen). cDNA libraries were prepared from polyadenylated RNA using the TruSeq RNA sample Kit V2 (Illumina) and were sequenced on the NextSeq 500. A total of 18 samples shared the same sequencing lane.
Viral gene expression profile was analyzed using CLC Genomics Workbench (QIAGEN) with an HSV-1 strain 17 (NC_001806.2) consensus sequence without the terminal repeat sequences as a reference (note: the genome sequence of strain 17 was only used in the CLC Genomics Workbench related analysis and all the exact splice site notations were based on HSV-1 KOS strain (JQ673480.1) as described below. CLC Genomics Workbench mapping of RNA-Seq data to genomes was performed without strand specificity. The cellular gene expression profile was analyzed using CLC Genomics Workbench and the human HG19 consensus sequence as a reference.
The RNA-Seq data were analyzed using MapSplice 2, software developed for mapping RNA-Seq data to a reference genome for splice junction discovery [66]. The HSV-1 KOS genome (JQ673480.1) was used as the reference sequence. Splice junctions with more than 30 reads were selected for further analysis.
The RNA-Seq data were further mapped to the exon-exon junction (22 bp from each adjacent exon) or splice site junction (22 bp from the exon and 22 bp from the 22 bp from the adjacent intron) reference sequences using CLC Genomics Workbench. Each mapping result was visually checked to avoid partial or false alignments. Relative splicing efficiency was calculated using the percentage of exon-exon junctions reads in the total reads mapped relative to the total exon-exon junctions and splice site (5’ or 3’) junctions for each splice. For further analysis of relative expression of ICP27 targeted genes, 44 bp sequences from the N-terminus of coding sequences or the sequence upstream of the splice site of targeted genes were used as references. Read counts were normalized with the highest reads of KOS or d27-1 infected cells to generate the relative expression levels between KOS and d27-1 infected cells (S3A Fig). Fold-reduction comparisons were generated based on relative expression levels (S3B Fig).
All identified novel 5’ss sequences (3 bases in exon and 6 bases in intron) and the 3’ss sequences (20 bases in the intron and 4 bases in the exon) were aligned to the genomic sequences of five commonly referenced laboratory and clinical HSV-1 strains including strain KOS, HSV-1 strain 17+ (NC_001806), strain F (GU734771), strain McKrae (JX142173), and strain H129 (GU734772). The strength of the splice sites were measured by MaxEntScan [67].
pICP27, an HSV-2 expression vector, was described previously [32]. HSV-2 ICP27 mutant plasmids including pΔR2 and pM15 were obtained from Dr. Masatoshi Hagiwara (Tokyo Medical and Dental University) [68]. The HSV-1 ICP34.5-specific DNA probe template (nt 125645–125827) containing 97 bp of the 5’ UTR sequence and 86 bp of the exon 1 sequence upstream of the novel 5’ss (as illustrated in Fig 5A) was prepared from plasmid constructed using PCR fragment by oST1076 and oST1075B. Oligonucleotide primers and synthesized DNA fragments are included in S3 Table.
HEK-293 cells, MRC-5 cells, Vero cells or L2-5 cells were infected with viruses indicated in the figures at a multiplicity of infection (MOI) of 5. Total protein or RNAs were prepared at different time points post inoculation. Western blot was performed using the antibodies described above. For RT-PCR, total RNAs were extracted using All-Prep DNA/RNA kits (Qiagen). The primer sequences are listed in S3 Table. The RT-PCR bands shown in the figures that correspond to novel splice junctions were further confirmed by Topo cloning and sequencing. HEK-293 cells were transfected with plasmids indicated in Fig 6D using Lipofectamine 2000 (Invitrogen). Total protein or RNAs were prepared 24 hours post transfection. For Northern blots, total RNAs were prepared from HEK-293 cells or Vero cells infected with HSV-1 KOS strain or ICP27 mutants by TRIzol (Invitrogen). Approximately 30 μg of total RNAs were separated in a formaldehyde denaturing 1.2% agarose gel (Life Technologies). After transfer to GeneScreen Plus hybridization transfer membrane (Perkin-Elmer), the membrane was incubated in NorthernMax hybridization buffer (ThermoFisher Scientific) at 58°C overnight with an HSV-1 ICP34.5-specific probe labeled with [α-32P] dCTP using a random priming kit (Promega).
|
10.1371/journal.pcbi.1002908 | Transformation of Context-dependent Sensory Dynamics into Motor Behavior | The intrinsic dynamics of sensory networks play an important role in the sensory-motor transformation. In this paper we use conductance based models and electrophysiological recordings to address the study of the dual role of a sensory network to organize two behavioral context-dependent motor programs in the mollusk Clione limacina. We show that: (i) a winner take-all dynamics in the gravimetric sensory network model drives the typical repetitive rhythm in the wing central pattern generator (CPG) during routine swimming; (ii) the winnerless competition dynamics of the same sensory network organizes the irregular pattern observed in the wing CPG during hunting behavior. Our model also shows that although the timing of the activity is irregular, the sequence of the switching among the sensory cells is preserved whenever the same set of neurons are activated in a given time window. These activation phase locks in the sensory signals are transformed into specific events in the motor activity. The activation phase locks can play an important role in motor coordination driven by the intrinsic dynamics of a multifunctional sensory organ.
| How sensory information is transformed into effective motor action is one of the most fundamental questions in neuroscience. As this question is difficult to assess experimentally, biophysical models of sensory, central and motor systems contribute to understand the information processing mechanisms involved in this transformation. Biophysical models can be informed by electrophysiological data in those situations where it is possible to record neural activity at all stages of sensory-motor processing. In this paper we use this approach to describe the dual dynamics of a multifunctional sensory organ in the mollusk Clione limacina and its transformation into two different motor programs. Our experimental and modeling results indicate that the sensory signals are modified to fit the changing behavioral context, and they are readily interpreted by the rest of the nervous system to produce the correct motor output.
| One of the most fundamental questions in neuroscience is how sensory information is transformed into effective motor action. This question is difficult to assess experimentally as it implies monitoring neural activity at different stages of the sensory-motor transformation, a task that can be more easily addressed in simple animals [1]–[4]. Experimental evidence points to the important role that intrinsic sensory dynamics, i.e., neural dynamics that does not directly correlate to the dynamics of a physical external stimulus, can serve in this transformation. In particular, there are several examples from neurophysiological studies that show complex intrinsic dynamics in sensory networks [5], [6]. In most cases, the dynamics observed is directly related to the information encoding mechanisms in these systems.
Complex intrinsic dynamics can also be related to multifunctionality, which has only been partially addressed in neuroscience research, mainly in motor networks [7]–[13], with a few examples in sensory systems [14]. One remarkable example of relationship between intrinsic sensory dynamics and multifunctionality has been discussed for the gravimetric organ of the mollusk Clione limacina [15], [16].
The marine mollusk Clione limacina (see insets in Figure 1) is a predator whose only prey is another mollusk, Limacina helicina. Because of the simplicity of its nervous system, this animal is a well known model to study both sensory and motor processing, and thus the transformation that occurs between them. During routine swimming, when water disturbance changes its head-up body orientation (see the left inset of Figure 1), Clione tries to correct for the change by actively moving the wings and the tail [17]. Several neural structures are involved in the control of the orientation of Clione's body during this routine swimming [17]–[19]. As the main sensory input, Clione uses its gravimetric organs, the statocysts. These are a pair of spherical structures located in the pedal ganglia which contain a stone-like structure –the statolith– that moves under the effect of gravity. The statolith exerts pressure on the internal wall of the sphere which is lined with statocyst mechano-receptor cells. The statocyst receptor cells (SRCs) react to the pressure of the statolith allowing the animal to determine changes in the orientation of its body. Sensory information about the orientation of the body is sent from the mechano-receptors to several groups of cerebro-pedal interneurons. These interneurons in turn control the central pattern generators that drive Clione's wing and tail motoneurons, which add steering (i.e., to induce transient corrective motions in the wings and tail to achieve the preferred head-up position during routine swimming) [18], [20]–[23]. In addition to gravimetric signals, chemical sensory information about the presence of prey is conveyed to the SRCs through excitatory input from a pair of cerebral hunting interneurons (CHI) [24], [25]. Clione does not have a visual system and although its chemosensors can detect the presence of Limacina, they are presumably nondirection-sensitive. When hunting behavior is triggered, the resulting hunting search consists of loops and turns in a complex trajectory to locate the prey. A quantitative analysis of the hunting search trajectories has been described in [15]. Hunting behavior typically occurs in different search episodes with resting times in between. The Videos S1 and S2 in the supplementary material illustrate Clione's typical routine swimming and hunting search behaviors. The two behavioral contexts described above make Clione a good animal model to study the mechanisms involved in sensory-driven motor activity.
There is a long tradition in building models of sensory-motor transformation based on electrophysiological recordings (e.g. see [26], [27]). In most cases, the sensory network is not included in the model as the intrinsic sensory dynamics are not taken into account. However, more complete computational models of the sensory-motor transformation can largely contribute to the understanding of how a given neural motor pattern is generated from sensory activity. In particular, theoretical efforts help to characterize the dynamics of sensory networks and to identify relevant features that are used in the organization of motor activity. We have previously developed a rate model to describe how a winnerless competition dynamics can arise within an inhibitory network [28] and pointed out its possible role in the organization of the complex hunting behavior of Clione [16], [29], [30]. Several predictions of the model regarding this type of intrinsic sensory dynamics have been tested in neurophysiological experiments: (i) the statocyst network produces a complex sustained spatiotemporal dynamics during hunting search behavior even when there is no motion in the in vitro experimental conditions, in contrast to the situation during routine swimming in which only a few neurons are active; (ii) during hunting behavior, the activity of the sensory neurons is correlated to the wing and tail motoneurons [15], [16].
In this paper we build a conductance based-model of the statocyst sensory network and the wing CPG. Our model illustrates how the multifunctional nature of the sensory network can drive the CPG activity in two different behavioral contexts. On one hand, the winner take-all dynamics in the gravimetric sensory network model is read by the wing CPG network and produces the steady rhythm during routine swimming. On the other hand, the same sensory network under a different stimulation produces a winnerless competition dynamics which is read by the motor network and results in the irregular patterns characteristic of wing motoneurons during hunting behavior. The model demonstrates that the spatiotemporal pattern of the sensory dynamics may be in the form of specific activation phase locks that emerge during hunting. These phase locks are transformed into specific motor events in the wing CPG model.
Experimental recordings have shown that during routine swimming, Clione uses information from the statocysts regarding body orientation to keep its preferred head-up posture (see inset in Figure 1, left panel) [21]. During this behavioral state a winner take-all (WTA) competition occurs between the SRCs and only the cell or group of cells pressed by the statolith persistently fire (see Figures 4 and 5 in [19]). This is in part due to the inhibitory nature of this network. The activated SRCs inform the rest of the nervous system about Clione's body orientation and, if a postural change occurs, wing and tail activity generates transient corrective motions to recover the preferred head-up position. On the other hand, during the hunting search behavior a sustained winnerless competition emerges in the statocyst network (see Figure 2) which consists of an irregular alternation of firing among the SRCs. The presence of Clione's prey evokes excitation of the CHI neurons, the hunting neurons, which send an excitatory input to the SRCs. When CHI neurons are activated, hunting behavior starts [24], [31], [32]. Hunting behavior can be evoked in vitro by applying the acetylcholinesterase inhibitor physostigmine to the animal (see “Methods and Models” section). In these experiments there is no motion and the main input to the SRCs comes from the external excitation of the hunting neuron and not exclusively from the statolith.
To model the gravimetric sensory network dynamics in these two different behavioral contexts (routine swimming and hunting behavior), we have used an inhibitory neural network with six SRCs under the action of the statolith and a CHI neuron (see Figure 1). Each cell of the network is implemented with a Komendantov-Kononenko conductance based model [33], a well characterized model for molluscan neurons that can qualitatively reproduce the spiking and spiking bursting behavior observed in Clione's neurons. The sensory network is built with an asymmetric inhibitory connection topology inspired by the current knowledge of the statocyst network [15], [28]. A detailed description of this topology and the parameters used in our simulations, for both the individual behavior of each cell and the connectivity, can be found in the “Methods and Models” section. During the simulation of routine swimming with our statocyst model, the CHI is silent and does not excite the SRCs. Thus, the only input that the statocyst neurons receive in this case is the excitation from the pressure of the statolith (simulated as a current injected in a specific SRC). The left panel in Figure 3A shows that in this situation, a winner take-all dynamics appears in the statocyst conductance-based network model, and only the SRC pressed by the statolith fires. A simulation of body orientation change as illustrated here by exciting another SRC (see arrow on the left panel of Figure 3A) causes a change of the active neuron in the statocyst network: the new pressed SRC starts firing immediately while the other neurons are silent.
In our model network we simulate the presence of a prey by activating the CHI neuron, which excites all SRCs. This activates the hunting behavioral state within the same sensory network model (Figure 1, right panel). Under the CHI neuron excitation, the dynamics of the model sensory network changes to a winnerless competition (WLC) among the SRCs and the action of the statolith hardly affects the network dynamics (see Figure 3A, right panel). This competition arises from the hunting neuron excitatory input to all SRCs and the inhibitory connections within the network. In this WLC dynamics the SRCs display switching activations of varying durations including some overlappings, as the inhibition among the SRCs is moderate. The activity of the network in this case is highly irregular. These irregular sequential activations in the model are qualitatively similar to those observed in the biological network during fictive hunting behavior evoked in vitro (c.f. right panel of Figure 3A and Figure 2, where four SRCs from the same statocyst are recorded simultaneously in an in vitro preparation). The level of irregularity of the sensory network model can be characterized by calculating the Lyapunov exponents. Figure 4 shows the evolution of the calculation of the positive Lyapunov exponents from the vector field (see “Methods and Models” section for details) describing the sensory network during hunting behavior. The presence of two positive Lyapunov exponents ( and ) in this mode of operation indicates that this network activity is chaotic, which reflects the richness of its dynamics. No positive Lyapunov exponents exist in the analysis of the statocyst activity during routine swimming.
Clione's sensory network has been studied in detail from a theoretical point of view using rate models, in particular the conditions to generate WLC dynamics [28], [30], [34]. Less attention has been paid to modeling the CPG responsible of the animal's movements and the transformation of the sensory dynamics into a motor activity. This can be addressed with more detailed biophysical models of neurons and synapses to better reproduce the connectivity and rhythms observed in the in vitro experiments.
Experimental evidence shows a significant correlation between the activity of the SRCs and the wing CPG cells (e.g. see Figure 5 in [15]). The dynamics of these sensory and motor networks have very similar time scales. Thus, we have developed a simple cerebral interface (CG1, CG2 and CG3) between the statocyst and a model wing CPG (see Figure 5 and “Methods and Models” section for details). The model cerebro-pedal interneurons integrate the activity of the sensory network and relays it to the wing CPG. In this interface CG1 and CG2 inhibit each other to avoid contradictory simultaneous left and right signals from the SRCs. Figure 3B shows the behavior of the model cerebral interface in response to a winner take-all mode (left panel) and a winnerless competition behavior (right panel) in the statocyst conductance based model. During simulation of routine swimming, only the SRC pressed by the statolith is active. This results in the activation of CG1 or CG2. During the simulation of a deviation from the preferred head up position, illustrated in the left panel of Figure 3A, CG1 stops firing while CG2 becomes active (left panel of Figure 3B). The activity of CG3 is modulated by this change, which leads to a short temporal variation in the regular beating pattern generated by the CPG (see below). During hunting behavior, the statocyst generates a stronger activity in all SRCs and, consequently, increases the cerebral interneuron activity, which maintains the competition of the CG1 and CG2 cells with an irregular bursting pattern.
Clione's wing CPG rhythm consists of the fire alternation of two half centers: the dorsal and ventral neural groups [18], [20]. In our CPG model (Figure 5) these groups are composed of three neurons. Each of these neurons represents an electrically coupled group of cells that fire synchronously. We use the same nomenclature as in [18]. The rhythm generators are neurons 7 and 8, while neurons 1A, 3, 2A and 4 are the motoneurons that innervate the wing muscles (see the network details in the “Methods and Models” section). All the neurons in the same group (7, 1A and 3 in the dorsal; and 8, 2A and 4 in the ventral) fire synchronously, so here we will analyze the rhythm generated by 1A and 2A cells. In the winner take-all mode of the model statocyst, the response pattern generated by the wing CPG model consists of the alternation of firing in interneurons 7 and 8 that, respectively, drive motoneurons 1A and 3 in the dorsal group; and 2A and 4 in the ventral group. This pattern in the models is similar to the one observed in in vitro electrophysiological recordings of Clione's motoneurons during routine swimming (c.f. left panel in Figure 3C and top panel in Figure 6). Only small changes in the mean rhythm frequency can be observed depending on the SRC pressed by the statolith (Table 1). In this behavioral mode, motoneuron 1A is active in the dorsal phase, and 2A motoneuron is active in the ventral phase, thus driving the wing flapping [18], [20]. The left panel in Figure 3C illustrates the response of the CPG model to the simulation of a transient change in the Clione's preferred orientation. In this situation, a fast change in the beating rhythm occurs. This fast motor response is always coherent with the change in the sensory network, i.e., the same sensory input (a specific change in the SRC pressed by the statolith) produces the same output in the CPG. These changes can presumably be translated into the wing movements required to generate the compensatory gravitational reflexes needed to correct small deviations in the animal's orientation [19], [21]. Note that, as can be observed in Video S1, the corrective beatings consist of successive dorsal or ventral movements. Immediately after that, the regular beating is restored.
On the other hand, although the sequential activations in the statocyst model network are highly irregular during hunting, the wing CPG produces a coordinated rhythmic pattern that could generate the complex motion observed during hunting in the behavioral experiments. As in the biological network (c.f. right panel in Figure 3C and bottom panel in Figure 6), this rhythm is non-regular and at higher frequency than during the routine swimming [18], [20]. The calculation of the Lyapunov exponents from the vector field of the wing CPG model in this situation yields two positive Lyapunov exponents: and . There are no positive Lyapunov exponents during the simulation of routine swimming. This means that the activity in the CPG model changes from regular to chaotic depending on the sensory context.
As Video S2 shows, hunting behavior in the living animals consists of different hunting episodes with resting times in between. Even though Clione generates a highly irregular sensory activity during these episodes, this activity has to be coordinated to be effective in driving the hunting search. During in vitro fictive hunting, specific activation sequences appear among the recorded SRCs (highlighted as grayed regions in Figure 2). The activity recorded during different hunting episodes in the physiological experiments can be classified into different types using a principal component analysis (PCA). In this representation, previous experimental results have shown that similar patterns in the sensory network induce similar patterns in the motor system [15], pointing to the significant correlation between the activity of the SRCs and the activity of the motor cells [15], [16]. To validate that our model reproduces these experimental results and assess how the different sensory activations are translated into motor commands, we used for the simulated data a PCA analysis similar to the one employed before for experimental recordings. With this kind of analysis we could display a high-dimensional dynamics in a three dimensional representation.
For our model analysis, we defined hunting episodes as time windows where similar sequences, in terms of the duration of specific patterns of sequential activations of the six SRCs, appear consecutively in the sensory dynamics at least three times. As an example, the top panels in each row of Figure 7 show three different representative types of hunting episodes in the statocyst model network. Type A corresponds to sequences of long activations of SRCs 1 and 4, and short activations of SRCs 2, 3, 5 and 6. Type B corresponds to sequential activations of similar length in all the SRCs. And type C corresponds to long activations of SRCs 2, 3 and 5, and short activations of SRCs 1, 4 and 6. In general, the specific activation pattern in the statocyst model during hunting arises from different departing activations of the SRCs and/or different CHI neuron inputs. However, if the simulation of hunting is long enough, the WLC dynamics in the statocyst network evolves to different activation patterns at different intervals. Therefore, in this situation, a given time series can contain hunting episodes of different types. For convenience, in our analysis of hunting behavior we used this approach while studying long time series of hunting. In these time series, first we identified and classified the different hunting episodes into different types according to the duration of specific patterns of sequential activations among the SRCs. Once different hunting episodes were identified, we analyzed the activity pattern of the model during these time windows with the PCA.
The first three principal components of the sensory and motor signals analyzed here explain more than 90% of the total variability of the signals, thus the PCA of these time series can be used to characterize the dynamics of the sensory and the motor network. The percentage of the variability explained by each of the three principal components is the following: first PC , second PC ; third PC, for the statocyst; and first PC , second PC ; third PC, for the wing CPG. Bottom panels in each row of Figure 7 show the PCA representation of the activity for three examples of hunting episodes in the statocyst model network, as well as, the corresponding representation for the motor activity. Note that similar patterns in the statocyst model network evoke similar patterns in the motor network as observed in the experimental results [15].
In our simulations of hunting behavior, we observe that the irregularity of each hunting episode of the sensory network is built out of sequential switchings among the SRCs, with activation phase locks of different durations involving a set of neurons activated in a given time window. By identifying specific activation phase locks in the sensory network (Figure 8, dashed rectangles), we saw that they are transformed into a specific fast irregular beating command of the wing CPG. To assess the response of the wing CPG to each of the activation phase locks, we first detected them (with the method described in the “Methods and Models” section) and then we analyzed the corresponding motor output. The motor response was characterized by the peristimulus time histograms (PSTH) of 1A and 2A neurons. In this analysis we searched for activation sequences of at least four SRCs which appeared a minimum of 30 times in different time series of 120 seconds, and we aligned the motoneuron spikes to the beginning of the activation sequence to calculate the PSTHs (Figure 8). Although in this search for activation phase-locks we allowed the duration of the activity to be different, the sequence of the switching among the SRCs activated in a given time window was preserved. During hunting, the activity is chaotic both in the sensory and the motor model networks (characterized by two positive Lyapunov exponents). Nevertheless, in 88% to 100% of the cases depending on the specific activation phase lock, a given sequence in the SRC network produces the same stereotyped motor activity in the motoneurons of the wing CPG. Note that the sequential activations are not exactly the same in terms of duration and preceding activity, which may be the source of the small number of missing events. In panels A and B of Figure 8 we illustrate two representative examples of sensory activation phase locks and the corresponding motor response in different hunting episodes. In the first example (panel A), we show (dashed rectangles) the activation phase locks among SRCs 1, 2, 4 and 5 during a hunting episode where the activations for neurons 1 and 4 are long. The corresponding PSTHs in panel C of Figure 8 show that these specific sequential activations induce a strong activity in motoneuron 1A at the end of the activation sequence. This response is produced in 71 out of 74 sequences during a 120 s simulation (6 out of 7 in the time series shown in the figure). The second example of activation phase lock shows another episode in which the activations have a similar duration in all neurons. The sequences shown in this case involve SRCs 2, 3, 5 and 6. As the corresponding PSTHs in panel D of Figure 8 indicate, these sequential activations induce the firing of 1A followed shortly after by 2A (32 out 32 sequences during a 120 s simulation). The fact that specific sequential activations in the sensory network may be interpreted by the CPG and lead to specific events in the motor activity is something that can be used by the system for the coordination of the wing beating.
The dual dynamics of the statocysts in two different behavioral contexts (routine swimming and hunting behavior) are directly translated into the corresponding characteristic motor behavior. The same CPG that controls the periodic wing beating during routine swimming reacts to the irregular commands from the sensory network to generate the hunting motor program. During the winner take-all phases, the sensory gravitational input from the statolith is used by the motor system to react to small deviations from the preferred head up orientation by generating compensatory gravitational reflexes [19], [21]. This behavior is reproduced by our model. During routine swimming the wing CPG model generates a regular pattern of activity able to control the wing beating until a deviation is simulated. This generates a fast transient response that could produce the required correction movements. After that, the regular pattern starts again. On the other hand, the activity generated during the winnerless competition phases acts on the same network to shape the irregular but coordinated search motor behavior [15]. In this case, the CPG model produces a motor output activity that is able to drive this complex motion.
It is difficult to experimentally assess the study of all the stages present in a sensory-motor transformation. Because of the lack of experimental results, there are also very few models that address the transformation of sensory dynamics into a motor program. Clione limacina is an experimental model in which this study is possible. In this paper we have discussed the dual role of a sensory organ in relationship to two types of motor behaviors during routine swimming and hunting behavior. To address this issue, we have built neural network models ranging from a single statocyst network to a system where the activity of the statocyst is transferred to a motor wing CPG through a simple model of cerebral ganglia. The statocyst network was used to reproduce the two types of dynamics observed during routine swimming and during hunting behavior, namely, winner take-all dynamics and winnerless competition. Our simulations show that a model built with conductance-based neurons and realistic inhibitory connections can display both types of dynamics depending on the stimulation from the statolith or from the hunting neuron, respectively. We have also shown that these two dynamics can be used by the wing CPG to generate the characteristic rhythmic motion during routine swimming, and a fast irregular motion that is observed during hunting behavior. The nature of the sensorimotor transformation cannot be described as a simple mapping but as a dynamical process that involves reading one spatio-temporal code (sensory) and translating it into a different spatio-temporal code in the CPG (motor).
Clione's hunting behavior is directed toward locating and capturing prey. The motor strategy during hunting is determined by the difficulty in detecting an odor source in the water. As it has been shown in the behavioral experiments reported in [15], after triggering the hunting, Clione's search movements are not directed by the prey since they continue even when the prey is taken out of the water (see Video S2). In an animal with an undeveloped visual system, such a motor strategy increases the chance of locating and capturing the prey.
Our modeling results suggest that, in spite of the intrinsic irregularity of the switching sensory dynamics (a network phenomena) during hunting, specific activation phase locks in the sensory WLC dynamics are transformed into specific motor events in the wing CPG activity. In this sense, we can consider these sequential activations as coordination patterns inside the irregular statocyst dynamics. This is particularly relevant in the context of a complex intrinsic sensory dynamics that has to be transformed into an effective motor program. Hunting search is highly irregular, but nevertheless organized and coherent. From this perspective, it makes sense that a complex sequential activation of sensory neurons contains coordination cues in the form of activation phase locks that can be interpreted and executed by the motor CPG to generate the motor program. Although we have not addressed it here because of lack of information regarding connectivity, the tail motoneurons could also use cues from the statocyst dynamics to contribute to an effective hunting search in coordination with wing motor activity. Tail movements do not have the repetitive pattern of a CPG output and thus these movements are less restricted and more prompt to be modulated by sensory input.
Beyond the specific role played by the statocysts in Clione's hunting behavior, the results discussed in this work, as well as the experimental results reported in [15], [16] and those obtained by [35] on the pulmonate snail Lymnaea, show that the statocysts can perform dual functions depending on the behavioral context. During routine swimming of Clione, the statocysts perform a purely sensory function and gravimetric reflexes are used for maintaining a vertical spatial orientation. In contrast, during hunting the statocysts participate in generating a hunting motor program. In both cases, the statocyst output is used to drive a CPG, hence, organizing motor behavior. Under our description, the sensory signals are modified to fit the changing behavioral context. In a sense, the statocyst network is fooled by the hunting neuron. However, this sensory dynamics is interpreted by the rest of the nervous system as during routine swimming, which results in a complex hunting search motor pattern. In spite of its irregularity, the statocyst activity can contain coordination cues to organize a complex motion, i.e., a hunting search program. There are some advantages in generating the motor program right at the sensory network, as the rest of the neurons in the sensory-motor transformation can just react normally to this signaling. Another alternative would be to generate the program at the cerebral ganglia. However a strong experimental fact goes against this hypothesis as fictive hunting search cannot occur without the statocysts [15]. The experimental and modeling results reported in this paper support the view that the dual dynamics of the statocyst network by itself can explain the two motor programs observed during routine swimming and during hunting behavior.
Preparations for electrophysiological experiments were made in ice-cold seawater to prevent excitation of nociceptive afferent fibers. The preparation, including cerebral, pedal, and abdominal ganglia with the wing nerves, was pinned to a Sylgard-lined Petri dish as described previously [15]. Extracellular recordings from nerves were made by using glass suction electrodes or stainless-steel electrodes. Intracellular recordings were made using glass electrodes () filled with 3 M KCl. The signals were acquired with a Digidata board (Molecular Devices, Union City, CA) and stored for later analysis with Dataview (http://www.st-andrews.ac.uk/wjh/dataview/). The spikes were sorted from the extracellular recordings in Dataview, using threshold and the spike template. Because there was little superposition in spike firing, we could typically sort four or five units in the statocyst nerves.
Fictive hunting behavior was induced by application of physostigmine as in [24] and [15]. To achieve fictive hunting, the seawater covering the isolated nervous system was replaced by seawater containing physostigmine.
All the equations of our models were numerically solved with a Runge-Kutta6(5) variable step method with a maximum error of .
|
10.1371/journal.pgen.1002094 | Mammalian BTBD12 (SLX4) Protects against Genomic Instability during Mammalian Spermatogenesis | The mammalian ortholog of yeast Slx4, BTBD12, is an ATM substrate that functions as a scaffold for various DNA repair activities. Mutations of human BTBD12 have been reported in a new sub-type of Fanconi anemia patients. Recent studies have implicated the fly and worm orthologs, MUS312 and HIM-18, in the regulation of meiotic crossovers arising from double-strand break (DSB) initiating events and also in genome stability prior to meiosis. Using a Btbd12 mutant mouse, we analyzed the role of BTBD12 in mammalian gametogenesis. BTBD12 localizes to pre-meiotic spermatogonia and to meiotic spermatocytes in wildtype males. Btbd12 mutant mice have less than 15% normal spermatozoa and are subfertile. Loss of BTBD12 during embryogenesis results in impaired primordial germ cell proliferation and increased apoptosis, which reduces the spermatogonial pool in the early postnatal testis. During prophase I, DSBs initiate normally in Btbd12 mutant animals. However, DSB repair is delayed or impeded, resulting in persistent γH2AX and RAD51, and the choice of repair pathway may be altered, resulting in elevated MLH1/MLH3 focus numbers at pachynema. The result is an increase in apoptosis through prophase I and beyond. Unlike yeast Slx4, therefore, BTBD12 appears to function in meiotic prophase I, possibly during the recombination events that lead to the production of crossovers. In line with its expected regulation by ATM kinase, BTBD12 protein is reduced in the testis of Atm−/− males, and Btbd12 mutant mice exhibit increased genomic instability in the form of elevated blood cell micronucleus formation similar to that seen in Atm−/− males. Taken together, these data indicate that BTBD12 functions throughout gametogenesis to maintain genome stability, possibly by co-ordinating repair processes and/or by linking DNA repair events to the cell cycle via ATM.
| Mutations in genes essential for genome maintenance during meiosis can result in severe disruptions to spermatogenesis and subsequent low fertility and/or birth defects in mammals. The mammalian homolog of yeast Slx4, BTBD12, plays a critical role in somatic cell repair in mice. Here, we show that this critical function extends to mammalian germ cells, by examining the effects of a Btbd12 gene disruption in mice. Btbd12 mutant mice show severely reduced fertility, as a result of both pre-meiotic spermatogonial proliferation defects and impairment of proper meiotic progression. BTBD12 appears to be required for normal progression of double-strand break repair events that result in the formation of crossovers between maternal and paternal homologous chromosomes, with Btbd12 mutants displaying an increase in unrepaired breaks, impaired homologous chromosome interactions, and a slight increase in the number of crossover intermediates. BTBD12 protein is also down-regulated in the testes of Atm null mice, supporting previous studies showing that BTBD12 is a target of ATM kinase. These data provide new evidence about the role of BTBD12 in mammalian gametogenesis and are critical to furthering the understanding of the molecular processes involved in meiotic DNA repair.
| SLX1 and SLX4 were identified, together with MUS81 and MMS4 (Eme1 in mammals), in a S. cerevisiae screen for genes required for the viability of sgs1-deficient cells [1]. Slx1 is the founding member of a family of proteins with a predicted URI nuclease domain whose activity is enhanced 500-fold by its interaction with Slx4 [2]. Slx4 can also form complexes with Rad1-Rad10 [3], [4] to effect DSB repair during single-strand annealing in yeast [3], [5]. However, Slx4 can also act independently of both Slx1 and Rad1-Rad10 [2], [6], and is phosphorylated by Mec1 and Tel1, the yeast orthologs of ATR and ATM, respectively, in response to DNA damage [4].
Human, C. elegans and D. melanogaster orthologs of SLX4 were described recently [7]–[10] and named BTBD12 (for BTB domain-containing protein-12), Him-18, and Mus312, respectively. These proteins are considerably diverged from their yeast counterpart; BTBD12 encodes a 1834 amino acid protein, approximately 2.5-times larger than the yeast protein, and resembles its lower eukaryotic orthologs mostly in its C-terminal SAP and CCD domains [7]. Like the yeast ortholog, the human protein is a substrate of the ATM/ATR kinases [11] and its depletion also results in DNA damage sensitivity [8]. Recently, a subset of Fanconi anemia (FA) patients were found to have biallelic mutations in BTBD12, making this gene a novel complementation group for this disorder [12]. A complex of BTBD12 and SLX1 displays robust Holliday Junction (HJ) resolvase and 5′ flap endonuclease activity in vitro, and mammalian BTBD12 also binds to, and enhances the activity of, several DNA repair proteins including MUS81 [7], [10] and the MSH2-MSH3 heterodimer of the DNA mismatch repair (MMR) family [8], suggesting a role for this protein as a docking platform for structure-specific endonucleases.
Recent reports in D. melanogaster and C. elegans indicate that the Btbd12 orthologs, Mus312 and Him-18, respectively are essential for normal meiotic progression [13], [14]. In addition, Him-18 appears to function pre-meiotically in the germ line, being required for repair at stalled replication forks [13], suggesting that Him-18 functions throughout germ cell development to maintain genomic integrity. Given these data, the primary goal of the current studies was to understand the function of BTBD12 in the germ line of mice, with the hypothesis that BTBD12 may be critical for the processing of homologous recombination intermediates, whether as the result of replication errors during pre-meiotic proliferation, or during the repair of double strand breaks (DSBs) that underlie meiotic recombination. For the latter, our studies were aimed at investigating the role of BTBD12 in the regulation of meiotic recombination events during prophase I in mammals, particularly those that ensure accurate segregation of maternal and paternal chromosomes at the first meiotic division. Principal amongst these is the formation of DNA crossovers between the homologs, as initiated by DSB induction through the activity of the SPO11 endonuclease [15], [16]. The resolution of DSBs can be achieved through the recruitment of various repair pathway complexes, to produce crossovers (CO) or noncrossovers (NCO). The fact that CO formation occurs with tight precision, coupled with the observation that only a small subset of total DSBs will become COs, suggests that orchestration of DSB repair events, and the various repair pathways that give rise to either CO or NCO events, is highly regulated at the molecular level (reviewed by [17]).
Two pathways have been defined for CO formation, the first involving the so-called Class I or “ZMM” pathway (for ZIP3, MSH4/5 and MER3), and the second, Class II pathway, involving the Mus81 endonuclease, which functions as a heterodimer with EME1 (Mms4 in yeast) [18]–[21]. In S. cerevisiae, and possibly in the mouse, this Class II pathway appears to be restricted to a subset of DSBs that may be aberrant in structure and/or that may be processed initially by the RecQ helicase, Sgs1/BLM (yeast/mouse ortholog; [20]–[23]). These aberrant DSB repair intermediates (or joint molecules, JMs) include a variety of structures that result from secondary strand invasion events, usually involving independent activities of each end of the DSB, and/or from closely spaced DSBs. These aberrant JMs have been demonstrated biochemically in S. cerevisiae [20]–[23], but not yet in mammals. Under wildtype conditions in budding yeast, Sgs1 can disassemble and/or process many of these aberrant DSB repair intermediates towards NCO or Class I CO fates [20], [21]. However, a small proportion of them cannot be processed in this manner and thus become the target of MUS81-driven crossing over.
In the mouse, the class I pathway accounts for some 90–95% of COs, and the major intermediate marker for these events is the accumulation, in pachynema, of the MutL homolog heterodimer, MLH1 and MLH3 [24]–[27]. The remaining events are processed via the MUS81-dependent Class II pathway [27], [28]. Interestingly, however, our studies in mouse have demonstrated that the loss of Mus81 results in the recruitment of 5–10% additional MLH1/MLH3 foci during pachynema, and that these additional foci may act to maintain CO rates at normal levels in the absence of MUS81 [28]. This suggests that a unique, possibly mammalian-specific, level of integration exists between the two crossover pathways. Importantly, BTBD12 interacts with many of the key players in both CO pathways, including BLM (reviewed by [29]), leading us to hypothesize that BTBD12 may functionally integrate the different CO pathways during mammalian meiosis.
We obtained a mouse line from the European Conditional Mouse Mutagenesis Program (EUCOMM), harboring a Frt-flanked βGeo cassette upstream of LoxP-flanked exon 3 of Btbd12 gene, and as initially described by Crossan et al [30]. We call this conditional genetrap allele Btbd12βGeoFlox. In wildtype mice, BTBD12 protein localizes to spermatogonial and spermatocyte populations of the adult testis and is dramatically down-regulated at these sites in the mutant animals. Weak staining against BTBD12 persists, however, in the Btbd12βGeoFlox/βGeoFlox testes, but the protein fails to be recruited to meiotic chromosome cores to detectable levels. Despite residual protein, the mice are sub-fertile as a result of a reduced spermatogonial population coupled with failure to progress normally through meiotic prophase I resulting in less than 15% normal numbers of spermatozoa. Importantly, BTBD12 localization to meiotic chromosomes appears to be dependent on ATM, since the presence of BTBD12 protein is lost in Atm−/− males, and Btbd12βGeoFlox/βGeoFlox mice show genomic instability similar to that seen in the absence of ATM. Taken together, our data suggest that BTBD12 is a key substrate of ATM in the mammalian germ line, playing a role in the spermatogonial stages of gametogenesis, as well as in the entry into, and progression through, prophase I.
Wildtype testis sections were stained with a commercial anti-BTBD12 antibody from Novus Biologicals that was raised against amino acids 1650–1700 of human BTBD12 at the c-terminus of the protein (Figure 1A, 1B). BTBD12 is expressed ubiquitously in the mouse, and is found in both adult testis and fetal ovary [8]. In adult testis of wildtype mice, the protein was found to localize strongly to spermatogonia and spermatocytes (Figure 1A, black arrowheads and arrows, respectively). At earlier stages of spermatogenesis, prior to entry into meiosis, the signal for BTBD12 protein appears to localize through the nucleus, with increased intensity of signal around the nuclear periphery. Upon entry into prophase I, the BTBD12 signal becomes more punctate in nature, associating with the increasingly condensed chromatin, but continuing to occupy the majority of the nuclear space. Additionally, BTBD12 protein localized to bivalent chromosomes on the meiotic spindle during metaphase of meiosis I (Figure 1A, white arrowheads and inset panel).
BTBD12 localization was evaluated on chromosome spread preparations of prophase I spermatocytes. In all chromosome “spread” preparations presented herein, we utilize the major protein component of the axial element of the synaptonemal complex (SC, the meiosis specific structure that assembles along and between each homologous chromosome pair), SYCP3, to visualize meiotic chromosome cores. We utilized an affinity-purified antibody that targets amino acids 1–350 of murine BTBD12 (antibody “NT”). The chromosome spreads from wildtype adult males (Figure 1 Ci-vi) showed accumulation of BTBD12 protein in a punctate staining pattern along the chromosome cores in zygonema (Figure 1 Cii), becoming more intense by pachynema (Figure 1Ciii). In some cases, increased staining was observed within the sex body (94% of wildtype cells having BTBD12 in this domain, compared to 0% in the Btbd12 mutants; n of 33 and 31, respectively), in line with the fact that BTBD12 is a target of ATM, which plays a significant role in the formation of this sub-nuclear domain during mid-prophase I [31], [32]. One autosomal chromosome from a pachytene cell of each genotype is shown, enlarged, with the BTBD12 staining offset from the SYCP3 staining to facilitate visualization of the punctate pattern of the Btbd12 signal (Figure 1Cvi). By diplonema, BTBD12 was no longer present on the cores, but remained associated with the centromeres (Figure 1Civ), and had disappeared from the nucleus by diakinesis (Figure 1Cv).
The Btbd12βGeoFlox allele contains an Frt-flanked βGeo insertion into the Btbd12 locus upstream of LoxP-flanked exon 3. The expectation, therefore, is that this allele would act as a gene trap, reducing expression of wildtype Btbd12 mRNA. Consistent with this, in the Btbd12βGeoFlox/βGeoFlox mutant testis sections, we observe weak BTBD12 protein staining persisting in some cells, particularly those close to the basal membrane of the tubules, which are presumptive spermatogonia (Figure 1B, red arrows). This persistent staining, however, was much fainter than that seen in the wildtype testis sections, indicating a significantly lower abundance of BTBD12 protein in the mutant animals, and suggesting a splicing event around the βGeo cassette to produce BTBD12 protein. Importantly, there was no staining evident on metaphase chromosomes in the Btbd12βGeoFlox/βGeoFlox testes.
In spermatocytes from Btbd12βGeoFlox/βGeoFlox mutant littermates, BTBD12 did not appear to localize to chromosome cores with as intense a signal as in wildtype cells (Figure 1Di-vi), although faint staining was observed at pachynema in the Btbd12βGeoFlox/βGeoFlox mutants. This staining, however, was barely detectable above background, even at higher magnifications (Figure 1Diii and vi). Thus, the persistent BTBD12 signal observed in testis sections from Btbd12βGeoFlox/βGeoFlox males is not associated with prophase I chromosome cores, or is present at levels that are undetectable on chromosome spreads.
To confirm the presence of BTBD12 protein in wild type testes, and to explore the status of BTBD12 protein in the homozygous mutant animals, western blots were performed using whole testis protein extracts from adult and juvenile animals (Figure 2). Two antibodies were utilized, one C-terminal and one N-(not shown) terminal, as described above. Both antibodies produced a band of similar size in protein extracts from wild type adult testis (lane 1), with decreasing amounts observed in Btbd12βGeoFlox/+ heterozygotes (lane 2) and further reduced expression observed in Btbd12βGeoFlox/βGeoFlox mutant protein extracts (lane 3). Quantitation of BTBD12 protein levels in testis extracts from all three genotypes revealed a decrease from wildtype levels of 23.5% and 45.6% for heterozygotes and homozygous mutant samples, respectively (Figure 2 graph). The presence of residual 130 kDa protein in the homozygous mutant animals is in line with the persistent protein signal observed in immunohistochemical staining of testis sections from these animals (Figure 1B), and suggests that the mutant allele is transcribed at a reduced level compared to the wildtype allele. As expected, the lanes containing Btbd12βGeoFlox/+ and Btbd12βGeoFlox/βGeoFlox protein (lanes 2 and 3 in Figure 2) also show positive staining for β-galactosidase at approximately the same size. No β-galactosidase band is observed in the wildtype testis protein lane. Taken together, these results show that the BTBD12 protein present in testis extracts from Btbd12βGeoFlox/+ and Btbd12βGeoFlox/βGeoFlox males encompasses some portions of both the N-terminus and C-terminus of the protein, and most likely arises as a result of splicing around the FRT-flanked cassette, but that expression of this protein is dramatically reduced by the presence of the βGeo cassette. Moreover, βGeo cassette itself is also transcribed from the targeted allele, either as a distinct protein and/or as a fusion with BTBD12. We favor the former option given that the size of the BTBD12 and βGeo bands are similar.
Btbd12βGeoFlox/βGeoFlox mutant mice are viable and are born at slightly lower than expected Mendelian rates. Out of a total of 395 pups born in 65 litters from heterozygote breedings, 66 of them were mutants, compared with an expected frequency of 98.75 (Table S1). These data are significantly different, at the p<0.05 level, from the expected percentages mutant or wildtype numbers (P = 0.012, χ2 test). This rate is slightly higher than that seen in the previous report describing these mice [30]. Occasionally we observe that mutants exhibit lower birth weights than their wildtype litter mates (not shown) but, by adulthood, these animals have regained weights comparable to their siblings. Thus, Btbd12βGeoFlox/βGeoFlox adult mice were not significantly smaller than wildtype littermates (average weights 19.5 and 21.3 g, respectively P = 0.25, unpaired T-test) but exhibited varying degrees of anophthalmia and microphthalmia from birth. 33 out of 66 (50%) mutants showed defects of one or both eyes. No such deformities were apparent in wildtype litter mates indicating a possible role for BTBD12 in eye development and in line with previous reports describing this EUCOMM mouse line [30].
It is well known that mammalian species with mutations in Atm show an increase in genomic instability (GIN), including defects in cell cycle regulation, sensitivity to DNA damage-inducing agents, and chromosomal aberrations (reviewed by [33]). Since BTBD12 is believed to be a direct target for phosphorylation by ATM/ATR, we assessed both wild type and Btbd12βGeoFlox/βGeoFlox mutant mice for GIN. A micronucleus formation assay was performed on peripheral blood from both Btbd12βGeoFlox/βGeoFlox mutants and wild type littermate controls [34], [35]. The Btbd12βGeoFlox/βGeoFlox mutant animals showed over a two-fold increase in micronucleus formation compared with wild type mice, which was statistically significant (Figure S1; p<0.0001). This result indicates that these mutants have a high level of GIN in their somatic cells approaching, though not as great, as that seen in Atm null mice [36].
Btbd12βGeoFlox/βGeoFlox mutant mice showed significantly decreased testis size when compared to wildtype littermates (approximately 25%, Figure 3A, 3B, p<0.0001). Sperm counts performed on the two cohorts of mice revealed that Btbd12βGeoFlox/βGeoFlox mice have only about 10% of the amount of epididymal sperm found in wildtype littermates (Figure 3C), and exhibit dramatically reduced fertility, with only 3 litters born to two different Btbd12βGeoFlox/βGeoFlox males over the period of 9 months, with the youngest male to sire a litter being approximately 7 weeks old. Female mutants were sterile, with two mutant females bred to a fertile male, over a period of a year, yielding no pregnancies, while wildtype cage mates produced healthy offspring from the same male.
H&E staining of testis sections from both wildtype and Btbd12βGeoFlox/βGeoFlox mutant mice at both 3 and 8 weeks of age (Figure 3D, 3E, 3H, 3I) showed that the seminiferous tubules of Btbd12βGeoFlox/βGeoFlox males are extremely variable in their cell density and also in the progression through meiosis. For example, in 8-week old Btbd12βGeoFlox/βGeoFlox mutant mice, neighboring tubules showed almost normal tubule morphology, juxtaposed to almost empty, abnormal tubules (Figure 3I). Immunohistochemical staining with the spermatogonial and early spermatocyte cell marker, GCNA-1, showed a severe depletion of early germ cells within the tubules of the Btbd12βGeoFlox/βGeoFlox compared with those within the wildtype litter mate mice, at both 3 week and 8 weeks of age (Figure 3F, 3G, 3J, 3K). By contrast, the proliferating cell marker, PCNA, showed a similar staining pattern in the majority of tubules in both wildtype and Btbd12βGeoFlox/βGeoFlox mice (Figure 3L, 3M) suggestive of normal progression through spermatogonial divisions and in self-renewal capabilities. The maintainenance of PCNA signal in spite of reduced PGC pool is suggestive of prolonged S-phase. In line with this, we observed increased TUNEL labeling of apoptotic cells in testis sections of Btbd12βGeoFlox/βGeoFlox, mostly during meiosis I. The majority of these cells are undergoing apoptosis at around the time of exit from prophase I, but some also appear to be in mid-prophase (Figure 3N, 3O, arrowheads). In a few instances, some TUNEL-positive cells appear to be in pre-meiotic stages (Figure 3O, arrows), but these are clearly fewer in number than those apoptotic cells in prophase I. Collectively, these results demonstrate a loss of germ cells from the testis of Btbd12βGeoFlox/βGeoFlox males, starting as early as the first wave of meiotic entry within the first three weeks of postnatal life.
Given the apparently normal proliferative capacity of spermatogonia in testes of 8-week old Btbd12βGeoFlox/βGeoFlox mice, we questioned why the tubules of 3 week-old mice were so heterogeneous with respect to cellular density. If a failure of spermatogonial proliferation is not the cause of the lack of cellularity of certain tubules in the Btbd12βGeoFlox/βGeoFlox males, then a second possibility is that the testis of these mice fail to be populated with appropriate numbers of spermatogonial precursors, known as pro-spermatogonia or gonocytes, during development. To investigate this option, testes were obtained for both wildtype and Btbd12βGeoFlox/βGeoFlox males between embryonic (e) day 18 and day 3 post-partum (pp), and prospermatogonia were visualized with antiserum against GCNA-1. In wildtype males, the G0-arrested prospermatogonia population is established around embryonic day 12.5–16.5 following migration of the primordial germ cells (PGC) to the genital ridge [37], [38], the exact timing being somewhat controversial, and remain quiescent until just prior to birth [39]. During the period between e18 and day 3pp, a large number of prospermatogonia are lost by apoptosis. During that time, approximately 1 to 7 prospermatogonia may be observed within the seminiferous cords of the developing testis, and these cells then start to proliferate from day 4 pp onwards [39]. In Btbd12+/+ males, these large round cells appeared separated from the basement membrane by the Sertoli cells, which are more columnated in appearance, and they stained readily with numerous markers including GCNA-1 (Figure 4A–4C, brown cells) and mouse Vasa homolog (MVH; not shown) from e16 onwards. By day 3 pp, every testis cord section contains on average 2.08±0.19 prospermatogonia per cord, representing a range of 1 to 10 cells (Figure 4C, 4M; Table S2), having declined dramatically at around the time of birth. In Btbd12βGeoFlox/βGeoFlox males at e16, normal numbers of GCNA-1 positive prospermatogonia are observed (Figure 4D), but by e18, their numbers have declined significantly (Figure 4E, 4M; Table S2). By d3 pp, the majority of testis cords contain no prospermatogonia, with a mean prospermatogonia content of 0.92±0.28 (Figure 4F, 4M; Table S2). The Sertoli cell populations in both Btbd12+/+ and Btbd12βGeoFlox/βGeoFlox males appeared normal throughout (Figure 4B, 4D, 4F). In line with this reduced cellularity within the testicular cords, we observed a marked increase in apoptosis, as measured by TUNEL labeling of testis sections from wildtype (Figure 4G–4I, 4N) and mutant (Figure 4J–4L, 4N, arrows) animals, particularly at e16 and e18 (Figure 4J, 4K, 4N; Table S2). These data demonstrate that the population of spermatogonia within the testes of Btbd12βGeoFlox/βGeoFlox males is markedly lower than that seen in wildtype as a result of early loss of these cells after arriving at the genital ridge, and suggesting that the proliferation of PGCs in the developing testis is dependent on BTBD12.
To examine meiotic prophase I progression, chromosome spreads from both Btbd12βGeoFlox/βGeoFlox and wildtype spermatocytes were stained with antibodies against SYCP3 and γH2AX (Figure 5A–5H), as a marker for DNA DSBs during prophase I. γH2AX accumulated on leptotene spermatocytes similarly in both wildtype and Btbd12βGeoFlox/βGeoFlox cells (Figure 5A, 5E), demonstrating the appearance and processing of DSB events. By zygonema, however, γH2AX localization began to diminish on the chromosome cores of wildtype spermatocytes (Figure 5B), coincident with the onset of DSB repair processes. By pachynema, and into diplonema, γH2AX localization was only restricted to a strongly-stained domain coincident with the sex body (Figure 5C, 5D and [40], [41]). By contrast, γH2AX staining is apparent at zygonema in the Btbd12βGeoFlox/βGeoFlox spermatocytes (Figure 5F), but persists along the autosomes well into pachynema, at which time this staining is limited to the sex body in wildtype cells (Figure 5C, 5G; ). γH2AX staining remains apparent well into diplonema in the Btbd12 mutants (Figure 5D, 5H; Figure S2).
To investigate the progression of DSB repair, we assessed RAD51 distribution along SCs during prophase I. Specifically, we were interested in observing persistence of RAD51 signal in Btbd12βGeoFlox/βGeoFlox spermatocytes as an indication of unrepaired, or delay in repair of, DSBs. As expected, we observed progressive loss of RAD51 from SCs in wildtype spermatocytes entering pachynema (Figure 5I; Table S3), whereas RAD51 focus numbers remained elevated through pachynema in Btbd12βGeoFlox/βGeoFlox spermatocytes (Figure 5M; Table S3). This difference between wildtype and mutant spermatocytes in terms of pachytene RAD51 foci (means of 26.2 and 45.1, respectively) was statistically significant.
The BRCT domain-containing protein, TOPBP1, functions in replication and DNA damage checkpoint processes, and in meiosis, it localizes to sites of DNA damage in response to DSBs [42], [43]. TOPBP1 is also known to be required for ATR binding/activation in a number of organisms [44]–[46] and, along with ATM/ATR kinases, may be a part of the machinery that monitors recombination during prophase I and activates the meiotic checkpoint. Indeed, TOPBP1 localizes exclusively to sites of SPO11-induced DSB, as demonstrated by co-localization with γH2AX [47].
Despite the massive increase in γH2AX staining, Btbd12βGeoFlox/βGeoFlox spermatocytes showed no difference in the TOPBP1 localization pattern compared to that seen in chromosome spreads from wild type spermatocytes (Figure 5J–5P). TOPBP1 accumulated on synapsed chromosomes during zygonema, and gradually decreased until it remained only at the sex chromosomes during pachynema, indicating that this signaling pathway is not affected by the loss of BTBD12 from the chromosome cores, in contrast to the persistent RAD51 observed on SCs from Btbd12βGeoFlox/βGeoFlox spermatocytes.
MLH1 and MLH3 localization was used to examine the progression of DSB repair events via the “ZMM”, Class I CO pathway (Figure 5Q–5T), which is overseen by key members of the DNA mismatch repair (MMR) family: MSH4, MSH5, MLH1 and MLH3. MLH1 and MLH3 form a heterodimer that binds to the MSH4/MSH5 heterodimer in pachynema [48]–[52]. MSH4/MSH5 assemble on DSB repair sites in zygonema in numbers that, in mice at least, far exceed the final tally of chiasmata [53], [54]. The number of these foci is then pared down through prophase I, but still maintains levels that are approximately two-fold higher than the final chiasmata count [53], [54]. Association of MLH1/MLH3 with a subset of these sites is thought to stabilize these events, resulting in the resolution of these structures via the class I CO pathway [24]–[26].
In spermatocyte spreads from both wildtype and Btbd12βGeoFlox/βGeoFlox males, MLH1 and MLH3 foci arise at pachynema, at frequencies of 1–2 foci per chromosome, which is comparable to that seen previously in wildtype [25], [26]. The temporal and spatial dynamics of MLH1 and MLH3 association with the SYCP3-positive chromosome core was similar for wild type and Btbd12βGeoFlox/βGeoFlox spermatocytes (Figure 5Q–5T), suggesting similar progression of class I CO events. When foci were counted and compared between wildtype and mutants, however, we observed a slight, yet statistically significant, increase in foci number for both MLH1 and MLH3 in Btbd12βGeoFlox/βGeoFlox spermatocytes, equating to approximately 1 additional focus per nucleus for each (Figure 5S, 5T, p = 0.0018 for MLH1 and p = 0.0389 for MLH3, unpaired T test). For MLH1, the mean number of foci was 23.28 and 25.56 for wildtype and Btbd12βGeoFlox/βGeoFlox males, respectively (Table S3). For MLH3, the mean number of foci was 24.62 and 25.57 for wildtype and Btbd12βGeoFlox/βGeoFlox males, respectively. MLH1 foci were also significantly elevated in meiotic spreads from female day e19 embryos, with average MLH1 focus numbers of 22.29 and 24.00 in wildtype and Btbd12βGeoFlox/βGeoFlox, respectively (p = 0.02, data not shown). The earlier recombination intermediate MSH4 remained the same in both wildtype and Btbd12βGeoFlox/βGeoFlox spermatocytes (data not shown).
To assess synapsis, chromosome spreads from Btbd12βGeoFlox/βGeoFlox spermatocytes were stained with an antibody against the central element component, SYCP1 (Figure 5U–5X). Interestingly, at pachynema, approximately 10% of cells harboring the mutant Btbd12 locus showed abnormal synapsis, compared to less than 1% for wildtype cells (not shown), characterized by frequent pairing between more than two chromosomes, incomplete synapsis at pachynema, and synapsis between chromosomes of differing lengths (Figure 5S–5U). In some cases, these synaptic errors appeared to persist into diplonema without first resulting in apoptosis (Figure 5X).
To assess the impact of loss of BTBD12 on the first meiotic division, we prepared air-dried diakinesis chromosome spreads and stained them with Giemsa. Chiasmata formation occurred in both wild type (not shown) and Btbd12βGeoFlox/βGeoFlox spermatocytes (Figure 6A). The number of diakinesis stage cells was severely depleted in Btbd12βGeoFlox/βGeoFlox mice, suggesting loss of these cells prior to completing prophase I and/or delayed progression through to diplonema. Of the diakinesis cells that we obtained from Btbd12βGeoFlox/βGeoFlox males, however, none showed any changes in chiasmata counts compared to that seen in wildtype litter mates (Figure 6B).
To ask whether the number of chiasmata present in Btbd12βGeoFlox/βGeoFlox mutant animals is sufficient to cause appropriate separation of chromosomes during the first meiotic division, an examination of oocytes undergoing metaphase I to anaphase I progression was undertaken. Oocytes from wildtype and Btbd12βGeoFlox/βGeoFlox mice were stained with an antibody against β-tubulin to show the meiotic spindle and DAPI to stain the DNA (Figure 6C). The arrangement of chromosomes on the meiotic spindle in Btbd12βGeoFlox/βGeoFlox mutant oocytes was similar to wild type controls, with 53 cells examined from each genotype. In both cases, occasional cells appear to show one or two misaligned chromosomes, but this occurs at similar rates in oocytes from wild type and Btbd12βGeoFlox/βGeoFlox females (Figure 6C, arrow; 3/53 cells for both wildtype and mutant). Taken together, these results indicate that, despite elevated MLH1/MLH3 focus numbers, chiasmata counts are unaffected in Btbd12βGeoFlox/βGeoFlox mutant animals. Moreover, these chiasmata can, and do, result in normal metaphase I progression, resulting in appropriate chromosome segregation at the first meiotic division. However, the fact that very few diakinesis cells are obtained suggests either a delay in prophase I completion or loss of cells through prophase I prior to diakinesis.
BTBD12 was first identified as a potential kinase target of ATM [11]. To investigate the functional interaction between these two proteins, we examined the localization of BTBD12 on meiotic chromosomes in the absence of ATM. When co-immunostaining for BTBD12 and SYCP3 was performed on spread preparations from Atm null spermatocytes, we observed a complete absence of BTBD12 protein on chromosome cores (Figure 7A), compared to the punctate pattern of BTBD12 staining observed on chromosome cores wild type spermatocytes (Figure 1C). TOPBP1, however, localizes normally to the cores in Atm null cells (Figure 7B) in line with the demonstration in budding yeast that Mec1(ATR) activation is dependent on Dpb11(TOPBP1), rather than vice versa, and that Mec1, in turn, mediates only the functional interaction between Slx4(BTBD12) and Dpb11(TOPBP1), rather than regulating directly the localization of TOPBP1 [44], [55], [56].
BTBD12 protein is also down-regulated in Atm−/− males, when compared to the wildtype littermate controls, with Atm+/− males showing intermediate levels of BTBD12 protein (Figure 2, lanes 4–6). Testis protein extracts from Atm null mice show a decrease in BTBD12 protein to 70.3% of that seen in wildtype males, compared to a decrease to 26.9% in Atm heterozygotes (Figure 2 graph). Since ATM deletion results in pachytene meiotic failure in mice [57], [58], western blots were also performed on juvenile testis extracts at day 17 post-partum to ensure that all protein extracts from Atm+/+, Atm+/−, and Atm−/− contained equivalent cell populations. Indeed, even with higher proportions of leptotene and zygotene cells present in the day 17 extracts, protein from Atm−/− males showed a depletion of BTBD12 by 67.5% compared to Atm+/+ males (Figure 2, lanes 7 and 8), with Atm+/−protein extracts showing an intermediate decrease of 43% compared to wildtype levels (lane 9). Thus, even when taking into account the earlier meiotic failure of Atm−/− males, BTBD12 protein is severely reduced in the absence of ATM.
The results presented herein describe, for the first time, the role of BTBD12 (SLX4) in mammalian gametogenesis. Our data show that BTBD12 plays dual roles in gametogenesis, firstly in facilitating primordial germ cell proliferation and establishment of the spermatogonial pool, possibly by ensuring genome stability, and secondly, in meiotic recombination events. These studies demonstrate that BTBD12 protein localizes to spermatogonia and spermatocytes of the testis. In the latter, BTBD12 is found along chromosome cores during prophase I, accumulating as early as zygonema and persisting through until late pachynema.
To explore the role of BTBD12 in mammalian gametogenesis, we obtained the Btbd12βGeoFlox/βGeoFlox mutant mouse line from the European Conditional Mouse Mutagenesis program (EUCOMM). The genetic disruption at the Btbd12 allele results in residual protein that appears on western blots and, to a lesser extent, on immunohistochemical sections. The detected protein could reflect either a truncated fusion protein consisting of BTBD12 and βGeo, or intact BTBD12 protein generated by a splicing event that removes the βGeo cassette, retaining separate and identiable β-galactosidase protein expression. We favor this latter possibility because the BTBD12 protein detected in the mutant animals can be detected with antibodies against either the N- or the C-terminus. Further, cDNA analysis reveals the presence of every exon of the mouse gene in reverse transcribed RNA from Btbd12βGeoFlox/βGeoFlox testis (data not shown). The reduced intensity of BTBD12 signal in the Btbd12βGeoFlox/βGeoFlox testis extracts compared to wildtype extracts indicates that the presence of the βGeo cassette dramatically reduces the efficiency of BTBD12 protein production and/or reduces the stability of the protein. Importantly, the residual BTBD12 protein does not localize appropriately to meiotic chromosome cores, or localizes at levels that are undetectable using standard chromosome spreads, leading us to conclude that the functional activity of BTBD12 is abnormal in these mice, at least in the context of prophase I.
Micronucleus formation in Btbd12βGeoFlox/βGeoFlox mice is very much elevated compared to wildtype litter mates, in line with a recent report that first described this mouse line [30]. Crossan et al report that the phenotype of Btbd12βGeoFlox/βGeoFlox mice bears some resemblance to human Fanconi anemia (FA), including blood cell cytopenia and numerous developmental deformaties such as the anophthalmia reported herein. This report is in line with another recent publication demonstrating that human SLX4 mutations are also found in a subset of FA patients [12]. Crossan et al also describe gonadal defects and subfertility in Btbd12βGeoFlox/βGeoFlox mice [30], but did not explore the origins of these phenotypes. They report that the histological appearance of the testes is consistent with a defect in meiosis, but they did not document pre-meiotic defects in these animals. They did, however, allude to similarities between the testicular phenotype of Btbd12βGeoFlox/βGeoFlox males and that of other FA-associated DNA repair proteins with which BTBD12 interacts, including Ercc1, Fancd2, Fancl, and Fanca [59]–. These latter mutations result in spermatogenic phenotypes ranging from spermatogonial proliferation defects, to meiotic defects, to defects in spermatozoa morphology. As will be discussed, results presented herein suggest that the phenotypic defects observed in the testes of Btbd12βGeoFlox/βGeoFlox males may be quite distinct from these other mutations.
Given the dramatic loss of testis weight we observe in the adult Btbd12βGeoFlox/βGeoFlox males, coupled with the severe paucity of cells in many seminiferous tubules as early as the period of the first wave of meiosis (days 13–26 pp), we reasoned that a large proportion of germ cells must be lost prior to entry into meiotic prophase I. Thus, these mice suffer from multiple germ cell defects, one involving spermatogonial proliferation and the other involving meiotic progression. The combined effect is a drop in testis size of 75% and a depletion of epididymal spermatozoa by about 90% of wildtype numbers. Given that we have observed only 3 viable pregnancies from females mated to Btbd12βGeoFlox/βGeoFlox males over a 9-month period, we conclude that these defects result in sub-fertility. Breeding data from Btbd12βGeoFlox/βGeoFlox females shows a more severe phenotype, with no pregnant females after 1 year of breeding. These data are more severe than the original description of these mice by the Crossan et al [30], but it should be noted that their analysis included many more mice over a much longer time period.
As predicted, histological examination of testis sections from e16 onwards revealed a dramatic decrease in the numbers of spermatogonial precursors from late gestation, known as prospermatogonia or gonocytes, within the developing seminiferous cords of Btbd12βGeoFlox/βGeoFlox male pups compared to their wildtype littermates. These data suggest that BTBD12 may function to promote DNA repair mechanisms during early proliferation of the PGC population within the developing gonad. PGCs originate at the posterior end of the primitive streak at e7, numbering around 45 [62]. They then migrate to the genital ridge, during which time they undergo a rapid cell proliferation to achieve a cell number of 3000 by e11.5 [63]. They continue to proliferate within the developing gonad until e13 when the testis/seminiferous cord structures form, trapping the now-mitotically arrested prospermatogonia within them [63]. They remain arrested until just prior to birth when they resume proliferation to provide the full complement of prospermatogonia to the post-natal testis. This late gestational wave of proliferation is also associated with increased germ cell apoptosis, even in testes from wildtype males. While the molecular pathways responsible for maintenance of genome integrity during this period are largely un-documented, it is likely that surveillance mechanisms exist similar to those found in somatic cells. Indeed, analysis of Atm mutant males reveals a requirement for ATM function in pre-meiotic spermatogonia [64]. Given the severe consequences of genomic instability within the PGC population for propagation of genetic mutations to offspring, it is plausible that particularly stringent DNA repair mechanisms must exist in these cells. Our results demonstrate that the PGCs arriving in the embryonic gonad appear normal in distribution and number since the testis of e16 mutants is similar in cellularity to that of wildtype littermates. From this time onwards, however, the numbers of prospermatogonia begin to decline rapidly in the testes of Btbd12βGeoFlox/βGeoFlox males such that the final number of these germ cells is dramatically lower in mutant males at e18 and at day 3 pp compared to wildtype males. This wave of apoptosis appears to coincide with the wave of proliferation that occurs around the time of birth because a large number of mitotic figures are observed in the testes of both wildtype and mutant males at this time (Figure 4, arrowheads).
Crossan et al have suggested that the germ cell defects present in Btbd12βGeoFlox/βGeoFlox mice may be similar to that of mutant mice for other key DNA repair genes, particularly those of various FA complementation groups [30]. Indeed, there are substantial similarities between these phenotypes described herein and those for proteins that have been shown to interact with BTBD12. For example, a targeted nullizygous mutation of Fancl also results in limited PGC populations in the embryonic gonad, similar to that seen in Btbd12βGeoFlox/βGeoFlox male embryos [65]. Interestingly, however, the residual PGCs can repopulate the testis gradually in the postnatal mouse such that, by 12 weeks of age, fertility is restored [65]. Thus, while the initial PGC defect may be similar in Btbd12βGeoFlox/βGeoFlox and Fancl−/− male embryos [65], the outcomes in terms of adult fertility are very distinct. In contrast to the age-related increase in fertility in Fancl nullizygous mice, and also to Fanca nullizygous mice, which show declining fertility with age [60], we see limited fertility in the Btbd12βGeoFlox/βGeoFlox males from 7 weeks of age onwards, with no subsequent restoration. Thus, it appears that mouse mutants for the FA complementation groups, while all showing similar anemia phenotypes, represent a spectrum of defects with respect to germ cell migration, proliferation, and differentiation. These differences underscore the importance of this family in genome stabilization within the germ line at all points in their development.
The pre-meiotic phenotype we observe in Btbd12βGeoFlox/βGeoFlox males is also temporally similar to that seen in Ercc1 nullizygous animals [66] in that both mutations result in significant loss of germ cells prior to entry into meiosis. Indeed, the testicular phenotype of Ercc1−/− males at day 3 pp is remarkably similar, if not identical, to that presented herein [66]. In the case of Ercc1, however, no restoration of fertility with age has been reported for nullizygous animals and, in fact, the mice have reduced numbers of epididymal spermatozoa throughout reproductive life, all of which show distinct morphological defects [66]. The limited spermatozoa observed in Btbd12βGeoFlox/βGeoFlox males appear to be morphologically normal and capable of fertilization, and could be a result of the fact that these mice retain residual BTBD12 protein.
Beyond spermatogonial stages, we present evidence of a role for BTBD12 in prophase I progression in the mouse, congruent with the localization of BTBD12 protein on synapsed meiotic chromosome cores. The early localization of BTBD12 at these sites implies an early function in recombination events. Accordingly, in the Btbd12βGeoFlox/βGeoFlox animals, there is persistence in the signal for γH2AX across the autosomes beyond pachynema in the absence of BTBD12. While some 10% of cells show synapsis errors in Btbd12βGeoFlox/βGeoFlox spermatocytes, synapsis is largely unaffected in these mutants, which is in contrast to that seen for FancD2 mutants in which spermatocytes show high levels of asynapsis in a subset of spermatocytes [59]. These data suggest a failure to repair DSBs in a timely fashion, leading to the persistent γH2AX and RAD51 observed at pachynema. However, we cannot rule out the possibility that the additional γH2AX signal is associated with persistent DNA damage arising during pre-meiotic replication events. Indeed, we did observe, by TUNEL labeling, a small fraction of spermatogonia undergoing apoptosis prior to meiotic entry, and so it is possible that if excessive DNA damage persists in the absence of BTBD12, then a proportion of these cells could avoid apoptosis and enter prophase I. Such a possibility can only be addressed through the use of prophase I-specific conditional knock-out strategies. Importantly, however, regardless of the timing of DNA damage induction (pre-meiotic or meiotic), these cells do not succumb to the pachytene checkpoint, as would be predicted from other knockout studies of early prophase I genes, including Msh4/Msh5 and Dmc1 [53], [54], [67]. Moreover, we do not see an overt increase in localization of TOPBP1, nor of MSH4 and MSH5 (data not shown), suggesting that the persistent DSBs observed at pachynema arise out of SPO11-induced events in early prophase I, and not as a result of DNA repair errors prior to meiosis.
Studies by the Sekelsky group were the first to indicate a role for vertebrate SLX4 orthologs in meiosis. These authors showed that the Drosophila Slx4 ortholog, MUS312, is essential for ensuring genomic stability and interacts with the MEI-9(XPF)/ERCC1 nuclease to produce meiotic COs [9], [14], [68], [69]. Thus mus312 mutant flies exhibit >90% reduction in meiotic crossovers [14], but this is unrelated to Mus81 events since Drosphila Mus81 does not participate in CO regulation during meiosis [70]. Similarly, Saito et al describe a role for the C. elegans Slx4 ortholog, named HIM-18, in meiotic recombination, since him-18 mutants exhibit reductions in crossing over of the order of 30–50%, depending on the chromosomal context [13]. Importantly, while Mus-81 does not appear to play a critical role during meiosis in worms, double mus-81;him-18 mutants show a more severe reduction in crossing over than him-18 alone, suggesting co-operative roles for MUS-81 and HIM-18 in meiotic recombination in C. elegans. It is interesting to note that a pre-meiotic function was also described for him-18 mutants, analogous to our observations for mouse BTBD12.
The role of BTBD12 in meiotic recombination is unclear at the current time, and it is interesting to note that the yeast ortholog, Slx4, does not appear to function in meiosis [71]. Given that BTBD12 interacts with components of both the Class I and Class II crossover pathways, as well as with BLM, it is well placed to play an important role in integrating CO decisions in mouse germ cells (Figure 8A). The loss of functional BTBD12 on meiotic chromosomes results in an increase in MLH1/MLH3 foci at mid-pachynema (Figure 5Q, 5R), similar to that seen in Mus81 nullizygous males [28]. Also similar to the Mus81 knockout mouse, we observe no increase in chiasmata at diakinesis in the Btbd12βGeoFlox/βGeoFlox males (Figure 6A, 6B). For Mus81 nullizygous mice, we hypothesize that the additional MLH1/MLH3 foci can restore/maintain chiasmata numbers in the absence of a functional Class II pathway (Figure 8C). By contrast, loss of MLH1 or MLH3, the major Class I mediators at pachynema, cannot be compensated for by MUS81 or any other pathway (Figure 8B). Given the similarity of the Btbd12 phenotype, described herein, to that of the Mus81 nullizygous phenotype, our data suggest that BTBD12 may drive recombination intermediates towards Class II events, thereby promoting MUS81-mediated crossing over. In the absence of either MUS81 or BTBD12 (Figure 8D), however, CO numbers are maintained because of a compensatory increase of MLH1/MLH3 foci, possibly suggesting that BTBD12 does not mediate this switch between the two pathways, but can promote Class II pathway choices under certain conditions. In this regard, it is possible that BTBD12 acts in concert with BLM helicase.
Studies in yeast have suggested that the BLM ortholog, Sgs1, acts to limit the formation of aberrant JMs that arise from strand invasion events that involve both ends of the DSB (as just one example), instead producing substrates for the ZMM crossover pathway or instead resulting in a NCO fate, whilst Mus81-Mms4 can process those events that are not efficiently processed by Sgs1 even in wildtype situations [20], [21]. The absence of Sgs1, therefore, results in an overload of substrates for the Mus81 pathway [20], [21].
Given the model that emerges from the yeast data, together with the phenotypic characterization of Btbd12βGeoFlox/βGeoFlox mice described herein, we propose that the function of BTBD12 is to drive events towards Class II COs, possibly in a BLM-dependent fashion. This raises the question of the fate of those aberrant JMs that are not directed towards Class I or NCO pathways by the actions of BLM; those intermediates that, ordinarily, would be the substrates for MUS81 processing but which, in the absence of Class II-promoting BTBD12, still require resolution. Since it is unlikely that these aberrant structures are responsible for the additional MLH1/MLH3 sites that (presumably) maintain the normal chiasmata count, this suggests two important points. Firstly, the additional MLH1/MLH3 sites must be generated through some other mechanism, perhaps taking advantage of the fact that there is already an excess pool of MSH4/MSH5 foci from which to select for subsequent MLH1/MLH3-driven maturation/stabilization (see model Figure 8). Secondly, those BLM-processed JMs that fail to be diverted towards other recombination fates may remain un-repaired into late pachynema, as suggested by the persistent γH2AX and RAD51 on meiotic SCs in Blm conditional knockouts [72] and in Btbd12βGeoFlox/βGeoFlox males, and could well account for the gradual loss of spermatocytes via apoptosis leading up to the first meiotic division. Conflicting with these suggestions are our previous data showing increased chiasmata-like structures in the absence of Blm, that appear to be MLH1/MLH3-independent [72]. In addition, it should be noted that the complex intermediate structures reported for budding yeast have not yet been demonstrated in mammalian meiosis and, as such, our models, by necessity, relies on extrapolation from a number of organisms, most notably budding yeast.
BTBD12 and its orthologs have been shown to interact with a number of key players in the meiotic machinery, including Rad1-Rad10 [73], ERCC-1 [9], [14], [30], XPF (and its ortholog in Drosophila, MEI-9) [7], [9], [10], as well as with MUS81 and SLX1 [8], [10]. Studies by Svendson et al further demonstrated interactions between human BTBD12 and components of the DNA mismatch repair (MMR) family, including MSH2 [8]. Thus, a model emerges by which BTBD12 and its orthologs can mediate DNA repair and/or modification by directing the activities of certain nucleases in a substrate and context-specific manner. While the precise role of BTBD12 in mammalian meiosis remains to be further clarified, therefore, and its interactions with these various partners in mammalian germ cells have yet to be described, it is tempting to speculate on the function of BTBD12 based on localization data presented herein, on the phenotype of these mice, and based on previous studies of BTBD12 orthologs in other species. For example, studies in vitro have revealed distinct cutting activities for mammalian BTBD12 that are specific to its interactions with MUS81-EME1 (flaps and replication forks) and with SLX1 (HJ cleavage) [8], [10], [29]. Interestingly, the cleavage of HJs in vitro by BTBD12-SLX1 occurs in a near-symmetrical fashion, and in a manner similar to that seen for GEN1 [74], and it is plausible that such an activity may well confer at least limited HJ resolvase activity on the BTBD12-SLX1 complex in mammalian cells. This is in contrast to that seen for yeast Slx4, which cleaves HJ structures asymmetrically [2], [75] thereby reducing the likelihood of an in vivo resolvase role for the yeast protein. This difference in the cutting symmetry between the yeast and mammalian orthologs may explain why slx4 mutants in yeast have no meiotic phenotype, while we observe a distinct meiotic phenotype in the Btbd12βGeoFlox/βGeoFlox mice.
BTBD12 was first identified as a candidate target of the ATM kinase [11] and, in line with this, Slx4 is a substrate of the yeast ATM/ATR orthologs, Mec1 and Tel1 [5]. The function of Slx4 in replication fork repair is dependent on this phosphorylation [55]. In line with this, our western blot analysis reveals a depletion of BTBD12 protein within the testis of adult and juvenile Atm−/− males compared to that seen in wildtype litter mates. BTBD12 protein is also lost from the chromosome cores of Atm−/− spermatocytes, while pre-meiotic spermatogonial proliferation appears to require both BTBD12 and ATM [64]. Thus, while it remains to be seen how ATM and BTBD12 co-ordinate their pre-meiotic functions in spermatogonia, we conclude that ATM activity is essential for the normal loading of BTBD12 onto meiotic chromosomes during prophase I and/or for the stabilization of BTBD12 at these sites. Interestingly, the increased loading of MLH1/MLH3 observed in Btbd12βGeoFlox/βGeoFlox spermatocytes mirrors the phenotype seen in Spo11+/−Atm−/− animals, the Spo11 heterozygosity in this case rescuing the zygotene loss of spermatocytes in Atm single nulls and allowing progression to metaphase [31], [32], again pointing to a functional interaction between the ATM and BTBD12. Spo11+/−Atm−/− spermatocytes also show defects in the dynamics of sex chromosome synapsis and in the formation of the obligate crossover on the pseudoautosomal region (PAR) of the XY [31]. This phenotype is not shared with the Btbd12βGeoFlox/βGeoFlox males, suggesting that BTBD12 is not involved in this aspect of ATM function.
The Btbd12βGeoFlox/βGeoFlox strain of mice we have used herein do not, in all likelihood, represent a complete null allele, due to the presence of residual BTBD12 protein, as measured by both western blot and by immunohistochemistry. However, given the failure of this residual protein to accumulate on meiotic chromosome cores, at least to detectable levels, we consider this mouse to represent a significant impairment in BTBD12 function in germ cells. The use of the Btbd12βGeoFlox/βGeoFlox mouse, while not removing all of the endogenous protein, and while not permitting a focus on meiotic events alone, has been fortuitous in this instance, as it has highlighted the primordial germ cell proliferation defect, which otherwise would not have been evident using a meiosis-specific conditional null mouse. Clearly, any future meiosis-specific research should utilize a conditional approach, which would ensure BTBD12 absence only in meiotic cells, and allow much easier characterization of any defects as a result of loss of BTBD12 protein. Such studies are currently ongoing but will, by necessity, be lengthy.
As discussed, BTBD12 has many potential roles in DNA repair in mammals, from its function in somatic cell repair in human cell lines [7], [8], [10], [29], to its roles in pre-meiotic PGC proliferation and meiotic crossover formation demonstrated herein. The ability of BTBD12 to process not only HJs robustly, but other DNA structures, such as 3′ and 5′ flaps, as well [7], [8], [10], [29], indicates it has the potential to have numerous roles in DNA repair, both in mitosis and meiosis. Moreover, the evidence of functional interactions with a myriad of DNA repair proteins such as MUS81, BLM, XPF-ERCC1, MSH2-MSH3, and the Fanconi anemia genes, along with its well established binding partner SLX1, show that BTBD12 may integrate with several DNA repair complexes to effect its HJ (and other) processing abilities [7], [8], [10], [29]. These interactions no doubt contribute to the similarities between phenotypes at discrete stages of spermatogenesis for Btbd12, Fancl, Ercc1, and other mutant mouse models. Thus a clearer understanding of the function of mammalian BTBD12, both in the context of its multiple roles in gamete formation, and in its function in general genome stability/DNA repair will require more detailed knowledge of these key interactions.
All animals used in this work were handled under strict guidelines imposed by Cornell Veterinary staff and by the Institutional Care and Use Committee (IACUC) under an approved protocol.
We obtained a line of Btbd12 mice with a cassette inserted into the Btbd12 gene, containing a β-galactosidase fused with a neomycin resistance gene, flanked by FRT sites, and LoxP sites flanking exon 3 of the Btbd12 gene from the European Conditional Mouse Mutagenesis (EUCOMM) program (EPD0028_7_A08; Btbd12tm1a(EUCOMM)Wtsi). We termed these mice Btbd12βGeoFlox, to indicate that the FRT flanked and LoxP flanked regions are intact. Mice were genotyped using primers Btbd12_F 5′ CACTGAGCCATCTCACCAGC 3′ and Cas_R1 5′ TCGTGGTATCGTTATGCGCC 3′ to amplify the mutant allele (Btbd12βGeoFlox), and Btbd12_F 5′ CACTGAGCCATCTCACCAGC 3′ and Btbd12_R2b 5′ GGAGCCCAGTCTGGGACTCTG 3′ to amplify the wildtype allele (Btbd12+).
The anti Rabbit polyclonal BTBD12 antibodies were against recombinant His-tagged murine BTBD12 peptide comprising amino acid residues 1–350 (“NT”) and 750–1100 (“CT”). For that, the corresponding cDNA fragment was cloned in pET-28 expression vector (Novagen) and recombinant proteins fused to a histidine tag were purified using Ni-NTA resin (Qiagen) following the manufacturer's instructions.
Epidiymides were removed from either Btbd12βGeoFlox/βGeoFlox or Btbd12+/+ adult mice, placed in human tubule fluid (HTF) culture medium containing BSA (Specialty Media, Millipore), ripped open using micro forceps and the contents squeezed into the medium. The spermatozoa were cultured for 20 minutes at 32°C, then a 20 µl aliquot was removed and fixed in 480 µl 10% formalin. The fixed cells were gently mixed then intact spermatozoa counted using a hemocytometer.
Testes were removed from pre-pubertal or adult mice and fixed either in Bouins fixative or 10% buffered formalin for 2–12 hours. Paraffin-embedded tissue was sectioned at 4 µm and processed for Hematoxylin and Eosin staining or immunohistochemical analyses using standard methods.
Testes were removed from adult Btbd12βGeoFlox/βGeoFlox or Btbd12+/+ mice for the meiotic spread analysis, as well as for the focus counts, and processed as previously described [26]. Briefly, testes were removed and decapsulated into hypotonic sucrose extraction buffer (HEB, containing 1.7% sucrose) and left on ice for 60 minutes. Tubules were macerated on glass depression slides in a bubble of 0.03% sucrose and added to slides coated in 1% paraformaldehyde. The slides were dried slowly in a humidified chamber for 3 hours and washed in PBS containing Photoflo (Kodak, EMS). Ovaries were removed from day e19 female embryos and incubated in HEB for 20 minutes, before being macerated on a depression slide in 0.03% sucrose and added to a bubble of 1% PFA on a well slide, before drying as above.
Slides were processed as described previously [76] using antibodies generated in this lab [26], generously donated by colleagues and available commercially. Immunohistochemistry was performed on formalin-fixed sections using rat monoclonal hybridoma supernatant against germ cell nuclear antigen-1 (GCNA-1), 10D9G11, for staining of germ cells [77], rabbit anti-BTBD12 antibody, rabbit anti-β-galactosidase or TUNEL staining (Chemicon) to detect cells undergoing apoptosis. γH2AX staining was described as either “normal” or “abnormal”, in both pachytene and diplotene cells from Btbd12βGeoFlox/βGeoFlox or Btbd12+/+ spermatocytes. Abnormal cells were classified by >1 γH2AX focus per homologous chromosome core in pachynema, and 1 or more SC-associated γH2AX focus per nucleus in diplonema.
Oocyte spindles were prepared using a modification of techniques described previously [78], [79] and used subsequently in our laboratory [80]. Briefly, ovaries were removed by puncturing ovaries from unstimulated females at 24–26 days of age, and placed in Waymouth's media (GIBCO, Invitrogen Corporation, Carlsbad, CA) supplemented with 100 units of penicillin (base) and 10 µg of streptomycin (base)/ml, 10% fetal bovine serum, and 0.23 mmol/l sodium pyruvate. Primary oocytes at germinal vesicle (GV) stage were cultured in 20 µl drops of KSOM (Millipore Corporation, Bedford, MA) overlaid with mineral oil (Chemicon, Millipore Corporation, Bedford, MA) and incubated at 37°C in an atmosphere of 5% CO2. After 2.5 hrs in culture, oocytes were transferred to fresh KSOM drops and scored for germinal vesicle break down (GVBD). In order to observe meiotic division at metaphase I, oocytes were cultured in KSOM for 8∼10 h and >18 h, respectively, and fixed in fibrin clots (below). Oocytes were fixed in fibrin clots, according to published methodology [78], prior to staining for β-tubulin (1∶500 mouse monoclonal antibody; Sigma-Aldrich, St. Louis, MO). Tubulin staining was visualized using a FITC-conjugated goat anti-mouse IgG (Jackson ImmunoResearch Laboratories, Inc., West Grove, PA), and counterstaining for DNA was achieved using 4′,6-Diamidino-2-phenylindole (DAPI).
Testis weights, spermatozoa numbers, TUNEL analysis, immunofluorescent focus counts and diakinesis spread counts were all analyzed for statistical significance by using an unpaired t-test using Prism 4.0 software.
Analysis of micronucleus formation in peripheral blood cells was performed as previously described [35]. Briefly, peripheral blood was collected from the retro-orbital sinus, fixed in methanol, and incubated in bicarbonate buffer containing RNase A and anti-CD71: FITC antibody (Biodesign International). After washing and staining with propidium iodide, the cells were analyzed on a FACSCalibur flow cytometer (Becton-Dickinson, San Jose, CA).
|
10.1371/journal.pcbi.1004789 | Quantifying and Mitigating the Effect of Preferential Sampling on Phylodynamic Inference | Phylodynamics seeks to estimate effective population size fluctuations from molecular sequences of individuals sampled from a population of interest. One way to accomplish this task formulates an observed sequence data likelihood exploiting a coalescent model for the sampled individuals’ genealogy and then integrating over all possible genealogies via Monte Carlo or, less efficiently, by conditioning on one genealogy estimated from the sequence data. However, when analyzing sequences sampled serially through time, current methods implicitly assume either that sampling times are fixed deterministically by the data collection protocol or that their distribution does not depend on the size of the population. Through simulation, we first show that, when sampling times do probabilistically depend on effective population size, estimation methods may be systematically biased. To correct for this deficiency, we propose a new model that explicitly accounts for preferential sampling by modeling the sampling times as an inhomogeneous Poisson process dependent on effective population size. We demonstrate that in the presence of preferential sampling our new model not only reduces bias, but also improves estimation precision. Finally, we compare the performance of the currently used phylodynamic methods with our proposed model through clinically-relevant, seasonal human influenza examples.
| Phylodynamics seeks to estimate changes in population size from genetic data sampled from individuals across a particular population. One approach to accomplish this task uses a model called the coalescent, which relates the shape of the individuals’ shared ancestral tree to genetic diversity, which is in turn related to population size. However, when analyzing genetic data sampled at different times, current techniques assume that sampling times are fixed ahead of time or are distributed randomly without any relation to the size of the population. Through simulation, we show that when sampling times are related to population size, a situation referred to as preferential sampling, those estimation methods may be systematically biased. To fix this problem, we propose a new model that explicitly accounts for and models the preferential sampling. We show that in the presence of preferential sampling our new technique not only fixes the bias, but also has improved precision in its population size estimates. We also compare the performance of the old and new techniques on several real-world seasonal human influenza examples.
| Phylodynamics—a set of techniques for estimating population dynamics from genetic data—has proven useful in ecology and epidemiology [1, 2]. Phylodynamics is especially useful in cases where ascertaining population sizes via traditional sampling methods is infeasible; e.g., in infectious disease epidemiology it is impossible to obtain the total number of infected individuals in a large population. Estimating population dynamics from a limited sample of genetic data is possible because changes in population size leave evidence in the molecular sequences of the population. Recently, techniques employing a nonparametric approach to inferring population trajectories have improved upon earlier models in terms of flexibility, accuracy, and precision by, e.g., employing Gaussian Markov random fields [3, 4] and Gaussian processes [5]. However, none of these state-of-the-art methods currently account for randomness in sampling time data, potentially introducing bias in studies where sampling times have a relationship to population dynamics. Through a simulation study we characterize this bias in a demographic scenario with seasonally varying population size. We also extend the state-of-the-art by incorporating a sampling time model into phylodynamic inference, mitigating the bias and improving precision.
Phylodynamic methods use Kingman’s coalescent model that, given a particular effective population size trajectory, defines the density of a genealogy relating the sampled individuals [6]. Effective population size measures genetic diversity present in the population and relates to census population size if certain assumptions are met [7]. Many early coalescent-based phylodynamic methods required strict parametric assumptions about the effective population size trajectory, such as constant through time [8] or exponential growth [9, 10]. A major alternative arose with the advent of nonparametric methods, one of the earliest and most influential being the piecewise constant classical skyline model [11]. This approach greatly increases the number of estimated parameters, leading to noisy effective population size trajectories. A number of algorithms seeking compromise between the relative stability of parametric approaches and the flexibility of nonparametric approaches have been implemented [3, 4, 12]. For a detailed comparison, see [13].
Many successful applications of phylodynamics methodology come from infectious disease epidemiology, where the effective population size is interpreted, albeit with caution, as the effective number of infections [14]. In these epidemiological applications, disease agent DNA or RNA sequences are collected at multiple times. When analyzing such heterochronous data, researchers implicitly assume that sampling times are either fixed or follow a distribution that is functionally independent of the effective population size trajectory. However, it is conceivable that the infectious disease agent DNA samples are collected more frequently when the number of infections is high and less frequently during time periods with few infections. Therefore, the implicit assumption of no relationship between sampling times and population dynamics, made by all state-of-the-art phylodynamic methods, is troublesome, since unrecognized preferential sampling leads to systematic estimation bias, as explored by Diggle et al. [15] in the context of spatial statistics. Furthermore, preferential sampling could be present in the sequence databases, but it could also be introduced accidentally or intentionally by filtering during database queries or data mining.
To test the effect of preferential sampling on phylodynamic inference we first perform a simulation study. We simulate sampling times according to multiple distributions, contrasting distributions functionally dependent on effective population size with a functionally independent distribution. We then simulate genealogies based on the sampling times and perform state-of-the-art phylodynamic analyses, and we find that ignoring preferential sampling can bias effective population size estimation and that the size of the bias depends on the local properties of the effective population size trajectory.
In order to account for preferential sampling, we formulate a new phylodynamic model in which sampling times are generated from an inhomogeneous Poisson process with intensity functionally dependent on effective population size. Our model is similar to the augmented coalescent model of Volz and Frost, who work with a specific parametric model [16]. In contrast, we work within a nonparametric framework by incorporating our Poisson preferential sampling model into a Gaussian process-based Bayesian phylodynamic method [3–5]. Applying our new sampling-aware method to our simulations shows that modeling preferential sampling eliminates the aforementioned bias and can increase precision of the phylodynamic inference. In all of our developments, we assume that the genealogy of the sample is known without error. This assumption allows us to use an integrated nested Laplace approximation (INLA) to make our Bayesian inference computationally efficient [17, 18], which is important for executing our simulation studies.
Finally, we examine the performance of our algorithm on two real-world examples. Rambaut et al.[19] explore the seasonal variation of genetic diversity in the genes that code for several of the most important proteins in the two most common influenza subtypes, H3N2 and H1N1. For the sake of brevity we only analyze the hemagglutinin gene in H3N2. We find evidence of preferential sampling in the dataset, and our sampling-aware method produces a large improvement in precision over the conditional (sampling un-aware) method. Zinder et al.[20] specifically explore the patterns of seasonal migration of genetic diversity of H3N2 influenza across the regions of the world. We examine the regions separately and find differing strengths of preferential sampling, but in all regions our method performs better than the conditional model. In some regions, we see stronger relationships between sampling frequency and population size, most often in regions with the most seasonal variation in incidence.
Consider a sample of individuals from a well-mixed population. Some individuals will share a common ancestor more recently than others. One pair of individuals in particular will have the pairwise most recent common ancestor. Moving backwards in time, we can consider those two individuals to have coalesced, treating the two individuals as one. We can then repeat this process of finding the pairwise most recent common ancestor and coalescing individuals until we reach the most recent common ancestor of the entire sample. If we keep track of the ancestral lineages and coalescences of the individuals, we see the data take the shape of a bifurcating tree, and we refer to this ancestry tree as a genealogy (illustrated in Fig 1).
We refer to the branching points of the genealogy tree as coalescent events. If the samples are all taken simultaneously, we refer to the genealogy as isochronous. Kingman’s original coalescent provided a density for isochronous genealogies with a fixed effective population size [6]. Later extensions to the coalescent allowed for parametric and nonparametric specifications of effective population size trajectories along with heterochronous sampling times. Heterochronous sampling times (also called sampling events) can occur at any time up to the present.
We consider first the case of a fixed, heterochronous genealogy [21]. The coalescent likelihood has sufficient statistics g = { t i } i = 1 n , 0 = t n < t n - 1 < … < t 1, representing the coalescent times, and s = { s i , n i } i = 1 m , 0 = s m < s m - 1 < … < s 1 , ∑ j = 1 m n j = n, representing the sampling times along with the corresponding number of lineages sampled (see Fig 1). We define the number of active lineages at time t as the number of lineages sampled between t and the present, minus the number of coalescent events between t and the present. In Fig 1, this appears as the number of horizontal lines that a vertical line at time t will cross.
We define a partition of (0, t1) with intervals Ii,k for k = 1, …, n. We let I0,k represent the intervals ending with a coalescent event and let Ii,k for i = 1, …, mk represent the mk intervals ending in a sampling event between the (k − 1)th and kth coalescent events (see Intervals in Fig 1). We let C i , k = ( n i , k 2 ), where ni,k is the number of active lineages in the interval Ii,k. Suppose s is fixed, then the coalescent likelihood is
Pr [ g | N e ( t ) , s ] ∝ ∏ k = 2 n C 0 , k N e ( t k - 1 ) exp - ∑ i = 0 m k ∫ I i , k C i , k N e ( t ) d t .
In Bayesian phylodynamic inference, our aim is to explore the posterior distribution of the effective population size trajectory Ne(t), so we employ a Gaussian process prior Pr[Ne(t)∣τ], where Ne(t) = exp[γ(t)], with γ ( t ) ∼ BM ( τ ) following a Brownian motion with precision parameter τ[18]. We assign a Gamma(0.01, 0.01) hyperprior to τ. This results in the posterior Pr[Ne(t), τ∣g]∝Pr[g∣Ne(t)]Pr[Ne(t)∣τ]Pr(τ).
The continuous case as written above involves an infinite-dimensional object—the function Ne(t)—which makes the problem as stated intractable. However, we can approximate the continuous function with a piecewise constant function. We construct a fine, regular grid x = { x j } j = 1 B with grid width w over the interval that supports the genealogy and let γj = log[Ne(xj)]. We construct a piecewise constant approximation N γ ( t ) = ∑ i = 1 B exp ( γ i ) 1 t ∈ [ x i - w / 2 , x i + w / 2 ). The discretized coalescent likelihood becomes
Pr ( g ∣ γ ) ∝ ∏ k = 2 n C 0 , k N γ ( t k - 1 ) exp - ∑ i = 0 m k ∫ I i , k C i , k N γ ( t ) d t , (1)
where γ = (γ1, …, γB) and the integrals are simple to compute over the step function Nγ(t). We discretize the Brownian process prior with an intrinsic random walk prior,
Pr ( γ ∣ τ ) ∝ τ ( n - 1 ) / 2 exp - τ 2 ∑ k = 1 B - 1 ( γ k + 1 - γ k ) 2 .
Finally, the discretized posterior becomes Pr(γ, τ∣g)∝Pr(g∣γ)Pr(γ|τ)Pr(τ).
With the posterior known (up to a proportionality constant), we can proceed with numerical integration techniques such as Markov chain Monte Carlo (MCMC) or INLA—a deterministic algorithm for approximating posterior distributions. We select INLA and name the implementation Bayesian nonparametric phylodynamic reconstruction (BNPR).
In the previous section we made the assumption that we could safely ignore any potential dependence of sampling times s on effective population size Nγ(t) in our calculations. In this section, we relax this assumption. We model sampling times according to an inhomogeneous Poisson process in a fixed sampling window [0, s0], with intensity λ ( t ) = exp ( β 0 ) [ N γ ( t ) ] β 1, i.e. proportional to a power of the effective population size, where β0 and β1 are unknown parameters. The sampling log-likelihood is
log [ Pr ( s ∣ γ , β 0 , β 1 ) ] = C + n β 0 + ∑ i = 1 n β 1 log [ N γ ( s i ) ] - ∫ s m s 0 exp ( β 0 ) [ N γ ( r ) ] β 1 d r .
To illustrate our parameterization, sampling with β1 = 1 would result in collecting genetic sequences with intensity directly proportional to effective population size, while higher β1 values result in more clustered samples. Conversely, β1 = 0 produces a uniform distribution of sampling times, with a Poisson distribution on the number of individuals sampled.
In many datasets, the sampling time data will have low enough resolution (for instance, only recording the date but not time of sampling) that some sampling times will appear to be coincident. Our sampling model is compatible with simultaneous sampling times because the model naturally bins the samples along our earlier discretization. The likelihood is proportional to a product of Poisson mass functions centered at the grid points x.
The genealogy depends on the sampling times, so we condition on s in the likelihood for g. We are treating s as random, so we insert the likelihood term for it as well as independent Normal priors for parameters β0 and β1—specifically βi ∼ N(mean = 0, variance = 1000) for i = 0, 1. We retain the same hyperprior for the precision parameter τ as above. This results in the posterior that accounts for preferential sampling,
Pr ( γ , τ , β ∣ g , s ) ∝ Pr ( g ∣ s , γ ) Pr ( s ∣ γ , β ) Pr ( γ ∣ τ ) Pr ( τ ) Pr ( β ) ,
where Pr(g∣s, γ) is defined by Eq (1), but now we add conditioning on s explicitly. In the case where the density of sampling times s is functionally independent of the vector of log effective population sizes γ, the posterior for g simplifies to the form it had in the previous section, because the likelihood for s becomes a constant in γ. We incorporate our sampling model into an INLA framework similar to BNPR and name the implementation Bayesian nonparametric phylodynamic reconstruction with preferential sampling (BNPR-PS).
Here we present a brief outline of the INLA methodology [17] in the context of our BNPR and BNPR-PS implementations. We first examine BNPR as the simpler model. In the end, we intend to estimate the marginal posteriors of the precision hyperparameter Pr(τ∣g) and the latent points Pr(γi∣g), i = 1, …, B, most often focusing on the posterior medians and the end points of the 95% Bayesian credible intervals. We approximate the marginal of τ with
Pr ^ ( τ ∣ g ) ∝ Pr ( γ , τ , g ) Pr ^ G ( γ ∣ τ , g ) γ = γ * ( τ ) ,
where Pr ^ G ( γ ∣ τ , g ) is the Gaussian approximation generated from a Taylor expansion around γ*(τ), the mode of Pr(γ∣τ, g) for a given τ. We can find γ*(τ) using the Newton-Raphson method.
Next, we need to approximate the distribution of γi conditional on τ. The simplest method of using the Gaussian approximations above can produce errors [17], so we briefly describe the use of nested Laplace approximations. The full implementation details can be found in [17]. We define
Pr ^ L A ( γ i ∣ τ , g ) ∝ Pr ( γ , τ , g ) Pr ^ GG ( γ - i ∣ γ i , τ , g ) γ - i = γ - i * ,
where Pr ^ GG ( γ - i ∣ γ i , τ , g ) is a Gaussian approximation of Pr(γ−i∣γi, τ, g) obtained by a Taylor expansion around γ - i * = E( G ( γ - i ∣ γ i , τ , g ), which is computed using Pr ^ G ( γ ∣ τ , g ). Finally, we normalize and combine the two approximations, then use numerical integration to calculate
Pr ^ ( γ i ∣ g ) = ∫ Pr ^ ( γ i ∣ τ , g ) Pr ^ ( τ ∣ g ) d τ .
The outline for BNPR-PS is very similar. The approximate marginal of the hyperparameters is
Pr ^ ( τ , β ∣ g , s ) ∝ Pr ( γ , τ , β , g , s ) Pr ^ G ( γ ∣ τ , β , g , s ) γ = γ * ( τ , β ) ,
for similarly defined factors. We take advantage of an INLA extension by Martins et al. [22] that allows for multiple likelihoods. The approximate distribution of γi conditional on τ, β becomes
Pr ^ L A ( γ i ∣ τ , β , g , s ) ∝ Pr ( γ , τ , β , g , s ) Pr ^ GG ( γ - i ∣ γ i , τ , β , g , s ) γ - i = γ - i * ,
and the final numerical integration is analogously more complex but still tractable, since we integrate over both τ and β.
We use the R-INLA package [17, 22] to perform the above calculations. We make INLA approximations of BNPR and BNPR-PS posteriors available, along with other phylodynamic tools, in the R package phylodyn which can be found at https://github.com/mdkarcher/phylodyn.
We investigate estimating effective population size in the presence of preferential sampling via simulated data. First, we seek to show where and how the model misspecification resulting from ignoring preferential sampling manifests itself in terms of posterior median and Bayesian credible interval width estimation. Our second goal is to show what we gain by properly modeling preferential sampling.
Our primary set of simulation results use the family of seasonally-varying effective population size functions characterized by
N e , a , o ( t ) = 10 + 90 / ( 1 + exp { a [ 3 - ( t + o (mod 12) ) ] } ) , if t + o (mod 12) ≤ 6 , 10 + 90 / ( 1 + exp { a [ 3 + ( t + o (mod 12) ) - 12 ] } ) , if t + o (mod 12) > 6 . (2)
For all of our experiments, the smoothness parameter a = 2 will be used. This family emulates a cyclical population time series with t in nominal months. The shape is loosely modeled after flu seasons, with o controlling which part of the year t = 0 represents (o = 0, 3, 6 emulates summer, spring, and winter, respectively). We simulate genealogies with varying tip sampling times using two sampling schedules. The uniform schedule distributes n sampling times uniformly throughout a given sampling interval. The proportional schedule distributes sampling times in the sampling interval according to an inhomogeneous Poisson process with intensity proportional to effective population size. The proportionality constant here is tuned to have an expected number of sampling times equal to n.
We explore the properties of our two methods using a Monte Carlo approach. To create a Monte Carlo iteration, we generate our sampling times according to their sampling schedules, then simulate our genealogies using coalescent theory via the rejection sampling method of [5]. Given the genealogy and the samples, we infer the effective population time series, using BNPR and BNPR-PS to approximate grids of marginal posteriors. For each iteration, this gives us approximate estimates of the posterior median and quantiles at each point in the effective population size time series. In Fig 2, we see outputs from BNPR and BNPR-PS on the same example iteration.
Our first set of experiments is aimed at determining the extent of the bias introduced by unaccounted preferential sampling. With r Monte Carlo iterations, we take two approaches to locating model misspecification error—time interval analysis and pointwise analysis. For time interval analyses, we calculate summary statistics for a pre-specified time interval (a, b) and average them over the set of r simulation iterations. For pointwise analyses however, we consider the time series of point estimates from each iteration, and then on a pointwise basis we calculate aggregate point estimates and confidence intervals.
Our time interval summary statistics are mean relative deviation,
MRD = 1 r ∑ i = 1 r 1 b - a ∫ a b | N ^ i γ ( t ) - N γ ( t ) | N γ ( t ) d t , mean relative width of the 95% Bayesian credible intervals,
MRW = 1 r ∑ i = 1 r 1 b - a ∫ a b N ^ i , 0 . 975 γ ( t ) - N ^ i , 0 . 025 γ ( t ) N γ ( t ) d t ,
where Nγ(t) is the discretized true effective population size trajectory, N ^ i γ ( t ) is the estimated posterior median of effective population sizes for iteration i, and N ^ i , q γ ( t ) is the estimated qth posterior quantile for iteration i. We also look at mean envelope, ME, the proportion of grid points where the credible interval contains the true trajectory, averaged over all grid points contained in [a, b] across all Monte Carlo iterations.
For a given grid of time points { t j } j = 0 k, pointwise analysis computes the means of pointwise posterior medians,
mpmedian ( t j ) = 1 r ∑ i = 1 r N ^ i , 0 . 5 γ ( t j ) , for j = 0 , … , k ,
pointwise mean relative errors,
mre ( t j ) = 1 r ∑ i = 1 r N ^ i , 0 . 5 γ ( t j ) - N γ ( t j ) N γ ( t j ) , for j = 0 , … , k ,
and a sequence of mean relative widths of the pointwise Bayesian credible intervals,
mrw ( t j ) = 1 r ∑ i = 1 r N ^ i , 0 . 975 γ ( t j ) - N ^ i , 0 . 025 γ ( t j ) N γ ( t j ) , for j = 0 , … , k .
We choose grid size k = 100, number of simulation iterations r = 512, and expected number of lineages per genealogy n = 500. We choose the sampling interval [0, 48] for all simulations.
Researchers who study measurably evolving populations [27], such as viruses, can inadvertently or purposefully preferentially select sequences in accordance to the changes in size of the population of interest. Failing to account for such an ascertainment bias can compromise the statistical properties of phylodynamic inference. Our simulation study shows that the effect of preferential sampling is particularly severe when the effective population size is decreasing. We propose an extension to the state-of-the-art in Gaussian process-based Bayesian phylodynamic methods, in which we assume that sampling times a priori follow an inhomogeneous Poisson process with intensity proportional to a power of the effective population size. This model extension eliminates the systematic estimation bias resulting from having unrecognized preferential sampling, and also gives us better population size estimates by incorporating sampling times as an additional source of information.
Applied to the real-world examples, our method produces improvements over the state-of-the-art. We see significantly improved precision, as well as more realistic estimation of seasonal variation of influenza diversity. In the presence of weaker preferential sampling, as in some of the regional influenza examples, we note that our method still performs better than the current state-of-the-art, with no loss of performance aside from a slightly longer computation time. In addition, by estimating β1, the effect of population size on the log-intensity of sampling times, we gain the ability to quantify the strength of the preferential sampling relationship in the different regions. Such quantification is scientifically useful in infectious disease phylodynamics, because researchers may want to know whether frequency of sampling times can be used as a proxy for incidence.
One avenue of future exploration is to intentionally guarantee preferential sampling during the sequence data collection phase. For example, if an epidemiological study contains noisy incidence data, we can subsample sequences with intensity proportional to incidence and apply our sampling-aware BNPR-PS model to the resulting sequence data. Such a procedure will indirectly combine sequence and incidence data to estimate the effective number of infections —a nontrivial task for the current methods [28]. We contrast this to the approach of [29], which examined the effect of sampling infectious disease agent sequences in batches at different points in an epidemic’s life-cycle compared to uniform and preferential sampling. They found that during epidemic declines their estimates had the largest mean squared error and benefited most in terms of this metric when samples were collected more frequently during the declines. This is consistent with our results, as we see the most error and widest credible intervals during effective population size declines. However, they did not consider the effect of the relationship between their proposed sampling intensity and population size trajectories on estimation of population dynamics—the primary goal of our work.
Our current implementation of the BNPR-PS model assumes a fixed, known genealogy. However, in practice, genealogies are inferred with inherent uncertainty from sequence data. We have found that point estimates produced by our method on the Regional influenza data are robust to genealogical uncertainty (see Regional influenza section in the Appendix), but a method that jointly estimates both genealogy and effective population is still necessary to properly assign uncertainty to population size estimates. One limitation of our method is that the INLA framework cannot be extended to include inference of genealogies. However, it should be straightforward to incorporate the core of our approach —the sampling times model— into an MCMC sampler that targets the joint posterior distribution of population size trajectory, genealogy of sampled sequences, and other parameters. We intend to implement such an MCMC approach in the software BEAST [24].
The main goal of this manuscript is to point out the danger of ignoring preferential sampling in phylodynamics. Providing a solution to this problem, in the form of BNPR-PS model, remains our secondary goal, but we emphasize that much work is still needed to refine our proposed approach. The main weakness of our new model lies in its rigid parametric form of dependence between effective population size Ne(t) and sequence sampling intensity λ(t). In our negative control simulations we see that BNPR-PS performance suffers, possibly greatly, when this assumption of a fixed relationship between effective population size Ne(t) and sampling intensity λ(t) is violated. Similar results under model misspecification are observed by Volz and Frost in the context of birth-death-sampling models for phylodynamic inference [16, 30].
Sampling times model misspecification is most likely to occur if other variables besides effective population size Ne(t) effect changes in the sampling intensity λ(t). For instance, not accounting for a lag between Ne(t) and λ(t) may cause a severe model misspecification. Similarly, not accounting for increases in sampling intensity on longer time scales due to decreases in the cost of sequencing will bias our BNPR-PS estimation. We plan to address these issues by modeling our sampling intensity λ(t) as a log-linear combination of effective sample size and other covariates:
log [ λ ( t ) ] = β T c ( t ) ,
where c(t)T = (1, Ne(t), c1(t), …, cp(t)) and ci(t), i = 1, …, p are covariates of interest. For example, the cost of genome sequencing over time and lagged population size Ne(t−l) are among prime candidates for covariates to be included into our BNPR-PS model. Another example of a promising time-varying sampling covariate is an indicator of ‘outbreak’ status, allowing for changes in sampling intensity during times of increased epidemiological oversight. We hope to explore these model extensions in our future research.
|
10.1371/journal.pgen.1006777 | Hybridization and polyploidy enable genomic plasticity without sex in the most devastating plant-parasitic nematodes | Root-knot nematodes (genus Meloidogyne) exhibit a diversity of reproductive modes ranging from obligatory sexual to fully asexual reproduction. Intriguingly, the most widespread and devastating species to global agriculture are those that reproduce asexually, without meiosis. To disentangle this surprising parasitic success despite the absence of sex and genetic exchanges, we have sequenced and assembled the genomes of three obligatory ameiotic and asexual Meloidogyne. We have compared them to those of relatives able to perform meiosis and sexual reproduction. We show that the genomes of ameiotic asexual Meloidogyne are large, polyploid and made of duplicated regions with a high within-species average nucleotide divergence of ~8%. Phylogenomic analysis of the genes present in these duplicated regions suggests that they originated from multiple hybridization events and are thus homoeologs. We found that up to 22% of homoeologous gene pairs were under positive selection and these genes covered a wide spectrum of predicted functional categories. To biologically assess functional divergence, we compared expression patterns of homoeologous gene pairs across developmental life stages using an RNAseq approach in the most economically important asexually-reproducing nematode. We showed that >60% of homoeologous gene pairs display diverged expression patterns. These results suggest a substantial functional impact of the genome structure. Contrasting with high within-species nuclear genome divergence, mitochondrial genome divergence between the three ameiotic asexuals was very low, signifying that these putative hybrids share a recent common maternal ancestor. Transposable elements (TE) cover a ~1.7 times higher proportion of the genomes of the ameiotic asexual Meloidogyne compared to the sexual relative and might also participate in their plasticity. The intriguing parasitic success of asexually-reproducing Meloidogyne species could be partly explained by their TE-rich composite genomes, resulting from allopolyploidization events, and promoting plasticity and functional divergence between gene copies in the absence of sex and meiosis.
| Sexual reproduction is evolutionary successful in animals as more than 99% of species reproduce sexually. The few animals that have abandoned sex are usually believed to be short-lived and outcompeted by their sexual relatives. Yet, in the root-knot nematodes, plant pests of global economic importance, an intriguing feature is observed. Species that have abandoned sex cause more damage and have a wider worldwide geographic distribution than their sexual cousins. To understand this puzzling success without sex, we have sequenced and analyzed the genomes of the three most devastating asexual nematodes and compared them to that of a sexual relative. We showed that their genomes are large, duplicated, transposon-rich and have hybrid origins. Due to this polylploid hybrid origin, most of their genes are in several copies with substantial sequence divergence. We detected signs of positive selection between these gene copies and confirmed functional divergence at the expression pattern level. We hypothesize that their peculiar hybrid genome structures provide these nematodes with a potential for adaptation and plasticity and could explain their paradoxical success in the absence of sex.
| Fully asexual reproduction occurs in only ~0.1% of animal lineages, which generally occupy shallow branches in the tree of life [1,2]. Although there are some exceptions [3–5], the majority of asexual lineages of animals seem to be recently derived from sexual lineages, suggesting they are generally short-lived. Asexual animals lack the possibility to combine advantageous alleles from different individuals via sexual recombination and in association with Hill-Robertson effect and linkage between conflicting alleles, selection is assumed to be less efficient [6,7]. Furthermore, Muller's ratchet [8] and Kondrashov’s hatchet [9] models of “clonal decay” predict that they progressively accumulate deleterious mutations. Supporting these models, different studies have demonstrated accelerated accumulation of harmful mutations in asexual lineages [10–14], or (short-term) increased accumulation of transposable elements (TE) in the absence of sex [15,16]. Hence, it is commonly postulated that obligate parthenogenetic animals have evolutionary and adaptive disadvantages compared to their sexual relatives and therefore represent evolutionary dead-ends. Consistent with the geographical parthenogenesis model, parthenogenetic populations of plant and animals are generally present at the edge of the geographical distribution of species, in marginal or anthropologically disturbed environments [17,18]. Their uniparental clonal reproductive mode is supposed to be advantageous for colonizing marginal environments where they escape competition with their sexual relatives. Indeed, parthenogenetic species are frequently found at higher latitudes and altitudes [17].
Root-knot nematodes (genus Meloidogyne) display a variety of reproductive modes ranging from sexual reproduction (amphimixis) to obligate asexual reproduction (apomixis) with intermediates able to alternate between sexual (amphimixis) and asexual (automixis) reproduction [19]. These notorious plant pests have been ranked number one in terms of economic threat to the agriculture among all nematodes [20]. Challenging the view that fully asexual lineages of animals are outcompeted by their sexual relatives, Meloidogyne species that reproduce without meiosis and without sex have a broader host range, a wider and more southern geographical distribution and are more devastating than their sexual relatives [21,22]. Whether some genomic singularities could account for their higher parasitic success despite the absence of sex remains unclear. In 2008, we coordinated the publication of the draft genome sequence of Meloidogyne incognita [23], an obligatory asexual nematode and the draft genome of M. hapla, a facultative sexual, was published the same year [24]. One singularity of the M. incognita genome was the presence of genomic regions in two or more copies that spanned several megabases and had an average nucleotide divergence of ~8% [23]. Such a structure was identified neither in the facultative sexual M. hapla nor in any nematode able to reproduce sexually, so far.
The possible origin of the duplicated and diverged genomic regions observed in M. incognita is still debated. Two main hypotheses for the origin of this duplicated genome structure are that (i) duplicated regions represent former paternal and maternal genomes that diverged and became rearranged after their diploid sexual ancestor became asexual or (ii) they result from interspecific hybridization events [21]. As early as 1983, observation of heterozygous patterns of isozymes had led to the hypothesis that M. incognita might have undergone hybridization [25]. Likewise, based on the presence of multiple divergent ITS nuclear markers within apomictic Meloidogyne despite closely related mitochondrial markers between species, it was suggested that these species had undergone hybridization from a set of closely related females with more diverse paternal lineages [26]. Recently, on the basis of the comparative analysis of the initial M. incognita genome and a draft of the meiotic asexual M. floridensis genome, it was also suggested that M. incognita is of hybrid origin [27].
Regardless of its origin, the potential functional impact conferred by the duplicated genome structure of M. incognita has never been assessed. Furthermore, in the absence of genomes for other apomictic Meloidogyne, it was impossible to state whether such a duplicated genome structure is a specificity of M. incognita or a more general signature of the most economically important root-knot nematodes that, intriguingly, all reproduce asexually.
Here, we aimed at characterizing the genome structures of asexual root-knot nematodes, their most likely origin and the potential consequences at the functional level. We have re-sequenced the genome of M. incognita at much deeper coverage and have sequenced de novo the genomes of M. javanica and M. arenaria, two other apomictic root-knot nematodes of high economic importance. We have assembled these three genomes and validated genome assembly sizes with experimental assays. We confirm that the genomes of M. incognita and of the two other mitotic asexual Meloidogyne are made of duplicated yet diverged and rearranged genome copies. We have annotated the protein-coding genes and TE of the three genomes and performed a comparative genomic analysis, including the genome of the facultative sexual species M. hapla and the meiotic parthenogenetic M. floridensis. We show that the genomes of asexual mitotic Meloidogyne have a higher abundance of TE than M. hapla and any other nematode genome published so far. Using a phylogenomic analysis of the duplicated genomic regions conserved between species, we deciphered the origin and evolutionary history of the peculiar genome architecture of mitotic asexual Meloidogyne. To assess the potential functional outcome of these duplicated regions at a whole-genome scale, we searched and found signs of positive selection between the gene copies defining these genomic blocks. Using RNAseq analysis of different life stages of M. incognita, we show that the majority of gene pairs forming duplicated blocks display diverged expression profiles. Furthermore, gene pairs detected as under positive selection show a significantly higher proportion of diverged expression profiles. Our results show that mitotic asexual Meloidogyne possess duplicated, highly diverged and TE-rich genomes, an ensemble of features unequaled in any other nematode genome so far. We propose that the peculiar genome structures of these nematodes offer potential for adaptive plasticity and might contribute to the paradoxical success of these plant-parasitic animals despite the absence of sex.
We sequenced the genomes of three asexually reproducing Meloidogyne species and the assemblies reached 184, 236 and 258 Mb, for M. incognita (Mi), M. javanica (Mj) and M. arenaria (Ma), respectively (Table 1). These genome assemblies are bigger than any Meloidogyne genome assembly reported so far. To confirm these genome sizes, we measured DNA content via flow cytometry experiments and obtained size estimates of 189 ±15, 297 ±27 and 304 ±9 Mb for Mi, Mj and Ma, respectively. The genome assembly of Mi was in the range of estimated size via flow cytometry whereas Mj and Ma assemblies were smaller by 34–60 Mb. To check whether these differences in sizes could be explained by duplicated or repetitive regions collapsed during genome assembly, we plotted the distribution of read coverage along the genome (S1 Fig). We estimated that 17.1 (Mi), 42.9 (Mj) and 21.6 (Ma) Mb of genome assemblies have a coverage twice higher than the rest of the genome sequence and might represent collapsed duplicated regions. Hence, part of the differences in sizes can be explained by these collapsed regions, as previously observed in the genome of the obligate mitotic rotifer Adineta vaga [5]. The genome assemblies contained 97% (Mi), 96% (Mj) and 95% (Ma) of the 248 Core Eukaryotic Genes (CEG) in complete length [28]. These are the highest scores for a Meloidogyne genome so far, suggesting these assemblies are the most complete available to date. We annotated 45,351 (Mi), 98,578 (Mj) and 103,269 (Ma) genes (including, protein-coding genes, ncRNAs, rRNAs and tRNAs). Protein-coding genes spanned up to 43.7 (24%), 75.2 (32%) and 82.2 (32%) Mb of the Mi, Mj and Ma genomes, respectively. Because genome assemblies of the asexual Meloidogyne were ~3–5 times bigger than the haploid genome size of the facultative sexual M. hapla, we suspected these genomes to be polyploid. As a proxy to estimate ploidy levels, we mapped back all the protein-coding sequences (CDS) to their respective genome assemblies and analyzed the proportion of CDS mapping one locus or several loci in the genomes (Fig 1). In the facultative sexual M. hapla, we observed a peak of CDS mapping one single locus in the genome, indicating no sign of whole genome duplication (WGD). In contrast, in the mitotic asexuals, we observed peaks of CDS mapping 3, 3 to 4 and 4 loci in the genomes of Mi, Mj and Ma, respectively. These CDS mapping multiple loci are consistent with the genome sizes of the asexual Meloidogyne (3-5x bigger than the M. hapla genome) and support their polyploid nature.
Transposable elements (TE) covered 50.0% (Mi), 50.8% (Mj) and 50.8% (Ma) of the genome assemblies. In comparison, only 29.2% of the M. hapla genome was covered by TE, using the same annotation protocol (Table 2). Due to its high fragmentation state, the genome of M. floridensis could not be annotated for TE. On average, 27–30% of the genes of mitotic parthenogenetic species are included within TE, whereas this proportion reaches only 17% in M. hapla (see TE section for more details).
Genome sizes as well as distribution of multi-mapping CDS strongly suggested polyploidy in the asexual Meloidogyne (see above). We used MCScanX [29] to further investigate the duplication relationships of the protein-coding genes in Meloidogyne genomes. MCScanX classifies protein-coding genes as (i) singleton when no duplicates are found in the assembly, (ii) proximal when duplicates are on the same scaffold and separated by 1 to 10 genes, (iii) tandem when duplicates are consecutive, (iv) WGD or segmental when duplicates form collinear blocks with other pairs of duplicated genes and (v) dispersed when the duplicates cannot be assigned to any of the other categories. In the three mitotic asexual Meloidogyne species, 93.0–94.1% of protein-coding genes were estimated to be duplicated whereas only 46.6% were duplicated in the facultative sexual M. hapla and 52.9% in the meiotic parthenogen M. floridensis (Table 3). We noted that the dispersed category was the most frequent in all Meloidogyne genomes. However, this proportion negatively correlated with N50 values in the mitotic Meloidogyne, suggesting that these duplicates might be re-classified in other categories in the future.
Interestingly, 12,445, 5,806 and 15,632 genes were classified in the WGD / segmental category in Mi, Mj and Ma, respectively (Table 3). They formed 933 (Mi), 581 (Mj) and 1,648 (Ma) pairs of segmentally duplicated genome regions. In contrast, there were only 90 genes forming 11 pairs of duplicated regions in M. hapla and only 12 genes forming one pair of regions in M. floridensis (Table 4). Collinear duplicated regions span up to 58.6, 14.8 and 59.0 Mb of Mi, Mj and Ma genomes, corresponding to 31.8%, 6.3% and 23.0% of their respective sizes (Table 4). Average nucleotide divergence between pairs of duplicated regions was 8.4%, 7.5% and 8.2% for Mi, Mj and Ma, respectively, indicating a similar average divergence of ~8%. The distribution of % divergence between duplicated regions presented one single to two almost totally overlapping peaks (S2 Fig). This observation holds for the three apomictic Meloidogyne and suggests the duplication events have occurred in a same time window. The divergence levels were substantially lower in coding regions (4.7, 6.0 and 5.9%) than in non-coding regions (9.7, 9.0 and 9.7% for intergenic and 11, 10.4 and 11.1% for introns) for Mi, Mj and Ma, respectively (Table 4, S3 Fig).
Median rates of synonymous substitutions (Ks) between gene pairs forming duplicated regions were 0.1 for the three apomictic Meloidogyne. Averaged Ks for duplicated pairs of regions were significantly negatively correlated with collinearity (P<10−8 for Mi, Mj and Ma), which we measured as the fraction of collinear genes within a pair of regions (S4 Fig). This indicates that divergence in terms of number of conserved genes between a pair of regions correlates with the nucleotide divergence of the coding sequences.
In M. incognita and M. arenaria, we found 5 instances of duplicated regions on a same scaffold (Fig 2). In M. incognita, this corresponded to 42 collinear genes present in 4 pairs of tandem regions and 1 palindrome, whereas in M. arenaria, we found 29 collinear genes present in 2 pairs of tandem regions and 3 palindromes. If the duplicated regions represent vestiges of homologous chromosomes, such tandem and palindrome structures appear consistent with absence of chromosome pairing and meiosis, such as in the genome of A. vaga [5]. No similar structure was found in the genomes of M. javanica, M. hapla or M. floridensis. Average Ks value of gene pairs forming tandem or palindromic regions were in the range of Ks values measured for gene pairs in the rest of duplicated regions, suggesting they have the same divergence times.
We found that 29–37% of gene copies forming duplicated regions in mitotic Meloidogyne were TE-derived, a proportion comparable to the proportion observed for the rest of the whole gene sets (27–30%). Furthermore, the distribution of Ks for pairs of TE-derived genes is not significantly different from the distribution of the rest of pairs of genes in duplicated regions, according to a Wilcoxon test. Thus, we can rule out the possibility that the observed duplicated regions are the results of TE multiplications.
In plant genomes, following WGD, fractionation biases can be observed. One genomic copy tends to retain more genes and to accumulate less mutations than the other copy [30]. For each pair of duplicated regions in Meloidogyne, we tested whether a bias of retention of ancestral genes could be observed (Methods). We found 24 (Mi), 1 (Mj) and 36 (Ma) cases where one region had significantly (Chi-square test, 5% level) retained more ancestral genes than its counterpart. By comparing all the genes in duplicated regions and the MCScanX classification of gene copies, we also estimated that only 6%, 4% and 5% of ancestral collinear genes had no more copies anywhere in the Mi, Mj and Ma genomes, respectively and had probably been lost after WGD.
To decipher the evolutionary history of the duplicated structure observed in the mitotic Meloidogyne, we conducted a phylogenomic analysis focused on the gene copies forming pairs of duplicated regions. We identified and used a dataset composed of 60 groups of homologous genomic regions defined as follows. The genomic regions must be conserved and contain at least 3 collinear genes in 2 copies in each of the mitotic Meloidogyne vs. one single copy in the amphimictic M. hapla (Fig 3 for an example). These 60 groups of genomic regions encompass 2,202 homologous genes distributed in 222 clusters (Fig 3A for an example, S5 Fig for the 60 conserved and duplicated homologous regions). Within each of the 60 groups of genomic regions, we generated multiple alignments of all the clusters of homologous genes individually. Although duplicated genes within a species are, on average, relatively distant (Ks = 0.1, 5–6% nucleotide divergence), gene copies between species can occasionally be identical. In order to maximize phylogenetic signal, the multiple alignments of each cluster of homologous genes were concatenated in each group of conserved duplicated regions. From the 60 concatenated multiple alignments, we successfully generated 54 maximum-likelihood (ML) phylogenies (6 failed because of short alignments, see Fig 3B for an example tree and S5 Fig for the 54 ML trees of conserved homologous regions).
Only three possible bifurcating monophyletic topologies exist to separate the three mitotic Meloidogyne (Fig 4): (1): (Mi, (Ma; Mj)), (2): (Ma, (Mi; Mj)) or (3): (Mj, (Ma; Mi)). We identified 60 such monophyletic clades and the most frequent corresponded to topology 1, observed 33 times. Topology 2 was observed 15 times and topology 3, 12 times (Fig 4). A total of 20 trees combined two of the three topologies mentioned above and allowed testing whether or not the two duplicated regions present the same evolutionary history across the 3 mitotic Meloidogyne. Among these 20 trees, a majority (13) combined two different topologies for the apomictic Meloidogyne, suggesting that the two regions have different evolutionary histories rather than a common ancestral duplication (Fig 5). The combination of topologies 1 and 2 was observed 7 times (see Fig 3B for an example). The combination of topologies 1 and 3 and the combination of topologies 2 and 3 were each observed 3 times. Only 7 of the 20 trees showed twice the same topology; and in all these cases this was twice topology 1.
Part of the genes forming duplicated regions were present in more than two copies in at least one apomictic species. We identified 387 groups of homologous collinear genes (total of 4,262 genes) forming 3 or 4 duplicated regions in at least one mitotic Meloidogyne and at least 2 duplicated regions in the other mitotic. To decipher the evolutionary history of these additional copies, we counted the number of times the third or fourth copies hold a recent in-paralog (or allele-like position) relative to another copy, vs. the number of times these copies were in a new independent branching position (Fig 6). For genes present in 3 copies in a given genome assembly, the number of allele-like relationship was significantly lower (binomial test, P<10−6) than the number of new phylogenetic position, for all three mitotic species (Table 5). Hence, genes present in 3 copies more frequently formed a new independent branch in the phylogenetic trees than species-specific recent paralogs or allele-like branches. For genes in four copies within a given genome assembly, the number of allelic-like relationship was lower than the number of new positions in all species but the difference was significant (binomial test, P<10−5), for M. arenaria only (Table 5).
Overall, this ensemble of results suggests that the duplicated genome regions have different evolutionary histories and thus probably result from allopolyploidization. These pairs of regions and the corresponding gene pairs can thus be considered as homoeologous [32]. This term refers here to pairs of genes that originated by speciation and were brought back together in the same genome by hybridization.
To reveal the maternal evolutionary history of Meloidogyne species included in our analysis, we performed a phylogeny based on mitochondrial protein-coding genes as well as the 12S and 16S rRNAs (S1 Text). The phylogenetic tree (Fig 7, S6 Fig, S1 Table) returned the following highly-supported topology: (Ma,((Mi,Mf),Mj))). This topology corresponds to topology 2, the second most frequently observed in the analysis of the 60 groups of homoeologous duplicated regions (Fig 4). This suggests that genomic regions displaying topology 2 correspond to the maternal contribution to the nuclear genome. We also measured the average nucleotide divergence of mitochondrial genes between the 3 apomictic Meloidogyne (Mi, Mj and Ma). On average, the inter-species nucleotide divergence was very low (0.17%) and ranged from 0.00 to 0.33%. In contrast, the average nucleotide divergence between the meiotic M. hapla and the three mitotic was 24.50% and ranged from 24.42 to 24.58%. Hence, mitochondrial phylogenetic analysis reveals a high similarity between mitochondrial genomes of the three mitotic species and a substantial distance to their sexual relative M. hapla.
We tested whether gene redundancy due to the duplicated genomic regions might result in a relaxation of selective pressure on the gene copies. We employed two different strategies to detect positive and episodic diversifying selection. One raw approach based on pairwise computation of the ratios of rates of non-synonymous (Ka) vs. synonymous mutations (Ks); and a phylogeny-based statistical approach. We found that 612 (8.8%) (Mi), 698 (22.4%) (Mj) and 2,061 (20.9%) (Ma) homoeologous gene pairs had a Ka / Ks ratio greater than 1, indicating possible positive selection (Fig 8). In a second, phylogeny-based approach, we looked for signs of episodic diversifying selection (EDS) in homoeologous genes shared by the 3 apomictic Meloidogyne genomes. We retrieved all homoeologous genes present in the three apomictic species and M. hapla and in 2 or more copies in at least one apomictic Meloidogyne. We found 1,735 such groups and used them to generate multi-gene alignments and their respective ML midpoint-rooted phylogenies. Using the random effects branch-sites model [33], we found 172 (Mi), 109 (Mj) and 208 (Ma) gene copies showing evidence of EDS (S7 Fig for an example) at the 0.05 confidence level (P_Holm: corrected for 9 tests using the Holm-Bonferroni procedure). Among these genes, 20 (Mi), 21 (Mj) and 47 (Ma) were also found to have Ka / Ks ratios >1.
To assess which functional categories were affected by positive selection or EDS, we examined Pfam domains and gene ontology (GO) terms associated to these genes (S1 Text). Overall, a large variety of Pfam domains and associated GO terms were identified among proteins encoded by genes under positive selection (Ka / Ks >1) or subject to EDS, in the three apomictic species. As many as 363, 304 and 674 distinct Pfam domains, corresponding to 177, 167, and 310 distinct GO terms were found in proteins encoded by genes with Ka / Ks >1 in Mi, Mj and Ma, respectively (S2 Table). Similarly, we identified 123, 78 and 174 distinct Pfam domains, corresponding to 93, 56 and 112 distinct GO terms, in proteins encoded by genes under EDS in Mi, Mj and Ma, respectively (S3 Table). Regardless of the dataset (Ka / Ks or EDS), few Pfam domains and GO terms were common to the 3 apomictic species and the majority of them were related to enzymatic, binding and metabolic activities as well as cell cycle-related and transport functions (S2 and S3 Tables). Mapping raw GO terms to the more generic GO-slim terms, revealed more overlap between the three mitotic species. The vast majority of GO-slim terms were shared by the three species in the EDS as well as in the Ka / Ks datasets (S2 and S3 Tables). Overlap between the EDS and Ka / Ks datasets was also high as 23 of the 28 GO terms shared by the three mitotic in the EDS dataset were also shared by them in the Ka / Ks dataset. These terms were mainly related to diverse enzymatic, catabolic, metabolic and biosynthetic functions. We identified significantly enriched GO and GO-slim terms in the Ka / Ks datasets as compared to the rest of homoeologous gene pairs (S2 Table). However, no significantly enriched GO or GO-slim term was common to the three species. No GO or GO-slim term was found to be enriched at the significance threshold (FDR<0.05) in the proteins under EDS as compared to the rest of homoeologous proteins.
Functional divergence between gene copies can be viewed at different levels, including the biochemical function, the biological process or the expression pattern. Gene copies featuring the same biochemical function but expressed in different tissues or time points can be involved in different biological processes (e.g. development of different organs). To biologically assess whether functional divergence actually occurs between homoeologous gene copies, we analyzed their expression patterns across four developmental life stages (eggs, J2 infective juveniles, J3-J4 larval stages and adult females) of M. incognita using RNAseq (methods). We generated between 60.5 (J3-J4 replicate 2) and 94.2 million (egg replicate 2) 2x75bp paired-end reads across the 12 libraries (4 stages x 3 replicates). After all the cleaning steps, between 11.8 (J3-J4 replicate 2) and 48.5 (egg replicate 2) million clean paired-end reads were mapped to the M. incognita reference genome. The proportion of paired-end reads aligned on the genome varied between 75.0% (female replicate 3) and 96.2% (egg replicate 1). The majority of the read pairs (57.2–76.1%) mapped to a unique position on the genome (Table 6). Overall, a total of 42,705 M. incognita protein-coding genes (or 97.7% of 43,718 in total) had a log10(FPKM+1) >1 in at least one sample and were considered as expressed. After filtering out low-signal values as well as low-complexity genes, a total of 38,870 expressed genes remained, including 6,767 homoeologous gene pairs (7,299 initially). This ensemble of expressed genes was classified into 24 distinct expression clusters (methods). We assessed whether homoeologous gene pairs tended to fall in the same expression cluster or in different expression clusters, indicating functional divergence. We found that 4,326 out of 6,767 (63.9%) expressed homoeologous gene pairs showed signs of diverged expression by being assigned to two different expression clusters (Fig 9). Interestingly, pairs of homoeologous genes showing evidence of positive selection in M. incognita tend to be more often in different expression clusters than those showing no sign of positive selection (74.1% vs 63.9%, p-value <10−7). This ensemble of results biologically confirms that the peculiar allopolyploid genome structure of asexual root-knot nematodes is associated to functional divergence between gene copies.
Although recombination can prevent the accumulation of TE, sexual reproduction can favor transmission of TE between individuals. In parallel, hybridization can initially favor TE multiplication by exposing naïve host genomes to TE uncontrolled by their inactivation machinery (e.g. chromatin modification or small RNAs). Thus, we investigated whether differences in TE abundance could be revealed between sexual and asexual Meloidogyne. With 29.2% of its genome occupied by TE, M. hapla has a relatively high TE abundance compared to other nematodes. Indeed, TE span 16.5% and 22.4% of the genomes of C. elegans and C. briggsae, respectively [34], 18% in Trichinella spiralis [35], 14–15% in Brugia malayi [36] and 22% in Bursaphelenchus xylophilus [37]. We found that TE span 50.0, 50.8 and 50.8% of the genome assemblies of the asexual Mi, Mj and Ma, respectively (Table 2). The genomes of the asexually reproducing Meloidogyne thus appear to be particularly rich in TE and 1.7 times richer than the only sexual Meloidogyne genome available to date. Consistent with this observation, Class I retro-elements are on average 1.5 times more abundant in the asexual species. Within Class I elements, DIRS-like (Dictyostelium intermediate repeat sequence) appear to have undergone a particular expansion in the asexuals as they are on average 5.5 times more abundant than in the sexual species. Class II DNA transposons are 1.9 times more abundant in the three apomictic species than in the M. hapla genome. Although Helitron occupy a comparable proportion in asexuals and in the sexual, all the other categories are more than 2 times more abundant in the asexuals. This includes Maverick-like and TIR (terminal inverted repeats) elements as well as “unclassified” TE that possessed characteristics of Class II elements but could not be further assigned to one family. The rest of the potential TE is in the “other” category, which gathers DNA fragments displaying contradictory features of both Class I and II elements. This category was also more abundant in the asexuals than in the sexual species (~1.8 times). This overall abundance of TE in asexual Meloidogyne has implications at the protein-coding level. While 27–30% of the protein-coding genes of asexual Meloidogyne are totally included within TE, only 17% of M. hapla genes are within TE. Hence, TE abundance partly explains the higher number of genes observed in the asexual Meloidogyne (43,718–102,269 compared to 14,207 in M. hapla).
We tested whether the higher gene numbers observed in asexual Meloidogyne were homogeneously distributed along all protein domain families. We plotted the abundance of protein domains in Mi, Mj and Ma as a function of their abundance in Mh (Fig 10). The abundances of protein domains in Mi, Mj and Ma were all positively correlated to the abundance in Mh (R2 = 0.92, R2 = 0.89 and R2 = 0.87 for Mi, Mj and Ma, respectively). The slopes of the linear regressions were 3.06, 4.49 and 4.80 for Mi, Mj and Ma, respectively, suggesting that most of the protein domains are between 3 and 5 times more abundant in the three asexuals as compared to M. hapla. We compared the abundance of Pfam domains known to be found in TE-related genes and important for their own transposition activity (e.g. reverse transcriptase, integrase, transposase) in the four Meloidogyne species. We found that, on average, these domains were 3.4 to 9.8 times more abundant in asexual Meloidogyne than in M. hapla (S4 Table). For instance, rve (integrase core domain) is present in 205–689 copies in the three asexuals while it is found in only 59 copies in M. hapla. Similarly, the DDE_ 3 (DDE superfamily endonuclease) domain is absent in the M. hapla protein set while it is found in 13–61 copies in the three asexuals. This suggests that expansion of at least some families of TE might be in part responsible for the higher number of protein-coding genes in the asexuals.
Meiotic pairing and segregation require high sequence identity and collinearity between homologous chromosomes. Usually, sequencing the genome of a diploid sexual animal involves performing repeated cycles of inbreeding to obtain lineages virtually homozygous at all loci. Genome assembly then results in collapsing all these virtually identical paternal and maternal variants into one single haploid reference sequence. We actually observed this in M. hapla which was assembled into a ~54 Mb genome [24], similar to previously reported measures of haploid genome sizes (~50 Mb) [38,39], and confirmed by our flow cytometry measures (~60 ± 1.5 Mb, Table 1). Concordance of genome assembly size with experimental measures, associated to the absence of extensive duplications of genomic regions, indicate a canonical sexual diploid genome. The haploid chromosome number of M. hapla is n = 16, similar to the putative ancestral haploid number of chromosomes (n = 18) in Meloidogyne species [19,22,40]. Hence, we can hypothesize that the ancestral haploid genome size for a Meloidogyne is ~55–60 Mb with n = 16–18 chromosomes. The genome assembly sizes of the three mitotic Meloidogyne species we describe here reach ~180 Mb, ~235 Mb and ~260 Mb for Mi, Mj and Ma, respectively. This represents ~3x, ~4x and >4x the expected haploid genome size for a Meloidogyne species. Flow cytometry measures of nuclear DNA content suggest an even larger genome size of up to ~300 Mb for Mj and Ma (Table 1). Hence, the genomes of apomictic Meloidogyne (~180–300 Mb) are 3 (Mi) to 5 times (Ma) bigger than the ancestral haploid genome size for a Meloidogyne species. Furthermore, Pfam domains are on average 3, 4.5 and ~5 times more abundant in the mitotic Mi, Mj and Ma genomes, respectively than in the M. hapla genome (Fig 10). Moreover, alignment of the CDS to the respective Meloidogyne genomes showed that while in the sexual M. hapla, most CDS map one single locus; we observe a peak at 3 matching loci for Mi, a peak between 3 and 4 for Mj and a peak at 4 matching loci for Ma (Fig 1). Finally, we showed that a substantial proportion of gene copies form collinear blocks of duplicated genome regions. Taken together, these results strongly suggest that the genomes of mitotic Meloidogyne are polyploid, with M. incognita being most likely triploid, M. javanica tetraploid and M. arenaria tetra- to pentaploid. Observation of chromosome numbers (supplementary discussion S1) further supports polyploidy of the three asexual Meloidogyne. Similarity between genome assembly sizes and measures of total nuclear DNA content via flow cytometry suggests that most of the former paternal and maternal donor genomes have been separately assembled, probably due to their high within-species divergence. Indeed, the duplicated collinear genome regions span several Mb and thousands of genes in each mitotic Meloidogyne genome and show a similarly high within-species average nucleotide divergence of ~8%, confirming initial analysis of the Mi draft genome[23]. Likewise, the per-site synonymous substitution rate (Ks) of collinear gene pairs that define these duplicated genome regions had a very similar median of 0.1 for all three species. This homogeneity of nucleotide divergence and Ks between pairs of collinear regions for the three mitotic Meloidogyne species suggests that they have duplicated in a same time window and thus separated for a similar amount of time.
Due to multiple synteny breakpoints, no long scaffold could be aligned on its whole length to another long scaffold in any of the three mitotic Meloidogyne. This rearranged chromosomal structure combined with the high average divergence between homoeologous blocks suggest chromosome pairing must be complicated if not impossible. Furthermore, in Mi and Ma, we observed collinear regions present in palindromic or tandem arrangement on a same scaffold. Such structures, similar to the ones observed in the ancient asexual bdelloid rotifer A. vaga [5], appear incompatible with segregation of homologous chromosomes in conventional meiosis. Both the difficulty of pairing homologous chromosomes and the impossibility to separate the genome in two equivalent chromosome sets are consistent with the absence of meiosis and the strict asexual reproduction of these organisms. Palindromes or tandem blocks were not observed in the genome of the meiotic M. hapla, and because this genome presents high contiguity (the highest for a Meloidogyne); this most probably represents true absence. Similar analysis could not be performed for the M. floridensis genome; due to its low contiguity and fragmented nature, (only 12 genes were found in one pair of duplicated regions).
The duplicated genome regions in mitotic Meloidogyne tend to be more similar across different species than they are to their other copies within the same species. Furthermore, when duplicated regions form two or more clades in phylogenomic analysis, these clades more frequently present distinct topologies. Thus, collinear duplicated regions within a species have different origins and evolutionary histories and probably do not originate from common ancestral allelic regions that accumulated mutations separately (i.e. no Meselson-White effect). Contrasting with the high within-species divergence of duplicated blocks in the nuclear genome (avg. divergence ~8%), mitochondrial genes are almost identical in Mi, Ma and Mj (avg. divergence ~0.17%). This confirms previous observations that these three species share virtually identical mtDNA markers [26,41,42] and suggests that Mi, Mj, and Ma share closely related or common maternal ancestors. The mitochondrial genome is expected to accumulate mutations faster than the nuclear genome (e.g. 100–1,000 times more rapidly than the nuclear genome in C. elegans [43–45]). Hence, the divergence time between the nuclear genome copies within a same species is assumed much higher than the divergence time between the different species themselves, based on mitochondrial data. Inter-specific hybridization is the most likely hypothesis that could resolve at the same time the discrepancy between low between-species mitochondrial divergence and high within-species nuclear divergence levels and the observed topologies in the phylogenomic analysis (alternative hypotheses are discussed in S1 Text). We propose that not only Mi but also Mj and Ma most likely originated from multiple hybridization events with a same or closely related maternal donor lineage and different paternal donors. This confirms, at a whole genome scale, a previous formulation of this hypothesis based exactly on the same kind of observed discrepancy between low divergence in mitochondrial markers between species and high divergence in ITS nuclear markers within species of apomictic Meloidogyne [26]. In this regard, the case of salamander in the Ambystoma genus constitute an interesting parallel. Indeed, some unisexual species in this genus are characterized by their absence of meiosis, their various ploidy levels (from 2n to 5n), their hybrid origin and their closely related mitochondrial genomes [46].
In apomictic Meloidogyne, because the mitochondrial divergence is very low, the speciation between Mi, Mj and Ma as well as the hybridization and associated loss of sex must be very recent. The very low proportion of collinear genes lost in duplicated genomic blocks further support recent whole genome duplication events (via hybridization). Indeed, after a whole genome duplication event, regardless whether it involves hybridization or not, most of the redundant gene copies are expected to be lost relatively rapidly. For instance, it has been shown in teleost fish that 70–80% of the genes have been rapidly lost after the latest WGD event [47].
Based on genome sizes, Pfam domain abundance and peaks of genes in 3, 4 or more copies, we estimated that Mi was most likely triploid while Mj and Ma were respectively most likely tetra to penta-ploids (see above). Some collinear regions are conserved in more than 2 copies within all the genomes, which allowed assessing the evolutionary histories of the third and fourth genome copies. For the three asexual Meloidogyne, the third copies of collinear genes were significantly more frequently similar to a cognate gene in another Meloidogyne species than to any of the other two copies found in the same species (Table 5). Thus, the third copies probably derive from a distinct hybridization event. This suggests a two-step hybridization process. First, homoploid hybridization (hybridization between two diploid AA and BB progenitors without associated genome doubling [48,49]) took place and led to a diploid AB hybrid. Then, a second hybridization between an unreduced AB gamete of the homoploid hybrid with a reduced C gamete of another sexual species led to the presence of three distinct copies (ABC) of nuclear genomes within a same species. It should be noted that unreduced gametes are frequently produced by inter-specific hybrids [50]. Although this hypothesis could explain the triploid genome of M. incognita, additional steps are needed to explain the tetra to penta-ploid genomes of M. javanica and M. arenaria. We hypothesize that the triploid bridge pathway described in several polyploid plants [51,52] could explain the transition between triploids (similar to M. incognita) and tetraploids (similar to M. javanica). Indeed, under this hypothesis, triploids can constitute a bridge towards tetraploidy by producing unreduced triploid gametes, that, by fusing with a haploid gamete, would lead to a tetraploid progeny. Those tetraploids would in turn produce diploid gametes that would combine with haploid gametes of other species, creating a new triploid. And this triploid, could in turn serve as an intermediate towards new tetraploid by fusing unreduced triploid gametes with haploid gametes, constituting a “triploid-tetraploid-triploid” circle as suggested in [53]. Finally, fusion of an unreduced triploid gamete with either a reduced gamete from a tetraploid or an unreduced gamete from a diploid, could lead to a pentaploid hybrid similar to M. arenaria.
In this perspective, the plant genus Boechera constitutes a good exemplary system for cases of hybridization at different ploidy levels and ecological success [54]. Indeed, it constitutes a genus in which both homoploid diploid and triploid hybrids are present with also less frequent tetraploids or species of higher ploidy levels. Similarly to the Meloidogyne, the hybrids are apomicts and highly heterozygous.
The whiptail lizards constitute an interesting similar example of animal with fully asexual reproduction. Like the asexual Meloidogyne, these lizards are of hybrid origin. Interestingly, several polylploid lineages were identified and they also present a fixed heterozygosity [55,56].
Following loss of sexuality, it has been hypothesized that TE could invade the genomes because recombination would tend to favor their elimination [57]. Alternatively, it has been suggested that the only asexual animals that survive are those that control TE multiplication in their genomes. Examples supporting these two different hypotheses exist in the literature. In Daphnia arthropods, sexual reproduction seems to be correlated with an initial slower accumulation of TE in genomes whereas, at the long-term sex is associated with higher TE loads [15]. In parasitoid wasps, it has been shown that TE are more abundant in Wolbachia-induced asexual lineages than in sexual lineages [58]. However, whether this is a consequence of sex loss or of Wolbachia infection remains to be clarified. In contrast, in the ancient asexual bdelloid rotifer A. vaga, TE occupy only 3% of the genome and while a high diversity of TE was found, they are generally present at very low copy numbers [5]. This suggests that TE proliferation might be under control in this species. Recently, a comparison of the TE load in five sexual vs. asexual lineages of arthropods showed no evidence for TE accumulation in the asexuals [59]. In Meloidogyne, we found that TE occupy ~50% of the genomes of the three mitotic species while they occupy only 29% of the genome of M. hapla. Although it appears that TE have proliferated in the genomes of the asexual Meloidogyne, this might be a consequence of their hybrid origin rather than of their mode of reproduction. Regardless their origin, this abundance of TE might constitute a potential for genomic plasticity in the absence of sexual recombination. Supporting this hypothesis, some canonical full length TE were previously experimentally identified in these Meloidogyne species [60]. Furthermore, a Tm1 transposon has been identified in apomictic Meloidogyne but no homolog with an intact transposase could be found in the sexually-reproducing relative M. hapla [61]. Interestingly, the Cg-1 gene, whose deletion is associated to resistance-breaking strains of M. javanica, has been identified within one of these Tm1 transposons. Thus, TE possibly have an adaptive impact on these nematodes, including on their host plant range.
Mi, Mj and Ma are exceptionally successful, globally distributed parasites of diverse agricultural crops [62,63]. Intriguingly, their geographical distributions and host ranges are wider than those of their sexual relatives. Furthermore, in controlled condition, they are able to overcome plant resistance within a few generations [22]. In the absence of sex and meiotic recombination to provide genomic plasticity and adaptability, their allopolyploid nature may provide benefits contributing to their parasitic success. First, polyploidy can provide the raw material for neo- and sub-functionalization of duplicated gene copies, resulting in novel genetic variation [64,65]. It has been shown in yeast that ploidy level is correlated to faster adaptation [66]. Also, it has been suggested that polyploidy could mask deleterious recessive alleles [67] and limit their accumulation via gene conversion between homologous regions [5]. Furthermore, allopolyploidy, by combining several genomes in one species, may lead to transgressive phenotypes that surpass those of the parent species via novel genetic combination and heterosis [50,67,68]. Ambystoma salamanders constitute one clear case of transgressive phenotype in animals. Indeed the hybrids between one native and one introduced species are ecologically fitter and more successful than the parental native species as well as other related species in the native environment [69,70].
Here, we tested whether the presence of several divergent genomic copies in a same species, could have functional consequences at the coding level. Hybridization brings together homoeologs chromosomes and therefore orthologous gene copies within an individual. Because the three mitotic Meloidogyne have very close mitochondrial genomes, their speciation was certainly recent. Hence, we can hypothesize that most of the high within-species nucleotide divergence between duplicated genomic regions is due to hybridization rather than long-term divergence. Most likely, the hybrid inherits orthologs that had retained similar function and following functional redundancy, selective pressure on these genes may relax and drive them to different evolutionary trajectories [71,72]. In some cases, the relaxation of selective pressure can allow emergence of new adaptive mutations. We have shown that ~8 to 20% of gene copies coming from the duplicated genomic regions harbor signs of positive selection. A diversity of Pfam domains and associated gene ontology terms were predicted in proteins encoded by positively selected genes. Although many terms and domains were related to enzymatic and other catalytic functions, there was a poor overlap between the three apomictic species, and different domains and functions were specifically enriched in positively selected genes in each species. These observations suggest that the functional consequences of the hybrid genome structure were different in each species.
In the model root-knot nematode M. incognita, we showed that more than 60% of homoeologous gene copies display diverged expression patterns. These gene copies resulting from hybridization have only single-copy equivalents in the sexual relative M. hapla. This biological confirmation of functional divergence suggests that additional genes in asexual root-knot nematodes are not just merely functionally redundant with their single-copy orthologs in the sexual relatives but actually support plasticity and variability. Thus, we can assume that the allopolyploid genome structures of asexual root-knot nematodes provide them with a reservoir of variability and adaptability that could partly compensate the absence of sexual reproduction. Noteworthy, these results are consistent with two other recent studies of hybrid animal genomes (the Atlantic salmon and the frog Xenopus laevis) that both also showed extensive functional divergence at the expression level between homoeologous gene copies [73,74]. Interestingly, we noted that the proportion of expression divergence is significantly higher (>70% vs. >60%, p-value<1.10−7) in homoeologous gene pairs that are under positive selection. These gene pairs combine both divergence at the expression level and accumulation of non-synonymous mutations that could lead to functional divergence at the biochemical level. They are thus the most obvious candidates for neo or sub functionalization.
How an animal can survive without sexual reproduction and compete with its sexual relatives remains an evolutionary puzzle. Intriguingly, asexually reproducing (apomictic) root-knot nematodes outcompete their sexual relatives as plant parasites of global economic impact. We have shown here that the genomes of the apomictic Meloidogyne are duplicated most likely because of a complex series of hybridization events. Although the parental lineages are unknown, they probably belong to the sexual relative clades. Hence, this parasitic success could be viewed as a case of transgressive phenotype, where the ecological success of the hybrid progeny surpasses those of the parents [27,68]. Furthermore, hybridization has been proposed as an important evolutionary phenomenon that could give rise to new parasites and pathogens. For instance, hybridization of two host-specific plant parasitic tephritid fruit flies gave rise to a new species able to parasitize a new invasive host plant [75]. Similarly, hybridization of two Blumeria fungal pathogens gave rise to a new species that is able to infest a host plant of economic interest resistant to both progenitor species [76]. In the asexual root-knot nematodes, the presence of duplicated and diverged genomic regions probably promotes functional novelty between resulting gene copies, following positive selection. We confirmed this functional divergence at the expression level at the whole genome scale. Furthermore, the TE-rich nature of their genomes might also foster genomic plasticity not only actively by TE movements across the genomes but also passively by promoting chromosomal shuffling between these repeated genomic regions. Such a TE-promoted chromosomal shuffling associated to adaptation to different host plants has already been shown in a plant-pathogenic fungus [77]. Part of the intriguing success of mitotic asexual Meloidogyne could thus reside in their duplicated, diverged and TE-rich genomes resulting from hybridization. It would be interesting to explore the potential for plasticity and adaptation in the genomes of other asexual animals, particularly parasites and pathogens, to assess whether convergent or independent genomic characteristics support this potential.
DNA samples preparation protocols for genome sequencing are detailed in the S1 Text. Genome assemblies were performed in four steps, following the same procedure as developed for resolving the degenerate tetraploid genome structure of the bdelloid rotifer A. vaga [5]: (i) assembly of 454 data into contigs, (ii) correction of the 454 contigs using Illumina data, (iii) scaffolding of 454 contigs and (iv) gap closing using Illumina data. For the first step, we used the multi-pass assembler MIRA [78] version 3.9.4 (normal mode, default options except the number of cycles) to generate contigs from the 454 genomic libraries (S5 and S6 Tables). Although computationally demanding, running MIRA with multiple cycles is particularly appropriate to separate heterozygous regions in genomes, as anticipated in polyploid species. Moreover, Sanger reads of the M. incognita first draft genome sequence [23] were used to generate the current assembly. Twelve (M. arenaria and M. javanica) or sixteen (M. incognita) cycles were performed to separate a maximum of repeats and heterozygous regions. We subsequently used Illumina data to correct the homopolymer errors of the 454 contigs following a standard procedure [79]. The corrected contigs were linked into scaffolds using the program SSPACE [80] with 454, Sanger and Illumina data. Finally, assemblies were gap-closed using GapCloser from the SOAPdenovo 2 package [81] with Illumina data. The statistics of the three genome assemblies are summarized in S5 Table. We assessed the completeness of the three genome assemblies by counting the number of Core Eukaryotic Gene (CEG) using CEGMA [82].
Flow cytometry was used to perform accurate measurement of nuclear DNA content in the three apomictic Meloidogyne (M. incognita, M. javanica and M. arenaria) as well as in the facultative sexual M. hapla, compared to internal standards with known genome sizes. Caenorhabditis elegans strain Bristol N2 (approximately 200 Mb at diploid state [83,84]) and Drosophila melanogaster strain Cantonese S. (approximately 350 Mb at diploid state [85,86]) were used as internal standards. Extraction of nuclei was performed as previously described [87]. Briefly, for each Meloidogyne species about two hundred thousand stage 2 juveniles (J2s) were suspended in 2 mL lysis buffer (1mM KCl, 30 mM NaCl, 10 mM MgCl2, 0.2 mM EDTA, 30 mM Tris, 300 mM sucrose, 5 mM sodium butyrate, 0.1 mM PMSF, 0.5 mM DTT, 40 μl Igepal), grinded for 10 min with a Dounce homogenizer and filtered through a 0.20 μm nylon mesh. Subsequently, this 2 mL suspension was overlaid on top of 8 mL suspension buffer (same as lysis buffer except for sucrose, 1.2 M, and without Igepal) so that the tubes were ready for centrifugation (10,000 rpm, 30 min, 4°C) to reduce the level of debris and to pellet nuclei. Supernatant was completely discarded and pelleted nuclei were re-suspended in suspension buffer. Then nuclei suspension was stained, at 37°C for 30 min, with 75 μg/mL propidium iodide and 50 μg/mL DNAse-free RNAse. The same nuclei extraction protocol was performed at the same time on the samples and on the two internal standards. Flow cytometry analyses were carried out using a LSRII / Fortessa (BD Biosciences) flow cytometer operated with FACSDiva v6.1.3 (BD Biosciences) software. Data were analyzed with Kaluza v1.2 software (Beckman Coulter) and cytograms exhibiting peaks for each phase of the cell cycle (G0/G1, S and G2/M) were obtained. Standards and samples were processed both alone and together. Only mean fluorescence intensity of the first peak (arbitrary units), corresponding to G0/G1 phase of the cell cycle of the cytograms, was taken into account to estimate DNA content. In this method [88,89], the amounts of DNA in the Meloidogyne samples were determined by interpolating the fluorescence signals generated from the standards using the following equation:
Meloidogyne DNA content (Mb)=(G0/G1Meloidogyne samplex Standard DNA content)/G0/G1standard.
The estimated DNA contents of the Meloidogyne samples were calculated by averaging the values obtained from three biological replicates (S8 Fig).
To check whether some nearly identical duplicated genomic regions had been collapsed during the assembly (as previously observed in the A. vaga genome [5]), we aligned the Illumina PE-reads of each species against their respective genome assembly sequence, using BWA [90] with default parameters. We computed the per base read coverage using BEDtools genomeCoverageBed [91] and plotted the distribution of the per-base coverage depth. This clearly showed 2 peaks for the three species, one systematically at twice the coverage of the first peak (S1 Fig). We calculated the number of bases with per base coverage comprised in the range of the second peak and summed it up to obtain the total size of the duplicated regions that had been collapsed during the assembly.
Predictions of protein-coding genes were performed using EuGene 4.1c [92], optimized and tested for M. incognita on a dataset of 301 non-redundant full-length cDNAs. Translation starts and splice sites were predicted using SpliceMachine [93]. Three datasets of M. incognita transcribed sequences were provided to EuGene to contribute to the prediction of gene models: i) Sanger ESTs (Genbank 20110419), ii) a dataset of seven Illumina transcriptomes obtained in our lab in a previous study [94], and iii) a dataset of nine Trinity [95] assemblies of RNAseq data, generated in this study (S7 Table, S1 Text). Transcribed sequences were aligned on the genome using GMAP [96]; spliced alignments spanning 80% of the transcript sequence length at a 90% identity cut-off were retained. Similarities to i) C. elegans release Wormpep221, ii) G. pallida, release 1.0 [97], and iii) Swiss-Prot release December 2013 (excluding proteins similar to REPBASE [98]) were searched using BLAST [99] and provided to EuGene to contribute to gene modelling. The gene modelling algorithm used the standard parameters for the 4.1c version, except for the fact that i) the gene finding algorithm was applied on both strands independently allowing overlapping gene models, ii) non-canonical GC/donor and AC/acceptor sites were allowed on the basis of transcriptional evidences, iii) a gene model was not allowed to span a gap (‘N’) longer than 1,000 nucleotides, iv) the minimum length of introns was set to 35 nucleotides, and v) the minimum CDS length cut-off was set to 150 nucleotides. For M. arenaria and M. javanica, the EuGene pipeline, with models and parameters tuned on M. incognita, was used to annotate both genomes. Two modifications were applied on the selection of reference datasets i) Swiss-Prot (excluding proteins similar to REPBASE) and the proteome of M. incognita were used as reference proteomes ii) assemblies of M. arenaria and M. javanica RNAseq data were used as sources of transcription evidences.
We annotated ncRNAs using RNAmmer [100], tRNAscan-SE [101], Rfam release 11 [102], and in house scripts to remove redundancy and consolidate results.
The predicted protein sequences of M. incognita, M. javanica, M. arenaria and M. hapla were scanned for the presence of Pfam protein domains using the program PfamScan [103] against the Pfam-A HMM domain library (release 27.0), using default thresholds and parameters. A gene ontology annotation was inferred from the Pfam protein domain annotation using the pfam2go mapping file maintained at the gene ontology portal and generated from the InterPro2GO mapping [104]. Gene ontology terms were also mapped on the generic GO-slim ontology using the GOSlimViewer utility developed as part of AgBase [105].
The duplicated structures of Meloidogyne species were estimated by detecting conserved blocks of duplicated genes. The protein sequences of each genome were initially self-blasted to determine a homologous relationship with an e-value threshold of 1e-10. Conserved blocks of duplicated genes were detected based on the gene locations in the genome using MCScanX [29] with default parameters. We required at least 3 collinear genes pairs for MCScanX to form a block. Using the perl script “add_ka_and_ks_to_colinearity.pl” included in the MCScanX package, we calculated Ks values for each homologous gene pairs between duplicated blocks. The median Ks value was considered a representative of the divergence between duplicated regions.
We used a custom python script (S2 Text) to compute the pairwise nucleotide identity between collinear blocks for each species. Briefly, pairs of duplicated genomic regions were extracted according to the GFF3 positions of their first and last collinear genes. They were then aligned using NUCmer from MUMmer v3.23[106] with default parameters. We then filtered out sub-alignment shorter than 50 nt (delta-filter -l 50) and summarized alignment using the dnadiff program from the MUMmer package. The average identity at the nucleotide level between duplicated regions was obtained from the output of dnadiff. Identity within coding and non-coding sequences was obtained by masking coding or non-coding sequences in each duplicated region before NUCmer alignment using BEDtools maskFastaFromBed v2.17.0 [91].
To analyze synteny conservation between genomes, we concatenated all the inter-/intra-species BLAST hits (e-value threshold of 1e-10) of M. incognita, M. javanica and M. arenaria and M. hapla protein sequences and fed MCScanX with this pooled BLAST result as well as with information on the location of the corresponding genes in the respective genomes, as recommended in the MCScanX manual for multi-species comparisons. The M. floridensis genome had to be discarded from this comparative analysis because only one pair of regions composed of 12 genes block was detected for this genome preventing any large-scale analysis of conserved synteny. We required at least 3 collinear genes pairs for MCScanX to detect a block. We parsed the results of the collinearity analysis between genomes of Meloidogyne species (HTML files output by MCScanX) to extract collinear genes forming duplicated regions conserved between Meloidogyne species. We used those homologous collinear genes to perform phylogenomics analyses (see below).
For each pair of duplicated regions, the genes present on each region were counted and the number of genes that were present in the ancestor of these two regions was calculated as the total number of genes on the two collinear regions minus the number of gene pairs. We then compared the number of genes in each region to the number of estimated genes in the ancestral region to statistically determine whether one region had retained significantly more genes than the other within a pair.
First, protein sequences were aligned using MUSCLE v3.8.31 [107,108]. Second, protein alignments were back translated into codon alignment using PAL2NAL v.14 [109] with the ‘nogap’ option. Third, codon alignments were trimmed using GBLOCKS [110] with default options. The fittest model of nucleotide evolution was searched using the function ModelTest as implemented in the R package phangorn [111]. We then used PhyML [112] (-d nt -b -4 -m GTR -f e -t e -v e -a e -s BEST) to build maximum likelihood phylogenies with SH-like branches support on these pruned alignments. We rooted the phylogenies using the midpoint function of R package phangorn. For the rest of the analyses, we only retained the trees in which M. hapla displayed an outgroup position relative to the other Meloidogyne species in the midpoint-rooted topologies. Tree topologies were classified and counted using a custom R script. Phylogenetic tree figures were formatted and edited using EvolView [113].
To compute the Ka / Ks ratios per pair of collinear homologous genes, we used the Nei-Gojobori method [114] implemented in MCScanX. We eliminated all cases where 0.01<Ks<1 to discard genes that were either evolving extremely slowly or extremely rapidly and could potentially yield erroneous Ka/Ks estimates. We performed tests of episodic diversifying selection (EDS, a form of positive selection) using the random effects branch-sites model [33] implemented in the HYPHY package [115]. We looped the branchSiteREL.bf script over the 1,735 multi-sequence alignments and their respective ML midpoint rooted trees. Each alignment contained at least one collinear protein-coding gene for all three apomictic species and M. hapla and a duplicate in at least one asexual species. We chose the adaptive version of BSRE and allowed branch-site variation in synonymous rates. Branch with length less than 0.01 were not considered because ω rate classes cannot be inferred reliably for very small branches (< 0.01).
Total RNAs were extracted from 4 M. incognita developmental stages (pre-parasitic J2s, parasitic J3-J4, adult females and eggs) using TRIzol Reagent (Invitrogen); three independent biological replicates were performed for each stage. Total RNA quality and quantity were assessed by a 2100 Bioanalyser (Agilent technologies). Samples with RNA integrity number (RIN) over 8.5 were kept for cDNA library construction, except for eggs samples for which RIN ranged between 6 and 7. An input of 100 ng total RNA was provided to construct cDNA libraries via the Ovation Universal RNAseq system (Nugen technologies). To eliminate unwanted rRNA transcripts, we designed 101 InDA-C primers to target M. incognita 28S and 18S transcripts for depletion. The 12 cDNA libraries (4 stages x 3 replicates) were quantified and equilibrated to 4 nM using Kapa QPCR (Kapa Biosystems). Finally, multiplexed libraries were sequenced on an Illumina NextSeq 500 sequencer on two High 150 flow cells (400M PE75 reads), on the UCA Genomix sequencing platform of Nice Sophia-Antipolis.
The quality of the raw read fastq files were manually checked using FastQC and the following series of filters were applied to all the files. We first eliminated possible remaining ribosomal RNA contamination using SortMeRNA [116]. We then used PRINSEQ [117] to trim sequence ends with quality score lower than 28, and only kept reads with an overall score >28 and a length >60 nucleotides. We aligned the cleaned reads to the M. incognita indexed genome assembly using the STAR 2-pass procedure [118].
We used RSEM [119] to estimate read counts, transcripts per million (TPM) as well as fragments per kilobase per million mapped reads (FPKM) for the M. incognita predicted protein-coding genes. RSEM takes into account multi-mapped reads and assigns them proportionally to the different loci according to probabilities estimated based on uniquely mapped reads.
We transformed the raw FPKM values in log10 (FPKM+1) values. To avoid the risk of spurious read counts due to low complexity regions in transcripts, we eliminated all the transcripts that had more than 1/3 of their length covered by low-complexity regions, as measured by RepeatMasker [120]. We also filtered out genes showing too much variability between replicates and showing an inside-replicate coefficient of variation of log10(FPKM+1) higher than 0.8. We finally averaged expression over each triplicate, and filtered out genes with log10(FPKM+1) mean expression values lower than 0.3 in all conditions, as this corresponded to low signal.
We then clustered genes according to their expression pattern in the four conditions. To be as robust and conservative as possible, we clustered together genes showing the same relative expression patterns in the four conditions, i.e. genes whose expression values are ranked in the same order between the four conditions, resulting in 24 (= 4!) groups. Homoeologous genes from a same pair that were located in two different gene expression clusters were considered as having diverged expression patterns.
|
10.1371/journal.ppat.1007795 | An NLRP3 inflammasome-triggered cytokine storm contributes to Streptococcal toxic shock-like syndrome (STSLS) | Infection with the Streptococcus suis (S. suis) epidemic strain can cause Streptococcal toxic shock-like syndrome (STSLS), which is characterized by a cytokine storm, dysfunction of multiple organs and a high incidence of mortality despite adequate treatment. Despite some progress concerning the contribution of the inflammatory response to STSLS, the precise mechanism underlying STSLS development remains elusive. Here, we use a murine model to demonstrate that caspase-1 activity is critical for STSLS development. Furthermore, we show that inflammasome activation by S. suis is mainly dependent on NLRP3 but not on NLRP1, AIM2 or NLRC4. The important role of NLRP3 activation in STSLS is further confirmed in vivo with the NLRP3 inhibitor MCC950 and nlrp3-knockout mice. By comparison of WT strain with isogenic strains with mutation of various virulence genes for inflammasome activation, Suilysin is essential for inflammasome activation, which is dependent on the membrane perforation activity to cause cytosolic K+ efflux. Moreover, the mutant strain msly (P353L) expressing mutagenic SLY without hemolytic activity was unable to activate the inflammasome and does not cause STSLS. In summary, we demonstrate that the high membrane perforation activity of the epidemic strain induces a high level of NLRP3 inflammasome activation, which is essential for the development of the cytokine storm and multi-organ dysfunction in STSLS and suggests NLRP3 inflammasome as an attractive target for the treatment of STSLS.
| The two large-scale human Streptococcus suis epidemics have caused unusual development of Streptococcal Toxic-Shock-like Syndrome (STSLS) and high incidence of mortality despite adequate treatments. However, how the epidemic strain causes STSLS remained to be elucidated. Because an excessive high level of inflammasome-regulated cytokine was detected in the blood of STSLS patients, we used a murine model to identify the role of inflammasome activation on the development of STSLS. We found that NLRP3 activation contributed to STSLS with the pharmacological inhibition and NLRP3-/- mice. We identified a novel mechanism of STSLS in that increased suilysin expression in S. suis highly virulent strain could induce high level of cytosolic K+ efflux, an essential event for NLRP3 inflammasome activation, and then further cause a cytokine storm, dysfunction of multiple organs and a high incidence of mortality, the characters of STSLS. Therefore, our study provides insights for STSLS development and highlights NLRP3 inflammasome as an attractive target for the treatment of STSLS.
| Streptococcus suis (S. suis) is a major swine pathogen that is responsible for severe economic losses in the porcine industry and represents a significant threat to human health [1–4]. To date, more than 1600 human S. suis infections have been reported worldwide [4, 5], and the infection has been identified as the leading and second-leading cause of adult meningitis in Vietnam and Thailand [2]. S. suis infection mainly induces meningitis, sepsis, arthritis, endocarditis, and endophthalmitis, and the pooled case-fatality rate is 12.8% [5]. However, two large-scale human S. suis epidemics in China (the first was 25 cases with 14 deaths in Jiangsu in 1998, and the second was 204 cases with 38 deaths in Sichuan in 2005) raised serious concerns for global public health and challenged the conventional perception that S. suis infections are sporadic in humans [2, 6, 7]. This infection causes unusual development of Streptococcal toxic-shock-like syndrome (STSLS), including the hallmarks of acute high fever, blood spots, hypotension, shock, and dysfunction of multiple organs, as well as acute death (mortality is more than 80% despite adequate treatment) [7, 8].
At present, how the epidemic strain causes STSLS and leads to high mortality remains unclear. A retrospective clinical investigation showed high tumor necrosis factor-alpha (TNF-α), interleukin (IL)-1β, IL-6, IL-8, IL-12, and interferon-γ (IFN-γ) levels in the blood of patients with STSLS [6]. Subsequent studies further confirmed that the induction of an inflammatory cytokine storm was essential for STSLS [9, 10], which was further supported by the finding that inhibition of the excessive inflammatory response with anti-inflammatory drugs improved survival against STSLS [11]. Together, these data highlight the great potential that comprehensive understanding of the molecular mechanisms by which S. suis induces a high level of inflammatory responses may contribute to identify new therapeutic targets for S. suis-caused conditions, including STSLS [11, 12].
IL-1β secretion is tightly controlled by the assembly of a multiprotein complex called the inflammasome [13, 14]. To date, a few types of inflammasomes (NLRP1, NLRP3, NLRC4, AIM2, etc.) have been described, and the NLRP3 inflammasome has been under intense investigation given its link with a vast number of diseases [13, 15, 16]. Upon activation, NLRP3 is recruited to the dispersed trans-Golgi network to form multiple puncta that induces ASC polymerization and makes pro-caspase-1 (pro-casp1) into an active protease [17]. In turn, caspase-1 (casp1) mediates the processing of several targets, including pro-IL-1β and pro-IL-18, into their biologically active forms and induces their secretion by triggering pyroptosis through cleaved gasdermin D (GSDMD) [18–21]. IL-1β and IL-18 secretion may further induce IL-6, IL-8, IL-17, and IFN-γ expression, thereby resulting in inflammatory conditions such as fever and septic shock [22]. Owing to the high levels of blood IL-1β and its inflammatory mediators in patients with STSLS [6], we hypothesized that the inflammasome could contribute to STSLS. Here, we demonstrated for the first time that a high level of inflammasome activation was essential for induction of the cytokine storm and the dysfunction of multiple organs—the hallmarks of STSLS.
STSLS is characterized by high bacterial burden, an inflammatory cytokine storm, multi-organ dysfunction, and ultimately acute host death [6–8]. In a murine model, S. suis epidemic strain SC-19 infection induced an acute and extremely high inflammatory cytokine response, including increased IL-1β, IL-18, TNF-α, IL-17A, and IFN-γ levels (Fig 1A), high bacterial burden (Fig 1B), and high CK (creatine kinase), ALT (alanine aminotransferase), AST (aspartate aminotransaminase), and LDH (lactate dehydrogenase) levels in the blood (Fig 1C), resulting in evident injury in multiple organs, such as severe congestion and dense infiltration of inflammatory cells in the lung, severe congestion in the spleen, and severe vacuolated degeneration and necrosis in the liver (Fig 1D). In addition, all infected mice presented with severe clinical signs and died within two days (n = 10) (Fig 1E and 1F). Moreover, the level of inflammatory response and organ damage caused by SC-19 is much higher than classical virulent P1/7 strain, which could also cause high mortality [10]. Thus, murine infection with SC-19 mimicked the STSLS observed in humans.
To evaluate the effect of the inflammasome on STSLS, an inhibitor (inh) of casp1, Ac-YVAD-CHO, was intraperitoneally injected into the infected mice 1 h after infection. Ac-YVAD-CHO treatment significantly reduced the IL-1β and IL-18 levels (Fig 1A), indicating that the secretion of IL-1β and IL-18 depended mainly on casp1 activity. In contrast, TNF-α production was not significantly inhibited by the treatment, which suggested that inhibition of the inflammasome with Ac-YVAD-CHO could not significantly inhibit the casp1-unrelated pro-inflammatory cytokine response (Fig 1A). IL-17A and IFN-γ induction was also inhibited by Ac-YVAD-CHO (Fig 1A) since these cytokines are reported as downstream effectors of the inflammasome [23–25].
Because the bacterial burden in the blood did not significantly decrease at the given time point (Fig 1B), the decreased inflammatory response was not due to a decreased bacterial load, the trigger for activation of this inflammatory signaling pathway. Furthermore, inhibition of casp1 activity also reduced the levels of CK and AST in the blood (Fig 1C), alleviated inflammation and injury in multiple organs (Fig 1D), reduced clinical signs and promoted survival (Fig 1E and 1F). Ac-YVAD-CHO was not an exclusive inhibitor for casp1, and it also exhibited some activity against caspase-4/5 [26], which directly recognized intracellular LPS for non-canonical inflammasome activation [27–29]. Therefore, these data indicate a potential critical role of casp1-based inflammasome activation in STSLS.
To understand the mechanism underlying STSLS development and to identify the type of inflammasome that is activated in response to S. suis infection, we constructed four types of inflammasome complexes in the 293T cell line (S1 Fig). S. suis could clearly induce cleavage of pro-casp1 and pro-IL-1β and secretion of IL-1β in 293T cells expressing the NLRP3 inflammasome complex but not in cells expressing the other three (NLRP1, NLRC4, or AIM2) inflammasome complexes (Fig 2A). In contrast, poly(dA:dT) mainly activated the AIM2 inflammasome, as described previously [30] (Fig 2A). These results indicated that NLRP3 was required for inflammasome activation in response to S. suis epidemic strain SC-19 infection.
To further confirm whether NLRP3 was indispensable for inflammasome activation induced by S. suis, an nlrp3-deficient human acute monocytic leukemia THP-1 cell line (THP-1-nlrp3-/-) and a control cell line (THP-1-nlrp3+/+) were constructed using clustered regularly interspaced short palindromic repeats (CRISPR) technology. Similar to cardiac glycosides ouabain, which activates the NLRP3 inflammasome [31], SC-19 infection induced cleavage of pro-casp1 and pro-IL-1β and secretion of IL-1β in THP-1-nlrp3+/+ cells, but the activation was significantly inhibited in nlrp3-/- cells (Fig 2B). This study was also performed using the murine macrophage cell line J774a.1 with nlrp3 gene knockout (J774a.1-nlrp3-/-) and the control cell line J774a.1-nlrp3+/+ (S2 Fig). Thus, NLRP3 was mainly responsible for inflammasome activation induced by S. suis epidemic strain SC-19 infection.
NLRP3 inflammasome activation can be attributed to several cellular events, including the presence of a P2X7 receptor agonist (extracellular ATP), ROS production, mitochondrial damage, lysosomal damage, formation of large nonspecific pores in the cell membrane, and cytosolic K+ efflux [32–34]. Activation of the inflammasome by SC-19 was not inhibited by the single treatment of the P2X7 antagonist KN-62, the ROS scavenger N-acetyl-L-cysteine (NAC), the phagocytosis inhibitor cytochalasin B, or the lysosomal inhibitor bafilomycin A (Fig 2C), indicating that inflammasome activation by S. suis was not dependent on the each single event or was dependent on these complicate events. However, the activation was significantly inhibited in the K+-rich media (Fig 2D). Although K+ efflux-independent NLRP3 inflammasome activation by small molecules targeting mitochondria had been observed [35], these results indicated that inflammasome activation in response to SC-19 infection was primarily dependent on K+ efflux, an essential process for recruitment of NLRP3 to the dispersed trans-Golgi network to cause K+-efflux-dependent NLRP3 activation [17].
Because SC-19 specifically activated the NLRP3 inflammasome in vitro, we further investigated the role of NLRP3 in STSLS with a small-molecule inhibitor of the NLRP3 inflammasome, MCC950, which blocks NLRP3-induced ASC oligomerization [36]. MCC950 effectively blocked inflammasome activation by SC-19 in vitro (S3 Fig). MCC950 treatment significantly reduced IL-1β level in response to SC-19 infection in mice (Fig 3A). As downstream effects of inflammasome activation, S. suis infection-induced IL-6 and IFN-γ levels were also significantly decreased by MCC950 treatment (Fig 3A). Therefore, NLRP3 inflammasome activation induced by S. suis significantly contributed to the inflammatory cytokine storm. MCC950 treatment also reduced the CK and AST levels in the blood (Fig 3B), alleviated injury in multiple organs (Fig 3C), decreased clinical signs (Fig 3D), and promoted host survival (Fig 3E), although the bacterial burden in the blood was not significantly changed at the given time point (Fig 3F). These indicated that blocking NLRP3 inflammasome could significantly inhibit STSLS caused by SC-19 infection.
To direct investigate the role of NLRP3 in STSLS, the comparison of infection was also performed on nlrp3-/- mice and nlrp3+/+ mice. Similar effects were observed for infection of SC-19 on nlrp3-/- mice, it induced significantly decreased levels of IL-1β and IFN-γ comparing to the infection on nlrp3+/+ mice (Fig 4A), while the bacterial burden in the blood did not significantly decrease at the given time point (Fig 4B). The infection on nlrp3-/- also caused significantly decreased levels of CK, AST, and LDH in the blood (Fig 4C), decreased injury in multiple organs (Fig 4D), decreased clinical signs (Fig 4E), and promoted host survival (Fig 4F). These results suggested that the NLRP3 inflammasome activation was essential for STSLS development following epidemic S. suis strain SC-19 infection.
Although we identified the NLRP3 inflammasome as being essential for STSLS development, it was also important to identify the component of S. suis involved in inflammasome activation. To identify the component of S. suis involved in inflammasome activation, we found that live, but not heat-inactivated, S. suis strain SC-19 induced very obvious cleavage of pro-casp1, pro-IL-1β, and GSDMD (Fig 5A and 5B), which resulted in pyroptosis and benefited the secretion of mature IL-1β and IL-18 [19–21]. Furthermore, the secretion of IL-1β was specific because treatment with either live or heat-inactivated S. suis did not induce significantly more TNF-α at the indicated time point (Fig 5C). Consistent with the results obtained in THP-1 cells, live, but not heat-killed, S. suis was required for IL-1β secretion, and IL-1β activation was inhibited by the casp1 inh in isolated murine peritoneal macrophages (S4A Fig) and bone marrow neutrophils (S4B Fig). Thus, live, but not heat-killed, SC-19 infection activated the inflammasome.
To further identify the component of S. suis that contributes to inflammasome activation, D-alanylation of lipoteichoic acid (DLTA) [37, 38], the capsular polysaccharides (CPS) structure [39], and SLY [38, 40–42], which are directly involved in the virulence of S. suis, were selected for evaluation of their roles in inflammasome activation. The isogenic mutants for dlta (Δdlta) (S5 Fig) or cpsEF (ΔcpsEF) induced pro-casp1, pro-IL-1β and GSDMD cleavage (Fig 5B) and IL-1β secretion (Fig 5C), similar to the wild-type (WT) strain. However, the isogenic sly mutant (Δsly) completely lost the ability to induce cleavage of pro-casp1, pro-IL-1β and GSDMD (Fig 5B) and secretion of IL-1β (Fig 5C), but it did not block TNF-α secretion (Fig 5C). In contrast, the complemental SLY strain could restore the ability for induction of inflammasome (S6 Fig). Furthermore, the purified recombinant SLY (rSLY) induced pro-casp1, pro-IL-1β and GSDMD cleavage and IL-1β secretion in a dose-dependent manner (Fig 5D and 5E and S2 Fig). These data indicated that SLY of S. suis activated the inflammasome.
Because SLY is a member of the pore-forming cholesterol-dependent cytolysin family of toxins [43, 44], we further evaluated the role of SLY in inflammasome activation by adding exogenous cholesterol, which can inhibit binding of SLY to host cells [45, 46]. Although cholesterol crystals induced the NLRP3 inflammasome [47], the addition of solubilized cholesterol at the given concentrations inhibited the pro-casp1, pro-IL-1β and GSDMD cleavage (Fig 5F) and IL-1β secretion (Fig 5G) induced by SC-19 in a dose-dependent manner. In contrast, the addition of solubilized cholesterol at the given concentration did not significantly inhibit the IL-1β secretion induced by the NLRP3 agonist ouabain (Fig 5F and 5G). These studies indicated that inflammasome activation in response to S. suis epidemic strain SC-19 infection required the binding of SLY to host cells.
Structural analysis of S. suis SLY indicated that P353L would result in a loss of hemolytic activity while retaining the biological activity of erythrocyte aggregation [43], which was further confirmed in a biological experiment using recombinant SLY [45]. To elucidate the mechanism underlying SLY-induced inflammasome activation, we constructed a mutant strain containing the P353L point substitution in SLY [msly (P353L)] to analyze the contribution of the membrane perforation activity of SLY to inflammasome activation (S7A Fig). Compared with WT strain inoculation, msly (P353L) strain inoculation failed to activate the inflammasome (Figs 2B, 5B and 5C and S2 Fig). The inability of msly (P353L) to activate the inflammasome was not due to failed SLY expression, because the amount of SLY in the supernatants of cells treated with msly (P353L) was not less than that in the supernatants of cells treated with the WT strain (S7C and S7D Fig). Therefore, our data strongly suggested that the membrane perforation activity of SLY was very important for inflammasome activation during S. suis infection.
Previous studies have indicated that SLY may confer bacterial resistance to complement-mediated killing [38, 48] and contribute to enhanced host inflammation [42], which ultimately contributes to S. suis virulence. The non-hemolytic mutant msly (P353L) retained its resistance to complement-mediated killing, while the Δsly mutant did not (S7E Fig). Therefore, the non-hemolytic mutant msly (P353L) could be used to further confirm the effect of NLRP3 inflammasome activation on STSLS.
As expected, msly (P353L) did not induce high levels of the inflammasome-regulated pro-inflammatory cytokines IL-1β and IL-18 or the downstream effectors, including IL-17A and IFN-γ, in contrast with the WT strain, but the mutant could still induce comparatively high levels of the inflammasome-unrelated cytokine TNF-α (Fig 6A). Notably, the trend in the induction of these inflammasome-related cytokines by the mutant was similar to the effect on nlrp3-deficient mice with SC-19 strain infection (Fig 4). These data suggested that membrane perforation activity was required for inflammasome activation in vivo and that inflammasome activation was essential for the development of the inflammatory cytokine storm following SC-19 infection. Interestingly, msly (P353L) infection did not result in high levels of ALT, AST, LDH and CK in the blood (Fig 6B), indicating that the mutant did not cause severe multi-organ injury, an essential aspect of STSLS. Furthermore, the bacterial burden was comparable in mice infected with the SC-19 or its mutant strain at the given time points (Fig 6C), which suggested that the decreased inflammasome activation was not attributable to differential bacterial load.
The SC-19 strain caused severe damage to multiple organs and acute death with severe clinical signs; in contrast, 90% of the mice infected with msly (P353L) survived, and only moderate clinical signs and alleviated organ damage were observed during the study (Fig 6D–6F). These data further confirmed that membrane perforation activity was required for inflammasome activation and full virulence of the epidemic strain SC-19, which can cause STSLS.
In summary, these experiments further supported our hypothesis that the membrane perforation activity of SLY leaded to NLRP3 inflammasome activation that was essential for the induction of STSLS following epidemic S. suis infection.
Highly virulent S. suis infection in humans, pigs, and mice induces STSLS, which is characterized by high bacterial burden, a cytokine storm, multi-organ dysfunction, and ultimately acute host death [8, 10, 49]. However, no superantigen responsible for toxic shock syndrome was detected in S. suis [7], indicating that the mechanism underlying STSLS is different from that of toxic shock syndrome. Although a few studies have indicated that an excessive inflammatory response is responsible for STSLS development [6] and that targeting the pathway may be a potential therapeutic strategy [11, 12], the precise mechanism underlying STSLS remains elusive.
In addition to being a characteristic of acute and fulminating infectious diseases, the “cytokine storm” plays an essential role in the associated high mortality [50, 51]. Therefore, suppression of inflammatory genes is an appealing strategy for preventing death due to severe infections [50]. The “cytokine storm” contributes to STSLS and high mortality [6]; however, the underlying mechanism was unknown. Among these cytokines, IFN-γ plays a broad and important role in severe inflammatory responses and organ injury during shock syndrome [10, 52, 53]. In the present study, NLRP3 inflammasome activation was responsible for high IFN-γ level, multi-organ dysfunction, and mortality in response to epidemic S. suis infection (Figs 3–4). These findings further demonstrated that NLRP3 inflammasome activation was important for S. suis-causing cytokine storm.
The pore-forming toxins have been reported to activate the inflammasome through various means [29, 54–56]. For extracellular Gram- bacteria, the toxins could help the bacterial outer membrane vesicles to escape from early endosomes [29], which was important for non-canonical inflammasome activation through caspase-11/4/5 to recognize the intracellular LPS [57, 58]. For extracellular Gram+ bacteria, the precise underlying mechanism remains unclear. Inflammasome activation by SC-19 was blocked in K+-rich media, which could also inhibit inflammasome activation by Streptococcus pneumonia [55]. The present study further indicated that the activation by this toxin could not be inhibited by any one of the inhibitors that block inflammasome activation by extracellular ATP and other stimulators (Fig 2), which indicated that the toxin activated inflammasome through various means. However, the activation by the toxin could be inhibited in the K+-rich media (Fig 2), providing a direct explanation for SLY activation of the inflammasome: SLY-induced formation of large pores might cause cytosolic K+ efflux-dependent NLRP3 inflammasome activation, which could further result in pro-casp1, pro-IL-1β, and GSDMD cleavage, leading to pyroptosis and facilitating the secretion of mature IL-1β and IL-18 [19–21], which ultimately leads to severe inflammation and STSLS.
In fact, the association of SLY with the virulence of S. suis has been known for decades [40–42, 59]. Although SLY does not seem to be a critical virulent factor for some strains [40], it is essential for the full virulence of the epidemic strain, which can cause STSLS [60]. SLY was first confirmed to be involved in resistance to complement-mediated killing [38, 48] and to contribute to the virulence of S. suis [42]. Recently, SLY was demonstrated to be the main stimulus for TNF-α production independently of its membrane perforation ability [61], and it was also involved in the invasive infection caused by S. suis [46, 62–64]. Here, we demonstrated that SLY was essentially responsible for the high level of inflammasome activation by S. suis (Fig 5) because the isogenic sly mutant showed no obvious ability to activate the inflammasome and inflammasome activation was significantly inhibited by soluble cholesterol, the target molecule in the cell membrane for SLY binding [44]. Furthermore, the membrane perforation activity of SLY was indispensable for inflammasome activation (Fig 5). Undoubtedly, all these pathogenic functions of SLY may contribute to the virulence of S. suis [46, 62, 63, 65, 66]. To further determine the significance of inflammasome activation by SLY for virulence, we constructed the mutant msly (P353L), which expresses SLY with a point mutation that resulted in a defect in hemolytic activity. The strain retained complement-mediated killing ability but lost its membrane perforation activity and the ability to activate the inflammasome (S7 Fig). Interestingly, the mutant maintained its ability to resist bacterial clearance and induced high levels of TNF-α, similar to the WT strain (the epidemic strain), but could not significantly induce high levels of inflammasome-related cytokines, which was similar to the effect of inflammasome inhibitors on S. suis infection. As a result, the mutant could not cause the cytokine storm and multi-organ failure (Fig 6). Therefore, the present study strongly indicates that the membrane perforation activity of SLY is important for causing high levels of NLRP3 inflammasome activation, which is essential for STSLS development.
However, it is still difficult to explain why the epidemic strain causes STSLS while other sly+ strains (such as the P1/7 strain) do not. Interestingly, the epidemic strain expressed higher levels of SLY [67], which further activated the inflammasome (S8 Fig). Surprisingly, a novel hemolysis-related gene was identified in the 89K pathogenicity island (89K PI), which could increase SLY expression [68]. Because the 89K PI was specifically present in the genome of the epidemic S. suis strain [69] and could be transferred in a T4SS-mediated horizontal manner [70], increased SLY expression due to the acquisition of the 89K PI might explain why the epidemic strain suddenly had the ability to cause high level of inflammasome activation and STSLS development. Therefore, it would be worthy to further elicit the mechanism underlying the regulation of SLY by the 89K PI.
In conclusion, we identified an important mechanism by which the epidemic S. suis strain causes STSLS (Fig 7). First, S. suis infection may activate the transcription of genes involved in the inflammasome through pattern-recognition receptors, such as Toll-like receptor (TLR) [9, 61, 71, 72]. Then, acquisition of the 89K PI enables the strain to increase SLY expression, the high membrane perforation activity of which causes several events, including cytosolic K+ efflux, an essential event for NLRP3 inflammasome activation. Thus, strong activation of the inflammasome is an important mechanism by which this strain causes the cytokine storm, multi-organ dysfunction, and a high mortality rate, which are hallmarks of STSLS. Therefore, our study provides an explanation for STSLS development and indicates that the NLRP3 inflammasome is an attractive target for the treatment of STSLS.
The S. suis epidemic strain SC-19, which shows high pathogenicity in humans, mice and pigs [11, 73], was used in the present study. The isogenic mutants for cpsEF (ΔcpsEF) [74], sly (Δsly) [75], dlta (Δdlta) and a mutant [msly(P353L)] containing a point substitution P353L were originally from strain SC-19 (S5 and S7 Figs). The S. suis strain P1/7, which induces only sporadic cases of meningitis and sepsis in pigs [76], was used as a non-STSLS-causing control. The sly gene and its predicted upstream promoter was constructed into a S. suis-E. coli shuttle vector pSET2 [77], and then introduced into Δsly strain to obtain the complemented SLY on Δsly strain (Δsly-Csly).
The experimental infectious studies were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals Monitoring Committee of Hubei Province, China, and the protocol was approved by the Scientific Ethics Committee of Huazhong Agricultural University (Permit Number: HZAUMO-2015-014). All efforts were made to minimize the suffering of the animals.
Five- to six-week-old Balb/c mice with similar body weights were randomly divided into groups of 10 mice and challenged with 0.5 mL of S. suis strains (8 × 108 CFU/mL) by an intraperitoneal (i.p.) injection to evaluate the pathogenicity of the different S. suis strains. To evaluate the effect of casp1 and NLRP3 signaling on S. suis infection, 100 μg of the casp1 inh Ac-YVAD-CHO (Merck Millipore, 400015-1MG, Germany) or PBS as a control; or 37.5 μg of MCC950 (Selleck, S7809, USA), a selective inh of NLRP3, or a PBS control were injected intraperitoneally 1 h post-infection with S. suis. The experimental infections were also performed on nlrp3-/- mice (C57BL/6 background, purchased from the Jackson Laboratory) and nlrp3+/+ mice (C57BL/6) to direct evaluate the effect of nlrp3 on STSLS development. All the mice were monitored three times a day for seven days for clinical signs and assigned clinical scores as follows [78]: 0 = normal response to stimuli; 1 = ruffled coat and slow response to stimuli; 2 = respond only to repeated stimuli; 3 = non-responsive or walking in circles; and 4 = dead. Mice exhibiting extreme lethargy or neurological signs (score = 3) were considered moribund and were humanely euthanized.
In addition to the evaluation of mortality, experimental infections were also performed with mice to evaluate the effect of various treatments on the cytokine response, blood biochemistry, and bacterial burden during S. suis infection. At the indicated time points post-infection with S. suis, mice in each group were euthanized by carbon dioxide inhalation, and blood was collected via cardiac puncture. Fifty microliters of blood was withdrawn for bacterial load analysis. The remaining blood was used to prepare plasma for analysis of the CK, ALT, AST, and LDH levels with a VITALAB SE Chemistry Analyzer and for analysis of the IL-1β (eBioscience, E09327-1647, USA), TNF-α (eBioscience, E09483-1670, USA), IL-6 (eBioscience, 88-7064-88, USA), IL-17A (eBioscience, 88-7371-88, USA), IL-18 (Sino Biological, SEK50073, China), and IFN-γ (eBioscience, 88-7134-88, USA) levels using commercial ELISA kits. Peritoneal lavage fluid was also collected from each mouse with 2 mL of PBS to analyze the bacterial load and cytokine levels. The lung, kidney, liver and spleen tissues were collected and fixed in 10% neutral buffered formalin and routinely processed in paraffin. Sections with a thickness of 2 to 3 mm were cut for hematoxylin and eosin staining for histopathologic evaluation as previously described [11].
The collected blood samples were serially diluted and then plated on Tryptic Soy Agar plates to evaluate the bacterial load.
The THP-1 nlrp3 knockout cell line (THP-1-nlrp3-/-) was constructed using CRISPR technology [79]. sgRNA (GGATCTTCGCTGCGATCAAC) for human nlrp3 was designed with an online CRISPR Design Tool (http://tools.genome-engineering.org) and then constructed into a lentiCRISPR v2 vector (Addgene, 52961) to produce the plasmid lentiCRISPR v2-hunlrp3. Then, HEK 293FT cells (ATCC source) were transfected with lentiCRISPR v2-hunlrp3, psPAX2 (Addgene, 12260), and pMD2.G (Addgene, 12259) to produce lentivirus for disruption of the nlrp3 gene. The lentivirus was then used to transduce THP-1 cells at an MOI = 0.5. After transduction, the THP-1 cells were cultured in the presence of 1μg/mL puromycin (Selleck, S7417, USA) for 5 days. The surviving THP-1 cells were diluted into 96-well plates at a concentration of 1 cell/200 μL and cultured in the presence of 1μg/mL puromycin. The THP-1-nlrp3-/- cell line was identified by a western blot assay with NLRP3 antibody (CST, 15101S, USA) and then by DNA sequencing of the nlrp3 gene. The control cell line (THP-1-nlrp3+/+) was also constructed according to the same procedure using the original lentiCRISPR v2 plasmid.
The nlrp3 knockout cell line derived from the murine macrophage cell line J774a.1 (J774a.1-nlrp3-/-) and its control cell line (J774a.1-nlrp3+/+) were constructed according to the same procedure used for THP-1 cells. The designed sgRNA targeted the murine nlrp3 gene and contained the sequence GAAGATTACCCGCCCGAGAA, and the concentration of puromycin for selection of J774a.1-nlrp3-/- or J774a.1-nlrp3+/+ cells was 2.5 μg/mL.
Cell supernatants were collected, and LDH release was quantified using a CytoTox 96 Non-Radioactive Cytotoxicity Assay (Promega, USA) according to the manufacturer’s instructions. The percentage of cytotoxicity was calculated based on LDH release in the total cell lysates.
THP-1 cells (ATCC source) were differentiated into macrophage-like cells by treatment with 50 nM phorbol myristate acetate (PMA) (Sigma, P8139-1MG) overnight. The differentiated cells (2 × 106 /mL) were primed with LPS (Sigma, L4391) at 0.5 μg/mL for 4 h and then infected with S. suis strains (2 × 107 /mL) or stimulated with ATP (Sigma, A2383) for 30 min or ouabain (Sigma, O3125) for 2 h in the presence of the following inhibitors: cholesterol (Sigma, C8667-1G), 5 μM cytochalasin B (Sigma, C274), 100 nM KN-62 (Santa Cruz, SC-3560, USA), 2.5 mM NAC (Sigma, 1009005), 50 nM bafilomycin A (InvivoGen, tlrl-baf1, USA), 100 μM casp1 inh, Ac-YVAD-CHO, or the controls containing the corresponding solvents. Then, 100-μL aliquots of the cell culture supernatants were collected to determine human TNF-α (Dakewe Group, DKW12-1720-096, China) and IL-1β (eBioscience, 88-7261-88, USA) secretion levels using commercial available ELISA kits.
The cellular proteins were extracted in Laemmli sample buffer. The proteins in the supernatants were precipitated with 20% trichloroacetic acid on ice for 30 minutes and then washed 3 times with ice-cold acetone. After the last wash, the acetone was removed by vacuum, and the pellets were allowed to air dry for 5 minutes and then dissolved in Laemmli sample buffer. The proteins were subjected to immunoblot analysis with antibodies for the detection of casp1 (Cell Signaling, 3866S, USA), GSDMD (Proteintech, 66387-1-Ig, USA), or IL-1β (Proteintech, 16806-1-AP, USA). Actin was also detected as an internal control using a specific antibody (Proteintech, 66009-1-AP, USA).
The THP-1-nlrp3-/- cell line and its control cell line (THP-1-nlrp3+/+) were also subjected to detection of inflammasome activation according to the procedure described for THP-1 cells.
Inflammasome activation was also performed using the murine macrophage cell line J774a.1 with the nlrp3 gene knockout (J774a.1-nlrp3-/-) and its control cell line J774a.1-nlrp3+/+ via western blotting with antibodies against casp1 (R&D MAB6215, USA) and IL-1β (BIO vision, 5129-30T) and with ELISA kits for TNF-α (eBioscience, E09483-1670, USA) and IL-1β (eBioscience, E09327-1647, USA).
Murine peritoneal macrophages and bone marrow neutrophils were isolated according to a procedure described previously [74]. Detection of inflammasome activation in isolated murine peritoneal macrophages and bone marrow neutrophils was also performed as described for THP-1 cells with ELISA kits for TNF-α (eBioscience, E09483-1670) and IL-1β (eBioscience, E09327-1647).
THP-1 cells (ATCC source) were differentiated into macrophage-like cells by treatment with 50 nM PMA (Sigma, P8139-1MG, USA) overnight. The differentiated cells (2 × 106 /mL) were primed with LPS (Sigma, L4391) at 0.5 μg/mL for 4 h and then treated with K+-rich media containing 45 mM KCL (Sigma, 746436) or Na+-rich media containing 45 mM NaCl (Sigma, S5886) for 1 h, followed by treatment with S. suis strain SC-19 for 2 h. The supernatants of the cells were collected for IL-1β and LDH detection.
293T cells (ATCC source) (1 × 106 /mL) were co-transfected with 0.3 μg, 0.1 μg, and 0.2 μg of expression plasmids encoding human Flag-tagged pro-IL-1β, Flag-tagged pro-casp1, and Myc-tagged ASC, respectively, and with 0.3 μg of plasmid for co-expression of GFP with NLRP3, NLRP1, NLRC4, or AIM2. The expression of these inflammasome components was confirmed by western blotting with a Myc-tag antibody (CST, 2272S, USA) and a FLAG-tag antibody (MBL, M185-3L, USA) and by examination of GFP expression with a fluorescence microscope (Nikon 80I; Tokyo, Japan).
At 24 h post-transfection, cells were infected with S. suis strain SC-19 for 2 h or transfected with poly(dA:dT) (Invivogen, tlrl-patn, USA) for 12 h. Then, cell supernatants were collected for the western blot assay with antibodies against casp1 (Cell Signaling, 3866S, USA) and IL-1β (Proteintech, 16806-1-AP, USA) and for determination of IL-1β (eBioscience, 88-7261-88, USA).
Unless otherwise specified, the data were analyzed using two-tailed, unpaired t-tests. All assays were repeated at least three times, and the data were expressed as the mean ± standard deviations. For the animal infection experiments, comparisons of survival rates and clinical scores were analyzed with a log-rank test or two-way RM ANOVA, respectively, using GraphPad Prism 6. For all tests, a value of p < 0.05 was considered the threshold for significance.
|
10.1371/journal.pgen.1004256 | DNA Glycosylases Involved in Base Excision Repair May Be Associated with Cancer Risk in BRCA1 and BRCA2 Mutation Carriers | Single Nucleotide Polymorphisms (SNPs) in genes involved in the DNA Base Excision Repair (BER) pathway could be associated with cancer risk in carriers of mutations in the high-penetrance susceptibility genes BRCA1 and BRCA2, given the relation of synthetic lethality that exists between one of the components of the BER pathway, PARP1 (poly ADP ribose polymerase), and both BRCA1 and BRCA2. In the present study, we have performed a comprehensive analysis of 18 genes involved in BER using a tagging SNP approach in a large series of BRCA1 and BRCA2 mutation carriers. 144 SNPs were analyzed in a two stage study involving 23,463 carriers from the CIMBA consortium (the Consortium of Investigators of Modifiers of BRCA1 and BRCA2). Eleven SNPs showed evidence of association with breast and/or ovarian cancer at p<0.05 in the combined analysis. Four of the five genes for which strongest evidence of association was observed were DNA glycosylases. The strongest evidence was for rs1466785 in the NEIL2 (endonuclease VIII-like 2) gene (HR: 1.09, 95% CI (1.03–1.16), p = 2.7×10−3) for association with breast cancer risk in BRCA2 mutation carriers, and rs2304277 in the OGG1 (8-guanine DNA glycosylase) gene, with ovarian cancer risk in BRCA1 mutation carriers (HR: 1.12 95%CI: 1.03–1.21, p = 4.8×10−3). DNA glycosylases involved in the first steps of the BER pathway may be associated with cancer risk in BRCA1/2 mutation carriers and should be more comprehensively studied.
| Women harboring a germ-line mutation in the BRCA1 or BRCA2 genes have a high lifetime risk to develop breast and/or ovarian cancer. However, not all carriers develop cancer and high variability exists regarding age of onset of the disease and type of tumor. One of the causes of this variability lies in other genetic factors that modulate the phenotype, the so-called modifier genes. Identification of these genes might have important implications for risk assessment and decision making regarding prevention of the disease. Given that BRCA1 and BRCA2 participate in the repair of DNA double strand breaks, here we have investigated whether variations, Single Nucleotide Polymorphisms (SNPs), in genes participating in other DNA repair pathway may be associated with cancer risk in BRCA carriers. We have selected the Base Excision Repair pathway because BRCA defective cells are extremely sensitive to the inhibition of one of its components, PARP1. Thanks to a large international collaborative effort, we have been able to identify at least two SNPs that are associated with increased cancer risk in BRCA1 and BRCA2 mutation carriers respectively. These findings could have implications not only for risk assessment, but also for treatment of BRCA1/2 mutation carriers with PARP inhibitors.
| Carrying an inherited mutation in the BRCA1 or BRCA2 gene increases a woman's lifetime risk of developing breast, ovarian and other cancers. The estimated cumulative risk of developing breast cancer by the age of 70 in BRCA1 and BRCA2 mutation carriers varies between 43% to 88%; similarly, between 11% to 59% of mutation carriers will develop ovarian cancer by the age of 70 [1]–[3]. These considerable differences in disease manifestation suggest the existence of other genetic or environmental factors that modify the risk of cancer development. The Consortium of Investigators of Modifiers of BRCA1 and BRCA2 (CIMBA), was established in 2006 [4] and with more than 40,000 mutation carriers currently provides the largest sample size for reliable evaluation of even modest associations between single-nucleotide polymorphisms (SNPs) and cancer risk. CIMBA studies have so far demonstrated that more than 25 SNPs are associated with the risk of developing breast or ovarian cancer for BRCA1 or BRCA2 carriers. These were identified through genome-wide association studies (GWAS) of breast or ovarian cancer in the general population or through BRCA1- and BRCA2-specific GWAS [5]–[8]. Cells harboring mutations in BRCA1 or BRCA2 show impaired homologous recombination (HR) [9]–[11] and are thus critically dependent on other members of the DNA repair machinery such as poly ADP ribose polymerase (PARP1) involved in the Base Excision Repair (BER) pathway. The BER pathway is crucial for the replacement of aberrant bases generated by different causes [12]. A deficiency in BER can give rise to a further accumulation of double-strand DNA breaks which, in the presence of a defective BRCA1 or BRCA2 background, could persist and lead to cell cycle arrest or cell death; this makes BRCA-deficient cells extremely sensitive to PARP inhibitors, as previously demonstrated [13]. We hypothesize that SNPs in PARP1 and other members of BER may be associated with cancer risk in BRCA1 and BRCA2 mutation carriers. SNPs in XRCC1, one of the main components of BER, have been recently evaluated within the CIMBA consortium [14], however a comprehensive study has not yet been performed of either XRCC1 or the other genes participating in BER.
In the present study, we used a tagging SNP approach to evaluate whether the common genetic variation in the genes involved in the BER pathway could be associated with cancer risk in a large series of BRCA1/2 mutation carriers using a two-stage approach. The first stage involved an analysis of 144 tag SNPs in 1,787 Spanish and Italian BRCA1/2 mutation carriers. In stage II, the 36 SNPs showing the strongest evidence of association in stage I, were evaluated in a further 23,463 CIMBA mutation carriers included in the Collaborative Oncological Gene-environment Study (COGS) and genotyped using the iCOGS custom genotyping array.
In stage I, 144 selected Tag SNPs covering the 18 selected BER genes were genotyped in 968 BRCA1 and 819 BRCA2 mutation carriers from five CIMBA centres (Spanish National Cancer ResearchCentre (CNIO), Hospital Clínico San Carlos (HCSC), Catalan Institute of Oncology (ICO), Demokritos and Milan Breast Cancer Study Group (MBCSG). Of those, 50 were excluded because of low call-rates, minor allele frequency (MAF)<0.05, evidence of deviation from Hardy Weinberg Equilibrium (p-value<10−3) or monomorphism. Associations with breast cancer risk were assessed for 94 SNPs, as summarized in Table S1. The 36 SNPs that showed evidence of association at p≤0.05 were selected for analysis in stage II. Of the 36 SNPs successfully genotyped in the whole CIMBA series comprising 15,252 BRCA1 and 8211 BRCA2 mutation carriers, consistent evidence of association with breast cancer risk (p-trend<0.05) was observed for six SNPs (Table 1). The strongest evidence of association was observed for rs1466785 in the NEIL2 gene (HR: 1.09, 95% CI (1.03–1.16), p = 2.7×10−3) for association with breast cancer risk in BRCA2 mutation carriers. We had observed a consistent association in stage I in BRCA2 mutation carriers (HR: 1.25, p = 0.06). The SNP was primarily associated with ER-negative breast cancer (HR: 1.20, 95%CI (1.06–1.37), p = 4×10−3), although the difference in HRs for ER-positive and ER-negative disease was not statistically significant. The evidence of association in Stage II was somewhat stronger when considering the genotype-specific models, with the dominant being the best fitting (HR: 1.20 95% CI: 1.09–1.37, p = 1×10−4). The associations remained significant and the estimated effect sizes remained consistent with the overall analysis when the data were reanalyzed excluding samples used in stage I of the study (data not shown). Imputation using the 1000 genomes data showed that there were several SNPs in strong linkage disequilibrium (LD) with rs1466785 showing more significant associations (p<10−3) (Figure 1).
Due to lack of power we did not perform analysis of associations with ovarian cancer in stage I. However, we performed this analysis for the 36 SNPs tested in stage II. Although they had been selected based on their evidence of association with breast cancer risk, under the initial hypothesis they are also plausible modifiers of ovarian cancer risk for BRCA1 and BRCA2 mutation carriers. We found four SNPs associated with ovarian cancer risk with a p-trend<0.01 in BRCA1 or BRCA2 mutation carriers (Table 1). The strongest association was found for rs2304277 in OGG1 in BRCA1 mutation carriers (HR: 1.12, 95%CI: 1.03–1.21, p = 4.8×10−3). The association was somewhat stronger under the dominant model (HR: 1.19, 95%CI: 1.08–1.3, p = 6×10−4). Although three other SNPs were found to be associated with ovarian cancer risk in BRCA2 mutation carriers (p-trend<10−3), these results were based on a relatively small number of ovarian cancer cases. Imputed data did not show any SNPs with substantially more significant associations with ovarian cancer risk except for rs3093926 in PARP2, associated with ovarian cancer risk in BRCA2 mutation carriers for which there was a SNP, rs61995542, with a stronger association (HR: 0.67, p = 4.6×10−4) (Figure S1).
Based on the interaction of synthetic lethality that has been described between PARP1 and both BRCA1 and BRCA2, we hypothesize that this and other genes involved in the BER pathway could potentially be associated with cancer risk in BRCA1/2 mutation carriers. Several studies have recently investigated the association of some of the BER genes with breast cancer, however, no definitive conclusions can be drawn, given that some publications suggest that SNPs in these genes can be associated with breast cancer risk with marginal p-values while others rule out a major role of these genes in the disease [15]–[21]. There is only one study from the CIMBA consortium which has evaluated the role of three of the most studied SNPs in the XRCC1 gene, c.-77C>T (rs3213245) p.Arg280His (rs25489) and p.Gln399Arg (rs25487), ruling out associations of these variants with cancer risk in BRCA1 and BRCA2 mutation carriers [14]. However, a comprehensive analysis of neither XRCC1 nor the other genes involved in the pathway in the context of BRCA mutation carriers has been performed. In the present study we have assessed the common genetic variation of 18 genes participating in BER by using a two stage strategy.
Eleven SNPs showed evidence of association with breast and/or ovarian cancer at p<0.05 in stage II of the experiment (Table 1). Of those, six showed a p-trend value<0.01 and were therefore considered the best candidates for further evaluation. Only one of those six, rs1466785 in the NEIL2 gene (endonuclease VIII-like 2) showed an association with breast cancer risk while the other five, rs2304277 in OGG1 (8-guanine DNA glycosylase), rs167715 and rs4135087 in TDG (thymine-DNA glycosylase), rs3093926 in PARP2 (Poly(ADP-ribose) polymerase 2) and rs34259 in UNG (uracil-DNA glycosylase) were associated with ovarian cancer risk.
The minor allele of NEIL2-rs1466785 was associated with increased breast cancer risk in BRCA2 mutation carriers; moreover, when considering the genotype-specific risks observed that the best fitting model was the dominant one. NEIL2 is one of the oxidized base-specific DNA glycosylases that participate in the initial steps of BER and specifically removes oxidized bases from transcribing genes [22]. By imputing using the 1000 genome data we found six correlated SNPs in strong LD with rs1466785 (r2>0.8), located closer or inside the gene and showing slightly stronger and more significant associations with the disease and therefore being better candidate causal variants. From those, we considered rs804276 and rs804271 as the best candidates given that they showed the most significant associations (p = 6×10−4 and p = 8×10−4 respectively) and there were available epidemiological or functional data supporting their putative role in cancer. SNP rs804276 has been associated with disease recurrence in patients with bladder cancer treated with Bacillus Calmette-Guérin (BCG) (HR: 2.71, 95%CI (1.75–4.20), p = 9×10−6) [23]. SNP rs804271 is located in a positive regulatory region in the promoter of the gene, between two potential cis- binding sites for reactive oxygen species responsive transcription factors in which sequence variation has been proven to alter the transcriptional response to oxidative stress [24]. Moreover, this SNP has been proposed to partly explain the inter-individual variability observed in NEIL2 expression levels in the general population and has been proposed as a potential risk modifier of disease susceptibility [25].
Several studies have been published showing associations between SNPs in NEIL2 and lung or oropharyngeal cancer risk [26], [27] but to our knowledge, no association with breast cancer risk has been reported. We hypothesize that the potential association observed in the present study could be explained by the interaction between NEIL2 and BRCA2, each of them causing a deficiency in the BER and HR DNA repair pathways, respectively. This would explain why the breast cancer risk modification due to rs1466785 would only be detected in the context of BRCA2 mutation carriers and not in the general population.
The strongest evidence of association found in BRCA1 carriers was between rs2304277 in the OGG1 gene and ovarian cancer risk. The association was more significant when considering the dominant model. OGG1 removes 8-oxodeoxyguanosine which is generated by oxidative stress and is highly mutagenic, and it has been suggested that SNPs in the gene could be associated with cancer risk [28]–[31]. This is an interesting result, given that to date only one SNP, rs4691139 in the 4q35.3 region, also identified through the iCOGS effort, has been found to modify ovarian cancer risk specifically in BRCA1 carriers [32]. SNP rs2304277 is located in the 3′UTR (untranslated region) of the gene and is probably not the causal variant, however, in this case imputations through the 1000 Genome did not show better results for a more plausible causal SNP.
We have identified four SNPs associated with ovarian cancer risk in BRCA2 mutation carriers, rs167715 and rs4135087 in the TDG gene, rs34259 in the UNG gene and rs3093926 in PARP2. However, these last results should be interpreted with caution given that the number of BRCA2 carriers affected with ovarian cancer is four-fold lower than for BRCA1 carriers and the statistical power was therefore more limited, increasing the possibility of false-positives. In the case of PARP2, imputed data showed a lower p-value of association (4×10−4) for another SNP, rs61995542, that had a slightly higher MAF than rs3093926 (0.074 vs. 0.067) (Figure S1). However, it must still be interpreted with caution due to small number of ovarian cancer cases in the BRCA2 group.
It is worth noting that, four of the five genes for which strongest evidence of association was observed, are all DNA glycosylases participating in the initiation of BER by removing damaged or mismatched bases. Apart from the already mentioned NEIL2 and OGG1, TDG initiates repair of G/T and G/U mismatches commonly associated with CpG islands, while UNG removes uracil in DNA resulting from deamination of cytosine or replicative incorporation of dUMP. We have not found strong associations with SNPs in genes involved in any other parts of the pathway, such as strand incision, trimming of ends, gap filling or ligation. It has been suggested that at least in the case of uracil repair, base removal is the major rate-limiting step of BER [33]. This is consistent with our findings, suggesting that SNPs causing impairment in the function of these specific DNA glycosylases could give rise to accumulation of single strand breaks and subsequently DNA double strand breaks that, in the HR defective context of BRCA1/2 mutation carriers would increase breast and ovarian cancer risk.
The fact that the SNPs tested are located in genes participating in the same DNA repair pathway as PARP1, make them especially interesting, not only as risk modifiers but also because they could have an impact on patients' response to treatment with PARP inhibitors. BRCA1/2 mutation carriers harboring a potential modifier SNP in DNA glycosylases could be even more sensitive to PARPi due to a constitutional slight impairment of the BER activity. This is a hypothesis that should be confirmed in further studies. The design of this study in two stages, the hypothesis-based approach adopted to select genes, and that it is based on the largest possible series of BRCA1 and BRCA2 carriers available nowadays, mean that the results obtained are quite solid However, the study still has some limitations such as the possible existence of residual confounding due to environmental risk factors for which we did not have information.
In summary, we have identified at least two SNPs, rs1466785 and rs2304277, in the DNA glycolylases NEIL2 and OGG1, potentially associated with increased breast and ovarian cancer risks in BRCA2 and BRCA1 mutation carriers, respectively. Our results suggest that glycosylases involved in the first steps of the BER pathway may be cancer risk modifiers in BRCA1/2 mutation carriers and should be more comprehensively studied. If confirmed, these findings could have implications not only for risk assessment, but also for treatment of BRCA1/2 mutation carriers with PARP inhibitors.
Eligible subjects were female carriers of deleterious mutations in BRCA1 or BRCA2 aged 18 years or older [6]. A total of 55 collaborating CIMBA studies contributed genotypes for the study. Numbers of samples included from each are provided in Table S2. A total of 1,787 mutation carriers (968 with mutations in BRCA1 and 819 with mutations in BRCA2) from the CNIO, HCSC, ICO, Demokritos and MBCSG were genotyped in the first stage of the study. Stage II included 23,463 CIMBA samples (15,252 with mutations in BRCA1 and 8,211 with mutations in BRCA2). All carriers participated in clinical and/or research studies at the host institution under IRB-approved protocols.
|
10.1371/journal.ppat.1005898 | Fis Is Essential for Yersinia pseudotuberculosis Virulence and Protects against Reactive Oxygen Species Produced by Phagocytic Cells during Infection | All three pathogenic Yersinia species share a conserved virulence plasmid that encodes a Type 3 Secretion System (T3SS) and its associated effector proteins. During mammalian infection, these effectors are injected into innate immune cells, where they block many bactericidal functions, including the production of reactive oxygen species (ROS). However, Y. pseudotuberculosis (Yptb) lacking the T3SS retains the ability to colonize host organs, demonstrating that chromosome-encoded factors are sufficient for growth within mammalian tissue sites. Previously we uncovered more than 30 chromosomal factors that contribute to growth of T3SS-deficient Yptb in livers. Here, a deep sequencing-based approach was used to validate and characterize the phenotype of 18 of these genes during infection by both WT and plasmid-deficient Yptb. Additionally, the fitness of these mutants was evaluated in immunocompromised mice to determine whether any genes contributed to defense against phagocytic cell restriction. Mutants containing deletions of the dusB-fis operon, which encodes the nucleoid associated protein Fis, were markedly attenuated in immunocompetent mice, but were restored for growth in mice lacking neutrophils and inflammatory monocytes, two of the major cell types responsible for restricting Yersinia infection. We determined that Fis was dispensable for secretion of T3SS effectors, but was essential for resisting ROS and regulated the transcription of several ROS-responsive genes. Strikingly, this protection was critical for virulence, as growth of ΔdusB-fis was restored in mice unable to produce ROS. These data support a model in which ROS generated by neutrophils and inflammatory monocytes that have not been translocated with T3SS effectors enter bacterial cells during infection, where their bactericidal effects are resisted in a Fis-dependent manner. This is the first report of the requirement for Fis during Yersinia infection and also highlights a novel mechanism by which Yptb defends against ROS in mammalian tissues.
| The pathogenic members of the genus Yersinia share a conserved virulence plasmid that primarily serves to encode a Type 3 Secretion System and its associated effector proteins. During mammalian infection, these effectors are targeted toward phagocytic cells, where they neutralize a multitude of functions, including oxidative burst. However, it has previously been reported that strains of Yersinia pseudotuberculosis lacking the virulence plasmid retain the ability to grow in mammalian tissue sites, suggesting that the Yersinia chromosome encodes a number of poorly appreciated factors that enable survival in mammalian tissue sites, even in the absence of a functional T3SS. Here, we further characterize a number of these factors, including the operon dusB-fis. Using a variety of in vitro and vivo approaches, we determined that Fis regulates the transcription of several genes implicated in ROS resistance and that dusB-fis is essential for preventing growth restriction by ROS produced by the NADPH complex of phagocytes, even in a T3SS-expressing strain. Combined, these data suggest a model in which, during tissue infection, Yersinia evade killing by ROS through both T3SS-dependent and independent mechanisms.
| Bacterial pathogens utilize both “defensive” and “offensive” strategies to survive in mammalian tissue sites and withstand the host immune response [1]. “Defensive” strategies often consist of physiological adaptations to stresses encountered in tissues, such as changes in pH or temperature, nutrient restriction, or influxes of toxic gases or proteins released by immune cells [2,3]. Many of these stresses are also found in other environments pathogens inhabit, such as soil, fomites, or in aerosol particles. By contrast, “offensive” strategies include the secretion of toxins or effector proteins that kill or block the actions of responding immune cells [1]. One such example is the Type 3 Secretion System (T3SS) used by many bacterial pathogens, including Yersinia, Salmonella, Shigella, Pseudomonas, and Chlamydia [4]. These systems translocate effector proteins into mammalian cells from the bacterial cytosol and, often, promote bacterial growth by neutralizing the anti-bacterial actions of these cells. Additionally, T3SS effectors are used by some pathogens to rearrange host cell processes to enable intracellular growth [4]. In other cases, effector proteins act to kill mammalian cells by targeting essential proteins or triggering cell death pathways [4].
T3SS effector proteins play critical roles in the virulence of the pathogenic Yersinia species, which include the pneumonic and bubonic agent Yersinia pestis, as well as the gastrointestinal pathogens Yersinia enterocolitica and Yersinia pseudotuberculosis (Yptb) [5]. These three organisms target the translocation of their T3SS effectors, called Yops, into responding phagocytic cells, particularly into neutrophils, where they dismantle a number of bactericidal responses, including the ability to phagocytose bacteria, release reactive oxygen species (ROS), and produce certain inflammatory cytokines [5–11]. The contributions of the T3SS and Yops to Yersinia pathogenesis have been extensively studied for more than two decades, and a number of reports have been published on Yop targets and their actions in mammalian cells [5,12]. However, during infection of mammalian tissue sites, not all immune cells are intoxicated with Yops [13,14], indicating that Yersinia must also employ additional, T3SS-independent strategies for surviving within the host and resisting the immune response, as remaining, non-intoxicated immune cells are competent to execute bactericidal functions. For example, it is known that at least one of these bactericidal functions, release of nitric oxide, restricts bacterial growth from a distance by cells not directly intoxicated with Yops [15]. Furthermore, Yptb lacking the pIB1 virulence plasmid, which encodes the T3SS and Yops, is capable of infecting and replicating within mouse tissues, in some cases at levels equivalent to a WT strain [16–18]. Even though infection with this strain seldom leads to death, these findings indicate that T3SS-deficient Yptb can withstand host defenses for several days and thus, likely encode “defensive” factors on its chromosome that allow for survival in harsh environments.
Indeed, some of these genes have been identified in high throughput screens in various Yersinia species [19–24], but all of these screens were performed in the presence of the virulence plasmid, which may mask the functions of some chromosomal factors. To determine which chromosomal factors contribute to infection of Yptb in the absence of the T3SS, we previously screened a library of 20,000 transposon insertions in a plasmid-deficient (pIB1-) Yptb strain during systemic infection and identified more than 30 mutants that were attenuated in livers [17]. One of these mutants, ΔmrtAB, was attenuated for virulence in the absence of the T3SS in all tissues, but only attenuated in the mesenteric lymph nodes in the presence of the T3SS, indicating that this screen uncovered factors that are redundant with the T3SS and/or that pIB1+ and pIB1- Yptb encounter distinct environments in some tissue sites [17].
Here, we follow up on our original work by employing a high-throughput, sequencing-based assay to evaluate the contribution of 18 additional genes identified in the screen to infection by both a plasmid-deficient and WT Yptb. In contrast to our findings with ΔmrtAB, we found that most of the genes evaluated were important for systemic infection by both pIB1+ and pIB1- Yptb, indicating that these genes play essential roles in virulence, regardless of the T3SS. Additional testing of the mutants in immunocompromised mice demonstrated that 4 loci were critical for virulence when interfacing with phagocytic cells. One operon, dusB-fis, prevented growth restriction by phagocytes and killing by ROS both in vitro and in vivo, in the presence of the T3SS. This work highlights the importance of studying both “offensive” (T3SS-dependent) and “defensive” (T3SS-independent) mechanisms of survival during Yersinia infection models, as both strategies are essential for the establishment of virulent infection.
By evaluating the virulence of 20,000 transposon mutants in a plasmid-deficient Yptb strain, we identified 33 mutants that were significantly defective for colonization of and/or growth within the liver, but were otherwise capable of growing in rich media at physiological temperatures [17]. However, using large pools of transposon mutants in animal infection models can sometimes result in “false negatives,” as libraries are subject to bottleneck constraints and transposon insertions can have polar effects on nearby loci [25]. Therefore, we devised and implemented a high-throughput, sequencing-based approach (a “mini” TnSeq) to simultaneously compare the survival of multiple in-frame deletion mutants in small infection pools to further evaluate the loci identified in our previous TnSeq screen (S1 Fig). Eighteen gene or operon deletion mutants containing identical in-frame scar sequences (S1A Fig) were generated in both a plasmid-deficient (pIB1-) YPIII strain (the parental strain of our original transposon library) and a plasmid-containing (pIB1+) IP2666 strain to determine whether the in vivo contributions of some of these genes may be influenced by the T3SS. We chose the IP2666 pIB1+ strain for further investigation because it encodes the known virulence factor, phoP, which is non-functional in YPIII due to a mutation [26]. The 18 operons and genes examined represent several broad functional classes, including biosynthesis of metabolic compounds, LPS synthesis and modification, and several other previously uncharacterized virulence factors (Table 1). In order to ensure that the bacterial pools used to infect mice contained an equal proportion of attenuated mutants and WT bacteria, we also constructed two deletions of “neutral genes” (Table 1), which were selected because transposon disruptions in these genes had no deleterious effects in the original TnSeq screen [17].
Mice were infected intravenously with 104 or 103 CFU of the pIB1- or the pIB1+ library, respectively. Following infection with these doses, equivalent bacterial loads were recovered from spleens at 3 days post-infection regardless of the presence of the T3SS (S2A Fig); however, livers infected with pIB1- bacteria contained lower bacterial loads than those infected with libraries generated in the WT background (S2B Fig). In each pool, the two neutral strains each comprised 25% of the inoculum, while the remaining 18 mutants each comprised ~3% of the population. Following recovery of bacteria from infected livers and spleens at 3 days post-infection, genomic DNA was processed for Illumina sequencing and fitness values were calculated for each mutant (Fig 1). Strikingly, 14 of the 18 mutants generated in the pIB1- YPIII background had statistically significant virulence defects in infected liver tissues. Of those 14 genes, all but one were also critical for growth of pIB1+ IP2666 within the liver (Fig 1A and 1B, Table 1), indicating that more than 70% of the genes evaluated were important for infection, regardless of the presence of the T3SS. Mutants attenuated for growth within the liver included the auxotrophic strains ΔaroA and ΔaroE, which are unable to produce aromatic amino acids [27,28], and ΔpurM, which lacks a component of the purine biosynthesis pathway [29]. With the exception of one strain (ΔYPK_3185), all of the strains with mutations in genes involved with LPS synthesis and modification were attenuated for virulence in at least one tissue site. Importantly, several factors that had not been previously characterized in Yersinia infection models, including YPK_2594, which has no predicted function, YPK_1920, which is predicted to encode a lipoprotein, YPK_3765, which is predicted to encode a murein peptide ligase, and the dusB-fis operon, which encodes the nucleoid associated protein Fis, were critical for infection. Six pIB1+ mutants, ΔaroA, ΔYPK_3184, ΔarnDT, ΔYPK_1920, ΔYPK_2594, and ΔpsaEFABC, were defective for growth in the liver (Fig 1B), but not the spleen (Fig 1D and Table 1), indicating tissue specific functions of these genes.
To evaluate whether mutants were attenuated when they were not a small minority of the input pool, traditional competition experiments were performed using bacterial mutants generated in the pIB1+ background. Mutants were mixed at a 1:1 ratio with a drug resistant WT strain, and C.I. values were obtained after intravenous infection (Fig 2). All of the mutants evaluated were attenuated in this assay. In conclusion, using our efficient and highly sensitive mini-TnSeq assay many bacterial mutants were attenuated in both WT and plasmid-deficient Yptb, indicating that most of these loci do not have redundant roles with the T3SS.
Infection with Yptb produces a pronounced inflammatory response, where bacteria growing in tissue sites are surrounded by phagocytic cells [15–17]. Therefore, we hypothesized that some of the genes evaluated in the mini-TnSeq assay may encode proteins that directly interface with phagocytic cells, or are important for surviving in the face of anti-microbial responses generated by these cells. To test this, mice pre-treated with the RB6-8C5 antibody, which depletes Gr1pos cells (Ly6Gpos neutrophils and Ly6Cpos inflammatory monocytes, dendritic cells, and lymphocytes) or the 1A8 antibody, which depletes Ly6Gpos cells (neutrophils only) were infected with the pIB1+ mini-TnSeq library. Surprisingly, very few significant changes in fitness scores were detected following infection of immunocompromised mice with these mutants (S3 Fig), suggesting that most of these genes are important for bacterial colonization and growth in animal tissues, regardless of the presence of these innate immune cells. However, four mutants displayed significantly altered fitness scores upon infection of immunocompromised mice (Fig 3). Growth of ΔYPK_3765 was restored to WT levels in the livers and spleens of both RB6-8C5- and 1A8-treated mice (Fig 3A–3D), indicating that this gene is required for resisting growth restriction by phagocytic cells. Surprisingly, depletion with the RB6-8C5 and 1A8 antibodies resulted in decreased growth of the ΔpsaEFABC mutant in spleens (Fig 3B and 3D), and depletion with 1A8 decreased the growth of ΔrfaH in livers (Fig 3A and 3C). These results suggest that neutrophils may protect these mutants from further growth restriction by other cells or factors in these tissue sites. Interestingly, the growth changes observed with ΔYPK_3765, ΔrfaH, and ΔpsaEFABC were specific to neutrophil depletion, as treatment with 1A8 was sufficient to alter the fitness of these mutants. By contrast, the fitness of ΔdusB-fis was restored in mice treated with the RB6-8C5 antibody (Fig 3A and 3B), but not in mice treated with the 1A8 antibody (Fig 3C and 3D). This result suggested that ΔdusB-fis is sensitive to all Gr1pos cells, to one or more Ly6Cpos cell types (inflammatory monocytes, dendritic cells, and lymphocyte subsets), or to Ly6Gpos neutrophils and a subset of Ly6Cpos cells.
To distinguish among these possibilities, we performed 1:1 co-infection experiments with ΔdusB-fis in mice treated with an antibody, MC-21, which blocks the chemokine receptor CCR2 and prevents recruitment of Ly6Cpos inflammatory monocytes to tissue sites during microbial infection [30–33]. Additional cohorts of mice were treated with 1A8, RB6-8C5, or with a combination of the 1A8 and MC-21 antibodies prior to infection. Depletion of cell subsets was confirmed by flow cytometry using Gr1 and Cd11b markers to distinguish between neutrophil and inflammatory monocyte populations (S4 Fig). Treatment with either 1A8 or MC-21 alone did not restore the virulence of the ΔdusB-fis mutant (Fig 4), indicating that the presence of either Ly6Gpos or Ly6Cpos cell type(s) at the site of infection was sufficient to restrict the growth of this mutant. By contrast, depletion with a combination of the MC-21 and 1A8 antibodies restored growth of ΔdusB-fis in livers and spleens, demonstrating that ΔdusB-fis is specifically sensitive to neutrophils and CCR2-recruited inflammatory monocytes during tissue infection. Importantly, growth of ΔdusB-fis was also restored when we complemented this mutant by re-introducing the dusB-fis genes into the deleted strain (Fig 4). To determine whether a dusB-fis mutant was more attenuated than fis, a deletion of fis was generated and evaluated in mice. The fis mutant was attenuated to the same extent as ΔdusB-fis (Fig 4A and 4B), indicating that Fis is essential for the virulence of Yptb. In summary, these results demonstrate that Fis promotes Yptb resistance to or evasion of killing by both neutrophils and inflammatory monocytes during mouse infection.
Because the virulence of the ΔdusB-fis mutant was restored in the absence of neutrophils and inflammatory monocytes, it is possible that these immune cells restricted survival of this mutant in the bloodstream after intravenous infection, thereby preventing high levels of tissue colonization. Alternatively, or in addition, neutrophils and inflammatory monocytes may restrict the growth of ΔdusB-fis in the systemic tissue sites once these cells surround the bacteria. To distinguish between these two possibilities, the growth kinetics of the ΔdusB-fis mutant were determined during systemic mouse infection at 4, 24, 48, and 72-hour time-points. In both co-infections with a WT strain (Fig 5A and 5B) and in single-strain infections (Fig 5C and 5D), the ΔdusB-fis mutant colonized tissue sites and grew for the first 24 hours post-infection with kinetics similar to WT Yptb. However, by 48 hours post-infection, the level of ΔdusB-fis failed to increase as rapidly as WT, indicating that the growth of this strain was not restricted until after initial seeding and expansion in tissue sites. Combined with our findings from the depletion experiments, these results suggest that ΔdusB-fis cannot adapt to a change in the tissue environment that likely occurs due to the influx and/or activities of neutrophils and inflammatory monocytes.
Fis serves as a transcriptional regulator of virulence factors in several pathogens [34]. Therefore, we speculated that the virulence defect of ΔdusB-fis was due to an inability to mount a transcriptional response to protect against the bactericidal actions of neutrophils and inflammatory monocytes in systemic tissue sites. Neutrophils and inflammatory monocytes use a variety of mechanisms to restrict bacterial growth upon recruitment to tissue sites, including the phagocytosis of bacteria, release of toxic granules and diffusible reactive gases (ROS and reactive nitrogen species), and chelation of metals [35,36]. Because T3SS effectors interfere with many of these processes [37], and because Fis regulates expression of the SPI-1 and SPI-2 pathogenicity islands in Salmonella [38], we first tested whether Fis positively regulated expression of the T3SS or its effectors. Under conditions that induce expression and secretion of T3SS effectors, ΔdusB-fis and Δfis mutants secreted effectors into culture supernatants at equivalent levels to WT Yptb (Fig 6A). Additionally, engineered strains of WT and ΔdusB-fis Yptb containing the beta-lactamase, TEM, fused to the first 100 amino acids of the T3SS effectors YopE or YopH [13,39] exhibited no difference in cleavage of the beta-lactamase substrate nitrocefin by those effectors (Fig 6B and 6C). Because Fis could regulate other factors, such as adhesins, which also contribute to efficient effector translocation into host cells [14], the ability of ΔdusB-fis to translocate T3SS effectors into cultured epithelial cells was measured using the CCF4-FRET based translocation assay. The ΔdusB-fis mutant had no defect in translocating YopE-TEM or YopH-TEM into cultured cells (Fig 6D and 6E), suggesting that Fis plays no role in regulating the expression of the Yptb T3SS machinery or in regulating the expression of other factors that promote efficient effector translocation through this system. Furthermore, a dusB-fis deletion generated in a strain lacking the T3SS needle was attenuated for virulence in the presence of Gr1pos cells (Fig 6F and 6G), indicating that dusB-fis is critical for preventing growth restriction by phagocytes, even in the absence of a functional T3SS.
To evaluate whether Fis promotes resistance to one or more of the bactericidal stresses imposed by neutrophils and inflammatory monocytes, the ΔdusB-fis mutant was exposed to several conditions that simulate the actions of these cells. These conditions included exposure to low pH (Fig 7A), low concentrations of iron (Fig 7B), nitric oxide (Fig 7C), and ROS (Fig 7D). While ΔdusB-fis was often delayed in entering into exponential growth compared to WT, its growth rate in broth with a low pH or titrated iron was not more impaired than a WT strain exposed to the same conditions (Fig 7A and 7B). Additionally, exposure to the nitric oxide donor DETA NONOate did not affect survival of ΔdusB-fis, but did result in killing of a mutant lacking hmp, which is known to play a role in nitric oxide detoxification by Yptb [15] (Fig 7C). By contrast, the survival of ΔdusB-fis and Δfis strains was significantly impaired after exposure to H2O2 (Fig 7D), suggesting that Fis is required for resistance to ROS.
To determine whether Fis protects against ROS by altering the transcription of one or more genes that are responsive to ROS in other organisms [40–43], we performed qRT-PCR on transcripts isolated from WT and Δfis following exposure to 20μM H2O2, a concentration that is sublethal to both strains (S5 Fig). In WT Yptb, these conditions were sufficient to induce transcription of four genes, katG, ahpC, grxA, and recA (Fig 7E). However, we observed significantly less transcriptional induction of these four genes in the Δfis mutant (Fig 7E), suggesting that Fis promotes expression of these genes during oxidative stress. By contrast, there were no differences between WT and the Δfis mutant in the expression of a non-ROS inducible gene, rpoC. In order to determine whether overexpression of a single ROS detoxifying protein was sufficient to restore growth of the Δfis mutant after exposure to lethal concentrations of H2O2, the coding regions of ahpC and katG, which encode an alkyl hydroperoxide reductase and a catalase, respectively, and have been shown to contribute to H2O2 detoxification in other organisms [44–46], were each fused to a constitutive tetracycline promoter on the plasmid pACYC184 and introduced separately into WT and ΔdusB-fis strains. Notably, while expression of these genes was enhanced in the ΔdusB-fis mutant (S6A and S6B Fig), the sensitivities of these strains to H2O2 were no different from isogenic strains expressing gfp downstream of the same promoter (S6C Fig), indicating that expression of more than one Fis-regulated gene may be required to resist killing by H2O2. Alternatively, other regulatory targets, such as the SOS-response regulator recA, may play an essential role in Fis-dependent protection against oxidative stress. Combined, these results indicate that Fis protects against killing by ROS by either directly or indirectly regulating the transcription of multiple genes required for resistance to oxidative stress.
To test the possibility that Fis protects Yptb from ROS in vivo, gp91phox-/- mice, which cannot assemble a productive NADPH oxidase complex [47], were infected with a mixture of WT and ΔdusB-fis. Strikingly, ΔdusB-fis was fully virulent in these mice (Fig 8A and 8B). Interestingly, gp91phox-/- mice were only slightly more susceptible to Yptb infection, with total bacterial loads in the spleen and liver that were only ~3.5x higher than those in tissues recovered from C57Bl/6 mice (Fig 8C and 8D). Furthermore, most of this increase was attributed to the relief in ΔdusB-fis growth restriction in the gp91phox-/- mice, as analysis of the CFU burden of each individual bacterial strain recovered from co-infected mice showed little difference in WT CFU levels between gp91phox-/- and C57Bl/6 mice (Fig 8E and 8F). This result, coupled with our earlier observations, indicates that the primary role of dusB-fis during Yptb infection within deep tissue sites is to protect against ROS produced by neutrophils and inflammatory monocytes, likely by initiating a transcriptional response that enables Yptb to resist killing by ROS that have entered the bacterial cell.
As bacteria evolve to adapt and grow in different niches, they often acquire new traits through the acquisition of plasmids, pathogenicity islands, or integration of phages or other mobile genetic elements [48]. However, these organisms can also exploit or rely on other traits they previously possessed to survive in these new environments [48]. These previously held abilities likely reflect the environments that had been central to the survival of the organism in prior niches. Thus, the observation that pIB1- strains of Yptb display a remarkable ability to grow and persist within several tissue sites during mammalian infection suggests that an ancestor of this bacterium may have relied on a number of chromosome-encoded factors to grow within mammalian tissue sites and withstand restriction by the host immune response prior to acquisition of the T3SS-encoding virulence plasmid. One of these host responses is the release of ROS, which are produced following the oxidative burst of phagocytic cells in response to fungal and bacterial infections in several tissue sites, including the GI tract, lungs, and systemic organs [49–51]. In phagocytic cells, oxidative burst occurs via the activation and assembly of the NADPH oxidase complex, usually in response to bacterial contact or pattern recognition receptor activation [52].
Our results, in aggregate, support a model (S7 Fig) whereby the dusB-fis operon in Yptb controls the transcription of genes critical for resisting killing by ROS that are generated by the NADPH oxidase complex of neutrophils and inflammatory monocytes surrounding bacteria. Specifically, (1) dusB-fis was important for defense against both neutrophils and inflammatory monocytes, as the growth of this mutant was only restored when both immune cell populations were depleted (Figs 3 and 4); (2) ΔdusB-fis initially colonized spleens and livers, but was unable to sustain growth in these tissue sites by 48 post-infection (Fig 5); (3) Fis was required for protection against oxidative stress (Fig 7D) and regulated the transcription of at least 4 genes, katG, ahpC, grxA and recA, which are predicted to contribute to resistance to ROS in Yersinia (Fig 7E); and (4) growth of ΔdusB-fis was restored in gp91phox-/- mice, whose immune cells lack a functional NADPH oxidase complex and thus cannot undergo oxidative burst (Fig 8). Remarkably, these mice contained equal bacterial loads of WT and ΔdusB-fis (Fig 8E and 8F), suggesting that the primary contribution of this operon to Yptb intravenous infection is to prevent against restriction by ROS.
The findings that dusB-fis was necessary for resisting killing by ROS in vitro and that this operon was dispensable for growth within gp91phox91-/- mice were unanticipated for several reasons. First, previous studies of Yersinia gene expression in animal models have observed little transcriptional induction of ROS-detoxifying genes during infection of lymphoid tissue sites, suggesting that Yersinia may not experience oxidative stress during growth in these organs [15,53]. However, these studies were largely directed at analyzing the transcriptional induction of ROS-responsive genes compared to in vitro growth, where bacteria may also encounter some endogenous oxidative stress, and did not assess the survival of Yptb mutants lacking one or more of these genes during animal infection. In fact, a prior study analyzing the phenotype of a mutant lacking the superoxide dismutase sodA determined that this gene was critical for growth of Y. enterocolitica within livers and spleens, suggesting that Yersinia do encounter ROS during tissue infection [54]. Second, it has been well established that two T3SS effectors, YopE and YopH, prevent oxidative burst in Yop-intoxicated phagocytic cells [7,11]. However, YopE and YopH can only function within the cells into which they have been delivered. As only a small fraction of immune cells are intoxicated with Yops during infection of some tissue sites [13], it is possible that in these tissues, Yptb encounters ROS produced by non-injected cells, and would therefore require mechanisms to resist killing by these species. Because Fis is dispensable for T3SS effector translocation, but is required for protection against ROS both in vivo and in vitro, our work suggests that Yptb does encounter ROS during infection of the spleen and liver, and that these species must be coming from neighboring immune cells not intoxicated with Yops. Furthermore, the observation that gp91phox-/- mice were no more sensitive to WT Yptb than C57Bl/6 mice (Fig 8E and 8F) suggests that WT Yptb is completely resistant to ROS produced by the immune response during infection. Therefore, we propose that Yptb utilizes both offensive and defensive measures to counteract ROS produced by phagocytic cells during mammalian infection by preventing oxidative burst in T3SS-intoxicated host cells, and also by initiating a Fis-dependent transcriptional response to protect against killing by ROS released by non-injected phagocytic cells (S7 Fig).
The dusB-fis operon is conserved in Enterobacteriaceae family members of the Gammaproteobacteria [55] and encodes the nucleoid-associated protein (NAP) Fis. While no published work has characterized a function for Fis in Yersinia, Fis and other NAPs have been well studied in E. coli and other organisms, where these small, histone-like proteins play important roles in modulating DNA architecture, as well as in directly and indirectly regulating transcription at a global level [34]. In E. coli, the two genes are co-regulated and transcribed from a single promoter upstream of dusB [56,57], where the dusB mRNA transcript is believed to play a regulatory role in promoting translation of Fis [55]. Interestingly, Fis serves as a transcriptional regulator of virulence factors in several mammalian pathogens, including Vibrio cholerae, Shigella flexneri, Pasteurella multocida, Salmonella typhimurium, and pathogenic Escherichia coli [34,38,58–65]. In these organisms, it activates a diverse range of virulence functions, including quorum sensing, capsule production, adhesion, and Type 3 Secretion [58–60,65]. Notably, a study performed in E. coli also characterized a role for Fis in protection against oxidative stress [66], suggesting that defense against ROS may be a conserved function of Fis across multiple bacterial species. However, the contribution of Fis to ROS resistance has not been examined in other pathogens.
Our findings indicate that, following exposure of Yptb to oxidative stress, Fis promotes the transcriptional induction of several ROS-detoxifying genes, as well the SOS response regulator recA. This suggests that Fis may prevent ROS-mediated killing of Yptb both by stimulating detoxification as well as by promoting the repair of DNA damage. Expressing the detoxifying genes ahpC and katG under the control of a constitutive promoter did not restore resistance of ΔdusB-fis to oxidative stress; however, this is not surprising because it is likely that Fis promotes expression of multiple genes that contribute to survival under these conditions. Future studies aimed at identifying global regulatory targets of this protein will further inform our understanding of how Fis promotes survival under these conditions.
Another surprise from our mini-TnSeq assay was the finding that the virulence defects of the ΔpsaEFABC and ΔrfaH mutants were exacerbated following neutrophil depletion during Yptb infections of the spleen and liver, respectively, indicating that these loci may promote survival in a non-inflammatory niche or in the presence of a host cell subset “unmasked” by neutrophil depletion. Consistent with this idea was the observation that the ΔpsaEFABC mutant was significantly attenuated in the pIB1-, but not the pIB1+, background in the spleen, as WT Yptb recruits a more robust, neutrophil-rich inflammatory response than its plasmid-deficient derivative in lymphoid tissue sites [16,17]. The psaEFABC operon encodes the fimbrial-like adhesin pH 6 antigen, which has been shown to contribute to lung colonization by Y. pestis [67]. Notably, Y. pestis undergoes an early “quiet” stage during infection by the pneumonic route, in which neutrophils are not recruited to the lungs until at least 24 hours post-infection [68]. Y. pestis may therefore require pH 6 antigen to colonize and grow within the lungs at early time-points because neutrophils have not yet been recruited. RfaH has been characterized as a global regulator of LPS synthesis in several gram-negative organisms, including Y. enterocolitica, where deletion of this gene results in a “rough” phenotype, in which core inner core LPS is exposed [69]. In Y. pestis, exposed core LPS promotes interactions with and uptake by dendritic cell subsets [70,71]. While neutrophils are the primary cells contacting bacteria during Yptb infection [15,17], upon neutrophil depletion, it is possible that bacteria encounter dendritic cell subsets. Thus, during this condition, the ΔrfaH mutant may come into contact with and be phagocytosed by certain dendritic cell subsets. In contrast to the ΔrfaH and ΔpsaEFABC mutants, the virulence of a mutant lacking YPK_3765 was restored in the absence of neutrophils, indicating that this gene is important for protection against clearance by these cells. YPK_3765 is predicted to encode a murein peptide ligase (Mpl), a class of proteins important for peptidoglycan synthesis and recycling in other organisms [72]. Peptidoglycan is a known activator of pattern recognition receptors [73], so a loss of this protein in Yptb may result in aberrant expression or release of peptidoglycan outside of the bacterial cell, which could further enhance killing by neutrophils during tissue infection.
Our mini-TnSeq assay offers a number of advantages in evaluating smaller cohorts of mutants initially identified in a large transposon-based screen, where extreme bottlenecks can inhibit the ability of an otherwise competent mutant to colonize tissues. In addition, transposon disruptions of genes can often have polar effects on the expression of nearby loci. Finally, screening individual mutants in single-strain and 1:1 co-infections often requires large numbers of mice. To address these issues, our assay utilizes deep sequencing as a read-out, where in-frame mutants contain scar sequences that can be used as a primer template for PCR amplification of Illumina libraries. This allowed us to use small pools of bacterial mutants in mouse infections, thereby bypassing bottleneck issues and also minimizing animal usage. Unlike our initial study, in which the operon containing the most significantly attenuated transposon mutant, mrtAB, was not required for systemic infection by a pIB1+ strain [17], the vast majority of mutants with defects in the absence of the virulence plasmid were also attenuated for infection of pIB1+ Yptb in the liver. In fact, only one mutant, ΔoppD, was attenuated for growth in livers only in the absence of pIB1. Curiously, both oppD and mrtAB encode components of transporters, suggesting that they could carry out functions that are redundant with the T3SS, or that they are critical for survival in niches that are not predominantly inhabited by WT Yptb.
Interestingly, six mutants, ΔaroA, ΔYPK_3184, ΔarnDT, ΔYPK_1920, ΔYPK_2594, and ΔpsaEFABC, were defective for growth of the WT strain in the liver, but not the spleen, reflecting the fact that the original screen was performed in the liver and suggesting that different tissue sites can influence the repertoire of bacterial virulence factors required during infection. Indeed, it has been established that mammalian organs differ in their mechanisms of sensing and responding to microbial infections [74,75] and that, consequently, bacteria may utilize genes to survive in some tissues that are dispensable in others. For example, the bacterial pathogen Francisella tularensis specifically requires tryptophan biosynthesis genes during infection of the lung in order to counteract restriction of this amino acid by a host-encoded enzyme expressed in this organ [76]. Likewise, Yptb may require certain factors, such as aroA, to survive in the liver because their products are limiting in this organ. Additionally, certain mutants, such as ΔYPK_3184 and ΔarnDT, may be more readily detected by pattern recognition receptors in the liver and would therefore fail to colonize or sustain growth in this organ. Future work with these mutants may help to uncover host immune mechanisms specific to this tissue site.
Altogether, our findings reinforce the argument that Yptb relies on a number of chromosome-encoded defense factors to grow within tissue sites and withstand restriction by immune cells. In particular, the small, histone-like protein Fis plays a critical role in protecting Yptb from ROS produced by phagocytic cells during tissue infection. Future work will be aimed at identifying the global network of Fis- regulated genes during conditions of oxidative stress to understand how this protein promotes bacterial adaptation to this condition.
This study was performed in accordance with the recommendations in the Guide for Care and Use of Laboratory Animals of the National Institutes of Health. The Institutional Animal Care and Use Committee (IACUC) of Tufts University approved all animal procedures. Our approved protocol numbers were B2012-54 and B2015-35. All efforts were made to minimize suffering; animals were monitored following infection and were euthanized upon exhibiting substantial signs of morbidity by CO2 asphyxiation followed by cervical dislocation.
Strains utilized in this study are listed in S1 Table and primers are listed in S2 Table. Yptb gene deletions were generated in pIB1- YPIII and pIB1+ IP2666, as indicated in S1 Table. Deletions replacing genes of interest with in-frame scar sequences were created using allelic exchange as follows: primers were designed to amplify ~800bp regions directly up and downstream of each targeted gene (S2 Table). These oligos also contained overlapping sequences necessary to create a ~60bp scar sequence after gene deletion. Overlapping products were combined using splicing by overlap extension (SOE) PCR and ligated into the sacB-based vector pCVD442 following restriction digestion. The resulting plasmids were introduced into E.coli DH5αλpir and integrated into the Yptb chromosome by mating in the presence of a third mating strain containing pRK600. Deletions were confirmed by PCR utilizing primers located 800bp up and downstream of the deleted gene. For fis deletion, primers with overlapping sequences were designed to amplify ~800bp regions directly up and downstream of the gene. These products were combined by SOE PCR, ligated into pCVD442, and the resulting plasmid was introduced into E. coli DH5αλpir and mated into Yptb as described above. Deletion of fis was confirmed by PCR. To complement dusB-fis, the entire operon and 800bp of both up and downstream sequences was amplified, and the product was cloned into pCVD442 by restriction digestion and ligation. The resulting plasmid was introduced into E. coli DH5αλpir and mated into Yptb ΔdusB-fis and successful restoration of the operon was confirmed by PCR. Strains containing YopE-TEM and YopH-TEM fusions were generated by mating Yptb strains with a SM10λpir strain containing the plasmid pSR47-YopETEM or pSR47-YopHTEM [13,39]. Following conjugation, bacteria were plated on kanamycin and irgasan to select for crossover of the chimeric YopE- or Yop-HTEM genes into the yopE and yopH loci. Successful crossover was confirmed by PCR. To generate strains constitutively expressing either ahpC or katG, the open reading frames of these genes were amplified by PCR and products were fused downstream of a constitutive tetracycline promoter on the plasmid pACYC184-gfp [17], using PCR [77] to replace the open reading frame of gfp with each respective product. Successful integration of ahpC and katG open reading frames was confirmed by sequencing, and plasmids were introduced into WT and ΔdusB-fis strains by electroporation and selection with 20mg/mL chloramphenicol.
All Yptb cultures were grown in L broth, with the exception of nitric oxide and H2O2 sensitivity assays (described below). Following mouse infections, tissue homogenates were plated onto L agar containing 0.5 μg/mL irgasan or a combination of 0.5 μg/mL irgasan and 50 μg/mL kanamycin to select for marked bacterial strains. During strain construction, 50 μg/mL carbenicillin and 0.5 μg/mL irgasan were used to select for strains containing integrated plasmids following matings, and 10% sucrose was utilized to select for strains that had resolved the integrated plasmid. With the exception of the T3SS and translocation assays, all cultures were incubated at 26°C with aeration. For animal infections, strains were inoculated into L broth 48 hours prior to infection. Following overnight growth, these strains were diluted 1:40 and incubated for ~8 hours, after which they were diluted 1:100 and incubated overnight.
All infections were performed by intravenous injection in 8–10 week C57Bl/6 or C67Bl/6 gp91phox-/- mice obtained from Jackson, NCI, and Taconic labs. For infections with strains constructed in pIB1- YPIII, mice were inoculated with 1 x 104 bacteria. For infections with strains constructed in pIB1+ IP2666, mice were inoculated with 1 x 103 bacteria. Competition experiments were performed using a 1:1 mixture of an unmarked strain and a strain harboring an insertion of miniTn5 KanR in a neutral locus [78]. Following infections, spleens and livers were isolated, weighed, homogenized, and plated on L agar containing 0.5 μg/mL irgasan. The quantity of CFU/gram of organ was determined by dividing the number of recovered CFU by the weight of the tissue sample extracted, or in cases where the entire organ was extracted, CFU/organ values were determined. For competition experiments, tissue homogenates were plated onto non-selective media as well as onto media containing 50 μg/mL kanamycin. The CFU count for each strain was determined by subtracting the number of KanR colonies from the total number of colonies recovered on non-selective plates. The proportion of each strain in the inoculum was confirmed using the same methods. C.I values were determined by the following equation: C.I. = (mutant/WT output ratio)/(mutant/WT input ratio). For Ly6G and Gr1 cell depletions, mice were intraperitoneally injected with 50 μg of 1A8 (Fisher) or RB6-8C5 (eBioscience) antibody 24 hours prior to and 24 hours post-infection. For inflammatory monocyte depletions, mice were intraperitoneally injected with 20 μg of MC-21 antibody [79] 1 day prior to infection and each day after until completion of the experiment. To confirm successful neutrophil and inflammatory monocyte depletion, spleen homogenates were stained with CD11b PE-Cy7 (eBioscience) and Gr1 PE Cy-5 (eBioscience) and analyzed by flow cytometry, as previously described [80].
Overnight cultures of individual strains were mixed so that each putatively attenuated mutant would represent ~3% of the inoculum, and the combined neutral mutants would represent ~50% of the inoculum. Libraries were intravenously injected into 10 untreated C57Bl/6 mice and 7–8 C57Bl/6 mice treated with either RB6-8C5 or with 1A8. At 3 days post-infection, tissues were isolated, homogenized, and plated for CFUs on 150mm agar plates so that each plate would contain ~1x104 CFUs. Bacteria were scraped off plates, mixed, and genomic DNA was extracted from a volume equivalent to ~2x109 CFUs using the Qiagen DNeasy Blood and Tissue kit. DNA libraries were prepared for sequencing using the homopolymer tail-mediated ligation PCR technique as previously described [81]. Briefly, genomic DNA was sheared by sonication and treated with terminal deoxytransferase in order to generate a 3’ poly C-tail sequence. Two rounds of nested PCR were then employed to amplify regions immediately downstream of deleted genes. These products were multiplexed using 6bp indexing primers and sequenced on the Illumina Hi-Seq 2500. Following sequencing, reads were mapped to the region immediately downstream of the deleted genes and the total number of reads for each mutant in a given organ or input pool was divided by the total number of reads obtained for that organ or input pool. Fitness values were obtained by dividing the abundance of a mutant in a given organ by its abundance in the input pool.
Strains were grown overnight at 26°C with aeration, then diluted 1:40 into L broth containing 20mM sodium oxalate + 20mM MgCl2. Cultures were grown for 2 hours at 26°C with aeration and then shifted to 37°C for 2 hours and grown with aeration. Following growth, the OD600 of each culture was measured and strains were diluted to achieve equivalent optical densities. Cultures were centrifuged and 10% trichloroacetic acid was added to culture supernatants to precipitate all secreted proteins. Precipitated proteins were pelleted by centrifugation, washed with acetone, and resolved by electrophoresis on a 12.5% SDS-polyacrylamide gel.
Strains containing chimeric YopE-TEM and YopH-TEM fusions were grown overnight at 26°C with aeration, then diluted 1:40 into L broth containing 20mM sodium oxalate + 20mM MgCl2. Cultures were grown for 2 hours at 26°C with aeration and then shifted to 37°C for 2 hours and grown with aeration. When performing T3SS assays, the OD600 of each culture was measured and strains were diluted to achieve equivalent optical densities. Cultures were centrifuged and 40 μL of the culture supernatant was applied to 10 μL of 500 μg/mL nitrocefin, for a final concentration of 100 μg/mL. After a 10-minute incubation, the A490 of samples was measured using a BioTek Synergy HT plate reader. When performing translocation assays, cultures were used to infect HEp-2 cells at the indicated multiplicities of infection. After 1 hour, cells were treated with gentamicin to stop the infection. Cells were lifted from plates using trypsin and then treated with 1 μg/ml CCF4 (Invitrogen) and 1.5 mM probenecid (Sigma). Following a 20-minute incubation, cells were analyzed by flow cytometry to quantify fluorescence following excitation at a 388 nm and blue fluorescence (450nm) and green fluorescence (530) were measured. Blue fluorescence indicated the presence of translocated effectors inside of the cell. The %blue cells were determined by dividing the number of blue cells by the total number of cells analyzed in a given sample.
For low pH growth assays, WT and ΔdusB-fis Yptb were grown overnight at 26°C with aeration, then diluted 1:100 into either L broth or L broth at pH 5.5. Cultures were grown at 26°C with aeration, and the OD600 of cultures was measured at 1-hour intervals for 12 hours. For low iron growth assays, cultures were grown overnight as described above and diluted 1:100 into a well of a 96-well plate containing L broth or L broth containing 250 μM 2,2’- Bipyridyl (Sigma). Plates were incubated for 20 hours in a BioTek Synergy HT plate reader at 26°C with aeration, and OD600 measurements were recorded for each well at 15-minute intervals.
Stationary phase cultures were diluted 1:40 into L broth and grown for 4 hours at 26°C with aeration. Cultures were then washed and diluted 1:50 into M9 glucose medium or into M9 glucose medium containing either 1.5mM H2O2 or 2.5mM of the nitric oxide donor DETA NONOate (Cayman Chemical). Samples were incubated at 26°C with aeration for 1 hour, and dilutions were then plated onto L agar in order to quantify surviving bacteria.
Stationary phase cultures were diluted 1:40 into L broth and grown for 4 hours at 26°C with aeration. Cultures were then washed and diluted 1:50 into M9 glucose medium or into M9 glucose medium containing 20 μM H2O2, and were incubated with aeration for 10 minutes. For experiments with strains containing pACYC184-ptet::katG and pACYC184-ptet::ahpC, cultures were incubated with 1.5mM H2O2 for 60 minutes prior to RNA isolation. H2O2 -treated samples were pelleted and resuspended in buffer RLT (Qiagen) + ß-mercaptoethanol, and RNA was isolated using the Qiagen RNeasy kit. DNA contamination was eliminated using the DNA-free kit (Ambion), and RNA was reverse transcribed into cDNA using M-MLV reverse transcriptase (Invitrogen), in the presence of RNase-OUT (Invitrogen). cDNA was utilized as a template in qPCR reactions with 0.5μM F and R primers (S2 Table) and SYBR Green (Applied Biosystems), using the BioRad CFX Real-Time PCR detection system. Samples were normalized to an endogenous 16S RNA control and relative expression was determined using the ΔCT and ΔΔCT methods (Applied Biosystems), when comparing treated to untreated samples.
Accession numbers for the genes described in this study in NCBI are: aroA, YPK_2670; aroE, YPK_0321; purM, YPK_1253; rfaH, YPK_3937; wecC, YPK_4030; arnDT, YPK_1834-YPK_1835; dusB, YPK_0453; fis, YPK_0452; flgD, YPK_2423; psaEFABC, YPK_2761-YPK_2757; katG, YPK_3388; ahpC, YPK_3267; grxA, YPK_2733; recA, YPK_3375; rpoC, YPK_0341.
|
10.1371/journal.pgen.1002363 | Capture of MicroRNA–Bound mRNAs Identifies the Tumor Suppressor miR-34a as a Regulator of Growth Factor Signaling | A simple biochemical method to isolate mRNAs pulled down with a transfected, biotinylated microRNA was used to identify direct target genes of miR-34a, a tumor suppressor gene. The method reidentified most of the known miR-34a regulated genes expressed in K562 and HCT116 cancer cell lines. Transcripts for 982 genes were enriched in the pull-down with miR-34a in both cell lines. Despite this large number, validation experiments suggested that ∼90% of the genes identified in both cell lines can be directly regulated by miR-34a. Thus miR-34a is capable of regulating hundreds of genes. The transcripts pulled down with miR-34a were highly enriched for their roles in growth factor signaling and cell cycle progression. These genes form a dense network of interacting gene products that regulate multiple signal transduction pathways that orchestrate the proliferative response to external growth stimuli. Multiple candidate miR-34a–regulated genes participate in RAS-RAF-MAPK signaling. Ectopic miR-34a expression reduced basal ERK and AKT phosphorylation and enhanced sensitivity to serum growth factor withdrawal, while cells genetically deficient in miR-34a were less sensitive. Fourteen new direct targets of miR-34a were experimentally validated, including genes that participate in growth factor signaling (ARAF and PIK3R2) as well as genes that regulate cell cycle progression at various phases of the cell cycle (cyclins D3 and G2, MCM2 and MCM5, PLK1 and SMAD4). Thus miR-34a tempers the proliferative and pro-survival effect of growth factor stimulation by interfering with growth factor signal transduction and downstream pathways required for cell division.
| microRNAs (miRNAs) are small RNAs that regulate gene expression by binding to mRNAs bearing a partially complementary sequence. miRNAs decrease the stability or translation of mRNA targets, leading to reduced protein expression. Understanding the biological function of a miRNA requires identifying its targets. Here we developed a sensitive and specific biochemical method to identify candidate microRNA targets that are enriched by pull-down with a tagged, transfected microRNA mimic. The method was applied to miR-34a, a miRNA that inhibits cell proliferation. We found that miR-34a can potentially regulate hundreds of genes. Computational analysis of these genes suggested a novel function for miR-34a—suppression of the pro-proliferative response to diverse growth factors. This function complements the previously known role of miR-34a in blocking cell cycle progression. Thus, by reducing the expression of an extensive network of genes, miR-34a dampens growth factor signaling as well as its downstream consequences, promotion of cell survival and proliferation.
| microRNAs (miRNAs) that promote cell differentiation, inhibit cell proliferation, or enhance DNA damage or stress-induced cell cycle arrest or death, and whose expression is reduced in some cancers, are candidate tumor suppressor genes [1]. One of the most well studied tumor suppressor miRNAs is miR-34a. Depending on cellular context [2], ectopic over-expression of miR-34a induces cell cycle arrest [3], senescence [4] or apoptosis [5]. miR-34a is up-regulated by p53 in response to DNA damage [6]–[8], but can also be transcriptionally activated independently of p53 [9], [10]. miR-34a is located on chromosome 1p36, a locus deleted in neuroblastoma, breast, thyroid, and cervical cancer [11], [12]. In other cancers, miR-34a expression is epigenetically reduced by hypermethylation [13]. miR-34a administration can inhibit tumor outgrowth in mice [4]. Thus miR-34a satisfies the criteria for a tumor suppressor gene.
The best way to understand the function of a miRNA is to identify the genes it regulates. In this study we sought to understand how miR-34a acts as a tumor suppressor by identifying its direct target genes. However, target gene identification is not straightforward because of the partial complementarity of the short ∼22 nt miRNA sequence with the miRNA recognition element (MRE) of the target gene [14]. MRE pairing to the miRNA seed region (nt 2–7) contributes significantly to target gene recognition and is the basis for the most successful target gene prediction algorithms [15], [16]. However, a perfect seed match is not necessary [17], [18] and does not guarantee targeting [19]. miRNA target prediction algorithms typically predict hundreds to thousands of putative miRNA target genes, but most predicted target genes are not bona fide targets and the best algorithms sometimes miss key targets [17], [19]–[21]. It is unclear how many target genes are in fact regulated by a given miRNA in any physiological context. Analysis of genes whose mRNA or protein expression decreases when a miRNA is overexpressed or increases when it is antagonized identifies genes that may be either direct targets or indirectly regulated [22]. Biochemical methods to capture RNA-induced silencing complex (RISC)-bound mRNAs potentially provide a more direct way to identify miRNA-regulated target genes [23]–[25]. However, immunoprecipitation has mostly been used to define the general features of miRNA-regulated mRNAs and their MREs, rather than to identify the targets of a particular miRNA.
Already 36 putative miR-34a targets have been validated by luciferase reporter assays. These targets strongly support miR-34a's role as a tumor suppressor. They include genes that promote cell cycle progression through the G1/S transition (CCND1, CCNE2, CDK4, CDK6, MYC, MYCN and E2F3) [3], [8], [10], [12], [26], enhance transcription (MYB, HNF4A and FOXP1) [9], [27], [28] or growth factor signaling (MET, MEK1, AXL and RRAS) [8], [29]–[32], inhibit apoptosis (BCL2) [33] or p53 activity (YY1, MTA2, SIRT1 and MAGE-A) [5], [32], [34], [35], and promote stem cell survival (NOTCH1, NOTCH2, LEF1 WNT1, DLL1, JAG1 and CD44) [32], [36]–[40]. The diversity of direct miR-34a targets suggests that miR-34a acts pleiotropically by regulating many genes.
To identify additional direct target genes of miR-34a without bias and understand better how miR-34a functions, we optimized a simple biochemical method to isolate mRNAs that bind to transfected biotinylated (Bi-)miR-34a [41], [42]. mRNAs significantly enriched in the Bi-miRNA pull-down with streptavidin relative to their cellular expression were candidate targets. The pull-down was performed in two unrelated cancer cell lines, K562 erythroleukemia cells and HCT116 colon carcinoma cells. p53 activates transcription of miR-34a [8]. Under basal conditions, p53-sufficient HCT116 cells highly express miR-34a, while p53-null K562 cells do not express it above background (data not shown). We selected disparate cell lines to identify genes that may be regulated in multiple cell types or more specifically in a particular context. Several thousand genes were significantly enriched in the miR-34a pull-down in each cell line and 982 were significantly enriched in both cell lines. Most known miR-34a target mRNAs expressed in these cells were pulled down with miR-34a. Despite the large number of genes significantly enriched in the miR-34a pull-down, 91% of a random list of 11 genes enriched in both cell lines contained miR-34a-regulated 3′UTR sequences. These results suggest that the pull-down is quite specific and that miR-34a potentially directly regulates hundreds of genes. Bioinformatic analysis of the pulled down genes or of genes down-regulated after miR-34a transfection suggested that miR-34a regulates a dense network of genes that transduce proliferative signals arising from growth factor stimulation. Multiple candidate target genes participate in RAS-RAF-MAPK signaling. In fact miR-34a knockout reduced sensitivity to growth factor withdrawal by serum starvation, while miR-34a transfection led to increased vulnerability. Fourteen novel miR-34a targets identified by the pull-down in both cell lines were experimentally verified, including ARAF and PIK3R2 in the RAS-RAF-MAPK pathway, and additional target genes required for cell cycle progression, including cyclins D3 and G2, MAD2L2, MCM2, MCM5 and PLK1.
We modified a method [5] for capturing miRNA-mRNA complexes using streptavidin-coated beads from cells transfected with miR-34a biotinylated at the 3′-end of the mature strand. Control samples were transfected with a biotinylated C. elegans miRNA (Bi-cel-miR-67) (Figure 1A). Biotinylation did not interfere with miRNA-mediated gene suppression as measured by luciferase reporter assay (Figure 1B). Over-expressing Bi-miR-34a or miR-34a in K562 cells also similarly suppressed expression of known miR-34a target genes (Figure 1C). Moreover, immunoprecipitation of HA-tagged Ago1 or Ago2 in K562 cells cotransfected with Bi-miR-34a specifically enriched for miR-34a by ∼4-fold and ∼6-fold, respectively (Figure 1D). Thus the Bi-miRNA is incorporated into the RISC and functions like the unbiotinylated miRNA.
We next optimized conditions to capture known target gene mRNAs. In the Bi-miR-34a pull-down of K562 cells, known miR-34a target transcripts CDK4 and CDK6, but not UBC (a housekeeping gene), were enriched 12 hr after transfection, and their capture plateaued at 24–48 hr (Figure 1E). Therefore, 24 hr was chosen for subsequent experiments. The specificity of the pull-down and applicability to other cell types was verified since CDK4, CDK6 and MYB mRNAs were consistently enriched by transfection of Bi-miR-34a, but not Bi-cel-miR-67, in K562 (Figure 1F) and HCT116 (Figure S1A) cells. Streptavidin beads did not enrich for non-target SDHA and UBC mRNAs, and the specific target mRNAs were not pulled down in cells transfected with unbiotinylated miR-34a (data not shown). miR-34a was specifically enriched >40-fold in the Bi-miR-34a pull-down compared to the input lysate (Figure S1B). Modifications of the pull-down to include formaldehyde cross-linking and/or pre-isolation of RNAs in high molecular weight cellular fractions reduced the amount of captured RNA, but did not improve the relative enrichment for known target gene mRNAs (data not shown). To confirm that association of Bi-miRNAs with target mRNAs was not a post-lysis artifact, we performed streptavidin pull-downs after adding Bi-miR-34a or Bi-cel-miR-67 to cytoplasmic extracts of untransfected K562 cells. CDK4, CDK6 and MYB mRNAs were not enriched when Bi-miR-34a was added post-lysis (Figure S1C). The general applicability of the pull-downs to enrich for miRNA target genes was also verified for another miRNA, miR-24 in HepG2 cells. Bi-miR-24 capture enriched for 3 known miR-24 targets (H2AFX, E2F2 and MYC [43]) by 2–5-fold (Figure 1G).
We next used gene expression microarrays to identify putative miR-34a targets captured by Bi-miR-34a in duplicate experiments from K562 (p53 deficient) and HCT116 cells (p53 proficient) (Table S1). mRNA abundance in the streptavidin pull-down and input in Bi-miR-34a-transfected cells were separately normalized to their levels in Bi-cel-miR-67-transfected cells. For each biological replicate, the ratio of the abundance of the pull-down mRNA compared to the input mRNA for cells transfected with Bi-miR-34a versus Bi-cel-miR-67 was calculated, averaged and used to define the enrichment ratio {Bi-miR-34a PD/Bi-cel-miR-67 PD}/{Bi-miR-34a input/Bi-cel-miR-67 input}. Normalizing to the input improved identification of true targets in 2 ways – by reducing the background caused by highly abundant mRNAs that associate with streptavidin beads nonspecifically and by incorporating a measure of mRNA knockdown into the denominator of the ratio.
The miR-34a pull-downs enriched for 2416 genes in HCT116 cells (by ≥1 standard deviation (SD), enrichment ratio ≥2.5) and for 2816 genes in K562 cells (≥1 SD, enrichment ratio ≥3.3) (Figure 2A). The overlap of genes enriched ≥1 SD in both of these unrelated cell lines was 982 genes. To determine the sensitivity of the pull-down, we first looked at how many of the 36 published targets of miR-34a were captured in the K562 or HCT116 pull-downs (Figure 2B). Of the known expressed targets, 22 of 31 mRNAs (71%) were enriched in HCT116 cells and 14 of 29 (48%) were enriched in K562 cells. It should be noted that the choice of cut-off is somewhat arbitrary. Two additional known targets had enrichment ratios of 2.5–3.2 in K562 cells. The enrichment ratio ranged from 2.7–85. 12 genes were identified in both pull-downs. The enrichment ratio for the shared hits was not significantly different in K562 cells, which do not express miR-34a, compared to HCT116 cells, which do, suggesting that the pull-downs efficiently captured miR-34a targets even in cells that express endogenous miR-34a.
To compare the mRNAs that associate with miR-34a to mRNAs that decrease with miR-34a over-expression, we measured mRNA abundance in cells transfected with miR-34a or cel-miR-67 by gene expression microarrays (Table S1). Genes whose mean mRNA level ratio decreased by at least 20% after miR-34a transfection were considered to be down-regulated either directly or indirectly by miR-34a. With this arbitrary cut-off (∼1 SD), 2087 genes were down-regulated in HCT116 cells and 945 genes were down-regulated in K562 cells (Figure 2C). About a third of these transcripts in both cell lines were also pulled down with Bi-miR-34a (30% in HCT116, 36% in K562).
Many miRNA targets contain a perfect match to the miRNA seed region in their 3′UTR. We examined the frequency of 3′UTR matches to all hexamer sequences in miR-34a in the pull-down and down-regulated gene sets relative to all genes probed on the microarray (Figure S2A). Hexamer matches to nt 2–7 in the miR-34a seed region were significantly enriched in the pull-down (HCT116 p = 1.8E-95; K562 p = 2.4E-11) and down-regulated (HCT116 p = 1.7E-24; K562 p = 1.0E-11) datasets. There was also significant enrichment in the HCT116 pull-down genes for nt 13–19 exact matches, suggesting that base-pairing there enhances miRNA binding, as has previously been shown [44]. In both cell lines, seed enrichment was greater for the overlapping set of genes that was both pulled down and down-regulated by miR-34a. For genes in this overlap, exact matches to nt 2–7 were 1.8–2.0-fold more frequent per kb of 3′UTR than for all genes on the microarray. These data suggest that genes in the overlap may be more likely to be direct targets than genes identified by only one method or that a perfect seed match might enhance miRNA-mediated mRNA decay.
We next examined hexamer enrichment in the 982 genes enriched ≥1 SD in pull-downs from both HCT116 and K562 cells (Figure S2B). Seed matches were most enriched in the 3′UTRs of these genes, with the nt 2–7 match being the most abundant (1.7 fold more abundant than in all genes on the microarray (p = 8.4E-39). The coding region (CDS) of these genes also contained a highly significant enrichment for hexamer seed matches (p = 6.1E-13). These results are consistent with recent cross-linked RISC pull-downs that suggest that 25–50% of MREs may be in the CDS [23], [25]. There was also a modest enrichment of hexamers matching the seed in the 5′UTR (p = 0.005). Thus the pull-down and down-regulated mRNAs were enriched for expected miRNA target sequence features.
We next analyzed whether mRNA expression of the enriched genes was reduced by miR-34a transfection in HCT116 cells (Figure 2D). The mRNAs of the 982 genes enriched in the miR-34a pull-down by ≥1 SD in both cell lines were significantly down-regulated after miR-34a transfection compared to the set of all genes expressed in the cell (p = 4.7E-80). The extent of down-regulation was comparable to the set of 469 TargetScan-predicted, evolutionarily conserved targets of miR-34a and significantly greater than in the larger list of 2904 poorly conserved, TargetScan-predicted genes (p = 1.6E-20). Increasing the cutoff for the enrichment ratio in the pull-down led to a greater proportion of highly down-regulated genes, indicating that a higher enrichment ratio correlates with more effective mRNA degradation and/or that highly enriched mRNAs are more likely to be miR-34a targets. Thus, the Bi-miR-34a pull-down enriches for known sequence and gene expression characteristics of bona fide miRNA targets.
To determine the specificity of the pull-down, we generated a random list (Table S2) of 11 genes enriched >2.5 fold in both pull-downs (median enrichment 3.5-fold, range 2.5–17.3). The random list contained 3 known target genes (AXL, CDK4 and FOXP1; AXL and FOXP1 were not known when the list was generated). First, qRT-PCR analysis verified that the random gene mRNAs are pulled down by Bi-miR-34a and not Bi-cel-miR-67. All 11 mRNAs were enriched (∼4–10 fold) by Bi-miR-34a pull-down in K562 cells, validating the microarray results (Figure 2E). miR-34a over-expression significantly down-regulated mRNA levels of 9 of 11 genes by 25–90% (Figure 2F). PCYOX1L expression declined by 20%, but the change was not significant. To test whether the 3′UTR of each gene could be regulated by miR-34a, the full 3′ UTR of each gene was cloned into a dual luciferase reporter plasmid. miR-34a repressed the 3′UTRs of 10 of 11 genes by ∼20–80% (Figure 2G). Thus, miR-34a could regulate the 3′UTR of 91% of a random set of genes enriched in both miR-34a pull-downs. These results suggest that the Bi-miRNA pull-down is highly specific for identifying direct miRNA targets. An important implication of the large number of genes in the overlapping target list and the low false positive rate is that miR-34a is capable of regulating hundreds of genes.
To understand miR-34a's biological functions, we analyzed the cellular pathways whose genes were most enriched in the Bi-miR-34a pull-downs (Figure 3A). In both K562 and HCT116 cells, Bi-miR-34a pull-downs enriched for genes in pathways related to growth factor signaling and cell cycle control. Bi-miR-34a pull-downs enriched significantly for genes in the EGFR, TGF-β, interleukin, estrogen, and androgen receptor signaling pathways (Figure 3A). Many of these pathways utilize common downstream signaling molecules and have a well-established link to cancer. Genes in the MAPK pathway, activated by most growth factors, were highly enriched in the pull-downs for both cell lines. Growth factor signaling also activates cell proliferation. Genes involved in cell cycle regulation, especially the G1/S transition, and the p53 response were enriched in both pull-downs, consistent with previously described targets and roles of miR-34a [3], [4], [7], [8].
We performed a similar pathway enrichment analysis for genes down-regulated by miR-34a (Figure 3B), which includes both direct and indirect miR-34a targets. The downstream effects of growth factor signaling on cell proliferation and p53 activation were more prominent in the down-regulated genes than in the pulled-down gene set, especially in p53-sufficient HCT116 cells. Cell cycle and DNA repair pathways were enriched in genes down-regulated by miR-34a in both K562 and HCT116 cells. These results suggest that miR-34a directly inhibits growth factor signal transduction and cell cycle progression pathways, culminating in reduced expression of genes needed for cell proliferation.
A pathway enrichment analysis of the TargetScan-predicted targets of miR-34a (Figure S3) also highlighted the most significantly enriched pathways in the experimental pull-down and down-regulated gene sets, notably TGFβ and MAPK signaling and cell cycle and G1/S transition. However, the significance of the enrichment was weaker and the strong role of miR-34a in growth factor signaling was less obvious.
To begin to understand regulation of growth factor signaling and cell proliferation at the gene level by miR-34a, an interactome of pulled down or down-regulated genes in HCT116 cells that participate in the significantly enriched pathways was generated (Figure 4). miR-34a potentially regulates the expression of critical genes involved in virtually every step and branch of growth factor signal transduction from ligand binding to downstream growth-promoting transcription factors. The putative direct targets included genes encoding multiple TGFβ and FGF isoforms, receptors for EGF, FGF, and insulin, and several oncogenic receptor tyrosine kinases, including MET and AXL. Several genes operating proximally in signal transduction, including SRC, PLCG1 and VAV2, were selectively pulled down. miR-34a targets also included protein kinase subunits that activate downstream signaling, including subunits of protein kinase A and C. In the RAS-RAF-MAPK signal transduction pathway, putative directly regulated genes included RRAS and RASA2, ARAF and BRAF, JAK2, and 11 MAPK genes. Although knockdown of most of the targets would be expected to inhibit cellular activation by diverse growth factors, the genes also encode for some important inhibitors, including the ubiquitin ligase CBLC, RASA2, and 5 DUSP genes (MAPK phosphatases). The pull-down also captured 76 transcripts of transcription factors, including some that orchestrate the transcriptional response to signal transduction (including STAT3, CREB1 and CREB3, SP1, ELK1 and SMAD4).
A major downstream effect of growth factor signaling and its activated transcription factors is to stimulate cell proliferation. miR-34a is already known to suppress E2F3 and some key cyclins and cyclin-dependent kinases that regulate the G1/S transition. The miR-34a pull-down enriched for additional cyclins (CCND3, CCNG2), but also for transcripts of genes that inhibit the kinases that promote exit from G1 (CDKN1C that encodes p57(KIP2), CDKN2A (p14(ARF)). Other enriched transcripts include MCM5, whose product is required to initiate DNA replication, and several genes required for mitosis (PLK1, MAD2L2 and CDC23). Ectopic miR-34a expression led to down-regulation of mRNAs for many genes needed to replicate DNA, including 2 members of the initiating complex that assembles at origins of DNA replication, 7 components of the MCM complex, 4 DNA polymerases, and 5 components of the RFC complex, a cofactor for DNA polymerase. These results suggest that miR-34a not only interferes with the signaling that transduces the growth factor response, but also directly and indirectly suppresses the expression of numerous genes needed for cell proliferation.
The Ras–extracellular signal-regulated kinase (ERK) and phosphoinositide 3-kinase (PI3K)–AKT pathways are key transducers of the cellular response to growth factors. Since many candidate miR-34a target gene products act in pathways converging on ERK and AKT activation, we analyzed the effect of miR-34a over-expression on ERK and AKT phosphorylation. miR-34a transfection reduced basal phosphorylation of ERK and AKT in HCT116 and HeLa cells (Figure 5A, 5B), but not in A549 cells (Figure S4A). miR-34a over-expression both reduced basal proliferation in the absence of serum and blunted the ability of HCT116 (Figure 5C), HeLa (Figure 5D) and A549 (Figure S4B) cells to proliferate in response to serum growth factors. Conversely, immortalized mouse embryonic fibroblasts (MEFs) genetically deficient in miR-34a were more resistant to serum starvation than WT MEFs (Figure 5E). Apoptosis measured by annexin V and propidium iodide staining was also significantly reduced in miR-34a−/− MEFs compared to wild-type MEFs after 24 hours of serum starvation (Figure 5F). Despite the strong difference in cell survival in cells deficient in miR-34a, expression of several known miR-34a targets did not differ significantly between wild-type and miR-34a−/− MEFs (data not shown). The lack of a notable difference may be due in part to compensatory up-regulation of miR-34b and miR-34c in miR-34a−/− MEFs (Figure 5G). These data suggest that miR-34a dampens the basal state of activation of proliferative and pro-survival pathways mediated by AKT and ERK by down-modulating multiple genes whose products contribute to their phosphorylation.
To determine whether some of the candidate miR-34a target genes identified in the pull-down that participate in growth factor signaling are bona fide targets, we next tested miR-34a targeting of selected receptor-proximal (AXL, MET and PIK3R2) and more downstream (ARAF and MEK1) components of ERK and AKT signal transduction pathways. These 5 genes were both pulled down with Bi-miR-34a and down-regulated by miR-34a in HCT116 cells. ARAF is a serine/threonine protein kinase that phosphorylates and activates MEK1, which in turn phosphorylates ERK [45]. AXL is a receptor tyrosine kinase that stimulates cell proliferation and also promotes metastasis [46], [47]. PIK3R2 is a regulatory subunit of PI3K [48] and MET is a tyrosine kinase receptor that activates both PI3K and RAS [49]. AXL, MET and MEK1 are described miR-34a targets [8], [29], [31], although AXL and MEK1 were not known when these studies were performed.
The transcripts of all 5 genes were enriched 3–15-fold in the Bi-miR-34a pull-down by qRT-PCR, validating the microarray results (Figure 6A). Furthermore, over-expression of miR-34a down-regulated both the mRNA and protein levels of all 5 genes (Figure 6B, 6C). All but ARAF are also predicted miR-34a targets by TargetScan. To determine whether these genes are direct miR-34a targets, we tested the 3′UTRs for 4 of the genes (ARAF, AXL, MEK1 and MET) by luciferase assay. miR-34a reduced reporter activity of these 3′UTRs by ∼40–75% (Figure 6D). Using the PITA algorithm [50] to identify potential MREs in their 3′UTRs, we found 1 potential MRE in AXL, 2 in ARAF, 3 in MEK1, 4 in PIK3R2 and 5 in MET (Figure S4). We tested repression of these MREs by miR-34a using luciferase assays. All 5 genes contained at least one miR-34a-responsive MRE (Figure 6E). Point mutations that disrupt the MRE-miR-34a interaction restored luciferase activity, validating their regulation by miR-34a. Therefore, these 5 important genes in PI3K and MAPK signaling are all directly regulated by miR-34a.
Ectopic expression of miR-34a reduces expression of multiple direct target genes whose products facilitate the G1/S transition (CDK4, CDK6, CCND1, CCNE2 and E2F3). The pull-down identified novel genes acting at the G1/S transition and genes involved in DNA replication and mitosis. Two cell cycle-regulating genes enriched in the miR-34a pull-down are in the random gene list and were already shown (Figure 2C–2E) to be miR-34a-regulated - CCNG2, which is most highly expressed in late S phase, and MAD2L2, a component of the mitotic spindle assembly checkpoint complex. To examine whether some of the other putative targets that participate in cell cycle progression are direct miR-34a targets, we focused on genes that were both pulled down and down-regulated by miR-34a in HCT116 cells (Table S2). Fourteen cell cycle-regulating genes (CDK4, CDK6, CCNE2, E2F2, E2F3, E2F5, HDAC1, CDKN2A, MCM5, PKMYT1, PLK1, SMAD4, MAD2L2 and CCND3) met these criteria. Four of these (CDK4, CDK6, CCNE2 and E2F3) are known miR-34a targets. We experimentally tested 5 of the 9 putative novel targets. These genes were CCND3, a cyclin that binds to CDK4 or CDK6 and regulates Rb phosphorylation; MCM5, a mini-chromosome maintenance (MCM) protein involved in initiating DNA replication, MYT1, a serine/threonine protein kinase that phosphorylates and inactivates CDC2, thereby negatively regulating cell cycle progression at the G2/M transition; PLK1, a serine/threonine protein kinase required for mitotic spindle maturation; and SMAD4, a TGFβ-activated transcription factor that induces G1 arrest and apoptosis. To determine whether these miR-34a pull-down genes are bona fide miR-34a target genes, we first verified that their transcripts associate with Bi-miR-34a (Figure 7A). After miR-34a over-expression, 3 of the 5 genes (MCM5, PLK1 and MYT1) had reduced mRNA by at least 2-fold (Figure 7B) and all 5 had significantly reduced protein (Figure 7C). Two other MCM genes, MCM2 and MCM4, also demonstrated a significant miR-34a-dependent reduction in mRNA, and their protein levels became undetectable in miR-34a-transfected cells.
To investigate whether these 5 genes are directly regulated, we measured changes in luciferase activity in HeLa cells after miR-34a co-transfection with reporters containing their 3′UTRs. The 3′UTRs of 4 of 5 of these genes (CCND3, MCM5, PLK1 and SMAD4) were significantly repressed 30–60% by miR-34a (Figure 7D). The 3′UTR of MYT1, which bound to Bi-miR-34a and was down-regulated by miR-34a over-expression (Figure 7A, 7B), was not regulated by miR-34a. MYT1 expression could be regulated by MREs outside the 3′UTR or indirectly. PITA and TargetScan were used to identify miR-34a MREs in the 3′UTRs of CCND3, SMAD4, MCM5, and PLK1 (Figure 7E, Figure S5). CCND3 MRE1, SMAD4 MRE1 and MCM5 MRE5 were significantly suppressed by miR-34a (Figure 7E, Figure S5). The CCND3 and SMAD4 MREs were predicted by TargetScan, while MCM5 MRE5 contains a miR-34a hexamer seed match. Mutations that disrupt base pairing with miR-34a rescued luciferase expression, further confirming that these genes are direct miR-34a targets. Because the enrichment ratios for MCM2 and MCM4 in the pull-down (∼2.3) were close to our cut-off, we also evaluated whether MCM2 and MCM4 might be direct targets. MCM2 is a direct target as verified by mRNA enrichment in the pull-down, decrease in mRNA and protein following miR-34 over-expression, miR-34a regulation of its 3′UTR by luciferase activity and MRE identification (Figure 7A–7E). However, the MCM4 3′UTR was not active in luciferase assays. Collectively, these findings suggest that miR-34a acts as a master regulator of cell proliferation, directly suppressing many key genes that control cell cycle progression.
Despite improvements in bioinformatic and experimental tools, distinguishing the direct targets of a miRNA from indirectly regulated genes remains challenging [14]. Here we describe a simple biochemical method to isolate candidate miRNA targets by streptavidin pull-down of mRNAs that associate with a transfected Bi-miRNA, and apply it to study miR-34a. Comparison of the set of mRNAs that directly associate with the Bi-miRNA with mRNAs down-regulated by miRNA over-expression makes it possible to distinguish the direct and indirect effects of a miRNA. Candidates identified by Bi-miR-34a pull-down have properties of validated miRNA targets: they are enriched for sequences complementary to the miR-34a seed and tend to decrease in expression with miR-34a over-expression. Genes that both decrease in mRNA abundance after over-expression and are isolated by Bi-miR-34a pull-down are further enriched for seed matches, indicating that either they are more likely true miR-34a targets or that a perfect seed match might enhance target mRNA degradation.
In our analysis we defined candidate direct targets using an arbitrary enrichment ratio cut-off of 1 SD, which corresponded to an enrichment of ≥2.5-fold for HCT116 cells and ≥3.3-fold for K562 cells. As the enrichment ratio cut-off was increased, mRNA suppression after ectopic miR-34a expression increased in tandem (Figure 2D). A more stringent cut-off would reduce the already low false positive rate, but also reduce the sensitivity to detect direct targets (Figure 2B). With this cut-off, we identify 71% of the known miR-34a targets expressed in HCT116 cells as “hits”, but only 48% of the known expressed targets in K562 cells. If we had also chosen a 2.5-fold cut-off for K562 cells, our sensitivity for picking targets would have increased to 55%, while a 2-fold cut-off would have increased it to 69%. Since 10 of 11 genes in the random list of genes enriched by ≥2.5 fold by Bi-miR-34a pull-downs in both cells have 3′UTRs regulated directly by miR-34a by luciferase assay, a lower cut-off for the enrichment ratio might have increased sensitivity without an unacceptable false discovery rate. Some bona fide target genes are only enriched in the pull-down by ∼2-fold; one of the novel genes we validated by identifying its MRE (MCM2) was only enriched by 2.3-fold in the pull-down of both cell lines. The low false positive rate of target identification demonstrated with the random gene list was also supported by the high degree of experimental validation of the growth factor signaling and cell cycle regulatory genes we chose to examine experimentally (Table S2). In all, we provided experimental evidence for 14 novel direct targets of miR-34a and identified 14 miR-34a MREs, of which 11 had a perfect hexamer seed match and the 3 others had perfect matches if G∶U wobbles were allowed. Thus, the majority of genes we identified as regulated by miR-34a contain canonical 3′UTR MREs with good seed pairing. In the setting of over-expression by transfection, protein levels of all 11 genes we analyzed by immunoblot declined substantially. The few target genes that we tested for which we did not find miR-34a regulation of the 3′UTR might be false positives or might be direct targets, regulated by sequences in the 5′UTR or CDS. In fact we found enrichment for hexamer seed matches in these regions in the mRNAs pulled down with miR-34a, consistent with MRE properties in recent cross-linking-RISC immunoprecipitation experiments [23], [25].
Known targets may not have been identified by the pull-down for a variety of reasons. First, not all of the targets in the literature may be correctly assigned. Second, some known targets, such as CD44, are only modestly regulated by miR-34a [40]. The ratio that defines a “hit” is arbitrary. We set a relatively high threshold for identifying “hits” to maximize the specificity of the method (especially given the large numbers of enriched mRNAs in the pull-down), which came at the cost of sensitivity. Some known targets, which we did not designate hits with our 1 S.D. threshold of the enrichment ratio (which corresponded to >3.3 in K562 cells) had enrichment ratios of 2.5–3.2 in K562 cells. Other bona fide targets may have low, but detectable expression levels, and could have been missed due to the low sensitivity and inter-assay variability of microarray experiments. In addition to cellular variation in endogenous miRNA expression and RISC abundance, other context-dependent biological factors, such as target site accessibility, might vary due to the expression of RNA binding proteins, which could influence the efficiency of miRNA target site binding and the mechanism of targeting [51], [52]. Cell-type specific expression of other MRE-containing genes that compete for miRNA binding could also influence the pull-down enrichment ratio [53]. Finally, some missed targets are likely to be false negatives.
Normalizing the pulled down mRNAs to their abundance in the input cellular mRNA was critical to eliminate from consideration highly abundant housekeeping mRNAs. Our pull-down method modified a previously developed protocol [41], [42], which did not normalize the pull-down mRNAs to the input RNA. Many of the “hits” pulled down with Bi-miR-10a included ribosomal mRNAs, which may represent background binding of very abundant transcripts. Moreover, the miR-10a “hits” were not enriched for mRNAs containing miR-10a 3′UTR seed matches and were not down-regulated by miR-10a over-expression. In other work to be presented elsewhere, the pull-down method was used to identify genome-wide targets of miR-200c and miR-21. Importantly, the miR-200c and miR-21 pulled down mRNAs are also enriched for known targets and for 3′UTR seed sequences.
An advantage to the Bi-miRNA pull-down method described here is its simplicity. In contrast to mRNA expression-based target identification methods, Bi-miRNA pull-downs should identify only direct targets, excluding genes whose expression is indirectly modulated by changes in miRNA expression. Because the degree of mRNA suppression mediated by miRNAs is often small relative to changes in protein, methods that rely on changes in mRNA expression in response to manipulation of miRNA levels will necessarily miss some direct targets. Although the enrichment ratio takes into account a reduction in target gene mRNA in its denominator, the pull-down should not only identify target genes whose mRNA levels decline, but also those that are regulated primarily by inhibiting translation. Unlike approaches based on Ago pull-downs, the Bi-miRNA pull-down identifies the mRNAs directly associated with a specific miRNA, simplifying analysis of biological processes regulated by the miRNA.
The method described here without cross-linking does not directly identify MREs. The streptavidin pull-down method might, however, readily be modified to include cross-linking, RNase digestion of unbound mRNA segments and sequencing, similar to the HITS-CLIP protocol [23], [24], to capture not only direct targets, but also identify MREs of an individual Bi-miRNA. Isolating RNAs associated with an individual miRNA rather than all RISC-associated RNAs in cells over-expressing the miRNA of interest might be a more direct way to define specific target sequences. Future bioinformatic studies of Bi-miRNA pull-down datasets could be used to better define in an unbiased manner the sequence features that dictate miRNA targeting, and could reveal non-canonical modes of targeting, such as those that contain only partial seed complementarity [17] or pairing to the central region of the miRNA [18] or that lie outside the 3′UTR. Indeed, in this work, we enriched for mRNAs with 5′UTR and CDS seed matches, indicating that some direct miR-34a targets may be regulated outside of their 3′UTR.
Only 29% of the 2416 enriched genes in the HCT116 pull-down had down-regulated mRNA levels by mRNA microarray analysis after over-expressing miR-34a for one day, while 10 of 11 randomly chosen genes in the pull-down had significantly decreased mRNA by qRT-PCR analyzed 72 hr after transfection. Thus although miRNAs may commonly lead to mRNA degradation, the degree of mRNA down-regulation of most genes is slight if cells are harvested within a day of transfection. mRNA microarrays may be too noisy to detect subtle changes in expression, unless the analysis is performed on many replicates. Our data also suggest that the kinetics of mRNA degradation may be slow. The early 24 hr time point used for the assay may have fortuitously enhanced our ability to capture miRNA-bound transcripts before too many had been degraded. Indirect effects of the miRNA are also likely to increase over time. The set of genes enriched in the miR-34a pull-down of both HCT116 and K562 cells contains 76 transcription factors or co-factors, whose suppression would reduce many mRNAs.
One important corollary of our results is that miR-34a likely directly regulates hundreds of genes. However, further experimental work is needed to assess how many of the hundreds to thousands of genes whose mRNAs associated with ectopic miR-34a are actually directly regulated by endogenous miR-34a. Possibly only a minority of potential targets is indeed directly regulated in an individual cell at any time. Based on our analysis (Figure 2D), the genes whose transcripts are most enriched in the pull-down may be the most significant targets in a given context. Additional experiments are needed to probe the functional consequences of miR-34a regulation of the genes we identified as targets. The directly regulated genes might vary considerably from cell type to cell type or even in the same cell lineage depending on differentiation state or environmental conditions. For this study we focused on the shared targets identified in two very different types of cells, rather than the ones that were unique to each cell-type. The pull-down method could be used in the future to compare miRNA target genes in different cellular contexts. Notably, the effect of miR-34a on cell signaling differed in the cancer cells we examined. Basal phosphorylation of AKT and ERK was reduced by miR-34a over-expression in HCT116 and HeLa cells (Figure 5), but not in A549 cells (Figure S4). Constitutively active RAS in A549 cells may override the effect of miR-34a in that context. Our results suggest that a dense network of genes that participate in common pathways, sometimes with opposing functions, is capable of being regulated by one miRNA. Although we observed a clear effect of genetic loss of miR-34a on the ability to cells to survive growth factor withdrawal, we did not see reduced expression in miR-34a−/− compared to wild-type cells of some of the key miR-34a target genes we identified. Since growth factor signaling is so central to cell survival and proliferation, the permanent loss of miR-34a expression likely led to myriad compensatory changes. This seeming paradox supports the conclusions of our study – namely that a single miRNA may exert its biological effect by regulating expression of hundreds of genes. The capacity of miR-34a to potentially regulate so many genes that affect growth factor signaling may enable it to exert an effect in diverse contexts.
The numbers of genes that are actually regulated by miR-34a in any setting will likely depend on how strongly miR-34a is expressed. In our pull-down, we greatly over-expressed miR-34a. However, the level of over-expression throughout this study was not greater than endogenous miR-34a expression in some physiological settings, i.e. in K562 cells stimulated with phorbol ester where miR-34a increases 1000-fold [9]. There may be a target gene hierarchy – some genes regulated by low levels of miR-34a, others regulated only by high levels.
The dense network of cell signaling genes captured in the pull-downs suggests that an important function of miR-34a is to regulate the proliferative and activation responses to extracellular growth factors. Despite its function in regulating growth factor signaling and cell proliferation, we did not find a significant variation in miR-34a expression after serum starvation or when cells were synchronized in different phases of the cell cycle (data not shown). In this study we experimentally verified as direct miR-34a targets 5 growth factor signaling genes (ARAF, AXL, MEK1, MET and PIK3R2). miR-34a was previously shown to inhibit the G1/S transition [3], [8]. Here we identified 7 novel cell cycle-regulating direct targets that included genes also required for DNA replication and mitosis. The ultimate anti-proliferative effect of miR-34a integrates both direct consequences of suppressing expression of genes required for progression through the G1/S transition and at other steps of the cell cycle as well as indirect anti-proliferative effects from repressing the growth factor signaling pathways that activate cell cycle progression. Consistent with our genome-wide target gene analysis, miR-34a expression resets the basal state of ERK and AKT phosphorylation in several cell lines, rendering cells less responsive to growth factor signaling (Figure 5). This was shown both by miR-34a overexpression as well as by genetic deletion. miR-34a may reduce cellular sensitivity to growth factor signaling by suppressing many genes in multiple signal transduction pathways. miR-34a candidate targets include genes that are universally involved in transmitting growth factor activation signals as well as some that participate in specific pathways. The particular signaling genes that are suppressed in a given cell line will likely vary from cell to cell, depending on the growth factors to which the cell responds. These types of differences likely contribute to the incomplete overlap between the enriched pathways captured in the two hematopoietic and colon cancer cell lines examined here.
HCT116, K562, A549 and HeLa cells were from ATCC. miR-34a+/+ and miR-34a−/− MEFs were generated from E14.5 littermate embryos. A full description of the mice will be published elsewhere. MEFs were transformed by infecting the cells with retroviruses encoding H-RAS-V12 and E1A and by selection with puromycin (1 µg/ml) and hygromycin (50 µg/ml). The plasmids for expression of H-RAS-V12 (plasmid 9051) and E1A (plasmid 18748) were obtained from Addgene. The VSV-G pseudotyped viruses were produced in 293T cells using the standard protocol. MEFs, HCT116, A549 and HeLa cells were grown in DMEM with 10% fetal bovine serum and supplemented with penicillin, streptomycin, HEPES, L-glutamine and β-mercaptoethanol, K562 cells were grown in RPMI containing 10% fetal bovine serum and the same supplements.
For most experiments, 2×106 HCT116 or K562 cells were transfected with 200 pmol hsa-miR-34a or cel-miR-67 miRNA mimics (Dharmacon), using Amaxa nucleofection according to the manufacturer's protocol. Biotin was attached to the 3′-end of the active strand. HeLa and A549 cells were transfected with Lipofectamine 2000 and miRNA mimics at a final concentration of 50 nM (Invitrogen). To study the association of Bi-miRNAs with HA-Ago1 or HA-Ago2, pIRESNeo (Clontech) or pIRESNeo-HA-Ago1 or pIRESNeo-HA-Ago2 (Addgene) plasmids were co-transfected in six-well plates (2 µg/well, 1×106 cells/well) with 200 pmol Bi-miR-34a or Bi-cel-miR-67 using Amaxa as per the manufacturer's instructions.
Total RNA was isolated using Trizol reagent (Invitrogen), treated with DNase I (Ambion) and reverse transcribed using random hexamers and superscript III reverse transcriptase (Invitrogen). qRT-PCR was performed in triplicate samples using SYBR Green FastMix (Quanta) on a BioRad CFX96. mRNA levels were normalized to housekeeping genes GAPDH, UBC or SDHA. miRNA was quantified in triplicate using the TaqMan MicroRNA Assay (Applied Biosystems) as per the manufacturer's instructions and normalized to U6. Primer sequences are listed in Table S3.
Whole cell lysates from transfected K562 or HCT116 cells were prepared using RIPA buffer. Proteins were analyzed by SDS-PAGE, transferred to nitrocellulose membranes and probed with the following antibodies: AXL [4566], ARAF [4432], MEK1 [9124], CDK4 [2906], MCM2 [3619], PKMYT1 [4282], PLK1 [4513], SMAD4 [9515], FOXP1 [2005], RBBP4 [4633], AKT [9272], pAKT ser-473 [4051], ERK [4370], pERK [9107] from Cell Signaling; MET [sc-161], MCM5 [sc-165995], E2F1 [sc-251], E2F3 [sc-879], CHEK1 [sc-8408] from Santa Cruz; ACSM3 [SAB1400253], MAD2L2 [SAB1400387], AGBL5 [AV53752], CCNG2 [AV03032], PSMD5 [WH0005711M1] from Sigma; MCM4 [06-1296] from Millipore; and PI3KR [610045], BD Biosciences. Western Blots were quantified by densitometry.
HCT116 or K562 cells (1×106) were transfected in triplicate with Bi-miR-34a or Bi-cel-miR-67 (Dharmacon) as described above and then cultured in six-well plates. Twenty-four hours later, the cells from 3 wells were pelleted at 500×g. After washing twice with PBS, cell pellets were resuspended in 0.7 ml lysis buffer (20 mM Tris (pH 7.5), 100 mM KCl, 5 mM MgCl2, 0.3% NP-40, 50 U of RNase OUT (Invitrogen), complete mini-protease inhibitor cocktail (Roche Applied Science)), and incubated on ice for 5 min. The cytoplasmic lysate was isolated by centrifugation at 10,000×g for 10 min. Streptavidin-coated magnetic beads (Invitrogen) were blocked for 2 hr at 4°C in lysis buffer containing 1 mg/ml yeast tRNA and 1 mg/ml BSA (Ambion) and washed twice with 1 ml lysis buffer. Cytoplasmic lysate was added to the beads and incubated for 4 h at 4°C before the beads were washed five times with 1 ml lysis buffer. RNA bound to the beads (pull-down RNA) or from 10% of the extract (input RNA), was isolated using Trizol LS reagent (Invitrogen). The level of mRNA in the Bi-miR-34a or Bi-cel-miR-67 control pull-down was quantified by qRT-PCR or mRNA microarray. For qRT-PCR, mRNA levels were normalized to a housekeeping gene (GAPDH, SDHA or UBC). The enrichment ratio of the control-normalized pull-down RNA to the control-normalized input levels was then calculated.
Total RNA (independently in two experiments) was amplified, labeled and hybridized to Affymetrix U133 plus 2.0 mRNA microarrays. The quality of the RNA was assessed before performing the microarray and the quality of the microarray data was assessed using affyPLM and Affy software. The replicate data sets for the 4 sets of samples (pull-down and input for miR-34a and cel-miR-67) were compared using an unsupervised hierarchical clustering algorithm, which verified the similarity of the duplicates. The microarray data were normalized using RMA [7] to reduce interarray variation. The enrichment ratio {Bi-miR-34a PD/Bi-cel-miR-67 PD}/{Bi-miR-34a input/Bi-cel-miR-67 input} was calculated for each probe. For genes represented by multiple probes, the mean ratio for all the probes was calculated. Genes for which none of the probe hybridization signals exceeded the background were considered not expressed and were disregarded in the analysis. For informatic analysis of the PD data, genes whose enrichment ratio were ≥1 SD above background based on a log-normal distribution were considered “hits”.
HCT116 or K562 cells were transfected in independent duplicate experiments as above with unbiotinylated miR-34a or cel-miR-67 (Dharmacon) and total RNA was harvested 24 hr later and analyzed as above by gene expression microarrays. After normalization, fold changes for each probe were calculated as the ratio of input RNA from miR-34a-transfected cells to the ratio of input RNA from cel-miR-67-transfected cells. Genes were considered down-regulated if the ratio decreased by at least 20%, which corresponded to ∼1 SD. To test the expression levels of putative target sets, each gene list was plotted in a cumulative distribution function (CDF) plot, and the Kolmogorov-Smirnov [KS] test was used for statistical comparisons between gene sets.
To determine whether a gene was also a predicted target of miR-34a, the presence of miR-34a binding sites was analyzed using TargetScan 4.2 (http://www.targetscan.org/) [39], [54], [55] or PITA (http://132.77.150.113/pubs/mir07/mir07_prediction.html) [50].
The mature hsa-miR-34a sequence was obtained from miRBase (http://mirbase.org/). All RefSeq human mRNA sequences were downloaded from NCBI in July 2009 (http://ftp.ncbi.nih.gov/). mRNAs were indexed by Entrez Gene ID; in cases where multiple sequences matched a gene ID, the sequence with the longest 3′UTR was selected. For each test gene list and miR-34a hexamer, the miR-34a hexamer frequency (hexamer matches per kb of sequence) was calculated. The frequency of hexamer matches for all genes on the microarray (the background set) was also determined. Gene IDs with no corresponding sequence in the database were excluded from analysis. Monte Carlo simulations of equally sized random gene sets (without replacement) were used to generate an empirical 2-tailed p-value for each gene set/hexamer combination. When p<1E-4, the p-value was calculated from curve fitting relative to the random background distribution.
For each of the lists of down-regulated and pull-down-enriched genes, the p-value of over-representation in a suite of canonical pathways (KEGG [56] and Wikipathways [57]) was determined using the hypergeometric distribution. A visualization of the relationship between the enriched pathways (p<0.001) based on the number of overlapping genes was rendered using Cytoscape [58]. The network of gene-gene interactions underlying these relationships was constructed based on interactions supplied by MetaCore (GeneGo Inc). Physical, predicted and genetic interactions were used to connect the down-regulated and pull-down enriched genes within the significant signaling, cell cycle or DNA repair pathways. Signaling pathway genes with no connection to any other node were removed and the network was arranged according to predicted sub-cellular localization.
HeLa cells were cotransfected in 24 well plates using Lipofectamine 2000 (Invitrogen) with 50 nM miR-34a mimic or control miRNA mimic and 50 ng of psiCHECK2 (Promega) vector containing the MRE or 3′UTR of indicated genes cloned into the multiple cloning site of Renilla luciferase. After 48 hr of transfection (unless otherwise indicated) luciferase activities were measured using the Dual Luciferase Assay System (Promega) and Top count NXT microplate reader (Perkin Elmer) per manufacturer's instructions. All experiments were performed at least in triplicate. Results were normalized to those obtained in cells transfected with an empty vector. For some experiments, a perfectly complementary antisense sequence to the active strand of miR-34a was inserted into the multiple cloning site for use as a positive control. Data were normalized to Firefly luciferase and results from 3 independent experiments were compared. Sequence of primers used for cloning 3′UTRs for miR-34a target genes are listed in Table S4. MREs sequences were cloned into psiCHECK-2 by annealing complementary oligomers matching each MRE sequence (Figures S4, S5) with overhanging ends complementary to the XhoI and NotI sites of psiCHECK-2.
HCT116, HeLa and A549 cells were transfected as described above. One day after transfection, cells were placed in serum-free medium or medium containing 10% fetal calf serum. 48 hours after the medium was changed, total cell numbers were counted. MEFs were plated at a density of 2.5×105 or 5×105 cells per well of a 6-well plate. The medium was changed to vary serum concentration 24 hr after plating. The MEFs were harvested 24 hr later and counted using Trypan blue staining or stained in PBS+0.4% BSA with annexinV-APC (Invitrogen) at a 1∶30 dilution, then washed once and stained with propidium iodide (4 µg/ml) (Sigma-Aldrich).
|
10.1371/journal.pgen.1007640 | Genome-wide association studies and CRISPR/Cas9-mediated gene editing identify regulatory variants influencing eyebrow thickness in humans | Hair plays an important role in primates and is clearly subject to adaptive selection. While humans have lost most facial hair, eyebrows are a notable exception. Eyebrow thickness is heritable and widely believed to be subject to sexual selection. Nevertheless, few genomic studies have explored its genetic basis. Here, we performed a genome-wide scan for eyebrow thickness in 2961 Han Chinese. We identified two new loci of genome-wide significance, at 3q26.33 near SOX2 (rs1345417: P = 6.51×10−10) and at 5q13.2 near FOXD1 (rs12651896: P = 1.73×10−8). We further replicated our findings in the Uyghurs, a population from China characterized by East Asian-European admixture (N = 721), the CANDELA cohort from five Latin American countries (N = 2301), and the Rotterdam Study cohort of Dutch Europeans (N = 4411). A meta-analysis combining the full GWAS results from the three cohorts of full or partial Asian descent (Han Chinese, Uyghur and Latin Americans, N = 5983) highlighted a third signal of genome-wide significance at 2q12.3 (rs1866188: P = 5.81×10−11) near EDAR. We performed fine-mapping and prioritized four variants for further experimental verification. CRISPR/Cas9-mediated gene editing provided evidence that rs1345417 and rs12651896 affect the transcriptional activity of the nearby SOX2 and FOXD1 genes, which are both involved in hair development. Finally, suitable statistical analyses revealed that none of the associated variants showed clear signals of selection in any of the populations tested. Contrary to popular speculation, we found no evidence that eyebrow thickness is subject to strong selective pressure.
| Hair plays an important role in primates and is clearly subject to adaptive selection. While humans have lost most facial hair, eyebrows are a notable exception. Eyebrow thickness is heritable and widely believed to be subject to sexual selection. Nevertheless, few genomic studies have explored its genetic basis. Here we performed genome-wide association studies for eyebrow thickness in multiple ethnic groups, including Han Chinese, Uyghurs, Latin Americans, and Caucasians. We found solid evidence that novel genetic variants near the SOX2, FOXD1 and EDAR genes could affect eyebrow thickness. After fine mapping, we prioritized four variants for experimental verification. CRISPR/Cas9-mediated gene editing provided evidence that the variants rs1345417 and rs12651896 affect the transcriptional activity of the nearby SOX2 and FOXD1 genes. This represents a successful example of a combination of GWAS and CRISPR/Cas9 technology to demonstrate how non-coding variants with regulatory functions may play an important role in common diseases and traits. Finally, suitable statistical analyses suggest that, contrary to popular speculation, eyebrow thickness should not be subject to strong selection pressure, including sexual selection.
| Hair has a range of important functions in primates and has been speculated to be subject to intense natural and sexual selection [1]. Although humans have lost most terminal body hair to allow the development of more efficient sweating as an adaptation to bipedal life, considerable facial hair remains, with great diversity between populations [2]. The eyebrow, an area of thick, short facial hair above the eye that follows the shape of the lower margin of the brow ridge, is one of the most conspicuous features of the face. It is thought that its main function is to prevent sweat, water, and other debris from getting into the eye [2]. As a major facial feature, the eyebrow plays an important role in human communication, facial expression, sexual dimorphism and attractiveness [3–6]. It has been suggested that eyebrows may be subject to sexual selection [7]. Notably, eyebrow thickness varies between and within human populations [8,9]. A recent genome-wide association study in Latin Americans found that common DNA variants in the FOXL2 gene are associated with eyebrow thickness. However, these variants have very low minor allele frequencies in East Asians and Europeans, suggesting that in these two populations, eyebrow thickness may well be affected by different genes.
To enhance our understanding of the genetic basis underlying the variation of human eyebrow thickness, we conducted a genome-wide association study in East Asians and Eurasians, followed by a trans-ethnic meta-analysis, to identify genetic variants that affect eyebrow thickness in humans. Moreover, in order to validate the findings from our association analyses, we performed fine mapping and conducted functional genetic experiments. Finally, we applied various statistical genetics methods to search for signals of positive selection in and around the identified candidate genes.
Our discovery GWAS included a total of 2961 subjects from the Taizhou Longitudinal Study [10] (TZL, Han Chinese). Replication cohorts were drawn from the Xinjiang Uyghur Study [11] (UYG, N = 721, Uyghur, an admixed East Asian-European population), the CANDELA study [8] (CANDELA, N = 2301, Latin Americans) and the Rotterdam Study [12,13] (RS, N = 4411, Northwestern Europeans; for sample characteristics see S1 Table). In all cohorts, eyebrow thickness was assessed on an ordered categorical scale using a self-developed photo numeric approach (S1 Fig). Inter-rater (for the CANDELA cohort, intra-rater) reliability was reasonable for all cohorts (Kappa: 0.48–0.66, S2 Table). There was a higher degree of diversity in eyebrow thickness in Uyghurs (D = 0.58, Gini-index) and Northern Europeans (D = 0.57) than in Han Chinese (D = 0.52) and Latin Americans (D = 0.49) (S1 Table). Eyebrows were significantly less thick in females than in males in TZL, UYG, and RS (PTZL = 7.23×10−70, PUYG = 7.31×10−33, PRS = 4.10×10−106; the CANDELA sample included only male subjects). Age was negatively correlated with eyebrow thickness in TZL, CANDELA, and RS (PTZL = 7.75×10−20, PCANDELA = 4.20×10−17, PRS = 1.33×10−10; the UYG sample included only young subjects between 18 and 24 years). The GWAS in Han Chinese was based on 7,119,456 SNPs; it identified association signals of genome-wide significance (P<5×10−8) for two genomic regions, located at 3q26.33 (rs1345417: P = 6.51×10−10) and 5q13.2 (rs12651896: P = 1.73×10−8; Fig 1). After conditioning on the genotypes of rs1345417 and rs12651896, no additional SNPs showed significant association at a genome-wide level (P<5×10−8). The heritability of eyebrow thickness in TZL was 37.6%. The SNPs rs1345417 and rs12651896 explain 3.11% and 2.55% of this heritability, respectively. Although the GWAS in the Uyghurs, which was based on 7,224,952 SNPs, did not detect any genome-wide significant signals (S2 Fig), the signals at 3q26.33 and 5q13.2 showed marginal significance (rs1345417: P = 3.78×10−4; rs12651896: P = 5.4×10−2), and allelic effects were consistent with those in TZL (Table 1).
The GWAS in Latin Americans has been published previously; it found that common variants at 3q22.3 are associated with eyebrow thickness [8]. We found that it also corroborated our findings at 3q26.33 and 5q13.2 (rs1345417: P = 1.04×10−7; rs12651896: P = 7.54×10−6). The meta-analysis of all three GWAS (TZL, UYG and CANDELA) identified four loci reaching genome-wide significance (P<5×10−8; Fig 2A; S3 Table). Apart from the loci described above 3q26.33 (rs1345417: P = 1.11×10−19), 5q13.2 (rs12651896: P = 2.52×10−13) and 3q22.3 (rs112458845: P = 2.24×10−9) there was one additional locus of genome-wide significance, at 2q12.3 (rs1866188: P = 5.81×10−11) (Fig 2). The respective quantile-quantile (Q-Q) plots showed no sign of inflation for any of the association tests described above, and the genomic control factor λ was relatively low in all cases (λTZL = 1.041, λUYG = 1.053, λCANDELA = 1.015, λMeta = 1.040, S3 Fig), indicating that the test statistics were not substantially confounded by population sub-stratification.
For all three novel loci (3q26.33, 5q13.2 and 2q12.3), the allelic effects acted in the same direction in all populations (Table 1). There was no significant effect size heterogeneity across populations for any of these three association signals (S3 Table). However, the signal at 3q22.3, which was highly significant in CANDELA, with P = 4.95×10−11 at rs112458845 and an effect allele frequency (EAF) of 0.127, did not reach nominal significance in our GWAS of TZL (EAF = 0.065, P = 0.729) and UYG (EAF = 0.032, P = 0.494). Furthermore, our meta-analysis revealed strong allelic heterogeneity for rs112458845 (Q = 16.284, I2 = 81.58%; S3 Table), indicating that the effect at 3q22.3 may be specific to Latin Americans.
We performed a further replication analysis for all SNPs described above in the Rotterdam Study sample, which comprises 4411 Dutch Europeans. With the exception of rs112458845 at 3q22.3, which was specific to Latin Americans, this analysis replicated all signals at a nominal significance level (rs1866188: P = 0.0141; rs112458845: P = 0.366; rs1345417: P = 4.03×10−3; rs12651896: P = 2.93×10−4; Table 1). The allelic effects at all associated SNPs acted in the same direction as in all other populations tested (Table 1).
Next, we sought to localize the variants driving the association signals at each of the novel loci that attained genome-wide significance in our trans-ethnic meta-analysis. We utilized PAINTOR [14,15] to calculate the Bayesian posterior causal probability for each variant included in the signal and identified the sets of variants that collectively explained 99% of the total probability (credible sets). The credible sets for 2q12.3 and 3q26.33 contained a single SNP each (S4 Table). The credible set for 5q13.2 comprised nine variants. Among these, the posterior causal probability was highest for rs12651896 (posterior probability = 0.744; S4 Table). Using CADD [16] and DeepSEA [17] to assess the functional consequences of each variant, we found that rs10061469 was consistently predicted to be functionally important (S4 Table). Our interrogation of ENCODE [18] and REMC [19] data revealed that the regions around rs1866188 and rs1345417 show distinct active enhancer signatures defined by epigenetic markers, such as the histone modifications H3K4me1 and H3K27ac, and DNase hypersensitivity in epithelial cells (Fig 3). The regions around rs12651896 and rs10061469 are involved in long-range chromatin looping interactions with FOXD1 (Fig 3B). Together, these data suggest the presence of putative regulatory regions in the neighborhood of these variants. We thus chose rs1345417, rs12651896, rs10061469 and rs1866188 for further experimental verification.
To further examine whether the four variants identified in our fine mapping analyses (rs1345417, rs12651896, rs10061469 and rs1866188) play a role in the expression of nearby genes, we conducted CRISPR/Cas9-mediated gene editing for each of them. We targeted the genomic regions surrounding each variant using two different sgRNAs (S4 Fig). Our qRT-PCR analysis of annotated transcripts near these SNPs found that mixed clones of A375 cells that were stably infected with rs1345417 targeting lenti-Cas9-sgRNAs displayed significantly reduced SOX2 expression and significantly elevated SOX2-OT expression compared to cells infected with control lenti-Cas9-sgRNAs (Fig 4A). Infection with rs12651896 targeting lenti-Cas9-sgRNAs led to a significant reduction in the expression of FOXD1 (Fig 4B), while infection with rs10061469 targeting lenti-Cas9-sgRNAs showed no effect on nearby genes (Fig 4C). Infection with rs1866188 targeting lenti-Cas9-sgRNAs resulted in a significant reduction in LIMS1 expression of (Fig 4D).
Among the genes showing significant changes in expression levels, only SOX2 and FOXD1 were reported to be related to hair growth [20,21]. We thus chose to focus on these two genes. For each of the two lenti-Cas9-sgRNAs targeting rs1345417 and rs12651896, we derived multiple independent single cell clones of infected A375 cells and characterized their exact deletion/mutation status at the target SNP sites (S4 Fig). We found that SOX2 expression was significantly reduced in A375 cell clones carrying deletions/substitutions encompassing rs1345417, in comparison with cells infected with empty lenti-Cas9 vector or with lenti-Cas9-control sgRNA (Fig 4E and 4F). Importantly, A375 cell clones that were infected with rs1345417 targeting lenti-Cas9-sgRNA but where the SNP site was not successfully edited, did not show reduced SOX2 expression (Fig 4E and 4F). Similarly, FOXD1 expression was significantly reduced in all A375 cell clones carrying a deletion encompassing the rs12651896 site, compared to both controls and unsuccessfully edited infected cell clones (Fig 4G and 4H). In one clone (896sg1m1), β-actin expression was also unexpectedly reduced, probably due to off-target effects.
To test if a single nucleotide substitution event is sufficient to affect endogenous SOX2 gene expression, we conducted a CRISPR/Cas9-mediated knock-in experiment at rs1345417, for which native A375 cells are homozygous (G/G). Consistent with our expectations, SOX2 expression levels were significantly lower in an edited, G/C heterozygous A375 clone than in the original G/G A375 cells (Fig 4I).
Additionally, we conducted a luciferase reporter experiment for the genomic region surrounding rs1345417 (S5 Fig). We found that a 1,495 bp genomic fragment encompassing the rs1345417 site and containing its common G allele can act as a transcriptional enhancer. Its insertion before a SOX2 promoter sequence led to 3 to 4-fold increase in reporter expression (S5 Fig). Moreover, a G>C mutation at rs1345417 on the reporter construct resulted in significantly reduced reporter activity, in line with our observations in the CRISPR/Cas9 experiments above (Fig 4I). Together, our data indicate that genetic variation at rs1345417 and rs12651896 can affect the transcription of SOX2 and FOXD1, respectively.
For some of the top associated loci, allele frequencies varied among populations (S6 Fig). This variation could be caused either by random genetic drift or by local positive selection. To assess whether positive selection could have helped shape variation in the genomic regions associated with human eyebrow thickness, we applied several statistical genetics methods. Using the Integrated Haplotype Score (iHS) [22] and the Composite of Multiple Signals (CMS) statistic [23], we did not find any evidence indicating that SOX2, FOXD1 and FOXL2 are subject to positive selection in East Asian (CHB), European (CEU) and African (YRI) populations (S7 and S8 Figs). However, there were highly significant signatures of strong positive selection in the EDAR region. This is not surprising, as our top signal near EDAR (rs1866188) is in high LD with rs3827760, a strongly selected functional SNP causing a number of ectodermal related phenotypic changes, as demonstrated previously [24]. Therefore, the signal of selection observed for rs1866188 may well be explained by the selective pressure on rs3827760. To test whether the differences in allele frequencies may stem from different local selection pressures, we applied a probabilistic approach described by He and colleagues [25] to quantify local inter-population differences in selection for the four top loci. Apart from rs1866188 at EDAR, we found no significant differences in selection for any other SNP in the associated regions (S9 Fig).
To test whether eyebrow thickness might be subject to polygenic selection, we applied the polygenic scores test developed by Berg and Coop [26]. This test measures the total frequency of associated alleles in a population, weighting each allele by its effect size. Loci that have undergone local positive selection will show greater divergence than expected under the neutral model. Previous studies using this method have found excess variance among populations for genetic scores associated to height, demonstrating that height is subject to local positive selection [27]. Here, we calculated the polygenic score for eyebrow thickness based on the top four associated variants. The resulting scores were similar between populations and showed no excess variance due to positive selection (P = 0.567, S10 Fig), indicating that the allele frequency differences observed among populations are better explained by random genetic drift than by selection. Our results thus provide no evidence that eyebrow thickness is under strong positive selection in human populations.
We found that eyebrow thickness is significantly associated with age and gender. As may be expected, older adults tend to have a significantly reduced eyebrow thickness due to the weakened biological function of older follicle cells. The eyebrows of males are generally thicker than those of females. Eyebrow plucking is widespread among females and may thus have affected our results; nevertheless, the large and consistent differences observed between genders are likely to be real, considering that androgens tend to stimulate hair growth [28,29].
The top associated SNP at 3q26.33, rs1345417, is located about 80 kb downstream of the SOX2 gene (Sex Determining Region Y-Box 2), an important transcription factor that is highly expressed in the dermal condensate (DC) and dermal papilla (DP) of growing follicles [30,31]. Interestingly, fine-tuning of SOX2 expression appears to play an important role in the regulation of hair thickness. In murine skin development, from E18.5 onwards, Sox2 expression becomes confined to DPs in thick guard/awl/auchene hairs, but not in thin zigzag hairs [20]. Moreover, conditional ablation of Sox2 from the DP resulted in significant reduction in the length of awl/auchene, but not zigzag hairs [20]. These findings are consistent with our hypothesis that the G>C mutation at rs1345417 may cause reduced eyebrow thickness by downregulating SOX2 expression. The top SNP within the second signal at 5q13.2, rs12651896, is located 242 kb downstream of FOXD1 (Forkhead Box D1). This gene belongs to the forkhead family of transcription factors, whose members are characterized by a distinct forkhead domain. FOXD1 is especially enriched in DC cells, a precursor of DP/dermal sheath niche cells within the mature follicle [21]. Finally, rs1866188 on chr2q12.3 is located 253 kb upstream of EDAR (Ectodysplasin A receptor), a gene which plays an important role in the development of ectodermal tissues such as hair, teeth, and sweat glands [24]. Previous studies have consistently found EDAR to be implicated in beard density [8] and hair thickness [24] and straightness [11,32,33].
It is notable that all of the signals identified by our GWAS fall into regulatory regions. To date, hundreds of GWAS have been conducted, resulting in the identification of a large number of genetic variants associated with common diseases and phenotypic variation. The majority (~93%) of variants associated with common traits lie within non-coding sequences, complicating their functional evaluation [34]. Recent studies have found that these variants are concentrated in regulatory DNA. This remains true even after adjusting for microarray SNP ascertainment bias and suggests that non-coding variants may act by causing changes in gene expression levels [34], as supported by several lines of evidence [35–39]. The association signals found here at 3q26.33 (rs1345417), 5q13.2 (rs12651896), and 2q12.3 (rs1866188) are all located within active enhancer regions characterized by epigenetic markers, such as H3K4me1 and H3K27ac histone modifications, and DNase hypersensitivity in epithelial cells. Our functional validation experiments provided evidence for enhancer activity around rs1345417 and rs12651896. It is thus highly plausible that these variants affect eyebrow thickness by regulating the expression of SOX2 and FOXD1, respectively. In particular, the CRISPR/Cas9-mediated knock-in experiment provided direct evidence that a single substitution at rs1345417 is sufficient to affect the endogenous gene expression of SOX2. Our study represents a successful example of how GWAS and CRISPR/Cas9 technology can be combined to demonstrate the involvement of non-coding variants with regulatory functions in common diseases and normal phenotypic variation.
The FOXL2 variant rs112458845 has previously been found to be associated with eyebrow thickness in Latin Americans. Here, we found no significant evidence of association of this variant with this phenotype in East Asians or Europeans. This may partly be due to the frequency distribution of the minor allele, which is rare in East Asians (4%) and absent in Europeans (0%), but relatively common in Native Americans (26%). Additionally, and perhaps more importantly, allelic effect sizes may be population specific due to a unique yet unidentified environmental or genetic background and may thus vary between Latin Americans on the one hand, and East Asians and Europeans on the other. A similar impact of population heterogeneity on allelic effect sizes has been reported for other traits, such as follicular lymphoma [40], body bone mineral density [41], and Type 1 diabetes [42].
It has long been debated whether eyebrow thickness is a selected trait or a ‘neutral feature’ with no apparent link to individual fitness [3,6,7]. We used several methods to detect potential signatures of positive selection for the variants associated with eyebrow thickness. First, we tested whether any of the associated regions overlapped with signals of positive selection. We then used a recently developed probabilistic method to test and estimate differences in selection of the associated variants between populations. We further tested the presence of polygenic selection by examining subtle allele frequency shifts across multiple loci. Previous studies using the above methods have found signatures of positive selection for a range of human traits, including height, skin color, and BMI [23,25,26,43]. However, apart from the EDAR region around rs3827760, a SNP known to be under strong selection, we found no significant signatures of positive selection for variants associated with eyebrow thickness. These results suggest that eyebrow thickness may not be subject to strong positive selection, at least not via the genes identified here. In this context, it is worth noting that most studies postulating sexual selection of the eyebrow were based on reconstructed attractiveness [3,5,7], and it may be argued that the perception of attractiveness can vary significantly over time [3]. Therefore, while sociological studies have indicated that eyebrow thickness may be subject to sexual selection, our study does not provide any support for this conclusion from an evolutionary or genetic perspective.
In conclusion, we identified three novel genetic variants near SOX2, FOXD1 and EDAR that influence eyebrow thickness. We demonstrated that rs1345417 and rs12651896 affect the transcriptional activity of the nearby SOX2 and FOXD1 genes. Furthermore, we found evidence for population heterogeneity in the genetics of eyebrow thickness. Finally, our results suggest that eyebrow thickness may not be subject to strong positive selection.
The Taizhou Longitudinal Study was carried out following protocols approved and oversight by the Institutional Research Board at Fudan University (Ethics Research Approval No.85). The Xinjiang Uyghur Study was conducted with the official approval of the Ethics Committee of the Shanghai Institutes for Biological Sciences (ER-SIBS-261410). The CANDELA ethics approval was obtained from: Universidad Nacional Autonoma de Mexico (Mexico), Universidad de Antioquia (Colombia), Universidad Peruana Cayetano Heredia (Peru), Universidad de Tarapaca (Chile), Universidade Federal do Rio Grande do Sul (Brasil) and University College London (UK, approval number 3351/001). The Rotterdam Study has been approved by the Medical Ethics Committee of the Erasmus MC (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG). All participants provided written informed consent in these four studies.
This study is based on data from four populations (S1 Table): Han Chinese, Uyghurs, Latin Americans and Europeans. 2,961 Han Chinese (including 1,060 males and 1,901 females, with an age range of 31–87) were enrolled in Taizhou, Jiangsu Province, as part of the Taizhou Longitudinal Study (TZL) [10], in 2014. 721 Uyghurs (including 282 males and 439 females, with an age range of 17–25) were enrolled at Xinjiang Medical University, Urumqi, Xinjiang Province, China, as part of the Xinjiang Uyghur Study (UYG) in 2013–2014. Additionally, we collected summary statistics for a previously conducted GWAS on eyebrow thickness in Latin Americans of the CANDELA cohort, details of which have been previously published [8]. The Rotterdam Study is a population-based prospective study including a main cohort (RS1) and two extensions (RS2 and RS3) [12,13]. All participants were examined in detail at baseline. Collection and purification of DNA have been described in detail previously [12]. The Rotterdam Study has been entered into the Netherlands National Trial Register (NTR; www.trialregister.nl) and into the WHO International Clinical Trials Registry Platform (ICTRP; www.who.int/ictrp/network/primary/en/) under shared catalogue number NTR6831. All participants provided written informed consent to participate in the study and to have their information obtained from treating physicians.
Eyebrow thickness (i.e., density) was rated by eye on a three-point scale (TZL, UYG, RS: scarce, normal, and dense; CANDELA: low, medium and high), following an established standard and based on photographic imagery (S1 Fig). For the TZL and UYG cohorts, each case was rated independently by two different investigators. The inter-rater reliability was evaluated with the Kappa statistic. Cases where ratings were inconsistent were reviewed by a third investigator, who made the final decision. For the CANDELA cohort, an interview of the volunteers had indicated that most women modified their eyebrows; in this cohort, eyebrow thickness was therefore only scored in men. Phenotyping was performed by an experienced investigator, and intra-rater reliability was assessed for 150 individuals. For RS, three investigators simultaneously and independently evaluated all photos, which were displayed on two identical screens with the same settings. Before evaluation, a total of 50 photos were openly discussed to reach a consensus between the three investigators. Inter-rater reliability was also evaluated with the Kappa statistic. The average score from the three independent ratings was used as the final phenotype in all subsequent analysis.
For TZL and UYG, blood samples were collected, and DNA was extracted. All samples were genotyped using the Illumina HumanOmniZhongHua-8 chip, which interrogates 894,517 SNPs. To control for genotype quality, we used PLINK 1.9 [44] to exclude individuals with more than 5% missing data, related individuals, and those that failed the X-chromosome sex concordance check, or for whom the available information on ethnicity was incompatible with their genetic information. We also excluded SNPs with more than 2% missing data, with a minor allele frequency (MAF) below 1%, and those that failed the test for Hardy-Weinberg equilibrium (P<1×10−5). The chip genotype data were phased using SHAPEIT [45], and IMPUTE2 [46] was then used to impute genotypes at non-genotyped SNPs using variant positions from the 1000 Genomes Phase 3 data as a reference. SNPs with an imputation quality score (INFO) below 0.8 or a MAF below 1% were eliminated from further analyses. For the Han Chinese population, a total of 6,343,243 imputed SNPs passed quality control and were combined with 776,213 genotyped SNPs for association analysis. For the Uyghur population, a total of 6,414,304 imputed SNPs passed quality control and were combined with 810,648 genotyped SNPs for further analyses. In RS1 and RS2, genotyping was carried out using the Infimum II HumanHap550K Genotyping Bead Chip version 3, which contains 6,787,905 probes. Complete information on genotyping protocols and quality control measures for RS1 and RS2 have been described previously [47,48]. In RS3, genotyping methods closely followed those established for RS1 and RS2, but a denser array, the Human 610 Quad Arrays of Illumina with 15,880,747 probes, was used. Individuals with a call rate < 97.5%, gender mismatch with typed X-linked markers, or excess autosomal heterozygosity (>0.33) were excluded, as were duplicates or 1st degree relatives identified using IBS probabilities, and outliers using multi-dimensional scaling analysis with reference to the 210 Hap Map samples. Genome-wide imputation in RS3 closely followed the methods used in RS1 and RS2, as described in detail previously [48]. Genotypes were imputed using MACH [49] based on phased autosomal chromosomes of the 1000 Genome reference panel, orientated on the positive strand.
The effects of possible population stratification were corrected using the EIGENSTRAT [50] tool from the EIGENSOFT package. To this end, TZL and UYG data were combined with 1000 Genomes Phase 3 data for YRI, CHB and CEU populations. 102,284 SNPs in low linkage equilibrium (r2<0.2) were selected for analysis. Principal component (PC) analysis did not find any outliers in TZL and UYG (S11 Fig).
Initial genome-wide association tests using multiple linear regression with an additive genetic model incorporating gender, age and four genetic PCs as covariates were performed in PLINK 1.9 [44]. Expected and observed association results for all tests were visualized in quantile-quantile (Q-Q) plots to assess systematic inflation in association resulting from population stratification or other systematic causes of bias. None of the Q-Q plots showed any sign of inflation, the genomic control factor λ being < 1.06 in all cases (S3 Fig). To evaluate the presence of additional independent signals at each locus, we performed conditional analyses by adding the dosages of the top SNP at each locus to the regression model. Q-Q, Manhattan and regional association plots [51] were created in R. The proportion of variance in eyebrow thickness explained by the genetic variants identified was estimated using GCTA [52]. The meta-analysis of the TZL, UYG and CANDELA data sets was performed using METAL [53]. Heterogeneity of SNP associations across studies was tested via Cochran’s Q statistic [54], and its magnitude was expressed by I2. For SNPs with significant heterogeneity, a random effects model was applied for meta-analysis using METASOFT [55].
We performed fine mapping of each locus for a 1 Mb genomic interval flanking the top SNP (500 kb upstream and 500 kb downstream) using PAINTOR [14,15]. For each SNP within this 1 Mb region, the posterior probability that this SNP is driving the region’s association signal was calculated by dividing the SNP’s Bayes factor (BF) by the sum of the BFs of all SNPs in the region [56]. A 99% credible set was then constructed by (1) ranking all variants according to their Bayes factor, and (2) including ranked variants until their cumulative posterior probability of representing the causal variant at a given locus exceeded 0.99 [57].
To further facilitate the prioritization of variants for functional analysis, we used CADD [16] (Combined Annotation-Dependent Depletion) and DeepSEA [17] (deep learning-based sequence analyzer) to evaluate the possible functional consequences of the variants in the 99% credible set.
Candidate genes were chosen based on their distance to the associated loci as well as their function, involvement in biochemical pathways, tissue expression, and involvement in similar phenotypes. The relevant information was obtained from NCBI [58] and Ensemble [59], as well as available published data. We used HaploReg v4.1 [60] to extract a variety of regulatory annotations, including histone modification (ChIP-seq tracks), chromatin state segmentations (15-state) and ChIA-PET [61] (Chromatin Interaction) from ENCODE [18] and the Roadmap Epigenomics Project [19], conserved regions from GERP [62] and Phastcons [63], and eQTLs from the GTEx [64] and GEUVADIS databases [65].
No cell lines were found in the database of commonly misidentified cell lines maintained by ICLAC and NCBI Biosample. Cell lines were not authenticated. NHEM (human melanocytes) and A375 (human melanoma) were purchased from the cell bank of the Chinese Academy of Sciences. HACAT (immortal keratinocytes) and SCC13 (skin squamous-cell carcinoma cell line) were purchased from ATCC. Cell lines were routinely tested for mycoplasma infection. Cells were cultured using DMEM+10%FBS+1%Penicillin-Streptomycin.
A 1,495 bp genomic fragment comprising the rs1345417 enhancer was amplified from human genomic DNA and cloned into a pGL3-promoter vector, in which the SV40 promoter was replaced by a SOX2 promoter. The G>C mutation was introduced by site-directed mutagenesis (Takara). The inserts in each construct were verified by sequencing. The detail information of primer sequences can be found in S5 Table. Constructs were transfected with equimolar amounts (500 ng) of luciferase reporter plasmids into A375 Melanoma cells using jetPEI (Polyplus), according to the manufacturer’s instructions. Luciferase expression was normalized to 200 ng Renilla luciferase expression (pRL-SV40). Cells were harvested after 48 h. Luminescence activity was measured with a Berthold Centro LB 960 Microplate Luminometer. Data represent at least three independent experiments. Student’s two-tailed t-test was used to determine statistical significance.
For each candidate variant, two sgRNAs were designed, cloned into the lentiCRISPR v2 vector, and packaged into lentivirus as previously described [66]. sgRNAs used for rs1345417: sgRNA1 (CCTGCTTTTGCCTCAGCCCACAT) and sgRNA2 (CCCACATCTTCTCTATTAGTAAG). sgRNAs used for rs12651896: sgRNA1 (CAAAATGTTCTTGCTAGCATATCCA) and sgRNA2 (GCATATCCATAACTAGCACAGG). sgRNAs used for rs10061469: sgRNA1 (ACAACCTGCAATAAACTATTAA). sgRNAs for rs1866188 site: sgRNA1 (GAGTGGCCACTCTCTTTTGC) and sgRNA2 (GAATGCATAAGGA TCAAATCG). Scramble Control sgRNA sequence is (CCCACATAGTCTCACTTAG TAAG).
For target site deletion via lentiCRISPR-sgRNA virus infection, stably infected cells were selected on puromycin. For clone selection, stably infected cells were diluted to allow colonial growth. Single colonies were individually picked for DNA sequencing of the target site. Because protospacer adjacent motives were located as far as 100 bp upstream and 22 bp downstream, the deletion at rs1866188 was considerably larger than for the other sites.
For G>C substitution at rs1345417 in A375 cells, we used a CRISPR-Cas9 mediated knock-in strategy. Specifically, in order to substitute the G/G of rs1345417 in A375 cell, a ~1000bp DNA fragment encompassing the rs1345417 site from a Chinese individual with a C/C homozygote genotype was PCR amplified and TA cloned into the pMD18T Vector (Takara) to construct a C-donor plasmid. We then co-transfected the C-donor plasmid and the lenticrispr-rs1345417-sgRNA2 plasmid into A375 cells using the Effectene reagent (Qiagen). 48 h after transfection, the cells were selected on puromycin for 48 h to eliminate untransfected cells. Successfully transfected cells were then diluted for colonial growth. Single colonies were individually picked for DNA sequencing to screen for substitution at the target site. In total, 44 clones were screened.
Sequencing of the rs1345417 site: Genomic DNA was extracted from each cell clone using ZR Genomic DNA-tissue MiniPrep (Zymo) following the manufacturer’s protocol. The rs1345417 genomic region was amplified from genomic DNA by nested PCR and TA-cloned into pMD18 T vector for sequencing. At least four individual TA-clones were sequenced for each cell clone. RNA extraction was conducted using Direct-zol RNA MiniPrep Plus (Zymo) following the manufacturer’s protocol. Reverse transcription was conducted using the iscript cDNA synthesis kit (Biorad) following the manufacturer’s protocol. Real-time PCR experiments were conducted on a VIIA7 Fast Real-Time PCR system (Applied Biosystem) using the iTaq universal SYBR Green supermix (Biorad). All statistical analyses were conducted using Microsoft Excel and the GraphPad Prism 6 software.
Genomic characteristics resulting from strong recent positive selection include low haplotype diversity and high linkage disequilibrium. We calculated extended haplotype homozygosity (EHH) [22,67] for all SNPs until EHH < 0.05 in CEU, CHB and UYG samples. Next, the integrated haplotype score (iHS) [68] was calculated for all SNPs, with an allele frequency bin of 0.05 to standardize iHS scores against other SNPs of the same frequency class within the region. Finally, we calculated P values assuming a Gaussian distribution of iHS scores under the neutral model, which was checked by plotting the values against a Gaussian distribution. The empirical significance cutoff was based on the top 0.1% iHS scores. We also performed genome-wide CMS [23] analysis in African (YRI), European (CEU), and East Asian (JPT+CHB) populations from the 1000 Genome Project [69] to validate our results. The empirical significance cutoff was based on the top 0.1% CMS scores. Differences in allele frequencies are indicators of possible differences in selection between populations. To test whether the differences in allele frequencies may result from different local selection pressure, we used a probabilistic method which was recently put forward by He and colleagues [25]. We used this approach to test and estimate selection differences between populations for the four top associated loci. We first estimated differences in selection coefficients between populations (CHB, CEU and YRI) using logarithmic odds ratios of allele frequencies. The variance of the estimation was then calculated based on genome-wide variants. Finally, we calculated P values assuming a Gaussian distribution of the statistic under the neutral model. The significance cutoff was P<0.005 after multiple testing correlation.
To investigate whether the loci associated with eyebrow thickness are more differentiated among populations than expected under neutral genetic drift, for each population m, we calculated the polygenic eyebrow thickness score (genetic score) as
Zm=2∑i=1Lβlpml
where βl is the effect size of the eyebrow thickness increasing allele l, and pml is the frequency of allele l in population m. We first used the four loci identified here in conjunction with allele frequency data from the 1000 Genome Project dataset to estimate the genetic score for eyebrow thickness in each population, with the effect sizes estimated in the meta-analysis. To test whether there was a signature of polygenic adaptation, we then adopted a framework developed by Berg and Coop [26], which builds a multivariate normal model based on matched, presumably neutral variants, to account for relationships among populations. Traits that have undergone local selection will show excess divergence among populations (significance cutoff: P<0.05).
|
10.1371/journal.pgen.1007185 | Allocation of distinct organ fates from a precursor field requires a shift in expression and function of gene regulatory networks | A common occurrence in metazoan development is the rise of multiple tissues/organs from a single uniform precursor field. One example is the anterior forebrain of vertebrates, which produces the eyes, hypothalamus, diencephalon, and telencephalon. Another instance is the Drosophila wing disc, which generates the adult wing blade, the hinge, and the thorax. Gene regulatory networks (GRNs) that are comprised of signaling pathways and batteries of transcription factors parcel the undifferentiated field into discrete territories. This simple model is challenged by two observations. First, many GRN members that are thought to control the fate of one organ are actually expressed throughout the entire precursor field at earlier points in development. Second, each GRN can simultaneously promote one of the possible fates choices while repressing the other alternatives. It is therefore unclear how GRNs function to allocate tissue fates if their members are uniformly expressed and competing with each other within the same populations of cells. We address this paradigm by studying fate specification in the Drosophila eye-antennal disc. The disc, which begins its development as a homogeneous precursor field, produces a number of adult structures including the compound eyes, the ocelli, the antennae, the maxillary palps, and the surrounding head epidermis. Several selector genes that control the fates of the eye and antenna, respectively, are first expressed throughout the entire eye-antennal disc. We show that during early stages, these genes are tasked with promoting the growth of the entire field. Upon segregation to distinct territories within the disc, each GRN continues to promote growth while taking on the additional roles of promoting distinct primary fates and repressing alternate fates. The timing of both expression pattern restriction and expansion of functional duties is an elemental requirement for allocating fates within a single field.
| A battery of transcription factors collectively called the retinal determination (RD) network controls the earliest steps in the specification of the fruit fly compound eye. Loss-of-function mutations lead to the loss of the compound eyes while over-expression of RD network members in non-retinal tissues induces the formation of ectopic eyes. These observations suggest that the network governs the growth, specification, and patterning of the eye field. Recent studies have also shown that the RD network represses the fates of the non-ocular tissues that are also derived from the disc such as the antenna, maxillary palp, and head epidermis. One inconsistency in the model for how this network controls eye specification is that many of its members are expressed throughout the entire eye-antennal disc. In this study, we show that early in development, the RD network is expressed throughout and promotes the growth of the entire eye-antennal disc. After the initial growth phase, the expression of these genes is restricted to just the eye field. This temporal and spatial limiting of the RD network to the developing eye is essential so that its role can expand to include promoting eye specification and repressing non-ocular fates.
| In organisms as diverse as flies and humans, it is common for multiple adult tissues/organs to originate from adjacent territories within a uniform precursor field. For example, the developing anterior forebrain in vertebrates gives rise to the eyes, hypothalamus, diencephalon, and telencephalon [1,2]. Likewise, the developing wing disc of Drosophila produces the adult wing, hinge, and thorax [3]. In these instances, it has been shown that gene regulatory networks (GRNs), consisting of signaling pathways and transcription factors, subdivide the precursor field and then separately specify the fate of the distinct territories that are contained therein. The GRNs that are involved in this process are thought to promote growth and specification of each tissue while acting to block tissues from adopting the wrong fate. This ensures that each organ reaches the appropriate size, adopts the correct fate, is positioned properly within the body plan, and has all the requisite cell types. This simple and straightforward model for establishing multiple fates from a single homogeneous precursor is at odds with observations that many GRN members, including master control genes, are expressed broadly within the precursor field. It is difficult to envision how GRNs function to allocate tissue fates if their members are uniformly expressed and competing with each other within the same populations of cells. We have examined the role that the retinal determination (RD) network plays in the Drosophila eye-antennal disc, and our results provide a mechanistic solution to this unresolved conflict.
During Drosophila embryogenesis, several different populations of cells coalesce to form the eye-antennal disc [4]. This monolayer epithelium, which initially is uniform, eventually gives rise to a number of head structures, including the visual system (compound eyes and ocelli), the olfactory system (antennae and maxillary palps), and the surrounding head epidermis. During the earliest stages of development, the eye-antennal disc does not have any obvious physical markings that would distinguish one region from another, nor are there many instances of genes being expressed in discrete patterns. In fact, a large number of genes that occupy very restricted patterns later in development are first ubiquitously expressed within the disc. As development proceeds, physical landmarks begin to distinguish the major regions of the eye-antennal disc from one another, while gene expression and protein distribution patterns can identify much more discrete domains. Indeed, mosaic clone analysis and transplantation experiments using disc fragments have identified the physical locations of each tissue within the late third larval instar disc [5–9].
Many members of the RD network are expressed throughout the entire disc before being segregated to the eye field. Two poignant examples are the Drosophila Pax6 genes, eyeless (ey) and twin of eyeless (toy). Starting at stage 15 of embryogenesis and continuing through the first larval instar, both ey and toy are expressed throughout the entire eye-antennal disc [10,11]. The simultaneous removal of these proteins early in development, when they are universally expressed in the eye-antennal disc, leads to the elimination of the entire disc and all associated head structures that are derived from the disc [12,13]. At the start of the second larval instar, the expression of ey and toy are restricted to the developing eye field [10,11,14]. If Ey/Toy are necessary for tissue proliferation early in development, then why is expression of both these genes lost from the antennal and head epidermal portions of the disc even as both domains continue to proliferate and grow? The answer may lie in the results from experiments involving the targeted expression of ey/toy in the antennal field; in these situations, ectopic eyes are induced [14,15]. It thus appears that Pax6 plays two roles in development: cell proliferation and tissue specification. The changing expression pattern of Pax6 genes over time is critical for separating the two functions of Pax6 in both time and space and ensures that development of the eye-antennal disc and formation of the entire fly head occurs correctly.
To better understand the mechanisms of how the RD network controls tissue specification we analyzed other members that are first expressed throughout the entire eye-antennal disc but are then restricted to just the eye field later in development. Of the many genes that fulfilled this criterion, we focused on teashirt (tsh) and its paralog tiptop (tio) for several reasons. First, the proliferation of the entire eye-antennal disc early in development is partially dependent upon Tsh [12]. Second, both Tsh and Tio induce ectopic eyes in the antenna when over-expressed [16–18]; therefore, their restriction to the eye field later in development is required for proper head formation. And lastly, although tsh and tio arose through a duplication of an ancestral gene, their temporal/spatial expression patterns are different, and they have distinct protein structures [19–21]. Therefore, we have an opportunity to functionally dissect the roles that these two key factors play within the eye-antennal disc.
Here we show that in the nascent eye-antennal disc Tsh and Toy cooperate to promote cell survival throughout the disc. The simultaneous loss of both factors leads to massive increases in apoptosis, which, in turn, leads to the complete loss of the eye-antennal discs. The adults are headless and die during the pharate pupal stage. The role for Tsh in early disc growth has remained hidden up to this point and is only revealed when Toy protein levels are simultaneously reduced. After the initial growth phase, tsh expression is restricted to the eye field. This sequestration is necessary since prolonged expression within the antennal field induces ectopic eye and leg formation [16]. We demonstrate here that during the induction of ectopic eyes Tsh not only initiates expression of retinal selector genes such as ey, sine oculis (so), eyes absent (eya), and dachshund (dac) but it also independently represses the expression of antennal/head epidermal selector genes such as cut (ct), Lim1, aristaless (al), and spineless (ss). This supports a model in which the RD network promotes the specification of the eye, in part, by “playing defense” through the suppression of alternate non-ocular fate choices. Forced expression of Tio within the antennal disc leads to similar changes in selector gene expression and tissue fate specification. However, unlike tsh, tio is never expressed within the antennal field but is always restricted to the undifferentiated cells of the developing eye. Thus, our data indicate that while Tsh/Tio proteins are functionally redundant later in the eye [17,18], sub and neo-functionalization at promoter/enhancer elements are likely to have played a key role in how these genes differentially affect early eye-antennal disc development.
Our findings shed new light on the role that the RD network plays in the development of the eye-antennal field. The RD network first promotes the growth of the entire precursor field before being rewired to promote specification of the retina. This shift in function necessitates a change in expression so that retinal development is confined to the eye field. As many other precursor fields in both flies and vertebrates are parceled out to produce multiple distinct organs, it is likely that similar mechanisms are in place to ensure the fidelity of growth, specification, and patterning.
The RD network controls both tissue fate specification and cell proliferation in the Drosophila eye (Fig 1A and 1B). Tsh, an important member of that network, is expressed throughout the entire nascent eye-antennal imaginal disc before being segregated to the eye field later in development. In the developing eye, Tsh has been implicated in driving both fate specification and cell proliferation [12,22–24]. These conclusions have relied mainly on the use of forced expression assays and protein-protein interaction studies because, due to several technical hurdles, the examination of eye development in tsh null mutants has proven to be quite difficult. As a result, there is little understanding of the molecular and developmental roles that Tsh plays during tissue specification and growth. Also, virtually nothing is known about the role that Tsh plays in the nascent disc, prior to the point of parceling of the eye-antennal disc into different individual domains.
To gain insight into the role that Tsh plays in the eye-antennal disc we used a set of RNAi lines to knockdown tsh expression during different stages of development. For this study, we used the Dorsal Eye (DE)-GAL4 line, which is an insertion in the (mirr) locus [25]. Early in development, DE-GAL4 drives expression in all cells of the eye-antennal disc, but later in development, its expression is restricted to the dorsal half of the eye and a subset of peripodial cells (S1A and S1B Fig) [12]. The use of this GAL4 line allows us to remove tsh from the entire eye-antennal disc during embryogenesis and the first larval instar during the period that tsh is normally expressed throughout the entire disc. Also, it allows us to, later in development, remove tsh from just the eye and compare Tsh negative (dorsal) and Tsh positive (ventral) tissue to each other within the same disc. We extended our study to include Tio since it is a paralog of Tsh and it is functionally redundant to Tsh in several developmental contexts [17,18,20,21]. We also targeted Ey and Toy since Tsh mediates a portion of their growth promoting activity [12].
We first set out to demonstrate the efficacy of the RNAi lines that we are using to target each candidate gene. When DE-GAL4 drives individual RNAi lines during the third larval instar, all four proteins are eliminated from the dorsal half of the eye disc (S1C–S1J Fig) [12]. We also examined transcript levels in late third larval instar eye-antennal discs and, as expected, we see a reduction in expression levels of approximately 50% for toy, tsh and tio (S1K Fig). Unexpectedly, ey transcript levels appear unchanged (S1K Fig). We assume that there must be compensation of gene expression within the ventral half of the eye disc through a mechanism that we do not yet understand. Tsh and Tio are thought to mutually repress one another’s expression [17]; therefore, we also looked at expression levels of tsh when tio is knocked down, and vice versa. There is a 10% increase in tsh expression when tio is knocked down, and a 32% increase in tio expression when tsh is knocked down (S1K Fig). Together, these controls demonstrate that the RNAi lines are efficient in eliminating expression of the target genes and that the transcription of tsh and tio is regulated as predicted by genetic studies.
Our initial results indicate that reducing ey, toy, tsh, or tio individually within the DE-GAL4 expression patterns has minimal, if any, effect on eye development (S1C, S1E, S1G and S1I Fig). Consistent, with this finding, transcript levels of the remaining RD genes are not reduced significantly (results are shown just for so S1K Fig). While it may seem surprising that the individual loss of these RD genes does not have a phenotype when removed with DE-GAL4, similar findings have been reported previously, and it has been demonstrated that the combined loss of multiple RD genes with this driver will reveal phenotypes that were absent with the single gene knockdowns [12]. Since genetic and biochemical interactions between the ey/toy and tsh/tio gene pairs exist, we decided to express all possible pairwise combinations of RNAi lines (toy/tsh, ey/tsh, toy/tio, ey/tio).
The concurrent reduction of Toy and Tsh eliminates the eye-antennal discs and results in headless adults that die during the pharate stage of pupal development (Fig 1C–1H). This is the very first evidence linking Tsh to early eye-antennal disc development. The headless phenotype is identical to the combined loss of Ey and Toy, and confirms our earlier suggestion that these Pax6 genes and Tsh cooperate to regulate growth of the eye-antennal disc gene [12]. Since our earlier report showed that tsh expression is lost when Toy and Ey are eliminated, we suggest that Tsh is functioning downstream of Toy and Ey. Surprisingly, the loss of the eye-antennal discs is specific to the Toy/Tsh knockdown combination, as the Ey/Tsh, Ey/Tio, and Toy/Tio dual knockdowns do not appear to affect eye-antennal disc development (Fig 1I–1K). Loss-of-function toy mutants exist and these show defects in the eye-antennal disc [12,26]; therefore, we are confident that the toy RNAi lines are behaving as expected. In contrast, eye development in tsh null mutants has not been characterized therefore we needed to express the tsh RNAi line in another tissue and confirm that the knockdown phenotype is similar to that of the tsh loss-of-function mutants. To accomplish this task, we expressed the tsh RNAi line within developing wing discs, and reassuringly the adult wings display an outstretched posture that is similar to the aeroplane-like allele of tsh (S4A Fig) [27,28]. Since the toy and tsh RNAi lines knock down the expression levels of the target genes and induce the expected loss-of-function phenotypes, our findings indicate that Toy and Tsh cooperate to control the development of the entire eye-antennal disc.
A previous study has proposed that Tsh is part of a tertiary biochemical complex that includes Ey and Homothorax (Hth) and controls cell proliferation of the developing eye field after segregation of the entire disc [22]. Here we are showing that Tsh and Toy may, together, control additional aspects of eye-antennal disc development. Our model is that Toy and Tsh are functioning prior to the segregation and rise of the different tissues within the eye-antennal disc. Since removal of Ey and Tsh together early in development doesn’t appear to have an effect on the eye-antennal disc, we propose that the Ey-Tsh-Hth complex may be working at a later stage of eye development. To make this determination, we introduced a temperature sensitive GAL80 construct [29] into the DE-GAL4>UAS-tsh RNAi, UAS-toy RNAi fly strain. GAL80 is an inhibitor of GAL4 activity, and the temperature sensitive nature of the GAL80ts construct allows us to use temperature to modulate the activity of GAL4 temporally and in turn regulate the timing of toy/tsh RNAi expression. At 18°C (the permissive temperature), GAL80 inhibits the activity of GAL4, thereby silencing the RNAi lines and allowing for normal expression of toy and tsh. Animals kept at this temperature throughout development emerge as adults that are indistinguishable from wild type. In contrast, GAL80ts is non-functional at 30°C (the restrictive temperature), which allows for the expression of the RNAi lines and the suppression of both genes. At this temperature, larvae lack eye-antennal discs, and the pharate adults are completely headless. By toggling between the permissive and restrictive temperatures, we can temporally control toy and tsh expression and determine when both genes are required together for disc development. When going back and forth between different temperatures, we had to take into account the dynamics of endogenous protein loss and recovery. We have determined that once animals have been shifted to 30°C and RNAi expression is initiated, it takes roughly 8hrs for Toy [12] and 12hrs for Tsh proteins to be erased from the disc (S2A–S2D Fig). Once flies are returned to 18°C, and the expression of RNAi lines has ceased, it takes approximately 40hrs for Toy [12] and 38hrs for Tsh (S3A–S3F Fig) to return to wild type levels.
We first kept animals at the permissive temperature (18°C) for varying periods of time before shifting to the restrictive temperature (30°C). Depending on when Toy/Tsh are removed, we see a range of phenotypes with headless flies on one extreme and wild type looking flies on the other (Fig 2A–2L). The percentage of flies with more severe defects increases in experiments where the removal of Toy/Tsh begins earlier in development (Fig 2M). We conducted the opposite experiment, in which animals were first kept at the restrictive temperature (30°C) for varying lengths of time before returning them to the permissive temperature (18°C). Again, depending upon the time at which Toy/Tsh proteins are restored, we see a range of mild to severe phenotypes (Fig 3A–3L). The complete loss of the eye-antennal discs is most often associated with longer knockdown periods of toy/tsh levels (Fig 3M). After taking into consideration the effect that different temperatures have on developmental rates [30], as well as the time that it takes for Toy/Tsh proteins to either be depleted or recovered, our findings suggest that the critical window for Toy/Tsh control of eye-antennal disc development is between stage 16 of embryogenesis and the end of the first instar (Figs 2M and 3M).
In order to confirm this timeline, we simultaneously knocked down toy/tsh using several different GAL4 drivers that initiate their expression at various times in development. Expression of the toy/tsh RNAi lines with ey-GAL4, which is first activated during embryogenesis, also results in larvae lacking eye-antennal discs and pharate stage pupae that are headless (Fig 4A–4C, 4G and 4J). This effect is synergistic, as expression of the individual RNAi lines with ey-GAL4 has much milder effects on the eye and head (S4B–S4E Fig). In contrast, removing Toy/Tsh simultaneously or individually with a tio-GAL4 driver, whose expression in the eye begins during the mid-second larval instar, does not affect the eye-antennal disc or the adult head (Fig 4D–4F, 4H and 4J, S4H–S4K Fig). tio-GAL4 also drives expression in regions within the leg disc that will give rise to the adult pleura and coxa (S4L Fig). If Tsh is removed from these regions (tio-GAL4, tsh RNAi), then the leg discs are duplicated (S4M–S4P Fig). The resulting adult legs are considerably smaller than their wild type counterparts and are often internalized within the adult fly (S4N–S4P Fig). This phenotype is consistent with the known role of Tsh in Drosophila leg development [31,32]. Based on the leg phenotypes associated with the tio-GAL4, UAS-tsh RNAi strain, the lack of observable phenotypes in the eye-antennal disc is due to the late onset of the tio enhancer.
We also tried to remove Toy/Tsh with an eya-GAL4 driver, which begins its expression at the start of the second larval instar (Fig 4I) [33]. While individual knockdown of toy with this driver does not affect the disc (S4F and S4G Fig), the knockdown of tsh resulted in embryonic lethality, therefore, we cannot draw any conclusions from this experiment. Unfortunately, there are no other suitable drivers to further test the temporal requirements of tsh/toy. However, the results from DE-GAL4, ey-GAL4, and tio-GAL4 are consistent with the GAL80ts time course experiments and indicate that Toy/Tsh function together during late embryogenesis and through the first larval instar to promote growth of the eye-antennal disc. This timeline for Toy/Tsh is consistent with the critical period for Ey/Toy function and is consistent with data that places tsh genetically downstream of both Pax6 genes [12]. Our findings are also consistent with a role for Tsh in promoting growth as proposed by [22], where Tsh is thought to work with Ey and Hth to keep the cells in proliferation. An important difference is that these two studies are addressing how Tsh interacts with distinct Pax6 proteins and at different times in development.
We then used the flp-out over-expression system to generate mosaic clones in which toy and tsh RNAi lines were simultaneously expressed in smaller cell populations. This allows us to measure the effect that the loss of both genes has on tissue growth. The size of the toy/tsh double knockdown clones and clones in which each gene had been knocked down individually were compared to GFP control clones. The growth of each type of clone was examined within the entire eye field (Fig 5A, dark blue), the antennal field (Fig 5A, purple), and just the retinal progenitor region (Fig 5B, light blue). Within the antennal field, control, individual RNAi, and double RNAi-expressing clones are recovered with equal frequency and are of nearly identical sizes (Fig 5C–5G). This is most likely due to the fact that expression of both toy and tsh become segregated to the eye field during the second larval instar, thus rendering the expression of the RNAi lines largely ineffective later in development. In contrast, within the eye field and retinal progenitor region, clones in which both toy and tsh were simultaneously knocked down are rare and significantly smaller than either control or individual RNAi-expressing clones (Fig 5C–5G). These findings are consistent with the above experiments using DE-GAL4, where we see that the loss of individual factors has no effect on growth but the concurrent loss of both Toy/Tsh appears to either block cell proliferation, induce apoptosis, or both.
Since several RD genes are known to participate in promoting growth and blocking cell death [12,22,23,33–39], we asked if the loss of Toy/Tsh proteins affects the expression of any of these other network members. This experiment is important for determining if Toy/Tsh are working directly on cell proliferation/survival pathways or if they are working through other RD network members. We analyzed the expression of ey, eya, eyg, dachshund (dac), and hth in clones lacking both Toy and Tsh, and in all cases, the expression patterns and levels of these genes were unaffected (S5A–S5F Fig), suggesting that the regulation of growth by Toy/Tsh is likely independent of the known RD gene network.
We then attempted to determine if the headless phenotype is caused by a lack of cell proliferation, an increase in apoptosis, or both. To do this, we examined the cell cycle profile of Toy/Tsh deficient cells. Since DE-GAL4 drives expression only in the dorsal compartment later in development, we can use EdU staining and PH3 to look at potential differences in S phase and M phase populations within the dorsal (mutant) and ventral (wild type) compartments. Despite detecting significantly fewer total cells in the mutant tissue (Hoechst positive), we do not observe any significant difference in the densities of S and M phase cells between wild type tissue (ventral) and tissues that lack both proteins (dorsal, S6A–S6H Fig), and as such did not pursue any rescue experiments with cell proliferation-promoting genes.
Using TUNEL staining and antibodies against the cell death marker Dcp-1 [40], we observed dying cells in the dorsal half of the eye when Toy/Tsh levels are reduced (Fig 6C–6F, yellow arrows), while wild type discs do not show any dying cells (Fig 6A and 6B). In this experiment, the RNAi lines were expressed only later in development to ensure that there is an actual disc to examine. By the time we start inducing RNAi expression, expression of the DE-GAL4 driver has been already segregated to the dorsal half of the eye, so that is why cell death is only observed in the dorsal compartment. We then asked whether we could restore eye and head development to the toy/tsh double knockdown animals by blocking cell death via expression of two apoptosis inhibitors, DIAP1 and P35 [41,42]. In both cases, substantial restoration of the eye and head were observed (Fig 6H, 6I, 6K, 6L, 6N, 6O and 6P). This is in contrast to the higher percentage of animals lacking any rescue that is seen when a control UAS-GFP transgene is expressed in the double knockdown background (Fig 6G, 6J, 6M and 6P). In total, this set of results suggests that Toy/Tsh are suppressing apoptosis during the early stages of eye-antennal disc development and that the reduced size of double mutant clones and headless phenotypes are mainly the results of the inappropriate induction of cell death. Our findings are consistent with the ability of the mammalian TSHZ3 protein to suppress the expression of the cell death caspase gene, CASP4 [43].
From the results we have shown above, we know that Tsh and Toy are required to promote growth of the eye-antennal disc by suppressing apoptosis. Then why are these proteins segregated to the eye field and removed from the antennal and head epidermis fields later in development? The answer appears to lie in the shifting roles that these proteins play in development. During the second larval instar, both Toy and Tsh proteins are restricted to the retinal progenitor zone ahead of the advancing morphogenetic furrow (Fig 7A–7C) [14,22,24]. If either protein is allowed to remain in the antennal field, then ectopic eyes are induced [14,16]. Having said that, we propose that there is a critical difference between how Ey/Toy and Tsh/Tio group of proteins bring about this tissue transformation. Ey/Toy are transcriptional activators, while Tsh/Tio are transcriptional repressors, and this inherent difference dictates how they act upon tissue-specific genes and ultimately, bring about alterations in tissue fate. Current models propose that induction of tsh/tio causes activation of ey and this further activates the expression of several downstream RD genes such as eya, so, and dac. This cascading set of events ultimately induces the formation of an ectopic eye [16,18]. An example is presented in S7A Fig, in which the forced expression of tsh induces both eya expression and ectopic eye formation.
To separate the role(s) of Tsh/Tio from other RD members in inducing ectopic eye formation, we used a dpp-GAL4 (Fig 7D) driver to induce tsh/tio expression within the antennal and head epidermal fields of wild type as well as eyLB, eya2, and so1 loss-of-function mutants. Each mutant is characterized by the absence or severe disruption of the eye that is caused by deletions of eye-specific enhancers [35,44] (S11 Fig). The loss of these enhancers completely eliminates expression of each gene within the developing eye. It also prevents ectopic eye formation from being induced by the expression of tsh/tio [16]. We observe that targeted expression of tsh/tio can induce the transformation of the arista into ectopic eyes, tarsal leg segments, or a mass of head epidermal tissue [16,18] (Fig 7E–7H). The ability to transform a portion of the antenna into the homologous leg segment is consistent with a report showing that Tio functions as a selector gene for promoting leg development in the milkweed bug, Oncopeltus fasciatus [45].
Surprisingly, blocking eye formation through mutations in ey, so, and eya does not revert the transformed tissue back to an arista as one might have expected. Instead, the percentages of the other types of homeotic transformations rise compensating for the loss of the arista-eye fate switch. For example, if ey is removed (while tsh/tio is still expressed), then arista–leg transformations increase to 80% from a starting point of 30% (Fig 7I). Similarly, if so or eya are removed, while expression of ey and tsh/tio is maintained (S9A–S9F Fig), then the number of arista–head epidermis transformations rises, while the arista–leg fate drops back to about 35% (Fig 7I).
With the knowledge that Tsh/Tio are transcriptional repressors, we sought out to seek answers to two main questions—how does continued expression of tsh/tio in the antennal field affect the antennal/head epidermis genes, and why is it that we observe these distinct phenotypes on induction of tsh/tio when different RD network members are removed. We used cut, (ct), a gene necessary for sensory organ formation in the antenna, as the readout for our primary assay [46]. Expression of either tsh or tio in the antennal field inhibits the expression of cut (Fig 8A–8D, yellow arrows) [47–49]. This finding is consistent with a study showing that Tsh/Tio negatively regulates ct expression during renal tubule development [50], suggesting that the observed cross-regulatory relationship is not tissue-specific and is in fact maintained throughout the multiple tissues. Moreover, the ability to repress ct is at least 250 million years old, as the tsh/tio homolog in the red flour beetle, Tribolium castaneum, can also inhibit ct expression when it is forcibly expressed within the Drosophila antennal disc (Fig 8E and 8F, yellow arrows). We also observe that the opposite is true–induction of ct expression inhibits expression of tsh in the eye progenitor region (Fig 8G, yellow arrows). This is also consistent with the regulatory relationship between Tsh-Cut in the renal tubules [50].
Tsh and Tio are zinc (Zn) finger transcription factors [21,51], and in a prior study, we deleted each Zn finger individually and showed the differential use of these DNA/protein binding domains in the context of inducing ectopic eyes and promoting cell proliferation [20]. Here, we expressed those deletion constructs within the antenna in an attempt to determine which Zn finger might be necessary for inhibiting ct expression. None of the individual deletion proteins were compromised in their ability to repress ct expression (S8A–S8G Fig, orange arrows), which suggests that the Zn finger domains may function redundantly in this context to repress antennal/head epidermis genes.
Apart from the zinc finger domains, Tsh/Tio each have an additional conserved protein motif, the PLDLS domain. The co-repressor protein, C-terminal Binding Protein (CtBP), interacts with the PLDLS domain to mediate the repressive activity of proteins that contain PLDLS and some limited variants. To test if Tsh/Tio induced repression of ct requires the PLDLS domain, we expressed modified Tsh/Tio proteins in which the PLDLS domain has been removed (Tsh ΔPLDLS/Tio ΔPLDLS). The removal of the PLDLS domain eliminates the ability of both Tsh and Tio to repress cut (Fig 9A and 9B). We then tested if CtBP is the actual repressor that mediates the inhibitory effect of Tsh/Tio. To do this, we used the MARCM method to express a full-length Tsh protein within CtBP null mutant cells. We found that Tsh is still capable of repressing ct (albeit at reduced levels) in the absence of CtBP (Fig 9C, green arrows). This suggests that there may be additional factors within the Drosophila genome that interact with the PLDLS domain and aid in the repressive activity of Tsh/Tio proteins.
Other RD member genes have been implicated in the inhibition of antennal/head epidermis genes such as ct [13,52–54], so we set out to determine if the repression of ct is due to Tsh/Tio directly or if it is the indirect consequence of activating one or more of these other retinal network genes instead. As is the case in wild type discs, ct is expressed normally within the antenna of eyLB, eya2, and so1 mutants (Fig 10A, 10D and 10G). However, it can be repressed in these mutant backgrounds when either tsh or tio is forcibly expressed (Fig 10B, 10C, 10E, 10F, 10H and 10I). This suggests that Tsh/Tio represses ct independently of the core ey/so/eya module. Further evidence to support this conclusion comes from three observations. First, although Ey has been implicated in the repression of ct, it can only do so when an ectopic eye is induced (S7B and S7C Fig). It should be noted that Tsh and Tio are activated in the ectopic eyes that are induced by Ey. Second, Ey is unable to repress cut in the absence of so (S7D Fig). And lastly, forced expression of either eya or so, individually, is insufficient to inhibit ct expression (S7E and S7F Fig). There is some debate surrounding the last point as very high levels of So can repress ct expression [13]. Overall, these data suggest that other RD genes may not be directly responsible for silencing ct in the eye disc. Instead, Tsh/Tio might be the primary agents responsible for preventing non-ocular selector genes from being expressed within the eye field.
As mentioned above, unlike other RD members, the induction of tsh/tio exhibits a unique set of phenotypes, including an arista-leg transformation. This was particularly interesting to us, so we set out to identify the molecular mechanism underlying this transformation. The Wingless (Wg) pathway is important for specifying head epidermal and antennal fates [55–58] and it is activated strongly within the aristal segment (S9G Fig), so we first asked if Tsh/Tio are silencing the Wg pathway. In disagreement with this model, the expression of Tsh/Tio actually induces Wg signaling (S9H Fig, blue arrow). A second potential mechanism might involve the Hox gene Antennapedia (Antp). Mis-expression of Antp within the antenna causes a complete antenna-leg transformation [59], we then asked if Antp is activated when tsh/tio is misexpressed. Antp is not normally expressed in either wild type or eyLB mutant eye-antennal discs [60] (Fig 11A). If tsh/tio are forcibly expressed in the antennae of wild type discs, Antp expression remains silenced (S9I and S9J Fig). However, Antp expression is activated within the antennal field when tsh/tio are expressed in eyLB mutants (Fig 11B and 11C green arrows). However, Antp is activated within the ventral head epidermis (Fig 11B and 11C green arrows), which is not the position from which the arista is formed. Thus, in this case, we don’t believe that Antp is mediating the arista–leg transformation. We also examined the possibility that the loss of ct is responsible for the arista-leg transformation, but expression of a ct RNAi construct with dpp-GAL4 did not induce a change in tissue fate (S9K–S9M Fig).
We actually believe that the arista–leg transformation that we are seeing is caused by a loss of spineless (ss) expression. We pursued this model since the Tsh/Tio induced arista–leg transformation that we observe is reminiscent of transformations that characterize the aristapedia allele of ss (ssa) [61,62]. Ss protein is distributed within the aristal segment in wild type discs (Fig 11D). However, if tsh/tio are expressed within the dpp-GAL4 domain, then ss expression is lost in the arista (Fig 11E and 11F, red arrows). A transcriptional reporter (ss-lacZ) that drives expression within the aristal segment [63] is also responsive to Tsh/Tio (Fig 11G–11I), suggesting that both genes are acting on the ss aristal enhancer. While it is not clear if either Tsh or Tio is directly binding to the enhancer, our model is that the arista–leg transformation is due to an inhibition of ss expression by Tsh/Tio. Additionally, we have evidence suggesting that the repression of ss is independent of Distalless (Dll) and hth, two key antennal selector genes that are upstream regulators of ss [62,64–66]. The expression of neither gene is affected when tsh/tio are expressed within the antenna (S9N–S9S Fig) thus Tsh/Tio are likely to repress ss directly.
We also considered the regulatory relationship between Tsh/Tio and Lim1 and aristaless (al), both of which are expressed in the distal-most segments of the antenna and leg [67,68]. Upon induction of Tsh/Tio, al and Lim1 are always lost within the ventral head epidermis in both wild type and eyLB mutant antennal discs (S9T–S9Y Fig, S10A–S10C Fig). However, expression of these two factors is sometimes maintained within the aristal segment. In looking at the aristal segments more carefully we noticed that if Ey is not activated in the arista, then both Lim1 and al expression are maintained in this segment (S10A and S10B Fig). However, if Ey is activated in the arista, then both al and Lim1 are lost. We predict that the adult heads derived from these discs will harbor the arista–eye transformation (S10C, Fig 7F). A more complicated situation involves Dac, which is lost in the A3 ring (S9Z–S9BB Fig, orange arrows) but activated in the ventral head epidermis (S9Z–S9BB Fig, green arrows). Our results suggest that transformation of ventral head epidermis into eyes and the arista into eyes, legs, and head epidermis is the result of two distinct but equally important events: (1) activation of the RD network and (2) inhibition of an endogenous non-retinal gene regulatory network. The type of transformation is dependent upon a combinatorial code of transcription factor activation and silencing.
Throughout the animal kingdom developing tissues often give rise to multiple distinct organs. Examples that we have mentioned earlier in this paper include the vertebrate anterior forebrain, as well as the wing and eye-antennal discs of Drosophila. Other examples include the mammalian diverticulum, which gives rise to the lungs, trachea, and larynx, and the Drosophila leg discs, which not only produce the legs themselves but also portions of the thorax. Our understanding of how initially uniform cellular fields generate multiple organs has been dominated by models in which different GRNs (which include batteries of transcription factors and multiple signaling pathways) become activated in discrete domains at key stages of development. Each GRN is then tasked with specifying the size and the primary fate of each organ, preventing each tissue from adopting the wrong fate, and with ensuring that all requisite cell types are specified and positioned correctly. Such models are challenged by several questions and observations. First, in a variation of the proverbial chicken versus egg conundrum–one must consider if the localized GRN is initially specifying the domain, or has the domain been already specified by upstream factors and the localized GRN is simply carrying out a pre-determined set of instructions. Second, it has been observed that in many instances members of the localized GRN are actually expressed throughout the entire tissue. Therefore, it is not clear how tissue fates are allocated if GRNs are uniformly expressed and competing with each other within the same populations of cells to activate and repress different fates?
In this paper, we provide one mechanistic solution to the latter paradigm. Here, we provide a genetic and developmental mechanism for how a uniform field can be parceled into distinct territories by gene regulatory networks that are initially expressed throughout the entire field. In our example, we have used the Drosophila eye-antennal disc to show that the retinal determination GRN is rewired temporally and spatially during development. In the nascent disc, several members of the RD network promote the growth of the entire field by promoting cell proliferation and cell survival (Fig 12, left panel). As development proceeds, the expression of these genes is sequestered within the developing eye field where they continue to promote tissue growth. In addition, these genes are now tasked with promoting the fate of the eye while simultaneously blocking the field from adopting non-ocular fates (Fig 12, right panel).
This last feature of the RD network, namely the repression of non-ocular fates, is a relatively new idea and is supported by recent findings in the Drosophila eye-antennal disc and the vertebrate anterior forebrain. In flies, mutations within the RD network result in a homeotic transformation of the eye into either an antenna or head epidermis [10,13,33,52–54]. These two non-ocular fates are also derived from the eye-antennal disc. Similarly, disruption of the equivalent RD network results in a transformation of the eye into the hypothalamus, diencephalon, and telencephalon [69–74]. These tissues are, along with the eye, derived from adjacent territories within the anterior forebrain. Together, these findings suggest that the RD network “plays defense” during development and prevents the eye from adopting the fates of adjacent tissues. It is likely that all GRNs play similar roles in their respective tissues. For example, hth is a critical member of the head epidermal GRN in Drosophila and hth loss-of-function mutations result in a head epidermis to eye transformation [75,76]. Given that the five different GRNs that operate in the eye-antennal disc (one each for the eye, antenna, ocellus, head epidermis, and maxillary palp) are likely promoting primary fates and repressing all other fates, our findings underscore the need for all GRNs members to be segregated to their respective domains prior to the tissue specification period.
In the example that we have provided here, the continued expression of Tsh/Tio proteins with the antenna and head epidermis (beyond the point at which they are normally restricted to the eye field) results in several distinct homeotic transformations: head epidermis to eye, arista to eye, arista to tarsal leg and arista to head epidermis (Fig 13 left). The wide range of transformation types is due to the diversity in selector gene activation and repression by Tsh/Tio (Fig 13, right). These results illustrate the drastic and deleterious consequences of inappropriate GRN member expression.
Our study demonstrates that nature has evolved an elegant mechanism for utilizing selector genes for different developmental tasks. In our example, members of the RD GRN are used early to promote the growth of multiple adjacent tissues while later the same GRN is used to promote the fate of one tissue while repressing all other tissue fates. This is gracefully achieved by allowing some RD GRN members to be expressed throughout the entire eye-antennal disc early in development while sequestering the entire network to the developing eye field during the process of tissue specification. Evidence from other GRNs within Drosophila and vertebrates suggest that this temporal and spatial rewiring may be a universally used genetic and developmental mechanism.
The following fly stocks were used in this study: (1) DE-GAL4 (Georg Halder), (2) ey-GAL4 (BDSC), (3) tio-GAL4 (Kwang Choi), (4) eya-GAL4, (5) dpp-GAL4 (Graeme Mardon), (6) hsFLP22 (BDSC), (7) Actin5C>GAL4 (BDSC), (8) Actin5C>y+>GAL4, UAS-GFP S65T (BDSC), (9) Actin5C>y+>GAL4, UAS-lacZ (BDSC), (10) UAS-tsh RNAi (BDSC), (11) UAS-tio RNAi (BDSC), (12) UAS-ey RNAi (BDSC), (13) UAS-toy RNAi (BDSC), (14) tub-GAL80ts (BDSC), (15) UAS-RedStinger, UAS-FLP, Ubi-p63E(FRT.STOP)Stinger (BDSC), (16) UAS-GFP (BDSC), (17) UAS-ey, (18) UAS-eya, (19) UAS-so, (20) UAS-tsh, (21) UAS-tio, (22) UAS-tsh ΔZnF1, (23) UAS-tsh ΔZnF2, (24) UAS-tsh ΔZnF3, (25) UAS-tio ΔZnF1, (26) UAS-tio ΔZnF2, (27) UAS-tio ΔZnF3, (28) UAS-tio ΔZnF4, (29) UAS-tsh ΔPLDLS (30) UAS-tio ΔPLDLS (31) UAS-Tc tsh/tio, (32) UAS-tsh (Amit Singh), (33) UAS-eyg (BDSC), (34) UAS-p35 (BDSC), (35) UAS-DIAP1 (BDSC), (36) eyg-GFP (BDSC), (37) eya2 (Nancy Bonini), (38) so1 (Larry Zipursky), (39) eyLB, (40) ss 522(1–5)-lacZ (Ian Duncan), (41) FRT82B tub-GAL80 (BDSC), (42) FRT82B CtBP87De-10 (Yutaka Nibu), (43) Wg signaling GFP reporter (3xGRH-4TH-GFP) (Ken M. Cadigan). BDSC = Bloomington Drosophila Stock Center. Stocks without a donor have been generated in our lab. All crosses (except for GAL80 temperature shift experiments) were conducted at 25°C.
The following primary antibodies and stains were used: chicken anti-β-gal (1:800, Abcam), guinea pig anti-Toy (1:500, Henry Sun), guinea pig anti-Ss (1:100), mouse anti-Ey (1:100, DSHB), mouse anti-Cut (1:100, DSHB), mouse anti-Eya (1:4, DSHB), mouse anti-Dac (1:100, DSHB), mouse anti-Antp (1:100), mouse anti-Dll (1:500, Dianne Duncan), rabbit anti-GFP (1:1000, Invitrogen), rabbit anti-cleaved Dcp-1 (1:100, Cell Signaling Technologies), rabbit anti-Tsh (1:3000, Stephen Cohen), rabbit anti-Hth (1:1000, Richard Mann), rabbit anti-PH3 (1:20,000, Abcam), rabbit anti-Lim1 (1:1000, Juan Botas), rat anti-ELAV (1:100, DSHB), rat anti-Al (1:1000, Gerard Campbell), guinea pig anti-dCtBP (1:500) Yutaka Nibu, Hoechst 33342 (1:2000, Invitrogen). Developmental Studies Hybridoma Bank = DSHB. Secondary fluorophore-conjugated antibodies and phalloidin-fluorophore conjugates (for detection of F-actin) are from Molecular Probes. Imaginal discs were prepared as described previously in [77]. TUNEL Assay (Sigma-Aldrich) was performed as per manufacturer’s instructions. For adult antennae, adult fly heads were removed using a surgical blade and were placed in a concave depression glass slide containing isopropanol. Using forceps, the antennae were then dissected away from the adult heads while the tissue incubated in isopropanol. The dissected antennae were then allowed to air dry and then mounted on a glass slide using Permount (Fisher Scientific). Adult eyes, heads, and antennae were imaged with a Zeiss Discovery Light Microscope or a Leica M205FA Stereo Microscope.
tub-Gal80ts; DE-GAL4, toy RNAi, tsh RNAi embryos were first collected for an hour at either 18°C or 30°C as per the experiment and then further incubated at the egg laying temperature for defined periods of time before being transferred to the opposite temperature. Eye-antennal imaginal discs were dissected at specific time points after the temperature shift or from third instar larvae.
The heat shock flp-out over-expression system was used to induce clones overexpressing tsh RNAi and/or toy RNAi. Embryos of the appropriate genotype were collected after an egg laying period of 1hr at 25°C. Clones were induced at 24hrs after egg lay (AEL) by a single heat shock pulse of 15min in a 37°C water bath. Larvae were returned to 25°C, and eye-antennal discs were dissected at specific time points. For each genotype clones in three different regions were measured: eye region, antennal region, and eye progenitor region (see Fig 5A and 5B). Adobe Photoshop CS 5.1 was used to outline and measure the area of the clones induced in the eye-antennal disc (in pixels). Statistical significance was calculated using one-way ANOVA with GraphPad Prism. tsh over-expression/CtBP mutant MARCM clones were induced with a 20-min heat pulse in a 37°C water bath 24 hrs AEL. Larvae were returned to 25°C, and eye-antennal discs were dissected from third instar larvae.
Click-iT EdU Alexa Fluor 555 imaging Kit (Invitrogen) was used to detect the cells in S phase. The protocol provided by the manufacturer was adapted and standardized for the eye-antennal imaginal discs. First, the eye-antennal discs were dissected in PBS and incubated in 50 μM EdU containing PBS for 15mins. This was followed by fixation in Paraformaldehyde-Lysine-Periodate (PLP) fixative as described in [77]. The fixed discs were then sequentially washed with 0.1% TritonX PBS and 3% BSA containing PBS. Next, eye-antennal discs were incubated with Click-iT reaction cocktail as per the manufacturer’s instructions followed by washing in 3% BSA containing PBS and 0.1% TritonX PBS. This was followed by standard immunostaining with pH3 antibody to detect the cells in M phase. Finally, for nuclear staining, the eye-antennal imaginal discs were stained with Hoechst 33342. Eye-antennal discs were imaged using Leica SP5 confocal. Total cell numbers, as well as EdU and PH3 positive cell density, were measured using Imaris Image Analysis Software. Statistical significance was calculated using a paired t-test.
RNA from wild type (control), DE>toy RNAi, DE>ey RNAi, DE>tsh RNAi, and DE>tio RNAi eye-antennal imaginal discs was isolated as described by [33] and subjected to RT-qPCR analysis as described in [78]. For each genotype, RNA isolation, cDNA synthesis, and qPCR were performed on three separate biological replicates with each sample consisting of approximately 50 third larval instar eye-antennal imaginal discs.
The following primers were used to detect toy, ey, tsh, tio and so transcripts using RT-qPCR: toy F: 5’-CCA GAG GCA CGT ATT CAG GTT TGG-3’; toy R: 5’-TTA TTT GCC GTG CTG GTT CGA C-3’ (QuantPrime) [79]; ey F: 5’-TGG TAG GTC AAT CAC CCA ACC-3’; ey R: 5’-GCT GCT GTA GTG CCT GAT GG-3’; tsh F: 5’-TCG CAC CAA TCT TTA TGG AAG G; tsh R: GTA CCT ACA GAG AGA TCG AGT GG-3’; tio F: 5’-GAG GCC GTC ATG CTG GAA AT-3’; tio R: 5’-ATG CGA CTC ATT CGA TGG ACA-3`; so F: 5’-GCC TGT GTT TGC GAG GTT CT-3-; so R: 5’-TGC AGC TTA TCA CAT TGT GGC-3’ (FlyPrimerBank) [80].
|
10.1371/journal.ppat.1005581 | Sulforaphane Inhibits HIV Infection of Macrophages through Nrf2 | Marburg virus, the Kaposi's sarcoma-associated herpesvirus (KSHV) and Dengue virus all activate, and benefit from, expression of the transcription regulator nuclear erythroid 2-related factor 2 (Nrf2). The impact of Nrf2 activation on human immunodeficiency virus (HIV) infection has not been tested. Sulforaphane (SFN), produced in cruciferous vegetables after mechanical damage, mobilizes Nrf2 to potently reprogram cellular gene expression. Here we show for the first time that SFN blocks HIV infection in primary macrophages but not in primary T cells. Similarly SFN blocks infection in PMA-differentiated promonocytic cell lines, but not in other cell lines tested. siRNA-mediated depletion of Nrf2 boosted HIV infectivity in primary macrophages and reduced the anti-viral effects of SFN treatment. This supports a model in which anti-viral activity is mediated through Nrf2 after it is mobilized by SFN. We further found that, like the type I interferon-induced cellular anti-viral proteins SAMHD1 and MX2, SFN treatment blocks infection after entry, but before formation of 2-LTR circles. Interestingly however, neither SAMHD1 nor MX2 were upregulated. This shows for the first time that Nrf2 action can potently block HIV infection and highlights a novel way to trigger this inhibition.
| Nrf2 turns on anti-oxidant genes in response to pharmaceuticals like oltipratz, environmental agents like heavy metals and cigarette smoke, endogenous agents like nitrous oxide and nitro-fatty acids and even plant products like sulforaphane (SFN) and epigallocatechin gallate (EGCG). An increasing body of work is showing that some viruses activate and benefit from Nrf2. In this work we tested the impact of Nrf2 on HIV. We used SFN, abundant in cruciferous vegetables and often used as a dietary supplement, to activate Nrf2. Here we show, for the first time, that in immune cells isolated from donor blood, SFN halts HIV infection in macrophages, but not in T cells. We further show that upon SFN treatment the virus is blocked after it has transcribed its RNA-encoded genome into DNA, but before this genetic material is inserted into host chromosomes. Importantly this block is indeed dependent on Nrf2. Interestingly, Nrf2 does not activate recognized anti-viral genes. Thus, unlike viruses recently found to benefit from Nrf2 activation, HIV can be blocked by its activation. This highlights the opportunity to activate a heretofore unrecognized anti-viral function by triggering an antioxidant response with a common dietary component.
| Highly active anti-retroviral treatment (HAART) is saving countless lives, however its application is accompanied by high financial costs, the emergence of resistant viruses and short- and long-term side-effects. A better understanding of how to activate cellular anti-viral defenses promises new therapeutic alternatives to overcome these limitations and to support prevention and cure strategies.
Here we show for the first time that sulforaphane (SFN), a natural product recognized for its health benefits, blocks HIV infection in macrophages. These cells play a critical role in HIV infection and pathogenesis, forming long-lived viral reservoirs [1, 2] and carrying virus into restricted compartments like the brain [3]. Most monocytes, precursors of macrophages, are largely refractory to HIV infection until they differentiate to replenish the macrophage pool [4, 5]. A small but specific subset of these cells however may be readily infectable, even in HAART-treated patients [6–9]. Infected macrophages have been observed in asymptomatic, untreated patients [10] and in HAART-treated individuals [11]. Other work, while not directly probing infection in situ, shows that rectal and vaginal macrophages share markers with HIV-susceptible macrophages [12, 13]. Importantly, macrophages can powerfully increase T cell infection [14] and may thereby exacerbate T cell depletion in infected individuals.
SFN is generated in cruciferous vegetables after mechanical damage, like chewing, causes admixing of glucoraphanin with the enzyme myrosinase. It reaches various tissues after consumption where it maintains biological function [15]. In humans SFN is absorbed in the gut and peak plasma levels are achieved within 90 minutes [16]. In mice SFN can reach the prostate after ingestion [17, 18] where biological activity is maintained [19–21].
SFN is a powerful mobilizer of the transcription factor nuclear factor erythroid 2-related factor 2 (Nrf2) [22–24]. Nrf2, a master regulator of the antioxidant response [25, 26], is held in check by its constitutive degradation through CUL3 ubiquitin ligase complexes using the protein KEAP1 as a substrate adaptor. SFN and other Nrf2-activators modify KEAP1 to prevent engagement of Nrf2 [27]. As Nrf2 accumulates, it translocates into the nucleus, partners with musculoaponeurotic fibrosarcoma protein (Maf) and initiates transcription from antioxidant response elements (ARE). The consensus Nrf2 binding motif, TGACTCAGCA, was derived from known anti-oxidant response genes and Nrf2 binding was reconfirmed in chromatin immunoprecipitation and transcription analyses [28]. Nrf2 activates several types of genes [29] which predominantly encode proteins that counter oxidizing conditions [30].
Three viruses, Marburg virus [31, 32] Kaposi's sarcoma-associated herpesvirus (KSHV) [33] and Dengue virus [34] trigger Nrf2 activation. KSHV and Dengue virus appear to cause Nrf2 mobilization indirectly. Marburg virus however boosts Nrf2 levels directly with a protein designated VP24. This protein assembles with Keap1 to, like SFN, hamper engagement of Nrf2 and thus Nrf2 turnover [31, 32]. All three viruses benefit from increased Nrf2 expression [31–34].
The impact of SFN and Nrf2 on HIV-1 and HIV-2 infection, in primary cells, has not been determined. Here we show that SFN, acting through Nrf2, triggers a strong block against HIV infection in myeloid-derived cells including primary macrophages, but not in primary T cells. SFN blocks infection after reverse transcription but before the formation of 2-LTR circles. Together this work shows that, in contrast to Marburg, Dengue and KSHV, HIV is blocked through Nrf2 and sets the stage for discovering the mechanism underlying this restriction.
SFN, in various contexts, has been found to impair bacteria, fungi and even viruses including Influenza A, respiratory syncytial virus (RSV) and Epstein-Barr virus (EBV) [35–39]. The impact of SFN on HIV infection however has not been tested in primary immune cells. Here, primary T cells were mock-treated or treated with 10 μM, 20 μM or 30 μM SFN for 24 hours. AZT-treated cells (5 μM) served as positive controls for infection inhibition. Cultures were infected with identical aliquots of VSV-G pseudotyped env(‒) HIV-1 luciferase reporter virus and harvested the following day. Cell lysates were tested for luciferase activity to quantify transcription from proviral DNA. While AZT suppressed infection, SFN did not impact infection throughout the dosage range tested (Fig 1A).
Dimethyl fumarate (DMF), a compound used to treat psoriasis and a known Nrf2 activator, slows HIV-1 spread in primary human monocyte derived macrophage (hMDM) cultures [40]. These experiments did not reveal how DMF interferes with HIV, but suggested to us that SFN could impact HIV infection in macrophages. This prompted us to test whether SFN blocks hMDM infection in single-round infection assays. Cells were mock treated or treated with 10 μM, 20 μM or 30 μM SFN for 24 hours as before. AZT-treated cells (5 μM) served as positive controls for infection inhibition. One day after infection with VSV-G pseudotyped reporter virus, culture lysates were assayed for luciferase activity. Here, unlike in primary T cell cultures, luciferase activity dropped sharply in cultures treated with 10 μM SFN and fell to background levels with higher SFN concentrations (Fig 1B). Thus, SFN acts to block single-round HIV-1 infections in primary hMDMs but not in primary T cells.
To gain a better understanding of the breadth of SFN anti-viral action, we tested HeLa, GHOST and HEK293T cells, as well as monocytoid U937 and THP1 cells that were differentiated into a macrophage-like state by incubation in PMA-supplemented media. The cells were pretreated for 24 hours, then infected and assayed as described above. We detected no inhibition of HIV infection after SFN treatment in HeLa, HEK293T or GHOST cells (Fig 2A–2C) but SFN markedly decreased evidence of infection in U937 and THP1 cultures (Fig 2D and 2E). Further titration of SFN in cultures of PMA-differentiated THP1 cells shows that infections can be reduced by 50% when cultures are treated with SFN in the 2.5 to 5 μM range (Fig 2F).
U937 cells do not produce detectable levels of SAMHD1, an anti-viral protein that is expressed and active in PMA-differentiated THP1 cells and in hMDMs [41]. HeLa and HEK293T cells express SAMHD1, but its anti-viral function is inactivated by cell cycle kinase-mediated phosphorylation [42]. Here we asked whether SAMHD1 could be responsible for SFN-mediated anti-viral action in U937 cells. SFN blocks HIV replication in U937 cultures, albeit at higher concentrations than in those of primary macrophages or of THP1 cells (Fig 2D versus 2E). This raised the possibility that SFN triggers SAMHD1 production in U937 cells to block infection. We immunoblotted lysates of mock- and SFN-treated U937 cells for SAMHD1 and found that while SAMHD1 was readily detected in THP1 lysates, it remained undetectable in U937 lysates under the conditions tested (Fig 2G). This indicates that SFN does not block HIV by increasing SAMHD1 levels.
Importantly, SFN does not appreciably affect the viability of the cell-types tested. In order to determine whether cytotoxicity contributes to the apparent anti-viral action of SFN, we assessed cellular dehydrogenase activity in the presence or absence of SFN. Primary hMDMs and T cells, HeLa, GHOST and HEK293T cells as well as PMA-differentiated U937 and THP1 cells were treated with SFN for 24 hours as indicated. The translation inhibitor blasticidin served as a positive control for toxicity. Blasticidin-treated cells all showed decreased dehydrogenase activity; SFN however did not reduce dehydrogenase activity in most of the cell types tested. A slight reduction in dehydrogenase function was seen in THP1 cells at the higher concentrations tested (S1 Fig). Overall these data indicate that SFN mediated impairment of HIV infection is not due to decreased cell viability as indicated by sustained metabolic activity.
SFN treatment could cause changes in cells that continue to impact HIV infection long after treatment. To address this possibility we tested the durability of the SFN-mediated anti-viral state in THP1-derived macrophages. All replicate cultures were treated with SFN for 24 hours. The SFN-containing media was then replaced with SFN-free media and new sets of cultures were exposed to freshly thawed aliquots of the same virus stock at 24-hour intervals. Each set of cultures was harvested for luciferase assays 24 hours after infection (Fig 3A). We found that infectability recovered over time in a dose-dependent manner (Fig 3B) with higher doses of SFN yielding a longer-lasting infection block. After removal from 10 μM SFN, cultures were as infectable as mock-treated controls in about 24 hours. These data show that the anti-viral effect induced by SFN treatment is reversible and that the SFN dose determines the duration over which the anti-viral state is maintained.
Here we tested whether SFN also impairs HIV-2 infection and whether SFN action is reporter-dependent. Cells were mock treated, or pretreated with vehicle, 5 μM AZT or 10 μM SFN, as indicated, and infected with VSV-G-pseudotyped HIV-1 or HIV-2 encoding luciferase in place of nef or with VSV-G-pseudotyped HIV-1 or HIV-2 encoding GFP in place of nef. Luciferase activity was measured in culture lysates and the percentage of GFP(+) cells was determined using flow cytometry. The loss of luciferase signal after SFN treatment in HIV-2-infected cultures paralleled that in the HIV-1-infected cultures (Fig 4A versus 4B). The decrease in the percentage of GFP(+) cells upon AZT or SFN treatment was also the same in HIV-1 and HIV-2 infected cultures (Fig 4A and 4B versus 4C and 4D). These results demonstrate that our observations, showing an SFN-mediated infection block, are not reporter-dependent. The similarity between the results for HIV-1 and HIV-2 further indicate that SAMHD1 is not responsible for the block because HIV-2 Vpx would be expected to at least partially overcome that restriction. The results also support a model, based on our flow cytometry data with the GFP reporter virus, in which fewer cells are becoming infected after SFN treatment, rather than one in which a similar number of infected cells produce less reporter protein.
We exploited the characteristics of VSV-G-pseudotyped, env(‒) reporter viruses to determine that SFN pretreatment blocks infection at or before reporter production from integrated proviruses. Here we tested the impact of continuous SFN treatment on infections with the replication competent, reporter-free, dual-tropic HIV-1 strain HIV-1 89.6 [43]. Primary hMDMs were cultured in media containing DMSO, the vehicle for SFN, 5 μM AZT or 10 μM SFN. Culture supernatant samples were collected 3, 6, 9 and 14 days after infection and tested to determine the concentration of viral capsid protein, p24. Both western blotting and antigen-capture ELISA were employed. The concentration of viral capsid in the cell supernatant increased steadily in the vehicle control, however in the AZT- and 10 μM SFN-treated samples it decreased from residual input levels to background levels (Fig 5A and 5B). The HIV-1 89.6 Env glycoprotein can mediate fusion between adjacent macrophages, making inhibition of virus replication by SFN and AZT readily apparent in the cell cultures (Fig 5C).
Replicate cultures from these experiments were maintained in the presence of vehicle (DMSO) or 10 μM SFN for 14 days to determine the impact of SFN on cell health as measured by mitochondrial dehydrogenase activity. Blasticidin-treated cultures served as positive controls for viability loss. Continuous SFN treatment did not reduce cellular dehydrogenase activity relative to the vehicle only control (Fig 5D).
Finally, we tested whether SFN also stops virus bearing HIV-1 89.6 envelope glycoprotein on its surface in single-round infections. This was of course to determine whether virus with an HIV-1 envelope also encounters a restriction that blocks viral gene expression. As in the other single-round infections, we generated virus by transfecting HEK293T cells with a plasmid expressing the env(‒) luciferase reporter virus. Here, however we co-transfected with an HIV-1 89.6 envelope glycoprotein expression vector rather than one for VSV-G glycoprotein. As in the experiments with the VSV-G-pseudotyped virus, SFN significantly hindered infection with the HIV-1 89.6 envelope glycoprotein-pseudotyped virus (Fig 5E).
Overall, this set of experiments shows that SFN hinders both single-round and spreading infections. It further demonstrates that SFN interferes with infection of virus bearing the HIV envelope glycoprotein as it also interferes with VSV-G-pseudotyped virus. Finally, it also supports that our observation of SFN-mediated HIV inhibition is independent from exogenous reporter genes and shows that the SFN mediated block cannot be reversed by the Nef protein, which was absent from the luciferase and GFP reporter viruses.
SFN acts through Nrf2 to execute the majority of its known functions [44–46]. We thus hypothesized that the SFN-mediated block to HIV infection also relies on Nrf2. We first tested whether other known Nrf2 activators, dimethyl fumarate (DMF) and epigallocatechin gallate (EGCG), also block HIV in single-round macrophage infections. DMF has been shown to temper spreading infections in primary macrophages [40] but its impact on single-round infections has not been tested. EGCG inhibits Tat-induced LTR transactivation in HeLa cell-derived MAGI cells [47] but was not tested in macrophages. EGCG has also been reported to block engagement of the CD4 receptor by HIV-1 gp120 [40, 47, 48]. Here we use VSV-G pseudotyped HIV for single round infections to avoid possible contribution of this effect to our investigation of the impact of Nrf2 on HIV infection.
We found that like SFN, DMF and EGCG hinder reporter virus expression in single-round HIV-1 infections of PMA-differentiated THP1 cells and exhibit minimal toxicity (Fig 6A–6C and S2A–S2C Fig). This extends the previous work with DMF [40] by showing that the block is at a replication step at or before transcription from the provirus. Our observation that EGCG blocked VSV-G-pseudotyped virus eliminates the possibility that EGCG is thwarting a specific gp120/CD4 interaction in our system [48]. While we did not see SFN-mediated infection inhibition in HeLa cells (Fig 2A) we tested whether reporter expression from the provirus is inhibited by EGCG or the other compounds, as this could reflect inhibition of Tat function or transcription from the viral promoter. PMA-differentiated THP1 cells were pre-treated with AZT, DMSO, SFN, DMF or EGCG and then infected with VSV-G-pseudotyped luciferase reporter-encoding HIV-1 (S3A Fig). Replicate cultures were infected with the same reporter virus and, 5 days later, treated for twenty-four hours with DMSO, AZT, SFN, DMF or EGCG. While pretreatment with AZT, SFN, DMF or EGCG significantly inhibited luciferase activity, treatment after infection had much less or no impact. This suggests that each compound is likely acting before transcription in THP-1 infections, although we cannot rule out some impact after integration, especially for SFN. Neither DMF nor EGCG suppressed reporter activity to the same degree as SFN in pretreated cultures, perhaps because SFN is one of the most potent naturally occurring Nrf2 activators [49–52] or because SFN may also have an additional, albeit smaller, impact after proviral integration. These observations prompted us to test directly whether SFN blocks HIV-1 through Nrf2.
To determine whether SFN requires Nrf2 to impair HIV infection of macrophages, we depleted primary hMDMs of Nrf2 with siRNA and assessed the impact of this treatment on infection with VSV-G-pseudotyped HIV-1 luciferase reporter virus, both in the presence and absence of 10 μM SFN. In the absence of SFN, depletion of Nrf2 caused a significant increase in luciferase activity indicating that Nrf2 is detrimental to HIV infection in macrophages and that the basal level of Nrf2 that is present in macrophages is sufficient for establishing some anti-viral activity (Fig 6D). In cells transfected with non-targeting siRNA, SFN significantly reduced luciferase activity, reconfirming that SFN triggers a block to HIV infection of macrophages (Fig 6D).
SFN application, in combination with Nrf2-specific siRNA treatment, yielded a level of infectability that was lower than in the Nrf2-depleted, vehicle-treated cultures but higher than in cultures treated with non-targeting siRNA and SFN together. A western blot of parallel replicate cultures revealed that primary human macrophages express readily detectable quantities of Nrf2 (Fig 6E). Specific siRNA transfection reduced Nrf2 protein levels, but SFN treatment restored Nrf2 expression to levels seen in untreated cultures. Thus, SFN partially counteracted the impact of Nrf2-specific siRNA. Importantly however, Nrf2 levels remained inversely proportional to infection levels even in the absence of SFN. This supports a model in which HIV infection is countered specifically through Nrf2, and SFN acts to block HIV by increasing Nrf2 levels.
If Nrf2 triggers expression of a protein or an RNA with anti-viral function, then we also expect levels of the Nrf2-dependent protein NQO1 to be inversely proportional to infectability. As expected in this scenario, NQO1 levels fell below constitutive quantities upon siRNA-mediated depletion of Nrf2 and increased well above the base line in non-targeting siRNA-transfected cultures treated with 10 μM SFN. Consistent with our hypothesis, NQO1 levels in Nrf2-siRNA transfected cells returned to baseline upon treatment with 10 μM SFN (Fig 6F). Thus infection of hMDMs appears to be Nrf2-dependent. The loss of Nrf2 and Nrf2 function, as reflected by Nrf2-dependent gene expression, enhanced reporter activity. Conversely, the gain of Nrf2 function reduced reporter activity. Overall these data are consistent with an Nrf2-dependent block to HIV infection in hMDMs that is enhanced, here, by SFN treatment.
SFN blocks single-round HIV-1 and HIV-2 infections (Fig 4). This limits the Nrf2-mediated block to steps that extend from attachment of the virus to the cell through transcription from the provirus. We found that SFN counters infection whether viral entry is mediated through HIV Env or VSV-G glycoproteins (Figs 1B versus 5E). This suggests a post-entry block as these glycoproteins engage different entry mechanisms. To further test whether SFN blocks after entry, we used real time PCR to quantify reverse transcription products in primary macrophage cultures infected in the presence or absence of SFN. We infected cells treated with vehicle, AZT or SFN with VSV-G-pseudotyped HIV-1 and quantified late reverse transcription products. Cultures exposed to the same quantity of heat-inactivated virus served to control for contamination of the viral stocks with the plasmid carrying the proviral clone used to generate the virus. The vehicle- and SFN-treated samples showed similar levels of late reverse transcription product, indicating that SFN, unlike AZT, did not hinder replication steps from viral entry through reverse transcription (Fig 7A). When we tested for 2-LTR circles, non-productive infection by-products that correlate with the transport of viral preintegration complexes into the cell nucleus, we found that these were clearly decreased in the AZT and SFN samples relative to the vehicle controls (Fig 7B). Alu-PCR, testing for integrated proviral sequences, also showed that these were similarly decreased in AZT- and SFN-treated cultures to levels that were significantly lower than in the vehicle-treated controls (Fig 7C).
Dengue virus infection, in monocyte-derived dendritic cells, activates the IRF3/7/STAT1 and NF-κB anti-viral and inflammatory pathways, as well as Nrf2-dependent antioxidant genes. Interestingly, silencing of Nrf2 with siRNA increased innate anti-viral responses linked to interferon treatment [34]. Our experiments, with HIV in macrophages, sharply contrast this trend. Specifically, boosting Nrf2 counters infection while Nrf2 depletion supports infection. The restriction that we observed however, after reverse transcription but before 2-LTR circle formation, parallels those imposed by the type I interferon-inducible proteins SAMHD1 and MX2 [53, 54]. Here we asked whether SFN-mediated upregulation of Nrf2 protein levels causes an increase in the expression of those anti-viral proteins. We mock treated PMA-differentiated THP1 cultures or treated them with vehicle (DMSO) or 10 μM SFN, or with 500 U/mL of IFNα. As expected, SFN significantly increased levels of Nrf2 as well as of the Nrf2-regulated genes NQO1 and GCLM (Fig 8). SFN did not significantly change levels of either SAMHD1 or MX2. Conversely, IFNα treatment increased levels of MX2 as expected. SAMHD1 levels, despite some donor-to-donor variation, did not change significantly in response to IFNα, nor did the levels of Nrf2, NQO1 or GCLM. Thus, in our system, we do not see an overlap in the set of proteins that are upregulated by SFN and those that are upregulated by IFNα.
Here we have, for the first time, shown that SFN stops HIV infection in primary macrophages by triggering a block to infection that impacts the virus after reverse transcription but before 2-LTR circle formation. The SFN-regulated restriction relies on Nrf2, a transcription regulator of anti-oxidant genes. Importantly, we found that macrophages express sufficient Nrf2 to maintain a partial restriction even in the absence of SFN. This underscores that the anti-viral activity, while significantly enhanced by SFN through its augmentation of Nrf2 levels, does not require SFN to function.
Our work showing that (1) SFN blocks macrophage infection mediated by either VSV-G or HIV-1 Env, (2) SFN does not block infection in all cell-types and (3) blocks replication after late reverse transcription products have been formed, all support a model in which the restriction is well after entry into the cell but likely before the viral pre-integration complex is imported into the cell nucleus. The lack of anti-viral SFN action in numerous cell types and the dependence in macrophages solely on Nrf2 also indicates that SFN is neither inhibiting HIV directly by modifying a viral component nor is it modifying and thereby inhibiting a cellular component that the virus needs for successful infection. Similarly it is unlikely that SFN is activating a protein, post-translationally, that blocks HIV directly.
Our experiments comparing SFN action on GFP reporter virus with that on luciferase reporter virus further support a model in which the block is before the level of transcription. We used the GFP reporter virus to quantify individual infected cells and the fraction of those infected cells closely paralleled the population-wide reporter activity measured with the luciferase reporter virus. This indicates that the SFN-mediated block reduced the number of infected cells rather than suppressing transcription from similar numbers of infected cells.
HIV-1 Tat was reported to increase Nrf2 levels in HeLa-derived MAGI cells which corresponded with decreased Tat-mediated HIV-1 LTR transactivation [47]. While we did not test MAGI cells, SFN did not detectably impact LTR-directed luciferase reporter expression in HeLa or other non-macrophage-like cells. Our finding that SFN strongly inhibits infection at or before 2-LTR formation could account for the entire SFN-mediated block however, based on data shown in S3 Fig, we cannot fully rule out the possibility that SFN also impacts transcription from virus that evaded the initial block.
Our findings also put into perspective previous work that tested the neuroprotective effects of di- and monomethyl fumarate (MMF). This study showed that DMF and MMF can attenuate spreading HIV infections in hMDMs [40]. Our work, employing single-round infections, allowed us to determine more specifically that DMF hinders a process before expression of new viral proteins. DMF and MMF, like SFN, increase levels of Nrf2 [55], suggesting that these compounds also block HIV in macrophages through Nrf2. Interestingly, in our experiments SFN blocked infections more effectively at lower concentrations than DMF did [40].
None of the genes known to be modulated by Nrf2 direct production of recognized anti-viral products [28]. It is however possible that Nrf2 initiates a different expression pattern in macrophages than it does in previously tested B lymphocyte-derived cell lines [28]. In considering gene expression changes that could occur in human macrophages after SFN treatment, our attention was drawn to work describing murine macrophages treated with oxidized phospholipids [56]. These cells, designated Mox, adopt an Nrf2-dependent phenotype that is distinct from that of both M1 and M2 macrophages. Like SFN treated B lymphocyte-derived cells [28] or the THP1 cells in our work, and unlike M1 macrophages, Mox cells do not upregulate expression of known anti-viral genes. SFN treatment, on the other hand, may elicit a transcription pattern distinct from that described for Mox cells as we observed increases in both GCLM and NQO1 (Fig 8) while only GCLM was upregulated after treatment with oxidized phospholipids.
Future experiments, measuring the infectivity of heterokaryons formed by fusing cells that block HIV in response to SFN and those that don’t, will help to define the SFN-mediated block as a lack of a factor required for infection or the induction of a factor that blocks infection. This will help to guide the search for the mechanism underlying the SFN-mediated block. Presently however, the lack of HIV-associated factors among SFN-modulated genes is not surprising since new cellular anti-HIV factors and HIV co-factors continue to be identified [53, 54, 57–67]. We thus have discovered in SFN, a new way to activate an anti-viral state that HIV may not have previously encountered. It is therefore unlikely that the virus has developed an effective countermeasure.
SFN, and other Nrf2 stimulators, could be used therapeutically to block HIV infection of macrophages and possibly of other important HIV targets. The capacity to hinder both HIV-1 and HIV-2 provides an advantage over several protease inhibitors as well as the fusion inhibitor enfuvirtide and first generation non-nucleoside reverse transcription inhibitors which only block HIV-1 [68]. Further investigation will be required to determine why SFN fails to block infection in T cells, whether the same pathway could be exploited there as well or whether the restriction can be activated directly by other means.
Drugs based on their electrophilic potential, like SFN, have been widely used. These include β-lactam antibiotics [69] and drugs that generate electrophilic intermediates through metabolism like the heartburn medicine omeprazole, or clopidogrel which is used against strokes and heart attacks [70–74]. Afatanib, another such drug, was approved by the FDA for the treatment of lung cancer [75]. Beloranib, also an agent with an electrophilic center, is in trials for the treatment of obesity [76]. The isothiocyante group of SFN can be replaced with other groups to further optimize this compound as a therapeutic agent [77]. Work in this regard has shown that substituting the sulfoxide group of SFN with a ketone group resulted in a compound that retained the biological activity of SFN itself [78].
Finally, the discovery that Nrf2 mobilization counters HIV infection in macrophages stands in contrast with recent work showing that Marburg and Dengue virus as well as KSHV initiate and benefit from increased Nrf2 expression [30, 32, 34, 41]. All four viruses share tropism for cells within the myeloid lineage but KSHV, Dengue and Marburg virus, unlike HIV, also share the capacity to infect endothelial cells [79–81]. The consequences of how HIV, KSHV and Marburg virus interface with Nrf2 may be a reflection of this tropism difference, especially if the putative Nrf2-regulated anti-viral defense is not active in endothelial cells.
All primary cells were obtained from de-identified donors at the University of Nebraska Medical Center under a category 4 exemption from consent procedures based on the anonymity of the donors. This exemption was approved by the Albany Medical College Committee on Research Involving Human Subjects.
HEK293T (ATCC), HeLa (ATCC) and GHOST R5/X4 cells (from Dr. Dan Littman and Dr. Vineet Kewalramani, NIH AIDS Reagent Program, Division of AIDS, NIAID, NIH) were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 5% fetal bovine serum (FBS), 1 mM glutamine, 50 U/ml penicillin and 50 μg/ml streptomycin at 37°C and 5% CO2.
Monocytic THP-1 (ATCC) and U937 (ATCC) cell lines were cultured in Roswell Park Memorial Institute 1640 media (RPMI), with 5% FBS, 1 mM glutamine, 50 U/ml penicillin and 50 μg/ml streptomycin. These cells were differentiated into a macrophage-like state in media containing 5 ng/ml phorbol myristate acetate (PMA) for 5 days (37°C and 5% CO2).
Human lymphocytes were isolated from whole blood as buffy coats by Ficoll density gradient centrifugation. Monocytes were isolated by adhesion and differentiated into macrophages (hMDM) for 10–14 days in DMEM supplemented with 10% human AB serum. Primary T cells were maintained in DMEM supplemented with 10% human AB serum, 2.5 μg/ml phytohaemagglutinin (PHA), and 10 U/ml interleukin-2 at 37°C and 5% CO2. Half of the culture media was replaced with fresh media every 3 days.
HEK293T cells were transfected with virus expression vectors encoding either HIV-1: pNL4-3env(‒)nef(‒)gfp(+) (a gift from Dr. Vicente Planelles), pNL4-3env(‒)nef(‒)luc(+) (a gift from Dr. Nathaniel Landau), HIV-2 (pGL-ANnef(‒)gfp(+) a modified version of a clone provided by Dr. Mikako Fujita in which the gene for GFP was inserted in place of nef), HIV-2 luc(+) (a gift from Dr. Lee Ratner) or dual-tropic, full length HIV-1 89.6 (AIDS Reagent Program, Division of AIDS, NIAID, NIH: HIV-1 89.6 from Dr. Ronald Collman). Where specified, vectors carrying env(‒) proviral clones were co-transfected with pCL-VSV-G, a vesicular stomatitis virus G protein (VSV-G) expression vector. Standard calcium phosphate transfection was used. Viruses for spreading or single-round infections were harvested 48 hours after transfection of the proviral clones. Virus in each experiment was produced in the absence of anti-viral agents. Viral titers were determined by serial dilution on the GHOST R5/X4 indicator cell line.
Sulforaphane (SFN) (catalog no.: S8044, LKT Laboratories, Inc., St. Paul, MN) and Dimethyl Fumarate (DMF) (catalog no.: 242926, Sigma-Aldrich, St. Louis, MO) and epigallocatechin gallate (EGCG) (catalog no.: E4143, Sigma-Aldrich, St. Louis, MO) were dissolved in dimethyl sulfoxide (DMSO). Azidothymidine (AZT) (AIDS Reagent Program, Division of AIDS, NIAID, NIH Germantown, MD) and epigallocatechin gallate (EGCG) (catalog no.: G6817, LKT Laboratories, Inc., St. Paul, MN) were dissolved in water (50 mM stock).
Luciferase activity was measured using the Promega Luciferase Assay System (Promega, Madison, WI) according to the manufacturer’s instructions. Culture medium was removed and cells were washed with phosphate-buffered saline (PBS). Samples were lysed and cellular debris was cleared by centrifugation at 18,000 × g for 1 minute. Twenty microliters of each sample were mixed with 80 μL of Luciferase Assay Reagent. Luciferase activity, in Relative Light Units (RLU), was measured using a Perkin Elmer VICTOR 3V 1420–040 multi-detection microplate reader.
Cells were seeded at 10,000 per well in a 96-well plate (BD Labware, Franklin Lakes, NJ) in triplicate. The cells were then either mock-treated or treated with 10 μM, 20 μM or 30 μM SFN for 24 hours. Cell viability was assessed by measuring metabolic activity using a 2-(2-methoxy-4-nitrophenyl)-3-(4-nitrophenyl)-5-(2, 4-disulfophenyl)-2H-tetrazolium monosodium salt (WST-8) assay which produces a water-soluble formazan-dye upon reduction in the presence of an electron mediator (Cell Counting Kit 8, Dojindo, Kumamoto, Japan). The absorbance at 450 nm was measured with a microplate reader (BioTek, USA). Treatment of cells with 10 μg/ml of the eukaryotic toxin blasticidin served as a positive control for reduced viability.
Short interfering RNAs (siRNA) were purchased from GE Life Sciences (Dharmacon): Non-targeting siRNA (catalog no.: P-002048-01-50) and Nrf2-specific siRNA (catalog no.: J-003755-09-0020). siRNA transfections were performed using Lipofectamine 2000 (Life Technologies, Grand Island, NY, catalog no.: 11668–019) according to the manufacturer’s instructions. siRNA transfection of hMDMs was performed three times with a recovery period of two days between transfections to ensure robust depletion of target mRNA.
Spreading HIV-1 89.6 infections in hMDM cultures were monitored daily, using a Zeiss Axio Observer Z1 microscope, to determine the integrity of the cells. 14 days after infection, phase-contrast images were acquired using the ZEN2012 software by Zeiss.
Flow cytometry was used to detect GFP production in cells infected with reporter viruses. Cells were fixed for 20 minutes at room temperature with 2% formaldehyde and washed with PBS. Ten thousand events were recorded with a FACScan flow cytometer (BD Biosciences, San Jose, CA). The data was analyzed using FlowJo v7.5 software (Tree Star Inc., Ashland, OR).
Proteins from cells lysed with Laemmli buffer (50 mM Tris-HCl (pH 6.8), 2% SDS, 10% glycerol, and 0.1% bromophenol blue) were resolved by SDS-PAGE, transferred onto PVDF membrane (Millipore, Billerica, MA) and probed with the indicated antibodies. The primary antibodies used were: anti-tubulin (catalog no.: N-356, Amersham, GE Healthcare, Pittsburgh, PA), anti-Nrf2 (catalog no.: sc-13032, Santa Cruz Biotechnology, Inc., Dallas, TX) anti-SAMHD1 (catalog no.: GTX83687; GeneTex, Irvine, CA), anti-NQO1 (catalog no.: 3187S; Cell Signaling Technologies, Danvers, MA), anti-GCLM (catalog no.: GTX114075; GeneTex, Irvine, CA), anti-Mx2 (catalog no.: D0214; Santa Cruz Biotechnology, Inc., Dallas, TX) and anti-HIV-1 p24 (catalog no.: 183–H12-5C, obtained from Bruce Chesebro and Hardy Chen through the NIH AIDS Research and Reference Reagent Program). Chemiluminescent blot imaging was done with an Alpha Innotech FluorChem HD2 Imaging system. Densitometric analysis was performed using the ImageJ image analysis software from the NIH.
Real-time PCR was performed, as described by Mbisa et al.[82], Yamamoto et al. [83], and Butler et al. [84] to determine where in the hMDM infection process HIV was blocked by SFN. Supernatant from virus producing cells was DNase treated (catalog no.: AM2238, TURBO DNase, Ambion, Life Technologies, Grand Island, NY) to minimize possible contamination from the proviral clones used to generate the virus. Cultures in which cells were incubated with heat inactivated (65°C for 10 minutes) virus served as controls for carryover of the proviral clones. Total DNA was isolated using a DNeasy kit (Qiagen, Germantown, MD) according to the manufacturer’s instructions. Sample DNA was amplified using TaqMan Universal PCR Master Mix (Thermo Fisher) reagent and the StepOnePlus Real-Time PCR System (Life Technologies, Grand Island, NY). The following primers were used:
Late RT. MH531: 5’- TGTGTGCCCGTCTGTTGTGT- 3’ [82].
MH532: 5’- GAGTCCTGCGTCGAGAGATC-3’ [82].
Probe LRT-P: 5’-(FAM)-CAGTGGCGCCCGAACAGGGA-(TAMRA)-3’ [82].
1st round PCR. First-Alu-F: 5’-AGCCTCCCGAGTAGCTGGGA-3’ [83]
First-Alu-R: 5’- TTACAGGCATGAGCCACCG-3’ [83]
First-gag-R: 5’- CAATATCATACGCCGAGAGTGCGCGCTTCAGCAAG-3’ [83]
2nd round PCR. Forward MH535: 5’- AACTAGGGAACCCACTGCTTAAG-3’ [83]
Reverse tag-R: 5’-CAATATCATACGCCGAGAGTGC-3’ [83]
Probe MH603: 5’-(FAM)-ACACTACTTGAAGCACTCAAGGCAAGCTTT-(TAMRA)-3’ [83]
MH535: 5’-AACTAGGGAACCCACTGCTTAAG-3’ [84]
MH536: 5’TCCACAGATCAAGGATATCTTGTC-3′ [84]
Probe MH603: 5’-(FAM)-ACACTACTTGAAGCACTCAAGGCAAGCTTT-(TAMRA)-3’ [84]
Reference gene for RT and 2-LTR. GAPDH forward 5’-GCATGGCCTTCCGTGTCCCC-3’
GAPDH reverse 5’- CCCTCCGACGCCTGCTTCAC-3’
Probe: 5’-(FAM)-GGTGGACCTGACCTGCCGTCTAGA-(TAMRA)-3’
hMDMs were pretreated with AZT, SFN, or DMSO (vehicle control) for 24 hrs prior to infection with dual-tropic HIV-1 89.6. hMDM cultures were subsequently maintained in media supplemented with AZT, SFN, or DMSO for the duration of the experiment (14 days). Viral supernatants were harvested 3, 6, 9 and 14 days after infection and stored at ‒80°C. HIV-1 p24 antigen capture ELISA assays were performed on all samples according to manufacturer’s instructions (catalog no.: 5421, ABL Inc. Rockville, MD).
The two-tailed Student’s t-test was used to determine statistical significance of differences between sample measurements.
|
10.1371/journal.pgen.1004334 | The Proper Splicing of RNAi Factors Is Critical for Pericentric Heterochromatin Assembly in Fission Yeast | Heterochromatin preferentially assembles at repetitive DNA elements, playing roles in transcriptional silencing, recombination suppression, and chromosome segregation. The RNAi machinery is required for heterochromatin assembly in a diverse range of organisms. In fission yeast, RNA splicing factors are also required for pericentric heterochromatin assembly, and a prevailing model is that splicing factors provide a platform for siRNA generation independently of their splicing activity. Here, by screening the fission yeast deletion library, we discovered four novel splicing factors that are required for pericentric heterochromatin assembly. Sequencing total cellular RNAs from the strongest of these mutants, cwf14Δ, showed intron retention in mRNAs of several RNAi factors. Moreover, introducing cDNA versions of RNAi factors significantly restored pericentric heterochromatin in splicing mutants. We also found that mutations of splicing factors resulted in defective telomeric heterochromatin assembly and mis-splicing the mRNA of shelterin component Tpz1, and that replacement of tpz1+ with its cDNA partially rescued heterochromatin defects at telomeres in splicing mutants. Thus, proper splicing of RNAi and shelterin factors contributes to heterochromatin assembly at pericentric regions and telomeres.
| Heterochromatin formation at specific genomic regions is critical for processes as diverse as gene expression and chromosome segregation. The formation of silent heterochromatin at repetitive DNA elements requires processing of transcripts by the RNA interference machinery. Curiously, factors involved in proper RNA splicing are required for heterochromatin assembly, and it was proposed that splicing factors provide a platform for the recruitment of RNAi complexes independently of their role in regulating splicing. In this study, we found several novel splicing factors involved in heterochromatin assembly. Our analysis of genome-wide splicing patterns by RNA sequencing showed that the mRNAs of RNAi factors are very sensitive to perturbations of RNA splicing machinery. Moreover, we found that splicing factors are critical for the production of a telomere shelterin component and proper telomeric heterochromatin assembly. Most importantly, we showed that introducing the cDNA versions of RNAi and shelterin components alleviates heterochromatin defects associated with splicing factor mutations. Thus splicing factors are required for heterochromatin assembly mainly by regulating the proper splicing of heterochromatin assembly factors.
| Eukaryotic genomic DNA associates with histone and non-histone proteins to form chromatin, which is necessary for the spatial and temporal organization of chromosomes. A distinction is commonly drawn between two types of chromatin: euchromatin and heterochromatin. Euchromatin is less condensed and often associated with genes that are actively transcribed. Heterochromatin is highly condensed and often forms over repetitive DNA elements such as transposons. The formation of heterochromatin prevents expression of transposons, improper recombination of repetitive genomic loci, and missegregation of chromosomes during mitosis and meiosis, thus maintaining genome stability [1], [2].
Histones within heterochromatin regions are usually hypo-acetylated and methylated at histone H3 lysine 9, which serves as a binding site for heterochromatin protein 1 (HP1) [3]–[5]. HP1 subsequently recruits diverse proteins to regulate cellular processes such as transcriptional silencing, recombination suppression, and chromosome segregation [1]. The mechanism that attracts histone methyltransferases and deacetylases to repetitive DNA elements is under intensive study. In certain cases, sequence-specific DNA binding proteins are directly involved in recruitment of these enzymes [6]–[9]. Alternatively, the repetitive nature may itself be sufficient to trigger heterochromatin assembly [10], [11]. Recent work has shown that DNA repeats are transiently transcribed, and the transcripts are processed by the RNA interference (RNAi) machinery into small interfering RNAs (siRNAs), which help target histone-modifying enzymes to repeat regions. However, the mechanistic details of RNAi-mediated heterochromatin assembly is not yet well understood [12]–[14].
The mechanisms by which heterochromatin is assembled and regulated have been extensively studied in the fission yeast S. pombe because it shares basic pathways of heterochromatin assembly with mammals, yet has the key advantages of facile genetics and single representative genes for most key families of mammalian chromatin-modifying factors. In this organism, heterochromatin is present mainly at pericentric regions, subtelomeres, and the silent mating-type region, all of which contain similar repeat sequences composed of dg and dh repeats [1]. These repeats are transcribed during S-phase of the cell cycle by RNA polymerase II. The transcripts are sliced by Ago1 and then reverse-transcribed by the RNA-directed RNA polymerase complex (RDRC: Rdp1, Cid12, and Hrr1) into double stranded RNAs, which are processed by Dcr1 into small interfering RNAs (siRNAs). These siRNAs are loaded onto Argonaute siRNA chaperone complex (ARC: Arb1, Arb2, and Ago1) and then onto the RNA-induced transcriptional silencing complex (RITS: Ago1, Chp1, and Tas3). RITS is targeted to the repeat regions through base pairing between siRNAs and the nascent transcripts and recruits the H3K9 methyltransferase complex CLRC, which contains SET domain protein Clr4 as its catalytic subunit. H3K9 methylation recruits HP1 family proteins Swi6 and Chp2, which in turn recruit histone deacetylases (HDACs) such as SHREC to further compact chromatin (see review [12]–[14]).
Surprisingly, in addition to these complexes, splicing factors are required for heterochromatin assembly at pericentric regions [15]–[17]. In fission yeast, 43% of genes contain introns [18], indicating the prevalence of splicing in this organism. The spliceosome and the splicing reactions of fission yeast are also highly conserved with those of higher eukaryotes, which utilize snRNAs U1, U2, U4, U5, and U6, as well as over one hundred protein components (see review [19]). The process of splicing starts when U1 and U2 snRNPs bind to the 5′ splice site and branch point on a pre-mRNA, respectively. The U4/U6.U5 snRNP and the Prp19 complex (also known as NineTeen Complex, or NTC) are then recruited to form the precatalytic spliceosome. After the release of U1 and U4, the spliceosome is activated and the 5′ splice site is cleaved and fused to the branch point to form a lariat structure. Next, 3′ intron cleavage is coupled to exon ligation in a post-spliceosomal complex of U2, U5, and U6. The mature mRNA is then released and the snRNPs recycled. Most notably, temperature-sensitive mutants prp10-1 (component of U2) and cwf10-1 (component of U5) exhibit silencing defects at pericentric regions [15]. Like dcr1Δ, these mutants lose most siRNAs derived from pericentric repeats [15]. Additionally, the spliceosome associates with RDRC component Cid12 [15], [20], indicating a possible direct role of splicing factors in connecting the nascent transcripts and the RNAi machinery during heterochromatin formation. It was hypothesized that splicing factors act in the RNAi pathway independently of RNA splicing because silencing defects are obvious when the splicing of a control tbp1 mRNA is intact and introducing cDNAs of ago1+ or hrr1+ was unable to rescue pericentric heterochromatin silencing defects of prp10-1 cells [15], [17]. However, the possibility that splicing factors regulate the proper processing of RNAi factors has not been rigorously tested. Notably, a number of recently identified factors involved in RNAi (arb1+, arb2+, ers1+, and dsh1+) contain introns and might require splicing factors for their proper expression [21]–[24].
In this study, we performed a screen of the S. pombe nonessential gene deletion strain library and discovered four new putative splicing factors involved in pericentric heterochromatin assembly. We demonstrated that the phenotype of the strongest of these, cwf14Δ, is similar to those of RNAi mutants in regulating pericentric heterochromatin assembly. RNA-seq analyses further found that cwf14Δ resulted in mis-splicing of a subgroup of genes, including a number of RNAi factors. Moreover, we showed that introducing the cDNAs of three RNAi factors, ago1+, arb2+, and ers1+, significantly alleviated silencing defects associated with cwf14Δ. Furthermore, we found that the mRNA of telomere shelterin protein Tpz1, which is involved in telomeric silencing, was also mis-spliced in splicing mutants and that introducing tpz1+ cDNA partially rescues telomeric silencing defects of splicing factor mutants. Thus splicing factors are involved in heterochromatin assembly mainly through regulating the proper splicing of heterochromatin assembly factors.
To comprehensively identify factors required for pericentric heterochromatin assembly, we performed a screen of the fission yeast haploid deletion library for mutants that affect silencing of a reporter inserted into pericentric heterochromatin, otr::ade6+ (Figure 1A–C) [25]. In wild-type cells, otr::ade6+ is silenced, causing red colony color on low adenine medium. However, when heterochromatic silencing is lost at the pericentric region, otr::ade6+ is expressed, and colonies are white. Strains with intermediate silencing defects show variable degrees of pink/red color, allowing rough phenotypic quantification (Figure 1C). In order to eliminate strains that have inherent metabolic defects causing lighter-than-red color, we performed a control screen with an ade6-M210 query strain lacking the otr::ade6+ reporter (Figure 1B), which allowed us to filter false positives out of the screen. Our finalized list of hits is shown in Figure 1C. Each hit colony was assigned a score between 1 and 4, with 4 indicating the strongest silencing defects.
Among the mutants identified, 25 were previously known to be required for pericentric heterochromatin assembly, validating the effectiveness of our screen. These were mutants in the complexes of CLRC, ARC, RITS, RDRC, CUL4-DDB1, HIRA, Clr6C, TRAMP, and SHREC, and in individual factors such as HP1 homolog Swi6, NAD+ histone deacetylase Sir2, CENP-B homolog Cbp1, and CHD1 remodeler Hrp1 (Figure 1C). There were also a number of previously reported heterochromatin mutants that are listed in the library but were missed in our screen, such as ago1Δ, rdp1Δ, and dcr1Δ. We confirmed by PCR that the null mutation was missing in these strains, indicating that false negatives are more likely the result of incorrect deletions present in the library than due to methodological bias.
Most interestingly, there were eight novel mutants identified in this screen, of which four are uncharacterized genes implicated in various steps of mRNA splicing: cwf14 (SPBC24C6.11), dre4 (SPAC13C5.02), cwf12 (SPBC32F12.05c), and the human SRRM1 homolog we named srm1 (SPCC825.05c). Each of these strains showed elevated levels of pericentric dh transcripts, a common phenotype of heterochromatin mutants, suggesting that these mutants indeed affected pericentric heterochromatin assembly (Figure 1D). As cwf14Δ showed the strongest phenotype among these four splicing mutants, we chose it as the focus of our subsequent experiments.
We first confirmed by serial dilution analysis that cwf14Δ cells containing otr::ade6+ formed white colonies similar to those of dcr1Δ (Figure 2A). We also examined the effect of cwf14Δ on silencing of another reporter inserted at the same location, otr::ura4+ [26]. The silencing of this reporter gene in wild-type cells allows them to grow on counterselective medium containing 5-fluoroorotic acid (FOA). Serial dilution analyses showed that cwf14Δ has comparable silencing defects to dcr1Δ, as indicated by attenuated growth on FOA medium (Figure 2A). Moreover, chromatin immunoprecipitation (ChIP) analyses showed increased enrichment of RNA Polymerase II at pericentric regions in cwf14Δ cells, indicating that the effect on pericentric transcript levels was at least in part due to increased transcription (Figure 2B).
Further ChIP analyses showed strong reduction in levels of heterochromatin hallmarks such as H3K9me2 and Swi6 at otr::ura4+ and to a lesser extent at the endogenous dh repeats in cwf14Δ cells (Figure 2C). This pattern is similar to RNAi mutants such as dcr1Δ, but less pronounced than that of clr4Δ, suggesting that Cwf14 might regulate RNAi-mediated heterochromatin assembly. Consistent with this idea, siRNAs derived from pericentric repeats were eliminated in cwf14Δ cells, similar to dcr1Δ cells (Figure 2D).
It was shown that a 1.6 kb fragment of pericentric dg repeats (termed L5) induces heterochromatin assembly and silencing of adjacent genes when inserted into an ectopic site in an RNAi-dependent manner [27], [28]. Silencing is lost in cwf14Δ cells as well, consistent with the idea that Cwf14 is involved in RNAi-mediated heterochromatin assembly (Figure 2E).
Since pericentric heterochromatin promotes proper loading of cohesin near centromeres to promote chromosome bi-orientation [29]–[31], most heterochromatin mutants show defects in chromosome segregation [26] and sensitivity to the microtubule-destabilizing drug thiabendazole (TBZ) [32]. As expected, cwf14Δ was also sensitive to TBZ (Figure 2A). Moreover, both the silencing and TBZ-sensitivity phenotypes were rescued by complementation of cwf14Δ with a plasmid containing an intact copy of cwf14+ under the control of its endogenous regulatory elements (Figure 2F). Since cwf14Δ cells had some growth defects, we also performed PCR-mediated random mutagenesis and isolated a mutant (cwf14-F26L) that resulted in silencing defects without significantly affecting growth (Figure 2G).
To test whether cwf14Δ affects H3K9 methyltransferase activity of CLRC, we used a system in which Clr4 is tethered via a fused Gal4 DNA binding domain (GBD) to an ectopically integrated ade6+ reporter adjacent to three copies of the Gal4 binding site (3xgbs-ade6+) [33] (Figure 3A). This tethering induces heterochromatin formation over a 6 kb locus, silencing transcription of the ade6+ reporter. RNAi factors are not required for this silencing, consistent with the idea that RNAi is required for the targeting of CLRC to DNA repeats [33]. Dilution analysis showed that cwf14Δ had no effect on Clr4-tethered silencing of 3xgbs-ade6+, and ChIP analyses showed that H3K9me2 was enriched at nearby loci 2 and 3 kilobases away in cwf14Δ cells, similar to wild-type cells (Figure 3B). Collectively, these data suggest that Cwf14 is involved in heterochromatin formation at pericentric repeats through the RNAi pathway.
Cwf14 is highly conserved across species. Its budding yeast homolog, Bud31p, co-purifies with the spliceosome [34], and BUD31Δ causes mis-splicing of the mRNAs of ARP2 and SRC1, two factors required for proper budding [35]. Cwf14 was initially identified in a purification of splicing factor Cdc5, together with a number of spliceosome components [36]. However, whether Cwf14 is a stable component of the spliceosome and involved in splicing has not been tested directly. In order to further specify the mechanism of Cwf14 action, we constructed C-terminally tagged versions of cwf14 at its endogenous locus. Cwf14-GFP and Cwf14-myc were fully functional, as they did not show silencing defects of otr::ade6+ (Figure 4A). Imaging of Cwf14-GFP showed that Cwf14 localizes predominantly in the nucleus (Figure 4B). Western blot analysis showed that Cwf14-myc is a 40 kD protein. Moreover, the F26L mutation resulted in reduced Cwf14 protein levels, indicating that it is a partial loss-of-function allele (Figure 4C). We also performed immunoprecipitation of Cwf14-myc followed by MudPIT mass spectrometry to identify its interacting proteins. Most factors that co-immunoprecipitated with Cwf14, but not in a control purification, were components of the spliceosome, especially from subcomplexes NTC and U5 (Figure 4D and Table S1). This result corroborates data that Cwf14 co-precipitates with Prp17, Prp19, and Cwf2, all members of U5-associated NTC [19], [37], as well as a component of U5 snRNP, Cwf10 [38].
Previous work suggests that the spliceosome is involved in heterochromatin assembly through tethering RDRC to pericentric transcripts [15]. This is because RDRC component Cid12 associates with the spliceosome [15], [20], and no splicing defects were observed in the well-characterized splicing substrate tbp1 in splicing factor mutants when silencing defects were apparent [15], [17]. However, purifications of spliceosome components, including our purification of Cwf14, have not identified any other heterochromatin components [36], [37], [39] (Table S1). These results indicate either that the physical connection between the spliceosome and RDRC is very weak, or that Cid12 is present in two separate complexes, RDRC and spliceosome. It remains a possibility that splicing factors regulate the correct processing of mRNAs involved in RNAi-mediated heterochromatin assembly.
In order to test whether cwf14Δ has a general splicing defect, we performed RNA-seq analyses of total cellular RNAs from wild-type, dcr1Δ, and cwf14Δ cells. The gene expression profiles of cwf14Δ and dcr1Δ showed strong overlap of significantly up-regulated genes (Figure 5A and Tables S2 and S3), consistent with the idea that Cwf14 functions in the RNAi pathway. We found that cwf14Δ indeed resulted in intron retention of a portion of genes (Figure 5B and Table S4), consistent with the finding that it is associated with the spliceosome. The majority of introns were properly processed, which indicates that Cwf14 only moderately affected the activity of the spliceosome. Interestingly, many RNAi factors contain introns. Although unbiased ranking of exon-exon junction ratios between wild-type and cwf14Δ RNAs did not show preferential enrichment of RNAi or heterochromatin assembly factors (Table S4), significant intron retention was observed within mRNAs from ago1, arb2, ers1, and dsh1 in cwf14Δ cells as compared to those in wild-type cells, despite similar sequencing depths genome-wide and similar levels of each gene transcript in both samples (Figure 5C). RT-PCR analyses confirmed that these mRNAs were indeed spliced inefficiently in cwf14Δ as well as in cwf10-1 and prp10-1 cells (Figure 5D). Moreover, Western blot analysis showed strong reduction of protein levels of Flag-Ago1, the RNAi factor most severely affected by cwf14Δ, in both cwf14Δ and cwf10-1 strains (Figure S1), indicating that the mis-splicing of ago1 mRNA, and possibly other RNAi factors as well, resulted in altered protein levels.
Previously, it was shown that introducing cDNA versions of ago1+ or hrr1+ failed to rescue silencing defects of prp10-1 [15]. We also generated a cDNA version of ago1+ at its endogenous chromosomal location and under the control of its endogenous regulatory elements (ago1+::cDNA). This cDNA construct showed no defects in silencing of otr::ura4+, indicating that this replacement created a functional Ago1 protein (Figure S2). However, ago1+::cDNA was unable to rescue otr::ura4+ silencing defects and TBZ sensitivity of cwf14Δ (Figure 6A), even though it restored Flag-Ago1 protein levels (Figure S1). We reasoned that the inability of ago1+::cDNA to rescue cwf14Δ defects is because other RNAi factors are also mis-spliced. We thus generated cDNA versions of arb2+ and ers1+ at their endogenous chromosomal loci, which were both functional (Figure S2). Neither arb2+::cDNA nor ers1+::cDNA alone showed any effect on silencing of otr::ura4+ in cwf14Δ cells (Figure 6A). However, when combinations of two cDNAs were introduced together into cwf14Δ cells, there was a detectable rescue of silencing defects and TBZ sensitivity, and the effect was stronger when all three cDNAs were introduced (Figure 6A). Further ChIP analysis showed that both H3K9me and Swi6 protein levels were significantly increased at both otr::ura4+ and pericentric dh repeats in cwf14Δ cells supplemented with ago1+, arb2+, and ers1+ cDNAs (3cDNAs) (Figure 6B).
We also found that introducing ago1+::cDNA was sufficient to rescue the silencing defects of otr::ade6+ associated with cwf10-1 (Figure S3A). Moreover, there is a significant increase in both H3K9me and Swi6 levels at both otr::ade6+ and dh repeats in cwf10-1 ago1+::cDNA cells (Figure S3B). Altogether, these results suggest that mutations in different splicing factors affect the splicing of diverse RNAi factors to regulate heterochromatin assembly at pericentric regions.
Thus our results clearly demonstrated that splicing factors mainly exert their effects on pericentric heterochromatin assembly by promoting the proper splicing of RNAi factors. However, we caution that the rescue of heterochromatin silencing defects of cwf14Δ cells was incomplete even with ago1+, arb2+, and ers1+ cDNAs. This probably reflects the requirement of Cwf14 for the proper splicing of other RNAi factors such as dsh1+, rdp1+, arb1+, hrr1+, or some unidentified factors involved in heterochromatin assembly. Such an idea is supported by the fact that the rescue of cwf14Δ silencing defects correlated with the number of cDNA constructs that were introduced. It is also possible that splicing factors have a direct contribution in pericentric heterochromatin assembly. However, the strong rescue of pericentric silencing in cwf14Δ cells with cDNA constructs suggests that the regulation of RNAi factor splicing is a major role of splicing factors in this process.
We also found that telomere shelterin component tpz1+ showed a strong reduction in exon-exon junction sequencing reads in cwf14Δ cells relative to wild-type cells, and this phenotype was confirmed by RT-PCR analyses in cwf10-1 and prp10-1 as well (Figure 7A). A C-terminally Flag-tagged version of Tpz1 [40] affected silencing of a reporter gene inserted near telomere repeats (Figure S4), suggesting that Tpz1 is required for telomere silencing. We then analyzed the effect of splicing factor mutations on silencing of a reporter inserted near telomere repeats of chromosome two (Tel2::ura4+) [41]. Indeed, cwf14Δ resulted in silencing defects at this reporter gene (Figure 7B), accompanied by a reduction of H3K9me levels at tlh1, which is embedded at telomeric heterochromatin, as well as the accumulation of tlh1 transcripts (Figure 7C and 7D). Both cwf10-1 and prp10-1 cells showed increased tlh1 transcript levels, indicating that loss of telomeric silencing is a general feature of splicing mutants (Figure 7C and 7D). Because multiple DNA sequences contribute to heterochromatin assembly at telomeres, including tlh1+, telomere associated sequences (TAS), and terminal TEL repeats [8], we further tested silencing at TEL::ade6+, which is inserted on the mini-chromosome Ch16 adjacent to telomere repeats [42]. We found that cwf14Δ, cwf14-F26L, and cwf10-1 resulted in loss of silencing of this reporter (Figure 7E). Interestingly, replacement of tpz1+ with its cDNA partially rescued telomeric silencing phenotypes of the cwf14-F26L and cwf10-1 mutants (Figure 7E and 7F). We could only marginally rescue the telomere silencing defects of cwf14Δ cells with tpz1+::cDNA (Figure 7F and data not shown). Thus it is possible that additional factors involved in telomere silencing might also contain introns and depend on the spliceosome to properly regulate their splicing. Alternatively, splicing factors might affect telomere silencing through additional mechanisms. Nonetheless, these results demonstrate that inefficient splicing of tpz1 contributes to telomeric silencing defects in splicing mutants.
The formation of heterochromatin requires RNAi-mediated processing of repeat-derived transcripts and the targeting of histone modifying activities to repeat regions, leading to H3K9me and the recruitment of HP1 proteins. It has been shown that splicing factors are required for RNAi-mediated heterochromatin assembly in fission yeast, although the mechanism by which they are involved is not well characterized. The previously prevailing model was that the spliceosome physically associates with RNAi factors to regulate heterochromatin assembly, rather than acting through its splicing activity [15], [16]. One of the main evidences for this idea is that splicing mutants show severe silencing defects even though the splicing of tbp1 mRNA was not affected [15], [17]. However, whether these splicing factor mutants selectively affect the splicing of RNAi factors has not been rigorously tested.
Our RNA-seq analyses showed prominent intron retention of a subgroup of mRNAs in cwf14Δ cells, even though the majority of introns (including those of tbp1) are still properly processed (Figure 5B and Table S4). Interestingly, a number of key RNAi factors were among the list of strongly mis-spliced introns, a result that is further corroborated by RT-PCR analyses of a selective set of RNAi factor mRNAs in other spliceosome mutants such as cwf10-1 and prp10-1 (Figure 5D). Most importantly, we found that introducing a combination of cDNAs of RNAi factors significantly alleviated pericentric heterochromatin defects associated with cwf14Δ and cwf10-1 (Figure 6 and S3). Thus splicing factors regulate the proper splicing of RNAi factors, which is a major, if not sole, contributor to heterochromatin assembly defects in splicing mutants. That introducing tpz1+::cDNA was able to partially rescue telomere silencing defects associated with splicing factors further supports the idea that mis-splicing of heterochromatin factors is the reason splicing factor mutants show heterochromatin assembly defects (Figure 7E and 7F).
It is noteworthy that cwf14Δ has only moderate splicing defects, with some introns show very strong sensitivity, whereas most others show little to no defects (Figure 5B and Table S4). This raises the question of whether specific intron sensitivity is a result of introns that are inherently difficult to splice. Consistent with this idea, our RT-PCR analyses showed prominent unspliced precursor mRNAs of RNAi factors even in wild-type cells (Figure 5D). It is also a striking pattern that heterochromatin factors in the same complexes tend to either have or not have introns. For example, many members of RNAi, such as ago1+, arb1+, arb2+, ers1+, dsh1+, rdp1+, and hrr1+, have introns, but none of CLRC (clr4+, rik1+, raf1+, raf2+, cul4+, stc1+), SHREC (clr1+, clr2+, clr3+, mit1+), or swi6+ have any introns, raising the possibility of selective regulation of the RNAi pathway through general or specific changes in splicing efficiency. In fact, an analysis of splicing changes during meiosis showed that one intron of arb1+ is induced to be spliced, while a different intron exhibits splicing repression [43]. Since splicing factors have been identified in screens that affect RNAi-based processes in worms, flies, and plants [44]–[48], it seems a conserved mechanism that proper splicing of the mRNAs of RNAi factors regulates the efficiency of RNAi inside the cell.
The Bioneer library is composed of strains of mixed endogenous ade6 alleles: ade6-M216, which forms pink colonies on low adenine medium, and ade6-M210, which give rise to red colonies, similar to ade6Δ. To avoid the complication of ade6-M216 alleles in our screen, we included in the query strain an ade6-M210-mCherry allele linked to a hphMX6 cassette, which confers resistance to the antibiotic hygromycin B, allowing us to generate a uniform ade6-M210 background. Screens were carried out according to previous protocols [49], with slight modifications. Yeast strains arrayed in 384 strains/plate format were pinned on YES agar medium containing 100 µg/ml G418, and otr::ade6+-natMX6 ade6-210-hphMX6 and ade6-210-hphMX6 strains were pinned on YES+100 µg/ml hygromycin B. After two days growth, strains were mated on SPA agar medium. Plates were incubated at 25°C for 3 days, then 42°C for 3 more days to kill vegetative cells. Strains were then germinated and correct genotype selected by pinning to YES+GNH (G418, Nat, and Hygromycin) or YES+GH (G418 and Hygromycin) medium, and then pinned to YE (low ade) medium for color readout.
Cwf14-GFP and Cwf14-13myc strains were constructed by a PCR-based module method. cwf14Δ, cwf12Δ, dre4Δ, and srm1Δ were derived from the Bioneer fission yeast deletion library, verified via PCR, and backcrossed. The p-cwf14+ plasmid was constructed by insertion of a PCR product containing cwf14+ promoter and coding region into SphI and XmaI sites of pREP41. The cwf14-F26L-myc mutant strain was constructed by integrating a cwf14+-myc-kanMX6 cassette amplified by error-prone PCR (high dNTP concentration) into the endogenous cwf14+ locus. ago1+::cDNA, arb2+::cDNA, and ers1+::cDNA strains were constructed by replacing the endogenous genes with intronless cDNA versions. All were sequenced to confirm full replacement and lack of mutation. cwf10-1 and prp10-1 strains were a kind gift from Robin Allshire. Genetic crosses were used to construct all other strains. The genotype of strains is provided in Table S5. For serial dilution plating assays, ten-fold dilutions of mid-log-phase culture were plated on the indicated medium and grown for 3 days at 30°C unless otherwise indicated.
Total cellular RNA was isolated from log-phase cells using MasterPure yeast RNA purification kit (Epicentre) according to the manufacturer's protocol. Quantification with qRT-PCR was performed with Power SYBR Green RNA-to-CT one-step Kit (Applied Biosystems). RNA serial dilutions were used as templates to generate the standard curve of amplification for each pair of primers, and the relative concentration of target sequence was calculated accordingly. An act1 fragment served as reference to normalize the concentration of samples. Sequence of DNA oligos is provided in Table S6. For RNA-seq, purified RNA was prepared by TruSeq Stranded Total RNA Kit (Illumina), which includes rRNA depletion and chemical fragmentation. Index adapters were added to allow for multiplexing. Paired-end sequencing with 100 bp read lengths was performed on Illumina HiSeq. Mapping was performed with the Tuxedo Suite consisting of Bowtie, TopHat, and Cufflinks. For cwf14Δ, 57,521,513 reads were obtained, and 83% mapped to the genome via Bowtie. For WT, 47,636,518 reads were obtained, and 82% mapped. For dcr1Δ, 64,972,007 reads were obtained, and 83% mapped. RNA-seq data have been deposited to the Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra/) with accession number SRP040479. For the dot plot, exon-exon junction ratios were filtered to remove several types: 1) junctions which mapped zero times in any sample (possible mapping noise), 2) junctions whose sum in WT, cwf14Δ, and dcr1Δ was less than 30 (to avoid randomness due to small sample sizes), and 3) ratios whose values were greater than 150 (to focus the diagram on splicing reduction). Northern blot of siRNAs was performed as described previously [50].
ChIP analyses were performed as described previously [51]. Antibodies used were H3K9me2 (Abcam 1220), Swi6 [52], and RNA Pol II (Covance 8WG16). Quantification with qPCR was performed with Maxima SYBR Green/ROX qPCR Master Mix (Thermo). Enrichment was calculated as: relative levels in ChIP/relative levels in total DNA. An act1 promoter fragment was used as a control for normalization unless otherwise indicated. Sequence of DNA oligos was provided in Table S6.
Exponentially growing yeast cells were harvested, washed with 2xHC buffer (300 mM HEPES-KOH pH 7.6, 2 mM EDTA, 100 mM KCl, 20% glycerol, 0.1% NP-40, 2 mM DTT, and protease inhibitor cocktail (Roche)) and frozen in liquid nitrogen. Crude cell extracts were prepared by vigorously blending frozen yeast cells with dry ice using a household blender, followed by sonication and incubation with 30 ml 1xHC buffer containing 250 mM KCl for 30 min. The lysate was cleared by centrifugation at 82,700×g for 3 hours. The supernatants were incubated with 50 µl of C-myc antibody (Sigma C3956) overnight, for three hours the next day with 50 µl protein G agarose beads, washed eight times with 1xHC containing 250 mM KCl, then two times with the same buffer without NP-40. For mass spectrometry analysis, bound proteins were eluted with 2×100 µl of 50 mM Tris pH 7.5, 5% SDS, 5% glycerol, 50 mM DTT. MudPIT mass spectrometry analysis was performed as described previously [53].
|
10.1371/journal.pgen.1000247 | Silent but Not Static: Accelerated Base-Pair Substitution in Silenced Chromatin of Budding Yeasts | Subtelomeric DNA in budding yeasts, like metazoan heterochromatin, is gene poor, repetitive, transiently silenced, and highly dynamic. The rapid evolution of subtelomeric regions is commonly thought to arise from transposon activity and increased recombination between repetitive elements. However, we found evidence of an additional factor in this diversification. We observed a surprising level of nucleotide divergence in transcriptionally silenced regions in inter-species comparisons of Saccharomyces yeasts. Likewise, intra-species analysis of polymorphisms also revealed increased SNP frequencies in both intergenic and synonymous coding positions of silenced DNA. This analysis suggested that silenced DNA in Saccharomyces cerevisiae and closely related species had increased single base-pair substitution that was likely due to the effects of the silencing machinery on DNA replication or repair.
| Many plants, fungi, pathogens, and animals have chromosome regions that are silenced. Special proteins change the chromosome structure in these domains, turning genes off or lowering their expression levels. We found an increased frequency of DNA mutations in these silenced regions of closely related yeasts. This increase is likely due to silencing proteins interfering with DNA repair or replication. Accurate replication of genetic information with minimal mutations is usually critical for the survival and fitness of an organism; however, there are examples where a high mutation rate is beneficial. The silenced regions of chromosomes are often associated with virus-like transposable elements, and with genes that are important in responding to environmental changes. Hence, it is possible that elevated DNA mutations in silenced regions contribute to genome defense against transposable elements or increased genetic diversity to cope with variation in surrounding conditions.
| The ends of chromosomes in yeasts, vertebrates, Drosophila, and eukaryotic pathogens such as Plasmodim falciparum diverge more rapidly than the rest of their genomes [1]. In budding yeasts of the genus Saccharomyces, chromosome ends contain a high density of repeated sequences and relatively few genes; they are more diverged between species than any other portions of the genomes, and are highly variable within species [2],[3]. The accelerated diversification of subtelomeric DNA is commonly attributed to the presence of transposons and the repetitive nature of these regions, as both contribute to recombination between different chromosome ends [4],[5]. However, subtelomeric regions in yeasts are also silenced, analogously to metozoan heterochromatin [6], raising the possibility that the formation and maintenance of a silenced chromatin state contribute to the observed rapid evolution.
In S. cerevisiae, the best characterized silenced regions are the HML and HMR transcriptionally inactive mating loci of chromosome III. They contain non-expressed copies of the MATa and MATα mating-type genes. During mating type interconversion, HML or HMR is copied into the MAT locus, also on chromosome III, where the resident allele is transcribed. Since haploid cells that express both MATa and MATα behave as non-mating diploids, it is crucial that HML and HMR are silenced. This is achieved through the E and I silencers that flank both of the silenced loci (Figure 1) and recruit Silent Information Regulator (Sir) proteins which then spread throughout the regions. The Sir proteins bind to and deacetylate the tails of histones H3 and H4, leading to silencing of HML and HMR [7].
The Sir2/Sir3/Sir4 protein complex that is responsible for HML and HMR silencing also binds to subtelomeric regions of S. cerevisiae chromosomes [8]. In contrast to the strong and robust silencing of HML and HMR, subtelomeric silencing is weaker [9]. Nevertheless, native telomere-proximal genes and reporter genes inserted near telomeres are reliably silenced [10]–[13].
The Saccharomyces sensu stricto species (S. paradoxus, S. mikatae, S. kudriavzevii, S. bayanus) genome sequences are sufficiently closely related to allow identification of conserved regulatory sequences [14]. Essentially all S. cerevisiae protein-coding genes are found in these other species, and most orthologous intergenic regions in the sensu stricto yeasts can be readily aligned [2],[15]. However, in analyzing the evolution of the HML and HMR silencers, we discovered a surprising lack of DNA conservation in all four flanking regions, motivating an in-depth exploration of the evolution of silenced regions within and between these yeast species. Our observations suggested an additional force in the shaping of these regions.
To identify the E and I silencers in the sensu stricto species, we searched for peaks of conservation in multiple sequence alignments. For both of the S. cerevisiae HML and HMR, we identified contigs in the sequenced sensu stricto species that contained a part of the locus and the adjacent gene. The right side of HMR was misassembled in S. paradoxus with two disjointed contigs with incorrect inverted ends, so we resequenced and assembled the region (GenBank EU597267). HML and HMR were conserved across all five species with clearly conserved orthologs of the neighboring genes (Table S1). However, unlike most intergenic sequences in the genome, the regions around HML and HMR were too diverged to allow multiple alignments. Moreover, local pairwise alignments of these flanking sequences between any of the ten species pairs were also unexpectedly dissimilar. The best pairwise alignments were between the two closest species S. cerevisiae and S. paradoxus, but instead of the genome-wide average of 80% identity for orthologous intergenic regions, the percent identities were: 46% left of HML, 55% right of HML, 52% left of HMR, 45% right of HMR. These alignments were almost as dissimilar as if the sequences were unrelated; 1000 random equal-length sequences with identical base composition that we generated had an averaged local pairwise similarity of 45%. BLAST-based comparisons also did not reveal matches for the sequences between HML or HMR and the nearest flanking genes, ruling out local inversions and rearrangements (Figure 2).
Translocations or transpositions could, in principle, have lead to poor alignments across the species in the HML and HMR flanking regions. In such a case, sequence searches from one species would be expected to produce matches in non-syntenic positions of other species. However, BLAST searches with the diverged intergenic segments around HML and HMR from each of the five species against the assembled genomes of the other species did not produce significant BLAST results outside of the syntenic contigs. The only exceptions were the S. cerevisiae to S. paradoxus matches in repetitive DNA (Figure S1); however these likely reflect homogenization of these repeated sequences by gene conversion rather than functional conservation [16]–[18]. We also excluded the possibility that systematic misassembly occurred in these regions in the sensu stricto by performing BLAST searches against the unassembled traces of each species. Therefore, sequence assembly issues and rearrangements did not explain the poor alignments of DNA sequences flanking HML and HMR.
We determined that the flanking sequences in the five species were indeed orthologous by analyzing conservation of the silencers that have been identified in S. cerevisiae. In three of the four cases (HMR-E, HMR-I, HML-I), there was clear conservation of the known functional binding sites in the silencers, despite the low sequence similarity throughout the intergenic regions. To the right of HML, an Abf1 binding site was present 319–321 base pairs past the HMLα1 stop codon in all five species. At HMR-I, the sequence of the Rap1 and Abf1 binding sites, their orientation, distance to HMR, and spacing between the binding sites were conserved between S. cerevisiae and S. paradoxus. Similarly, the Abf1 and Rap1 binding sites in HMR-E were conserved in all five species, with virtually the same spacing between the sites (39–43 bp), and the distance to HMR was identical in S. paradoxus and S. cerevisiae (Figure 3, Figure S2).
To test if the observed sequence conservation reflected functional conservation, we deleted a 140-bp fragment containing known Abf1p and Rap1p binding sites from the presumptive HMR-E in haploid S. bayanus. The deletion abolished silencing at the HMR locus to the same extent as did deletion of the SIR2 gene (Figure 4). This experiment, together with the in silico observations of the conservation of binding sites and silencer architectures in the HML and HMR silencers, established that the regions from the five species were orthologous and suggested that the DNA flanking the HM loci evolved more rapidly than other intergenic DNA.
Intrigued by the unusual divergence around HML and HMR, we sought to determine if other silenced regions were enriched for diverged sequences. We searched all 6,217 S. cerevisiae intergenic regions for DNA sequences without significant matches to any of the other sensu stricto genomes (Table S2). In subtelomeric regions, defined as the 50 kb internal to each telomere [2], there was an unmistakable enrichment of these non-conserved intergenic sequences. Of the 344 S. cerevisiae intergenic regions with no matches to the sensu stricto, over 40% were subtelomeric, even though subtelomeric DNA constituted less than 20% of the total analyzed S. cerevisiae intergenic DNA (p<10−10 by χ2-statistic).
In principle, unequal recombination between repetitive elements and transposon activity might have caused sufficient insertions and deletions to result in segments of subtelomeric DNA in S. cerevisiae that lacked counterparts in S. paradoxus. Therefore we counted intergenic regions with detectable homology but less than 70% identity between the two species (Table S3). If the enrichment of unique sequences in subtelomeric regions were due to insertions and deletions, we would not expect to also see a subtelomeric enrichment of low-identity regions. However, similarly to the excess of unmatched segments, 12% of intergenic subtelomeric DNA had low-identity matches between S. cerevisiae and S. paradoxus, compared to 7% in the rest of the genome (p<10−10 by χ2-statistic). Therefore, an excess of insertions and deletions could not be the sole reason for the enrichment of diverged intergenic sequences in subtelomeric regions.
Unmatched and poorly conserved subtelomeric intergenic regions were found on all chromosomes (Table S2, Table S3). Therefore, the higher-than-expected divergence was not unique to HML, HMR, or the chromosome that bears them, but was a general phenomenon common to silenced regions.
If rapid divergence were an inherent property of silenced DNA, more intra-species polymorphisms in these regions would also be expected. We measured genome-wide average intergenic SNP frequencies in S. cerevisiae and S. paradoxus [19] and compared them to the frequencies in sequences flanking HML and HMR. Although the HML and HMR loci, per se, and the four neighboring genes exhibited SNP frequencies typical of genome-wide averages, the intergenic silenced DNA around HML or HMR had SNP frequencies two to three times higher than average in both species (Figure 5).
A similar pattern of SNP frequencies to that observed at the HM loci was also detected for telomere-proximal intergenic regions among S. cerevisiae isolates. To avoid counting polymorphisms arising from recombination between repetitive DNA sequences, only SNPs in single-copy intergenic regions were considered. SNPs were significantly more frequent in subtelomeric regions, within 0–20 and 20–40 kilobases of telomere edges, than in the rest of the genome (Figure 6, upper panel). The subtelomeric regions were the only ones that deviated strongly from the genome-wide frequencies.
Increased polymorphisms in subtelomeric and HML and HMR-flanking DNA could result from accelerated base-pair substitutions or from decreased selective constraint on these regions. To distinguish these two possibilities, we analyzed polymorphisms in synonymous positions of codons. If subtelomeric intergenic regions were diverging faster than non-subtelomeric ones because of lower functional constraint, then higher SNP frequencies would be expected for the intergenic but not for synonymous coding positions of subtelomeric DNA.
We counted SNPs at fourfold-degenerate synonymous sites of single-copy genes in S. cerevisiae; dubious genes were excluded. Synonymous SNP frequencies in subtelomeric genes were significantly elevated, compared to the rest of the genome, and the level of increase was similar in the synonymous coding and in intergenic positions (Figure 6, lower panel). For the analyzed subtelomeric and non-subtelomeric genes, there was no significant difference in protein-level conservation of orthologs between S. cerevisiae and S. paradoxus (Wilcoxon–Mann–Whitney p = 0.10) (Figure S3). For the codons of the four genes flanking HML and HMR in S. cerevisiae, the fourfold-degenerate synonymous SNP frequency was also elevated compared to the genome-wide average (7% versus 4.4%), however due to the small number of total synonymous sites, the difference was less statistically impressive (p = 0.01 by χ2-statistic).
Presumably, fourfold-degenerate synonymous sites of similarly conserved genes are under the same selection, regardless of chromosome position. The concordance between SNP frequencies in intergenic regions and in synonymous codon positions in functional genes implied that the higher SNP frequency closer to chromosome ends resulted from hyperdivergence rather than relaxed selective constraint.
Transcription-coupled repair is a type of the general nucleotide excision repair that targets repair machinery to highly transcribed genes [20]. One possible model is that silenced DNA, by virtue of its lack of expression, is deficient in transcription-coupled repair, resulting in increased substitutions. We tested this possibility by analyzing the effect of expression on SNP frequencies for intergenic and coding regions.
A genome-wide RNA-sequencing dataset [21] was used to assign median expression level for each gene and intergenic region. The extent of expression of intergenic DNA was indistinguishable between the most telomere-proximal and non-subtelomeric regions (Figure 7A). As would be expected from the observation, there was no correlation between intergenic expression and SNP density (Figure 7B). For genes, there was a definite decrease in median expression of subtelomeric genes (Figure 7A). However, as for the intergenic regions, there was no increase in SNP frequencies for highly expressed genes (Figure 7C).
Therefore the lack of coding or non-coding correlation between expression and SNP frequencies indicated that transcription-coupled repair was not likely to have contributed to the hyperdivergence of DNA sequence in silenced regions.
Chromosome ends vary widely among the sensu stricto species due to transposons, gene families, and other repetitive elements [2]. By focusing on orthologous sequences that flank the HML and HMR loci in these species and on unique subtelomeric DNA, we identified an additional contribution to diversification of these regions: increased base-pair substitutions.
The data in this paper were based upon SNP frequencies, which reflect the combined effect of the rate of nucleotide change and repair, and the strength of selection. Because the elevated SNP frequency was also found in silenced regions in synonymous coding positions, the most parsimonious view was that selection had little if any impact on these frequencies. Therefore, we inferred that the increased SNP frequency in silenced chromatin reflected an increased mutation rate; whether that increased rate resulted from increased rates of substitution or repair, or both, could not, at present, be determined.
Our analysis of inter- and intra-species variation detected a clear and compelling correlation between Sir-silenced regions and those that exhibited hyperdivergence. In S. cerevisiae, the increase in SNP frequencies was higher in constitutively silenced HML and HMR regions than in the transiently silenced subtelomeric DNA. We considered a myriad of other explanations including proximity to tRNAs, transposons, LTRs, and autonomous replicating sequences and also base composition; however, none of these genomic features explained the dramatic increase in divergence within subtelomeres and in regions flanking HML and HMR. Because silencing can interfere with DNA repair, Sir-based silencing appeared to be the most likely mechanism for this rapid sequence diversification. DNA at the expressed MAT locus is repaired 2.5 times faster than identical DNA at the silenced HML locus [22], and silencing interferes with both photolyase and nucleotide excision repair pathways at a subtelomeric position, independently of transcription [23]. Although Sir-based inhibition of repair was an adequate explanation of these data, we could not exclude the possibility that silenced chromatin may have intrinsically reduced replication fidelity. We considered other possible explanations, of which transcription-coupled repair seemed most plausible, since it should be rendered less useful for genes subject to silencing. However, upon genome-wide analysis, we found no correlation between the level of expression and the frequency of SNPs. Hence, trascription-coupled repair was an unlikely explanation for the increased mutation rate in silenced regions of the genome.
In principle, it should be possible to test whether Sir-based silencing were responsible for the rapid diversification of sequences near and within silenced regions by evolving Sir+ and Sir− strains over a sufficiently long time, and then sequencing the genomes. However, our best estimate of the time that would be required suggested this approach was impractical. There is little doubt that the URA3 gene, if inserted in silenced regions, could be used to detect a higher frequency of ura3 mutations in silenced versus non-silenced regions of the genome. However, the phenotypic lag introduced by the higher expression level of URA3 in the Sir− cells would give the expected correlation of Sir genotype to mutation rate, but for the wrong reason.
Regardless of the underlying mechanism, the potential benefit or detriment to the cell of elevated substitutions in subtelomeres is an intriguing question. Subtelomeric regions are gene poor; therefore the cost of increased mutation rate in these regions might merely be tolerated by the yeasts. However, certain characteristics of heterochromatin in many different organisms and of subtelomeric DNA in yeasts and eukaryotic pathogens raised the possibility that an increased mutation rate may have selective advantage. Heterochromatin in many fungi, animals, and plants commonly contains transposable elements [24],[25]. In budding yeasts, silenced DNA is a hotspot for Ty5 retrotransposon insertion [16], and the Sir4 silencing protein directly interacts with the integrase of Ty5, targeting it to silenced DNA [26]. Silenced chromatin could serve as a decoy to attract an invading transposon to that portion of the genome where its expression would be inhibited, while increased rates of substitution would help to inactivate the newly incorporated transposon [27].
An alternate hypothesis for a beneficial role of hyperdivergence is inhibition of deleterious recombination. Ectopic recombination between repetitive subtelomeric DNA sequences destabilizes the genome. Of the 19 reciprocal translocations identified in the Saccharomyces species, 11 are in subtelomeric regions [2]. Subtelomeric sequences may also promote proper segregation of chromosomes by decreasing meiotic recombination in chromosome ends [28]–[30]. Increased divergence and subsequent reduction in sequence identity would be expected to lower both ectopic recombination between subtelomeric repeat elements and meiotic crossovers in chromosome ends.
It is also possible that residence within hyperdivergent regions may facilitate diversity of certain classes of genes. In S. cerevisiae, many of the subtelomeric genes play a role in adapting to changes in environmental conditions [31],[32]. Antigenic variation of most eukaryotic pathogenic parasites relies on subtelomerically positioned genes [33]. If silencing-based hypersubstitution also occurs in these pathogens, it may aid in host immune evasion. More broadly, transient subtelomeric silencing combined with accelerated DNA evolution may increase phenotypic diversity, allowing organisms to cope with environmental changes. Of course, increased diversity in perpetually silenced genes would have questionable evolutionary value. However, most subtelomeric genes are only partially silenced, with the level of silencing both variable on a cell-to-cell basis and heritable through multiple cell divisions. The striking exception to hypermutation in heterochromatic genes in our data were the HML and HMR loci themselves. Because these loci are in frequent recombinational communication with the MAT locus, the powerful selection exerted on MAT was presumably the force that, through recombination, removed the variation in HML and HMR that would be expected, based upon our hypothesis.
Two recent studies indicate an elevated substitution rate in X chromosome subtelomeric regions and Troponin C gene family members of Drosophila melanogaster [34],[35]. Our study established the generality of this effect across taxa, extended it to the full genome analysis, and excluded all proposed mechanisms except for elevated mutation in silenced regions. Given the conservation of heterochromatic hyperdivergence across taxa, it is presumably beneficial and it may be that increased base-pair substitutions contribute simultaneously to genome stability and to adaptive evolution.
All of the yeast strains used in this study are listed in Table 1.
The URA3 gene was replaced in S. bayanus strain JRY7880 with the hph gene (EUROSCARF plasmid pAG32, [36]), producing the ura3Δ::hph strain (JRY8772). The resulting strain was crossed to JRY7890 to give JRY8774 and JRY8775 (from two different tetrads). Next, the 138-bp fragment of the putative S. bayanus HMR-E, containing matches to the Abf1 and Rap1 binding sites, was deleted through transformation and homologous recombination with a loxP-K. lactis URA3-loxP construct (EUROSCARF plasmid pUG72, [37]). In the resulting strains (JRY8781 and JRY8782), the K. lactis URA3 sequence was excised by expressing the Cre recombinase (EUROSCARF plasmid pSH62, [37]). The hmr-e deletion in the final strains (JRY8785 and JRY8786) was confirmed by sequencing. As a result of these manipulations, the original 138-bp putative HMR-E sequence was replaced with 134-bp sequence from pUG72, containing one copy of a loxP site and flanking nucleotides from the vector (hmr-eΔ::loxP).
The phenotypic consequence of the hmr-e deletion in S. bayanus was assayed by comparing mating ability of the hmr-eΔ::loxP MATα strains (JRY8785 and JRY8786) to the parental HMR-E strains (JRY8781, JRY8782). The S. bayanus strains were patched onto synthetic dextrose minimal medium plates [38], overlapping patches of S. cerevisiae Mata mating tester (JRY2726). Only diploid hybrids resulting from mating would be histidine prototrophs and able to grow. The disruption in HMR silencing changed the MATα mating type to the non-mating phenotype of MATa/MATα diploids, interfering with the haploid's ability to mate with the S. cerevisiae Mata tester.
S. paradoxus genomic DNA was isolated from JRY7910 using the Qiagen Miniprep kit. 5 kb fragment from HMRa1 to GIT1 was amplified with LongTemplate DNA polymerase PCR (forward primer: CTCCACTTCAAGTTAGAGTTTGGG; reverse primer: TTATTAGCAGTGAGGCGTCAGCCA). 12 primer sets were used in sequencing reactions to produce overlapping fragments along the 5 kb sequence, and the fragments were subsequently manually assembled based on overlap and deposited in GenBank (EU597267).
Multiple alignments were made using the ClustalW program [39]. Local pairwise Smith-Waterman alignments [40] between S. cerevisiae and S. paradoxus sequences flanking HML and HMR were performed using the EMBOSS “water” program [41] with DNA-matrix, gap-open penalty of 9 and gap-extension penalty of 1. The flanking regions to the left and right of the HML and HMR loci were based on the annotations in Table S1, using full intergenic regions from the edge of each flanking gene to the nearest HML/HMR edge. Estimation of percent identity in local pairwise alignments of unrelated DNA sequence was based on 1000 alignments between 4,000 base-pair, randomly generated DNA sequences with AT content, matching that of the left side of HMR (67%).
All BLAST searches were performed using NCBI BLAST [42] without repeat masking (−F F), and with mismatch penalty of −1 (−q −1). For HML/HMR BLASTs, e-value cutoff was set at 10−3; for all other searches, the cutoff was 10−5. The “blastp” program was used for S. cerevisiae and S. paradoxus orthologous protein comparisons; and the “blastn” program was used for all other intergenic and coding DNA BLASTs.
Intergenic regions of S. cerevisiae were defined as sequences between transcript edges of all SGD-annotated genes, including uncharacterized, dubious, and coding regions. Transcript edges were defined using the annotations from the RNA-sequencing dataset [21], to exclude 5′ and 3′ untranslated regions from the intergenic sequence. Overlapping BLAST matches to S. paradoxus were merged into contiguous blocks, regardless of synteny. S. cerevisae intergenic sequences 250 base-pairs or longer without BLAST results were considered unmatched. In analysis of poorly conserved intergenic DNA, BLAST matches with less than 70% identity were compared to matches with greater than 70% identity.
S. cerevisiae and S. paradoxus SNP positions were downloaded from http://www.sanger.ac.uk/Teams/Team71/durbin/sgrp [19]. SNPs within 50 kilobases of chromosome ends were counted as subtelomeric, and those at greater distances as non-subtelomeric. Single-copy genes and intergenic DNA were defined as S. cerevisiae sequences that produced only a single significant BLAST match to themselves. If any part of an intergenic region or a gene had additional BLAST matches, the whole region or gene was excluded from the SNP analysis. Genes classified as “dubious” in the Saccharomyces Genome Database were not considered.
Expression levels were obtained from the genome-wide RNA-sequencing dataset [21]. For each transcript and intergenic region, expression level was defined as the median of all the mapped RNA sequencing reads from that segment. SNP frequencies, as described above for the intergenic and synonymous coding regions, were graphed against the respective expression levels, as indicated on the x-axes of Figure 7B and Figure 7C.
S. paradoxus orthologs of S. cerevisiae genes were determined based on best-reciprocal BLAST matches. All possible peptide sequences longer than 50 residues were extracted from six-frame translation of the S. paradoxus genome. Verified and uncharacterized SGD-annotated S. cerevisiae proteins were BLASTed against all the potential S. paradoxus peptides. For each S. cerevisiae protein (XC), the best S. paradoxus match (XP) was then BLASTed back against all S. cerevisiae proteins, and if the best match for XP was also XC, the pair was defined as orthologous. For the genes used in SNP analysis (non-dubious and single-copy in S. cerevisiae), distribution of protein percent identity of subtelomeric S. cerevisiae—S. paradoxus orthologs was compared to orthologs positioned greater than 50 kilobases from chromosome ends in S. cerevisiae.
All statistical tests were performed using R [43].
|
10.1371/journal.ppat.1001030 | Unexpected Inheritance: Multiple Integrations of Ancient Bornavirus and Ebolavirus/Marburgvirus Sequences in Vertebrate Genomes | Vertebrate genomes contain numerous copies of retroviral sequences, acquired over the course of evolution. Until recently they were thought to be the only type of RNA viruses to be so represented, because integration of a DNA copy of their genome is required for their replication. In this study, an extensive sequence comparison was conducted in which 5,666 viral genes from all known non-retroviral families with single-stranded RNA genomes were matched against the germline genomes of 48 vertebrate species, to determine if such viruses could also contribute to the vertebrate genetic heritage. In 19 of the tested vertebrate species, we discovered as many as 80 high-confidence examples of genomic DNA sequences that appear to be derived, as long ago as 40 million years, from ancestral members of 4 currently circulating virus families with single strand RNA genomes. Surprisingly, almost all of the sequences are related to only two families in the Order Mononegavirales: the Bornaviruses and the Filoviruses, which cause lethal neurological disease and hemorrhagic fevers, respectively. Based on signature landmarks some, and perhaps all, of the endogenous virus-like DNA sequences appear to be LINE element-facilitated integrations derived from viral mRNAs. The integrations represent genes that encode viral nucleocapsid, RNA-dependent-RNA-polymerase, matrix and, possibly, glycoproteins. Integrations are generally limited to one or very few copies of a related viral gene per species, suggesting that once the initial germline integration was obtained (or selected), later integrations failed or provided little advantage to the host. The conservation of relatively long open reading frames for several of the endogenous sequences, the virus-like protein regions represented, and a potential correlation between their presence and a species' resistance to the diseases caused by these pathogens, are consistent with the notion that their products provide some important biological advantage to the species. In addition, the viruses could also benefit, as some resistant species (e.g. bats) may serve as natural reservoirs for their persistence and transmission. Given the stringent limitations imposed in this informatics search, the examples described here should be considered a low estimate of the number of such integration events that have persisted over evolutionary time scales. Clearly, the sources of genetic information in vertebrate genomes are much more diverse than previously suspected.
| Vertebrate genomes contain numerous copies of retroviral sequences, acquired over the course of evolution. Until recently they were thought to be the only type of RNA viruses to be so represented. In this comprehensive study, we compared sequences representing all known non-retroviruses containing single stranded RNA genomes, with the genomes of 48 vertebrate species. We discovered that as long ago as 40 million years, almost half of these species acquired sequences related to the genes of certain of these RNA viruses. Surprisingly, almost all of the nearly 80 integrations identified are related to only two viral families, the Ebola/ Marburgviruses, and Bornaviruses, which are deadly pathogens that cause lethal hemorrhagic fevers and neurological disease, respectively. The conservation and expression of some of these endogenous sequences, and a potential correlation between their presence and a species' resistance to the diseases caused by the related viruses, suggest that they may afford an important selective advantage in these vertebrate populations. The related viruses could also benefit, as some resistant species may provide natural reservoirs for their persistence and transmission. This first comprehensive study of its kind demonstrates that the sources of genetic inheritance in vertebrate genomes are considerably more diverse than previously appreciated.
| The integration of a DNA copy of the retroviral RNA genome into the DNA of infected cells is an essential step in the replication of these viruses. Portions of DNA tumor virus genomes can also become integrated into cellular DNA, but this is a relatively rare event, detected by selection of a clone of cells that express the viral oncogene(s). While such integration events occur routinely in somatic cells, retroviral DNA sequences are also integrated in the germlines of many hosts, giving rise to inherited, endogenous proviruses. It has been reported that sequences from viruses that contain RNA genomes and do not replicate through a DNA intermediate, may also be copied into DNA and become integrated into the germline cells of plants and insects [1], [2], [3]. That such events can have biological impact was demonstrated in the case of sequences derived from the positive strand RNA genome of a Dicistrovirus (Israeli acute paralysis virus), which were integrated into the germline of bees from different hives [2]. Bees with genomes that contain sequences encoding a portion of the structural protein of this virus are resistant to infection by this same virus. Similar observations have been made in mice with endogenous retroviral sequences related to a capsid gene (Fv-1 locus) which confers resistance to infection by some retroviruses [4]. These observations suggest that chronic infections of a host with both retroviruses and non-retro RNA viruses can result in germline integration events that produce a host expressing some viral functions that confer an advantage to the species; resistance to subsequent infection by that virus.
With these ideas in mind, we undertook a search in the germline genomes of vertebrates for DNA sequences that may be related to any of the known non-retroviral families of viruses that contain single-stranded RNA genomes. As our analyses were being completed, an independent group of investigators reported that sequences derived from the nucleocapsid gene (N) of ancient relatives of such a virus, the Borna disease virus (BDV), are integrated in the genomes of several mammalian species [5]. Here we report the results of our comprehensive search in which 5,666 sequences from non-retroviruses with RNA genomes were compared with the DNA sequences in the genomes of 48 vertebrate species. Our studies have not only confirmed the integration of BDV N-related sequences, but they have also revealed that sequences related to the matrix and polymerase genes of this virus have been integrated into the germlines of various vertebrate species. In addition, we have discovered genome integrations of viral gene sequences from other members of the order Mononegavirales, with the most prominent related to Ebolaviruses and Lake Victoria Marburgvirus. It is noteworthy that these viruses exhibit extremely high mortality rates in some susceptible species, for example reaching 80% in horses that develop Borna disease, and up to 90% in humans infected with Ebolavirus [6].
In addition to possessing linear non-segmented, negative sense single-stranded RNA genomes, the Mononegavirales have several other features in common, including a similar gene order and transcription strategy in which genes are flanked by specific transcription start and stop sites and are expressed in a gradient of decreasing abundance (Figure 1, for review see: [7]). The 8.9 Kb BDV genome encodes information for at least six proteins. These viruses form a unique family, the Bornaviridae, and they are the only viruses in the Order to replicate and transcribe their genomes within the nucleus of the infected cell [8]. Sheep, horses, and cows are among the natural hosts for this enzootic virus; while there are a number of other experimental hosts, virus replication under such conditions is poor, chronic, and slow [8]. Many tissues can be infected in susceptible hosts, but disease symptoms are commonly neurological. Natural infections of humans are at best controversial, and infectious virus has been isolated from this source only infrequently [9]. Given that the BDV is an RNA virus, its genome sequence conservation among isolates of many mammalian species, separated in both time and geographic locations, is surprisingly high. This suggests strong selection pressure to retain a core sequence for virus viability in a reservoir species with which an evolutionary equilibrium has been established.
The Ebola (EBOV)- and Marburg (MARV)- viruses comprise the two genera of the family Filoviridae. Their approximately 19 Kb genomes are replicated and transcribed in the cytoplasm of infected cells. EBOV and MARV cause highly lethal hemorrhagic fever in humans and have high potential for individual-to-individual transmission. Several strains of EBOV are known, including the Zaire and Sudan strains in Africa, and the Reston strain in the Philippines. The latter has only been associated with monkeys, but a recent report also found infection by this strain in domestic swine, and the presence of antibodies in six exposed farm workers [10]. Recent evidence suggests that bats are the natural reservoir of these zoonotic agents [[11], and references therein,[12]].
To conduct this survey, a BLAST program (see Methods) and the NCBI viral Refseq database of virus sequences were employed (October 2009 release) which, at the time, contained a total of 79,001 viral protein sequences, among them 5,666 sequences from viruses with single-stranded RNA genomes that replicate without a DNA intermediate. The latter sequences included all 4 known Orders of animal viruses?with single-stranded RNA genomes, and represented all 38 recognized families, as well as 9 additional unclassified viral genera with such genomes. These viral sequences were compared with 48 complete vertebrate genomes, to determine if any could be identified in the vertebrate genomes. The results were striking, revealing numerous genomic sequences related primarily to two currently circulating virus families with single, negative strand RNA genomes, the Bornaviruses and Filoviruses (Table 1). Selected examples are listed in Table 2, with a complete list provided in Supporting Tables S1, S2, S3, S4, S5, S6 and S7 and Figures S3 and S4. The most numerous of these virus-like sequences were related to the nucleocapsid N (p40) gene of BDV, but sequences related to the BDV RNA-directed RNA polymerase (L), and to the genes encoding the major nucleocapsid protein, NP, and the minor nucleocapsid, polymerase complex cofactor protein (VP35) of EBOV/MARV, were also detected in several vertebrate genomes. Sequences related to the matrix protein (M) gene of BDV were detected in the lemur and medaka genomes, and to the L gene of EBOV/MARV in the opossum genome. Altogether, we discovered BDV-like sequences in at least 13 species, and EBOV/MARV-like sequences in at least 6 species. A single, high confidence example of sequences related to the L gene of Midway/Nyamanini virus was detected in the zebrafish genome. A sequence related to the Tamara Bat virus in the medaka genome was the lone example related to a positive strand RNA viral genome. In many of these examples no synteny was observed among chromosome locations of the sites in different related vertebrate genomes and we conclude that most represent independent integration events, possibly taking place over extended time periods. In other cases, both synteny of chromosomal locations and copy number stability in a genome is observed for virus-related sequences, through lines of inheritance.
The genes of viruses in the Order Mononegavirales are transcribed as mono- or dicistronic mRNAs (Figure 1). The distribution of endogenous virus-like sequences that were detected here, appear to be limited to one or very few per specie. This, and the fact that single genes are represented in diverse locations, is suggestive of a mechanism that involved the reverse transcription and integration of DNA copies of viral mRNAs by LINE elements, much as cellular pseudogenes are produced. Indeed, we found several cases in which landmarks, or remnants of landmarks, characteristic of Line element-mediated insertion are associated with specific Bornavirus- and Filovirus-related integrations. These include direct repeats flanking transcription start sites and 3′ polyA sequences (Table 3). In many additional cases, only 3′ polyA sequences were observed (data not shown). The fact that direct repeats are not found for some endogenous sequences is not surprising, as these repeats may be just 2 nucleotides long and likely have experienced numerous mutations from the time of initial integration. However, from the informative examples in Table 3 we conclude that some, if not all, RNA virus-related sequences have been integrated into their host genomes by LINE elements via target-primed reverse transcription from ancient viral mRNAs.
In some cases, the integrations of virus-related genes were observed in closely related species descended from each other, allowing an estimate of the oldest common ancestor of these integrations. For example, a rodent lineage (including mice and rats) contains BDV gene N- and L-related endogenous sequences, and a separately derived primate lineage (comprising marmosets, macaques, chimps, and humans) contains endogenous BDV gene N-related sequences integrated into seven different places in the genomes. The rodent and primate lines differ from each other in their integration sites, but within both lineages identical sites of integration and stable copy numbers of genes are observed, indicating decent through lineages of viral genes integrated in the past. In the primate line these sites first appear in the present day marmosets and have been retained over forty million years from a common ancestor of marmosets and humans (Figure 2). Based on the degree of sequence homology of BDV-related genes in different host genomes, most of these integrations seem likely to have originated in the same time frame, with the exception of the integration in squirrels, which has much higher sequence homology to the present day virus (Table S1). We stress that integration events illustrated in Figure 2 appear to have been independent events, and do not come from a single ancient integration: no synteny in integrated sequences and adjacent chromosome is observed across species.
The timing of integrations of the EBOV/MARV-related sequences is less clear. The examples of these viral gene sequences fail to distinguish between the present day strains of EBOV and Lake Victoria MARV suggesting an ancient ancestor of both (Figure 3 and Figure S1). Because the integration events appear to predate the split between these genera, we consider them together, and have estimated their ages indirectly. We start with the assumption that at the time of integration, functional protein-coding sequences were free of stop codons. Some of these integrated viral sequences appear to be under positive selection to the present day, because they have retained their open reading frames. Other integrated viral gene sequences have not retained open reading frames and have mutation rates that are measurable. We can employ the latter to estimate the age of an integration event. The typical rate of vertebrate genetic drift ranges from 0.12% of nucleotides per million years in primates to 2–4 times that value in rodents [13], [14], [15], [16]. There are three stop codons and nineteen codons that can become stop codons with a single base change. Assuming an equal frequency of all possible single nucleotide changes, there is a 12% probability that a random codon change will produce a stop codon in one mutational step. Genomic sequences that once encoded proteins, but are now non-functional pseudogenes, are therefore expected to develop stop codons at a rate of one per 1/(0.12×3×0.0012)≈2310 positions for each million years of evolution of primates, and 2–4 times more frequently in rodents.
We next analyzed virus-derived integrations for the presence of stop codons in the stretches of aligned peptide sequences, as shown in Table 4 (additional integrations are listed in Table S8). According to the calculations described above, the two least conserved, near full-length integrations of BDV-related genes in humans, hsEBLN-3 and hsEBLN-4, appear to be 48 and 40 million years old respectively, consistent with our earlier estimates based on primate phylogeny. Integrations in rodents appear to be more recent, or have lost their protein coding ability at a later time, about 21 million years ago for rodEBLL and 19 million years for rodEBLN-2 and rodEBLN-4. Interestingly, the mouse integrations appear to be under stronger selection that those in rats. The EBOV/MARV-related integrations in the opossum genome appear to be 32–53 million years old (assuming 0.13% neutral rate for nucleotide drift per million years [17]). The ages cited here are rough estimates, as rates of genetic drift vary in time and across different stretches of DNA. Other integrations have similar sequence identity with the present day viruses and appear to originate from the same time in history. However, we do not explicitly cite their ages due to the preliminary nature of the scaffold assemblies for carrier species (Table 4 and Table S8).
The absence of the stop codons in some integrations points to strong selective pressures towards maintenance of full-length open reading frames. This is in contrast to the actual peptide sequences that appear to be undergoing neutral drift. Over the 20 million years of evolution in rodents and 40 million years in other mammals, we expect a 5–10% nucleotide change or approximately 15–30% codon change, if there is no selective pressure against fixation of such events in the population. Accordingly, one would expect to observe a stop codon in 1.8–3.6% of the codons. This is, indeed, the case for the majority of the integrations (Table 4 and Table S8). In contrast, several integrations show signs of strong positive selection, namely those related to the BDV N gene in humans, microbats, rodents, and other animals, and both the EBOV/MARV NP and VP35 gene-related integrations in bats and tarsier. Some integration events, including the BDV N-like sequences in humans (e.g. hsEBLN-1) and the EBOV VP35-like sequences in microbats (mlEEL35) have maintained nearly full-length open reading frames (Table 2). The probability of having no stop codon in the longest of these, the BDV gene N-like integration in humans, is one in eight hundred, suggesting that at some time, past or present, there was strong selective pressure to keep and express this ancestral viral gene.
Expressed sequence tags (EST) were identified for four integrated copies of the BDV N-related genes in humans (hsEBLN-1 through hsEBLN-4). The chromosome 3 integration (hsEBLN-2) is actually tiled on Affymatrix chips to detect mRNAs from human tissues. Analysis of a very large diversity of tissue types show low levels of this transcript in most tissues tested, intermediate levels in thymus, olfactory bulb, fetal thyroid, liver, prefrontal cortex, CD34 cells, endothelial cells and dendritic cells, and high levels in CD4 and CD8 T-cells (Figure S2). In susceptible species, BDV replicates mainly in cells of the nervous system, but viral nucleic acids and proteins have been isolated from peripheral blood mononuclear cells. It is clear that several BDV N-like endogenous sequences are expressed as mRNAs in human tissues. Expression of mRNA from these endogenous sequences was also detected in several cell lines in cell culture [5].
BDV is an enzootic virus, with natural infections occurring in sheep, horses, and cattle [18], in which serious, often fatal, neurological symptoms are observed. These animals have no detectable copies of the BDV-related endogenous sequences. Furthermore, species in the primate and mouse/rat lineages, which contain endogenous N-like sequences, are generally resistant to the virus, or the virus is observed to replicate poorly with little or no symptoms in these animals [19] (Table 5). In cows, which do have endogenous sequences related to the BDV N gene, there is apparently no present day selection for its coding capacity (Table 4), and cows are known to be susceptible to Borna disease. Thus, there appears to be a general correlation between natural resistance to the pathogenic effects of the virus and the potential for expression of BDV N-like endogenous sequences in a host. However, as has been observed with Fv-1 in mice [20], natural resistance can be overcome under experimental conditions in which animals or cell cultures may be subjected to large doses of the virus (Table 5).
The X-ray crystal structure of the N protein of BDV has been solved, and a number of critical features determined [21]. The protein is organized in two domains, separated by a short linker, and assembles into a homotetramer. We find that open reading frames in two endogenous human inserts, hsEBLN-1 and hsEBLN-2, are long enough to encode folded N-terminal domains (Figure 4), while an open reading frame in hsEBLN-1 also encodes a complete C-terminal domain. When expressed, either of these proteins could conceivably affect the proper assembly of the BDV ribonucleoprotein complex. Production of N-related antibodies might also inhibit virus replication. An open reading frame is also observed in the integration in squirrels (stEBLN), encompassing a complete C-terminal domain. The BDV gene N-related integration in the genome of the microbat Myotis lucifugus (mlEBLN-1), might also be found to carry a full-protein open reading frame when the preliminary assembly of this genome undergoes final revision.
EBOV and MARV are zootropic viruses that cause infections with some of the highest mortality rates in humans, primates, and pigs. Recent studies have suggested that megabats, specifically Hypsignathus monstrosus, Epomops franqueti, and Myonycteris torquata, could be potential natural reservoirs for EBOV [22]. Later studies also identified microbat Mops condylurus, as well as several other megabats, as potential reservoirs [11]. Some of the bats actually carry live virus, yet exhibit no visible symptoms of disease. There are more than 1,100 recognized species of bats, comprising about a fifth of all mammalian species [23], but the genomes of only two bat species have been sequenced. Our results show that at least one of them, the microbat Myiotis lucifugus, has detectable integrations of EBOV/MARV-like sequences, with several of these showing strong selective pressure for maintaining open reading frames (Table 4).
The most widespread EBOV/MARV integrations observed in this study are derived from the major viral nucleocapsid gene NP and the minor nucleocapsid and polymerase complex cofactor gene VP35. The endogenous sequences related to the NP protein align with the amino-terminal region (Figure 5), which is conserved among these viruses and the Paramyxovirus family, and is critical for NP-NP protein interactions [24], [25]. The microbat sequence mlEELN-1, for example, covers most of this region, including a highly conserved stretch of amino acids and part of a structurally disordered acidic region, which is thought to play a role in the incorporation of the protein into virus particles [24].
Determination of the X-ray crystal structure of the interferon inhibitory domain (IID) of the EBOV VP35 protein has identified two interacting sub-domains, the C-terminal of these includes a cluster of basic residues, centering on R312, which are critical for RNA binding [26]. The microbat endogenous sequence mlEEL35 encompasses the entire IID domain as well as a good portion of the N-terminal domain, which is required for VP35 oligomerization as well as viral replication and transcription (Figure 6). A comparison of the sequences shows that residues important for interactions between the IID sub-domains are largely conserved in mlEEL35 [27], [28]. However, while an arginine residue corresponding to R312 is retained in microbats and the tarsier, two or more of the surrounding acidic residues are substituted in each of these endogenous sequences. Substitution of these residues in EBOV VP35 diminishes RNA binding and abrogates the interferon antagonist function of this protein [26], [27]. Furthermore, viruses that carry these relevant mutations are non-pathogenic in normally susceptible guinea pigs, and animals infected with this mutated virus develop antibodies that render them resistant to subsequent challenge [29].
Our sequence search also uncovered what appear to be remnants of ancient integrations of virus-like glycoprotein genes (G), which are most similar to the glycoproteins from the Order Mononegavirales (Table 6). A BDV gene G-like integration in primates was acquired sometime before the split between humans and old world monkeys, and there are several integrations that most resemble the Filovirus glycoprotein genes (GP). In the Filoviruses, the GP precursor protein is cleaved to form two bound peptides, GP1 and GP2. We found no traces of receptor-binding GP1 [30] in the vertebrate genomes analyzed. However, we identified several sequences related to the second peptide, GP2, which is involved in glycoprotein trimerization [31], and is highly conserved among known Filoviruses (Table 6). Because GP2 shares sequence elements with the avian sarcoma/leukosis virus, the flanking regions of the top BLAST glycoprotein hits were checked for retroviral sequences, LTR elements and gag-pol genes (as described in Methods), and integrations that show no known adjacent retroviral elements were identified. Nevertheless, some ambiguity remains due to the preliminary nature of several of the vertebrate genome assemblies.
Assuming that the endogenous glycoprotein encoding sequences are, indeed, related to viruses in the Order Mononegavirales, their integration may also play role in virus resistance. For example, expression of a GP2 peptide from endogenous sequences may affect the trimerization of GP from a related infecting virus. Recent studies have indicated that over-expression of Filovirus GP in host cells may prevent subsequent infection with the virus [32]. Whether expression of integrated GP-like sequences can stimulate such cellular immunity or other types of resistance to infection remains to be explored.
This survey has uncovered a fossil record for currently circulating RNA virus families that stretch back some 40 million years in the evolution of host species. The error rate per replication of the DNA genomes of the hosts is much lower than the error rates of RNA-dependent RNA synthesis, the mechanism by which these viruses replicate their genomes. Consequently, the host genome contains a more accurate record of the archival genes of viruses with RNA genomes than the related present-day viruses. Considering the relatively high rate of mutation in RNA viruses, and the stringent criteria we utilized to detect homologies, what is reported here should be taken as an underestimate of such viral gene integration events. The most common events we detected derive from certain viruses that contain negative single strand RNA genomes. This might be a reflection of some unusual properties of such viruses and their hosts. For example, the viruses could have high sequence conservation or the hosts could have been selected to retain specific viral sequences that confer resistance to subsequent infection. However, the results of this search are as interesting for what was not found as what was found.
The endogenous viral sequences that were identified with highest confidence are all related to currently circulating viruses in the Order Mononegavirales, which contain single negative strand RNA genomes. Furthermore only two of the four recognized families in this Order are represented, the Bornaviruses (BDV) and Filoviruses (EBOV and MARV). In one species, zebrafish, we also found endogenous sequences related to members of a possible new Taxon in this viral Order, comprising Midway and Nyamanini viruses [33]. These results seem especially noteworthy, as the genomic insertions reported in plants and insects are all derived from viruses with plus strand RNA genomes, such as the Flaviviruses and the Picornaviruses [1], [2], [3]. Furthermore, the data presented here (Tables 3 and S1) indicate that the endogenous sequences in vertebrate genomes were likely integrated via target-primed reverse transcription of ancestral viral mRNAs by LINE elements. As all viruses produce mRNAs during active infection, the selection or retention of endogenous sequences from mainly one viral Order, is all the more striking.
The cellular location of viral replication does not appear to be a critical factor in the insertion of endogenous sequences, because the Bornaviruses replicate in the nucleus and the Filoviruses, in the cytoplasm. We note, in addition, that no endogenous sequences were found that are related to viruses in the Orthomyxovirus family, such as the influenza viruses, which contain segmented negative strand RNA genomes and also replicate in the nuclei of infected cells. However, it is possible that some feature of the mRNAs produced by these viruses is recognized preferentially by LINE machinery, or can promote access to such machinery in the nucleus, and such notions can now be tested. LINE elements are known to be active in the germline [34], and it is possible that the germline cells of some infected vertebrates may have been especially susceptible to infection by the ancestors of these viruses. Finally, DNA copies of mRNAs from other RNA viruses may, indeed, have been integrated into the germlines of infected vertebrates, but are no longer recognizable. Once DNA copies are inserted into the host genome one would expect the mutation rate of these sequences to be reduced by about four orders of magnitude compared to the genes in replicating RNA viruses, rapidly separating the virus sequences of today from the those of the past. Indeed, a DNA copy of an RNA viral genome trapped in a host chromosome is a window on the RNA virus sequences of the past. In this context, the high conservation of the BDV genome [35], [36] may partially explain our ability to detect the related endogenous sequences.
By far the most readily observable endogenous virus-like elements uncovered in our study were related to BDV. For example, these germline integrations persisted for millions of years as recognizable copies of the N gene in primate and rodent lineages, and of the N and the L genes in bats. Furthermore, an initial event appears to slow or stop further integration events, suggesting that the viral gene product(s) can inhibit further virus infection, or eliminates the need to further select for the new integration event. Several integrations also appear to have been selected for their protein coding capacity, with no stop codons emerging over the past forty million years. This is particularly striking because the amino acids in these genes appear to be undergoing the expected frequency of neutral drift, at least among shared integrations in the primate lineage.
There are several possible mechanisms by which an endogenous viral gene product may inhibit the subsequent infection of a cell or animal by the same virus. For example, synthesis from the endogenous sequence of an RNA molecule that is partially complementary to the infecting viral RNA could trigger an early interferon or RNA interference response. In addition, translation of an mRNA from the endogenous viral sequence would lead to production of a protein or peptide that is similar, but not identical to that of the infecting viral protein. In the case of nucleocapsid-like proteins (N, NP), such an endogenous gene product could block virus replication or result in the assembly of faulty, non-infectious particles. This would require genetic drift to produce missense mutations but no stop codons, which is the case for some endogenous sequences that we have discovered. Because the function of these proteins requires appropriate multimerization, even a small number of abnormal or defective, endogenously produced monomers could exert a substantial biological effect. Sequence differences in proteins expressed by the endogenous L- and VP-35-like genes could also result in assembly of defective virus particles. Such particles might then become good immunogens, providing immune protection in the host. It is also possible that production of glycoprotein peptides encoded in endogenous viral sequences might block infections by viruses with similar glycoproteins. Examples of the various resistance mechanisms cited above have been shown to exist with several virus groups. This includes experiments in rats, where ectopic expression of individual proteins of the Bornavirus N, X, and P genes, but not their mRNA, inhibits virus replication [37].
There is likely strong selection pressure to establish a resistance mechanism against Bornavirus and Ebolavirus/Marburgvirus, given their high mortality rates in susceptible species. We have noted that the natural hosts of BDV, such as cows and horses, have no detectible sequences related to the BDV N gene (Table 1), or that the integration is under no present-day selection (Table 4). It has also been reported that resistance to the neurological symptoms of BDV is genetically inherited in rats and is encoded in an unknown host gene [38]. It would now be quite interesting to test whether or not that gene is the BDV-related rodEBLN sequence. It would also be interesting to examine the endogenous sequences in the human population in greater detail, to determine if there are polymorphisms or deletions that might correlate with neurological diseases, which could lead to a re-examination of the role of BDV in such conditions.
Natural resistance to currently circulating EBOV and MARV may allow species to serve as asymptomatic reservoirs for these viruses. In microbats, we identified endogenous sequences related to the NP and VP35 genes of these Filoviruses, in addition to the N and L genes of BDV. Bats of different species have been identified as possible natural reservoirs of EBOV and MARV in areas of human outbreaks in Africa [39], [40], [41]. Recent studies confirm that these viruses co-circulate in Gabon, where bats infected by each virus are found. It should now be possible to ask if there is any correlation between the presence and properties of the endogenous sequences in the various bat species and their ability to serve as natural reservoirs for these negative strand RNA viruses.
In summary, our studies have made it clear that ancient relatives of some RNA viruses have left DNA copies of their sequences in the germline cells of their vertebrate hosts. The sources of vertebrate genetic inheritance are, therefore, considerably more diverse than previously appreciated. A number of recent reports from tissue culture experiments or clinical studies have presented evidence for the incorporation of DNA sequences corresponding to all or part of the genomes of a variety of infecting RNA viruses into host cell DNA [e.g. 5] [42], [43], indicating that such events might occur in somatic tissues with some frequency. However, the mechanisms of integration seem to be varied, and the biological impacts have yet to be elucidated. Whether the germline integrations that we have identified are simply accidents or, as we suspect, may sometimes provide the host with an important selectable advantage, can now be tested.
Analysis of genome integrations was conducted based on viral protein sequences available at NCBI FTP website (ftp://ftp.ncbi.nih.gov/refseq/release/viral/). Most recent sequences were downloaded on October 28, 2009. A total of 79,001 sequences were included in that distribution, with each representing an individual viral protein. This number slightly overestimates the actual number of unique sequences, as some proteins may be part of a polyprotein. However, the discrepancy is small, as a total of only 561 sequences are labeled as polyproteins. Finally, every individual virus encodes more than one protein.
The complete list of viral proteins was further narrowed down to include only single stranded RNA viruses with no known DNA phase in their replication. For this purpose, we used the list in the NCBI taxonomy database, downloaded on the same date as the viral protein sequences (http://www.ncbi.nlm.nih.gov/Taxonomy/). This screening procedure yielded 5,666 independent viral protein sequences. Again, small overlap is possible due to dual representation in polyproteins.
Viral sequences were then screened against publicly available genomic assemblies of 48 sequenced vertebrates and a few close siblings. Vertebrate sequences were downloaded from the UCSC genome website, and when not available, directly from the sequencing center websites or from the Ensembl database (release 56). The list of species considered is given in Table S9. The initial search was performed using BLAST 2.2.17 with parameters -p tblastn -M BLOSUM62 -e 1e-4.
A direct search produced 14,281 results, with BLAST E-value cutoff at 10−04. The vast majority of hits arose from homology between viral proteins and a few host proteins. By far the most widespread homology was between the gene for a 60–70 kDa protein in plant viruses and vertebrate heat shock proteins (HSP70 in humans). Similarly, several viral genes had homologies with GIMAP8, BIRC8, PARP14, and the DNAJC14 families of genes. We removed from further consideration any viral protein that had homology with known mRNAs in humans, cows and mice at the same time. Any integrations in this group would likely represent host pseudogenes, rather than integrations of viral origin.
As a final crosscheck, all integrated sequences were reverse-searched against all known nucleotide and protein sequences in the NCBI database using BLAST algorithm, to ensure that a putative integration is indeed from a virus with a single strand RNA genome, and is not a homologous protein from another virus or organism. Additionally, all reported sequences have 30–50% identity with the present-day virus proteins. These values are common for many homologous proteins in Ensembl database, and support an evolutionary relationship between the integrated sequences we have identified and present day virus proteins.
Altogether we identified strong hits from seven viral proteins from three different viruses/families (Table 1), all within the Order Mononegavirales (non-segmented single stranded negative sense RNA viruses). The sole exception that resembles a Flavivirus-derived gene is discussed below. All of the Mononegavirales-derived hits come from nucleocapsid (N, NP), and matrix (M) proteins, as well as the viral RNA-dependent DNA polymerase (L), and the polymerase complex cofactor (VP35). Additionally, weaker hits were associated with glycoproteins (G, GP) of the same viruses. Extra care has to be taken here, as glycoproteins are encoded in many viral genomes including retroviruses, which are commonly integrated in the germ lines. We did the following checks to eliminate potential retroviral glycoproteins from further consideration: regions of 10 kb extending both downstream and upstream of each potential glycoprotein-like integration were downloaded and checked for retroviral gag- and pol-genes, as well as for LTR-signatures. Retroviral pol genes were chosen for their highest conservation among all retroviral genes. Altogether, gag- and pol- genes were downloaded from approximately 50 different retrovirus families, and searched using blastx algorithm of the BLAST program, with E-value threshold of 10−3. Search for LTR-sequences was conducted using LTR-FIND tool (http://tlife.fudan.edu.cn/ltr_finder/) [44].
While all aforementioned integrations were related to members of the Mononegavirales, one putative integration on scaffold 1104 of medaka is most similar to a virus with a positive strand RNA genome, the Flavivirus, Tamana Bat virus. Integration with putative coordinates 26500-2900 on scaffold 1104 has low sequence similarity to Tamana Bat virus and several other Flaviviruses. However, sequence similarity of this integration is fairly low (BLAST value 10∧-7 for a 190 amino acid fragment of a 600 amino acid protein, with sequence identity of just 28%). Additionally, the entire scaffold is not yet mapped to a chromosome, has no known genes, and is not readily aligned with other species. It therefore remains to be seen if this is an actual integration of a positive-sense virus, some accidental sequence, or the result of laboratory contamination. The possibility of somatic cell integration, as opposed to germ-line integration, also remains open, as medaka sequencing relies on genomic DNA from adult bodies [45].
|
10.1371/journal.pntd.0001579 | Sensitive and Specific Target Sequences Selected from Retrotransposons of Schistosoma japonicum for the Diagnosis of Schistosomiasis | Schistosomiasis japonica is a serious debilitating and sometimes fatal disease. Accurate diagnostic tests play a key role in patient management and control of the disease. However, currently available diagnostic methods are not ideal, and the detection of the parasite DNA in blood samples has turned out to be one of the most promising tools for the diagnosis of schistosomiasis. In our previous investigations, a 230-bp sequence from the highly repetitive retrotransposon SjR2 was identified and it showed high sensitivity and specificity for detecting Schistosoma japonicum DNA in the sera of rabbit model and patients. Recently, 29 retrotransposons were found in S. japonicum genome by our group. The present study highlighted the key factors for selecting a new perspective sensitive target DNA sequence for the diagnosis of schistosomiasis, which can serve as example for other parasitic pathogens.
In this study, we demonstrated that the key factors based on the bioinformatic analysis for selecting target sequence are the higher genome proportion, repetitive complete copies and partial copies, and active ESTs than the others in the chromosome genome. New primers based on 25 novel retrotransposons and SjR2 were designed and their sensitivity and specificity for detecting S. japonicum DNA were compared. The results showed that a new 303-bp sequence from non-long terminal repeat (LTR) retrotransposon (SjCHGCS19) had high sensitivity and specificity. The 303-bp target sequence was amplified from the sera of rabbit model at 3 d post-infection by nested-PCR and it became negative at 17 weeks post-treatment. Furthermore, the percentage sensitivity of the nested-PCR was 97.67% in 43 serum samples of S. japonicum-infected patients.
Our findings highlighted the key factors based on the bioinformatic analysis for selecting target sequence from S. japonicum genome, which provide basis for establishing powerful molecular diagnostic techniques that can be used for monitoring early infection and therapy efficacy to support schistosomiasis control programs.
| Schistosomiasis remains a serious parasitic disease worldwide. Adequate case-finding is important and crucial for the effective control programs as schistosomiasis control programs are chiefly based on treatment of infected populations. However, the currently available diagnostic assays are not ideal, and DNA detection can provide useful alternative approaches for the sensitive and specific diagnosis of schistosomiasis, provided that reliable genetic markers are employed in the tests. However, different target sequences used for DNA detection of samples showed different sensitivity. In previous studies, we identified a 230-bp sequence of high sensitivity and specificity from the retrotransposon SjR2 of Schistosoma japonicum. Here, new primers based on 25 novel retrotransposons were designed and their sensitivities and specificities for detecting S. japonicum DNA were compared. Of these, a new 303-bp DNA sequence from the non-LTR retrotransposon (SjCHGCS19) had high sensitivity and specificity in detecting S. japonicum DNA. More importantly, we also found that both the SjCHGCS19 (303-bp sequence) and SjR2 (230-bp sequence) have high repetitive copies, higher genome proportions and active ESTs in the genome of S. japonicum. Our findings provide new insights into selecting suitable target sequences which may play a key role for the sensitive and specific detection of S. japonicum DNA.
| Schistosomiasis is caused by Schistosoma haematobium, S. mansoni, S. japonicum, and less frequently, S. mekongi and S. intercalatum. It occurs in the tropics and subtropics and it is among the most important parasitic diseases worldwide, with a significant socio-economic impact [1]. The disease affects people living in 74 endemic countries, with approximately 120 million individuals being symptomatically infected and 20 million being severely affected [2], [3]. Moreover, schistosomiasis represents an increasing problem in non-endemic areas, due to the growing number of immigrants and tourists [4], [5]. As schistosomiasis control programs are chiefly based on treatment of infected populations, adequate case-finding is important for the effective consecution of the control programs. Herein, diagnosis plays a crucial role in the monitoring of early infection and therapy efficacy. However, the currently available diagnostic assays are not ideal, since the examination of eggs in stools, such as Kato-Katz assay, and detection of circulating antigens lack sensitivity due to low disease prevalence, post-treatment situations where chemotherapeutic agents mask the presence of existing disease and methodologically, antibody detection lacks specificity [6]. In addition, current ELISA methods cannot be used to evaluate the efficacy of chemical treatment as IgG antibody levels remained elevated despite the fact that the disease was cured, indicating that there may be false positive results [7].
In the last several years, several research groups have developed specific and sensitive PCR-based methods for detecting S. japonicum, S. mansoni and S. haematobium DNA from humans and the intermediate hosts [7]–[15]. These PCR assays have proven useful alternatives for the accurate diagnosis of human schistosomiasis. Numerous factors influence the sensitivity and specificity of PCR assays for the diagnosis of schistosomiasis, in particular, the target sequences selected for PCR amplification. Several research groups have developed PCR methods using various target sequences and these PCR assays showed different sensitivities for detection of Schistosoma genomic DNA. An 121-bp rDNA sequence was the major target sequence for detecting S. mansoni [8], [9], [11], [13]. Mitochondrial NADH I gene (nad1) has been used as genetic marker for detection of S. mansoni and S. japonicum DNA [15]–[17].
SjR2, a new RTE-like, non-long terminal repeat retrotransposon from S. japonicum, is firstly described by Laha et al. [18], and its 230-bp sequence was first used as target sequence in PCR and LAMP assays for detecting S. japonicum DNA [7], [12]. Recently, 29 retrotransposons were identified in the genome sequence of S. japonicum [19], including known Gulliver, SjR1, SjR2 and Sj-pido elements as well as 25 novel elements. In the present study, primers targeting these 25 novel retrotransposons were designed and used in PCR assays for detecting S. japonicum DNA, and their sensitivities and specificities were compared.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committ-ee on the Ethics of Animal Experiments of the Soochow University (Permit Number: 2007–13).
All surgeries were performed under sodium pentobarbital anesthesia, and all efforts were made to mini-mize suffering of animals.
Serum samples of healthy individuals were obtained from Suzhou, Jiangsu Province, China. The protocol was approved by the Scientific Committee of the Soochow University. Serum samples of patients were obtained from the endemic area of Hunan Province, China. Written informed consent was obtained from all donors. Ethical clearance for the project was obtained from the Scientific Committee at the Hunan Institute of Parasitic Diseases, which is responsible for schistosomiasis control within the Hunan Province where targeted villages are located.
Schistosome-infected snails were obtained from Jiangsu Institute of Parasitic Diseases, China. The institute provided live Oncomelania hupensis snails exposed to the Chinese strains of S. japonicum. All living snails were putting into a tray which was filled with 4/5 volume of water and exposed to a light source to induce shedding of live S. japonicum cercariae.
Clonochis sinensis adult worms were provided by Prof. Kuiyang Zheng, Xuzhou Medical College, China, Trichinella spiralis adult worms were provided by Prof. Zhongquan Wang, Medical College of Zhengzhou University, China, and S. mansoni adult worms were obtained from Dr. Donato Cioli, Institute of Cell Biology, Monterotondo, Italy.
Six female New Zealand rabbits, weighing 1.8–2.2 kg, were randomly divided into three groups of two rabbits each. Blood collected before infection from all rabbits was served as negative control. Group I and Group II were percutaneously infected with 200 mixed sexual cercariae of S. japonicum (light infection), and Group III was percutaneously infected with 500 mixed sexual cercariae of S. japonicum (medium infection). On the seventh week, the eggs were detected in the feces. The EPG was 96 (Group I), 92 (Group II) and 397 (Group III), respectively. In Group I, between 1 and 7 weeks post-infection, blood were collected weekly. On the seventh week, the rabbits were anaesthetized with sodium pentobarbital and the adult worms and liver samples were collected. Between 1 and 24 weeks post-infection, blood was collected from rabbits in Group II. Serum of each rabbit was separated by centrifugation (1500 g for 15 min) and kept at −80°C until use.
The rabbits infected with S. japonicum in Group III were treated with two doses of 150 mg/kg praziquantel on the seventh and eighth week post-infection. In this group, blood was collected weekly between 1 and 30 weeks post-infection (23 weeks post-treatment) and all of the rabbits were sacrificed on the last week, and no adult worms were found in portal system. Serum of each rabbits was separated by centrifugation (1500 g for 15 min) and kept at −80°C until use.
Forty-three S. japonicum-positive serum samples of patients with positive Kato-Katz stool examination results were obtained from the endemic areas in Hunan Province, China, and the egg counts of these patients ranged from 8 to 1160 eggs per gram (EPG). 51 serum samples from healthy individuals were collected from healthy donors at the Center of Health Examination, the First Affiliated Hospital of Soochow University, Suzhou, Jiangsu Province, China, and were used as negative control to evaluate the specificity of the PCR assay.
DNA from all of the collected samples was extracted using the method described by Steiner et al [20]. as modified by Xia et al. [7]. Briefly, 0.2 g liver homogenate samples was mixed with 200 µl distilled water, grinded and dissolved in 200 µl Triton-X-100 (10 mMol/L Tris-HCL; 0.45% Triton-X-100; 0.45% NP-40; 300 µg/mL protease K, pH 7.4) and were incubated at 60°C for 2 h with vigorous agitation for 10 min on a shaker, and then centrifuged at 12,000 g for 10 min. 300 µl supernatant of the liver homogenate was collected in a separate tube and incubated at 100°C for 10 min, centrifuged at 12,000 g for 10 min and the supernatant was collected.
DNA from adult worms was also extracted using the method modified by Xia et al [7]. Five adult worms of each S. japonicum, S. mansoni, C. sinensis and T. spiralis were homogenized in 300 µl physiological saline and then digested with equal volume of extraction buffer containing 3 mg/ml proteinase K, 0.1 mol/L Tris-HCl, pH 8.5, 0.05 mol/L EDTA, and 1% SDS. Then the mixture was incubated at 60°C for 1 h.
200 µl serum of infected rabbits or humans were diluted in 400 µl serum extraction buffer containing 150 mol/L NaCl, 10 mol/L EDTA, 10 mol/L Tris-HCl, pH 7.6, 2% SDS, 5 µg/ml salmon sperm DNA, 4 µg 25 mg/ml proteinase K, and were incubated at 37°C overnight.
All of the extraction mixtures (liver homogenate, adult samples and serum) were extracted twice with phenol, chloroform, isoamyl alcohol (25∶24∶1), once with chloroform, isoamyl alcohol (24∶1), and then precipitated with dehydrated alcohol and 3 M sodium acetate. The supernatant was discarded and the pellet was washed twice with 1 mL of 75% ethanol, finally the pellet was left to dry at 37°C for 30 min and then re-suspended with 100 µl of TE (10 mmol/L Tris-HCl, 1 mmol/L EDTA, pH 8.0) for DNA of liver homogenates and adult worms, and 20 µl of TE (10 mM Tris-HCl, 1 mM EDTA, pH 8.0) for DNA of serum samples, respectively.
The DNA extracted from adult worms, liver and serum were used as the template. Primers were designed targeting the 25 novel retrotransposons repeat DNA sequences of S. japonicum (Table S1). PCR reaction (25 µl) contained 2.5 µl of buffer, 1.5 µl of 25 mmol/L MgCl2, 2 µl of 2.5 mmol/L dNTP, 0.5 µl of each 20 pmol/L primer, 0.4 µl of 5 U/L Taq polymerase (Takara) and 4 µl of template. The conditions for PCR were as follows (with the exception of SjCHGCS4 and SjCHGCS6 using 2-step PCR: 94°C for 4 min, followed by 30 cycles of 94°C for 30 s, 68°C for 60 s, and a final extension of 72°C for 7 min): 94°C for 3 min, followed by 35 cycles of 94°C for 60 s, Tm/(list in Table S1) for 60 s, 72°C for 60 s and a final extension of 72°C for 7 min. The PCR was performed using GeneAmp PCR System (Eppendorf, Hamburg, Germany). Finally, a 5 µl aliquot of the PCR product was run on a 2.0% agarose gel along with DNA ladder marker (Takara, Dalian) in TBE buffer containing 0.5 µg/ml ethidium bromide, and the bands were visualized under UV light on a transilluminator.
The nested-PCR reaction system was similar to the normal PCR as described above. The first-round amplification was carried out in the same manner except that the degenerate temperature was 55°C. Primers were designed on the basis of the SjCHGCS19 retrotransposons repreat DNA sequence of S. japonicum, the primers employed were P3 (5′-CCAAATCGCAACACTACGC-3′ (forward) and P4 (5′-ATCGGATTCTCCTTGTTCAT-3′) (reverse). DNA samples extracted from serum of rabbits and humans were used as the template. The expected length of the amplification product was 607 bp. Second-round amplification (nested-PCR) was carried out in the same manner as the first-round except that the DNA sample was a 1∶10000 dilution of the first-round PCR product and the degenerate temperature was 65°C. The sequences of these primers are listed in Table S1. The expected length of the amplification product was 303 bp.
The first-round PCR product targeting SjCHGCS19 was cloned into plasmid by means of a pMD20-T II cloning reagent Kit (Tiangen, Beijing, China). Plasmid purification was done with a TIAN pure Mini Plasmid Kit (Tiangen, Beijing, China). Plasmids were quantified by spectrophotometry. Sequencing of the cloned amplification product confirmed that it was identical to part of the S. japonicum retrotransposon SjCHGCS19. The standard plasmid was tested in 10-dilution series by nested-PCR.
Primers were designed from the 25 novel retrotransposons of S. japonicum and a series of diluted genomic DNA of S. japonicum adults were used as the template. The target fragments were amplified by nested-PCR assay and the minimum amounts detectable were different among 25 new retrotransposons and SjR2. In addition to the 230-bp fragment from SjR2, a new 303-bp fragment from non-LTR retrotransposon (SjCHGCS19) displayed high sensitivity and specificity. Bioinformatic analysis showed that both the SjCHGCS19 (303-bp sequence) and SjR2 (230-bp sequence) have higher genome proportions (4.09% and 4.43%), higher repetitive complete copies (793 and 400) and partial copies (17,373 and 23,755), and higher EST numbers (39 and 79) than the other retrotranposons in the genome of S. japonicum (Table 1).
Furthermore, to determine the limit of the 303-bp DNA fragment for detecting S. japonicum DNA, 10-fold serial dilutions of the standard plasmid clones of SjCHGCS19 and SjR2 were amplified by nested-PCR assay. The minimum detectable amount of standard plasmid from SjCHGCS19 was only 2.02 copies, whereas the 230-bp fragment from SjR2 was 10.2 copies.
In addition, the 303-bp DNA fragment was amplified by nested-PCR assay from male and female adults of S. japonicum, from liver homogenate and sera of infected rabbits (Figure 1). Interestingly, the target DNA was amplifiable from both S. japonicum and S. mansoni, but no cross-reaction was detected with DNA samples representing C. sinensis and T. spirals (Figure 2).
To clarify the potential diagnostic value of the 303-bp DNA fragment for early infection and chemotherapy evaluation in the rabbit model, a series of rabbit sera in Group II (light infection) and Group III (praziquantel treatment) were examined by nested-PCR assay. As shown in Figure 3, the 303-bp DNA fragment was amplified at 3 d sera post-infection in Group II and could be amplified until 24 weeks post-infection by nested-PCR assay. The 303-bp DNA fragment was amplifiable between 1 and 24 weeks post-infection and it became negative from 25 weeks post-infection (17 weeks after praziquantel treatment) in Group III (Figure 4).
To illuminate the clinical utility of the 303-bp DNA from SjCHGCS19 as the target sequence for the diagnosis of human schistomomiasis, 43 human serum samples from patients with S. japonicum infection confirmed by stool examination and 51 serum samples from normal healthy individuals were examined. Of the 43 patient serum samples, 42 (97.67%) were positive by the nested-PCR assay. Of the 51 serum samples from healthy individuals, 49 (96.07%) were detected as negative by the nested-PCR assay. the percentages of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for nested-PCR assay were 97.67%, 96.07%, 97.67% and 96.07%%, respectively. For patients with heavy infection with S. japonicum (EPGs≥400) and medium infection (100≤EPG<400), the percentage of sensitivity was 100%. For patients with light infection (EPG<100), the percentage of sensitivity was 96.87% (Table 2, Figure S1).
Because schistosomiasis control programs are chiefly based on treatment of infected populations, adequate case-finding is important for the effective consecution of the control programs. Herein, accurate diagnosis plays a crucial role in the monitoring of early infection and therapy evaluation. Nucleic acid-based diagnosis has been used in the clinical testing of a wide variety of pathogenic infections, such as human immunodeficiency virus [21], Mycobacterium tuberculosis [22], Plasmodium falciparum [23], Trypanosoma cruzi [24] and Leishmania braziliensis [25]. In parasitic diseases such as schistosomiasis, it was possible to detect cell-free parasite DNA circulating in plasma, and this could be used to diagnose schistosomiasis. Importantly, nucleic acid-based diagnostic methods have the same diagnostic value as parasitological diagnostic methods. Recently, several PCR-based methods for detecting Schistosoma DNA from various samples have been developed [7]–[15]. Numerous factors may influence the sensitivity and specificity of PCR assays for the diagnosis of schistosomiasis, in particular the target sequences selected.
A 121-bp tandem repeat rDNA sequence was the major target sequence for detecting S. mansoni DNA in mouse serum samples and in human plasma by PCR [11], [13]. Although a real-time PCR assay using nad1 as target sequence for detecting S. japonicum DNA has proven highly sensitive, even for samples containing less than 10 EPG, it was negative for examining serum and urine samples from the infected pigs [14].
While previous efforts of PCR diagnosis of S. mansoni and S. japonicum infections have been relied on mitochondrial or rDNA sequences, we have focused on the sequence of the highly repetitive retrotransposon SjR2 in S. japonicum. In our previous studies, the 230-bp sequence from the highly repetitive retrotransposon SjR2 of S. japonicum was used as target sequence and it showed high sensitivity and specificity in detecting S. japonicum DNA, and the 230-bp fragment was amplified with DNA equivalent of 1.1 egg from feces [7]. In particular, the 230-bp sequence was able to be amplified from the sera of the rabbit model at 1 week post-infection, which is one week earlier than that of the 121-bp sequence in the mouse-S. mansoni model [7], [11], [12].
SjR2 was 3.9 kb in length and was constituted of a single open reading frame encoding a polyprotein with apurinic/apyrimidinic endonuclease and reverse transcriptase domains [18]. Phylogenetic analyses based on conserved domains of reverse transcriptase or endonuclease revealed that SjR2 belonged to the RTE clade of non-long terminal repeat retrotransposons and SR2 elements are members of a non-LTR retrotransposons typified by the RTE-1 non-LTR retrotransposon of Caenorhabditis elegans[26]. According to the phylogenetic tree, both SjCHGCS19 and SjR2 belonged to RTE clade of non-LTR retrontransposon (Figure S2). SjCHGCS19 showed the closest relationship with Perere-3 of S. mansoni, while SjR2 located in the same branch with SR2. At amino acids level, SjCHGCS19 showed 74% identity with Perere-3, while only 30% identity with SR2 and SjR2. Between SR2 and SjR2, the polyprotein showed 55% identity. Furthermore, hybridization analyses indicated that 10,000 copies of SjR2 were dispersed throughout the S. japonicum genome, accounting for up to 14% of the nuclear genome [18]. In our previous study, 29 retrotransposons were identified, including the known Gulliver, SjR1, SjR2 and Sj-pido elements as well as 25 novel elements, together constituting 19.8% of the genome. Of the 25 novel retrotransposons, 18 were LTR forms, four were non-LTR forms and three were Penelope-like elements—enigmatic retroelements that retain introns. Each type of retrotransposons was represented by 1 to 793 intact copies or hundreds to thousands of partial copies [19]. The non-LTR retrotransposons such as SjR1, SjR2, Sj-pido, SjCHGCS19, SjCHGCS20, SjCHGCS21 and SjCHGCS22 have significantly higher copy numbers, constituting 12.6% of the genome. We then wanted to determine whether there are any associations between the numbers of copies of target sequences and the sensitivity of the PCR detection.
In the present study, primers were designed to amplify the 25 novel S. japonicum retrotransposons. The target fragments were amplified by nested-PCR assay for comparing sensitivity and specificity of detecting S. japonicum DNA of the target sequences. The results showed that a new 303-bp fragment from the highly repetitive retrotransposon, SjCHGCS19, had high sensitivity for detecting S. japonicum DNA in addition to the 230-bp fragment from SjR2. Importantly, bioinformatic analysis of 26 S. japonicum retrotransposons showed that both the SjCHGCS19 (303-bp sequence) and SjR2 (230-bp sequence) have higher genome proportions, repetitive complete copies and partial copies, and active ESTs than the others in the chromosome genome (Table 1).
The minimum amount of the standard plasmid detectable using nested-PCR assay was 2.02 copies per reaction using the 303-bp target sequence from SjCHGCS19, whereas 10.2 copies per reaction of S. japonicum DNA was detected targeting the 230-bp sequence from SjR2 by the nested-PCR assay. This indicated that the 303-bp fragment from SjCHGCS19, as target sequence for detecting S. japonicum DNA, was more sensitive than the 230-bp fragment from SjR2 previously identified.
Additionally, as shown in Figure 1, we evaluated the specificity of the 303-bp target sequence. The expected product was amplified by the nested-PCR assay from S. japonicum male and female adults, from liver homogenate and from sera of infected rabbits. We did not find any false positive bands in non-infected rabbit sera. Interestingly, the target DNA was amplified from both S. japonicum and S. mansoni, but no cross-reaction was detected in DNA samples representing C. sinensis and T. spirals (Figure 2).
Furthermore, in the present study, the 303-bp sequence was amplified in rabbit sera at 3 d day post-infection, which is 4 d earlier than that of the 230-bp sequence in the rabbit-S. japonicum model by nested-PCR (Figure 3). The results showed that higher sensitivity is achieved using the 303-bp sequence for the detection of S. japonicum DNA in serum samples. In addition to the obvious advantage of an early diagnosis, our findings showed that the 303-bp target sequence might be valuable for the evaluation of chemotherapy efficacy, and it became negative at 25th week post-infection (17 weeks after praziquantel treatment) by nested-PCR, and was 7 weeks longer than detection by using the 230-bp sequence from SjR2, indicating a higher sensitivity of the 303-bp sequence than that of the 230-bp sequence (Figure 4).
The effectiveness of the 303-bp target sequence was validated by examining serum samples from patients infected with S. japonicum (Figure S1). The findings showed that the sensitivity was 97.67%, and the specificity was 96.07% (Table 2). In particular, the PPV was 100% in patients with heavy infection (EPGs≧400) and medium infection (100≦EPG<400), indicating its high sensitivity. For patients with light infection (EPG<100), the percentage of sensitivity was 96.87%. Only 1 serum sample was detected as false negative, and this sample was from a patient with particularly low infection (EPG = 56). This could be due to little S. japonicum DNA in human blood circulation and/or DNA loss during template extraction procedures. There appeared to be a correlation between PCR results and the EPGs of the patients (Table 2). However, in this study, the examined number of patient serum samples was small. It is imperative to carry out further investigation using a large number of patient serum samples.
Since the prevalence of human infection with S. japonicum has been decreasing year by year [27], the rapid and reliable diagnosis of Schistosoma infection is central to the control as well as to the environmental monitoring and disease surveillance, especially for evaluation of treatment efficacy. DNA amplification assays provide alternative approaches for sensitive and specific diagnosis of Schistosoma infection, provided that reliable genetic markers are employed in the tests. Our results demonstrated that this new 303-bp sequence from non-LTR retrotransposon (SjCHGCS19) had high sensitivity and specificity for the detection of S. japonicum DNA, which may provide the new target sequence useful for the early diagnosis and for the evaluation of chemotherapy efficacy of schistosomiasis. More importantly, although many factors may affect the sensitivity and specificity in the target sequence selection, such as the length of target sequence, the specificity of the conserved sequence, the present study highlighted the key factors based on bioinformatic analysis for selecting a new perspective sensitive target sequence from genome sequences, which provides new insights into selecting suitable target sequence which may play a key role for the sensitive and specific detection of Schistosoma DNA. These findings would provide basis for establishing powerful molecular diagnostic techniques that can be used in clinical settings and as laboratory tools for surveillance and for environmental monitoring to support schistosomiasis control programs.
|
10.1371/journal.pbio.3000041 | An alternative mode of epithelial polarity in the Drosophila midgut | Apical–basal polarity is essential for the formation and function of epithelial tissues, whereas loss of polarity is a hallmark of tumours. Studies in Drosophila have identified conserved polarity factors that define the apical (Crumbs, Stardust, Par-6, atypical protein kinase C [aPKC]), junctional (Bazooka [Baz]/Par-3), and basolateral (Scribbled [Scrib], Discs large [Dlg], Lethal [2] giant larvae [Lgl]) domains of epithelial cells. Because these conserved factors mark equivalent domains in diverse types of vertebrate and invertebrate epithelia, it is generally assumed that this system underlies polarity in all epithelia. Here, we show that this is not the case, as none of these canonical factors are required for the polarisation of the endodermal epithelium of the Drosophila adult midgut. Furthermore, like vertebrate epithelia but not other Drosophila epithelia, the midgut epithelium forms occluding junctions above adherens junctions (AJs) and requires the integrin adhesion complex for polarity. Thus, Drosophila contains two types of epithelia that polarise by fundamentally different mechanisms. This diversity of epithelial types may reflect their different developmental origins, junctional arrangement, or whether they polarise in an apical–basal direction or vice versa. Since knock-outs of canonical polarity factors in vertebrates often have little or no effect on epithelial polarity and the Drosophila midgut shares several common features with vertebrate epithelia, this diversity of polarity mechanisms is likely to be conserved in other animals.
| The Drosophila midgut is lined by a single-layered epithelium that acts as a barrier to the environment while allowing for nutrient uptake and related physiological processes. To fulfil these roles, midgut epithelial cells are highly polarised, with a pronounced asymmetric distribution of cellular components. Previous work in Drosophila revealed a conserved set of factors governing cell polarity, and it is thought that this network of proteins underlies all examples of polarity in Drosophila (and other organisms). Here, we demonstrate that the epithelial cells in the Drosophila midgut are not polarised by these canonical polarity factors but instead rely on the integrin adhesion complex. Thus, there are two types of epithelia in Drosophila that polarise using fundamentally different mechanisms. This diversity may reflect a difference in developmental origin (endodermal versus ectodermal), a difference in junctional arrangement, or the direction in which the respective cells polarise. Since knock-outs of canonical polarity factors often have little or no effect on epithelial polarity in vertebrate model systems, this diversity of polarity mechanisms might be conserved in other organisms.
| Most animal organs and tissues are composed of epithelial cells that adhere laterally to each other to form sheets that act as barriers between compartments. The formation of epithelial monolayers depends on the coordinated polarisation of each cell along its apical–basal axis, and this polarity underlies all aspects of epithelial biology [1,2]. For example, the function of epithelia as barriers to fluids and pathogens depends on the correct positioning of the occluding cell–cell junctions (septate junctions [SJs] in invertebrates and tight junctions in vertebrates), whereas the adhesion between cells depends on the lateral localisation of cadherin-dependent adherens junctions (AJs).
Much of our understanding of how epithelial cells polarise comes from studies of Drosophila melanogaster that have identified a conserved set of epithelial polarity factors that define different cortical domains along the apical–basal axis of the cell. The apical domain is specified by the transmembrane protein Crumbs, the adaptor protein Stardust, and the Par-6/atypical protein kinase C (aPKC) complex; the boundary between the apical and lateral domains is defined by Bazooka (Baz, Par-3 in other organisms), which positions the apical-most lateral junction; and the rest of the lateral domain is marked by Scribbled (Scrib), Discs large (Dlg), and Lethal (2) giant larvae (Lgl) [3]. Null mutations in any of these factors disrupt epithelial polarity in the primary epithelium that forms from the cellular blastoderm of the Drosophila embryo and gives rise to most of the structures of the larva and adult [4–11]. Similarly, loss of any of these genes disrupts the secondary epithelium formed by the follicle cells that surround the developing oocyte [12–14]. In each tissue, Baz seems to play a pivotal role in positioning the apical AJs and in localising the apical factors, which then exclude Baz from the apical domain [15–19]. The identity of the apical and lateral domains is then maintained by mutual antagonism between apical and lateral factors [20,21]. The requirement for some of these factors becomes less stringent in polarised epithelia as they mature. For example, Crumbs is particularly important in epithelial tissues that are remodelling their cell junctions as they undergo morphogenetic rearrangements, and Scrib, Dlg, and Lgl are not required to maintain polarity in mid-embryogenesis, as the Yurt group of lateral proteins takes over the antagonism of the apical factors, although Scrib and Dlg are still required for the formation of the SJs [20–24].
Epithelial cells are thought to have evolved at the origin of multicellularity, as cells first started to adhere to each other to form sheets, suggesting that apical–basal polarity is an ancient invention that is controlled by a conserved mechanism [25]. In support of this view, all of the canonical epithelial factors in Drosophila are conserved in vertebrates and localise to the equivalent cortical domains [26–33]. Knock-downs or knock-outs of some of these factors impair polarity in certain contexts, such as in Xenopus embryonic blastomeres and in some cultured cell lines [34–37]. In most cases, however, knock-outs of canonical polarity factors have little or no effect on polarity or cause unrelated phenotypes. For example, mice homozygous for a null mutation in PAR-3 die in mid-gestation from a heart defect caused by a failure of the epicardial cell migration, but other embryonic epithelia appear to form normally [38]. Similarly, knock-out of PAR-3 in mouse mammary stem cells disrupts the morphogenesis of the mammary gland but not the ability of the cells to polarise [39,40]. Finally, Scrib functions as a planar cell polarity gene in mice but has no obvious effect on apical–basal polarity [41]. Although this lack of polarity phenotypes in mammals could be a result of redundancy between paralogues, it raises the possibility that at least some vertebrate epithelia polarise by different mechanisms from the model Drosophila epithelia. In support of this view, many mammalian epithelia require integrin adhesion to the basement membrane to orient their polarity, whereas the well-characterised Drosophila epithelia do not [42–44]. Furthermore, vertebrate epithelial cells have an inverted arrangement of junctions compared to insect epithelia: the apical junction in vertebrates is the occluding, tight junction, which forms at the apical/lateral boundary above a lateral zonula adherens (ZA), whereas insect cells form lateral SJs, below an apical ZA [45–47].
The possibility that the polarisation of some epithelia is independent of the canonical polarity system prompted us to ask if all Drosophila epithelia polarise in the same way. We therefore examined polarity in the Drosophila adult midgut epithelium, which is mainly absorptive rather than secretory and is endodermal in origin, unlike the well-characterised epithelia, which are secretory and arise from the ectoderm or mesoderm [48]. The adult midgut is a homeostatic tissue in which basal intestinal stem cells (ISCs in Fig 1F) divide to produce new cells that integrate into the epithelium to replace dying enterocytes (ECs), which are shed into the gut lumen. This has the advantage that one can generate homozygous mutant stem cell clones in heterozygotes at the late pupal or adult stage to produce clonal patches of mutant ECs in the adult midgut without disrupting the development of the tissue. Our results reveal that the polarisation of the midgut epithelium does not require any of the canonical polarity factors and has several features in common with vertebrate epithelia, making it a useful model for investigating alternative polarity pathways.
The midgut is a typical epithelium with an apical brush border marked by F-actin (Fig 1A) and phospho-Moesin (pMoe) (Fig 1B) as well as Myosin IA (Fig 1E) and an apical domain marked by Myosin 7a(Fig 1C). However, our analysis led us to rediscover an interesting property of this epithelium: the smooth SJs, which form the occluding barrier to paracellular diffusion, form at the apical side of the lateral domain, above AJs, which are diffusely distributed over the lateral domain [49–52] (Fig 1D and 1F). This is the opposite way around compared to other Drosophila epithelia, in which the AJs condense to form a ZA around the apical margin of the cell, with the SJs, if present, positioned more basally in the lateral membrane. The organisation of junctions in the midgut therefore resembles the junctional arrangement in mammals, in which the occluding tight junctions form above the AJ.
In secretory epithelia, the Crumbs/Stardust complex defines the apical and marginal region and anchors the Par-6/aPKC complex in this domain [5,9,53]. Apical Crb and aPKC then exclude Baz/Par-3 to define the apical/lateral boundary by positioning the apical AJs [17–19]. This raises the question of whether these factors mark the same positions or the same structures when the junctions are reversed in the midgut.
Crumbs is not detectable in the adult midgut epithelium, as has previously been observed in the embryo [4] (Fig 2A). We used the mosaic analysis with a repressible cell marker (MARCM) technique [54] to generate positively marked clones in the adult midgut epithelium for null mutations in crumbs or stardust (crb8F–105, crb11A22; sdtk85). The adults were then dissected 8 to 10 days after clone induction to allow the stem cells, which divide about once a day, to go through multiple divisions because this rules out any perdurance of wild-type proteins expressed before the clones were induced. However, clones of crb and sdt null mutations give no obvious phenotypes (Fig 2B, S1A, S1D and S1E Fig). Overexpression of Crumbs expands the apical domain in other Drosophila epithelia [55]. By contrast, ectopic expression of Crumbs in the adult midgut epithelium does not affect EC polarity and the Crumbs protein does not localise apically, concentrating instead in the basal labyrinth (BL), an extensive set of tubular membrane invaginations from the EC basal surface (Fig 2C). Baz/Par-3 is also not detectable in ECs, although it is expressed in the ISCs, in which it localises apically as previously reported [56] (Fig 2D). MARCM clones homozygous for baz null alleles (baz815–8; baz4) occur at a similar frequency to wild-type clones and contain a normal number of cells, arguing against a role for Baz in asymmetric stem cell division, consistent with the view that these divisions are random and largely symmetric [57]. More importantly, baz null mutant ECs integrate into the epithelium and develop normal apical–basal polarity, in contrast to other epithelia in Drosophila (Fig 2E, S1B, S1C and S1F Fig). Baz-green fluorescent protein (GFP) localises apically and to the AJs when ectopically expressed in the midgut epithelium, consistent with binding to the Par-6/aPKC complex and to E-cadherin complexes (Fig 2F). However, ectopic Baz expression has no effect on the polarity of the ECs or on the formation of an apical SJ.
Both aPKC and Par-6 are expressed in the midgut and localise apically, as they do in all other epithelia (Fig 3A and 3B). In most polarised cells, the apical localisation of the Par-6/aPKC complex depends on Baz/Par-3, and in epithelia, this also requires Crumbs and Stardust [2]. Consistent with our observation that these proteins are absent from ECs, neither Baz nor Crb are required for the localisation of Par-6, indicating that the latter must be targeted apically by a distinct mechanism (S1A and S1B Fig). Surprisingly, the apical domain forms normally in par-6Δ226 and aPKCk06043 mutant clones, and the morphology of the cells is unaffected (Fig 3C, S1H and S1J Fig). Although aPKCk06043 is considered a null allele, the corresponding P element insertion does not disrupt the shorter isoforms of aPKC. Thus, it is conceivable that some aPKC activity remains in aPKCk06043 homozygous mutant cells. We therefore used CRISPR to generate a complete null, aPKCHC, a frameshift mutation resulting in a premature stop codon and a truncation of aPKC at amino acid 409, which is located in the middle of the kinase domain (S1L Fig). Homozygous aPKCHC clones also show no phenotype, forming normal actin-rich brush borders and SJs, confirming that aPKC is dispensable for EC polarity (Fig 3D and 3E). Nevertheless, Par-6 is lost from the apical domain in aPKCk06043 clones and aPKC is not apical in par-6Δ226 clones, showing that their localisations are interdependent (Fig 3F and 3G). Thus, the Par-6/aPKC complex still marks the apical domain of the midgut epithelium but is not required for its formation or maintenance.
The lateral polarity factors Scrib, Dlg, and Lgl are all expressed in the midgut and colocalise with each other to the SJs, marked by the conserved SJ component, Coracle [58] (Fig 1B and 1C, S2A Fig). Since the SJs form at the apical side of the lateral membrane in the midgut, in the position occupied by the AJs in other Drosophila epithelia, these proteins mark a conserved structure rather than a conserved position. The lateral epithelial polarity factors are required for the formation of pleated SJs in the embryo [20,21,59]. However, the smooth SJs form normally in scrib, dlg, and lgl mutant clones or when these factors are depleted by RNA interference (RNAi) (Fig 4A and 4C, S2B Fig). The apical domain is also unaffected in scrib, dlg, and lgl mutant or knock-down cells, in contrast to other epithelia in which apical factors are mislocalised to the basolateral domain (Fig 4E and 4F, S2C and S2D Fig). In stage 13 embryos, the Yurt complex excludes apical factors from the lateral membrane in place of Scrib, Dlg, and Lgl [23,24]. We therefore also examined the role of Yurt in the midgut. Yurt localises to the SJs, as it does in the embryo, but yurt mutant clones still polarise normally and form SJs that recruit Lgl (Fig 5). Finally, we examined Par-1, which localises laterally in other epithelia in which it plays a role in localising AJs through phosphorylation of Baz and in organising the microtubule cytoskeleton [60,61]. Par-1 is expressed in the ISCs, in which it localises laterally but is not detectable in the ECs. Consistent with its lack of expression, par-19 mutant ECs show normal apical–basal polarity (S3A and S3B Fig). Thus, all the canonical epithelial polarity factors are dispensable for the polarisation of the midgut epithelium, even though Par-6, aPKC, Scrib, Dlg, Lgl, and Yurt are expressed and localise to equivalent positions to secretory epithelia.
The relationships between the lateral factors has been difficult to assess in other epithelia because mutants in scrib, dlg, and lgl give rise to round, unpolarised cells without an identifiable lateral domain [11]. We took advantage of the normal EC polarisation in these mutants to investigate the interdependence of their recruitment to the SJs. Neither Dlg nor Lgl are recruited to SJs in cells depleted of Scrib by RNAi (Fig 4B and S2D Fig). Scrib localises normally in dlg mutant clones, whereas both Scrib and Dlg localise normally to the SJs in lgl mutant clones (Fig 4C–4F). Thus, there is a simple hierarchical relationship between these factors in the midgut epithelium, in which Scrib is required to recruit Dlg, which is needed for Lgl localisation.
The surprising observation that none of the classical epithelial polarity factors are required for the apical–basal polarisation of the midgut epithelium raises the question of how polarity is generated and maintained. Given the similar junctional arrangement to mammalian epithelia, we addressed whether polarity in midgut ECs depends on integrin-dependent adhesion to the extracellular matrix, as it does in several mammalian epithelia [42–44]. Components of the integrin adhesion complex, such as the α-integrin Mew and the essential cytoplasmic adaptor proteins Talin (Drosophila Rhea) [63] and Kindlin (Drosophila Fit 1 [64]; Fit 2 is not detectable expressed in the midgut) are highly localised to the basal side of the midgut epithelium (Fig 6A). The expression of two α-integrins and two β-integrins in the midgut complicates the genetic analysis of their function, so we focused on the cytoplasmic components of the integrin adhesion complex. Clones of cells homozygous for null alleles of rhea (rhea79a and rheaB128) detach from the basement membrane and fail to polarise, forming irregularly shaped cells that do not form SJs or an apical domain (Fig 6B). Most rhea mutant cells remain within the epithelial layer, below the SJs of the wild-type cells, probably because they do not form SJs themselves (Fig 6B and S4A Fig). Despite their inability to polarise and integrate into the epithelium, the mutant cells still appear to differentiate: they become polyploid, the size of their nuclei is not significantly different from that of wild-type cells (rhea nuclear long axis: 6.0 μm ± 1.0 μm versus heterozygous cells: 6.9 μm ± 0.9 μm), and they express the marker for differentiating ECs, Pdm1 [65] (Fig 6B). Cells mutant for both Fit1 and Fit2 show a similar phenotype to Talin mutants. Mutant cells have normal nuclear dimensions (Fit1 Fit2 nuclear long axis: 6.6 μm ± 1.2 μm versus heterozygous cells: 6.9 μm ± 1.0 μm) and express Pdm1 but fail to localise apical markers or form SJs (Fig 6C and S4B Fig).
The ISCs lie beneath the epithelium and differentiating ECs must therefore integrate into the epithelium from the basal side, inserting between the SJs of the flanking ECs while maintaining an intact barrier. We therefore examined the effects of mutations in the core SJ components Tsp2a and Mesh [51,52]. More than 90% of mutant cells fail to integrate through the SJs of the neighbouring wild-type cells, and the clones form clusters on the basal side of the epithelium (Fig 7A, S3A and S4C Figs). Nevertheless, the mutant cells still appear to differentiate normally, as shown by their nuclear size (Tsp2a nuclear long axis: 6.0 μm ± 0.9 μm versus heterozygous cells: 6.6 μm ± 0.8 μm) and Pdm1 expression (S4D Fig). Wild-type cells start to form an apical domain as they integrate, before they have a free apical surface, as shown by the enrichment of apical components such as Myo7a (arrowhead in Fig 7B). By contrast, apical markers only weakly localise in cells mutant for SJ components and never form a clear apical domain, even if the cells are extruded from the apical side of the epithelium (Fig 7B and 7C). Thus, the midgut epithelium appears to polarise in a basal to apical manner, in which adhesion to the ECM is required for the formation of the SJs and the SJs are needed for the formation of the apical domain (Fig 7D).
Our results reveal that the intestinal epithelium polarises by a fundamentally different mechanism from other Drosophila epithelia, as none of the classical epithelial polarity genes are required for its apical–basal polarisation. This cannot be attributed to redundancy between paralogues, as might be the case in vertebrates, because all of the polarity factors are single-copy genes in Drosophila. Thus, our observations provide strong evidence against the idea that there is a universal system for polarising epithelial cells. Nevertheless, a core set of the polarity factors (Par-6, aPKC, Scrib, Dlg, and Lgl) is expressed in the midgut epithelium, and these proteins localise to the equivalent positions to other Drosophila epithelia. Thus, they may still serve important functions in the midgut epithelium that are not essential for the overall apical–basal polarisation of the cells.
Given that all other epithelia in Drosophila use the canonical polarity pathway, our observations raise the question of why the midgut epithelium is different. This is unlikely to reflect the fact that the midgut is absorptive rather than secretory, as secretory cells in the midgut, such as the enteroendocrine (ee) cells and the acid-secreting copper cells, polarise in the same way as the ECs (S1J and S1K Fig). One key difference between the midgut epithelium and other epithelia is that it is the only epithelial tissue of endodermal origin in the fly, whereas all other epithelia are ectodermal or mesodermal. Thus, it is possible that endodermal epithelia are intrinsically different in the way that they polarise. In support of this view, it has recently been found that PAR-3, PAR-6, and aPKC are degraded in the invaginating endomesoderm of the Cnidarian Nematostella vectensis and are not required for this tissue to form an epithelium [67]. Thus, the difference between endodermal and ectodermal polarity systems may have evolved before the origin of Bilateria. The Drosophila midgut arises from the cellular blastoderm of the embryo and initially polarises in the same way as other embryonic epithelia before it undergoes an epithelial-to-mesenchymal transition (EMT) under the control of the endodermal GATA family transcription factor, Serpent [68]. Serpent drives EMT at least in part by inhibiting the transcription of crumbs and stardust and might therefore contribute to the switch in polarity mechanisms. However, Serpent is turned off as the midgut primordium migrates and is not expressed when the cells reform an epithelium to generate the midgut tube. Indeed, continued expression of Serpent blocks the cells from re-epithelising after migration. Thus, a pulse of Serpent expression may trigger the switch in polarity mechanisms, perhaps through downstream transcription factors, but Serpent itself prevents epithelial polarisation.
A second important difference between the adult midgut epithelium and other epithelia in Drosophila is its reversed arrangement of occluding junctions and AJs, with apical SJs forming above lateral AJs. In other Drosophila epithelia, Baz plays a key role in concentrating the AJs at the apical margin of the lateral membrane to form the ZA [69]. Thus, the absence of Baz in midgut ECs may contribute to the absence of an apical ZA. This cannot be the only factor making the midgut different, however, as ectopic expression of Baz in midgut ECs does not alter the position of the SJs. One reason why the SJs may have evolved to form above the AJs in the midgut is that this places the barrier to paracellular diffusion apically, thereby preventing the contents of the gut lumen from accessing the lateral sides of the cell, which is presumably important because the gut is full of digestive enzymes and potential pathogens. Other Drosophila epithelia that face the external environment—such as the epidermis, trachea, foregut, and hindgut—secrete an impermeable cuticle, which provides a protective covering to prevent pathogens from accessing the cell surface [70]. The development of the typical arthropod cuticular exoskeleton may therefore have freed these epithelia from the need to place their occluding junctions apically, allowing the apical positioning of the ZA.
The third important difference between the midgut epithelium and other epithelia is that the ECs polarise in a basal-to-apical direction as they integrate into the epithelium, whereas all other Drosophila epithelia polarise in an apical-to-basal direction. For example, the primary embryonic epithelium forms during cellularisation, as the furrow canals grow in from the apical surface of the embryo between the nuclei, while the follicle cells polarise in response to an apical cue from the germ line [14,71]. Thus, it is possible that the mechanism by which cells polarise depends on the order in which they generate the basal, lateral, and apical domains. Since the midgut is the only endodermal epithelium in Drosophila, the only epithelium with apical SJs, and the only epithelium that polarises from basal to apical, it is not possible to determine which of these characteristics underlies its alternative mechanism of cell polarisation, and this will require analysis in other organisms with a greater diversity of epithelial cell types.
Whatever the reason for the alternative polarity mechanism in the Drosophila midgut epithelium, its polarity is much more similar to that of well-characterised vertebrate epithelia than other Drosophila epithelia. Firstly, the midgut and vertebrate epithelia have apical occluding junctions above lateral AJs, whereas other Drosophila epithelia have a reversed arrangement of junctions. Secondly, the midgut epithelium does not require the canonical epithelial polarity factors that are essential in other Drosophila epithelia, and this also seems to be the case in some vertebrate epithelia, although this may be due to redundancy between paralogues. Thirdly, the midgut epithelium and a number of vertebrate epithelia depend on signals from the integrin adhesion complex to polarise correctly [42–44]. These similarities suggest that the Drosophila midgut epithelium may prove a better in vivo model for at least some types of vertebrate epithelia. It will therefore be important to determine whether vertebrates also contain distinct types of epithelia whose polarity is controlled by different factors.
w1118 or y2 or OreR flies were used as wild type unless otherwise specified. Other stocks used in this study were as follows:
The following stocks were used to generate (positively labelled) MARCM [54] clones:
Negatively marked clones on the X chromosome were generated using the following stock:
Standard procedures were used for Drosophila husbandry and experiments. Flies were reared on standard fly food supplemented with live yeast at 25 °C. For the conditional expression of UAS responder constructs (e.g., RNAi) in adult flies, parental flies were crossed at 18 °C and the resulting offspring reared at the same temperature until eclosion. Adult offspring were collected for 3 days and then transferred to 29 °C to inactivate the temperature-sensitive GAL80 protein. To generate MARCM or GFP-negative clones, flies were crossed at 25 °C and the resulting offspring subjected to heat shocks either as larvae (from L2 until eclosion) or as adults (5–9 days after eclosion). Heat shocks were performed at 37 °C for 1 h (twice daily). Flies were transferred to fresh food vials every 2–3 days and kept at 25 °C for at least 9 days after the last heat shock to ensure that all wild-type gene products from the heterozygous progenitor cells had turned over. For this study, all (midgut) samples were obtained from adult female flies.
Samples were dissected in PBS and fixed with 8% formaldehyde (in PBS containing 0.1% Triton X-100) for 10 min at room temperature. Following several washes with PBS supplemented with 0.1% Triton X-100 (washing buffer), samples were incubated in PBS containing 3% normal goat serum (NGS, Stratech Scientific Ltd, Cat. #005-000-121; concentration of stock solution: 10 mg/ml) and 0.1% Triton X-100 (blocking buffer) for 30 min. This fixation method was only used for samples in which F-actin was stained with fluorescently labelled phalloidin, as phalloidin staining is incompatible with heat fixation.
The heat fixation protocol is based on a heat–methanol fixation method used for Drosophila embryos [82]. Samples were dissected in PBS, transferred to a wire mesh basket, and fixed in hot 1X TSS buffer (0.03% Triton X-100, 4 g/L NaCl; 95 °C) for 3 s before being transferred to ice-cold 1X TSS buffer and chilled for at least 1 min. Subsequently, samples were transferred to washing buffer and processed for immunofluorescence stainings.
After blocking, samples were incubated with the appropriate primary antibody/antibodies diluted in blocking buffer at 4 °C overnight. Following several washes, samples were incubated with the appropriate secondary antibody/antibodies either at room temperature for 2 h or at 4 °C overnight. Samples were then washed several times in washing buffer and mounted in Vectashield containing DAPI (Vector Laboratories) on borosilicate glass slides (No. 1.5, VWR International). All antibodies used in this study were tested for specificity using clonal analysis (MARCM) or RNAi.
Primary antibodies:
Mouse monoclonal antibodies: anti-Dlg (4F3), anti-Cora (c615.16), anti-αSpec (3A9), anti-Arm (N2 7A1), anti-Talin (A22A, E16B), anti-Pros (MR1A), anti-Crb (Cq4), anti-Nrv (Nrv5F7), anti-Mys (CF.6G11). All monoclonal antibodies were obtained from the Developmental Studies Hybridoma Bank and used at 1:100 dilution.
Rabbit polyclonal antibodies: anti-pEzrin (NEB Cat. #3726S, 1:200 dilution); anti-Lgl (Santa Cruz Biotechnoloy Inc., d-300, Cat. #SC98260, 1:200 dilution); anti-aPKC (Santa Cruz Biotechnoloy Inc., Cat #SC216, 1:100 dilution); anti-βHSpec [83] (gift from C. Thomas, The Pennsylvania State University, USA, 1:1,000 dilution); anti-Baz [6] (gift from A. Wodarz, University of Cologne, Germany, 1:200 dilution); anti-Par6 [84] (gift from D. J. Montell, UCSB, USA, 1:500 dilution); anti-Mesh [51] and anti-Tsp2A [52] (gift from M. Furuse, 1:1,000 dilution); anti-Scrib [85] (gift from C. Q. Doe, University of Oregon, USA, 1:1,000 dilution); anti-Pdm1 [86] (gift from F. J. Diaz-Benjumea, Centre for Molecular Biology "Severo Ochoa" (CBMSO), Spain, 1:1,000 dilution); anti-Cno [87] (gift from M. Peifer, UNC, USA, 1:1,000 dilution).
Other antibodies used: chicken anti-GFP (Abcam, Cat. #ab13970, 1:1,000 dilution); guinea pig anti-Yurt [88] (gift from U. Tepass, University of Toronto, Canada, 1:1,000 dilution); guinea pig anti-Myo7a [89] (gift from D. Godt, University of Toronto, Canada, 1:1,000 dilution); guinea pig anti-Shot [90] (1:1,000 dilution); rat anti-Mesh [51] (gift from M. Furuse, 1:1,000 dilution).
Secondary antibodies:
Alexa Fluor secondary antibodies (Invitrogen) were used at a dilution of 1:1,000.
Alexa Fluor 488 goat anti-mouse (#A11029), Alexa Fluor 488 goat anti-rabbit (#A11034), Alexa Fluor 488 goat anti-guinea pig (#A11073), Alexa Fluor 488 goat anti-chicken IgY (#A11039), Alexa Fluor 555 goat anti-rat (#A21434), Alexa Fluor 555 goat anti-mouse (#A21422), Alexa Fluor 555 goat anti-rabbit (#A21428), Alexa Fluor 568 goat anti-guinea pig (#A11075), Alexa Fluor 647 goat anti-mouse (#A21236), Alexa Fluor 647 goat anti-rabbit (#A21245), Alexa Fluor 647 goat anti-rat (#A21247). Only cross-adsorbed secondary antibodies were used in this study to eliminate the risk of cross-reactivity.
F-Actin was stained with phalloidin conjugated to Rhodamine (Invitrogen, Cat. #R415, 1:500 dilution).
Images were collected on an Olympus IX81 (40× 1.35 NA Oil UPlanSApo, 60× 1.35 NA Oil UPlanSApo) using the Olympus FluoView software Version 3.1 and processed with Fiji (ImageJ).
Endogenously tagged aPKC with EGFP fused to the N-terminus was generated by CRISPR-mediated homologous recombination. In vitro synthesised gRNA [91] to a CRISPR target approximately 60 nucleotides downstream from the aPKC start codon (target sequence GAATAGCGCCAGTATGAACATGG) and a plasmid donor containing the ORF of EGFP as well as appropriate homology arms (1.5 kb upstream and downstream) were coinjected into nos-Cas9–expressing embryos (Bloomington #54591; also known as CFD2) [92]. Single F0 flies were mated to y w flies and allowed to produce larvae before the parent was retrieved for PCR analysis. Progeny from F0 flies in which a recombination event occurred (as indicated by PCR) were further crossed and analysed to confirm integration. Several independent EGFP-aPKC lines were isolated. Recombinants carry the EGFP coding sequence inserted immediately downstream of the endogenous start codon and a short linker (amino acid sequence: Gly-Ser-Gly-Ser) between the coding sequence for EGFP and the coding sequence for aPKC. Homozygous flies are viable and healthy.
We used the CRISPR/Cas9 method [91] to generate a null allele of aPKC. sgRNA was in vitro transcribed from a DNA template created by PCR from two partially complementary primers: forward primer: 5′-GAAATTAATACGACTCACTATAggattacggcatgtgtaaggGTTTTAGAGCTAGAAATAGC-3′; reverse primer: 5′- AAAAGCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3′. The sgRNA was injected into Act5c-Cas9 embryos [92]. Putative aPKC mutants in the progeny of the injected embryos were recovered, balanced, and sequenced. The aPKCHC allele contains a small deletion around the CRISPR site, resulting in one missense mutation and a frameshift that leads to stop codon at amino acid 409 in the middle of the kinase domain, which is shared by all isoforms (S1L Fig). The aPKCHC allele was subsequently recombined onto FRTG13 to generate MARCM clones. No aPKC protein was detectable by antibody staining in both midgut and follicle cell clones, and follicle cells homozygous for aPKCHC display a phenotype that is indistinguishable from that observed in follicle cells homozygous for the aPKCK06403 allele.
The proportions of rhea, Fit, and Tsp2a mutant cells inside the epithelial layer were calculated as follows: images were taken of different regions of several midguts containing MARCM clones stained with an apical marker. Cells that were above the neighbouring cells or had a clear apical domain were counted as ‘cells NOT inside the layer’, whereas cells without a detectable free apical surface were counted as ‘cells inside the layer’. Data were analysed with Graphpad Prism software. The graph in S4A Fig shows the average percent of cells inside the layer ± SD%.
The nuclear long axis of rhea, Fit, and Tsp2a mutant clones was manually measured in 20 random mutant cells (stained with DAPI, excluding obvious ISCs) and 20 surrounding heterozygous cells using Fiji.
All experiments were repeated multiple times with independent crosses: Baz-EGFP (4 independent experiments), EGFP-aPKC (9 independent experiments), Mew-YFP (5 independent experiments), Par6-EGFP (4 independent experiments), Crb-EGFP (5 independent experiments), Fit1-EGFP (4 independent experiments), Myo31DF-YFP (4 independent experiments), MyoIA[ts] > UAS-Scrib-RNAi (11 independent experiments), MyoIA[ts] > UAS-Crb (6 independent experiments), and MyoIA[ts] > UAS-Baz-GFP (3 independent experiments).
The phenotypes of homozygous mutant clones were analysed in multiple guts from independent experiments as follows: baz4 (13 independent experiments, 3,078 mutant cells analysed); baz815-8 (4 independent experiments, 734 cells analysed); aPKCk06043 (8 independent experiments, 3,681 mutant cells analysed); aPKCHC (9 independent experiments, 15,984 mutant cells analysed); par6Δ226 (4 independent experiments, 2,558 mutant cells analysed); crb11A22 (4 independent experiments, 1,478 mutant cells analysed); crb8F105 (5 independent experiments, 3,288 mutant cells analysed); lgl4 (6 independent experiments, 6,790 mutant cells analysed); dlg114 (5 independent experiments, 3,092 mutant cells analysed); rhea79a, rheaB28, rheaB128 (4 independent experiments for each genotype, 184 mutant cells analysed in total); Fit118Fit283 (7 independent experiments, 608 double mutant cells analysed); Fit118 (4 independent experiments, 854 mutant cells analysed); ilk54 (5 independent experiments, 65 mutant cells analysed); Tsp2a1-2, Tsp2a2-9, and Tsp2a3-3 (7, 4, and 6 independent experiments, respectively; a total of 1,205 Tsp2a mutant cells were analysed); and meshf04955 (5 independent experiments, 643 mutant cells analysed).
|
10.1371/journal.pgen.1004841 | PRL1, an RNA-Binding Protein, Positively Regulates the Accumulation of miRNAs and siRNAs in Arabidopsis | The evolutionary conserved WD-40 protein PRL1 plays important roles in immunity and development. Here we show that PRL1 is required for the accumulation of microRNAs (miRNAs) and small interfering RNAs (siRNAs). PRL1 positively influences the processing of miRNA primary transcripts (pri-miRNAs) and double-stranded RNAs (dsRNAs). Furthermore, PRL1 interacts with the pri-miRNA processor, DCL1, and the dsRNA processors (DCL3 and DCL4). These results suggest that PRL1 may function as a general factor to promote the production of miRNAs and siRNAs. We also show that PRL1 is an RNA-binding protein and associates with pri-miRNAs in vivo. In addition, prl1 reduces pri-miRNA levels without affecting pri-miRNA transcription. These results suggest that PRL1 may stabilize pri-miRNAs and function as a co-factor to enhance DCL1 activity. We further reveal the genetic interaction of PRL1 with CDC5, which interacts with PRL1 and regulates transcription and processing of pri-miRNAs. Both miRNA and pri-miRNA levels are lower in cdc5 prl1 than those in either cdc5 or prl1. However, the processing efficiency of pri-miRNAs in cdc5 prl1 is similar to that in cdc5 and slightly lower than that in prl1. Based on these results, we propose that CDC5 and PRL1 cooperatively regulate pri-miRNA levels, which results in their synergistic effects on miRNA accumulation, while they function together as a complex to enhance DCL1 activity.
| PRL1, a conserved WD-40 protein, is required for plant development and immune responses. However, its functional mechanisms are not well understood. Here, we show the positive impact of PRL1 on the accumulation of miRNAs and siRNAs, which are key regulators of plant growth and immunity. PRL1 interacts with multiple DCLs (the processors of miRNAs and siRNAs) and is required for their optimal activities, suggesting that PRL1 acts as a general factor to facilitate the production of miRNAs and siRNAs. In addition, PRL1 is an RNA-binding protein, binds pri-miRNAs in vivo and positively influences the levels of pri-miRNAs levels without affecting the promoter activities of genes encoding pri-miRNAs. These results suggest that PRL1 may also stabilize pri-miRNAs. We further show that RPL1 and its interactor CDC5 (a DNA-binding protein) synergistically regulate pri-miRNA levels, resulting in enhanced effects on miRNA accumulation, although they function together as a complex to facilitate DCL1 activity.
| In plants and animals, microRNAs (miRNAs), ∼20–25 nucleotides (nt) in size, regulate gene expression in various biological processes such as development and metabolism [1]–[3]. They are produced as duplexes through precise excision from imperfect fold-back primary transcripts (pri-miRNAs) [1]–[3]. In the miRNA duplex, the miRNA strand is loaded into ARGONAUTE (AGO) proteins to repress the expression of target genes containing its complementary sequences while the other strand (passenger strand; miRNA*) is degraded [1]–[3]. Plants and animals also use small interfering RNAs (siRNAs) to repress gene expression. siRNAs are chemically identical to miRNAs [2]. However they are produced from long double stranded RNAs. The two major classes of plant siRNAs are siRNAs derived from repeated DNAs (ra-siRNAs) and trans-acting siRNAs (ta-siRNAs) [4], [5].
In plants, most pri-miRNAs are transcribed by DNA-dependent RNA polymerase II (Pol II) from endogenous miRNA encoding genes (MIR) [1], [2]. The mediator complex and Negative on TATA less2 (NOT2; a transcription factor) regulate the transcription of MIR [6], [7]. After generation, pri-miRNAs are proposed to be stabilized by DAWDLE (DDL), an RNA binding protein [8]. Pri-miRNAs are then processed to stem-loop precursors (pre-miRNAs) and finally to the miRNA/miRNA* duplex by Dicer-LIKE 1 (DCL1; an RNAse III enzyme) in the nucleus in plants [9], [10]. The C2H2 zinc-finger protein SERRATE (SE) and the RNA binding proteins HYPONASTIC LEAVES 1 (HYL1) and TOUGH (TGH) form a complex with DCL1 to ensure efficient and accurate process of pri-miRNAs [9], [11]–[17]. To ensure its proper function, HYL1 needs to be dephosphorylated during pri-miRNA processing [18]. Several other proteins including DDL, Cap-Binding Protein 20 (CBP20), CBP80, RACK1 and NOT2 are associated with the DCL1 complex to facilitate miRNA maturation [7], [8], [19]–[21]. NOT2 and MODIFIER OF SNC1, 2 (MOS2; an RNA binding protein) have been shown to guide the correct localization of the DCL1 complex [7], [22]. SICKLE (SIC; a proline rich protein) is shown to regulate the accumulation of some miRNAs [23]. Besides protein factors, the structure of pri-miRNAs plays essential roles in regulating DCL1 activity [24]–[27]. For instance, an imperfectly paired lower stem of ∼15 bp below the miRNA:miRNA* duplex is crucial for the initial pri-miRNA cleavage [25]–[27].
We previously showed that Cell Division Cycle 5 (CDC5), a DNA-binding protein, positively regulates miRNA biogenesis in Arabidopsis [28]. CDC5 interacts with Pol II and MIR promoters [28]. Lack of CDC5 reduces the occupancy of Pol II at MIR promoters and pri-miRNA levels, suggesting that CDC5 is a positive transcription factor of MIR [28]. Besides acting as a transcription factor, CDC5 functions as a co-factor of the DCL1 complex to participate pri-miRNA processing [28]. CDC5 is a component of the conserved MOS4-associated complex (MAC). MAC was first identified as a suppressor of snc1, which carries a gain-of-function mutation in the SNC gene and show constitutive resistance to a wide spectrum of pathogens [29]. Loss-of-function mutations in the MAC complex reduce plant immunity to bacterial infections and cause multiple developmental defects such as reduced fertility and delayed growth [29]. The counterparts of MAC in yeast and Human associate with spliceosome and function in splicing [29]. Other components of MAC include MOS4 (a coil-coil domain containing protein), PRL1 (a WD-40 protein), MAC3A and MAC3B (two functionally redundant U-box E3 ubiquitin ligases). Among these proteins, PRL1 and MOS4 have been shown to interact with CDC5 directly [29].
In this study, we show that PRL1 but not MOS4 plays important roles in the accumulation of miRNAs and siRNAs. Lack of PRL1 in prl1 reduces miRNA accumulation and pri-miRNA processing efficiency. In addition, PRL1 interacts with the DCL1 complex, suggesting it may function as co-factor of DCL1 to promote miRNA maturation. Pri-miRNA levels are reduced in prl1 relative to wild-type plants. However, MIR promoter activity is not affected by prl1, despite of the association of PRL1 with Pol II. Based on the facts that PRL1 is an RNA-binding protein and binds pri-miRNAs in vivo, we propose that PRL1 may stabilize pri-miRNAs. Furthermore, the levels of both miRNAs and pri-miRNAs are further reduced in cdc5 prl1 relative to either cdc5 or prl1. However, CDC5 and PRL1 do not show additive effects on the processing of pri-miRNAs. These results suggest that CDC5 and PRL1 may synergistically influence pri-miRNAs levels and act together as a complex to promote miRNA maturation. PRL1 also interacts with DCL3 and DCL4, which produces siRNAs, and is required for their optimal activities, suggesting that PRL1 may be a general accessory factor for the production of small RNAs.
Given the role of CDC5 in miRNA biogenesis, it is possible that other components of the MAC complex may also be required for miRNA accumulation. Therefore, we examined the effect of the mutants mac3b (SALK_050811), mos4 (SALK_0090851C) and prl1-2 on miRNA abundance using Northern blot. We also included snc1 (SALK_047058C) in the analysis since SNC1 is related to the MAC complex and snc1 causes development defects. These mutants are likely null since the transcripts of corresponding genes could not be detected by RT-PCR (Figure S1A). Like in cdc5-1, the abundance of all four tested miRNAs (miR167, miR171, miR172 and miR173) was decreased in prl1-2 compared to Col (wild-type control). In contrast, miRNA levels in mos4, mac3b and snc1 were comparable with those in Col (Fig. 1A). We examined the accumulation of additional miRNAs in prl1-2 and found that all these miRNAs were reduced in abundance in prl1-2 relative to Col (Fig. 1B). In addition, expression a wild-type copy of PRL1 fused with a YFP tag under the control of its native promoter (pPRL1::PRL1-YFP) fully recovered miRNA levels in prl1-2 (Fig. 1B). These results demonstrated that PRL1 but not MOS4 and MAC3b is required for miRNA accumulation. We next analyzed the transcript levels of several miRNA targets (ARF3, CUC1, MYB33, MYB65 and PHV) in prl1-2 and Col by quantitative RT-PCR (qRT-PCR) in order to test the effect of prl1-2 on miRNA function. The transcription levels of these targets were slightly increased in prl1-2 relative to Col (Figure. S1B). The PRL1 transgene fully recovered miRNA function in prl1 (Figure S1B). We also asked if PRL1 has a role in siRNA biogenesis. The levels of nine examined siRNAs (three ta-siRNAs and six ra-siRNAs) were reduced compared to those in Col (Fig. 1B and 1C), which was complemented by the expression of pPRL1::PRL1-YFP. These results revealed that like CDC5, PRL1 positively regulates the levels of miRNAs and siRNAs in Arabidopsis.
The PRL1-CDC5 interaction suggests that similar to CDC5, PRL1 may act as a co-factor of Pol II and DCL1 to regulate miRNA accumulation. To test these two possibilities, we first examined the PRL1-Pol II association using co-immunoprecipitation (co-IP) assay. In this experiment, anti-YFP and anti-RPB2 that detects the second largest subunit of Pol II (RPB2) [6] were used to capture the PRL1-YFP and Pol II complex, respectively, from the protein extracts of prl1-2 complementation line expressing the pPRL1::PRL1-YFP transgene. After IP, PRL1-YFP was detected in the Pol II precipitates whereas RPB2 existed in the PRL1-YFP complex (Fig. 2A and 2B). In contrast, no interaction was detected in the control reactions (Fig. 2A and 2B), demonstrating the PRL1-Pol II association.
We next tested the association of PRL1 with the components of DCL1 complex using a bimolecular fluorescence complementation (BiFC). In the BiFC assay, transient co-expression of PRL1 fused with C-terminal fragment of cyan fluorescent protein (cCFP) with DCL1, SE, HYL1 or CDC5 fused with the N-terminal fragment of Venus (nVenus) produced yellow fluorescence signals (Fig. 2C), suggesting that PRL1 might associate with the DCL1 complex. To verify this result, we tested co-IP of PRL1 with DCL1 and SE. After PRL1-YFP or YFP was transiently co-expressed with DCL1-MYC and SE-MYC fusion proteins in N. benthamiana, respectively, IPs were performed with anti-YFP or anti-MYC antibodies. Western blots detected PRL1-YFP in the DCL1-MYC and SE-MYC complexes and DCL1-MYC/SE-MYC in the PRL1-YFP precipitates, suggesting that PRL1-YFP and DCL1/SE reciprocally pulled down each other (Fig. 2D, 2E, 2F and 2G). As a control, YFP did not show interaction with either DCL1 or SE. These results demonstrated the association of PRL1 with DCL1 and SE.
The interaction of PRL1 with Pol II suggests that PRL1 may positively regulate MIR transcription. If so, lack of PRL1 will impair MIR transcription, resulting in reduced levels of pri-miRNAs. To test this, we compared the pri-miRNA levels in prl1-2 with those in Col using qRT-PCR. In fact, the levels of all seven examined pri-miRNAs in prl1-2 were less than those in Col (Fig. 3A), which were recovered in the complementation line of prl1-2 (Fig. 3A). To test whether the reduction of pri-miRNA levels is due to impaired MIR promoter activity, we introduced the prl1-2 mutation into a Col transgenic line containing a single cope of GUS transgene driven by MIR167a promoter (pMIR167a::GUS), which was previously used to test the function of the mediator complex in regulating MIR transcription [6]. However, GUS staining and qRT-PCR analysis showed a similar GUS expression level in PRL1+ (PRL1/PRL1 or PRL1/prl1 genotype) and prl1-2 containing the pMIR167a::GUS transgene (Fig. 3B and 3C). This result demonstrated that PRL1 does not affect MIR promoter activity. Consistent with this notion, prl1 did not show obvious effect on MIR172b promoter activity (Figure S2A).
We next evaluated the effect of prl1-2 on the half-lives of pri-miRNAs using cordycepin as a transcriptional inhibitor [30]. Two-week old plants were transferred to medium containing cordycepin. After incubation was stopped at various time points, we measured the levels of pri-miR164a and pri-miR167a using qRT-PCR. The results showed that the degradation rate of pri-miR164a and pri-miR167a in prl1-2 is similar to that in Col (Figure S2B).
We next asked whether PRL1 has a role in processing of miRNA precursors through an in vitro processing assay [13], [31] since it is associated with DCL1 and SE. In this experiment, a portion of pri-miR162b that contains the stem-loop of miR162b with 6-nt arms at each end (MIR162b; Fig. 4A) and pre-miR162b (Fig. 4B) were first produced through in vitro transcription in the presence of [α-32P] UTP. [32P]-labeled MIR162b and pre-miR162b were then processed in the protein extracts of young flower buds of prl1-2 or Col. The production of miR162b from both MIR162b and pre-miR162b in prl1-2 at various time points was less than that in Col (Fig. 4C and 4D). The processing of MIR162b and pre-miRNA162b was recovered in the PRL1 complementation line (Figure S3) The levels of miR162 produced from MIR162b and pre-miRNA162b in prl1 at 80 min were ∼40% of those produced in Col (Fig. 4E and 4F). These results suggested that PRL1 might have a role in promoting miRNA maturation.
We next asked the role of PRL1 in siRNA biogenesis, as prl1-2 reduces the accumulation of siRNAs. By analog, we examined the interaction of PRL1 with DCL3 and DCL4 and the effect of prl1-2 on dsRNA processing. To test the PRL1-DCL3/DCL4 interaction, we expressed a recombined PRL1 fused with a maltose-binding protein at its N-terminus (MBP-PRL1) and MBP in E.coli. The protein extracts containing MBP-PRL1 or MBP were mixed with protein extracts containing DCL3-YFP or DCL4-YFP, which were transiently expressed in N. benthamiana. Then the DCL3-YFP or DCL4-YFP complex was IPed with anti-YFP antibodies. MBP-PRL1, but not MBP, was co-IPed with DCL3-YFP and DCL4-YFP (Fig. 5A). In addition, YFP did not interact with MBP or MBP-PRL1. These results demonstrated that PRL1 interacts with DCL3 and DCL4 (Fig. 5A).
To test the effect of prl1 on dsRNA processing, we generated ∼460 bp dsRNAs through in vitro transcription of a DNA fragment (5′ portion of UBIQUITIN 5) containing the T7 promoter at the end of each strand under the presence [α-32P] UTP. The radioactive labeled dsRNA then incubated with prl1 or Col protein extracts. The production of both 21 nt and 24 nt small RNAs was impaired in prl1 compared with that in Col and that in the complementation line (Fig. 5B). This result indicated that multiple DCL activities are impaired by prl1-2, because DCL3 is responsible for the production of 24 nt small RNAs and DCL1/DCL4 is involved in the production of 21 nt small RNAs from dsRNAs.
CDC5 and PRL1 have been shown to directly interact with each other. Both CDC5 and PRL1 interact with DCL1 and positively regulate miRNA processing. These results raise a possibility that CDC5 and PRL1 may act as a complex to regulate DCL1 activity. In addition, CDC5 regulates MIR promoter activity while PRL1 does not. These suggest that PRL1 and CDC5 might act additionally in miRNA pathway. To test these two possibilities, we constructed a cdc5-1 prl1-2 double mutant by crossing prl1-2 into cdc5-1 and compared miRNA levels in cdc5-1 prl1-2 with those in cdc5-1 and prl1-2, respectively. The cdc5-1 prl1-2 double mutant displayed more severe developmental defects than either cdc5-1 or prl1-2, suggesting that PRL1 and CDC5 function additionally in regulating development (Fig. 6A). Northern blot analyses showed that the levels of several examined miRNAs in cdc5-1 prl1-2 were lower than those in either prl1-2 or cdc5-1 (Fig. 6B), indicating that PRL1 and CDC5 function synergistically in miRNA pathway.
There are at least two possible explanations for the further reduction of miRNA levels in cdc5-1 pr1l-2 relative to either cdc5-1 or prl1-2 based on the fact that both PRL1 and CDC5 positively regulate pri-miRNA levels and miRNA maturation. One is that pri-miRNA levels might be further reduced in cdc5-1 prl1-2. The other is that the processing efficiency of miRNA precursors might be lower than either cdc5-1 or prl1-2. To test these two possibilities, we first determined the pri-miRNA levels in cdc5-1 prl1-2, cdc5-1 and prl1-2 through qRT-PCR. The levels of several pri-miRNAs were decreased in cdc5-1 prl1-2 when compared with those in either cdc5-1 or prl1-2 (Fig. 6C), demonstrating that CDC5 and PRL1 indeed act synergistically in regulating pri-miRNA levels. Next, we evaluated the in vitro processing of pre-miR162b in cdc5-1 prl1-2. The amount of miR162b produced in cdc5-1 prl1-2 was similar to that in cdc5-1 and slightly lower than that in prl1-2 (Fig. 6D), suggesting that PRL1 and CDC5 may not act additionally in promoting miRNA maturation.
Given the role of PRL1 in RNA metabolism, it is reasonable to speculate that PRL1 might have an RNA-binding activity. We therefore performed an in vitro RNA pull-down assay to test this possibility. In this assay, recombinant PRL1 fused with a maltose-binding protein at its N-terminus (MBP-PRL1) and MBP were expressed in E.coli and purified with amylose resin (Fig. 7A). MBP-PRL1 and MBP were then incubated with [32P]-labeled MIR162b or pre-miR162b, respectively. MBP-PRL1 but not MBP bound MIR162b and pre-miR162b after incubation. In addition, when excess amount of unlabeled MIR162b or pre-miR162b was added in the reaction, radioactive labeled MIR162b or pre-miR162b could not be retained in the MBP-PRL1 complex. These results suggested that PRL1 binds RNAs in vitro. We next tested the binding ability of MBP-PRL1 to dsRNA, ssRNA and DNA using the in vitro RNA pull-down assay described above. MBP-PRL1 was able to bind a ∼100-nt RNA corresponding to a portion of the 5′ end of the UBIQUITIN 5 (UBQ5) [13], but not an in vitro synthesized ∼50 bp DNA fragment [32] and a dsRNA generated through in vitro transcription vitro transcription of a DNA fragment (5′ portion of UBQ5, ∼460 bp) containing the T7 promoter at end of each strand [13] (Fig. 7B).
Next, we performed an RNA immunoprecipitation (RIP) assay to test whether PRL1 binds pri-miRNAs in vivo [13]. Seedlings of the prl1-2 complementation line expressing pPRL1::PRL1-YFP transgene and the control plants harboring YFP were used for RIP. After PRL1-YFP or YFP complex were precipitated with anti-YFP antibody, pri-miRNAs were detected with RT-PCR. Several tested pri-miRNAs (pri-miR159a, pri-miR167a, pri-miR171 and pri-miR172a) existed in the PRL1-YFP complex but not in the YFP complex and no Anti-body (NoAb) controls (Fig. 7C). These results suggested that PRL1 associates with pri-miRNAs in vivo.
In this study, we identify PRL1, a WD-40 protein, as an important regulator of miRNA accumulation. Several evidences including reduced accumulation of pri-miRNAs and miRNAs in prl1, PRL1-DCL1 interaction and PRL1-pri-miRNA association demonstrate that PRL1 positively impacts miRNA biogenesis. It has been suggested that PRL1 influences plant immunity and development through its impacts on RNA processing [29], [33]. Given the essential roles of miRNAs in plant immunity and development, it is possible that reduced miRNA levels in prl1 may partially contribute to the observed phenotypes.
PRL1 likely has a role in promoting miRNA maturation, as lack of PRL1 reduces processing of MIR162b and pre-miR162b. PRL1 interacts with the DCL1 complex and does not positively regulate the transcription of genes involved in miRNA biogenesis (Figure S4), suggesting that PRL1 may act as a co-factor to regulate DCL1 activity. CDC5, a direct interactor of PRL1 also regulates the DCL1 activity through its interaction with the helicase and dsRNA binding domains of DCL1. The effect of PRL1 on pri-miRNA processing appears to be weaker than that of CDC5. The processing efficiency of MIR162b and pre-miR162b in cdc5-1 prl1-2 is similar to that in cdc5-1 and slightly lower than that in prl1-2. This result suggests that PRL1 and CDC5 may act together as a complex to regulate DCL1 activity. Furthermore, gel filtration analysis suggests that PRL1 may not affect DCL1-CDC5 association (Figure S5). Thus, it is possible that PRL1 may act as accessory factor to facilitate CDC5 function.
PRL1 also positively regulates the pri-miRNA levels since prl1 reduces the accumulation of pri-miRNAs. We previously showed that CDC5 interacts with Pol II and positively regulate MIR transcription [28]. Since PRL1 associates with Pol II as well, it is possible that PRL1 acts as a component of the CDC5 complex to regulate MIR promoter activity. However this seems not to be the case, as loss-of-function of PRL1 does not affect the GUS levels driven by the MIR167a promoter. Consistent with this notion, the levels of pri-miRNAs are further reduced in cdc5-1 prl1-2 compared with cdc5-1 or prl1-2. Given the fact that PRL1 binds pri-miRNAs in vitro and vivo, we propose that PRL1 may stabilize pri-miRNAs. Indeed, the fact that the half-life of pri-miR164a and pri-miR167a in prl1 is similar to that in Col suggests that the degradation of pri-miRNAs may be increased in prl1, because less efficient processing may lead to increased abundance of pri-miRNAs in prl1. However, we cannot rule out the possibility that PRL1 acts in MIR transcription after initiation, as it associates with Pol II.
In summary, we reveal that PRL1 positively regulates miRNA levels through its impacts on pri-miRNA levels and processing. PRL1 functions additively with its interactor CDC5 as miRNA abundance is lower in cdc5-1 prl1-2 than in cdc5-1 or prl1-2. The synergistic effect of CDC5 and PRL1 on miRNA levels can be explained by their different roles in controlling pri-miRNA levels rather than their function in promoting miRNA maturation. Besides CDC5 and PRL1, the core components MAC complex includes MOS4, MAC3A and MAC3B [29]. We show that MOS4 and MAC3b have no impact on miRNA levels. However, whether MAC3B has a role in miRNA accumulation needs to be further explored since it acts redundantly with MAC3A [29]. The MAC complex appears to have a role in siRNA biogenesis. Both CDC5 and PRL1 promote the accumulation of siRNA [28] while MOS4 is required for the accumulation of ra-siRNAs [34]. How does MAC participate in siRNA biogenesis? We have showed both PRL1 and CDC5 interact with the DCL1 complex and regulate its activity. By analogy, it is possible that the MAC complex associates with the DCL3 complex to regulate its activity. In fact, DCL3 interacts with PRL1. prl1 also reduces the abundance of ta-siRNAs, whose production requires DCL4 and DCL1-dependent miRNAs. Since PRL1 interacts with DCL4 and is required for the accumulation of DCL1-dependent miRNAs, it may promote ta-siRNA production through facilitating DCL4 function and miRNA production. The MAC complex is an evolutionarily conserved complex [29]. As many aspects of small RNA pathway are conserved, it is tempting to propose that the counterparts of MAC play some roles in small RNA pathways in other organisms.
The mac3b (SALK_050811), mos4 (SALK_0090851C), prl1-2 (Salk_008466), snc1 (SALK_047058C) and cdc5-1 (SAIL_207_F03) mutants were ordered from Arabidopsis Biological Resources Center (ABRC). All of them are in the Columbia-0 genetic background. Transgenic line containing a single copy of pMIR167a::GUS was crossed to prl1-2. In the F2 population, PRL1/PRL1, PRL1/prl1-2 and prl1-2/prl1-2 harboring pMIR167a::GUS were identified through PCR genotyping for prl1-2 and GUS.
PRL1 genomic DNA was amplified from Col genome and cloned to pMDC204 binary vector to generate pPRL1::PRL1-YFP construct. The construct was transformed to prl1-2. The full-length PRL1 cDNA was amplified by RT-PCR and ligated to pMAL-c5x (NEB) to generate MBP-PRL1. To generate the cCFP-PRL1 fusion vector, the PRL1 cDNA was first cloned into the pSAT4-cCFP-C vector [35]. The DNA fragment containing cCFP-PRL1 was released by I-SecI restriction enzyme and subsequently cloned into the pPZP-RCS2-ocs-bar vector. All the primers are listed in Table S1.
In the PRL1-PoII co-IP experiment, proteins were extracted from the transgenic plants harboring the pPRL1::PRL1-YFP transgene and incubated with anti-YFP (Clontech) or anti-RPB2 antibodies coupled to protein G agarose beads (Clontech) for 4 h at 4°C. After the beads were washed five times with protein extraction buffer, proteins were resolved by SDS/PAGE. Anti-YFP and anti-RPB2 antibodies were then used to detect PRL1-YFP and RPB2, respectively. To test the interaction of PRL1 with components of the DCL1 complex, PRL1-YFP (YFP) was co-expressed with DCL1-MYC or SE-MYC in N. benthamiana. Protein extracts were then incubated with anti-YFP or anti-MYC antibodies coupled to protein G agarose beads. Anti-YFP and anti-MYC (MBL) antibodies were used to detect PRL1-YFP/YFP and DCL1-MYC/SE-MYC, respectively.
In vitro dicer activity assay was performed according to Qi et al and Ren et al [13], [31]. MIR162b and pre-miR162b RNAs were produced by in vitro transcription under the presence of [α-32P] UTP. In the dicer activity assay, protein extractions were incubated with [32P] labeled MIR162b or pre-miR162b in reaction buffer containing 100 mM NaCl, 1 mM ATP, 0.2 mM GTP, 1.2 mM MgCl2, 25 mM creatine phosphate, 30 µg/ml creatine kinase and 4 U RNase inhibitor at 25°C. After the reactions were stopped at 40, 80 or 120 mins, respectively, RNAs were extracted and resolved on PAGE gel. Radioactive signals were detected with a PhosphorImager and quantified by ImageQuant version 5.2.
Paired cCFP and nVenus constructs were co-infiltrated in N. benthamiana leaves for 40 h. YFP signals were then detected with a confocal microscopy (Fluoview 500 workstation; Olympus) at 488 nm with a narrow barrier (505–525 nm, BA505–525; Olympus).
Northern blot was used to detect small RNA abundance as described [13]. qRT- PCR was performed to detect the levels of pri-miRNAs, transcripts of miRNA targets and GUS using cDNA templates reverse transcribed by the SuperScript III (Invitrogen) and oligo dT18 primer. qRT-PCR was run on an iCycler apparatus (Bio-Rad). RNA pull-down were performed according to Ren et al [13]. MBP and MBP-PRL1 were expressed in E.coli. MIR162b, pre-miR162b, dsRNAs and ssRNA were produced by in vitro transcription with T7 RNA polymerase at the presence [α-32P] UTP whereas DNA was synthesized at IDT and labeled with T4 PNK at the presence [γ-32P] ATP. [32P]-labeled probes are incubated with amylose resin beads combined MBP or MBP-PRL1 at 4°C for 1 hour. After 4 times wash with washing buffer, DNA or RNA are extracted and resolved on PAGE gel. Radioactive signals were detected with a PhosphorImager and quantified by ImageQuant version 5.2. RIP was performed according to [13], [36]. Seedlings of transgenic plants harboring the pPRL1::PRL1-YFP transgene or YFP were used to examine the RNA binding activity of PRL1 in vivo. All the primers are listed in Table S1.
RNA half-life assay was performed according to Lidder et al [30]. Two-week-old Col and prl1-2 seedlings were transferred to flask with incubation buffer (1/2 MS medium), respectively. After 30 min incubation, 3′-deoxyadenosine (Cordycepin, Sigma) was added to final concentration of 0.6 mM (time 0). Seedlings were collected at various time points (0, 15, 30, 60, 90, 120 and 240 min). qRT- PCR then was performed to detect the transcript levels of pri-miRNAs and DDL. For quantification, the transcript levels of pri-miRNAs and DDL at various time points were normalized to that of eIF4a, respectively. Value of time 0 was set to 1. Error bars indicate standard deviation of three technical replications. Three biological repeats were performed and similar results were obtained.
The gel filtration was performed on an HPLC system and a HiPrep 16/60 Sephacryl S-300 HR column (GE Healthcare) at a rate of 0.5 ml/min, and 0.5 ml fractions were collected every minute. Fractions were separated by 8% SDS–PAGE and analyzed by Western blotting using antibodies recognizing CDC or YFP. The protein standards (Bio-Rad, http://www.bio-rad.com/) were used to calibrate the column contain five size standards.
|
10.1371/journal.pntd.0005110 | Migrant Workers in Malaysia: Current Implications of Sociodemographic and Environmental Characteristics in the Transmission of Intestinal Parasitic Infections | A cross-sectional study of intestinal parasitic infections amongst migrant workers in Malaysia was conducted. A total of 388 workers were recruited from five sectors including manufacturing, construction, plantation, domestic and food services. The majority were recruited from Indonesia (n = 167, 43.3%), followed by Nepal (n = 81, 20.9%), Bangladesh (n = 70, 18%), India (n = 47, 12.1%) and Myanmar (n = 23, 5.9.2%). A total of four nematode species (Ascaris lumbricoides, Trichuris trichiura, Enterobius vermicularis and hookworms), one cestode (Hymenolepis nana) and three protozoan species (Entamoeba histolytica/dispar, Giardia sp. and Cryptosporidium spp.) were identified. High prevalence of infections with A. lumbricoides (43.3%) was recorded followed by hookworms (13.1%), E. histolytica/dispar (11.6%), Giardia sp. (10.8%), T. trichura (9.5%), Cryptosporodium spp. (3.1%), H. nana (1.8%) and E. vermicularis (0.5%). Infections were significantly influenced by socio-demographic (nationality), and environmental characteristics (length of working years in the country, employment sector and educational level). Up to 84.0% of migrant workers from Nepal and 83.0% from India were infected with intestinal parasites, with the ascarid nematode A. lumbricoides occurring in 72.8% of the Nepalese and 68.1% of the Indian population. In addition, workers with an employment history of less than a year or newly arrived in Malaysia were most likely to show high levels of infection as prevalence of workers infected with A. lumbricoides was reduced from 58.2% to 35.4% following a year’s residence. These findings suggest that improvement is warranted in public health and should include mandatory medical screening upon entry into the country.
| Neglected intestinal parasitic infections (IPIs) such as soil-transmitted helminthes (STH) have been recognized as one of the main causes of illnesses especially among disadvantaged communities. The last survey of parasitic infections among migrant workers in Malaysia was conducted more than a decade ago. Although it provided useful methodological enquiries, the accrued data were not designed for policy recommendations. Over the last decade, the number of migrant workers has grown exponentially. There is an acute need for accurate information on the epidemiology of parasitic infections and infectious diseases especially as they affect urban communities in Malaysia. A particular gap has been identified in understanding the presence, persistence and epidemiology of infections among longer-term residents and immigrants who have settled in Malaysia. Hence, there is a need for a comprehensive study to establish the composition and transmission of parasitic infections in these communities with a view to developing effective methods of prevention, control and treatment of these infections. Therefore, this study is timely in adopting a scientific approach to address an important public health problem and to provide conclusions that can inform the design of effective public health policies.
| Mass migration from less developed to more developed countries have created a shift in the global population. Urbanization and extensive industrialization of developing nations have resulted in millions of migrants travelling to major urban cities around the globe to join the expanding workforces. The International Labor Organization (ILO) estimates that there are approximately 232 million international migrant workers worldwide. Globalization, demographic shifts, conflicts, income inequalities and climate change are some of the influences that drive workers and their families to cross borders in search of better employment and security [1]. In Malaysia, the robust economic growth of the different sectors has led to the mushrooming of small to large enterprises requiring high demand of a low-skilled workforce primarily in sectors such as construction, domestic and food services, manufacturing and plantation. This has attracted many to flock to the country both legally and illegally [2,3] from South East Asian (Indonesia, Cambodia, Vietnam, the Philippines and Myanmar) and West Asian countries (Nepal, India and Bangladesh) [2,3] where endemic infections are very much prevalent and most likely to pose public health problems to the local community [4,5,6,7].
Malaysia is a middle-income country whose economy has transformed into an emerging multi-sector economy and since the 1970s it has been facilitated largely by imported migrant workers. Malaysia has a higher standard of living compared with other neighboring countries in the South East Asian and West Asian region. A total of 74.7% of the population in Malaysia has undergone urbanization with 2.66% annual rate of change (2010–2015) [8]. Access to sanitation facilities in Malaysia has improved also in both urban and rural areas for up to 96.0% of the population. Meanwhile drinking water sources have improved for up to 98.2% of the population [8]. The percentage of the population in Malaysia still living below the poverty line is 3.8%, considerably lower than that of other nationalities recruited in the present study. Myanmar has reported the highest percentage of its population living below the poverty line (32.7%), followed by Bangladesh (31.5%), India (29.8%), Nepal (25.2%), Vietnam (11.3%) and Indonesia (11.3%). The push factors for migration include poor remuneration and slim employment opportunities in their home country. Meanwhile the main factors for choosing Malaysia as a destination country are perceived to be abundant opportunities, high wage levels and attractive job offers [9].
Neglected intestinal parasitic infections (IPIs) such as soil-transmitted helminthes (STH) have been recognized as one of the main causes of illnesses especially among disadvantaged communities [10, 11]. According to the World Health Organization (WHO), STH have been identified as one of 17 neglected tropical diseases, with more than 1.5 billion people or 24% of the world’s population infected [12] with roundworm (Ascaris lumbricoides), whipworm (Trichuris trichiura) and hookworms (Necator americanus and Ancylostoma duodenale) primarily through soil contaminated by human feces. These infections can cause anemia, vitamin A deficiency, stunted growth, malnutrition, intestinal obstruction and impaired development [13]. Mild infections in adults normally are asymptomatic however in serious ascariasis infections symptoms include shortness of breath, coughing/gagging/whizzing, irregular stool, abdominal pain, nausea and vomiting. While symptoms of heavy hookworm infections include itchy rash, blood in stool and abdominal pain and infections with trichuriasis include abdominal pain, inflammatory bowel and rectal prolapse. It is estimated currently up to 800 million people are infected with A. lumbricoides, 600 million people with T. trichiura and 600 million with hookworms [10,14,15].
In addition, common human intestinal protozoan infections such as Entamoeba histolytica/ dispar, Giardia duodenalis and Cryptosporidium spp. [10, 15] are also widespread. It is estimated that there are 50 million cases of invasive E. histolytica disease each year, resulting in as many as 100,000 deaths. In several parts of the world, Entamoeba infection affects 50% of the population especially in areas of Central and South America, Africa, and Asia [16]. Whilst G. duodenalis, a parasite that is frequently associated with cases of diarrheal disease throughout the world, affects approximately 200 million people worldwide [17, 18]. On the other hand, Cryptosporidium spp. infection has been reported in every region of the United States [19] and throughout the world, with approximately 4% of people in developed countries infected [20]. Intestinal protozoan infections are spread by the fecal-oral route, so infections are widespread particularly in areas with inadequate sanitation and water treatment [10,15,21,22].
There is continuous migration of populations from rural to urban areas as well as mass influx of immigrants from neighboring countries to big cities. This sudden influx of people has contributed to the mushrooming of numerous mega urban slums where the environment is conducive for the transmission of intestinal pathogens [11]. Studies on parasitic infections amongst migrant workers have been conducted worldwide particularly in Asia, for example in Thailand [23,24,25], Taiwan [26,27,28,29], Taipei [30] and in the middle east primarily in the Kingdom of Saudi Arabia; Abha district [31], Riyadh [32], Al-Khobar [33], Makkah [34], Al-Baha [35] and Medina [7]. In Qatar, Abu-Madi et al. [4,5,6] have also extensively studied the parasitic infections in migrant workers. They reported the occurrence of parasitic infections in three subsequent years among migrant workers; 33.9% in 2008, 10.2% in 2009 and 21.5% in 2011. In 2008, they recorded intestinal parasitic infections (IPI) amongst food handlers and housemaids from different geographical regions or origin. In 2009, Abu-Madi et al. looked into the trends of IPI among long-term-residents and settled immigrants after introduction of routine albendazole treatment as a condition of entry, residence and issuance of a work permit. Results reported low infection rate (10.2%) with at least one species of intestinal parasite (2.6% with helminthes and 8.0% with protozoan species). While in 2011, Abu-Madi et al. compared IPI between newly arrived and resident workers in Qatar and results showed that 21.5% of the subjects were infected with at least one of the species recorded.
In Malaysia, Suresh et al. [36] conducted a similar study more than a decade ago among migrant workers however, the study only involved clinically ill subjects from University Malaya Medical Centre. The findings of this study provided useful data but the study was not robustly designed to identify priorities for policy recommendations to the health and political authorities. Studies on intestinal parasitic infections have been conducted also among the Malaysian population and infections continue to be a public health problem especially among the poverty- stricken communities. Studies analyzing parasitic infections among various communities in Malaysia include; the Orang Asli (indigenous group) [10,11,37,38,39,40,41], plantation and rural communities [11,42,43,44,45], slum dwellers [46,11], fishing communities [47,48,49,50,51] and flat dwellers [52,53,54,11].
The current study is timely as in the past decade, the number of migrant workers has grown exponentially with a percentage increase of 49% between 2002 (1.06 million) and 2014 (2.07 million) [3]. The global DALY values of intestinal nematode infections particularly due to intestinal obstruction of Ascariasis increase from 0.024 to 0.03 in 2010 (95% uncertain interval: 0.016–0.048) [55]. Despite compulsory medical screening for workers prior to entering the Malaysian workforce, screening for parasitic infections is grossly inadequate or lacking. Therefore, there is an acute need for more accurate and up-to-date information on the parasitic infections in this particular group of workers and an understanding of the factors associated with transmission of these infections, especially as they are likely to impact significantly upon the local community through close contact, lost productivity and the heightened cost of healthcare. The addition of this screening will benefit both the government and employers in particular, due to the improved general health for the worker that further translates to better productivity.
Migrant low skilled and semi-skilled workers can only be employed in Malaysia in five working sectors, namely manufacturing, food services, agriculture and plantation, construction and domestic services. Workers, willing to participate in the current study, were recruited from September 2014 to August 2015 from various agencies and companies around Peninsular Malaysia. A minimum sample size was calculated using a formula by Leedy and Ormrod [56] based on earlier estimates of infection prevalence (36%) values in Malaysia [36]. A total of 388 migrant workers from different categories were recruited.
Questionnaires were distributed to participants to gather relevant information related to the study. An individual clinical interview with questionnaire was performed in order to collect individual information including socio-demographic data (nationality, sex, age, religion, marital status, educational level and working sector), migration history (region in country of origin, years of living in Malaysia, mode of entry, working history), environmental health (current residential area, type of accommodation, amenities), life-style habits (smoker, consumer of alcohol and user of illegal drugs), recent episodes of illness (health care utilization, mode of payment, health history) and occupational health and safety (safety hazard, personal protective equipment). In the survey, participants were also questioned on their history of taking anthelminthic drugs. The interview process was performed through an interpreter especially if migrant workers had difficulty in understanding Malay or English. All participants were fully informed of the nature of the study in order to enable maximum co-operation and completion of consent forms.
After consent was obtained and the questionnaire completed, each individual was provided with a plastic container marked with a specific identification number and the name of the participant. The participant was instructed to scoop a thumb size fecal sample into the container, ensuring that the sample was not contaminated with urine. All samples were preserved in 2.5% potassium dichromate solution and brought back to the laboratory at the Institute of Biological Science, Faculty of Science, University of Malaya. For the formalin ether concentration technique, approximately 1 to 2g of sample were mixed with 7 ml of formalin and 3 ml ethyl acetate and centrifuged for 5 minutes at 2500 rpm. After centrifugation, 4 layers were seen, composed of ethyl acetate, debris, formalin and pellets containing parasites. A drop of pellet was taken and stained with Lugol’s iodine on a clean glass slide. The slide was examined under a light microscope at 10x and 40x magnification for helminthes and protozoa, respectively. For Cryptosporidium sp., modified Ziehl-Neelsen staining technique was conducted. A smear was made on a glass slide and allowed to dry. Then the smear was fixed with methanol for about 5 minutes and afterwards flooded with cold strong neat carbol fuchsin for 5 to 10 minutes. The slide was washed in tap water and differentiated in 1% acid alcohol until color ceased to leach. The smear was next rinsed under tap water, again followed by counter staining with malachite green for 30 seconds. Slides were blotted dry and examined using 1000x oil immersion objective. Three slides per sample were examined by experienced microscopists and further confirmed by their supervisors in the Department of Parasitology, Faculty of Medicine in the University of Malaya.
Prevalence data (percentage of subjects infected) are shown with 95% confidence limits (CL95), as described by Rohlf & Sokal [57] using bespoke software. Prevalences were analyzed using maximum likelihood techniques based on log linear analysis of contingency tables using the software package SPSS (Version 22). Analysis was conducted in two phases. In the first phase, full factorial models were fitted with the intrinsic factors sex (2 levels, males and females), age (5 age classes comprising those <25 years old, 25–34 years old, 35–44 years old, 45–54 years old and those >54 years) and nationality (5 countries, Indonesia, Bangladesh, Myanmar, India and Nepal). Infection was considered as a binary factor (presence/absence of parasites). These explanatory factors were fitted initially to all models that were evaluated. For each level of analysis in turn, beginning with the most complex model, all possible main effects and interactions were investigated and those combinations that did not contribute significantly to explaining variation in the data were eliminated in a stepwise fashion beginning with the highest-level interaction (backward selection procedure). A minimum sufficient model was then obtained, for which the likelihood ratio of χ2 was not significant, indicating that the model was sufficient in explaining the data. The importance of each term (i.e. interactions involving infection) in the final model was assessed by the probability that its exclusion would alter the model significantly and these values relating to interactions including presence/absence of infection are given in the text. The remaining terms in the final model that did not include presence/absence of infections are not given but can be made available by the authors upon request.
In the second phase, models were fitted with four environmental factors (employment sector [Construction, manufacturing, plantation, food services and domestic services], educational level [no formal education, primary education only, education to high school level and to university level], accommodation [hostel/employer provided or own/rented] and years of residency [less than one year or more than 1 year] and presence/absence of infections). The most significant of the intrinsic factors detected in the first phase of the analysis was also included and the model re-run as explained above.
An ethical clearance was obtained from the ethics committee, University Malaya Medical Centre, Malaysia prior to commencement of the study (Reference number: MECID NO: 20143–40). All adults provided written, informed consent to participate in the study and a parent/guardian gave consent on behalf of any child participant. Furthermore, all individual tested positive were notified of their condition through their respected employers.
A total of 388 volunteers of migrant workers provided stool specimens. The socio-demographic profile of this subset comprised 304 males (78.4%) and 84 females. Among the males, 37.4% were between 25 to 34 years old (n = 145), 29.4% were younger than 25 (n = 114) and 23.2% older (n = 90 for 35 to 44 years). Most respondents were from Indonesia (n = 167, 43%) followed by Nepal (n = 81, 20.9%), Bangladesh (n = 70, 18%), India (n = 47, 12.1%) and Myanmar (n = 23, 5.9%). The majority were involved in the domestic service sector (n = 105, 27.1%), followed closely by the food service sector (n = 104, 26.8%), while, only a small proportion were from among those working on plantations (n = 71, 18.3%), manufacturing (n = 61, 15.7%) and construction (n = 47, 12.1%) sectors.
The demand for low and semi-skilled workers in several sectors in Malaysia has seen a dramatic rise in the number of workers entering the country from 1.06 million in 2002 to 2.07 million in 2014 [3]. The presence of such a substantial foreign work force originating from countries where parasitic infections are endemic is a major concern especially as this community is highly dynamic, and the emerging and re-emerging infectious diseases that they may carry are a great concern. For the present study we successfully recruited 388 migrant workers from their workplace who provided stool specimens compared to 173 stool specimens of clinically ill subjects from the University Malaya Medical Centre in the previous study [36]. Recruiting workers to participate in the present study was challenging mainly because this screening was not mandatory by FOMEMA (the agency responsible for the implementation, management and supervision for the nationwide mandatory health screening programme for all legal migrant workers), Ministry of Health and Immigration Department of Ministry of Home Affairs upon entry / residing in Malaysia. Other reasons included lack of interest, disgusted with feces and preoccupied with work.
Our study identified a two-fold increase of IPIs (62.9%) among workers compared to a decade ago (36.0%) [36]. Studies reporting analyses of parasitic infections among various communities in Malaysia have been conducted also among the Orang Asli (44.33%-99.2%) [10,11,37,38,39,40,41] plantation and rural communities (32.3%-70.0%) [11,42,43,44,45], slum dwellers (20.6%-90.9%) [46,11], fishing communities (54.2%-98.0%) [47,48,49,50,51] and flat dwellers (5.1%-57.0%) [52,53,54,11]. Our findings based on migrant workers are in agreement with other studies on poverty- stricken communities in Malaysia although some studies have reported fluctuations in prevalence values especially among the slum dwellers (90.9% in 1978 to 20.6% in 2014) [46,11], flat dwellers (57% in 1983 to 5.5% in 2014) [52,11] and rural communities (90.0% in 1970 to 32.3% in 2014) [43,11]. A total of 8 species of parasites were identified (A. lumbricoides, T. trichiura, hookworm, E. vermicularis, H. nana, Entamoeba sp., Giardia sp. and Cryptosporidium spp.), compared to only 6 species recorded previously (A. lumbricoides, T. trichiura, hookworm, H. nana, Giardia sp. and Blastocystis sp.) among migrant workers [36]. It is noted that prevalence studies has many limitations and in order to evaluate the impact of IPI on the workforce, further studies are necessary to guide changes in the government policy.
Soil-transmitted helminth (STH) (68.3%) infections were more prevalent compared to protozoan infections (25.5%). Of the three common intestinal nematodes, A. lumbricoides (43.3%) infections were the most frequently identified, followed by hookworm (13.1%) and T. trichiura (9.5%). In contrast, a study more than a decade ago highlighted hookworm infections as the most prevalent [36]. However, our result concurs with global findings highlighting A. lumbricoides infections as the most common helminth among the underprivileged communities [12]. A high presence of A. lumbricoides eggs contaminating public parks in Peninsular Malaysia has also been reported recently [58].
The demographic profiles of respondents comprised predominantly volunteers from rural areas in their respective countries of origin where IPIs are still very much prevalent and a major concern among the poor and in deprived communities, particularly among workers from India and Nepal where prevalence can exceed 80%. The latest study in the low socio-economic areas of South Chennai, India documented a prevalence of 75.7% with IPI [59], especially in children from rural and urban locations among whom prevalence with A. lumbricoides ranged between 60 to 91% [60]. This was the most common helminth infection in this community (52.8%). Both studies suggest that inadequate sanitation and poor drainage is likely to have contributed to disease prevalence. Similarly, parasitic infections in Nepal have also been reported as being linked to rapid, unplanned urbanization, open defecation and other unhygienic habits, as well as a lack of health awareness [61, 62, 63]. It is unlikely that workers acquired the infection in Malaysia as all the workers in this study were provided accommodation with adequate facilities such as clean water and flush toilets. It is believed that the infections continue to persist long after entry into the country and during employment and maintained due to bad hygiene practices.
Among the significant explanatory factors associated with the high prevalence of parasitic infections in this country were two main factors i.e, the number of working years in Malaysia and anthelmintic treatment. Workers with an employment history of less than a year or newly arrive workers in Malaysia were those who were most likely to be infected. In addition, they were also most likely to have no history of taking any anthelmintic drugs in the last 12 months. The improvements in health shown by the workers with over a year of residency was possibly due to the impact from better quality of life from the provision of clean water and sanitation. In the event an introduction of anthelminthic treatment is implemented on workers upon entry, this can further reduce infection and improve their overall health. This is not surprising as the mandatory medical screening procedure upon entry to this country excludes examination for IPIs and does not require administration of anthelmintic drugs to newly arrived workers [64]. Therefore our findings call for an improvement in health screening in future to include screening for parasitic infections and compulsory administration of anthelmintic drugs to workers upon entering Malaysia for employment. Such requirement is already implemented in some countries, that depend on an immigrant workforce, as for example in Qatar where currently prospective workers are required to undergo health checks at approved health clinics in their country of origin and if infection with helminths is detected, are routinely given albendazole prior to arrival as a condition for entry, residence and issuance of a work permit [5,6]. Moreover those working in the food service industry have to undergo subsequent annual compulsory examinations by the Medical Commission as a condition of the continuation of their work permits.
Transmission of intestinal nematode infections within the community is predominantly dependent on human behavior, particularly during eating and defecation, personal hygiene, and cleanliness. The high prevalence of parasitic infections among the immigrant community sampled in this study provides an insight into the conditions under which the subjects live, and reflects the availability of environmental sanitation as well as the socioeconomic status of this sector of the population in Malaysia [12]. Despite some of the workers obtaining high education levels, high disease prevalence was still observed amongst the workers possibly acquired due to the lack of sanitation and clean water in their home country compounded with behavioral factor such as bad hygiene practice that continues to persist after entry into the country. Therefore, not only screening is necessary but there is a need for workers to be further educated on good hygiene practices and knowledge of disease transmission.
These findings highlight the urgent need to refine current health polices for Malaysia and especially to include in the future mandatory screening for parasitic infections, as well as STH, of those applying for entry, work permits and residence in Malaysia. Moreover, this should be accompanied by health education campaigns and programs aimed at increasing in the community awareness of the importance of personal hygiene, sanitation, cleanliness and healthy behaviors in controlling parasitic infections [10,11,15].
|
10.1371/journal.ppat.1004467 | CD4 Depletion in SIV-Infected Macaques Results in Macrophage and Microglia Infection with Rapid Turnover of Infected Cells | In rhesus macaques (RMs), experimental depletion of CD4+ T-cells prior to SIV infection results in higher viremia and emergence of CD4-independent SIV-envelopes. In this study we used the rhesus recombinant anti-CD4 antibody CD4R1 to deplete RM CD4+ T-cells prior to SIVmac251 infection and investigate the sources of the increased viral burden and the lifespan of productively infected cells. CD4-depleted animals showed (i) set-point viral load two-logs higher than controls; (ii) macrophages constituting 80% of all SIV vRNA+ cells in lymph node and mucosal tissues; (iii) substantial expansion of pro-inflammatory monocytes; (iv) aberrant activation and infection of microglial cells; and (v) lifespan of productively infected cells significantly longer in comparison to controls, but markedly shorter than previously estimated for macrophages. The net effect of CD4+ T-cell depletion is an inability to control SIV replication and a shift in the tropism of infected cells to macrophages, microglia, and, potentially, other CD4-low cells which all appear to have a shortened in vivo lifespan. We believe these findings have important implications for HIV eradication studies.
| CD4+ T-cells are both mediators of antiviral immune response and critical targets for HIV replication. We have previously shown that experimental depletion of CD4+ T-cells prior to SIV infection in rhesus macaques results in higher viremia and the emergence of CD4-independent SIV-envelopes. The findings reported in this new study of CD4 depletion address key unanswered questions about the phenotype, location, and lifespan of the sources of the increased viral replication in the absence of CD4+ T-cells. Altogether, our new data indicate that depletion of CD4+ T-cells prior to SIV infection results in activation of monocyte and massive infection of tissue-resident macrophages, which appear to be the predominant population of productively infected cells. Furthermore, our analysis of the slope of viremia decline after initiation of antiretroviral therapy suggests that the lifespan of these virus targets is markedly shorter than those previously estimated for macrophages. In summary, in the context of CD4+ T-cell depletion macrophages can be highly infectable, exhibit rapid turnover, and short in vivo lifespan. These finding raises a suggestive hypothesis that eradication of HIV from this reservoir could be enhanced by therapeutics able to modulate monocyte/macrophage turnover.
| The interaction between HIV and CD4+ T-cells is complex, and may result in contrasting effects with respect to virus replication. On the one hand, CD4+ T-cells have a beneficial role as mediators of antiviral immune responses, both directly and by providing help for HIV-specific CD8+ T-cells and B cells [1]–[4]. On the other hand, CD4+ T-cells are key targets for infection and sustain virus replication [5], [6]. To better understand the relationship between CD4+ T-cell availability and HIV replication, we recently conducted a CD4+ T-cell depletion study in rhesus macaques (RMs) prior to SIV infection [7]. This previous study showed that antibody-mediated depletion of CD4+ T-cells was associated with increased virus replication and rapid disease progression [7]. Furthermore, using in vitro systems we demonstrated the emergence of CD4-independent SIV envelopes capable of mediating entry into cells expressing CCR5 without CD4. The absence of antibodies targeting conserved CD4-inducible epitopes has been proposed as one of the mechanisms allowing CD4-independent SIV to emerge in CD4-depleted RMs [8]. Of note, in that study one RM with the least effective CD4+ T-cell depletion showed the lowest viremia and survived throughout the entire study, suggesting that intermediate levels of CD4+ T-cells may be the ideal balance between the beneficial and harmful contribution of CD4+ T-cells to disease progression. This previous study raised some critical questions, including: (i) is partial depletion of CD4+ T-cells beneficial? (ii) What cells are the main sources of virus replication in the absence of CD4+ T-cells, and where are they located? (iii) What is the in vivo lifespan of these productively infected cells? And finally, (iv) can we identify correlates of the high viremia associated with CD4 depletion?
To answer these questions, we designed a new study where we used a concentration of the CD4 depleting antibody CD4R1 which generated variable levels of CD4+ T-cell depletion allowing us to test how CD4+ T-cell availability impacts SIV infection. Furthermore, we performed an extensive combined immunohistochemistry (IHC) and in situ hybridization (ISH) analysis on colon, jejunal and brain tissues collected at necropsy from the eight SIV-infected RMs - four CD4-depleted and four controls - included in Ortiz A.M et al [7]. In contrast to our current study, these animals were not ART-treated when euthanized. As such, they represent the ideal samples to investigate the presence and the phenotype of productively infected cells. Of note, both the present and the former studies were performed at the same facility, using the same animal species, virus, route and dose of infection.
We found that depletion of RM CD4+ T-cells prior to SIV infection is associated with dramatic changes in the course of disease, including post-peak viral load two-logs higher than undepleted controls, expansion of pro-inflammatory monocytes, and massive activation and infection of macrophages and microglia that appear to be the predominant population of productively infected cells. Finally, our analysis of the slope of viremia decline after initiation of antiretroviral therapy (ART) suggests that, in the absence of CD4+ T-cells and in the presence of high levels of activation, the lifespan of these virus targets is significantly longer than controls, but markedly shorter in comparison to those previously estimated for macrophages.
CD4+ T-cells were depleted in eight RMs using a single administration of CD4R1 antibody at 50 mg/kg, as recommended by the “NIH Nonhuman Primate Reagent Resource” protocol (Figure 1a). Of note, this regimen generated variable levels of CD4+ T-cell depletion in our pilot studies. Four untreated animals were included as controls. All 12 RMs were infected with SIVmac251 (i.v. 3,000 TCID50) six weeks post CD4R1 treatment, in order to avoid the possible confounding effect of direct antiviral activity of the CD4R1 antibody. At day-52 post-infection (p.i.) all animals were treated with a three-drugs antiretroviral regimen (PMPA, FTC, raltegravir). Blood, bone marrow aspirate (BM), lymph nodes (LN) and rectal biopsies (RB) were collected longitudinally and at necropsy (Figure 1a).
The efficacy of CD4+ T-cell depletion was determined in blood and tissues by flow cytometry. Figure 1 shows the longitudinal levels (mean±S.E) of the circulating CD4+ T-cells expressed as the absolute number (b) or fraction of CD3+ T-cells (c). Treatment with 50 mg/kg CD4R1 resulted in a severe depletion in three RMs (orange), but an intermediate depletion in the remaining five treated animals (blue). Of note, there was no association between the extent of CD4+ T-cell depletion and the age, sex, and weight of the animals. Representative flow plots of the percentage of CD4+ T-cells pre- and post-depletion are shown in Figure 1d for a severely (top) and an intermediately (bottom) depleted RM. Severe depletion was characterized by a nadir CD4+ T-cell frequency of 3.18±0.356% (93.5% depletion from baseline), and CD4+ T-cell counts of 25.64±6.877 cells/µL (Figure 1b,c). Intermediately-depleted animals had a nadir CD4+ T-cell frequency of 18.6±3.021% (57% depletion from baseline) and CD4+ T-cell counts of 181.2±58.98 cells/µL. In both severely- and intermediately-depleted RMs, the percentage and absolute count of CD4+ T-cells were significantly lower (P<0.01) than controls at all experimental points post-depletion (D-35, D-11, D0; Figure 1b,c). The same gradient of CD4+ T-cell depletion was confirmed in BM and LN tissues (Figure 1e). Since we used an anti-CD4 antibody clone for immunophenotyping that is not cross-blocked by CD4R1, and additionally found a significant reduction in the fraction of CD3+ T-cells and no selective increase in the fraction of CD3+CD4−CD8− T-cells, we are confident that the loss of CD4+ T-cells observed following CD4R1 administration is indeed a true depletion of these cells, rather than a masking. As expected based on previous studies [7], depletion of CD4+ T-cells was minimal at mucosal sites. In severely-depleted RMs, depletion of CD4+ T-cells induced high levels of proliferation, with in average 72% of the remaining CD4+ T-cells expressing Ki-67 at D0 (Figure 1f). Although we cannot exclude the contribution of other mechanisms, including reactivation of latent infections, we interpreted this high proliferation as an attempt of the immune system to reconstitute the depleted CD4+ T-cell compartment.
In summary, the variable efficacy of CD4+ T-cell depletion achieved in this study allowed us to study SIV viral dynamics and disease progression in RMs with normal, low or extremely low levels of circulating and tissue resident CD4+ T-cells, therefore representing an ideal model to address how CD4+ T-cells impact SIV infection.
Six weeks post CD4R1 administration all twelve RMs were infected with SIVmac251 (i.v. 3,000 TCID50). Peak viral load (day 10 p.i.) was similar between CD4-depleted and control RMs (Figure 2a,b). However, whereas control animals showed a rapid drop in viral load after the peak, CD4-depleted animals maintained viremia close to peak levels and ∼2-logs higher than controls in later infection. This difference in viral load was statistically significant (P<0.001) at each experimental point between day 21 and day 52 p.i. (Figure 2a,b). Intermediately-depleted animals exhibited viral kinetics remarkably similar to those of severely-depleted animals (Figure 2b). Furthermore, following SIV infection the absolute numbers of CD4+ (a), CD4+Ki-67+ (b) or CD4+CCR5+ (c) T-cells were comparable in intermediately-depleted RMs when compared to severely-depleted animals, and significantly lower than those found in controls at all the experimental points (Figure S1). Since severely- and intermediately-depleted RMs had comparable viremia and numbers of activated/proliferating T-cells, for the remaining of the study all eight treated animals were grouped together and defined as CD4-depleted RMs.
These results show that a partial depletion of CD4+ T-cells is sufficient for the establishment of an infection phenotype of persistently high viremia. Furthermore, the presence of a 2-log higher viremia together with a dramatic loss of CD4+ T-cells suggests a different source of viral burden in CD4-depleted animals.
Increased activation and turnover of monocytes predict progression to AIDS in SIV-infected RMs even better than CD4+ T-cell number [9]–[11]. Hence, we quantified the levels and Ki-67 expression of the monocyte subsets in CD4-depleted and control RMs. Monocytes were defined as classical (CD14+CD16−), pro-inflammatory (CD14+CD16+), and non-classical (CD14−CD16+) based on the expression of CD14 and/or CD16 (Figure 3a). In the control animals, stable levels of all monocyte subsets at day 52 p.i. followed their initial expansion. However, classical and pro-inflammatory monocytes in CD4-depleted RMs continued to increase, with numbers of CD14+CD16+ monocytes significantly higher than those of controls at days 28 (P = 0.0283) and 52 (P = 0.0283) p.i. (Figure 3b). Furthermore, and consistent with an activated/pro-inflammatory status and an increased output from bone marrow, the number of CD14+CD16+ monocytes expressing Ki-67+ in CD4-depleted RMs was significantly higher as compared to controls at days 28 (P = 0.0162) and 52 (P = 0.0485) p.i. (Figure 3b). At day 52 p.i. CD4-depleted RMs also have a higher number of proliferating CD14+CD16− and CD14−CD16+ monocytes than controls, although the difference was not statistically significant (Figure 3b). In CD4-depleted RMs plasma levels of soluble CD163 (sCD163), a marker of monocyte activation associated with rapid disease progression [9], [12], [13], were significantly higher than in controls at day 28 (p = 0.016) and day 52 (p = 0.048) p.i. (Figure 3c). Of note, at day 52 p.i., levels of sCD163 strongly correlated with the numbers of total (r = 0.7483, P = 0.0070) and Ki-67+ (r = 0.6923, P = 0.0155) CD14+CD16+ monocytes (Figure 3d), as well as with viremia (r = 0.8322; P = 0.0013) (Figure S2). Thus, depletion of CD4+ T-cells prior to SIV infection results in increased number, activation and turnover of monocytes during early SIV infection. As indicated in Figure 1a, all 12 SIV-infected RMs were started on ART at day 52 p.i. The numbers of pro-inflammatory monocytes (CD14+CD16+; P = 0.0145) and their levels of Ki-67 expression (P = 0.0189) after 12 days of ART (day 64 p.i) were significantly decreased as compared to pre-ART (day 52 p.i.) levels, and become comparable to those found in controls (Figure 3b). Classical (CD14+CD16−) and non-classical (CD14−CD16+) monocyte levels were not significantly affected by ART (Figure 3b). Levels of sCD163 remained significantly higher in CD4 depleted animals than controls after 12 days of ART (Figure 3c). Of note, the interpretation of the effects of ART in the aforementioned parameters, in particular for sCD163, is complicated by the fact that we were able to only treat the animals for a short period. Indeed, and perhaps as a consequence of the fact that CD4-depletion resulted in very high virus replication, seven of the eight depleted RMs had to be euthanized for AIDS-related reasons briefly after ART initiation (Table S1).
We next investigated the levels of SIV infection of macrophages in peripheral lymph node and intestine from CD4-depleted and control RMs by immunofluorescence staining for cell lineage markers combined with fluorescence in situ hybridization (F-ISH) for SIV vRNA. Since monocyte/macrophage express CD4, we used CD3 to determine the infection frequency of CD4+ T-cells. At day 42 p.i. SIV vRNA+ cells were more frequent in the LN of CD4-depleted animals as compared to controls (Figure 4a), consistent with the ∼2-log higher plasma viremia found at the same experimental point. The vast majority of SIV vRNA+ cells expressed the T cell marker CD3 in undepleted controls, but macrophage markers (CD68 and/or CD163) in CD4-depleted RMs. We then performed the same F-ISH staining on day 42 rectal biopsies from the RMs included in the current study as well as on colon and jejunal tissues at necropsy from the four CD4-depleted animals included in Ortiz A.M et al [7]. In contrast to our current study, those latter animals were not ART-treated and thus showed high viremia when euthanized. The same phenomena of increased levels of total SIV vRNA+ cells and expression of macrophage markers by infected cells were observed in the intestine of CD4-depleted (Figure 4b), but not of control RMs. Quantitative image analysis of LN and intestinal tissues showed that in undepleted controls more than 80% of SIV vRNA+ cells were CD3+ T-cells, while in CD4-depleted RMs ∼80% of SIV vRNA+ cells were CD68+ and/or CD163+ macrophages (Figure 4c).
Although the majority of infected cells in CD4-depleted animals were macrophages, it is possible that infection of these cells was a result of high viral loads, and they did not contribute substantially to the total viral load because they produced little virus per infected cell. In order to determine whether viral production from macrophages was likely to contribute to the total viral load, we sought to compare the relative abundance of SIV vRNA in productively infected CD4+ T cells to that of macrophages in four CD4-depleted RMs. We measured the volumetric sum of the SIV vRNA intensity integration values in defined productively SIV infected CD3+ and CD68/CD163+ cells in situ from confocal images collected under identical laser settings. Using this approach we were able to demonstrate that, within a particular host, macrophages have similar or even higher (on average two-fold more) SIV vRNA content per cell than productively infected CD4+ T-cells (Figure 4d).
Altogether, these data indicate that depletion of CD4+ T-cells prior to SIV infection results in activation of monocyte and massive infection of tissue-resident macrophages. These infected macrophages have higher levels of SIV vRNA than infected CD4+ T cells, and, since they constitute 80% of all infected cells in CD4-depleted animals, they are most likely the major source of systemic viremia in these animals.
We then investigated if the observed high viremia, monocyte activation, and infection of macrophages in peripheral tissues were associated with central nervous system (CNS) virus dissemination and pathology. First, we investigated the presence of infected cells in brain sections collected at necropsy from the CD4-depleted and non-depleted (SIV+ controls) SIV-infected RMs included in our previous study (in contrast to the current study, these animals were not treated with antiretroviral drugs at the time of necropsy [7]). For the SIV negative controls, we used brain sections from historically SIV-uninfected RMs present at the Tulane National Primate Research Center (SIV− controls). In CD4-depleted RMs SIV vRNA+ cells were found throughout the parenchyma (Figure 5a) and had a stellate morphology typical of microglia. The high number of infected cells led us to evaluate, in the same brain sections, the levels of activation by IHC/IF staining for CD163, HLA-DR, and proliferating cell nuclear antigen (PCNA, a marker of perivascular macrophages that has been used as a marker of DNA repair in macrophages [14]). As shown in the representative images (b) and in the quantitative analysis (c) of Figure 5, the expression of CD163 (P = 0.0013; P = 0.0047), HLA-DR (P = 0.0571; P = 0.0571), and PCNA (P = 0.0026; P = 0.0121) in cells with location and stellate morphology typical of microglia were significantly higher in CD4-depleted than SIV-uninfected or SIV-infected controls. The surprising finding that a significant number of microglia are activated and productively infected was confirmed by triple fluorescent labeling for SIV vRNA (red), CD163 (blue) and the microglia-specific marker IBA-1 [15] (green) (Figure 5d). Quantification of these stainings showed numbers of SIV vRNA+ IBA-1+ microglia markedly higher in CD4-depleted compared to non-depleted SIV-infected animals (11.3±8.4 vs. 0.2±0.2; P = 0.0286). Of note, we found a direct correlation between the number (cells/mm2) of SIV vRNA+ cells and PCNA+ cells in the eight CD4-depleted animals (r = 0.8289, P = 0.0302).
Collectively, these results indicate that depletion of CD4+ T-cells in RMs results in massive activation and infection of microglia.
One way of testing the longevity of infected cells is to block new rounds of infection with ART and measure the rate of viral decay (which reflects the death rate of productively infected cells present at the time of treatment). Thus, if infected macrophages have a longer survival than infected CD4 T cells, we should expect to see a much slower decay of virus under therapy in CD4-depleted RMs, where infected macrophages are the major source of systemic viremia.
To directly estimate the in vivo lifespan of the productively infected cells, all 12 SIV-infected RMs were started on ART at day 52 p.i. We then measured and modeled the slope of viral load decay during ART as described in previous studies [16]–[19]. As described above, seven of the eight depleted RMs had to be euthanized for AIDS-related reasons briefly after ART initiation (Table S1). As a result of this, we were able to perform only three sample collections, one at ART initiation (day 0) and the other two at day 7 and day 12 on-ART (Figure 6a). Since at day 12 on-ART we already lost three CD4-depleted animals, the lifespan of productively infected cells was estimated based on viral load changes between day 0 and day 7, although the same conclusions were reached if we used the smaller subset of animals alive at day 12. As expected, the viral decay in control animals was rapid, with an estimated half-life of infected cells of 0.775±0.01 days, which was consistent with previous estimates. The lifespan of productively infected cells was significantly longer in the CD4-depleted animals (1.33±0.108 days, P = 0.0238; Figure 6b), but was considerably shorter than expected for HIV infected macrophages [20]–[22]. Thus, in CD4-depleted RMs, macrophages constitute the main productively infected cell type, produce the majority of virus, and show a relatively short in vivo lifespan. Further analysis of cell death in a subset of animals showed a statistically significant higher number of TUNEL positive cells (the vast majority of which express CD163) in the brain parenchyma in CD4-depleted RMs than in controls (p = 0.0439) and CD4-depleted animals on-ART (p = 0.0002) (Figure S3a). The increased cell death was not Caspase-3 mediated since the number of cells positive for the active form of Caspase-3 was comparable between SIV-infected CD4-depleted RMs and controls (Figure S3b).
We previously showed that a severe depletion of CD4+ T-cells is associated with increased viremia, emergence of CD4-independent viruses and rapid progression to AIDS in SIV-infected RMs [7]. In this new study we used CD4R1, a rhesus recombinant anti-CD4 depleting antibody, to achieve a gradient of CD4+ T-cells in vivo. This strategy allowed us to experimentally infect with SIVmac251 RMs with normal, low or very low level of CD4+ T-cells in blood and lymphoid tissues. The critical questions we aimed to answer in this study were the following: (i) is partial depletion of CD4+ T-cells beneficial? (ii) What cells are the main sources of virus replication in the absence of CD4+ T-cells, and where are they located? (iii) What is the in vivo lifespan of these productively infected cells? And (iv) can we identify correlates of the high viremia associated with CD4 depletion? Answering these questions may have important implications for our understanding of HIV pathogenesis and latency.
Viral loads were indistinguishable between CD4-depleted and control RMs for the first 14 days p.i. However, severely- and intermediately-depleted RMs maintained viremia ∼2-logs higher than undepleted controls at all subsequent experimental points. Thus, partial depletion of CD4+ T-cells prior to infection did not result in a more benign course of infection and was sufficient to generate the higher viremia phenotype. CD4R1 was efficient in depleting CD4+ T-cells from blood, BM, and LN but not from mucosal sites. Thus, mucosal CD4+ T-cells may be responsible for maintaining peak viral loads in depleted RMs similar to that in controls, before being severely depleted in the first weeks of infection due to the high levels of SIV replication. Of note, our data on the level of proliferating cells does not support the possibility that CD4-depleted RMs have higher numbers of T-cells that can support viral replication as a result of the high level of activation/proliferation in the remaining CD4+ T-cells. Thus, in RMs the association of viremia that is ∼2-logs higher with a profound loss of CD4+ T-cells is consistent with an alternative critical source of virus replication.
The observed high viremia in the context of very low levels of CD4+ T-cells prompted us to investigate the susceptibility of macrophages/myeloid cells to SIV infection. Remarkably, we showed directly in situ that CD4-depletion induces a significant shift in the nature of productively infected cells, with 80% of SIV vRNA+ cells in LN and mucosal tissues being CD3+ T-cells in undepleted controls but CD68/CD163+ macrophages in CD4-depleted RMs. It has recently been described that increased activation and rapid turnover of monocytes, in particular those with a CD14+CD16+ phenotype, associates with macrophage destruction in tissues and predicts the tempo of progression to AIDS in SIV-infected RMs [9]–[11]. In response to activation, CD163 is cleaved from the cellular surface of monocytes/macrophages and is shed as sCD163 [23]. Recently, sCD163 has been indicated as a strong correlate of monocyte activation and turnover as well as disease progression in SIV-infected RMs [9]. Consistent with these studies, CD4-depleted RMs, but not undepleted controls, experienced a substantial increase in the number of circulating CD14+CD16+ monocytes. Furthermore, plasma levels of sCD163 were significantly higher in CD4-depleted than in undepleted control RMs, and directly correlated with viral load as well as with the numbers of total and proliferating CD14+CD16+ monocytes. Although in vivo BrdU labeling studies were not performed, the findings of higher frequency of Ki-67+ monocytes and sCD163 plasma levels are consistent with a model in which CD4-depletion prior to SIV infection results in increased activation/turnover of monocytes.
The key findings we described in CD4-depleted animals - high viral load, expansion of pro-inflammatory monocyte, increased level of sCD163, and massive infection of tissue-resident macrophages/myeloid cells – are all potential markers of CNS infection [7], [9], [13], [24]–[27]. In this study, CD4-depleted animals were euthanized during ART, thus the analyzed brain tissues did not represent the ideal samples to investigate the presence of productively infected cells. To overcome this limit, we stained brain sections collected in our previous study of CD4+ T-cell depletion that did not include ART [7]. It is worth mentioning that both the present and the former studies were performed at the same facility, using the same species, virus, route and dose of infection. Furthermore, the high viremia and rapid disease progression phenotype was remarkably similar between the two studies. Our combined IHC/ISH approach provided three surprising results. First, a large fraction of microglial cells from CD4-depleted animals expressed high levels of activation/proliferation/DNA repair markers, such as CD163, HLA-DR and PCNA, that were absent or exclusively present in perivascular macrophages in brain tissue of uninfected RMs and undepleted SIV-infected RMs. Second, the amount of SIV vRNA+ cells was remarkably higher in the brain of CD4-depleted RMs when compared to undepleted SIV-infected controls, thus consistent with what we found in LN and RB tissues. Third, and uniquely for the CD4-depleted RMs, SIV vRNA+ cells include not only perivascular macrophages but also cells with anatomic location, morphology, and phenotype typical of microglia. Demonstrable in vivo infection of microglia in SIV-infected RMs, particularly at the extent found here, is an exceedingly rare event even in the well-developed models of accelerate SIVE induced by depletion of CD8+ T-cells or by joint infection with an immune-suppressive and a neurovirulent SIV-variants [9], [15], [28].
Based on the paucity of CD4+ T-cells and the infection of macrophages/myeloid cells, which are typically thought to be long-lived cells, we hypothesized that in CD4-depleted RMs viral load decay during ART would be slower when compared to undepleted controls. In CD4-depleted RMs the lifespan of the main productively infected cells was 1.3 days, considerably shorter than what is generally expected for macrophages [20], [21], [29].
Two caveats must be considered when discussing this result. The first is that the decay model assumes that viral load is at steady-state and ART is equally effective at inhibiting viral replication in depleted and control animals. The second is related to a specific limitation of our study. Due to the poor clinical conditions of CD4-depleted RMs at ART initiation and their rapid disease progression, we could not sample these animals as frequently and as long as originally planned. Thus, the estimates of viral decay rates are based on a few relatively early experimental time points on-ART. However, given the rapid rate of decay in both groups, we were able to measure a substantial drop in viral load over seven days in both cases, and observed a significant difference in the decay rate of infected cells between control and CD4-depleted animals. These caveats notwithstanding, we concluded that the bulk of the available data are consistent with the idea that– in the context of CD4+ T-cell depletion and high levels of activation/inflammation—macrophages can be highly infectable, exhibit rapid turnover, and short in vivo lifespan when infected. In this regard, our data are consistent with recent findings indicating that rapid turnover of monocyte/macrophages determined by in vivo Brdu labeling predicted AIDS progression better than viral load or lymphocyte activation [9], [10], [30]. An alternative interpretation of the short lifespan of productively infected cells in CD4-depleted RMs is that the few remaining productively infected CD4+ T-cells produce much more virus than productively infected macrophages on a per cell basis. Thus in this scenario, T-cells could still constitute the major source of virus production measured in the plasma. However, this seems extremely unlikely because the differences in virus production would need to be dramatically skewed in favor of productively infected CD4+ T-cells, considering that they represent only 20% of all productively infected cells. In fact, when we measured in situ the relative volumetric abundance of SIV vRNA within CD3+ T cells and CD68/CD163+ macrophages, we found that macrophages on average have even higher per cell SIV vRNA content (∼2-fold greater) compared to productively infected CD4+ T-cells within the same host. With macrophages representing approximately 80% of all productively infected cells in all examined tissues, and with higher levels of SIV vRNA per cell in infected macrophages, it is safe to conclude that macrophages are the main productively infected cells in CD4-depleted animals. It is important to note that, due to the short length of ART-treatment, our data do not exclude the existence of a subpopulation of SIV-infected, long-lived macrophages. However, this would constitute a very small percentage of total viral production at most, because of the observed rapid viral load seen in CD4-depleted animals.
In conclusion, our study demonstrates that, in SIV-infected RMs, the net effect of CD4+ T-cell depletion is the inability to control SIV replication and a shift in the pattern of infected cells from CD4+ T-cells to macrophages, microglia, and, potentially, other CD4-low cells. These findings have important implications for functional cure and eradication studies as they indicate that macrophages and microglia can be critical target for virus infection in the context of a severely compromised immune system. Furthermore, our finding that HIV-infected macrophages can be short-lived if highly activated raises a suggestive hypothesis that eradication of HIV from this reservoir could be enhanced by therapeutics able to modulate monocyte/macrophage turnover.
All animal experimentations were conducted following guidelines established by the Animal Welfare Act and the NIH for housing and care of laboratory animals and performed in accordance with Institutional regulations after review and approval by the Institutional Animal Care and Usage Committees at the Yerkes National Primate Research Center (YNPRC). All efforts were made to minimize suffering. All the blood and tissue collections were obtained from RMs housed at the Yerkes National Primate Research Center, which is accredited by American Association of Accreditation of Laboratory Animal Care. RMs are fed standard monkey chow (Jumbo Monkey Diet 5037, Purina Mills, St Louis, MO) twice daily. Consumption is monitored and adjustments are made as necessary depending on sex, age, and weight so that animals get enough food with minimum waste. SIV-infected RMs are singly caged but have visual, auditory, and olfactory contact with at least one social partner, permitting the expression of non-contact social behavior. The YNPRC enrichment plan employs several general categories of enrichment. Animals have access to more than one category of enrichment. IACUC proposals include a written scientific justification for any exclusions from some or all parts of the plan. Research-related exemptions are reviewed no less than annually. Clinically justified exemptions are reviewed more frequently by the attending veterinarian. This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, a national set of guidelines in the U.S. and also to international recommendations detailed in the Weatherall Report (2006). This work received prior approval by the Institutional Animal Care and Use Committees (IACUC) of Emory University (IACUC protocol #2000353, entitled “Homeostasis of CD4+ T cells in non-human primates”). Appropriate procedures were performed to ensure that potential distress, pain, discomfort and/or injury was limited to that unavoidable in the conduct of the research plan. The sedative Ketamine (10 mg/kg) and/or Telazol (4 mg/kg) were applied as necessary for blood and tissue collections and analgesics were used when determined appropriate by veterinary medical staff.
Twelve female RMs were included in this study. Eight of them were treated with a single administration of the anti-CD4 antibody CD4R1 at 50 mg/kg (intravenous), as recommended by the “NIH Nonhuman Primate Reagent Resource” protocol. CD4R1 has rhesus constant regions and rhesus variable framework sequences. Only the CDRs (and a few amino acids critical for Ig conformation) are derived from the original mouse antibody. Four untreated animals were included as controls. All 12 RMs were infected with SIVmac251 (i.v. 3,000 TCID50) six weeks post CD4R1 treatment. At D52 post-infection all animals were treated once a day with Tenofovir (30 mg/kg; s.c.), Emtricitabine (30 mg/kg; s.c.) and Raltegravir (100 mg; oral). Furthermore, we used paraffin embedded tissues, including brain, collected at necropsy from the eight SIV-infected RMs (four CD4-depleted and four controls) included in Ortiz A.M et al [7]. All animal experimentations were conducted following guidelines established by the Animal Welfare Act and the NIH for housing and care of laboratory animals and performed in accordance with Institutional regulations after review and approval by the Institutional Animal Care and Usage Committees at the Yerkes National Primate Research Center (YNPRC). All efforts were made to minimize suffering.
Collections and processing of blood, bone marrow aspirate (BM), lymph nodes (LN) and rectal (RB) biopsies were performed longitudinally and at necropsy as previously described [31]–[34]. Briefly, blood samples have been used for a complete blood count, and plasma separated by centrifugation within 1 h of phlebotomy. Peripheral blood mononuclear cells were prepared by density gradient centrifugation. BM aspirates were performed using an aspiration kit to remove approximately 1 mL. For rectal biopsies, an anoscope has been placed a short distance into the rectum and up to 20 pinch biopsies obtained with biopsy forceps. RB-derived lymphocytes have been isolated by digestion with 1 mg/ml collagenase for 2 h at 37°C, and then passed through a 70-µm cell strainer to remove residual tissue fragments. For lymph node biopsies, the skin over the axillary or inguinal region have been clipped and surgically prepared. An incision has been made over the LN, which has been exposed by blunt dissection and excised over clamps. Biopsies have been homogenized and passed through a 70-µm cell strainer to mechanically isolate lymphocytes. All samples were processed, fixed (1% paraformaldehyde), and analyzed within 24 hours of collection.
Twelve-parameter flow cytometric analysis was performed on WB, LN, RB and BM derived cells according to standard procedures using a panel of monoclonal antibodies that we and others have shown to be cross-reactive with RM [35]–[37]. Predetermined optimal concentrations were used of the following antibodies: anti-CD3-Alexa700 (clone SP34-2), anti-CD3-APC-Cy7 (clone SP34-2), anti-CD4-Pacific Blue (clone OKT4), anti-CD8-APC-Cy7 (clone SK1), anti-CD95-PE-Cy5 (clone DX2), anti-CCR5-APC (clone 3A9), anti-Ki-67-Alexa700 (clone B56), anti-Ki-67-FITC (clone B56), anti-CD14-Pe-Cy7 (clone M5E2), anti-CD16-BV421 (clone 3G8), anti-CD62L-PE (clone SK11), anti-CCR7-PE-Cy7 (Clone 3D12) (all from BD Pharmingen); anti-CD28-ECD (clone CD28.2) (Beckman Coulter); anti-CD8-Qdot705 (clone 3B5) and Aqua Live/Dead amine dye-AmCyan (Invitrogen). Intracellular staining for Ki-67 was performed at room temperature for 30 minutes following permeabilization with cytofix/cytoperm (BD Bioscience). Flow cytometric acquisition was performed on an LSRII cytometer driven by the FACS DiVa software. Analysis of the acquired data was performed using FlowJo software (TreeStar) and graphs were prepared using Prism version 6.0 (GraphPad).
Quantitative real-time RT-PCR assay to determine SIV viral load was performed as previously described [38].
To detect viral RNA in the tissues, in situ hybridization for SIV was performed using riboprobes as described previously [39]. Briefly, 7 µm-thick formalin-fixed, paraffin-embedded tissue sections were de-paraffinazed through a graded series of xylenes and ethanol and rehydrated in water threated with diethylpyrocarbonate (DEPC; Sigma; Aldrich, St. Louis, MO) before antigen retrieval by boiling in a microwave for 20 minutes in citrate buffer pH 6. The tissue sections were blocked with hybridization buffer containing 50% formamide with denatured herring sperm DNA and yeast tRNA at 10 mg/ml each in a humidified chamber at 45°C for 1 hr. Hybridization was performed with SIV-digoxigenin-labeled anti-sense riboprobes (Lofstrand Labs Ltd; Gaithersburg, MD) that were applied to the tissue sections at 10 ng/slide in hybridization buffer and incubated overnight at 45°C. After hybridization, slides were washed sequentially with 2× SSC, 1× SSC, and 0.1× SSC. The slides were incubated in TBS (tris buffer saline, 10 mM, pH 7.4) and followed with blocking solution for 1 h. Alkaline phosphatase-conjugated sheep anti-digoxigenin antibody diluted at 1∶200 (Roche; Penzberg, Germany) was used to detect hybridized digoxigenin-labeled probes. Either Dako Liquid Permanent Red or NBT/BICP substrate-chromogens system (Dako, Inc. Carpinteria, CA) were prepared according to manufacturer's instructions and added to the tissue for 20 min at room temp to develop the reaction. Controls included matched positive and negative tissues hybridized with digoxigenin-labeled sense RNA-labeled probes and processed in parallel.
Formalin-fixed, paraffin-embedded tissue sections (5–6 µm thick) were examined by immunohistochemical staining using the following primary antibodies: Anti CD163 (clone EDHu-1, AbD Serotec), HLA-DR (clone LN3, eBioscience), and PCNA (clone PC10, DAKO). Reactivity to primary antibodies was detected using the MACH3 alkaline phosphatase polymer detection kit from Biocare Medical (Concord, CA) with either NBT/BCIP substrate system for light microscopy or Permanent red for fluorescent microscopy both from Dako (Carpinteria, CA) after which sections were counterstained using YO-PRO nuclear stain (Life Technologies, Grand Island, NY). As controls, duplicate sections were processed in the absence of primary antibodies.
Triple label confocal microscopy was performed to co-localize SIV-RNA with cell type specific markers to determine the immunophenotype of infected cells as described previously [39]. Immunofluorescent labeling for microglia (rabbit polyclonal against IBA-1, Wako) and macrophages (Mouse IgG1 monoclonal to CD163, novocastra) was performed after ISH as previously described [15]. After incubation with the primary antibodies and subsequent washes, the appropriate species-specific secondary antibodies were applied; AlexaFluor 488 (green) conjugated goat-anti-rabbit (Invitrogen, Carlsbad, California) and AlexaFluor 647 (far-red, shown in blue) conjugated goat-anti-mouse IgG1 (Invitrogen, Carlsbad, California), respectively. To image the sections a Leica TCS SP2 confocal microscope equipped with 3 lasers (Leica Microsystems, Exton, PA) with a resolution of 512×512 pixels was used. The confocal imaging was performed using sequential mode to separately capture the fluorescence from the different fluorochromes. Volocity Software (v 5.5, Perkin-Elmer) was used to render the images from the Leica z stacks. Adobe Photoshop software (Version CS6; Adobe Systems) was used to assemble the images.
Quantitation of positive cells was performed manually for the SIV-RNA detection with NBT/BICP (purple-black color) by counting positive cells per field in 10 randomly selected fields encompassing a minimum of 120 mm2. The numbers of infected cells are expressed as cells per mm2. Quantitation of cells labeled for the cell type specific antigens was performed on immunolabeled slides using Inform v2 1.2 software after capturing 10 randomly selected fields encompassing a minimum of 30 mm2 with a CRI-multispectral camera using Nuance software v 3.0.2 (Perking-Elmer). The positive cell numbers are express as cells in mm2. Prism v5 software (GraphPad software) was used to present the data in a graphic form.
Fluorescent in situ hybridization (F-ISH) was performed using our newly designed SIV riboprobes as previously published [7]. In brief, 5-µm tissue sections were mounted on Superfrost Plus Microscope Slides (Fisher Scientific), heated at 60°C for 1 h, dewaxed in xylenes and graded ethanols, and rehydrated in double-distilled H2O. Heat-induced epitope retrieval was performed by placing slides in in 10 mM Citrate (pH 6.0) containing 0.05% Tween-20 or Pretreatment-2 buffer (Advanced Cell Diagnostics, Inc.) at 100°C for 5 min followed by immediate immersion into HyPure molecular biology–grade H2O (Hyclone; Thermo Scientific) at room temperature. The slides were then incubated for 5 minutes at 37°C in a Tris-buffered solution containing 2 mM CaCI2 and proteinase K (1.25 µg/ml) then washed in HyPure molecular biology–grade H2O (Hyclone; Thermo Scientific), acetylated (0.25% acetic anhydride) for 20 minutes, and placed in 0.1 M triethanolamine (pH 8.0) until hybridization. Sections were then covered with hybridization solution (50% deionized formamide, 10% dextran sulfate, 0.6 M NaCI, 0.4 mg/ml yeast RNA (Ambion Inc.), and 1× Denhardt medium in 20 mM HEPES buffer (pH 7.2) with 1 mM EDTA) containing 100–400 ng/ml pooled SIV riboprobes and hybridized for 18 hours at 48°C. After hybridization, slides were washed in 5× SSC (1× SSC = 0.15 M NaCL+0.015 M Sodium Citrate) at 42°C for 20 minutes, 2× SSC in 50% formamide at 50°C for 20 minutes, and 1× RNA wash buffer (RWB: 0.1M TRIS-HCL PH 7.5, 0.4M NaCL, 0.05M EDTA PH 8.6) at 37°C with ribonuclease A (25 µg/ml) and T1 (25 U/ml) for 30 minutes. After washing in RWB buffer, 2× SSC, and 0.1× SSC at 37°C for 15 minutes each, sections were transferred to 1× Tris-buffered saline (TBS; Boston BioProducts) containing 0.05% Tween-20 (TBS-Tw). Tissues were blocked in TBS-Tw containing 2% donkey serum for 1 hour at room temperature, then incubated with goat anti-digoxigenin-DyLight-594 (Vector Labs; 1∶1,1000), mouse anti-CD68 (1∶400; clone KP1, Dako), mouse anti-CD163 (1∶400; clone 10D6; Novocastra/Leica), and rabbit monoclonal anti-CD3 (clone SP7; Labvision; 1∶200) in TBS-Tw containing 2% donkey serum overnight at 4°C. Slides were washed in TBS-Tw, incubated with donkey anti-goat–Alexa594, donkey anti-mouse–Alexa488, and donkey anti-rabbit–Alexa647 (all Invitrogen; 1∶400) for 1 hour at room temperature in the dark and washed in TBS-Tw. Slides were incubated with 0.1% Sudan Black B in 70% ethanol (Cat. No. 4410; ENG Scientific) and 1× TBS for 30 minutes at room temperature to quench autofluorescence then incubated with 300 nM DAPI for 10 min. Slides were washed, mounted with Prolong® Gold (Invitrogen) and imaged on an Olympus FV10i confocal microscope using a 60× phase contrast oil-immersion objective (NA 1.35) imaging using sequential mode to separately capture the fluorescence from the different fluorochromes at an image resolution of 1024×1024 pixels.
Volumetric confocal images (Z-stack) were taken from lymph node and jejumal tissues following SIV F-ISH using the Nyquist sampling method using an Olympus FV10i confocal microscope with a 60× phase contrast oil-immersion objective (NA 1.35) with imaging using sequential mode to separately capture the fluorescence from the different fluorochromes at an image resolution of 1024×1024 pixels. Laser settings (gains and power) were set to ensure no pixel saturation in the SIV vRNA (Alexa594) channel and kept constant for all sample collections for each animal imaged. Images were opened in the Olympus FV10-ASW software (v3.1) and well-defined cellular regions were manually drawn around SIV vRNA+ macrophages and T cells from the compressed Z-stack image. The integrated intensity within these defined cell regions was measured in each Z-plane. The integrated intensity (or intensity integration) is the sum of the pixel intensities for a channel of interest within a specified object, and thus, the sum of the integrated intensities from each Z-stack is the total of the pixel intensities within each object volume. Thus, the “relative abundance” of SIV vRNA within the volume of each cell of interest was determined by summing the integrated intensities from each Z-plane.
Soluble CD163 (sCD163) plasma levels were quantified by ELISA according to manufacturer's protocol (Trillium Diagnostics) using a 1∶50 dilution of plasma samples.
Repeated-measures analyses for each outcome (CD4+ T cells; CD4+Ki-67+ T cells; CD14+CD16+; CD14+CD16+Ki-67+; and viral load) were performed with a means model with SAS Proc Mixed (version 9) providing separate estimates of the means by weeks post-depletion and infection between groups. A compound-symmetric variance-covariance form in repeated measurements was assumed for each outcome and robust estimates of the standard errors of parameters were used to perform statistical tests and construct 95% confidence intervals [40]. T-test or Mann Whitney test were used to compare the differences between the model-based treatment means (least-squares means) at each time point within the framework of the mixed effects linear model. Statistical tests were 2-sided. Pearson product-moment correlation coefficients were utilized to estimate linear associations for normally distributed data and Spearman rank correlation coefficients were used for skewed and other non-normal distributions. A P value ≤0.05 was considered statistically significant for the correlation analyses. The mean ± SEM were used as descriptive statistics for each continuous outcome. The rate of decay of virus was estimated by taking the linear slope of the natural log-transformed data of viral load from day 0 to 7 of treatment. The half-life is calculated as ln(2)/decay rate.
|
10.1371/journal.ppat.1006849 | The structure of FIV reverse transcriptase and its implications for non-nucleoside inhibitor resistance | Reverse transcriptase (RT) is the target for the majority of anti-HIV-1 drugs. As with all anti-AIDS treatments, continued success of RT inhibitors is persistently disrupted by the occurrence of resistance mutations. To explore latent resistance mechanisms potentially accessible to therapeutically challenged HIV-1 viruses, we examined RT from the related feline immunodeficiency virus (FIV). FIV closely parallels HIV-1 in its replication and pathogenicity, however, is resistant to all non-nucleoside inhibitors (NNRTI). The intrinsic resistance of FIV RT is particularly interesting since FIV harbors the Y181 and Y188 sensitivity residues absent in both HIV-2 and SIV. Unlike RT from HIV-2 or SIV, previous efforts have failed to make FIV RT susceptible to NNRTIs concluding that the structure or flexibility of the feline enzyme must be profoundly different. We report the first crystal structure of FIV RT and, being the first structure of an RT from a non-primate lentivirus, enrich the structural and species repertoires available for RT. The structure demonstrates that while the NNRTI binding pocket is conserved, minor subtleties at the entryway can render the FIV RT pocket more restricted and unfavorable for effective NNRTI binding. Measuring NNRTI binding affinity to FIV RT shows that the “closed” pocket configuration inhibits NNRTI binding. Mutating the loop residues rimming the entryway of FIV RT pocket allows for NNRTI binding, however, it does not confer sensitivity to these inhibitors. This reveals a further layer of resistance caused by inherent FIV RT variances that could have enhanced the dissociation of bound inhibitors, or, perhaps, modulated protein plasticity to overcome inhibitory effects of bound NNRTIs. The more “closed” conformation of FIV RT pocket can provide a template for the development of innovative drugs that could unlock the constrained pocket, and the resilient mutant version of the enzyme can offer a fresh model for the study of NNRTI-resistance mechanisms overlooked in HIV-1.
| The majority of anti-AIDS drugs target the reverse transcriptase (RT) enzyme of the HIV-1 virus. RT catalyzes the central step in the virus replication cycle converting the viral RNA genome into DNA for subsequent integration into the host genome. As with all anti-AIDS treatments, continued success of RT inhibitors is persistently disrupted by the occurrence of resistance mutations. To explore latent resistance mechanisms potentially accessible to therapeutically challenged HIV-1 viruses, we examined RT from the related feline immunodeficiency virus (FIV). FIV closely parallels HIV-1 in its replication and pathogenicity however is resistant to all non-nucleoside inhibitors of HIV-1 RT. We resolved the crystal structure of FIV RT, and using mutational and biochemical analyses, we show that specific differences in the FIV RT structure inhibit the binding of non-nucleoside inhibitors. We further show that mutating the protein to facilitate binding of the inhibitors does not confer sensitivity to these inhibitors, suggesting that other variances inherent in FIV RT modulate a second layer of resistance. These insights can help in the development of novel drugs against evolving HIV-1 RT resistance.
| Reverse transcriptase (RT) is the most common target for anti-AIDS drugs being the enzyme that catalyzes the central step in the HIV-1 replication cycle converting the viral RNA genome into DNA for subsequent integration into the host genome [1]. While the relative contribution is still undetermined, errors made by the RT enzyme provide one source of genetic variances emerging in the replicating viral genomes and facilitating the development of resistance to all anti-AIDS drugs [1]. RT inhibitors are mainly nucleoside/nucleotide analogues (NRTI), which target the catalytic site acting as competitive chain terminators in the enzymatic reaction, or non-nucleoside inhibitors (NNRTI) targeting a hydrophobic pocket key in allosteric regulation of RT structural rearrangements [1]. RT is a heterodimeric protein of two subunits, p51 and p66, encoded by the p66 template and, therefore, identical in sequence except for lacking the C-terminal RNase-H domain in p51 as a result of proteolytic processing. The structure of p51 is rigid and provides structural support to the more flexible p66 subunit that undergoes functionally crucial conformational rearrangements. The unliganded p66 predominantly folds into a “closed” conformation of a “right-hand” shape with the “thumb” crumpled down on the “fingers” (Fig 1). Upon nucleic acid binding, the thumb lifts up and fingers fold down to hold an incoming nucleotide for a productive reaction. Within the “palm” subdomain, and adjacent to the flexible thumb, resides a hydrophobic non-nucleoside binding pocket (NNBP) (S5 Fig). By targeting this pocket, NNRTIs restrict the structural flexibility of RT and abolish the DNA polymerization activity of the enzyme [1–3]. Although inhibition mechanisms have yet to be specifically defined, NNRTIs have been suggested to act by restricting the mobility of the thumb, distorting the catalytic triad, repositioning the primer grip and loosening the thumb and fingers clamp (reviewed in [4]).
As with all anti-AIDS treatments, resistance mutations persistently disrupt the continued success of RT inhibitors. While the mechanistic bases are not yet clearly defined, the prevailing view establishes that NNRTI-resistance mutations mainly alter the association and dissociation kinetics of NNRTIs (reviewed in [4]). Resistance mutation K103N has been suggested to restrict the conformational plasticity of NNBP [5], by forming additional hydrogen bonds that stabilize Y188 into a “closed” form of the pocket [6–8], reducing the rate of inhibitor binding [6, 9]. A recent mutational study, however, has shown that while the additional hydrogen bonds may have contributed to resistance (~ 2 fold enhancement), hydrophobic and electrostatic interactions contributed by the K103N played the fundamental role in impacting NNRTI binding kinetics [10]. Similarly, E138K has been shown to enhance the dissociation rate of NNRTIs [11]. Mutations such as E138K, K101E and V179D can alter the chemical environment around the NNBP and its entryway and, therefore, can impact NNRTI association and dissociation rates [4, 5]. E138K and K101E have also been proposed to modulate the conformational plasticity by disrupting the formation of an inter-subdomain salt-bridge, proposed to enhance the inhibitory effect of NNRTIs [5, 9]. Nevertheless, earlier kinetic studies highlight a controversy with this prevailing “binding kinetics” dogma by showing that even at saturating concentrations of NNRTI, a slow but significant polymerization activity continues by the fully inhibited enzyme and not due to the uninhibited enzyme recovered by the slow release of inhibitor [12, 13]. This challenge has recently been emphasized by a study showing that the K103N resistance mutation does not inhibit NNRTI binding to RT, and that NNRTI binding does not prevent the formation of a productive reverse transcription complex [9]. Therefore, we may not completely rule out a potential role of resistance mutations in modulating the conformational plasticity of RT subdomains, enabling the formation of a productive reverse transcription complex of an NNRTI-bound RT.
Current NNRTIs are HIV-1 specific. RT enzymes from HIV-2, simian (SIV) and feline (FIV) immunodeficiency viruses are intrinsically resistant to HIV-1 NNRTIs. The intrinsic resistance of FIV RT (fRT) is particularly interesting since FIV harbors the Y181 and Y188 NNBP sensitivity residues absent in both HIV-2 and SIV (S1 Fig) [14]. Unlike HIV-2 RT, or SIV RT (V181Y/L188Y) [15], where a simple I181Y/L188Y mutation bestowed NNRTI susceptibility [16], previous efforts have failed to make the fRT susceptible to NNRTIs, attributing resistance to distinct structural features potentially crafting a more rigid enzyme [14, 17]. It has been speculated that this rigidity could make the creation of the NNRTI-pocket, and consequent NNRTI penetration or binding, energetically unfavorable, or, alternatively, could adjust subdomain plasticity to mitigate inhibitory impacts of bound NNRTIs [14].
FIV RT inherently combines numerous substitutions frequently observed in HIV-1 resistance to NNRTIs including (HIV-1 numbering used; FIV variances in brackets) M41L, V75M(I), K101E/Q(Q), V118I, I132L, E138K/A(A), Q145M/L(V), V179D, F227L/C(Y), K238T, L283I(M), N348I, I393L(T), T/A376S(R), and T369I/V(L) (S1 Fig) (reviewed in [2, 4, 18], and Stanford Database https://hivdb.stanford.edu/). While the exact role and significance of each of these variances, separately and combined, remains to be determined, fRT presents a pertinent model for probing the NNRTI-resistance mechanisms. The FIV model is the only non-primate model of a lentivirus that induces AIDS-like syndrome in its natural host [19], and has greatly advanced our understanding of fundamental questions in retrovirology [19–21], including drug-resistance in HIV-1 RT and the identification of novel RT inhibitors [22]. The remarkable NNRTI-resistance of FIV RT has stimulated earlier studies exploiting the feline enzyme in probing resistance determinants of RT [14, 17]. The decreased catalytic activity of FIV/HIV-1 chimeric RT of these studies has underscored the significance of the delicately tuned p51/p66 inter-subunit interactions, and pinpointed NNRTI-sensitivity to the p66 subunit [14, 17]. Interestingly, whereas HIV-1 RT has been made entirely insensitive to NNRTIs by exchanging a fragment containing residues 97–205 of FIV RT, the reverse swapping did not bestow NNRTI-susceptibility to FIV RT [14], suggesting a more complex resilience. While the structure and function of FIV proteins are highly conserved with their HIV-1 analogues [20, 21], they greatly diverge in their amino acid sequences. Therefore, exploring how orthologous proteins coevolved in their natural environment can highlight crucially conserved patterns for drug and vaccine targeting, and more importantly, can uncover conceivably latent escape patterns accessible to challenged HIV-1. Here, we report the first crystal structure of RT from a non-primate lentivirus, the FIV, and provide insights into its complex mechanism of NNRTI resistance. The structure reveals a “closed” pocket configuration restricting NNRTI binding potentially by the distinct rigidity and electrostatic features of the pocket. We show that FIV RT is unable to bind NNRTIs and whilst mutating the loop at the NNBP-entryway confers binding ability, it does not provide sensitivity to the inhibitors. Therefore, the feline enzyme appears to harbor a more complex combination of variances that specifically constrain the NNBP pocket attenuating NNRTI binding, and further variances that potentially could enhance the dissociation of bound inhibitors, or, perhaps, modulate protein plasticity to overcome inhibitory effects of bound NNRTIs. Together, the more “closed” conformation of FIV RT pocket providing a unique example of impeded NNRTI-binding, and the resilient mutant version of FIV RT, which offers a model for NNRTI-resistance based on distinct variances inherent in FIV, make the FIV RT an invaluable model for the study of HIV-1 RT resistance and drug development.
To elucidate the mechanisms employed by fRT and promote our general understanding of NNRTI-resistance, we determined the 2.94 Å crystal structure of fRT (Table 1). With 48% identity and 67% similarity [23] (S1 Fig), the fold of feline RT is conserved with the HIV-1 RT. The p51/p66 heterodimeric enzyme embraces the famous “right-hand” shape of p66 [1, 5, 24], which superimposes on the “closed” form of unliganded HIV-1 RT (~ 2 Å RMSD) (Fig 1).
The NNBP of unliganded RT is rimmed at the p51/p66 interface (residues 100–105 of p66 and E138-loop of p51) and largely filled by the primer grip (β9-β10) and the catalytic loop (β7-β8) [1, 25]. The pocket opens upon NNRTI binding, which causes the side chains of Y181 and Y188 to flip upwards moving the primer-grip (β9-β10) away from the catalytic β7-β8 loop [1, 4, 5, 24, 26] (Fig 2A). The fRT NNBP structure is conserved with HIV-1 with variances around the NNBP of fRT making no global change to the pocket size or shape. Both FIV and HIV-1 RT NNBPs are of comparable volumes (~635 Å3) and superimposable NNRTIs in a model of open fRT NNBP (Fig 2B).
Although the NNBP of FIV and HIV-1 RT are conserved, several neighboring variances could modulate the chemical features and plasticity of the rims walling the NNBP and its entrance, perhaps making it unfavorable for NNRTIs to penetrate and bind. Because p51 is a proteolytic product of p66, variances or mutations will be identical in both subunits. However, we refer to variances specific to each subunit where structurally relevant (e.g. at the p51/p66 interface). Intriguingly, a group of FIV variances clusters at the rim of the NNBP entryway. fRT is altered to K101Q, D192N, D177Q and V179D in the p66 subunit, and E28N, T139G, E138A, I135K and S134R, in the p51. The prominent contributors to key physiochemical alterations at the entryway appear to be K101Q and V179D of p66, and E138Ap51 and I135Kp51 of p51 (Figs 2C and 3A).
Structural analysis suggests that variances at the fRT entryway could render a rigid pocket with a more of a “closed” conformation that could inhibit NNRTI binding. Particularly, Y181, which must flip upwards to open the NNBP, appears to be trapped by interactions with the conserved N136p51 and potentially with I135Kp51. In HIV-1 RT, the negatively charged and bulky side chain of E138p51 may shield Y181 against these interactions (Figs 2C and 3A). In an unliganded HIV-1 RT structure (PDB code: 1HMV) E138p51 is positioned between Y181 and N136p51 (spaced at 6.7 Å), preventing their interaction. In another HIV-1 RT structure (PDB code: 1DLO) E138p51 is rotated away from Y181, allowing a 3.4 Å interaction between Y181 and N136p51, 0.5 Å farther, and consequently weaker, than the interaction in fRT (2.9 Å) (Fig 3A and 3B). Moreover, I135Kp51 stabilizes the p51/p66 interface with potential interactions (< 4 Å) to carbonyl oxygen of three p66 residues (98, 382 and 383). The bulkier side chain of S134Rp51 may further contribute to a tighter entryway by filling more space at the p51/p66 interface and interacting (3.6 Å) with the carbonyl oxygen of I180 (Fig 2). fRT also forms three additional bonds around the entryway: one salt bridge (K173R/T39Ep51) and two hydrogen bonds (K166N/K49Rp51, and backbone amine of Q182 with carbonyl oxygen of E138Ap51) (Fig 3B). Consequently, the pocket entrance of FIV p66 is constricted by ~ 2 Å as compared to that of HIV-1 (Fig 3C). FIV variances, especially at the p51 loop (S134R, I135K, E138A and T139G), have also altered the electrostatic properties around the entryway from negative to positive charges, which may interfere with initial drug channeling into the entrance and subsequent penetration and binding (S2 Fig).
Additionally, a cluster of substitutions in fRT creates extra interactions that could restrain the flexibility of the primer grip (β9-β10) rimming the NNBP. Particularly, Gln of H221Q in fRT forms an extra interaction to backbone amine of L228T. Also, OH of F227Y interacts with backbone amines of V106 or K223E (Fig 2A). These interactions may further stabilize the fRT primer-grip, making its displacement and Y188/W229 dislocation out of the NNBP energetically unfavorable. Collectively, FIV variances at the p51/p66 interface appear to have promoted a more rigid and “closed” pocket conformation impeding NNRTI penetration and binding. Whether the “locked” configuration is a consequence of the extra interactions making the pocket more rigid, or a consequence of the electrostatic features at the entryway, needs to be validated by future studies.
The closed architecture of the NNBP entryway readily indicates that NNRTI binding to fRT would not be favorable. Previous studies documented the failure of NNRTIs to inhibit fRT activity but did not establish if this is a result of fRT inability to bind NNRTIs [14, 17]. In order to clarify this point, we measured the binding affinities of fRT to Delavirdine (DLV), Efavirenz (EFZ) and Rilpivirine (RPV). DLV is a first generation NNRTI inhibitor, susceptible to resistance, whereas EFZ and RPV are potent second generation inhibitors [1]. We found that DLV and EFZ inhibitors fail to bind to fRT (Fig 4A), supporting our model of a restricted NNRTI-pocket in the feline enzyme. Surprisingly, the more flexible RPV inhibitor, which has not previously been assessed with feline enzyme, was able to bind to fRT with comparable affinity to HIV-1 RT (Fig 4B), providing a promising molecule that could overcome the physiochemical barriers imposed by fRT variances at the pocket entryway.
To further validate the restricted-pocket configuration, we mutated the fRT p51-loop to match its HIV-1 RT equivalent (R133S, K134I, A137E, G138T and R141I in FIV p51) and found that the mutated fRT was able to bind EFZ and RVP with comparable affinities to HIV-1 RT (Fig 4B & 4C) and binds DLV with twice the affinity as compared to HIV-1 RT (Fig 4D). RPV bound to the mutant with twice the affinity as compared to fRT (Fig 4B). These results support our proposed “closed” pocket configuration of the feline enzyme, which was opened by mutating residues gating the pocket entryway.
Noteworthy, the fRT mutant had reduced thermal stability (~ 2.5°C), presumably as a result of the interactions lost through this mutation (S3 Fig).
The structure of the intrinsically resistant fRT provides a novel model of a constrained NNRTI pocket, which when combined with the numerous other NNRTI-resistance mutations inherent in fRT could result in a resilient enzyme. Although mutating the p51-loop opened the NNBP for drug binding, this was not sufficient to render fRT sensitive to these drugs. While potently inhibiting HIV-1 RT, none of the three inhibitors EFZ, DLV or RPV reduced the activity of the mutated fRT: the mutant retained the same activity as wild type fRT in both the presence and absence of these NNRTIs (Fig 4E, 4F and S4 Fig). Although RPV was able to bind the wild type fRT enzyme (Fig 4B), it was unable to inhibit its polymerization activity (Fig 4E), further supporting our observation that the fRT mutant, rendered susceptible to NNRTI binding, continued to resist inhibition. Whereas both K101Q and V179D variances inherent in fRT, similar to HIV-1 NNRTI-resistance mutations (see introduction), could modulate the chemical environment gating the pocket entryway (Fig 2B) enhancing inhibitor dissociation, a Q101K/D179V mutation of fRT was not sufficient to render it susceptible to NNRTIs [14], and therefore may not alone, explain fRT resistance to RPV that we observed (Fig 4F). A comprehensive mutational analysis coupled with detailed kinetic study would be required to establish the mechanistic basis underlying the continue resistance of the fRT, which binds RPV, and the fRT mutant, which binds all three NNRTIs. Whereas the pocket-opening mutation promoted inhibitor binding, analysis of dissociation kinetics is required to establish the role of the many other fRT variances in enhancing NNRTI dissociation or, perhaps, accentuating an overlooked role of these variances in facilitating a catalytically competent form of an inhibitor-bound enzyme, which has previously been documented [9, 12, 13]. The fRT and its p51-mutant offer pertinent models for such kinetic and mechanistic studies that could advance our understanding of the mechanisms underlying NNRTI resistance of lentiviral RT.
Variances distinct to fRT clustering around the nucleotide-binding site may also add to the overall resilience of the feline enzyme. A cross-species study investigating FIV virus evolution revealed that the majority of positive-selection mutations mainly cluster in the nucleotide-binding pocket and sites contacting nucleic acid substrates [27]. Modeling fRT bound to a nucleotide shows that, additional to more common variances (I47V, K73M, V75I, V111I and L214F), FIV contains unique substitutions, e.g. L12M, M41L, D67-lacking, F124Y, Y146W, V148S, K154I, G155L and F160Y, and three potential salt-bridges to form by T215E (to K46), K219D (to K70) and K223E (to I195K and R199K) (S1 Fig and S5 Fig). For example, V111I mutation in HIV-2 has been proposed to constrain the mobility of a flexible loop resulting in a productive increase in nucleotide pocket accessibility [28]. Therefore, while the exact role of these variances remain to be determined, their contribution to NNRTI-resistance should not be ignored and is reminiscent to NNRTI-resistance mutations of HIV-1.
Given its importance for productive virus infectivity, it is not surprising to find that RT accumulates the majority of positive-selection mutations during a cross-species study investigating virus evolution [27]. NNBP residues are not essentially conserved, therefore, HIV-1 has a rather low genetic barrier for developing NNRTI-resistance mutations [1]. The 2016 Los Alamos database (4416 entries) (https://www.hiv.lanl.gov/) shows that distinct substitutions of fRT p51 can occur in HIV-1 p51: K101Q (0.32%, K101R: 0.45%), I135K (0.38%, I135R: 1.54%), E138A (1.88%), and S134R (3 occurrences combined with E138A or S134N). Of the 14 records of K101Q, one was combined with E138A and 4 with I135R. There was yet no record of I135K/R with E138A. The resistant mutant version of FIV RT, combining these mutations together, may now provide a model to predict novel resistance mechanisms that a challenged HIV-1 may exploit in escaping inhibition of bound NNRTIs. The resilient FIV mutant RT offers a model for NNRTI-resistance based on distinct variances inherent in FIV but overlooked in HIV-1, HIV-2 and SIV. These intrinsic species-specific differences, which apparently hinge on variations as subtle as few or single residue mutations, highlight the importance of cross-species analysis for discovering potential unknown alternative routes awaiting HIV-1 exploitation. However, while NNBP residues are not crucially conserved and recombinant RT mutants were found enzymatically active, future studies need to evaluate the replication competence of FIV and HIV-1 viruses containing the p51-loop mutations.
Prominently, the wild type FIV RT offers a unique model to a more constrained pocket than has previously been observed in other lentiviruses and may, therefore, pave the way for the development of novel NNRTIs that need to disrupt such a locked pocket that may emerge in HIV-1. Whereas RPV appears to be one such inhibitor, further studies need to investigate whether the observed RPV binding to wild type fRT (Fig 4B) is specific to NNBP and what makes the fRT enzyme RPV resistant (Fig 4F). This study opens new horizons and opportunities to elaborate on minimal p51 variances required to block NNBP entryway, as well as detailed binding kinetics of present and novel NNRTIs to further our understanding of NNRTI-resistance mechanisms.
The two subunits of FIV RT (Petaluma) were subcloned into separate plasmids to facilitate site-specific mutagenesis. The p66 subunit (residues 1–554) was cloned into pET22b (amp resistance, Novagen) and p51 subunit (residues 1–430) in pET28b (kan resistance, Novagen) with a thrombin-cleavable His-tag at the N-terminus. FIV RT mutant (FIVm) was generated by introducing the specific mutations into the p51 subunit using QuikChange Kit (Agilent Technologies Genomics). HIV-1 RT construct, containing p66 and protease coding regions in pT5M plasmid, was a generous gift of Prof. Stephen H. Hughes (National Cancer Institute- Frederick, MD, USA).
FIV RT constructs were expressed in ArcticExpress cells (Agilent Technologies Genomics) with 0.5mM IPTG induction overnight in 16°C. HIV-1 RT was expressed in BL21(DE3) with 1mM IPTG induction for 4 hours in 37°C.
RT proteins from FIV and HIV-1 were purified using HisPur Cobalt Resin (Thermo Fisher Scientific) in a buffer containing 500 mM NaCl, 20mM potassium phosphate pH 7.5, 5 mM β-mercaptoethanol and dialyzed into a final buffer containing 500mM NaCl, 20mM Hepes 6.0, 2 mM DTT. Thrombin (2 units/mg, Novagen) was used to cleave the His-tag of FIV RT (4°C, overnight), and was subsequently removed using benzamidine sepharose resin (GE Healthcare Life Sciences). FIV RT proteins were further purified using HiTrap SP HP cation exchange (GE Healthcare Life Sciences) eluting at 350 mM NaCl.
For crystallization, FIV RT was further purified using Superdex-200 size exclusion column (GE Healthcare Life Sciences) equilibrated with 300 mM NaCl, 20mM Hepes pH 7.4 and 2mM DTT, and concentrated to 10 mg/ml. Protein crystals were grown using the hanging-drop vapor-diffusion method at 4°C. Initial crystals were obtained in C9 condition of crystals screen (Hampton Research) containing 4.0 M sodium formate. Diffraction quality crystals were obtained in 3.8 M sodium formate supplemented with 0.2 M ammonium formate. Cryoprotectant solutions contained an additional 20% glycerol.
Diffraction data were collected at beamline I-24 of Diamond Light Source Ltd (Harwell Research and Innovation Campus, UK), and processed with XDSAPP [29]. The structure was solved by molecular replacement using BALBES [30] and 3DOL [31] as a search model. The resulting model was fitted into the electron densities using COOT [32] and refined using REFMAC5 [33] in CCP4 suite [34]. Table 1 summarizes data collection and refinement statistics.
DNA-dependent DNA polymerization activity of RT enzymes was assessed using a DNA primer/template prepared by annealing of 70-mer ssDNA template with a complementary 25-mer ssDNA primer labeled with 6-FAM at 5’-end (Integrated DNA Technologies).
Reaction mixtures contained 100 mM Tris-HCl pH 7.5, 75 mM KCl, 0.5 mM dNTP mixture, 60 mM MgCl2, 5 mM DTT and 0.15 μM RT, 0.5 μM DNA substrate. Reactions were incubated for 30 minutes at 37°C and stopped using urea sample loading buffer and heating at 95°C for 10 minutes. Reaction products were resolved using denaturing urea polyacrylamide gel electrophoresis (S5B Fig). Percentage RT activities represent percentile of band-intensities of polymerized product divided by those of template/primer substrate.
RT proteins were labeled with Monolith NT protein labeling kit BLUE-NHS (NanoTemper Technologies) and eluted in microscale thermophoresis (MST) buffer (50 mm Tris-HCl, pH 7.0, 300 mM NaCl and 0.05% Tween-20). Labeled RT proteins were incubated with serial dilutions of Efavirenz or Delaviridine, and MST measurements were carried out using the Monolith NT.115 in standard capillaries with 40% LED and 20% MST power.
RT proteins were diluted (into 20 mM Hepes pH 7.4 and 300 mM NaCl) to reach a final concentration of 0.15 mg/ml and subsequently filled into standard treated capillaries. Label-free differential scanning fluorimetry (nanoDSF) measurements were carried out using Prometheus NT.48 (NanoTemper Technologies) and thermal unfolding was monitored with temperature changing from 15°C to 95°C, with the rate of 1°C/min.
Protein sequences were aligned using T-Coffee [35] and annotated using ESPript [36]. PyMOL Molecular Graphics System (Schrödinger, LLC) was used for preparing structural figures and for structural superposition. Comparative pocket volumes were analyzed using the CASTp server [37] using HIV-1 RT bound to NVP without nucleic acid (PDB code: 5HBM) or bound to NVP with nucleic acid (PDB code: 4PUO). Both HIV-1 RT structures show comparable volumes validating the use of this measure. Next, the fRT was modeled into an open conformation using homology modeling and employing these two HIV-1 RT structures as templates. Pocket volumes of fRT and HIV-1 RT were, therefore, compared using one specific NVP inhibitor, and the analysis is only comparative and is not made to indicate an exact volume of NNRTI pockets. Homology modeling was performed using the Swiss Model server [38]. Analyses of interacting interfaces were performed using PDBsum [39].
|
10.1371/journal.ppat.1001251 | Characteristics of the Earliest Cross-Neutralizing Antibody Response to HIV-1 | Recent cross-sectional analyses of HIV-1+ plasmas have indicated that broadly cross-reactive neutralizing antibody responses are developed by 10%–30% of HIV-1+ subjects. The timing of the initial development of such anti-viral responses is unknown. It is also unknown whether the emergence of these responses coincides with the appearance of antibody specificities to a single or multiple regions of the viral envelope glycoprotein (Env). Here we analyzed the cross-neutralizing antibody responses in longitudinal plasmas collected soon after and up to seven years after HIV-1 infection. We find that anti-HIV-1 cross-neutralizing antibody responses first become evident on average at 2.5 years and, in rare cases, as early as 1 year following infection. If cross-neutralizing antibody responses do not develop during the first 2–3 years of infection, they most likely will not do so subsequently. Our results indicate a potential link between the development of cross-neutralizing antibody responses and specific activation markers on T cells, and with plasma viremia levels. The earliest cross-neutralizing antibody response targets a limited number of Env regions, primarily the CD4-binding site and epitopes that are not present on monomeric Env, but on the virion-associated trimeric Env form. In contrast, the neutralizing activities of plasmas from subjects that did not develop cross-neutralizing antibody responses target epitopes on monomeric gp120 other than the CD4-BS. Our study provides information that is not only relevant to better understanding the interaction of the human immune system with HIV but may guide the development of effective immunization protocols. Since antibodies to complex epitopes that are present on the virion-associated envelope spike appear to be key components of earliest cross-neutralizing activities of HIV-1+ plasmas, then emphasis should be made to elicit similar antibodies by vaccination.
| A fraction of those infected with HIV develop broadly neutralizing antibodies (bNAbs) capable of preventing cell-infection by diverse HIV isolates; the type of antibodies we wish to elicit by vaccination. Identifying factors associated with the natural development of bNabs, and defining the timing of their emergence and their epitope specificities, will assist the development of more effective immunogens and vaccination protocols. Here we performed a neutralization screen of plasma samples collected longitudinally from HIV-1-infected subjects and determined that on average, cross-neutralizing antibody responses emerge 2–3 years, but as early as one year, following infection. A significant portion of the earliest cross-neutralizing antibody response to HIV targets epitopes that are present on the virion-associated trimeric Env spike, but not the corresponding soluble monomeric versions of that viral protein. Our study highlights the importance of eliciting by vaccination antibodies with this type of complex epitope specificities.
| The initial antibody response to the HIV-1 viral envelope glycoprotein (Env) manifests itself within the first 2 weeks of infection and is non-neutralizing [1], [2]. Autologous neutralizing antibodies develop during the first months after infection [3], [4], [5] and recent studies indicated that approximately 10%–30% of chronically-infected HIV-1 subjects develop cross-reactive neutralizing antibody responses of significant breadth [6], [7], [8]. These latter responses are the ones an effective vaccine should elicit [9]. Several studies indicated that the breadth of plasma cross-neutralizing antibody responses is positively associated with plasma viral load [6], [7], [10], [11], [12], but very little is known about the time course of these responses. A recent study by van Gils et al, using samples collected at 2 and 4 years following infection, indicated that a greater number of infected subjects displayed cross-neutralizing activities at 4 than at 2 years [12]. However, the earliest timing of the development of such responses was not determined. Defining the timing of emergence of cross- neutralizing antibody responses following HIV-1 infection and identifying factors associated with their development, will advance our understanding of the complex interaction of HIV-1 with the immune system, will improve our understanding on how HIV-1 infection leads to immune dysfunction, and will also be useful to the development of immunization protocols that hopefully would elicit similar antibody responses.
The epitope specificities of the anti-HIV-1 cross-reactive neutralizing antibody responses in HIV-1+ plasmas collected during chronic infection are complex, with many specificities remaining undefined. Although there is general consensus that these neutralizing activities rarely target the transmembrane subunit gp41, but mostly the extracellular gp120 subunit [7], [13], [14], [15], [16], [17], there remains quite an uncertainty whether the overall cross-neutralizing activities of HIV-1+ plasmas are due to a single, a limited number of, or many different epitope specificities [7], [13], [14], [15], [16], [18], [19], [20], [21], [22]. The above studies were conducted with samples from chronically-infected subjects and very little, if anything, is known about the epitope specificities of the earliest cross-neutralizing antibody responses in HIV-1+ plasmas. Defining these epitope specificities would be informative for future immunogen design efforts.
Here we analyzed the cross-neutralizing antibody responses in longitudinal plasmas collected soon after and up to seven years after HIV-1 infection. We found that the subset of HIV-1-infected subjects that develop cross-neutralizing antibody responses do so on average within the first 2.5 years of infection, although in rare cases such responses became detectable as early as 1 year after infection. Epitope-mapping analyses indicated that the earliest cross-neutralizing antibody responses target primarily epitopes within and around the CD4-BS of gp120, or epitopes that are present on the virion-associated trimeric Env, but not on the corresponding monomeric gp120 or gp41 Env subunits. In contrast, the neutralizing activities of plasmas from subjects that did not develop cross-neutralizing antibody responses, target epitopes on monomeric gp120, other than the CD4-BS. These observations are indicative of the presence and long-term survival of B cells that recognize complex but conserved epitopes on the viral Env in those HIV-infected subjects that develop cross-neutralizing antibody responses.
To define the earliest period following HIV-1 infection when cross-neutralizing antibody responses appear in plasma we determined the neutralizing activities of plasmas collected within a few months and up to several years post HIV-1 infection from anti-retroviral naïve subjects infected with clade B viruses, against 20 heterologous clade A, B and C primary isolates (Figures 1 and 2). Plasma samples from two independent cohorts were examined. The samples from the Vanderbilt cohort (VC) were, for the most part, collected within the first year of infection (Figure 1). The breadth of cross-neutralizing activity (i.e., the percentage of viruses neutralized by any given plasma out of the total number of viruses the plasma was tested against) was minimal (less than 50%), in agreement with previous observations [2], [5]. In most cases, these ‘early’ plasmas efficiently neutralized the ‘easy-to-neutralize’ primary SF162.LS virus, but not other primary viruses examined here. In the few cases where neutralizing activity against viruses other than SF162 was observed, the potency of neutralization was for the most part very weak and the neutralizing activities targeted clade B viruses. In two cases (subjects VC20017 and VC20027) the ‘early’ plasmas also neutralized a few non-clade B viruses. Plasma VC20027 collected within the first year of infection neutralized 6/9 clade B, 3/6 clade C and 1/4 clade A viruses. Plasma sample collected from subject VC20017 during the first year of infection neutralized 4/9 clade B viruses, the clade A virus Q259d2.17, and the clade C viruses ZM214M and Du422.1. These observations indicate that cross-neutralizing antibody responses begin to emerge during the first year of HIV-1 infection, but that such responses are weak in potency and narrow in breadth; rarely targeting viruses from clades other than the one the patient is infected with.
In the case of the MGH Acute HIV Infection Cohort (AC), plasma samples were collected longitudinally within a few months after infection and up to approximately 7 years post infection, with an average follow-up of 3.31 years. Here, too, samples collected during the first year of infection did not display broad cross-neutralizing activities (Figure 2). In only one case (subject AC128), a plasma sample collected approximately 1.4 years after infection neutralized 65% of the heterologous viruses tested (7/10 clade B, 5/6 clade C and 1/4 clade A). A plasma sample collected a year later from the same subject neutralized 90% of the viruses tested with greater potency: an indication of a continuous evolution and increase in the breadth of the cross-neutralizing antibody responses during the first 2 years of infection in this subject. Overall, plasma samples from 7/17 subjects (41%) (AC049, AC053, AC071, AC089, AC128, AC131, and AC180) displayed cross-neutralizing activities against 50% of the isolates tested against at some point during the period of observation. Samples from 5/17 subjects (29%) (AC049, AC053, AC128, AC131, and AC180) displayed broad cross-neutralizing activities against at least 75% of the viruses tested at some point during the period of observation. This percentage is in agreement with numerous previous reports on the frequencies of broad cross-neutralizing activities in sera collected during chronic HIV-1 infection [6], [7], [8]. In these 5 cases, a gradual increase in the breadth of cross-neutralizing activities was recorded over time, even though plasmas collected longitudinally from individual subjects did not always neutralize the same isolates, nor with the same potency. Such changes in potency by samples collected over time from individual subjects could be due to changes in the number of epitopes recognized by the circulating antibodies (i.e., the relative proportions of NAbs with diverse epitope specificities change over time), and/or due to changes in the plasma concentrations of antibodies with epitope specificities that do not change over time. Collectively, the above results indicate that the mean time it took for the breadth of cross-neutralizing activities to reach 50% was 2.13 years and the mean time to reach 75% was 3.08 years. Although the percentage of subjects developing cross-neutralizing antibody responses increased during the first 3 years of infection, that percentage did not further increase in year 4 (Figure 3). Clearly, additional longitudinal analysis is required to determine whether cross-neutralizing antibody responses increase past that time.
In chronic HIV-1 infection the breadth of serum cross-neutralizing antibody activities positively correlates with the levels of plasma viremia [6], [7], [11], [12]. Here, we recorded a positive correlation (p = 0.0026; R = 0.3615) between the breadth of the earliest cross-neutralizing antibody responses in HIV+ plasmas and the levels of plasma viremia. Because of the association between plasma viremia and immune activation during early HIV infection [23], we examined potential associations between the development of serum cross-neutralizing activities and markers of immune activation and exhaustion (Figure 4). Specifically, we compared the percent of CD4+ and CD8+ T cells expressing Ki67, CD57, CD38, PD1, and HLADR in subjects that developed cross-neutralizing antibody responses and those who did not. The immune activation status of subjects who developed broad cross-neutralizing antibody responses (at least 75% breadth at some point during the period of observation) was determined at the earliest time point when cross-neutralizing antibody responses were evident: for AC049 at 2.62 year post-infection (ypi), for AC053 at 3.29 ypi, for AC128 at 1.41 ypi, for AC131 at 1.52 ypi, and for AC180 at 2.19 ypi. Similar time points of infection were used for those subjects who did not develop cross-neutralizing antibody responses (AC093, AC110, AC167, AC183, AC194, and AC212).
A trend towards higher percentages of CD8+ T cells expressing Ki67 (p:0.122), CD38 (p:0.0823), and PD1 (p: 0.0823) was recorded in subjects with breadth. It is likely that, because the number of subjects who developed cross-neutralizing antibody responses is small, these differences did not reach statistical significance. A similar trend towards higher expression of Ki67 (p:0.0823) and CD38 (p:0.0823) was recorded in the case of CD4+ T lymphocytes. A statistically significant difference (P:0.0173) was, however, recorded in the percent of CD4+ PD1+ T cells between those subjects that developed cross-neutralizing antibody responses and those who did not. In addition, we performed correlation analysis between the degree of breadth and the frequencies of T cells expressing the various activation markers. A statistically significant positive association was observed between breadth and the frequency of CD4+ PD1+ T cells (p: 0.0174, Pearson r: 0.6961), and CD4+ CD38+ T cells (p: 0.0306, Pearson r: 0.6494). Overall, these results link for the first time the state of immune activation (within approximately 2 years of infection) to the development of cross-reactive neutralizing antibody responses.
Taking advantage of the availability of longitudinal samples from the MGH Acute HIV Infection Cohort, we performed epitope-mapping studies to determine: (a) whether the initial cross-neutralizing antibody responses developed by subjects infected with different viruses were due to the emergence of antibodies that target one or multiple epitopes on heterologous Env, and (b) whether the initial epitope specificities of cross-reactive neutralizing antibody responses in HIV-1+ plasmas evolve over time.
It is now recognized that 10–30% of HIV-1 chronically infected subjects develop cross-neutralizing antibody responses of significant breadth [6], [7], [8], [12]. We and others have previously discussed that the duration of HIV infection is positively associated with the breadth of cross-neutralizing antibody responses [6], [7], [11], [12]. Here we show that such anti-viral responses become detectable in the blood of these subjects, on average, at 2.5 years after infection. In rare cases, cross-neutralizing antibodies appear as early as 1 year post-infection. A recent study indicated that the development of cross-neutralizing antibody responses does not delay the onset of AIDS [43], however, it is currently unknown whether an unusually early emergence of such responses will offer a long-term clinical benefit to the patient or not. The observation that the development of cross-neutralizing antibody responses is associated with higher levels of plasma viremia potentially indicates a more efficient viral escape from autologous anti-viral responses (neutralizing antibody responses and/or cellular-mediated anti-viral responses) in those subjects that eventually develop cross-neutralizing antibodies. Our data indicate that those subjects who develop cross-neutralizing antibodies have higher frequencies of CD4+ T expressing PD1. This observation is intriguing and potentially of high importance. A fraction of CD4+ T cells that express high levels of PD1 (termed follicular T helper cells, TFH) have a distinct gene expression profile from other effector T cells and develop independently of the classic TH1 or TH2 lineages [44], [45] They are not ‘exhausted’, they secrete IL-4 and IL-21 for extended periods of time, and they are crucial for the formation of germinal centers and the proliferation and survival of circulating plasma cells [46]. One possible reason for the development of broad neutralizing antibody responses in only a subset of HIV-1-infected subjects is that optimal interactions between the TFH and B cells are taking place in those subjects who develop such antibody responses, while the TFH-B cell interactions are limited by the smaller number of TFH cells in those subjects who do not develop such responses. Clearly, follow up studies are required to determine whether the CD4+PD1+ T cells found in the periphery actually behave like TFH cells and whether or not the early development of cross-neutralizing antibody activities impacts the rate of disease progression.
The subjects examined here were infected with clade B HIV-1 viruses and it is not known whether the timeline for the emergence of cross-neutralizing antibody responses in non-clade B HIV-1 infections is similar, or whether infection with a particular HIV-1 subtype elicits an earlier or a delayed development of cross-neutralizing antibody responses. In the ‘early’ cases examined here, 29% of subjects developed cross-neutralizing antibody responses. This percentage is in agreement with several previous studies conducted with sera collected during chronic infection [6], [7], [8]. Potentially our data, in combination with data on the frequency of broad cross-neutralizing antibody responses in sera collected during chronic HIV-1 infection [7], [8], [47], suggest that if cross-neutralizing antibody responses are not generated during the first 2–3 years of infection, they may not emerge later. However, further follow up of these subjects is required to address this important point. It is currently unknown whether the emergence of cross-neutralizing antibody responses, at that particular time period of HIV-1 infection, in only approximately a third of those infected, is the result of a stochastic event or due to genetic predisposition, and whether it is related to particular evolutionary pathways the virus follows in response to other types of anti-viral immune responses.
In agreement with publications with sera from chronic HIV-1 infection [13], [22], we found that the earliest cross-neutralizing antibody response targets only a few regions of Env. It appears, therefore, that a few Env regions are targeted early and late during HIV-1 infection by cross-neutralizing antibodies. The fine specificities of such antibodies within these Env regions may evolve over time.
A significant portion of, or the entire, ‘early’ cross-neutralizing antibody response was due to antibodies that target virion-associated Env, rather than epitopes present on monomeric gp120 or gp41. The fact that the antibodies that bind such complex epitopes were elicited in response to infection by viruses unrelated to the heterologous viruses used to assess the cross-neutralizing potentials of HIV-1+ plasmas is strongly suggestive that these epitopes are common and most likely present on the viruses circulating even in those subjects who do not develop such antibody responses.
Since such epitopes are conserved among diverse viruses, we assume that they are also present on the transmitted viruses. Then why is the appearance of antibodies that target these epitopes delayed by 2–3 years? Potentially, antibodies that recognize epitopes which are exclusively present on the virus but not the monomeric form of HIV-1 Env could very well be generated earlier following infection, but may specifically target the autologous virus. In fact, MAbs with such complex epitope specificities that display only autologous virus neutralizing activities, or activities only against SF162 and related viruses, have been isolated from SHIV-infected macaques and from a chronically HIV-1-infected human [48], [49]. As infection progresses and in response to a continuous viral evolution, the B cell response to such complex epitopes may also evolve, and this evolution may eventually lead to the generation of antibodies with broader cross-neutralizing activities [22], [36].
The fact that anti-CD4-BS antibodies contribute to the initial cross-neutralizing activities of diverse HIV-1+ plasmas is not surprising since the CD4-BS is one of the most conserved regions of the HIV-1 Env. Numerous studies have already reported the contribution of such antibodies in defining the cross-neutralizing activities of plasmas collected during chronic HIV-1 infection [7], [13], [14], [15], [18], [19]. However, anti-CD4-BS antibodies are present in plasmas with and without ‘breadth’, and several anti-CD4-BS MAbs displaying very narrow cross-neutralizing activities have been isolated from HIV-1-infected individuals [50], [51]. Presently, it is not known why only a subset of HIV-1-infected subjects generates anti-CD4-BS antibodies that are cross-neutralizing, while the majority of subjects generate anti-CD4-BS antibodies of narrow neutralizing breadth. The angle of recognition of the CD4-BS by anti-CD4-BS antibodies with narrow and broad neutralizing activities is different [52], which implies that the CD4-BS is recognized differently by the B cell receptors (BCRs) of subjects who develop cross-neutralizing anti-CD4-BS antibodies and the BCRs of subjects who develop anti-CD4-BS antibodies of narrow neutralizing activities.
It is important to note that a fraction of the cross-neutralizing activities in some subjects could be adsorbed on both gp120 and the D368R mutant. Such specificities may be similar to those reported by Scheid et al [16] and more recently by Pietzsch et al [53] that target the ‘core’ part of gp120. Overall therefore, our data indicate that the ‘earliest’ cross-neutralizing antibody response to HIV is primarily comprised of antibodies that target the CD4-BS, the core of gp120, and epitopes present on the trimeric Env. The positive association, however, between plasma viremia levels and the breadth of the earliest cross-neutralizing antibody responses suggests that HIV is able to escape the action of the antibodies that recognize conserved regions of Env. Viral escape from antibodies that preferentially bind the Env trimeric spike may involve changes in the V1V2 region of Env, since the epitopes of this type of antibodies include elements of the V1V2 Env region [22], [36]. In fact, the V1V2 region of Env undergoes extensive alterations (including increases in length and in glycosylation) early following infection [54], [55], [56]. These changes are associated with early escape from autologous neutralizing antibody responses. Our data suggest that such changes may also be involved in the escape from the early cross-neutralizing antibody responses.
Our results provide information that may guide the development of effective immunization protocols. Since antibodies to complex epitopes that are present on the virion-associated envelope spike appear to be key components of the earliest cross-neutralizing activities of HIV-1+ plasmas, then emphasis should be made to elicit similar antibodies by vaccination. As a first step, HIV envelope glycoproteins that readily display such complex epitopes must be identified and tested as immunogens. However, if the development of such cross-neutralizing antibodies is somehow linked to genetic factors, then the outcome of immunizations with such immunogens will largely depend on the population the immunogens are evaluated, since only those vaccinees with the appropriate genetic makeup will respond appropriately.
Patients from the Vanderbilt University and the Ragon Institute of Massachusetts General Hospital ‘acute / early’ HIV infection cohorts (also referred to ‘primary’ cohorts) were used in this study. The subjects selected for the present study were infected with clade B HIV-1, had no AIDS-defining illnesses, and were not on antiretroviral therapy at the time of sample collection. In the MGH Acute HIV Infection Cohort, ‘primary infection’ was defined by detectable HIV RNA in the presence of either (i) a negative p24 ELISA, or (ii) a positive ELISA but evolving WB, or (iii) documented negative HIV ELISA within past 6 months. The plasma samples from the ‘Vanderbilt Cohort’ were collected mostly during the first year of seroconversion. All early infection subjects in this cohort had a documented negative HIV antibody test within one year of their first positive western blot result. In the case of the ‘MGH Acute HIV Infection Cohort’ the date of infection was known and samples were collected longitudinally from a few months post infection to up to 7 years post infection. In total, 53 plasma samples (collected longitudinally up to 2.5 years post-infection) from 21 HIV+ subjects from the ‘Vanderbilt Cohort’ and 69 plasma samples from 17 HIV+ subjects from the ‘MGH Cohort’ were evaluated.
The Ragon Institute's and Vanderbilt University's Institutional review boards approved the study. Written informed consent was provided by all study participants and/or their legal guardians. The data were analyzed anonymously.
Plasma anti-HIV Env antibodies were adsorbed on beads coated with either recombinant SF162 gp120 or HxB2 gp41 (amino acids 541–682, Viral Therapeutics, Inc Ithaca NY) as previously described [7], [19]. The proteins were coupled to MyOne Dynabeads Tosylactivated (Invitrogen) following the manufacturer's instructions. Briefly, 40 mg of magnetic beads were reacted with 1 mg protein ligand overnight at 37°C with gentle rotation. After collecting the beads on a magnet, the supernatant was removed and the beads were incubated overnight at 37°C in PBS, 0.5% BSA, 0.05% Tween 20. The magnetic beads were washed twice with PBS, 0.1% BSA, 0.05% Tween 20, and stored at 4°C in the same buffer, with the addition of 0.02% Sodium Azide. Bead-coupled Env proteins were tested for antigenic integrity by flow cytometry using known MAbs b12, 447–52D, 2G12, IgG-CD4, and 4E10, followed by detection with goat-anti-human-IgG-FITC secondary antibody (data not shown). Mock adsorption/elution experiments using several anti-HIV Env MAbs at a concentration of 10 µg/ml in naïve plasma were performed as a positive control (data not shown). 500µl of plasma, diluted 1∶5 in DMEM/10%FBS, were incubated with 200µl Env protein-coupled beads at room temperature for 120 min with gentle rotation. The samples were placed on a magnet and the beads were isolated.
The antibodies bound to the bead-coupled Env proteins were eluted in a series of increasingly acidic solutions as previously described [19]. The beads from each serial adsorption were combined and incubated in 0.1M Glycine-HCl, pH 2.7 for 30 seconds with vortexing. The beads were collected by brief centrifugation and held in place by a magnet. The supernatant was removed and adjusted to pH 7.5 with 1M Tris (pH 9.0). The process was repeated with the beads in 0.1M Glycine-HCl, pH 2.3 and then again in pH 1.7. The final supernatants were buffer-exchanged in PBS and washed over a 30kD Amicon Ultra centrifugation concentrator (Millipore). Concentration of immunoglobulin was determined by absorbance at 280 nm (NanoDrop Spectrophotometer ND-1000, Thermo). The depleted plasmas and the antibodies that were eluted from gp120-coated beads were tested by ELISA for reactivity to gp120, and for neutralizing activity.
The neutralizing activities of plasmas were determined using the Tzm-bl-based neutralization assay [57]. Briefly, plasma dilutions (starting at 1∶20) were pre-incubated with single-round competent virions (pseudovirus) for 60 minutes at 37°C. The plasma / pseudovirus mixture was added to TZM-bl cells (3000 cells per well in a 96-well plate) for 72 hrs at 37°C. The supernatant was removed and 100µl of Steady-Glo Luciferase Assay Substrate (Promega) was added to each well. Plates were incubated for 15 minutes at room temperature and 75µl of the lysate was transferred to micro titer plates. The cell-associated luciferase activity for each well was determined on a Fluoroscan Luminometer (Thermo). Percent neutralization was calculated at each plasma dilution as the percent inhibition of viral entry by the plasma sample compared to the absence of plasma. For each plasma/virus combination tested, a neutralization curve (percent neutralization versus plasma dilution) was generated using GraphPad Prism version 4.03 for Windows (GraphPad Software, San Diego California, USA), and the plasma dilution at which 50% neutralization was recorded (IC50) was determined by transforming the data to a log10 scale with fitted sigmoidal dose-response curves.
Neutralization breadth of a plasma sample is defined as the percent (0%–100%) of the 20 isolates neutralized by that sample.
All plasmas were tested against single round competent virions expressing Envs from 10 Clade B, 6 Clade C and 4 Clade A primary viruses. The clade B SF162.LS (EU123924), JRFL (U63632) and YU2 (M93258) viruses were isolated during chronic HIV-1 infection and the remaining isolates were isolated during acute infection, with published accession numbers [58], [59], [60], [61]. All plasma samples were also screened for non-HIV-specific neutralization using the murine leukemia virus (MLV) pseudotyped into the HIV backbone. Neutralization activity was not detected against MLV at 1∶20 by any of the plasma samples (data not shown).
In certain cases, competition neutralization experiments were performed in the presence of the D368R gp120 or an MPER-derived peptide. Serially diluted MAbs or HIV+ plasmas were pre-incubated with D368R (25 µg/ml) or the MPER peptide (10 µg/ml) for 1 hour at 37°C and then the mixture was incubated with virus for another hour at 37°C, and subsequently with cells as described above. The fold decrease in log10 IC50 neutralization titers of each plasma tested against each virus in the presence of D368R or the MPER peptide was determined.
Logarithmic transformation was used for viral load, and nonparametric regression with two-tailed p-value analysis was used to determine correlations between the breadth of cross-neutralizing antibody responses in HIV-1+ plasmas and plasma viremia levels. Mann-Whitney Test and Pierson correlation and linear regression analysis were used to determine correlations between immune activation and breadth of neutralizing activities.
|
10.1371/journal.pgen.1007124 | An Arabidopsis Nucleoporin NUP85 modulates plant responses to ABA and salt stress | Several nucleoporins in the nuclear pore complex (NPC) have been reported to be involved in abiotic stress responses in plants. However, the molecular mechanism of how NPC regulates abiotic stress responses, especially the expression of stress responsive genes remains poorly understood. From a forward genetics screen using an abiotic stress-responsive luciferase reporter (RD29A-LUC) in the sickle-1 (sic-1) mutant background, we identified a suppressor caused by a mutation in NUCLEOPORIN 85 (NUP85), which exhibited reduced expression of RD29A-LUC in response to ABA and salt stress. Consistently, the ABA and salinity induced expression of several stress responsive genes such as RD29A, COR15A and COR47 was significantly compromised in nup85 mutants and other nucleoporin mutants such as nup160 and hos1. Subsequently, Immunoprecipitation and mass spectrometry analysis revealed that NUP85 is potentially associated with HOS1 and other nucleoporins within the nup107-160 complex, along with several mediator subunits. We further showed that there is a direct physical interaction between MED18 and NUP85. Similar to NUP85 mutations, MED18 mutation was also found to attenuate expression of stress responsive genes. Taken together, we not only revealed the involvement of NUP85 and other nucleoporins in regulating ABA and salt stress responses, but also uncovered a potential relation between NPC and mediator complex in modulating the gene expression in plants.
| Nuclear pore complex (NPC) mediates the traffic between nucleus and cytoplasm. This work identified NUCLEOPORIN 85 (NUP85) as an important factor for the expression of stress-responsive luciferase reporter gene RD29A-LUC in response to ABA and salt stress from a forward genetics screen. Mutation in NUP85 and other NPC components such as NUP160 and HOS1 resulted in decreased expression of several stress responsive genes such as RD29A, COR15A and COR47. Proteomics data uncovered a list of putative NUP85 associated proteins. Furthermore, NUP85 was demonstrated to interact with MED18, a master transcriptional regulator, to control the expression of stress responsive genes. The study has added a new layer of knowledge about the diverse functions of NPC in abiotic stress responses.
| Nucleocytoplasmic transport plays vital roles in eukaryotic systems [1,2]. The exchange of macromolecules such as RNAs and proteins is predominantly regulated by highly conserved nuclear pore complexes (NPCs), which consist of multi-nucleoporins (Nups) arranged in distinct sub-complexes [3,4]. Although the composition and functions of NPCs have been extensively characterized in yeast and vertebrate systems, our knowledge on the functions of plant NPCs remains poor.
Recent genetic screens have identified the involvement of several Nups from Arabidopsis and Lotus japonicus (lotus) in a variety of developmental processes, hormone signaling pathways and environment adaptation [3,5,6]. For instance, NUP160 (SUPPRESSOR OF AUXIN RESISTANCE1) and NUP96 (SUPPRESSOR OF AUXIN RESISTANCE3) have been indicated to play a role in auxin signaling as they were identified as suppressors of auxin-resistant 1(axr1) mutants [7]. NUP160 was also shown to be involved in cold stress responses since the knock-out of NUP160 impaired cold-induced expression the CBF3-LUC reporter gene and several endogenous genes; and thus resulted in hypersensitivity to chilling and freezing stresses [8]. Another important NPC component [9], HIGH EXPRESSION OF OSMOTICALLY RESPONSIVE GENES1 (HOS1), which encodes a RING finger E3 ubiquitin ligase, is well-known as a negative regulator in the cold signaling [10]. Cold-responsive genes such as RESPONSIVE TO DESICCATION 29A (RD29A), COLD-REGULATED 47 (COR47), COLD-REGULATED 15A (COR15A) and KIN1 were reported to be induced to higher levels in hos1 mutants than in wild type plants [11,12]. HOS1 was further shown to attenuate cold signaling by ubiquitination and degradation of INDUCER OF CBF EXPRESSION 1 (ICE1), which encodes an important positive transcription factor critical for the cold-induction of C-REPEAT BINDING FACTORs (CBFs) [11–13]. Moreover, the mutation of NUP160, HOS1 and NUP96 caused early flowering phenotypes, indicating that NPCs are also involved in flowering time regulation [5]. Interestingly, only HOS1 was demonstrated to interact with specific chromatin such as the floral repressor FLOWERING LOCUS C (FLC) chromatin to regulate FLC transcription under low temperature [14–16].
Besides their involvement in hormonal and cold stress signaling and flowering time control, some Nups such as NUP160 and Seh1 are also important for disease resistance in Arabidopsis [17,18]. Similarly, mutation of either of NUP85, Seh1 and NUP133 led to defective responses to symbiotic microorganisms in Lotus Japonicus [19,20]. Most recent genetic studies have revealed NUP85 (also termed as SBB1, suppressor of bak1 bkk1-1) as a suppressor for BRI1-ASSOCIATED KINASE 1 (BAK1) and BAK1-LIKE 1 (BKK1)-mediated cell death control [21]. These studies suggested that Nups also participate in plant defense and cell death control. Nevertheless, the biological functions of many Nups remain elusive in plants.
Recently, a proline-rich protein gene, SICKLE-1 (SIC-1), was isolated because sic-1 mutants exhibited enhanced expression of stress-inducible RD29A-LUC reporter in response to abiotic stresses such as cold and salt treatments [22]. To identify new regulatory components involved in the response to abiotic stresses, we performed a forward genetic screen after EMS mutagenesis of the RD29Apro-LUC line in the sic-1 mutant background. NUP85 was identified as its mutation caused significantly reduced expression of RD29A-LUC in response to ABA and salt stress. Moreover, we discovered a list of putative NUP85 interacting proteins by affinity purification followed by mass spectrometry and further validated the interaction between NUP85 and MED18.
To identify new factors involved in abiotic stress responses, we performed a forward genetics screen using an EMS mutant population generated in the sic-1 mutant background with the stress-inducible proRD29A-LUC reporter gene. Putative mutants with altered luciferase (LUC) activities under abiotic stress treatments such as ABA treatment and salt stress were selected. One suppressor mutant was identified which exhibited reduced bioluminescence in response to ABA and salt stress compared to sic-1 plants (Fig 1A). As revealed in Fig 1A and 1B, the luminescence representing RD29A-LUC expression was highly induced in sic-1 by ABA and salt stress, whereas the luminescence was considerably weaker in the double mutant than sic-1 after either ABA or salt stress treatment. The luminescence intensities of the suppressor mutant were similar to those in the Col gl1 parental line harboring the proRD29A-LUC transgene (referred to as WT), which showed little LUC activity likely due to progressive transgene silencing [22,23]. After genetic mapping and whole genome re-sequencing, we discovered a mutation causing a premature stop codon in the 4th exon in NUP85 (AT4G32910), which encodes a nucleoporin protein (Fig 1C).
Gene expression data show that the LUC expression was highly induced in sic-1 by ABA, but this induction was impaired in the suppressor (i.e. sic-1 nup85) mutant (Fig 1D). The transcript level of endogenous RD29A was induced by ABA treatment in WT, sic-1 and the double mutant. However, the induction was significantly lower in the WT and sic-1 nup85 double mutant compared to that in sic-1 (Fig 1E). We also tested the expressions of other stress-responsive genes and the results showed that ABA-induced expression of COLD-REGULATED 15A (COR15A) and RD29B was lower in sic-1 nup85 double mutants than that in sic-1 mutant plants (S1 Fig).
To verify if the suppressor phenotype was caused by the NUP85 mutation, we cloned the genomic sequence of NUP85 with its native promoter and generated two independent complementation lines by transforming sic-1 nup85 double mutants. As shown in S2 Fig, the leaves of sic-1 nup85 double mutant were bigger than sic-1. The wild type NUP85 gene rescued the bigger leave size phenotype of the double mutant. We also examined the expression of RD29A-LUC reporter gene in the complementation lines after ABA and NaCl treatments (Fig 2). The diminished luminescence in sic-1 nup85 double mutant was rescued in the NUP85 complementation lines, which exhibited comparable LUC activity to sic-1. These genetic evidences demonstrated that the NUP85 mutation is responsible for the double mutant phenotypes.
To investigate if NUP85 affects the expression of endogenous stress responsive genes in response to ABA and salt stress, we obtained two independent T-DNA insertion homozygous mutants, nup85-1 (SALK_133369) and nup85-2 (SALK_113274) (Fig 1C). PCR analysis confirmed that the two T-DNA insertion mutant lines are homozygous (S3A Fig). Gene expression data further showed that NUP85 expression was significantly lower in those two mutant lines compared to the Col-0 wild type (S3B Fig). We then treated the Col-0 wild type and two nup85 mutants with 50 μM ABA for 3h, and found that the expression of several stress responsive genes such as RD29A, COR15A and COR47 was significantly lower in the nup85 mutants than that in the wild type after ABA treatment (Fig 3A). In addition to ABA, we also investigated if high salinity induced expression of stress responsive genes was also altered by the NUP85 mutation. As shown in Fig 3B, salt stress induced expression of RD29A, COR15A and COR47 was evidently impaired in the two nup85 mutants after treated with 0.3M NaCl for 5h.
To better understand the genome-wide effects of NUP85, we performed RNA-sequencing experiments. Col-0 wild type and nup85-1 mutant seedlings were treated with mock or 50 μM ABA for 3 hours. Under mock treatment, there were 197 differentially expressed (DE) genes which showed more than 1.5-fold changes in nup85-1 mutant compared to Col-0 wild type plants (S1 Table). Gene ontology (GO) analysis revealed that the DE genes regulated by NUP85 under mock conditions were enriched in categories such as responses to stimulus, response to stress and response to hormone, indicating an important role of NUP85 in plant responses to environmental stresses (Fig 3C). Upon ABA treatment, there were totally 1389 ABA-responsive genes (759 ABA induced genes and 630 ABA repressed genes), whose expressions were significantly induced or repressed more than 4-fold changes by ABA treatments in Col-0 wild type (S2 Table), whereas there were 1357 ABA-responsive genes in nup85-1 mutants after ABA treatments with 982 overlapping genes in Col-0 wild type (S3 Table). Out of 1389 ABA responsive genes in Col-0, 178 ABA responsive genes including RD29B exhibited significantly altered expressions in nup85-1 mutants in comparison with Col-0 wild type, suggesting that about 13% of ABA responsive genes are regulated by NUP85 (S4 Table). GO analysis further revealed that those NUP85 regulated genes were also enriched at categories such as response to stimulus, response to chemical and response to abiotic stimulus (S4 Fig). It is possible that Nups may have redundant function in regulating ABA responsive genes since nup85 single mutants did not alter most ABA-responsive genes. The heat map generated with NUP85 regulated ABA- responsive genes in both Col-0 wild type and nup85-1 mutant showed that the induction of a group of ABA-responsive genes was partially suppressed by the NUP85 mutation (Fig 3D). In contrast, a small portion of DE genes were up-regulated in nup85-1 mutant compared to Col-0 wild type. These results indicate that NUP85 is involved in regulating gene expressions in response to ABA.
Since we have shown that NUP85 is important for the expression of several stress responsive genes in response to ABA and salt stress, we next examined if nup85 mutants may show altered stress phenotypes to ABA and salt stress. In plants, the NUP107–160 sub-complex consists of eight members, NUP160, NUP133, NUP107, NUP96, NUP85, NUP43, SEH1 and SEC13 [3,21,24]. Thus, we isolated homozygous mutant plants for the other seven nucleoporins and tested their stress phenotypes. As shown in S5 Fig, mutation of NUP43, NUP107, NUP133, NUP96, SEC13 and SEH1 did not cause obvious changes in their responses to ABA and salt stress when compared to Col-0 wild type control plants. The average root length of those mutants was similar to the Col-0 wild type control after being transferred to ABA and NaCl containing medium. Nevertheless, two alleles of nup85 mutants were slightly more sensitive to ABA and salt stress as their root length was shorter than the wild type control after being transferred to ABA- and NaCl-containing medium (Fig 4A and 4B). In contrast, the root length of Col-0 wild type and nup85 mutants was comparable when they were transferred to control medium. In addition to the nup85 mutants, hos1 and nup160 mutants also showed increased sensitivity to ABA and salt stress as their root growth was significantly less compared to the Col-0 wild type after being transferred to the ABA and NaCl containing medium (Fig 4C and 4D). We also examined the responses of nup85-1 nup160 and nup85-1 hos1 double mutants to ABA and salt stress. As illustrated in Fig 5A, nup85-1, nup160, hos1 single mutants and those two double mutants all displayed hypersensitivity to ABA and NaCl because their root length was shorter than Col-0 wild type after being transferred to the indicated ABA and NaCl containing medium. The root length of the double mutants was not evidently shorter than the single mutant (Fig 5B), indicating a lack of additive phenotypes in the double mutant, and thus suggesting that the Nups may function in the same genetic pathway in regulation of the tested stress responses. These genetic results suggest that NUP85, NUP160 and HOS1 are involved in ABA and salt stress responses.
To investigate if NUP160 and HOS1 affect the expression of stress responsive genes in response to ABA and salt stress, we treated Col-0 wild type, nup160 and hos1 mutants with mock, 50 μM ABA or 0.3 M NaCl. The results showed that ABA and salt stress induced expression of RD29A, COR15A and COR47 was expressively reduced in nup160 and hos1 mutants after ABA and salt treatment when compared to the Col-0 wild type (Fig 6A and 6B). Additionally, we also tested the expression of RD29B in the Col-0 wild type and those mutants. We found that the expression of RD29B was significantly inhibited in nup160, hos1 and nup85-1 mutants compared to Col-0 wild type (S6 Fig). These results suggest that HOS1 and NUP160 are also required for proper expression of stress responsive genes.
Recent proteomics studies have revealed more than 22 Nups in plants [24]. To better understand the biological functions of NUP85 and its associating proteins, we performed anti-MYC immunoprecipitation and subsequent mass spectrometry (IP-MS) in WT and Nup85pro: NUP85-MYC transgenic plants. From two independent IP-MS experiments, we totally identified 55 putative NUP85 interacting proteins following two criteria to rule out the nonspecifically interacted proteins (1) proteins are present in the two independent replicates of IP-MS data; (2) proteins that have unique peptides or significantly more peptides (at least 5-fold more) identified in NUP85 transgenic plants compared to WT samples. The full list of putative NUP85 interacting proteins are listed in S5 Table, NUP160, NUP133, NUP43, NUP96, NUP107 and Seh1 as well as Sec13 were present in the NUP85 immuno-complex with abundant unique peptides, confirming the conserved configuration of NUP107-160 sub-complex in plants. HOS1 was also present in the NUP85 immuno-complex with comparable abundance to other Nups in NUP107-160 complex. Besides Nups, some Transducin/WD40 repeat-like superfamily proteins such as Sec13A, which are known to be involved in mRNA and protein transport, were found in the NUP85 immuno-complex. Interestingly, several mediator subunits including MED16, MED14 and MED18 were also precipitated by NUP85, suggesting that possible involvement of NUP85 and other Nups in transcriptional regulation. To validate some of the candidate interacting proteins, we cloned HOS1 and Sec13A, and tested their interactions with NUP85 in split-luciferase (LUC) complementation assays. The results indicated that NUP85 could directly interact with Sec13A but not HOS1 (S7 Fig).
To investigate if there is an interaction between NUP85 and mediator subunits, we selected MED18, which has been reported to be involved in ABA signaling pathway [25,26]. As shown in Fig 7A, the co-expression of MED18-Cluc and NUP85-Nluc resulted in strong LUC activity in the split-LUC complementation assay. In contrast, the co-expression of NUP85-Nluc and empty Cluc or PYL1-Cluc did not cause any detectable LUC activities. Co-immunoprecipitation assay in protoplasts showed that MED18-HA was precipitated by NUP85-GFP (Fig 7B), further confirming the interaction between NUP85 and MED18. In addition, we also investigated the temporal and spatial expression patterns of NUP85 and MED18 in Arabidopsis electronic fluorescent pictograph (eFP) browser. As illustrated in S8 Fig, both NUP85 and MED18 are widely expressed in most of tissues such as seeds, leaves, and stems, especially in shoot apex and flowers with higher expression levels.
We next examined ABA and salt stress induced gene expression in Col-0 wild type and med18 mutants (Fig 7C). Similar to nup85 mutants, the ABA and salt stress induced expression of RD29A, COR15A and COR47 was also inhibited in med18 mutants relative to the wild type Col-0, signifying overlapping functions of MED18 and NUP85. To further explore the mechanism of how Nups regulate the expression of stress responsive genes, we examined ABI5 expression in wild type Col-0, nup85-1, nup160 and hos1 as well as med18 mutants under mock or ABA treatments. ABI5 is not only one of the MED18 targeted genes [25], but also a key regulator for the expression of stress responsive genes [27]. As illustrated in S9 Fig, ABI5 expression was obviously inhibited in those Nup mutants and med18 mutants after ABA treatments when compared to the wild type Col-0. Consistent with the interaction between NUP85 and MED18, med18 mutants also displayed hypersensitivities to ABA and salt stress because their primary root length was shorter than Col-0 wild type after transferred to MS medium containing indicated concentrations of ABA and NaCl (Fig 8).
Although core factors in abiotic stress signaling and responses have been identified with extensive genetic and biochemical studies in plants [28], abiotic stress responses remain intricate as many other factors yet to be discovered. In our present study, we have identified NUP85 as a regulator of ABA and salt stress responsive genes. The mutation of NUP85 impaired the ABA- and salt stress-induced expression of RD29A and several other stress-responsive genes, suggesting that NUP85 may be a positive regulator for ABA and salt responses. Similarly, mutation of other Nups such as HOS1 and NUP160 also reduced the expression of stress responsive genes, thereby indicating that NUP85 and other Nups components have overlapping functions in regulating the expression of stress responsive genes. Consistent with the altered expression of stress responsive genes, nup85, nup160 and hos1 mutants are hypersensitive to ABA and salt stress. Thus, our results indicate the involvement of Nups in ABA signaling and salt stress responses.
Currently, one of the best characterized NPC components is HOS1, which has been shown to associate with the chromatin of FLOWERING LOCUS C (FLC) and the promoter region of MIR168b, to regulate flowering time and miRNA biogenesis [15,23]. Nevertheless, how other Nups regulate gene expression remains largely unknown. Based on previous and our present studies, NPC components were suggested to play a role in gene expression regulation through various mechanisms. In fact, a number of previous studies have revealed that mutations in NPC components resulted in the accumulation of mRNAs in the nuclei [3,7,8,29,30], which could subsequently affect the expression of genes such as stress responsive genes in plants. Moreover, increasing evidence supported NPC as gene-gating organelles which could recruit actively transcribed genes to them and regulate gene expression [31]. It is likely that transcription of group of stress responsive genes is dependent on certain NPC components in response to different environmental stresses. Our results showed that the expression of RD29A and several stress responsive genes was impaired at the transcriptional level in nup85, hos1 and nup160 mutants, indicating that these NPC components are required for high level transcription of the stress responsive genes in response to ABA and salt stress. Additionally, plant NPC was recently demonstrated to undergo conformational switch in response to pathogen infection, allowing significant activation of nucleocytoplasmic transport and multiple stress-related signaling pathways [32]. The study raised the possibility that NPC components may facilitate the transcription of stress responsive genes in response to various environmental stresses by controlling the transport of critical signaling components and transcriptional regulators between the nuclei and cytoplasm.
In addition to these evidence, proteomics data from our studies further revealed several putative NUP85 interactors, which not only broadened our understanding of NPC functions in plants, but also indicated some potential mechanisms of how NPC components regulate the gene expression. It is worth noting that all the proteins identified from our proteomics data are putative NUP85 interacting proteins until they are validated by further experiments. Our IP-MS results suggested that NUP85 forms a complex with HOS1 and seven other nucleoporins located in the NUP107-160 sub-complex in plants, thus confirming the conserved configuration of NUP107-160 sub-complex in eukaryotic cells. Additionally, some Transducin/WD40-repeat proteins were co-purified with NUP85 including Sec13A, which are involved in assembling NPC domains, signal transduction and mRNA and protein transport [24]. Another interesting discovery from our proteomics data was that several mediator subunits were also present in the NUP85 immuno-complex, providing an alternative way for NUP85 and other Nups to regulate transcription. Recent studies in yeast cells showed that nuclear pore-associated TREX-2 complex directly interacts with mediators to regulate gene expression through the RNA Pol II transcription machinery [33]. However, whether NPC could regulate transcription through the mediator complex remains unknown in plants. Our data suggested the possibility of NUP85 and other NPC components in accessing the core transcriptional machinery through modulating the mediator complex and RNA Pol II association. Among the mediator subunits co-purified by NUP85, MED18 has been known to be involved in ABA signaling, flowering control and plant defense [25,34]. The ABA and salt stress induced expression of several stress responsive genes is significantly attenuated in med18 mutants, which was similarly observed in nup85 mutants. Importantly, we further demonstrated that NUP85 and MED18 are both important for the expression of ABI5, which could activate the expression of downstream stress responsive genes [35]. In summary, our study illustrates that NUP85 and some Nups such as NUP160 and HOS1 regulate stress-responsive gene expression through cooperating with mediator complex. Thus, our findings provide new insights into the role of Nups in ABA signaling and salt responses.
To screen for new regulatory components involved in abiotic stresses, we first performed EMS mutagenesis on sic–1 mutants expressing RD29Apro-LUC as described in [23]. The mutants with significant reduced luminescence in response to 100μM ABA or 200 mM NaCl were selected. The wild type (WT) for LUC assays refers to Col–0 ecotype with the gl–1 mutation harboring the proRD29A-LUC transgene. For abiotic stress phenotypic analysis and gene expression, WT refers to Col-0. The T-DNA insertion mutants nup85-1 (SALK_133369C), nup85-2 (SALK_113274C), nup160 (SALK_126801C), hos1 (SALK_069312C), nup43 (SALK_095344C), nup107 (SALK_057072), nup96 (SALK_135920C), nup133 (SALK_092608C), sec13 (SALK_045825C), seh1 (SALK_022717C) and med18 (SALK_027178C) were in Col-0 background and were obtained from the Arabidopsis Information Resource Center (ABRC). nup85-1 nup160 double mutant was described in [36]. nup85-1 hos1 double mutant was generated by crossing the nup85-1 and hos1 single mutant. The seeds were surface sterilized and sown on half Murashige and Skoog (MS) medium containing 1% sucrose and 0.8% agar. After 2 days in 4°C, the plates were moved to growth chamber under photoperiod of 16 h light/8 h dark. Root growth inhibition assays were performed as described previously [37]. For post-germination root growth assays, 3 or 4-days-old seedlings were first germinated on vertical half MS medium and then were transferred to ABA-(20 μM) or NaCl (100 mM) containing medium and the primary root growth was documented and quantitatively measured at 7 days after transfer.
To generate NUP85 complementation lines, the genomic sequence of NUP85 with about 2 kb upstream sequence before its start codon was amplified by high-fidelity DNA polymerase (PrimeSTAR HS DNA Polymerase, Clontech). The PCR product was first inserted into pENTR vector (Invitrogen), and then was transferred to destination vector pGWB-16 (MYC tag) via LR reaction. After verified by sequencing, the construct was introduced into Agrobacterium GV3101 for plant transformation. All indicated constructs were transformed by floral dip method [38]. The homozygous NUP85 complementation lines were used for affinity purification.
Twelve-day-old seedlings of wild-type, sic–1 and sic–1 nup85 were sprayed with distilled water or indicated concentrations of ABA and NaCl solutions for 3h and 5h, respectively. The LUC images were captured using a low-light video imaging system with WinView software.
Map-based cloning was performed as described [23]. The F2 mutants with suppressed LUC phenotype were selected by genotyping for sic-1 mutant. The results indicated that the mutation was located in chromosome 4 between 12,980 Kb and 16,500 Kb. Whole genome sequencing was performed, and a mutation was found in At4g32910 locus.
RNA was extracted from 7-day-old seedlings by RNeasy Plant mini kit (QIAGEN) followed by DNase (Turbo) digestion. For testing gene expressions, 1 μg total RNA was used for reverse transcription using M-MLV reverse transcriptase (Promega) according to the manufacturer's instructions. Real-time PCR was performed using iQ SYBR Green Supermix (Bio-Rad) on a CFX96 real-time PCR detection system (Bio-Rad). Actin 2 (ACT2) was used as the internal reference for all reactions. The relative gene expression value was calculated as described previously [39]. All primers were listed in S6 Table.
wild type and nup85 mutant seeds were germinated on 1/2 MS medium in growth chamber at 24°C for 7 days and were treated with mock or 3 hr of 50 μM ABA at room temperature. The total RNA was isolated with Trizol reagent (Invitrogen) according to the manufacturer’s instruction and RNA-sequencing were carried out by the Core Facility of Genomics in Shanghai Plant Stress Biology Center, China. Quality control was checked using FastQC (www.bioinformatics.babraham.ac.uk/projects/fastqc). RNA-Seq reads were trimmed using the fastx_trimmer command FASTX-Toolkit (hannonlab.cshl.edu/fastx_toolkit/index.html) with parameter “-f 11 -l 80” before alignment. Trimmed reads were mapped to Arabidopsis reference genome (TAIR10) using TopHat2 using “—b2-very-sensitive” option [40]. Read count for each gene were obtained using featureCounts [41]. Differentially expressed (DE) genes were identified using DESeq2 [42]. NUP85-regulated genes were defined under the criteria: 1) genes were differentially expressed after ABA in Col-0 with 4-fold change; 2) gene expressions in nup85 mutants were at least 1.5-fold higher or lower than those in wild type after ABA treatments. GO enrichment analysis were performed at http://geneontology.org/.
For affinity purification of NUP85 and its associated proteins, about two grams of 12-day-old NUP85pro: NUP85-MYC transgenic seedlings were collected and same amount of WT seedlings was used as a negative control. Total protein extraction and affinity purification were performed as described previously [39]. The protein supernatants were incubated with 30 μL of anti-MYC agarose beads (Abcam), which had been pre-equilibrated with lysis buffer. After incubation at 4°C with rotation for 4 hours, the agarose beads were washed four times with lysis buffer followed by one time wash with 1 mL of PBS buffer. The agarose beads were finally resuspended in 100 μL of PBS buffer. After trypsin digestion, the mass spectrometer was operated in the data-dependent mode in which a full MS scan (from m/z 350–1500 with the resolution of 30,000 at m/z 400), followed by the 10 most intense ions being subjected to collision-induced dissociation (CID) fragmentation. CID fragmentation was performed and acquired in the linear ion trap (normalized collision energy (NCE) 30%, AGC 3e4, max injection time 100 ms, isolation window 3 m/z, and dynamic exclusion 60 s) according to [43]. The raw files were searched directly against the Arabidopsis thaliana database (TAIR10) with no redundant entries using Proteome Discover 2.1. Precursor mass tolerance was set at 10 ppm, and the MS/MS tolerance was set at 0.6 Da. The searches were performed with trypsin digestion and allowed a maximum of two missed cleavages on the peptides analyzed from the sequence database. The false discovery rates for proteins and peptides were set at 0.01, and the minimum peptide length was six amino acids. To identify significantly changed proteins from IP-MS results, we performed two biological replicates. The putative interacting proteins of NUP85 were selected based on two criteria:(1) proteins were identified in both replicates of IP-MS data or (2) proteins that have unique peptides or significantly more peptides in NUP85 transgenic plants than in control samples.
The full-length coding sequences of NUP85, HOS1, Sec13A and MED18 were amplified by PCR using primers listed in S2 Table. The PCR products were first cloned into pENTER vector (Invitrogen) and then transferred to nLUC/cLUC vectors via LR reactions. Split-LUC complementation assay was performed in Arabidopsis protoplasts per [39]. The reconstitute LUC activity was detected in the dark by cooling camera. The image and quantification of LUC activities were analyzed with Winview software.
The NUP85-GFP and MED18-HA or CDK8-HA were transiently co-expressed in Arabidopsis protoplasts per [39]. After transformation, the protoplasts were collected and suspended in 1 mL lysis buffer containing 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 1 mM phenylmethylsulfonyl fluoride (PMSF), 2 mM DTT, 0.1 (v/v) Triton X-100 and Protease Inhibitor Cocktail (Sigma-Aldrich) in ice for 20 min. The lysis was then centrifuge at 13000 rpm for 15 min at 4°C and was incubated with pre-equilibrant GFP-Trap beads (Chromo Tek) for at least 4 hours with continuous rotation. The beads were washed at least four times with lysis buffer at 4°C and boiled in 4× SDS loading buffer for 10 min. Protein samples were separated by SDS-PAGE and were further detected with polyclonal anti-HA (Abcam) and anti-GFP antibody (Roche).
|
10.1371/journal.pgen.0030165 | Linkage Analysis of a Model Quantitative Trait in Humans: Finger
Ridge Count Shows Significant Multivariate Linkage to 5q14.1 | The finger ridge count (a measure of pattern size) is one of the most heritable
complex traits studied in humans and has been considered a model human polygenic
trait in quantitative genetic analysis. Here, we report the results of the first
genome-wide linkage scan for finger ridge count in a sample of 2,114 offspring
from 922 nuclear families. Both univariate linkage to the absolute ridge count
(a sum of all the ridge counts on all ten fingers), and multivariate linkage
analyses of the counts on individual fingers, were conducted. The multivariate
analyses yielded significant linkage to 5q14.1 (Logarithm of odds
[LOD] = 3.34, pointwise-empirical
p-value = 0.00025) that was predominantly driven by
linkage to the ring, index, and middle fingers. The strongest univariate linkage
was to 1q42.2 (LOD = 2.04, point-wise p-value
= 0.002, genome-wide p-value = 0.29). In
summary, the combination of univariate and multivariate results was more
informative than simple univariate analyses alone. Patterns of quantitative
trait loci factor loadings consistent with developmental fields were observed,
and the simple pleiotropic model underlying the absolute ridge count was not
sufficient to characterize the interrelationships between the ridge counts of
individual fingers.
| Finger ridge count (an index of the size of the fingerprint pattern) has been
used as a model trait for the study of human quantitative genetics for over
80 years. Here, we present the first genome-wide linkage scan for finger
ridge count in a large sample of 2,114 offspring from 922 nuclear families.
Our results illustrate the increase in power and information that can be
gained from a multivariate linkage analysis of ridge counts of individual
fingers as compared to a univariate analysis of a summary measure (absolute
ridge count). The strongest evidence for linkage was seen at 5q14.1, and the
pattern of loadings was consistent with a developmental field factor whose
influence is greatest on the ring finger, falling off to either side, which
is consistent with previous findings that heritability for ridge count is
higher for the middle three fingers. We feel that the paper will be of
specific methodological interest to those conducting linkage and association
analyses with summary measures. In addition, given the frequency with which
this phenotype is used as a didactic example in genetics courses we feel
that this paper will be of interest to the general scientific community.
| Finger ridges and ridge patterns are highly heritable, durable, and age-independent
human traits and have been studied as a model quantitative trait in humans for over
80 years [1]
. They develop between approximately the 13th and 18th weeks of gestation, and in
the absence of trauma remain essentially unchanged throughout life. The cutaneous
mechanoreceptive afferent neurons that innervate the fingertips develop in alignment
with the ridges [2], lending support to the theory that fingerprints play a role in
gripping [3]
and tactile perception [4]. While relatively little is known about the developmental
processes underlying fingerprint patterns, these results suggest that factors
influencing the direction and complexity of ridge pattern formation also influence
the receptive fields of the mechanoreceptors.
The development of ridge patterns coincides with the regression of embryonic volar
pads on fingers, and the type and size of patterns are largely determined by the
size and timing of subsidence of these pads [5]. Any genetically or environmentally
determined growth disturbances that affect the limbs in the critical period of ridge
formation may also affect normal development of ridges and ridge patterns. Finger
ridge count is also subject to a sex chromosome dosage effect, with the largest
count encountered in females with X monosomy (Turner's syndrome) and the lowest in
the X, Y polysomies [5]. Hence, dermatological traits can assist in determining the
nature and timing of developmental disturbance.
Traditionally, the ridge count is defined as the number of ridges that intersect or
touch the line drawn from the easily recognized triradius (where three ridges meet)
to the center of the pattern [6]. The most common pattern, a simple loop
(60%–70% of all patterns [6]), characterized by a
single triradius, is most advantageous for tactile perception [7] and precision grip
[3].
Whorls have two triradii yielding two counts, while simple arches have no true
triradii, resulting in a zero count. When the ridge count is used as a measure of a
maximum pattern size on fingers, only the largest count from each finger is scored,
and their sum is defined as the total ridge count. Alternatively, the sum of all
possible counts on all ten fingers can be calculated yielding an absolute ridge
count (ARC) [5], a measure of the total pattern size.
Both total ridge counts and ARC are highly heritable. Genetic effects have been found
to account for 90%–95% of the variation on these
measures. Estimates using either traditional correlation-based methods
[5,6] or structural
equation models fitted to twin and sibling [7] or family [8–10] data have found additive genetic
effects to account for around 90% of the variation. That the remaining
genetic variation arises from dominance and/or higher order genetic effects was
initially suggested by skewness in the distribution of these measures
[6] and
supported by the modeling of twin and sibling data [7].
The ridge counts of individual fingers are interrelated (correlations range from
∼0.4 to 0.8) and highly heritable, with the lowest heritability
(∼0.50) observed for the thumb and little fingers [5,6,8]. These findings have led to the development of a variety of
models to explain the genetics of finger ridge count, the simplest of which
postulates pleiotropic gene effects that assume a single genetic factor determining
the general magnitude of the counts and random influences accounting for
between-finger variation [6]. More complex models, such as those developed by Martin et al.
[7], have
found shared genetic effects common to all digits, in addition to patterns of
covariation suggestive of developmental fields acting across the fetal hand,
implying heterogeneous gene action between digits.
The aim of this study was to identify loci influencing finger ridge counts by
conducting a genome scan on 2,114 twin and singleton offspring from 922 twin
families (equivalent to 2,826 quasi-independent sib pairs). The data were collected
from two twin cohorts; an adolescent sample (which included non-twin siblings)
[11,12] and an adult sample
[13]. As
described in the Materials and Methods section below, prints were not available for
digits IV and V for participants from the adult study.
Given the evidence for developmental field effects from quantitative genetic
analysis, we conducted both univariate (ARC) and multivariate (simultaneously
modeling ridge counts (radial + ulnar) for each of the ten fingers)
variance components linkage analyses. The aim was to determine whether loci
influencing ridge count acted in a simple pleiotropic fashion (i.e., all fingers
were influenced by the same loci to the same extent), or if more complicated
pleiotropic patterns indicative of field effects were present (e.g., a quantitative
trait locus's (QTL's) maximum influence was seen on the little finger and the
effects tapered off towards the thumb).
As shown in Figure 1, the
strongest evidence for univariate linkage for ARC was seen at 1q42.2 (250 cM,
Logarithm of odds (LOD) = 2.04; point-wise p-value
= .002, genome-wide p-value = .29). At this
position, the QTL explained 21% of the variance in ARC, while the
multivariate test revealed low QTL factor loadings across the five fingers,
explaining 7.7%, 12.7%, 10.0%, 13.8%,
and 9.3% of the variation in individual counts from thumb to little
finger, respectively (Figure
2A). As may be expected for a locus at which small pleiotropic effects were
found for all digits, the multivariate test for linkage was less powerful than the
univariate test (multivariate LOD = .46), because the pattern of factor
loadings is most economically summarized by the mean (or total) score with a single
degree of freedom [14]. The next highest univariate LOD scores were found at 15q26.1
(95cM, LOD = 1.52; point-wise p-value = .006,
genome-wide p-value = .62) and 7p15.3 (35cM,
LOD= 1.26; point-wise p-value = .01,
genome-wide p-value = .79). Loci with LOD scores greater
than 1 are listed in Table 1.
The strongest evidence for multivariate linkage was seen at 5q14.1 (95 cM,
multivariate LOD = 3.34). As shown in Figure 2B, the pattern of loadings is consistent
with a developmental field factor whose influence is greatest on the ring finger,
falling off to either side. It is interesting that the highest heritability for
ridge count is seen for the middle three fingers [6,8]. The QTL loading was strongest for the ring finger
(24.1% of the variation), explained around 6.6% and
11.2% of variance in ridge counts on the index and middle fingers,
respectively, 2.6% in little finger ridge count, and less than
1% in thumb ridge count. Post-hoc univariate linkage analyses for each
finger individually (shown in Figure
3) confirmed this pattern of factor loadings. Pointwise simulations
(described below) revealed that a LOD score this extreme arose by chance in 1/4,000
simulations, yielding a pointwise empirical p-value of 0.00025.
There was no evidence for linkage to the X chromosome. The highest LOD score observed
(LOD = 0.25) was for the univariate analyses at a locus at 25 cM. It is
possible that the assumption of dosage correction may have obscured linkage to a
gene acting in a pseudoautosomal manner. However, a post-hoc univariate analysis
(not shown here) of the total absolute ridge count, in which there was no assumption
of dosage compensation, did not yield any further evidence for linkage.
As this is the first linkage analysis for finger ridge count, these peaks are novel
and there are numerous candidate genes lying under them that warrant investigation.
The linkage region on Chromosome 5 contains a number of zinc-finger genes
(ZFYVE16, ZCCHC9, and ZBED3). Similarly, the
peak on Chromosome 19 (85 cM) spans a cluster of zinc-finger genes. The Chromosome
15 (55 cM) peak is within 5 cM of the Fibrillin 1 gene (FBN1),
which is involved in maintenance of elastic fibers and anchoring epithelial cells to
the interstitial matrix. Mutations in this gene result in severe developmental
malformations of the hands (among other phenotypes) and are a major cause of Marfan
syndrome and a range of other conditions that result in arachnodactyly or
brachydactyly. Given findings that ridge count is correlated with finger length in a
sample of individuals with Marfan syndrome [15], it seems plausible that polymorphic
variations in this gene that influence the development of the digits may have a
secondary influence on the normal development finger pads and dermal ridges.
Early studies attempted to link the presence of an arch pattern (i.e., a zero ridge
count) on any finger to blood groups, finding no linkage to the rhesus
(1p36.2–1p34) or P1 (22q13.2) blood groups but some evidence for linkage
to the haptoglobin locus (16q22.1) [16]. The small peak in both the
univariate and multivariate genome scans on Chromosome 16 at 110 cM is approximately
15 cM distal to the haptoglobin locus, so it is possible that this peak may in fact
replicate these early findings.
As evident from Table 1, a
variety of factor loading patterns were observed during the multivariate genome
scan. At the multivariate level, the genetic and environmental factor loadings
(summarized in Table 2) showed
patterns consistent with developmental fields, such as the peaks on Chromosome 5 (95
cM) and Chromosome 1 (35 cM), in which the covariation between adjacent fingers was
higher than between more distal fingers. At other loci, such as the second highest
multivariate linkage peak, on Chromosome 15 (55 cM), the peak loaded predominantly
on a single pair of digits. These findings were anticipated by earlier multivariate
genetic analyses of ridge counts that found evidence of both common and group (or
field) genetic factors responsible for the high level of covariation between ridge
counts on different digits, including some of the QTL factor patterns seen here
[7].
Although the multivariate peak on Chromosome 5 was primarily driven by the fourth
digit, it is unlikely that this result is a consequence of the missing data in the
adult sample. The missing data from digits IV and V were not imputed, and may be
considered missing completely at random (as the decision not to collect prints on
digits IV and V was unrelated to the ridge counts that would have been observed)
[17]. The
analytic approach used in these analyses (raw continuous data maximum likelihood
analyses implemented in Mx [18]) has previously been shown to yield unbiased results when
the distribution in normal or slightly skewed and missing values are missing
completely at random [19]. In addition, the post hoc univariate analyses (Figure 3; in which all available
data for each digit were analyzed in separate univariate analyses) provide support
for the multivariate results. As expected, the highest peak was observed for the 4th
digit, followed by the 3rd and the 2nd.
The absence of significant univariate results and the general dearth of peaks from
the univariate analysis of ARC suggest that the simple pleiotropic model specified
by using a sum score such as ARC alone is not sufficient to characterize the rich
biological interrelationships influencing finger ridge count. However, in some
areas, such as the peak on Chromosome 1, we did observe QTL loadings that were
consistent with pleiotropic effects. In addition, given the disparity between the
digits in their contributions to the multivariate peak on Chromosome 5, assessing
the individual contributions of the fingers may also provide information that can
aid in the interpretation of multivariate linkages. These analyses have shown that
even for conceptually and theoretically simple phenotypes such as finger ridge
count, using a single approach to linkage analyses (such as a sum score) may place
biologically implausible restrictions upon the model, significantly reducing the
power to detect effects and the interpretability of results.
A limitation of our study is the well-known low power of sib-pair linkage analysis
for unselected samples, which can in part be ameliorated through multivariate
analysis [14,20]. The apparent
advantages of multivariate analysis revealed here are somewhat exaggerated by our
inability to calculate the total ARC for adult twins because only digits I to III
were counted in that study. However, as discussed above, using the raw data maximum
likelihood option in Mx, joint analysis of the six fingers for which adult data are
available, together with the almost complete adolescent set, increases power of the
multivariate analysis by making use of every measured data point while providing
estimates of factor loadings unbiased by missing values [17].
In conclusion, we report the first linkage scan for finger ridge count, finding a
significant peak on Chromosome 5 and suggestive peaks on Chromosomes 1 and 15. Both
pleiotropic QTL effects consistent with development fields and nonpleiotropic
effects influencing single fingers were observed. In addition, we have demonstrated
that a comprehensive approach involving both multivariate analyses of constituent
phenotypes and univariate analyses of a sum score can be more informative than
simple univariate analyses in the presence of complex pleiotropic models.
The data were collected within the context of an adolescent twin family study
[11,12] and an adult twin
study [13].
Characteristics of the phenotypic and genotypic data are provided in Table 3.
Rolled fingerprints were collected from twin pairs and their available siblings
by research nurses trained to ensure that prints contained the centre of the
pattern and all triradii. In the adult study, because of a shortage of time in
the protocol, prints were only collected from the first three fingers of each
hand (it is much harder and takes longer to obtain good prints of digits IV and
V (the ring and little fingers) using traditional methods). From 1994 to 2005,
prints were collected using a fingerprint ink pad and archival quality paper.
From 2005 onwards, prints were collected using an electronic rolled fingerprint
scanner (Smiths Heimann Biometrics ACCO1394, http://www.shb-jena.com/ACCO1394_scanner_AQ.pdf).
The majority of ridge counts analyzed here were scored from the inked prints and
counted by eye using a binocular dissecting microscope (counting was performed
by three of the authors, BM, DZL, and SEM). The number of ridges lying between
the center of the pattern (core) and the triradius/triradii (delta) were counted
using standard conventions [6]. Ridge counts for ten individuals
were scored from the electronic prints using a purpose-built software package
[21]
. Summary statistics are given in Table 3. Since ARC is calculated by summing the ridge counts of all
ten fingers, it could not be calculated for individuals with missing data,
including all the adult twins. The phenotypic correlations between digits and
ARC are shown in Table 4.
To improve computational efficiency phenotypes were corrected for mean
differences between males and females, and transformed to
z-scores prior to analysis.
Genotypic data were available for 1,230 twins and sibs with fingerprint
phenotypes from the adolescent and 664 twin individuals from the adult study (as
detailed in Table 3). In
addition, 110 nongenotyped monozygotic pairs from the adolescent study were
included in these analyses to allow the within-family shared variance to be
partitioned into that due to additive and dominant genetic effects, respectively
[22].
The cleaning and error checking of the genotypic data for the adolescent and
adult studies have been described in detail elsewhere [23,24]. In summary, the adolescent
genotypic data was composed of three waves of genotyping. Each wave of
genotyping included overlapping samples and markers in order to check data
quality. Following error checking and data cleaning [24], the resulting
genotypic dataset comprised 796 markers (35 of which are duplicates) at an
average spacing of 4.8 cM (Kosambi). Similarly, the adult genotypic data was
composed of four waves of genotyping, with duplicate individuals and markers
between each wave of genotyping. Following error checking and data cleaning
[23],
the resulting genotypic dataset comprised a dense map of 1770 markers (394 of
which are duplicates) at an average spacing of 6.1 cM (Kosambi). Since the
adolescent and adult marker sets only partly overlapped, Identity By Descent
(IBD) estimates and information contents were obtained separately for the two
samples using a standard map[25] at 5cM intervals across the
genome using MERLIN 1.0.1[26]. As the IBD estimates were made at fixed points along
the chromosomes, the data from the two samples could then be jointly modeled at
5-cM intervals. Characteristics of these participants and the genotypic
information available are summarized in Table 3.
All linkage analyses were conducted by variance components analysis using raw
data maximum likelihood methods implemented in Mx1.63 [18]. For the
autosomal univariate variance components QTL analysis of ARC, a likelihood ratio
chi-square test
was used to compare the fit of the alternate model, in which the total variance
was modeled as the sum of the additive genetic, dominant genetic, QTL, and
unique environmental variances
to a null model in which variance due to the QTL was set to zero
.
Significant and suggestive thresholds and genome-wide empirical
p-values were obtained by simulation. Data for 1,000 simulated
unlinked genome scans that preserved the pedigree structures, information
content, and missing data patterns were obtained using MERLIN–simulate
[26].
IBDs were extracted from the simulated data, each replicate was analyzed in the
same way as the observed data, and the highest LOD score for each chromosome was
recorded. Empirical significant and suggestive thresholds were then estimated by
extracting the LOD scores that were obtained with a probability of 0.05 (i.e.,
once in every 20 null replicates) and once per genome scan, respectively.
Pointwise empirical p-values were obtained by calculating how
often a result as extreme as that which was observed, for the given map
position, occurred by chance within the simulated data. Genomewide empirical
p-values were obtained by extracting the highest LOD score
from each simulation replicate and recording how often a result as extreme as
that which was observed occurred by chance within these simulated data.
For the autosomal multivariate analysis of individual finger ridge counts, use of
raw data within Mx analyses allowed the inclusion of individuals with missing
data (in particular, all the adults twins missing prints for digits IV and V). A
similar alternate model was used
.
To provide the most conservative test of QTL variance, saturated factor models
were fitted for A, D, and E components, with the constraint for A and D matrices
that the factor loadings, patterned as a 5 × 5 Cholesky decomposition,
be equal for corresponding fingers on left and right hands. In this way, we
sought to capture the major QTL features while minimizing capitalization on
chance. As summarized in Figure
4, the alternate model thus contained five additive genetic factors, five
dominant genetic factors, a single QTL factor, and ten unique environmental
factors (patterned as a full 10 × 10 Cholesky decomposition). The
single QTL factor influenced all phenotypes and contained five estimated
parameters, one for each of the five pairs of digits. In the null model
,
the QTL factor loadings were set to zero. A likelihood ratio chi-square test
with five degrees of freedom
was
used to test the difference in fit of these two models. To obtain LOD score
equivalents for this test,
was converted to a p-value, which was then transformed to a
standard LOD score. Given the time required to conduct the multivariate linkage
analysis (over an hour per marker on our Linux server), calculation of empirical
significance values was not feasible. Instead, a pointwise empirical
p-value was calculated at the highest peak using simulated IBD
data information derived from ped files made using MERLIN–simulate
(4,000 replicates).
For X chromosome linkage analyses, we implemented a simple extension of the
X-linked variance components model [27], in which an extra additive
genetic variance component is modeled with the coefficient of relatedness
(usually set to 1/2 in the autosomal case) corrected for the sexes of the
siblings for each sib-pair combination. As for the autosomal additive
polygenetic case, covariation among relatives due to additive X-linked variance
arises because of alleles shared IBD. Assuming complete random X-inactivation
(lyonization), then
,
the coefficient of X-chromosome relationship [28] was set to 1/2 for
brother–brother pairs, 3/8 for sister–sister pairs, and 1/4
for opposite-sex pairs. In the multivariate analyses, the additional X-linked
additive genetic variance component was patterned as a 5 × 5 Cholesky
decomposition, with parameters constrained to be equal for corresponding fingers
on left and right hands. As for the autosomal case, in the multivariate analyses
a single QTL factor was modeled that loaded on all phenotypes and contained five
estimated parameters, one for each of the five pairs of digits. The QTL model
also assumed complete random X-inactivation, so that the X-linked QTL variance
in females was set to be half that of males. Following the suggestion of Kent et
al. [29,30], separate unique
environmental effects were estimated for males and females to allow for genotype
or environment interactions with sex. Based on the results of these analyses, a
post hoc decision was made not to obtain empirical p-values for
these X-linked analyses.
The National Center for Biotechnology Information (NCBI) Online Mendelian
Inheritance in Man (OMIM) database (http://www.ncbi.nlm.nih.gov/sites/entrez?db=OMIM)
accession numbers for the syndromes discussed in this paper are: Marfan
syndrome, #154700 and Dermatoglyphics—arch on any digit,
125570.
The NCBI OMIM database accession numbers for the genes discussed in this paper
are Fibrillin 1 (FBN1), *134797; Haptoglobin blood group,
*140100; P1 blood group, #111400; and Rhesus blood group,
+111700.
The NCBI Entrez (http://www.ncbi.nlm.nih.gov/sites/gquery) database accession
numbers for the genes discussed in this paper are ZBED3, 84327;
ZCCHC9, 84240; and ZFYVE16, 9765.
|
10.1371/journal.pcbi.1003992 | The Social Bayesian Brain: Does Mentalizing Make a Difference When We Learn? | When it comes to interpreting others' behaviour, we almost irrepressibly engage in the attribution of mental states (beliefs, emotions…). Such "mentalizing" can become very sophisticated, eventually endowing us with highly adaptive skills such as convincing, teaching or deceiving. Here, sophistication can be captured in terms of the depth of our recursive beliefs, as in "I think that you think that I think…" In this work, we test whether such sophisticated recursive beliefs subtend learning in the context of social interaction. We asked participants to play repeated games against artificial (Bayesian) mentalizing agents, which differ in their sophistication. Critically, we made people believe either that they were playing against each other, or that they were gambling like in a casino. Although both framings are similarly deceiving, participants win against the artificial (sophisticated) mentalizing agents in the social framing of the task, and lose in the non-social framing. Moreover, we find that participants' choice sequences are best explained by sophisticated mentalizing Bayesian learning models only in the social framing. This study is the first demonstration of the added-value of mentalizing on learning in the context of repeated social interactions. Importantly, our results show that we would not be able to decipher intentional behaviour without a priori attributing mental states to others.
| A defining feature of human social cognition is our insight that others' behaviour is driven by their beliefs and preferences, rather than by what is objectively true or good for them. In fact, a great deal of our social interactions are concerned with guessing others' mental states. But is such "mentalizing" of any help for predicting others' behaviour? After all, most animal species seem to cope with this problem without appealing to any form of sophisticated "Theory of Mind". Here, sophistication refers to the depth of recursive beliefs, as in "I think that you think that I think…" Although we are likely to engage in such recursive beliefs whenever our interests are tied up with others' (e.g. in the aim of deceiving them), it is unclear how these beliefs are updated and whether this gives us any advantage when we learn. These are the questions we address in this work, by combining computational and experimental approaches.
| What is so special about the way we select the most appropriate action in a social context? We make decisions on the basis of their expected consequences, which we may have to learn from trial and error. However, when this involves predicting other peoples' overt reactions, we almost irrepressibly engage in rich and complex representations of their hidden mental states, such as beliefs, emotions, intentions… In fact, one of the most critical aspects of social inference may be our insight that people's behaviour is driven by their beliefs rather than by physical reality, even if these beliefs happen to be false [1]. In this work, we ask whether this specific aspect of social cognition makes a difference when we learn.
We acquire this insight during early childhood [2], from our developing ability to attribute mental states to others, known as "Theory of Mind" (ToM) or "mentalizing" [3]. ToM is concerned with the interpretation of social signals, from eye gazes and facial expressions to overt behaviour and language, which is why it lies at the core of human social cognition [4]. We know that ToM engages large-scale specific brain networks [5], [6] and that severe neuropsychiatric disorders such as schizophrenia or autism are associated with its impairment [7], [8]. However, current research falls short of an understanding of the computational mechanisms underlying mentalizing, or of a clear demonstration of its added-value for decision making in social exchanges [1]. Here, we take inspiration from recent works in behavioural economics and experimental psychology, which investigate sophisticated mentalizing processes, of the sort that adaptive social skills such as persuading or deceiving proceed from. On the one hand, it has been shown that decisions made in the context of economic games entail recursive thinking of the sort "I think that you think that I think, etc…" [9], [10]. This is essentially because if others' reward depends upon your action, what they believe you will do is relevant for you to predict their behaviour. On the other hand, it has been suggested that simple forms of action understanding conform with Bayesian models of intention recognition [11], [12]. This means that our interpretation of others' actions is optimal, under the insight that others behave according to common sense. Taken together, these ideas yield the "social Bayesian brain" hypothesis, namely: our (Bayesian) brain a priori assumes that others are Bayesian too (i.e. others also learn about ourselves) [13]–[15]. In the context of mutual social exchanges, this implies that mentalizing may involve the update of recursive beliefs from the repeated observation of others' overt behaviour. From a modelling perspective, one can define optimal learning rules that are rooted in information theory and are specific to the sophistication of mentalizing agents (i.e., the depth k of their recursive beliefs). This is important, because one can now evaluate the added value of some form of mentalizing sophistication, in terms of its ability to decipher intentional behaviour. Critically, our k-ToM model predicts that the performance of agents engaged in competitive repeated interactions increases with their ToM sophistication [14].
We test these ideas in the following experiment: we had participants believe either that they were playing a competitive game with each other, or that they were performing a gambling task. In fact, in both conditions, participants were competing against artificial k-ToM agents with different ToM sophistication levels. Critically, the task-relevant information (available actions and correct/incorrect feedback), is identical in both framings. Our prediction is twofold: (i) the social framing of the task induces participants to mentalize and thus to engage in recursive inference, and (ii) domain-general learning heuristics that prevail in the non-social framing are vulnerable to artificial mentalizing agents (whose sophistication people cannot grasp). This implies that people should perform better in the social than in the non-social framing of the task, because artificial ToM agents would outsmart learners who do not engage in mentalizing.
Our analysis involved de-identified participants' data and was approved by the ethics committee of the Laboratoire d'Economie Expérimentale de Paris (LEEP, Paris Experimental Economics Laboratory). In accordance with the Helsinki declaration, all subjects gave an informed consent.
All statistical data analyses (including ANOVAs) were performed using the VBA toolbox (http://code.google.com/p/mbb-vb-toolbox/) [34]. Note: although we report summary statistics that are not corrected for multiple comparisons, we indicate the family-wise error rate threshold (FWER5%) when necessary.
Fig. 3 summarizes the group results on the behavioural performance in the main task. Overall, the pattern of mean performances follows our predictions.
Let us first consider the top-left panel of Fig. 3, which depicts the dynamics of the group mean cumulative earnings (averaged across opponents) for both framing conditions, overlaid on the chance 5% false positive rate threshold. One can see how the effect size unfolds over time. In particular, it is reassuring to see that participants' performance tends to reach statistical significance almost from the start of the experiment onwards. When summarizing the performance in terms of final earnings: people significantly win in the social framing (ū60 = 1.79, p = 0.008), whereas they significantly lose in the non-social framing (ū60 = −1.28, p = 0.047) despite positive earnings against RB in the non-social condition (cf. Fig. 3, bottom-left panel). The framing effect is even clearer on the top-right panel of Fig. 3, which depicts the dynamics of the difference in average cumulative earnings between framings. In brief, the framing effect becomes significant at about trial t = 15, and increases in size as time unfolds (to reach ū60 = 3.07, p = 0.002 at the end of the game). We refer the interested reader to Figure 2 in Text S1 for further information regarding the dynamics of condition-specific earnings.
Now, as one can see on the bottom-left panel of Fig. 3, participants' final earnings seem to depend upon both the framing and the opponent type. More precisely, in the social framing, participants seem to win against all artificial agents except 2-ToM (null earnings). In contrary, in the non-social framing, participants seem to lose against all mentalizing opponents, be even with 0-ToM, and win against RB. This view is largely consistent with results of the ANOVA on peoples' final earnings: In addition to the main effect of framing (F = 7.49, p = 0.007), participants' performance significantly decreases with the sophistication of their opponent (F = 6.96, p = 0.009), but show no interaction of framing and opponent (F = 0.89, p = 0.35). Including participants' performance in the seven secondary tasks (as well as their age and gender) as confounding factors in the ANOVA did not change these results.
When looking more closely at condition-specific effects (cf. Fig.3 bottom-right panel), we found that the opponent, against which participants' performance showed the strongest framing effect was 1-ToM (t = 2.9, p = 0.003; FWER5% = 0.0032). This makes sense, if we assume that peoples' effective ToM sophistication is higher (resp. lower) than 1-ToM in the social (resp., non-social) framing. Note that the mean performance in the control condition (RB) shows no difference between the social and non-social framings (t = 0.1, p = 0.43). This is important, because it implies that the difference in mean performance against 1-ToM is unlikely to be due to motivational or attentional confounds (which would also induce differences against RB).
At this point, we looked at inter-individual differences to strengthen our results' interpretation. First, we asked whether any inter-individual variability in peoples' performance could be explained by inter-individual differences in the seven secondary cognitive tasks. Interestingly, we found no significant effect on peoples' performance in the main task, irrespective of the task framing or the opponent's sophistication (see Text S1 for further details). This is important, because this implies that peoples' capability to outsmart artificial mentalizing opponents is not influenced by executive functions or empathy. Next, we asked whether idiosyncratic differences in motivational and/or attentional states could drive the inter-individual variability in our main task. We reasoned that if this was indeed the case, people who win more than others in the social framing should also win more in the non-social framing. We thus focused on the correlation between peoples' performance in the social and in the non-social framings. To begin with, we found no correlation between average performances in the social and non-social framings (r = 0.24, p = 0.23). Furthermore, when testing the correlation for each opponent's sophistication separately, we found that it was significant only in the control condition (r = 0.48, p = 0.0100, FWER5% = 0.0102). Recall that RB is the only opponent, against which mentalizing should yield no advantage. Against other opponents, differences in performance induced by individual variability in attentional or motivational states are negligible, when compared to, e.g., differences induced by peoples' ToM sophistication. In brief, the inter-individual variability of peoples' performance against artificial mentalizing agents is unlikely to be driven by cognitive requirements (such as behavioural flexibility, working memory, inhibitory control, etc…) or attentional/motivational confounds. Rather, our analysis of peoples' earnings seems to indicate that peoples' ability to reliably predict the behaviour of artificial mentalizing agents critically depends upon whether or not they engage in (potentially automatic) sophisticated ToM inferences.
Next, we asked whether we could find evidence for framing-specific learning rules that could explain the observed differences in peoples' performances across framings. We thus performed Volterra decompositions of peoples' trial-by-trial choice sequences, i.e. we looked at how much trial-by-trial variance in peoples' choice sequences can be explained by the history of both players' actions.
Average Volterra's fit accuracy in each of the 4×2 conditions is given in Table 3 below. One can see that Volterra decompositions of participants' and artificial ToM agents' choices have similar fit accuracies. More precisely, they yield about 75% of correct choice predictions, which is significantly above chance level. This is a prerequisite for interpreting the estimated Volterra kernels as a summary of participants' average response to the history of players' actions.
Fig. 4 depicts the group mean Volterra kernels against each opponent, in the social and in the non-social framing condition. For each opponent, we superimposed the Volterra kernel of the corresponding "best k-ToM response", i.e. one ToM sophistication level above participants' opponents. For completeness, results of a parametric Volterra decomposition are exposed in Figure 5 of Text S1. In the non-social framing, it seems that people have a strong tendency to imitate their opponent's last action (cf. positive Volterra weight ). They also tend to perseverate, i.e. to reproduce their last choice (cf. positive Volterra weight ). In the social condition, people rather seem to alternate their own actions (cf. negative Volterra kernels ) and to imitate their opponent's choices less often than in the non-social framing (cf. small Volterra kernels ). In addition, Volterra decompositions of peoples' choice sequences in the social framing seem much closer to the "best k-ToM response" than in the non-social framing (except maybe in the control condition).
First, we consider the impact of our experimental factors onto peoples' response to feedback history. The ANOVA on peoples' Volterra kernels confirms that both weights and significantly decreased in the social framing, when compared to the non-social framing (: F = 6.6, p = 0.01; : F = 13.7, p = 0.0003). Also, peoples' response to their opponent's past actions shows a main effect of opponent. More precisely, participants' tendency to replicate their opponents' actions decrease with the sophistication of their opponent (: F = 11.5 p = 0.001, : F = 6.8 p = 0.01). Note that there was no significant interaction between framing and opponent on Volterra weights (irrespective of the lag). This is interesting, because this means that our experimental factors have a similar effect on behavioural performance and on peoples' response to feedback history. Moreover, the observed change in Volterra kernels is consistent with the idea that peoples' effective ToM sophistication increases in the social framing, when compared to the non-social framing. This is because Volterra weights of mentalizing k-ToM agents are systematically smaller than those of 0-ToM (cf. Fig. 1).
Next, we focus on the similarity to the "best k-ToM response", which we take as a proxy for the optimality of peoples' response. We measured the correlation between each participant's Volterra kernel and the appropriate "best k-ToM response" in each of the 4×2 conditions. This analysis is summarized on Fig. 5. One can see that the optimality score seems to mimic peoples' final earnings (cf. Fig. 3, bottom panels). In fact, people's optimality significantly correlated with their final earnings (r = 0.25, p = 0.0001), even after having removed the effect of the experimental factors (p = 0.002). We then performed an ANOVA on the Fisher-transformed correlation coefficients. Results showed that people's optimality significantly increased in the social framing, when compared to the non-social framing (F = 5.62, p = 0.02), and significantly decreased with the opponent's sophistication (F = 18.5, p = 0.0001). There was no significant interaction (F = 0.126, p = 0.723). Taken together, these results suggest that the effect of our experimental factors onto behavioural performance is mediated through peoples' similarity to the "best k-ToM response". A classical Sobel test [37] confirmed this for both framing (p = 0.010) and opponent (p = 0.013) factors.
In summary, our analysis of Volterra kernels demonstrates that the social framing induces a systematic change in peoples' behavioural response to feedback history. Importantly, this change is reminiscent of sophisticated meta-Bayesian inference, i.e. peoples' similarity to the "best k-ToM response" increases in the social framing, when compared to the non-social framing. This eventually drives peoples' behavioural performance against artificial mentalizing agents.
Lastly, we performed a formal model-based analysis of peoples' trial-by-trial choice sequences, in the aim of identifying the most likely learning scenario in both social and non-social framings. In brief, we performed a group-level random-effect Bayesian model comparison (RFX-BMS, [36]) of fourteen different models (cf. Table 2). These include meta-Bayesian ToM models (1-ToM, 2-ToM and 3-ToM), non-Bayesian ToM models (1-Inf and 2-Inf), Bayesian no-ToM models (0-ToM, hBL, 1-BSL, 2-BSL and 3-BSL), as well as non-Bayesian no-ToM models (RL, WSLS, Nash and Volterra decompositions). In what follows, we will exploit these two orthogonal partitions of our model set, namely: T+/T- (which refers to models that include mentalizing or not) and B+/B- (which refers to models that rely upon Bayesian belief updates or not). Note that all models include a bias term that can capture a systematic tendency to prefer one alternative option over the other (within games/sessions). First, we performed Bayesian hypothesis tests to assess the stability of models attribution across conditions. To begin with, we tested the hypothesis that the model family (T+ versus T-) used in the social framing was the same than in the non-social framing, for each opponent. Evidence for the null hypothesis was found for the control condition RB (EP = 95%). However, evidence for a difference in model families across framings was found for both 0-ToM (EP = 23%) and 1-ToM (EP = 0%) opponents. The test was inconclusive for 2-ToM (EP = 53%). Then, we tested whether the same family of model was used across opponents in a given framing. In this case, we found strong statistical evidence in favour of stability of model attributions. More precisely, the null hypothesis was strongly supported for all between-conditions comparisons (EP>83%), with the exception of comparisons between 2-ToM and RB in the social framing, which yielded weaker evidence (EP = 69%). Overall, this analysis indicates that people's learning rule is mostly framing-dependent (but not opponent-dependent). This motivates our final analysis, which essentially is a framing-specific RFX-BMS. The result of this procedure is depicted on Fig. 6, which shows the exceedance probability of model families in both the social and non-social conditions. We refer the interested reader to Text S1 for quantitative diagnostics of the RFX-BMS approach (cf. fixed-effect analysis and confusion matrices).
One can see that, in the social condition, peoples' trial-by-trial choice sequences are more likely to be explained by T+ models than by T- models (EP = 100%). In contradistinction, peoples' behaviour in the non-social condition is more likely to be explained by models that do not rely on mentalizing (EP = 96%). This is strong statistical evidence that any realistic mechanistic description of peoples' policy in the social framing has to rely upon recursive mentalizing processes. We then asked whether we could find more specific evidence regarding the information-theoretic nature of peoples' belief updates. Thus, we further divided our T+ and T- families into B+ and B- subfamilies. We then used RFX-BMS to perform a comparison of the two corresponding subfamilies (T-B-, T-B+ in the non-social condition T+B+ and T+B+ in the social condition. We found that T+B+ models were the most likely explanations to peoples' trial-by-trial choices (EP = 98%) in the social condition, whereas T-B- was the most likely family in the non-social condition (EP = 99%). This is important, because this means that mentalizing processes are likely to follow meta-Bayesian belief update rules (as opposed to other non-optimal heuristics). In other terms, the way we learn about how others learn is near-optimal (from an information-theoretic point of view).
Let us now focus on the estimated models' frequency distribution in the social condition (cf. upper panel of Fig. 7). First, one can see that 2-ToM is the most prevalent model (well above reference models such as Nash or RL). Second, we restricted the model comparison to the T+B+ family, in the aim of deriving efficient estimates of the distribution of ToM sophistication in the human population. We found that 2-ToM agents are about two times more frequent than 1-ToM agents (3-ToM being almost negligible). This suggests that the natural inter-individual variability of ToM sophistication exists but is rather narrow. In addition, it is likely to be upper-bounded.
For completeness, Fig. 7 also shows the equivalent estimated models' frequency distribution in the non-social condition (cf. bottom panel). One may infer that WSLS is the most likely explanation for peoples' behaviour in this condition. However, it turns out that RFX-BMS may confuse Bayesian sequence learning with WSLS (more precisely: 2-BSL or 3-BSL). Although such statistical confusion does not compromise the interpretation of other potentially likely models, it renders the comparison of the families T-B- and T-B+ slightly unreliable. Thus, the estimation of model frequencies within the winning family (T-B-) is provided only as an indication (see Text S1 for further details).
Our study combined a computational modelling approach with an experimental investigation of Theory of Mind (ToM) in a situation of social interaction. We demonstrated a strong social framing effect, whereby the ability of participants to predict the behaviour of artificial mentalizing agents was conditional on whether or not they believed they were playing against another human being. Using data-driven analyses, we showed that this social framing effect was due to a difference in peoples' trial-by-trial response to feedback. In addition, we found that our meta-Bayesian model is a more plausible explanation of people's trial-by-trial choice sequences than other non-Bayesian and/or non-social (non-mentalizing) learning heuristics only in the social condition. Finally, we found statistical evidence that ToM sophistication is variable across people, and is likely to be upper-bounded (2-ToM).
Recall that our experiment aimed at revealing the specificity of social inference indirectly, by simulating behavioural data that conform to peoples' natural prediction of others' actions, and then measuring a difference in performance that originates from the task framing. Here, the framing induces priors that determine how people process the feedback information, which shapes their predictions regarding the next best move. Critically, such a manipulation only works if (i) the underlying model realistically simulates peoples' hidden social prior beliefs, and (ii) people are unlikely to appeal to these priors in the non-social framing. In our case, social priors essentially induce a sophisticated interpretation of the game's outcome, which involves mentalizing about others' beliefs. In turn, people engage in recursive belief updates, which we claim is very specific to human social interactions. To support this claim, we have provided two complementary pieces of evidence: (i) people could win over sophisticated (artificial) mentalizing agents only in the social framing condition, and (ii) the most likely explanation for people's trial-by-trial choices involves mentalizing only in the social condition. Note that the qualitative change in people's perspective induced by the framing is confirmed by the short debriefing we conducted at the end of the main experiment. In brief, most participants reported "having tried to adapt their strategy to their opponent's" in the social framing, whereas they were "looking for feedback temporal patterns" in the non-social framing. Some participants even reported that they perceived well that hiders were "responding to their own choices", whereas slot machines "followed complex, predetermined, sequences". Taken together, these results validate our meta-Bayesian model of mentalizing in repeated social interactions.
Perhaps the most shocking result of this work is the fact that people are clearly fooled by mentalizing (artificial) agents in the non-social condition. This happens despite repeated negative feedback that signals persistent prediction error. Note that this does not mean that people disregard this prediction error in the non-social condition; however, prediction error does not serve to learn the relevant variables. Our analyses suggest that the non-social framing of the task induces implicit priors that obscure the evidence for intentional behaviour. This is important, because this may explain why we engage in mentalizing as soon as we interact with social agents [1].
Note that one could argue that with sufficient training, participants would eventually learn the best response to their opponent, without having to mentalize. This is in principle possible, since k-ToM agents are reducible (up to 80% accuracy) to a linear convolution of competing players' actions (cf. Volterra decompositions in Fig. 1). However, there is hardly any sign of performance improvement over the entire session duration (cf. Figure 2 in Text S1).
A slightly more severe criticism of our interpretation of the social framing effect appeals to some form of systematic order effect between the social and the non-social conditions (the former was always performed after the latter). An example of this is [12], which shows that, e.g., pedagogical learning is facilitated when people are primarily engaged in teaching others. In our context, such order effect could not be driven by training or priming, which would rather improve peoples' performance in the non-social condition. In other words, our current (imbalanced) design could detect a net performance decline from the social to the non-social condition, above and beyond potential training and/or priming effects. Note that order effects could also be due to the impact of cognitive fatigue. Under the assumption that mentalizing is an effortful mental activity, one could argue that people may be less motivated to engage in sophisticated mentalizing in the second (non-social) condition, which would lead to performance losses. We will discuss motivational confounds below.
Even more problematic is the concern that the social framing effect might be confounded by some trivial difference in the understanding of the task structure (as induced by, e.g., peoples' assumptions regarding the way casino slot machines work). In particular, this implies that participants might have performed better in the non-social condition, had they been "warned" about the existence of some form of hidden sophisticated rule. Instead, we chose to favour a balanced design that relied on rather non-informative instructions. Critically however, participants' answers to our debriefing questions seem to indicate that they were well aware of the existence of some structure in the feedbacks' sequence (cf. above). Note that model comparisons of participants' trial-by-trial choices in the non-social framing yield ambiguous evidence either in favour of simple heuristics like "win-stay/lose-switch" or in favour of more sophisticated Bayesian sequence learning schemes (cf. confusion matrix in Figure 10 of Text S1). In addition, our analyses show that non-ToM sophisticated learning models do not seem to provide a likely explanation for peoples' trial-by-trial choices in the social condition. This means that sophisticated inferences induced by the social framing were specifically stemming from adopting the intentional stance [38], i.e. they assumed that the feedback sequence was the (potentially complex) result of their opponent's reaction to their own choices. Although this is certainly reassuring, we cannot entirely rule such potential confound out. We will address this potential design imbalance in forthcoming experiments.
Let us now briefly discuss potential attentional and/or motivational confounds. In brief, one could argue that the prospect of outsmarting some conspecifics (as opposed to some uninteresting machine) incites us to invest the mental effort required for performing sophisticated inferences (typically: mentalizing). In fact, our results rather speak against such attentional/motivational effects on peoples' performance (e.g., no framing effect against RB, no correlation between peoples' performance in the social and in the non-social framings…). In addition, we found no effect of framing on peoples' reaction times (see Text S1), which is surprising under such motivational interpretation (because one would expect people to respond faster in the social than in the non-social condition). In any case, such potential issues do not confound our main result, namely that one is unlikely to decipher intentional behaviour without a priori adopting the intentional stance [38].
Given the apparent added value of ToM sophistication, one might be surprised by its apparent limitation. In other words, one may wonder why evolution has not made all of us smarter. In fact, one can show that, in theory, competitive and cooperative social interactions induce both a lower and an upper bound on ToM sophistication [14]. Interestingly, the empirical estimate of the distribution of ToM sophistication levels (cf. bottom panel in Fig. 4) is very similar to the predicted equilibrium we derived from evolutionary game theory. Although this is certainly reassuring, it is yet unclear how such results would generalize over contexts that induce different incentives for sophisticated mentalizing. For example, the effort cost incurred when mentalizing in very complex settings might overcome the expected gain in performance. Thus, the cognitive process that yields the best complexity/accuracy trade-off might not involve ToM at all. This may explain why people tend to resort to rather heuristic behavioural policies in some complex social interactions. One can note however, that our upper bound on ToM sophistication (2-ToM) is consistent with results from behavioural economics regarding limited depth in strategic thinking. Experimental investigations of the cognitive hierarchy model, for instance, typically demonstrate that only a small proportion of people (around 20%) would exceed 2 steps of recursive thinking in strategic games (e.g., "beauty contest" games) [9]. Having said this, we would argue such strategic games are essentially different from our main task. This is because they monitor some form of explicit reasoning about others, whereas the time limitation on each trial of our main task rather reveals participants' intuitive "first guess" on their opponent (as is evident from peoples' short reaction times and the lack of effect of, e.g., working memory and inhibitory control on their performance in the main task). This relates to the current debate regarding the implicit/explicit dichotomy of mentalizing processes [39].
Let us now briefly discuss how novel or consistent our results are, when compared to to existing studies in both experimental psychology and behavioural economics. First, on the theoretical side, we bridged the gap between the literatures on strategic thinking in games [9], [10], [40], [41] and action understanding [11], [42], [43]. More precisely, we extended inverse planning models to situations of reciprocal social interactions, which may induce recursive beliefs. We also extended cognitive hierarchy models to repeated games, which may involve the (Bayesian) recognition of others' intentions and beliefs. The key point is that we can now mimic different sophistications of mentalizing. Second, on the experimental side, our results are consistent with the idea that learning in a social context relies on very specific cognitive processes, which are engaged for predicting others' behaviour (see, e.g., [22], [44]). In particular, previous neuroscientific studies have demonstrated that specific neural systems are activated when performing classical ToM tasks [45], [46] and during recursive thinking in games [22], [47]–[50]. In this context, our critical contribution was to demonstrate the added-value of (some form of) sophisticated mentalizing, in terms of its ability to decipher intentional behaviour. That is, we showed that, peoples' ability to predict goal-oriented choices critically depends upon whether they adopt the intentional stance [38] or not. This is not trivial, as one could think that domain-general learning heuristics could have performed well against mentalizing agents. Among the existing literature, the closest example to our work is [12], which shows that learners who know they are being explicitly taught (by a teacher) learn more from the data than when assuming otherwise. Taken together, our work and this recent study tend to contradict other existing studies that concluded that social learning (such as advice taking behaviour) was driven by non-specific reinforcement-like processes [44], [51]. Note however that no recursive learning models was considered for comparison purposes in these works.
Of course, our k-ToM model does not embrace all mentalizing processes. For example, it cannot be used to model how people "read others' mind" from low-level social signals such as eye gaze, bodily posture or facial expression [52]. Although it comprises the basic building blocks for modelling false beliefs (cf. beliefs about beliefs), it would still require some modification to capture the difference between people who pass and people who fail the false belief test [53] (but see [54]). We note that extending k-ToM in order to explain the various phenomena observed across the literature is well beyond the scope of the present study. We will pursue this in subsequent publications.
Finally, we would like to highlight a few promising applications of this work. Given the simplicity of the task that participants have to perform (namely: choosing between two alternative options, one of which is leading to a reward), one could argue that it could be used to address three aspects of mentalizing. First, one could assess its developmental aspect by quantifying the drift in ToM sophistication that occurs when we age. Second, our approach could be adapted to perform ethological inter-species comparisons of ToM sophistication (e.g. monkeys, great apes and humans). Third, in line with ideas from the emerging field of computational psychiatry [55], [56], one may wish to quantify pathological impairments of mentalizing in neuropsychiatric disorders such as autism or schizophrenia. We are currently pursuing these ideas. In these contexts, the main added-value of our approach lies in its ability to capture quantitative differences in ToM sophistication through its impact on behaviour, without being confounded by linguistic skills.
|
10.1371/journal.pcbi.1004445 | Task-Driven Activity Reduces the Cortical Activity Space of the Brain: Experiment and Whole-Brain Modeling | How a stimulus or a task alters the spontaneous dynamics of the brain remains a fundamental open question in neuroscience. One of the most robust hallmarks of task/stimulus-driven brain dynamics is the decrease of variability with respect to the spontaneous level, an effect seen across multiple experimental conditions and in brain signals observed at different spatiotemporal scales. Recently, it was observed that the trial-to-trial variability and temporal variance of functional magnetic resonance imaging (fMRI) signals decrease in the task-driven activity. Here we examined the dynamics of a large-scale model of the human cortex to provide a mechanistic understanding of these observations. The model allows computing the statistics of synaptic activity in the spontaneous condition and in putative tasks determined by external inputs to a given subset of brain regions. We demonstrated that external inputs decrease the variance, increase the covariances, and decrease the autocovariance of synaptic activity as a consequence of single node and large-scale network dynamics. Altogether, these changes in network statistics imply a reduction of entropy, meaning that the spontaneous synaptic activity outlines a larger multidimensional activity space than does the task-driven activity. We tested this model’s prediction on fMRI signals from healthy humans acquired during rest and task conditions and found a significant decrease of entropy in the stimulus-driven activity. Altogether, our study proposes a mechanism for increasing the information capacity of brain networks by enlarging the volume of possible activity configurations at rest and reliably settling into a confined stimulus-driven state to allow better transmission of stimulus-related information.
| Task- or stimulus-related changes of brain dynamics have been the subject of intense investigation during the last years. One of the most robust hallmarks of task/stimulus-driven brain dynamics, as measured using diverse recording techniques, is the decrease of variability with respect to the spontaneous level. This has led several researchers to focus on the second-order statistics of evoked activity and to study their functional consequences for information processing. In particular, it was observed that the trial-to-trial variability (related to variable responses to an identical stimulus from one presentation to the next) and the temporal variance of functional magnetic resonance imaging (fMRI) signals decrease in the task-driven activity. Here, we built a computational model of the whole brain to understand how local and large-scale brain dynamics contribute to these effects. The model allowed us to derive equations for the network statistics of both spontaneous and evoked activity. We observed that, as a consequence of single node and network dynamics, stimulus input impacts network statistics in such a way that the entropy of the stimulus-driven activity is lower than that during spontaneous activity. We confirmed this model prediction using empirical fMRI data and we further discuss its functional implications.
| How spontaneous brain dynamics are altered under stimulation or task conditions remains an important open question in neuroscience. Empirically, one of the most robust hallmarks of task-driven brain activity is the decrease of variability following an external stimulus input, a phenomenon observed across a variety of species, cortical areas, tasks, stimulus and attentional conditions, and using brain signals observed across multiple spatiotemporal scales including neuronal membrane potentials, neuronal firing rates, field potentials and functional magnetic resonance imaging (fMRI) signals [1–5]. A recent fMRI study showed that trial-by-trial variability of BOLD signals decreases following stimulus onset in a visual detection task and that the magnitude of variability reduction was correlated with the magnitude of trial-averaged response [3]. Moreover, the temporal variance of BOLD signals is significantly smaller during the same task as compared with the resting condition [6]—an effect that has also been reported in brain field potentials, neuronal membrane potentials, and neuronal spiking activity [7–9]. This suggests that the multidimensional space outlined by cortical activity is reduced following the stimulus onset [10].
Yet, a detailed mechanistic explanation of these effects is still lacking. In the present work we aimed to model the empirical observations of the fMRI study of [3], by studying the effect of external inputs on the first- and second- order statistics of a large-scale model of the brain [11]. This model is composed of N local E-I nodes, with one excitatory and one inhibitory neural subpopulations, representing N brain regions that are interconnected through an empirical large-scale connectivity matrix obtained using diffusion imaging data of healthy human subjects [12]. The dynamics of each of the E-I nodes follows the mean field equations derived by [13] and the excitatory firing rate is clamped around 3 Hz by adjusting the connection weight from the I population to the E population, a procedure known as Feedback Inhibition Control (FIC) [11]. This large-scale model has been shown to provide an efficient description of resting-state fMRI functional connectivity together with realistic stimulus-evoked activity [11]. Here we assumed that different tasks can be modeled by sets of inputs that co-activate different brain regions. Furthermore, we focused on synaptic activity, since it has been shown that BOLD signals relate to local field potentials (LFPs) more closely than to neuronal firing rates [14–17].
Using this model we observed that, as a consequence of single node and network dynamics, the application of an external input impacts the network statistics, so that the entropy of the stimulus-evoked activity is lower than that during spontaneous activity. We confirmed this model prediction using empirical fMRI data and further discussed its functional implications.
In this work, we examined how external stimulation impacts the statistics of both single E-I nodes and a large-scale model composed of interconnected E-I nodes. A priori, the nonlinear and stochastic nature of the dynamical equations hinders analytical progress, and the calculation of the network’s statistics relies on numerical simulations of the stochastic differential equations (SDEs), which are time consuming and subject to sampling issues. But, in the case of weak noise, one can linearize the stochastic fluctuations and derive deterministic differential equations for the network’s statistics. This so-called linear noise approximation is described in the Methods section. In the following, unless otherwise specified, we used this method to approximate the network’s covariances, autocovariances, and power spectral densities both in the spontaneous condition and when an external input is applied to the network.
We first evaluated the variability of the synaptic activity of single E-I nodes (Fig 1A). We calculated two types of variability: i) the variance across stochastic realizations of synaptic activity (trial-by-trial variance), noted σ2, and ii) the autocovariance (temporal variance) of synaptic activity, defined as the covariance of the synaptic activity with itself at pairs of time points and noted Fu(t+τ,t). Explicitly these statistics are given by:
σ2=Var[u]=〈[u(t)−〈u(t)〉]2〉,
(1)
Fu(t+τ,t)=〈[u(t+τ)−〈u(t+τ)〉].[u(t)−〈u(t)〉]〉,
(2)
where u is the synaptic activity and the angle brackets <.> denote the average over stochastic model realizations (i.e. the average over simulated trials). The autocovariance measures the strength of the influence of the past dynamics of the system on its future dynamics [its normalized version, Fu(t+τ,t)/Fu(t,t), which is insensitive to the absolute amount of fluctuation, is the autocorrelation function (ACF)]. The equations governing these statistics can be analytically calculated by assuming that the noise is sufficiently weak to allow for a linearized treatment of the fluctuations or linear noise approximation (see Methods).
We examined how the application of an external stimulus Iext to the E population (Fig 1A) changes the variability of synaptic activity. The stationary trial-by-trial variance of the synaptic activity under external input was compared to its stationary spontaneous level (Iext = 0), and the relative change was quantified by:
Δσ2(Iext)=100×[σ2(Iext)σ2(0)−1].
(3)
The relative change of the mean synaptic activity (Δm) was also computed:
Δm(Iext)=100×[m(Iext)m(0)−1].
(4)
Fig 1B shows that an external input monotonically reduces the trial-by-trial variance of the synaptic activity of both E and I populations, and increases the mean synaptic activity for both populations. In S1 Text we explicitly solved the equations for the variance and showed that the decrease of synaptic activity’s variance in response to an external input is determined by nonlinearities and connectivity parameters (see also S1 Fig).
Moreover, the external input reduces the spread of the autocorrelation function (ACF) of the synaptic activity of both E and I populations (Fig 1C and 1D). In conclusion, application of an external input attenuates the trial-by-trial fluctuations and shortens the temporal memory of the synaptic activity of an E-I local node.
We next evaluated the first- and second-order statistics of task-driven activity in a large-scale network composed of local E-I nodes interconnected through empirically derived anatomical connectivity (see Methods; see also [11]). The model has a single free parameter G that determines the strength of connectivity, called global coupling parameter (see Methods), which in the following is fixed to G = 2.15, this value falls in the range of G values (between 1–4.45) for which the model fits closely to the resting-state functional connectivity of fMRI data [11]. Given the previous results for an isolated node, we predict that external inputs to local nodes propagate through the dynamical system, reducing the trial-by-trial variance of other nodes in the network via direct or indirect pathways. Fig 2A shows the response of the large-scale network when eight brain regions receive an external input (equal to Iext = 0.02 nA). To simulate the results of [3], in which human subjects performed a visual detection task, the selected brain regions receiving external inputs are related to visual processing. Two observations can be made: First, as expected from the response of isolated nodes, trial-by-trial variance reduces under simulated task condition for nodes directly receiving external inputs (Fig 2B). Second, consistent with the above prediction, many nodes not directly receiving external inputs also exhibit trial-by-trial variance reduction upon external stimulation to (other nodes in) the network. Notably, the change of trial-by-trial variance with respect to the spontaneous activity (Δσ2) is negative for all nodes (Fig 2B) and Δσ2 is negatively related to the change of synaptic activity Δm (Fig 2C). This negative relation is consistent with the empirically observed negative correlation between the magnitude of variability reduction and the amplitude of evoked response in fMRI signals [3]. This relation is expected for a large variety of connectivity matrices, since it arises from the propagation of the stimulus to nodes separated by direct and indirect links. However, using synthetic connectivities with different levels of clustering, we found that the relation holds for connectivity matrices with low or intermediate clustering, as it is the case of human connectomes, but it breaks for excessively clustered connectivity matrices for which recurrent connections highly dominate (see S2 Fig).
The temporal dynamics of the model (Fig 2D) show that, during the application of the stimulus, the mean synaptic activity increases, while its variance decreases, and, after a period of relaxation of ~1–2 s, the system settles into a stable stimulus-evoked state. In the stationary spontaneous and stimulus-induced states, the power spectral density (PSD) of fluctuations of the system in the presence of stochastic perturbations can be calculated using the linear approximation (see Methods, Eq 26). The change of variance in the frequency domain is given by the relative change of the power spectral density (ΔPSD) in the task-driven synaptic activity with respect to the spontaneous condition, defined as:
ΔPSD=100×[PSDtaskPSDspont−1].
(5)
Interestingly, the effect of imposing an external input is different for different frequencies and, as a result of network interactions, the PSDs of the brain regions are differently affected by the external input (Fig 2E). For both excitatory and inhibitory units, most of the brain regions directly receiving the external input showed reduced power in frequencies lower than 100 Hz, with a maximal reduction at ~9Hz (9.65 Hz for excitatory units; 7.42 Hz for inhibitory units), but those not directly receiving external input showed increased power in frequencies below 0.9 Hz and decreased power in frequencies between 0.9 and 100 Hz, with a maximal reduction at ~9Hz (9.35 Hz for excitatory units; 8.42 Hz for inhibitory units). These results are consistent with empirical electrophysiological findings of prominent desynchronization in alpha/beta frequency ranges during task performance [18] and human ECoG observations of decreased power in <1 Hz range only in task-relevant brain regions [7].
To show that the above results are not specific to the particular hypothetical “visual” task, we produced a large set of hypothetical tasks, by imposing an external input (equal to 0.02 nA) to the excitatory population of 8 randomly selected brain regions. The negative relation between Δσ2 and Δm was found for all tested stimuli (Fig 2F). Interestingly, while the external stimulus highly impacts the covariances with respect to the spontaneous case (Fig 2G), with a tendency to increase them, it only slightly changes the correlations between nodes (Fig 2H). This indicates that functional connectivity amongst nodes, classically measured using correlation matrices, is not dramatically changed by imposing an external stimulus.
We next allowed the global parameter G to vary and observed that the above results are qualitatively the same for a large parameter space (within G = 1 and 3) (Fig 3), namely that Δσ2 and Δm are negatively related (Fig 3A), that the task-driven functional connectivity is very similar to the spontaneous functional connectivity (Fig 3B), and that the input prominently reduces the power of frequency fluctuations lower than 40 Hz (Fig 3C and 3D). Within this parameter range the model captures both the observed behavior of the stimulus-driven activity and the resting functional connectivity (as shown previously in [11]). In contrast, for G>3, the model correctly predicts the resting functional connectivity, but the behavior of the stimulus-driven activity is not consistent with the empirical observations.
We next calculated the dynamic change of the temporal variance (autocovariance) of the synaptic activity when an external input is applied to the large-scale model. The external input was applied at time t = 0 and lasted for 2 s. We used direct stochastic simulations of the network to estimate the time evolution of the autocovariance (Fig 4A). During the application of the external input, the temporal correlation length is reduced. To quantify this effect we calculated the characteristic time scale of the ACF, noted T95, given by the time lag at which its value is equal to 0.05 (i.e. 95% percent of correlation decay). T95 was calculated using the linear approximation (Eqs (21–24)) in stationary spontaneous and stimulus conditions (Fig 4B). We found that temporal correlations lasted more than twice as long in the spontaneous state than in the task state (for the excitatory synaptic activity: T95 = 290 ms vs. T95 = 140 ms; for the inhibitory synaptic activity: T95 = 170 ms vs. T95 = 30 ms). Hence, the temporal memory of the synaptic activity of the large-scale model is shortened after the stimulus onset.
Up to now we have focused on the dynamics of the synaptic activity. Because BOLD fMRI is widely used to study brain dynamics under both resting state and cognitive tasks, an important question pertains to whether the previous results apply to the dynamics of BOLD signals. To test this, we used a hemodynamic model to convert the total synaptic activity (the sum of excitatory and inhibitory synaptic activity) into BOLD activity. We used the Balloon-Windkessel model for the Hemodynamic response that describes the transduction of neural activity to BOLD changes, though non-linear dynamic equations of blood flow and deoxyhemoglobin content [19]. The model parameters were chosen as in [3]. Using this nonlinear model we found that an external stimulus input increases the trial-averaged BOLD activity, while reducing the averaged trial-by-trial variance of BOLD signals (Fig 5A), leading to a linear negative relation between the relative change of trial-averaged BOLD activity and the relative change of its trial-by-trial variance during the application of the external input (Fig 5B). However, the relative change of variance is positive for some of the brain regions (23 over 66). In the model, this is due to the low-pass filtering of the hemodynamic model, since the Balloon-Windkessel model acts as low-pass filter of the synaptic activity that passes frequencies under 1 Hz [20, 21]. As shown in Fig 2E, the stimulus-induced decrease of the synaptic variance is not negative for all brain regions for frequencies under 1 Hz. As a consequence, those brain regions for which the synaptic activity presents an elevation of the spectral power under 1Hz have a positive relative change of the variance of the BOLD activity (Fig 5C). The stimulus-induced reduction of the autocovariance (Fig 5D) is moderate for the same reason: the memory of the BOLD signal is highly dominated by the slow hemodynamic response.
We next investigated the functional implications of the change in network statistics induced by external inputs. To this end, we calculated the differential entropy H of the synaptic activity. The differential entropy is an extension of the Shannon entropy for a continuous random variable and it is related to the volume occupied by the continuous random variable. H can be easily calculated for a multivariate normal distribution, an assumption that is met in our case for the level of noise used in this study (S3 Fig). In such cases, H depends on the covariance matrix which can be calculated using the linear noise approximation (see Methods). We evaluated the differential entropy of the spontaneous activity and of the stimulus-driven activity for different model tasks determined by external inputs to a given subset of brain regions (Fig 6A). We found that external stimulation systematically reduces the entropy of the synaptic activity (Fig 6B).
We next asked how much entropy (or uncertainty) in the synaptic activity is explained by the intrinsic noise present at each node of the model. In other words, we asked how much uncertainty is produced by the dynamical system due to the intrinsic noise of each node propagating into the network. To answer this question we calculated the Kullback-Leibler divergence (KLD), also called relative entropy, between the distribution of intrinsic noise and the distribution of synaptic activity. Because the intrinsic noises are normally distributed with covariance Qn and the distribution of synaptic activity is normally distributed (for weak noise) with covariance Cv, the KLD can be calculated using Eq (32) (see Methods). We found that the relative entropy of the spontaneous synaptic activity is systematically higher than that of the stimulus-driven synaptic activity, indicating that in the spontaneous state the dynamical system adds more uncertainty to the intrinsic stochastic process than it does in the stimulated condition (Fig 6C).
Thus far we have considered that the intrinsic noise of each brain region is independent between nodes (i.e., Qn is diagonal). However, it is reasonable to think that during a task and even at rest different brain regions share some noise, possibly due to shared sensory/proprioceptive background inputs. We thus calculated the entropy and the relative entropy in the case of non-diagonal noise covariance matrices. As for the diagonal case, we found that the stimulus-driven synaptic activity has lower differential entropy and lower relative entropy than the spontaneous activity (S4 Fig). Thus, external stimulation reduces the entropy of synaptic activity even in the presence of common noise.
We tested the model prediction of higher entropy in the spontaneous activity than in the task-driven activity using empirical data from [3]. The data consists of fMRI time-series from 33 ROIs, covering five cortical networks, as well as the hippocampus, thalamus and cerebellum, acquired in 17 healthy subjects (see Methods). Each subject completed 8 fMRI runs, each lasting ~7 min, including 4 runs in resting-state conditions and 4 runs in a visual detection task condition. Each task run contains 20 stimulus presentations that the subject was to detect by pressing a button as quickly as possible. The inter-stimulus interval ranged from 17.3–30.2 s. First, for each subject and condition, we concatenated the time-series of the different runs and estimated the entropy using two methods: i) by assuming normality and using Eq (30), and ii) by using the Nilsson-Kleijn non-parametric estimator (see Methods). Using both methods, we found that the differential entropy in the resting activity is significantly higher than that of the task-driven activity (p<0.01, Wilcoxon signed-rank test) (Fig 7A). Second, we performed a time-resolved analysis in which the differential entropy was calculated using sliding windows of 5 frames (10.8 s) shifted in steps of 1 frame (2.16 s). Using Eq (30), we computed the time course of the differential entropy, averaged across subjects, in the task condition, Htask(t), and during rest, Hrest(t) (Fig 7B). The entropy values were referenced to the rest entropy H0 averaged across subjects and across time windows, i.e. H0=1T∑t=1THrest(t), where T is the total number of time steps in a run. When the task data was aligned to the stimulus onset, we found that the differential entropy significantly decreases after the stimulus onset (p<0.01, paired t-test) and, after ~8 s, it recovers its resting level H0 (the significant difference at –17.3 s is due to the previous stimulus).
Interestingly, the resting-state functional connectivity and the task functional connectivity were very similar, with differences in correlation coefficients ranging between ±0.02 (Fig 7C), a feature that is captured by the model (see Fig 2H).
The above results show that the task-driven synaptic activity has lower trial-by-trial variance, lower temporal variance, and lower entropy than the spontaneous synaptic activity. Altogether, this indicates that the space occupied by the synaptic activity is reduced when external inputs are impinging upon the network. To illustrate this effect, we represented the synaptic activity at a given time point or in a given trial as a point in the state space. Fig 8A shows the simulated synaptic activity of three brain regions in a three-dimensional space defined by the activity of these brain regions when no external input is applied (spontaneous condition) and when an external input is applied (task condition). The mean activity was removed for each brain region; thus, here, the activity represents deviations from the mean. In the spontaneous condition, the network explores a volume of the state space that is larger than the volume explored in the task condition. In the temporal domain, the space occupied by the synaptic activity is also reduced in the task condition compared to the spontaneous condition (Fig 8B). This is shown by plotting the synaptic activity of a given brain region i in the three-dimensional space, or Poincaré map, defined as the synaptic activity in three different time points t, t+τ, and t+2τ. The volume of the space in the Poincaré map outlined by the spontaneous synaptic activity is larger than that occupied by the task-driven synaptic activity.
We have shown that external stimulation to a large-scale brain model attenuates the synaptic fluctuations, increases the covariances, and reduces temporal memory in brain regions receiving the input directly or indirectly through the anatomical connectivity. Furthermore, we showed that the spontaneous activity has more entropy and more relative entropy than the task-driven activity. More entropy and more relative entropy means that the brain network produces a larger number of possible activity configurations that are not explained by the intrinsic noise. In other words, as shown in Fig 8 and in accordance with empirical observations, the multi-dimensional synaptic activity space is larger in the spontaneous state than under external inputs.
Reducing the space occupied by the synaptic activity, as a consequence of reducing the trial-by-trial and the temporal variability and increasing covariances, has relevant implications for information processing. It has been shown that temporal variance and entropy of BOLD signals change with chronological age and that young adults who are also faster and more consistent performers exhibited significantly higher brain variability across tasks [22–24]. In addition, the reduction of trial-by-trial variability is highly predictive of better performance [4]. These observations are likely complementary. Indeed, in the view of Information Theory, the mutual information between the brain activity and a given stimulus can be decomposed as the difference between the entropy of the full set of response patterns for all stimuli (total entropy) and the entropy conditioned to one stimulus (evoked entropy). Thus, there are two ways of increasing the information carried by the brain activity: by increasing the total entropy or by decreasing the evoked entropy. There is growing evidence that the entropy at rest is an upper bound of the total entropy, since stimulus-evoked patterns reoccur during spontaneous activity [10, 25, 26]. In other words, more variability at rest is associated with a larger repertoire of potential brain states and greater information capacity [27] while the ability to reliably settle in a stimulus-evoked brain state allows better transmission of the information about the stimulus.
Information Theory provides quantification of the amount of potential information that is available given the distribution of brain activity. How the brain decodes this available information is a topic of active research. Classification of multivariate fMRI patterns has been used to decode different stimuli or behavioral conditions from the fMRI signals [28–30]. In this context, the reduction of trial-by-trial variability under task would improve the discriminability of the fMRI multivariate patterns, which in turn improves the decoding performance. Moreover, if multivariate patterns have to be estimated using short time windows, as is likely during dynamical task processing, reducing the temporal correlations of the fluctuations would improve the estimation of the patterns (since the reduction of the autocorrelation leads to an increase of the effective number of independent samples within the time window). It is possible that the brain uses similar coding schemes to efficiently represent the incoming sensory information and evolving mental states, although exactly how such decoding schemes are implemented by neural systems remains an open question.
In the present study we focused on the dynamics of the synaptic activity to model the empirical BOLD fMRI signals. Concentrating on the synaptic activity is justified since it has been shown that BOLD signals relate more closely to Local Field Potentials (LPF) rather than neuronal firing rates [14–17]). As in previous studies of large-scale models [11, 21, 31], we converted the synaptic activity into simulated BOLD signals via a non-linear hemodynamic model, known as the Balloon-Windkessel model [19]. We found prominent stimulus-induced decrease of BOLD variance and a negative correlation between the relative change of trial-averaged BOLD responses and the relative change of trial-by-trial BOLD variance, as reported empirically by [3]. This supports a previous conclusion [3] that the observed BOLD variability reduction is unlikely to be an effect of the nonlinearities in the hemodynamic response but rather is likely due to the underlying synaptic activity.
Nonetheless, for some brain regions the simulated BOLD activity has slightly more variance in the stimulus-driven activity than in the spontaneous activity (Fig 5B). Using a voxel-wise analysis across the whole brain on empirical fMRI data, it was found that whereas some voxels showed increased variance after stimulus onset, none of them were statistically significant after correction for multiple comparisons (Figure 4 in [3]). Thus, whether the observation of task-induced increase in trial-to-trial variance in selected regions in our model has physiological importance remains to be seen. As mentioned above, the increase of variance in the present model is due to the low-pass filtering of the hemodynamic model that suppresses the fluctuations frequencies above 1 Hz. Indeed, consistent with previous empirical observations using human ECoG recordings [7], the present model shows a prominent task-induced decrease of synaptic variance for frequencies >1Hz, but, for frequencies <1Hz, this is mostly evident in directly activated brain regions only. This suggests that the present dynamic mean field model might be too simple to reconcile these two features and should be extended to consistently reproduce the change of spectral power in both synaptic and BOLD activities. Another alternative is that the hemodynamic model needs to be refined to completely describe the neurovascular coupling between the BOLD signal and the synaptic activity at different frequencies. Indeed, experimental evidence shows that BOLD fluctuations correlate with broadband LFP signals and that the alpha (8–12 Hz), beta (18–30 Hz), and gamma (40–100 Hz) LFP bands were informative about the spontaneous BOLD signals from an individual brain area [32].
The mechanism underlying stimulus-induced decrease of neural variability has been recently studied in theoretical works. Among the proposed mechanisms, spontaneous multi-stability has received much attention [5, 33, 34]. Under this scenario, the spontaneous activity of local neural networks with an underlying clustered connectivity is highly variable due to transitions through multiple spontaneous states. These transitions render the spontaneous activity more heterogeneous, but are suppressed when a stimulus stabilizes the network in a single evoked state and, as a result, the variability decreases in the stimulus-driven activity. This scenario naturally predicts an important feature of spontaneous activity, namely that the different spontaneous states are similar to the stimulus-evoked states [35, 36], a phenomenon reported in studies of neuronal membrane potentials and spiking activity at the microcircuit level [10, 25] and in resting-state fMRI studies at large-scale network level [26, 37–40]. By contrast, in the present study, the reduction of variability is due to single node synaptic dynamics (Fig 1) without the need of multi-stability originating from clustered connections. We showed that variance decrease results from nonlinearities and local E-I connectivity (see S1 Text). When the local nodes interact through long-range connections a pattern of stimulus-induced variance reduction is observed as a result of direct and indirect inputs—a phenomenon that is expected for a large variety of connectivities, as soon as large-scale recurrent connections do not strongly dominate (S2 Fig). The model for local nodes presented herein is a mean-field model that describes the mesoscopic dynamics of synaptic activity. This model can be extended by introducing multi-stability in the local dynamics, a direction that requires further investigation.
In the present data and model the functional connectivity is only slightly changed between rest and task. Several studies have reported high similarity between resting and task-related functional connectivity [37–40]; however, other studies have demonstrated reorganization of functional networks during task performance [41–43]. Brain dynamics might be engaged into task activity through diverse mechanisms. Here we modeled task-driven activity by imposing sets of inputs that co-activate different brain regions. There are other possible models for task effects on the brain, such as neuromodulation-mediated changes of network parameters that modify the neural excitability, the synaptic efficacy, or the gating of inputs. How these mechanisms alter the statistics of task-driven activity in a large-scale model have only recently been examined and awaits further investigation [44–46].
Finally, we here focused on the effect of imposing a stationary input to the large-scale brain model. A natural extension of the present work would be to study the effect of time-varying (sinusoidal) inputs and compute the frequency-dependent response function of different network statistics. Considering an input of small amplitude would allow to linearize the response and to study the eventual resonances. Moreover, these resonances may be partly determined by transmission delays, given by the experimental distance matrix between the different brain regions, a scenario that is not consider in the present work.
In conclusion, we have shown that the stimulus-driven shrinkage of cortical activity space can be understood as a property of mesoscopic dynamics embedded in large-scale brain networks, a property that has important implications for information processing.
This research was conducted in agreement with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and informed consent was obtained from all subjects before performing the study, in accordance with institutional guidelines. The study design was approved by the Human Studies Committee of Washington University in St. Louis and the local Ethics Committee of Lausanne University.
Blood-oxygen-level dependent (BOLD) fMRI data (4x4x4 mm3 voxels, TE 25 ms, TR 2.16 s) were acquired in 17 normal right-handed young adults (9 females, age 18–27 years) using a 3T Siemens Allegra MR scanner. All subjects gave informed consent in accordance with guidelines set by the Human Studies Committee of Washington University in St. Louis. Each subject completed 8 fMRI runs, each 194 frames (~7 min) in duration. They consisted of two alternating run types. The first run type was a resting-state study in which a white crosshair was presented in the center of a black screen. Subjects were instructed to look at the crosshair, remain still, and to not fall asleep. The second run type was a task study in which the identical crosshair was presented, but now it occasionally changed from white to dark gray for a period of 250 ms, at times unpredictable to the subjects, with an inter-stimulus interval of 17.3–30.2 sec. The subjects were instructed to press a button with their right index finger as quickly as possible when they saw the crosshair dim. This data set has been previously used in [3, 6, 47]. Thirty-three regions of interest (ROIs) covering five cortical networks—the attention, default-mode, motor, saliency and visual networks, as well as the hippocampus, thalamus and cerebellum were defined based on previous task-related functional neuroimaging studies. The preprocessing of the fMRI data and definition of ROIs are described in detail in [3].
We used the model of [11] to describe the global dynamics of the whole cortex. This model binds the dynamics of N local nodes, composed of excitatory—inhibitory subnetworks (E—I networks), through the underlying anatomical structure which is estimated using diffusion-imaging data from healthy human subjects. The stochastic differential equations of the model describe the time evolution of the mean synaptic activity of each local node (i.e., brain region) and there are given by:
ui(E)=I0,E+wEESi(E)+G∑jCijSj(E)−wEI,iSi(I)+Iext,i,
(6)
ui(I)=I0,I+wIESi(E)−wIISi(I),
(7)
ri(E)=ΦE(ui(E))=aEui(E)−bE1−exp(−dE(aEui(E)−bE)),
(8)
ri(I)=ΦI(ui(I))=aIui(I)−bI1−exp(−dI(aIui(I)−bI)),
(9)
dSi(E)dt=−Si(E)τE+(1−Si(E))γri(E)+βηi(E)(t),
(10)
dSi(I)dt=−Si(I)τI+ri(I)+βηi(I)(t),
(11)
where SiE,I denotes the average excitatory or inhibitory synaptic gating variable (i.e., fraction of open channels) at the local area i (i ∈ [1,…,N]). In Eqs 10 and 11 ηi(E)(t) and ηi(I)(t) are uncorrelated Gaussian noises and the noise amplitude at each node is β = 0.01. riE,I denotes the population firing rate of the excitatory (E) or inhibitory (I) population in the brain area i. The population firing rates are sigmoid functions (ΦI and ΦE) of the input synaptic currents to the excitatory or inhibitory population i is given by uiE,I. Synaptic currents are the sum of i) local currents within the local E—I networks, ii) excitatory currents from the other local nodes, and iii) external inputs Iext. The local currents in node i are the sum of constants inputs to excitatory and inhibitory populations, noted I0,E and I0,I, respectively, local excitatory-to-excitatory currents wEESi(E), local inhibitory-to-excitatory currents wEI,iSi(I), local excitatory-to-inhibitory currents wIESi(E), and local inhibitory-to-inhibitory currents wIISi(I). The weights of these local connections are given by: wEE = 0.21; wIE = 0.15; wII = 1; and the feedback inhibition weight,wEI,i, is adjusted for each node i so that the firing rate of the local excitatory neural population is clamped around 3Hz, whenever nodes are connected or not—this regulation is known as Feedback Inhibition Control (FIC) and the algorithm to achieve it is described in [11]. It has been shown that the FIC constrain leads to a better prediction of the resting functional connectivity and a more realistic network evoked activity [11]. Local E—I networks interact through excitatory connections given by the N-by-N anatomical connectivity matrix, noted C. The connectivity matrix is scaled by a single global parameter, G, that changes the network from weakly to strongly connected and determines the dynamical state of the system. As shown in [11] the model has one single stable fixed point of low firing activity in all cortical areas, for all values of G within the region where the FIC regulation can be achieved. For larger values of G, long-range interactions are too strong to be compensated by FIC and the activity diverges. Finally, Iext represents external stimulation for simulating task evoked activity: it is zero for all neural populations under resting state condition, and Iext>0 for those populations excited in the task condition.
The values of all parameters are taken from [11] and are presented in S1 Table.
Neuroanatomical structure was obtained using Diffusion Spectrum Imaging (DSI) data and tractography from five healthy right-handed male human subjects [12]. The grey matter was subdivided into 998 regions of interest (ROIs) which are grouped into 33 cortical regions per hemisphere (66 areas in total) according to anatomical landmarks (S2 Table). White matter tractography was used to estimate the fiber tract density connecting each pair of ROIs, averaged across subjects. Anatomical connectivity among the 66 cortical regions was calculated by summing all incoming fiber strengths to the corresponding ROIs of the target region, and dividing it by its region-dependent number of ROIs, resulting in a non-symmetric connectivity matrix. This normalization by the number of ROIs—which have approximately the same surface on the cortex, i.e. the same number of neurons—is required because neuronal activity is sensitive to the number of incoming fibers per neuron in the target region. As the dynamical model of one region already takes into account the effect of its internal connectivity (see below), the connection of a region to itself was set to 0 in the connectivity matrix for the simulations.
In the following we derive approximated equations for the statistics of the gating variables and the synaptic activity. To estimate the network’s statistics, we assume that the noise is sufficiently weak so that the state variables fluctuate around their mean value and, by linearizing the equations, we concentrate on linear fluctuations. In this way, we express the system of stochastic differential Eqs (6–11) in terms of the first- and second-order statistics of the distribution of synaptic gating variables: μi(m), the expected mean gating variable of a given local neural population of type m (where m = E or I) of the cortical area i, and Pij(mn), the covariance between gating variables of neural populations of type m and n of local cortical areas i and j, respectively. The statistics are defined as:
μi(m)(t)=〈Si(m)(t)〉,
(12)
Pij(mn)(t)=〈[Si(m)(t)−μi(m)(t)][Sj(n)(t)−μj(n)(t)]〉,
(13)
where the angular brackets <.> denote the average over realizations or “trials”. Note that, for the model, a “trial” means a realization of the system of differential Eqs (6–11). In vector form, the system of equations writes:
ddt(S→(E)S→(I))=(f(E)(S→(E),S→(I))f(I)(S→(E),S→(I)))+η→(t),
(14)
where S→={S→(E),S→(I)}={S1(E),…,SN(E),S1(I),…,SN(I)}, η→ is uncorrelated Gaussian noise, fi(E)(S→(E),S→(I))=−Si(E)τE+(1−Si(E))γΦ(E)(ui(E)), and fi(I)(S→(E),S→(I))=−Si(I)τI+Φ(I)(ui(I)) for i = 1,..,N.
In the following we use a linear approximation of the fluctuations. As shown in [11], Taylor expanding S→ around μ→=〈S→〉, i.e. Si(m)=μi(m)+δSi(m), up to the first order, we obtain the differential equations for the means of the gating variables and the covariance of the fluctuations around the mean. For the mean values:
dμi(E)dt=ddt〈S→i(E)〉=−μi(E)τE+(1−μi(E))γΦE(ui(E)),
(15)
dμi(I)dt=ddt〈S→i(I)〉=−μi(I)τI+ΦI(ui(I)),
(16)
where ui(m) is the mean input current to the neural population m = E,I of cortical area i, defined as:
u→=(u→(E)u→(I))=WS→+I→0+I→ext,
(17)
where W is a block matrix defined as:
W=[wEEIN+G.C−D(w→EI)wIEIN−wIIIN],
where C is the NxN anatomical matrix, G the global coupling parameter, IN is the NxN identity matrix, D(w→EI) is a NxN diagonal matrix containing the weights of the feedback inhibition wEI,i as diagonal elements, and I→0 and I→ext are the vectors containing the constant and external inputs.
Let P being the covariance matrix between gating variables S→. P is a block matrix defined as:
P=[P(EE)P(EI)P(IE)P(II)].
The differential equation of the covariance matrix is [11]:
dPdt=AP+PAT+Qn,
(18)
where the superscript T is the transpose, Qn is the covariance matrix of the noise, given by Qn=〈η→(t)η→(t)T〉, and A is the Jacobian matrix given by first-order partial derivative of the nonlinear function f with respect to each variable S, evaluated at μ→. A is a block matrix defined as:
A=[A(EE)A(EI)A(IE)A(II)],
where
Aij(mn)={∂fi(m)(μ→)∂Sj(n)}.
Note that the Jacobian matrix depends on the point μ→ at which it is evaluated.
The synaptic input variables u→ are a linear combination of the gating variables S→ and, thus, covariance matrix between synaptic input variables u→ is given by:
Cv=WPWT.
(19)
Knowledge of the Jacobian matrix and the stationary covariance gives the stationary autocovariance of the gating variables S→, defined as the covariance of the process with itself at pairs of time points and given as:
FS(t+τ,t)=〈[S→(t+τ)−μ→(t+τ)][S→(t)−μ→(t)]T〉.
(20)
In the stationary regime FS(t+τ,t) depends only on τ and is given by:
FS(τ)=eτAFS(0)=eτAP,
(21)
where the exponential matrix is defined as:
eτA=I+τA+12!(τA)2+13!(τA)3+…
(22)
The stationary autocovariance of the synaptic input variables u→ is, thus, given by:
Fu(τ)=WFS(τ)WT.
(23)
The autocorrelation function (ACF) of the i-th synaptic input variable is given by:
ACFi(τ)=Fu,i(τ)/Fu,i(0).
(24)
Finally, the power spectral density (PSD) of fluctuations around the fixed points is also determined by the Jacobian matrix. The cross-spectrum of the gating variables S→ is given as [11]:
ΠS(ω)=〈δS˜(ω)δS˜(ω)†〉=(A+iω)−1Qn(AT−iω)−1,
(25)
where δS˜(ω) is the Fourier transform of δS→(t) and the superscript † is the conjugate transpose. The cross-spectrum of the synaptic input variables u→ is, thus, given by:
Πu(ω)=〈δu˜(ω)δu˜(ω)†〉=WΠS(ω)W†.
(26)
The PSD of synaptic activity as a function of the frequency ω is given by the diagonal of ∏u(ω).
Note that the different network’s statistics (variances, covariances, and PSD) are determined by the Jacobian matrix A that depends on the state of the nonlinear system (the elements of the A are derivatives evaluated at μ→). Because the application of an external input changes the state of the system, therefore changing the derivatives, the network’s statistics are also changed. In other words, the nonlinear nature of the system renders the network’s statistics state-dependent.
In summary, to get the stationary network’s statistics we simulated the deterministic Eqs (15–18) and, once the stationary values of the mean synaptic gating variables (μ→), the covariance matrix (P), and the Jacobian matrix (A) were reached all other statistics were computed using Eqs 19–26. All differential equations used in the present study were solved using the Euler’s method with a time step equal to dt = 0.1 ms. The total number of simulation steps was 105, this simulation length ensures that the system reaches the stationary regime.
Once we have obtained the linear prediction of the covariance we can estimate the extent of all possible configurations of the network given by the differential entropy H, which expresses the entropy of a continuous variable with n-dimensional probability density function (p.d.f.) f, and writes:
H(f)=−∫Df(x→)lnf(x→)dx→,
(27)
where D ∈ ℝn is the support set of f, i.e., D = {x|f(x) > 0}. The entropy is related to the spread of the p.d.f., i.e., it relates to the volume occupied by a continuous random variable. The volume of the support set D is defined as:
Vol(D)=∫Ddx1dx2…dxn.
(28)
The volume of the smallest set that contains most of the p.d.f is approximately 2nH(f) [48]. Thus, low entropy implies that the random variable is confined to a small effective n-dimensional volume and high entropy indicates that the random variable is widely dispersed.
For a n-dimensional normal distribution (μ, ∑) with covariance matrix ∑, the differential entropy in bits is given by the following form [48]:
H=12ln[ (2πe)ndet(Σ) ]/ln(2)=n2ln(2)(1+ln(2π))+12ln(2)det(Σ),
(29)
where det(Σ) is the determinant of the covariance matrix. We also calculated the differential entropy for the fMRI time-series used in [3]. For these empirical data we used two different calculations of the differential entropy. The first measure assumes that the data follows a n-dimensional multivariate normal distribution (n = 33) and is given by, first, estimating the covariance matrix of the fMRI signals for each subject (averaged across runs of the same condition, rest or task), noted Σ^, second, calculating the determinant of Σ^ as the product of the k non-zero singular values (λ) to elude singularity, and, finally, calculating the entropy as follows:
H=k2ln(2)(1+ln(2π))+12ln(2)∑j=1kln(λj).
(30)
For 16/17 subjects we found that k = 29 for both rest and task. For only one subject we found that k = n = 33 for both rest and task. As a second measure we used the Nilsson-Kleijn non-parametric estimator that does not assume normality and calculates the differential entropy based on nearest neighbors of a sample set [49]. Both ways of calculating the differential entropy H gave very similar results: the values of H obtained using the two methods were highly correlated (rc = 0.91 for rest data and rc = 0.90 for task data).
Following [50], we defined the relative entropy as the Kullback-Leibler divergence between the intrinsic noise and the synaptic activity of the network. In its general form the Kullback-Leibler divergence between two distributions f and g is defined as:
KLD=∫flnfg.
(31)
The intrinsic noise and the synaptic activity are normally distributed (see S3 Fig) and, in this case, it can be shown that the relative entropy between the intrinsic noise and the synaptic activity writes [50]:
KLD(u→,η→)=12[trace(Qn-1Cv)−lndet(Cv)det(Qn)−2N]/ln(2).
(32)
The relative entropy can be seen as the amount of uncertainty that is produced by the dynamical system.
|
10.1371/journal.pbio.1001231 | Differentiation of the Lateral Compartment of the Cochlea Requires a Temporally Restricted FGF20 Signal | A large proportion of age-related hearing loss is caused by loss or damage to outer hair cells in the organ of Corti. The organ of Corti is the mechanosensory transducing apparatus in the inner ear and is composed of inner hair cells, outer hair cells, and highly specialized supporting cells. The mechanisms that regulate differentiation of inner and outer hair cells are not known. Here we report that fibroblast growth factor 20 (FGF20) is required for differentiation of cells in the lateral cochlear compartment (outer hair and supporting cells) within the organ of Corti during a specific developmental time. In the absence of FGF20, mice are deaf and lateral compartment cells remain undifferentiated, postmitotic, and unresponsive to Notch-dependent lateral inhibition. These studies identify developmentally distinct medial (inner hair and supporting cells) and lateral compartments in the developing organ of Corti. The viability and hearing loss in Fgf20 knockout mice suggest that FGF20 may also be a deafness-associated gene in humans.
| A large proportion of age-related hearing loss is caused by loss or damage to outer hair cells in the organ of Corti. The organ of Corti is a highly specialized structure in the inner ear that is composed of inner hair cells, outer hair cells, and associated supporting cells. Although we understand some of the mechanisms that regulate hair cell versus supporting cell differentiation, the mechanisms that regulate differentiation of inner versus outer hair cells are not known. One potential candidate is fibroblast growth factor (FGF) signaling, which is known to regulate the morphogenesis of many sensory organs, including the organ of Corti. In this study, we find that FGF20 signaling is required at a specific time during development to initiate differentiation of cells in the mouse lateral cochlear compartment (which contains outer hair cells and supporting cells, but not inner hair cells). In the absence of FGF20, mice are deaf, and lateral compartment cells remain undifferentiated and unresponsive to mechanisms that regulate the final stages of differentiation. These findings are significant given the importance of outer hair cells during age-related hearing loss. Our studies also suggest that genetic mutations in FGF20 may result in deafness in humans and that FGF20 may be an important factor for the repair or regeneration of sensory cells in the inner ear.
| Congenital hearing loss is one of the most common hereditary diseases, affecting 2–3 infants per 1,000 live births [1]. Acquired age-related hearing loss affects one-third of people over the age of 65 [2]. A large proportion of age-related hearing loss is sensorineural and is caused by loss or damage to outer hair cells (OHC) in the organ of Corti (OC) [3],[4]. The OC is the sensory transducing apparatus in the cochlea and is composed of one row of inner hair cells (IHC) and three rows of OHCs that are separated by two pillar cells (PCs) that form the tunnel of Corti. Each sensory hair cell is associated with an underlying supporting cell (SC). Although there has been progress in understanding mechanisms of hair cell (HC) and SC differentiation [5],[6], the cellular signals that specify the distinct phenotypes of cochlear IHCs and OHCs are not known [7].
Fibroblast growth factor (FGF) signaling has essential functions at several stages of inner ear development. In the embryonic day 9–10 (E9–E10) developing mouse, FGF3, FGF8, and FGF10 are essential for development of the otic vesicle [8]. These ligands signal through FGF receptor (FGFR) 2b in otic epithelium, and mice lacking Fgfr2b show impaired otic vesicle development [9]. At later stages of development, FGF signaling is required for morphogenesis of the organ of Corti. At E11.5, Fgfr1 is expressed in the ventromedial wall of the otocyst, the region that will give rise to the cochlea [10]. At E15, Fgfr1 expression is observed in the sensory epithelium of the developing cochlea [11],[12]. Conditional disruption of Fgfr1 in sensory epithelial progenitor cells (with Foxg1cre) resulted in a severe reduction in HC number, possibly due to reduced proliferation of progenitor cells [10]. A similar phenotype was also observed in organ cultures treated with FGFR inhibitors [11]. Fgf20 is expressed in the presumptive epithelial domain of the developing cochlea at E13.5 and antibody inhibition of FGF20 in cochlear organ culture resulted in fewer SCs and HCs [11]. These studies suggest that FGF20 might be the ligand for FGFR1 during the early growth and differentiation stages of cochlear development.
At later stages of organ of Corti development (after E15), inhibition of FGF signaling results in loss of PCs, suggesting an additional stage-specific role for FGF signaling [13]. Genetic and gene expression data suggest that this function is mediated by FGF8 signaling to FGFR3. Fgfr3 is expressed after E15.5 in undifferentiated postmitotic progenitor cells that are thought to have the capacity to form OHCs, Deiters' cells (DCs), PCs, and Hensen's cells (HeCs) [12]–[15]. Genetic disruption of Fgfr3 prevents the differentiation of PCs and the formation of the tunnel of Corti and results in deafness [13],[15],[16]. FGF8 is expressed in IHCs where it induces differentiation of PCs and formation of one row of OHCs through signaling to FGFR3 [10],[17],[18].
The mechanisms that regulate the formation of OHCs are particularly significant, given the importance of OHCs for hearing function and age-related hearing loss. Although mouse mutants lacking FGFR1 suggest a role for FGF signaling in OHC development [10], the underlying mechanisms regulating OHC development are not known. Here we generated mice lacking FGF20 (Fgf20βGal/βGal). We show that Fgf20βGal/βGal mice are viable, healthy, and congenitally deaf, specifically lack OHCs and outer supporting cells, and have patterning defects throughout most of the cochlear sensory epithelium. These studies show that the organ of Corti can be subdivided into developmentally distinct medial (IHCs and inner SCs) and lateral (OHCs and outer SCs) compartments that are under the control of distinct developmental programs. This model posits the existence of distinct progenitor cells that give rise to medial and lateral compartments of the OC.
To study the function of Fgf20 in vivo, we generated Fgf20 null mice in which exon 1 was replaced with a β-galactosidase gene (Fgf20βGal) (Figure S1). Homozygous Fgf20βGal/βGal mice were viable, fertile, and healthy. However, Fgf20βGal/βGal mice lacked auditory perception (no ear twitch response to loud noise) and had auditory brainstem response (ABR) thresholds greater than 40 db above controls in the 5–20 kHz range (Figure 1A). Histological sections of the adult inner ear of Fgf20βGal/βGal mice showed normal gross morphology of the temporal bone and cochlea (Figure S1E–G); however, the OC showed significant dysmorphology, with variability in the degree of disorganization along the length of the cochlea (Figure 1B). Some sections showed almost complete absence of sensory HCs and SCs, while other sections showed loss of OHCs and DCs. In contrast, wild type and heterozygous littermates showed normal cochlear organization, with one IHC, three OHCs, one inner and outer pillar cell (IPC, OPC), and three DCs (Figure 1B).
To identify spatial and temporal patterns of Fgf20 expression in the developing inner ear, we stained whole mount preparations for β-galactosidase (βGal) activity. In Fgf20βGal/+ embryos, βGal was first detected in the anterio-ventral region of the otic vesicle at E10.5, the region of the otic vesicle where sensory progenitor cells are located (Figure 2A) [19]. In histological sections of the otic vesicle, Fgf20-βGal was expressed within the domain of Sox2+ sensory progenitor cells at E11.5 (Figure 2B). At E14.5, the time of sensory cell specification, Fgf20-βGal was expressed in the Sox2+, p27+ sensory domain (Figure 2C,D), in an apical to basal graded expression pattern, similar to previously reported expression patterns for Fgf20 [11]. At postnatal day 0 (P0), a time when almost all sensory cells have completed differentiation, Fgf20-βGal was expressed throughout all inner ear sensory epithelia (Figure 2E and Figure S2). In the cochlea, Fgf20-βGal expression was restricted to SCs and was expressed in a graded medial to lateral pattern, with highest levels in the inner phalangeal cells (IPhC) and lower levels in PCs (Figure 2F and Figure S2A). Fgf20-βGal was also expressed in the vestibular sensory organs of the inner ear, including the maculae of the utricle and saccule and cristae of the semicircular canals (Figure S2B–F). Fgf20βGal/βGal mice did not show any vestibular dysfunction (unpublished data).
To further analyze the hair cell phenotype in the cochlea, we isolated whole cochleae at P0 and stained with phalloidin, as well as with antibodies against Myo6 and Calretinin (Figure 3 and Figure S3) [20],[21]. Wild type cochleae showed three rows of OHCs and one row of IHCs throughout the OC (Figure 3A). However, in the cochlea of Fgf20βGal/βGal newborn pups, the proximal base region contained only two rows of OHCs and one row of IHCs. In the middle and apical regions, patches of HCs were observed (Figure 3B). Such patches typically contained three rows of OHCs and two rows of IHCs, and there were no HCs in the regions between the patches. Finally, no distinct phalloidin or Myo6 positive HCs were present in the most apical region. Because HC differentiation progresses from the base to the apex of the cochlea, we sought to determine whether differentiation of HCs in the distal apex region of Fgf20βGal/βGal cochlea was delayed or whether HCs were lost. At P7, expression of HC markers in the distal apex of Fgf20βGal/βGal and Fgf20βGal/+ cochlea were comparable (Figure 3 and Figure S3), indicating that HCs were not lost but rather delayed in differentiation. Consistent with delayed HC differentiation, at E16.5, HCs were undifferentiated in the middle region of the cochlea of Fgf20βGal/βGal compared to Fgf20βGal/+ cochlea (Figure S3A,B).
To identify whether SCs were properly formed, we stained P0 cochlea with antibodies against Prox1 and p75. At P0, Prox1 is expressed at high levels in DCs and PCs, while p75 labels PCs and HeCs [22],[23]. In Fgf20βGal/+ cochleae, there were two rows of PCs, three rows of DCs, and one row of HeCs (Figure S3G). Immunolabeling of Fgf20βGal/βGal cochleae showed that all the SC types existed, although with dysmorphology. Similar to HC patterns, two rows of DCs and two rows of PCs were formed at the base region of Fgf20βGal/βGal cochlea (Figure S3H). In the middle region, where HCs were clustered in patches, SCs were formed in accordance with the hair cell pattern (three rows of DCs and two rows of PCs). No SCs were observed in the space between the sensory patches. Interestingly, within a patch, OHCs were surrounded by PCs and HeCs, as indicated by continuous p75 staining (Figure S3J). Unlike HCs, apical SCs were differentiated, as indicated by Prox1 staining (Figure S3G,H), suggesting that apical SCs may develop normally in the absence of FGF20.
Because of the delay in HC differentiation and patch formation, the total numbers of cochlear HCs were quantified at P4, a time when both Fgf20βGal/+ and Fgf20βGal/βGal cochlea appeared fully differentiated (Figure S3K). Surprisingly, although there were two rows of IHCs in patches and no IHCs in the region between patches, the total number of IHCs was the same in Fgf20βGal/+ and Fgf20βGal/βGal cochlea (680±18, n = 4 in Fgf20βGal/+ and 685±58, n = 3 in Fgf20βGal/βGal, p = 0.4). However, the number of OHCs was decreased by 70% in Fgf20βGal/βGal compared to Fgf20βGal/+ cochlea (2035±42 in Fgf20βGal/+ and 601±60 in Fgf20βGal/βGal, p<0.001) (Figure 3D). In addition, cochlear length was decreased by 10% in Fgf20βGal/βGal compared to Fgf20βGal/+ mice (Figure 3E). We also counted the number of SCs (DCs, OPCs, and IPCs) normalized to 100 µm intervals. Similar to the large decrease in the number of OHCs, the number of DCs+OPCs was decreased by 52% in Fgf20βGal/βGal compared to Fgf20βGal/+ cochlea (55±3, n = 6 in Fgf20βGal/+ and 26±4, n = 8 in Fgf20βGal/βGal per 100 µm, p<0.001) (Figure 3F). In contrast, the number of IPCs was decreased by only 15% in Fgf20βGal/βGal compared to Fgf20βGal/+ cochlea (19±1 in Fgf20βGal/+ and 16±1 in Fgf20βGal/βGal per 100 µm, p<0.01) (Figure 3F). Next, we compared the ratio of different cell types in Fgf20βGal/βGal and Fgf20βGal/+ cochlea. In Fgf20βGal/+ cochlea, the ratio of OHC/IHC was 3.4±0.3. However, in Fgf20βGal/βGal cochlea this ratio was decreased (by 62%) to 1.3±0.6 (p<0.001). Additionally, the ratio of DC+OPC/IPC in Fgf20βGal/+ cochlea was 2.9±0.1, but was decreased (by 45%) to 1.6±0.2 in Fgf20βGal/βGal cochlea (p<0.001). Interestingly, within the lateral and medial compartments, the ratio of DC+OPC/OHC (1.4±0.0 in Fgf20βGal/+ and 1.5±0.1 in Fgf20βGal/βGal, p<0.01) and IPC/IHC (1.6±0.1 in Fgf20βGal/+ and 1.2±0.1 in Fgf20βGal/βGal, p<0.001) was slightly increased (by 7%) or decreased (by 25%), respectively, in Fgf20βGal/βGal compared to Fgf20βGal/+ cochlea. These ratios indicate that absence of FGF20 primarily affects lateral compartment cells (i.e., OHCs and DCs).
Next we asked whether loss of the lateral compartment was due to loss of sensory domain progenitor cells. To do this, we labeled E13.5 or E14.5 cochlea for Sox2, a marker for sensory progenitors, or Jag1, a marker for Kölliker's organ [24],[25]. The expression pattern of Sox2 and Jag1 was comparable in Fgf20βGal/βGal and Fgf20βGal/+ cochlea at E13.5 and E14.5 (Figure S4A–F). Additionally, cell proliferation was comparable in Fgf20βGal/βGal and Fgf20βGal/+ E13.5 cochlea (Figure S4G,H), indicating that the sensory domain had formed normally. Next, we hypothesized that FGF20 may play a role in lateral compartment differentiation. To test this, we isolated E13.5 or E14.5 cochlea and treated explants with 1 µM FGF9 (which shows similar biochemical activity in vitro compared to FGF20) [26] beginning at E13.5, E14.5, E15.5, and E16.5. Control cultures were maintained in parallel, but did not receive FGF9. Cochlear explants were cultured in these media for 5 d and then stained with Myo7a antibodies (to identify HCs) [20],[27]. Untreated Fgf20βGal/+ explants showed normal patterning, with one row of IHCs and 3–4 rows of OHCs (Figure 4A,E and Figure S5A,E). Also, untreated Fgf20βGal/βGal explants showed the expected patterning defects (patches of HCs and SCs towards the apical cochlea) and loss of HCs (Figure 4C,G and Figure S5C,G). Notably, however, treatment of Fgf20βGal/βGal explants with FGF9 at E13.5 and E14.5 resulted in rescue of the cochlear phenotype, with the cochlea showing a normal and contiguous pattern of sensory cells and increased numbers of OHCs (162±53, n = 2 without FGF9 and 486±122, n = 3 with FGF9 treatment at E13.5, p<0.05, and 376±96, n = 3 without FGF9 and 725±100, n = 3 with FGF9 treatment at E14.5, p<0.01) compared to untreated explants (Figure 4C,D,I and Figure S5C,D,I). Finally, treatment of Fgf20βGal/βGal explants with FGF9 at E15.5 or E16.5 did not affect the number of HCs (168±0, n = 2 without FGF9 and 176±90, n = 3 with FGF9 treatment at E15.5, p = 0.9, and 238±40, n = 4 without FGF9 and 279±47, n = 4 with FGF9 treatment at E16.5, p = 0.3) (Figure 4G,H,I and Figure S5G,H,I), indicating that normal differentiation of lateral compartment cells requires active FGF signaling prior to E14.5. Patterning and numbers of SCs were also rescued by FGF9 treatment (640±49, n = 3 without FGF9 and 1001±39, n = 4 with FGF9 treatment at E14.5, p<0.001) (Figure 4J–N). BrdU labeling of these cultures indicated that treatment of Fgf20βGal/βGal cochlea with FGF9 did not induce renewed proliferation within the sensory epithelia (Figure S5J and K) at this stage, which indicated that FGF9 treatment functioned to induce lateral compartment cell differentiation into HCs and SCs. Also, treatment of Fgf20βGal/+ explants with FGF9 did not change the morphology or number of HCs or SCs, indicating that FGF signaling is not sufficient to induce ectopic HC or SC formation (Figure 4B,F,I,K and Figure S5B,F,I). Similar experiments were repeated with recombinant FGF20 protein with qualitatively similar results, although FGF20 was less active than FGF9 in this assay (unpublished data).
To determine whether the missing outer compartment cells were lost or were still present in an undifferentiated state, we stained whole cochleae of P0 pups with E-Cadherin antibodies, which marks lateral compartment cells at late gestational and postnatal stages of development [6]. In Fgf20βGal/+ cochlea, E-Cadherin labeled all lateral compartment cells including OHCs, DCs, and HeCs (Figure 5A, upper). Interestingly, in Fgf20βGal/βGal cochlea, E-Cadherin was highly expressed in the region between the sensory patches where there were no HCs or SCs (Figure 5A, lower), identifying these as potential lateral compartment cells. We also labeled specimens for Sox2 and with phalloidin. At P0, Sox2 labels all supporting cells, but at earlier stages of sensory domain formation (E14.5), Sox2 labels undifferentiated sensory cells [24]. We observed normal patterns of Sox2 expression in Fgf20βGal/+ cochlea. However, in Fgf20βGal/βGal cochlea, Sox2 was expressed both in supporting cells and in the regions between the sensory patches (Figure 5B arrows). We also examined the expression of p27, a marker of SCs and undifferentiated sensory progenitors [28]. The expression pattern of p27 was similar to that of Sox2, with high expression in cells in the region between the sensory patches (Figure 5C). Although the identity of the cells within these gaps in the sensory epithelium is not known, these expression studies suggest that these cells may be an arrested progenitor-like cell or a differentiated non-sensory cell. To determine whether the lineage precursors of these cells could be rescued, E14.5 Fgf20βGal/βGal cochlea explants were treated with or without FGF9 and co-stained for Sox2 and Prox1 expression after 5 d in culture. In explants not exposed to FGF9, the region between the patches was Sox2+; Prox1− (Figure 5D, left). However, following exposure to FGF9, these cells became Sox2+; Prox1+ (Figure 5D, right), indicating that the lineage precursors of these cells are undifferentiated sensory cells and that exposure to FGF9 induced their differentiation into lateral HCs and SCs, such as OHCs and DCs. This finding indicates that FGF signaling is required to induce differentiation of cells in the lateral cochlear compartment.
Next, we asked whether FGF20 functions to induce differentiation of a specific cell phenotype in the lateral compartment versus functioning as a gate to permit lateral compartment differentiation. To answer this question, we treated E14.5 cochlea explants with DAPT, a γ-secretase inhibitor, which inhibits the Notch signaling pathway [29]. At this stage of development, the Notch pathway functions to prevent SC differentiation into HCs [30]. In Fgf20βGal/+ explants treated with DAPT, the domain of IHCs and OHCs expanded at the expense of SCs, compared to untreated explants (Figure 6A and B). In Fgf20βGal/βGal explants, treatment with DAPT also expanded the IHC domain, similar to heterozygous explants treated with DAPT. However, the domain of OHCs in DAPT-treated Fgf20βGal/βGal explants was still smaller than the OHC domain of DAPT-treated Fgf20βGal/+ explants and also contained patches of undifferentiated sensory progenitor cells (Figure 6C and D). This finding indicates that DAPT treatment did not induce differentiation of otherwise FGF20-dependent precursor cells. Also, we observed a dramatic reduction of SCs following DAPT treatment of either genotype, indicating that all SCs were converted into HCs (Figure 6F and H). Together, these data support a model in which FGF20 functions as a permissive factor that is required to initiate lateral compartment differentiation before E14.5 (Figure 7). Without FGF20, lateral sensory cells remained in an undifferentiated state.
The mechanisms that differentially regulate formation of inner versus outer hair cells are poorly understood. In this study, we show that Fgf20βGal/βGal mice have a specific deficiency in the formation of OHCs and outer supporting cells that make up the lateral compartment of the organ of Corti. These observations suggest that FGF signaling may regulate the growth or differentiation capacity of a progenitor cell that gives rise to lateral compartment cells and that medial (IHCs and inner SCs) and lateral compartment development may be controlled by distinct mechanisms (Figure 7). Additionally, since Fgf20βGal/βGal mice are viable and healthy, but are congenitally deaf, FGF20 is likely a candidate gene for hereditary deafness in humans. Interestingly, the FGF20 gene is located on human chromosome 8p22–21.3 within the autosomal recessive non-syndromic hearing impairment locus, DFNB71 [31].
To identify the developmental time when FGF20 functions to regulate lateral compartment differentiation, rescue experiments were performed in which Fgf20βGal/βGal cochlea were placed in culture prior to differentiation (E13.5) and then FGF9 was added to the culture at different time points. These experiments showed that the lateral compartment differentiation defects in Fgf20βGal/βGalmice could only be rescued if FGF9 was added at or before E14.5. However, treatment with FGF9 at or after E15.5 failed to rescue the phenotype of Fgf20βGal/βGal mice. This is interesting because E14.5–15.5 corresponds to the time when sensory cell specification is completed and HC and SC differentiation begins [5],[32].
The changes in the cochlear epithelium that renders it non-responsive to FGF signaling after E14.5 are not known. Possibilities include loss of FGFR1 expression, uncoupling of FGFR1 to cellular signal transduction pathways, or loss of cofactors required for ligand activation of FGFR1. In lung development, a feed forward signaling loop couples FGF9 with Wnt/β-catenin signaling and maintenance of FGFR expression. Loss of Fgf9 resulted in loss of Fgfr1 and Fgfr2 expression and subsequent loss of responsiveness of explanted lung to exogenous FGF9 [33]. If a similar feed forward loop functions in the inner ear prosensory epithelium, loss of FGFR expression in Fgf20βGal/βGal mice could explain the loss of responsiveness to exogenous FGF after E14.5. However, in the inner ear, FGF20 continues to be expressed in IPhCs and at low levels in PCs until early postnatal ages (Figure 2). This suggests that FGF20 signaling may have additional roles in cochlear development. At P0, βGal staining indicates that Fgf20 is expressed at highest levels in the cochlear apex (Figure 2). Since differentiation of the apical cochlea is delayed in Fgf20βGal/βGal mice (Figure 3), FGF20 may function at later stages of development to promote sensory cell maturation.
Because damage or loss of OHCs is thought to be a major cause of sensorineural hearing loss, efforts to restore hearing in some patients with sensorineural hearing loss will require regeneration of OHCs. Understanding the changes that occur in sensory progenitor cells between E14.5 and E15.5 is important because they may provide clues about pathways required for reactivation of OHC progenitors in the adult or protecting OHCs from ototoxic damage. Although FGF20 signaling alone may not be sufficient to induce regeneration of OHCs, it may be required in combination with other signaling molecules. For example, in lung development, responsiveness of lung tissue lacking FGF9 can be restored by simultaneously treating Fgf9−/− explants with activators of Wnt/β-catenin signaling and with FGF9 [34].
The Fgf9 subfamily includes Fgf16 and Fgf20 [35]. Consistent with the conserved sequences within this subfamily, the biochemical activities of FGF20 are similar to that of FGF9 and FGF16 [26]. In vitro, FGF20 binds and activates c splice variants of FGFR1, FGFR2, and FGFR3, which are generally expressed in mesenchymal cells, and b splice variants of FGFR3, which are expressed in epithelial cells [26]. However, the phenotype of Fgf20βGal/βGal mice is most similar to that of Fgfr1 conditional deletion mutants, in which epithelial Fgfr1 was inactivated in the developing inner ear sensory epithelium with Foxg1cre [10]. The phenotypic similarities strongly suggest that FGFR1 is the physiological receptor for FGF20. Because, in vitro, FGF20 activates FGFR1c to a much greater extent than FGFR1b [26], the FGFR1c variant may be expressed in the developing cochlear sensory epithelium. Alternatively, unique cofactors within the cochlear sensory epithelium may allow FGF20 to activate FGFR1c.
The sensory epithelium of the mammalian cochlea cannot regenerate following ototoxic or noise damage; however, the avian and amphibian inner ear responds to ototoxic or noise-induced injury with a robust regenerative response that results in complete functional recovery [36]. The underlying mechanisms accounting for this difference in regenerative capacity are not understood. However, in principal, therapeutic reactivation of appropriate signaling pathways in the mammalian inner ear should be able to recapitulate the avian response, resulting in both functional repair and prevention of further pathology. Our observation that FGF20 functions as a permissive factor for lateral compartment differentiation suggests that FGF signaling may be a necessary factor for promoting inner ear regeneration. Additionally, zebrafish lacking FGF20 are viable and healthy, but have defects in their ability to regenerate damaged fins [37]. These observations suggest that FGF signaling, and specifically FGF20 or related FGFs, may be important factors for regeneration of a variety of tissues, including the inner ear. Inducible genetic systems in the mouse and the identification of signaling pathways that interact with FGF20 will be required to test the protective or regenerative potential of FGF20 in noise or ototoxic damaged mammalian inner ear.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Washington University Division of Comparative Medicine Animal Studies Committee (Protocol Number 20100223). All efforts were made to minimize animal suffering.
The Fgf20 targeting construct was made using recombineering methods as previously reported [38]. Briefly, exon1 of Fgf20 was replaced with a βGal–LoxP-neomycin-LoxP cassette to generate Fgf20βGal(neo)/+ mice. The neomycin gene was eliminated by mating with β-actincre mice to generate Fgf20βGal/+ mice. Fgf20βGal/+ males and females were crossed to generate Fgf20βGal/βGal mice. Genotyping was performed using PCR1: CTGCATTC GCCTCGCCACCCTTGCTACACT; PCR2: GGATCTGCAGGTGGAAGCCGGTGCGGCAGT; PCR3: GGCCTTCCTGTAGCCAGCTTTCATCAACAT primers, which amplify wild type (335 bp) and mutant (498 bp) PCR fragments as indicated in Figure S1A. Mice were maintained on a 129X1/SvJ;C57B6/J mixed background. Fgf20βGal/+ and Fgf20βGal/βGal mice were viable and fertile.
Mice were anesthetized by i.p. administration of ketamine (80 mg/kg) and xylazine (15 mg/kg), and maintained at 37°C throughout the testing. ABR testing was carried out in a single walled sound-attenuating room. Testing was similar to what has been previously described [39]. Briefly, stimulus presentation and data acquisition were performed with TDT System 3 equipment using SigGen and BioSig software (Tucker Davis Technologies). An ES-1 electrostatic speaker was placed 7 cm from the animal's right ear. Toneburst stimuli (5, 10, 20, 28, and 40 kHz) were 5 ms in length with a 1-ms rise/fall time. Stimuli were presented at decreasing intensities in 5 dB steps until Wave I was no longer observed. Auditory profiles were recorded using platinum subdermal needle electrodes (Grass Technologies) placed with the recording electrode behind the right pinna, the reference electrode at the vertex, and ground electrode in the skin of the back. Responses were amplified and filtered (X 100,000 and low filter: 100 Hz, high filter: 3,000 Hz) using a Grass P 55 preamp. Tonebursts at each frequency and intensity were presented 1,000 times. Stimulus levels were calibrated using SigCal (Tucker Davis Technologies) program with an ACO Pacific ¼ inch microphone placed where the mouse's ear would be located.
For detection of Fgf20 mRNA, E14.5 embryos were dissected and inner ear tissue was isolated. RNA was extracted with the RNeasy kit (Qiagen). cDNA synthesis used the SuperScript III First-Strand Synthesis System for reverse transcription–polymerase chain reaction (Invitrogen), following the manufacturer's protocol. Mouse Fgf20 mRNA levels were quantified using a TaqMan gene expression assay (ABI Mm00748347_m1). TaqMan assays were run in an ABI7500 fast real-time PCR machine.
Cochleae were dissected from P0 pups in PBS and fixed overnight in Mirsky's Fixative (National Diagnostics). For whole mount staining, samples were washed three times in PBT (PBS, 0.1% Tween-20) and incubated in βGal staining solution (2 mM MgCl2, 35 mM potassium ferrocyanide, 35 mM potassium ferricyanide, 1 mg/mg X-Gal in PBT) at 37°C until color reaction was apparent. Samples were washed in PBS fixed in 10% formalin and imaged under a dissecting microscope. For staining histological sections, samples were cryosectioned, washed with PBS, and incubated in βGal staining solution. Samples were washed in PBS, embedded, and photographed.
For adult histology, mice were sacrificed with an overdose of pentobarbital (200 mg/kg). Temporal bones were dissected free from the skull and broken in half to expose the cochlea, which was perfused via the round window with a solution containing 4% paraformaldehyde and 0.1% glutaraldehyde. Cochleae were further fixed in this solution overnight and then rinsed free of aldehyde with several changes of PBS. They were then post-fixed in 1% osmium tetroxide (30 min) and rinsed and dehydrated through a series of acetones. Tissues were then infiltrated and embedded in an epon-araldite mixture and polymerized overnight at 60°C. Ten adjacent, 4 µm thick sections were saved from the midmodiolar plane from each cochlea, counterstained with toluidine blue, and coverslipped.
For frozen sections, embryos were fixed with 4% paraformaldehyde overnight and washed with PBS. Samples were soaked in 30% sucrose and embedded in OCT compound (Tissue-Tek). Samples were sectioned (12 µm) and stored at −80°C for immunohistochemistry.
For total hair cell counting, P4 cochleae stained with phallodin were used because both Fgf20βGal/+ and Fgf20βGal/βGal cochleae were completely differentiated at this stage. For supporting cell counting, P0 cochleae stained with either phallodin or Prox1 were used. Because of the incomplete staining pattern of Prox1 from the base to the apex, we could not count all of the supporting cells. Instead, we counted more than 300 µm regions of the base, middle, and apex of the cochlea and normalized counts to 100 µm. Inner and outer hair cells were identified by location and morphology of phalloidin staining. Inner pillar cells were distinguished by location and morphology among Prox1+ cells. Deiters' cells and outer pillar cells were counted by exclusion of inner pillar cells from Prox1+ cells. Cell counting was performed using Image J software.
Embryonic mouse cochlear cultures were established as described previously [40] with minor modifications. In brief, cochleae from Fgf20βGal/+ and Fgf20βGal/βGal embryos at various ages (E13.5–E16.5) were dissected, to expose the sensory epithelium, in ice-cold M199 Hanks solution, transferred to a Ma-Tek dish (Ma-Tek Corporation), coated with Matrigel (BD Biosciences), and maintained at 37°C in vitro in experimental (FGF9 or DAPT) or control culture media for 3–6 d. Recombinant FGF9 and FGF20 protein was obtained from PeproTech Inc. DAPT was obtained from Sigma. To activate FGF signaling, FGF9 culture media (1 µg/ml FGF9+1 mM Heparin in MEM+10% FBS) was added to explant cultures at E13.5, E14.5, E15.5, or E16.5 for 6, 5, 4, or 3 d, respectively (until age of E19.5). Control culture media contained (1 µg/ml heparin in MEM+10% FBS). To inhibit Notch signaling, DAPT (N-[(3,5-Difluorophenyl)acetyl]-L-alanyl-2-phenyl]glycine-1,1-dimethylethyl ester) culture media was added to explant cultures at E14.5 for 5 d (E19.5). DAPT media: 10 µM DAPT (reconstituted in DMSO) in MEM+10% FBS. Control culture media contained DMSO in MEM+10% FBS. Cochleae were treated in pairs (i.e., cochlea from left ear received experimental media while the cochlea from the right ear of the same embryo received control media). The appropriate culture media was replaced every 24 h for all explants. Following incubation, explants were fixed in 4% paraformaldehyde for 30 min and analyzed by immunohistochemistry.
Immunohistochemistry was described previously [41]. Briefly, for whole mount immunofluorescence, cochleae were isolated and fixed in 4% PFA overnight at 4°C. Samples were washed with PBS and blocked with PBS containing 0.1% triton X-100 and 0.5% goat serum. Primary antibody was incubated overnight at 4°C. Samples were washed with PBS and incubated with a secondary antibody for 2 h at room temperature. Samples were washed, placed on a glass microscope slide, coverslipped, and photographed using a Zeiss LSM 700 confocal microscope. For section immunofluorescence, frozen sections (12 µm) were washed with PBS and blocked with 0.1% triton X-100 and 0.5% donkey serum. Sections were incubated with primary antibodies in a humidified chamber overnight at 4°C. Sections were then washed and incubated with secondary antibody for 1 h at room temperature. Samples were washed, coverslipped with Vectashield Mounting Media (Vector lab), and photographed using a Zeiss LSM 700 confocal microscope. Primary antibodies used: Phallodin (R&D Systems, 1∶40), Myo6 (Proteus Biosciences, 1∶500), Myo7a (Proteus Biosciences, 1∶500), Calretinin (Millipore, 1∶500), Prox1 (Covance, 1∶500), p27 (Neomarkers, 1∶500), p75 (Chemicon, 1∶500), β-Galactosidase (Abcam, 1∶500), Sox2 (Millipore, 1∶500, Santa Cruz 1∶200), BrdU (BD Biosciences, 1∶500), E-cadherin (Invitrogen, 1∶500), and Jag1 (Santa Cruz 1∶200).
Number of samples is indicated for each experiment. All data are presented as mean ± standard deviation (sd). The p value for difference between samples was calculated using a two-tailed Student's t test. p<0.05 was considered as significant.
|
10.1371/journal.pntd.0006972 | Emergence of Madariaga virus as a cause of acute febrile illness in children, Haiti, 2015-2016 | Madariaga virus (MADV), also known as South American eastern equine encephalitis virus, has been identified in animals and humans in South and Central America, but not previously in Hispaniola or the northern Caribbean. MADV was isolated from virus cultures of plasma from an 8-year-old child in a school cohort in the Gressier/Leogane region of Haiti, who was seen in April, 2015, with acute febrile illness (AFI). The virus was subsequently cultured from an additional seven AFI case patients from this same cohort in February, April, and May 2016. Symptoms most closely resembled those seen with confirmed dengue virus infection. Sequence data were available for four isolates: all were within the same clade, with phylogenetic and molecular clock data suggesting recent introduction of the virus into Haiti from Panama sometime in the period from October 2012-January 2015. Our data document the movement of MADV into Haiti, and raise questions about the potential for further spread in the Caribbean or North America.
| Madariaga virus (MADV) is the name given to what used to be called South American eastern equine encephalitis virus (EEEV), based on recent studies suggesting that MADV is distinct genetically from the EEEV circulating in North America. Until now, MADV has been found primarily in animals in South and Central America, with a limited number of human cases reported (most of whom had encephalitis). Our group has been responsible for a series of studies assessing the etiology of acute febrile illness (AFI) among children in a school cohort in Haiti. Unexpectedly, in April, 2015, we identified MADV on viral culture of plasma from a student with AFI in this cohort; an additional seven cases were identified on culture of samples from children with AFI in this same cohort in February, April, and May 2016. On sequence analysis, all strains were very similar genetically, and appear to have come from a strain introduced into Haiti from Panama sometime in the period from October 2012- January 2015. Symptoms of children were similar to those seen with dengue; none had encephalitis. Our data indicate that this virus, which has the potential for causing serious illness, has been recently introduced into Haiti, and raises the possibility that it might move into other parts of the Caribbean or North America.
| Madariaga virus (MADV), also known as South American eastern equine encephalitis virus, is an alphavirus in the family Togaviridae. Recent ecologic and genetic studies of eastern equine encephalitis virus (EEEV) have demonstrated clear separation between North and South American EEEV strains: North American EEEV cluster in a single genetic lineage (lineage I, in the system proposed by Arrigo et al. [1]), with South American EEEV strains, or MADV, clustering in EEEV lineages II, III, and IV. MADV can cause outbreaks in horses, and appears to infect a variety of mammals, including rats and bats, and possibly birds and reptiles [1–3]. However, less than a dozen human cases of MADV infection have been documented, and almost all were encephalitis cases seen as part of an outbreak in Panama in 2010 [3,4]. In population-based serologic surveys in Panama and the Peruvian Amazon, between 2 and 5% of the general population had evidence of prior infection [2,3,5], suggesting that mild or asymptomatic human infection is relatively common. In support of the latter hypothesis, we recently reported isolation of MADV from a child with acute febrile illness (AFI), but no evidence of encephalitis, during the Zika virus (ZIKV) epidemic in Venezuela [6]. The virus has not been previously recognized in Hispaniola or other parts of the northern Caribbean. We report here the apparent recent introduction of MADV into Haiti.
Our group maintains regular surveillance for AFI through a school clinic serving a cohort of approximately 1,250 children who attend schools at the four campuses within the school system operated by the Christianville Foundation (a U.S.-based non-profit organization): this includes the main campus, housing grades K-13 (School A), and three small satellite elementary school campuses (Schools B, C, and D)(Fig 1). Clinic services are free, and serve as the primary source of medical care for students [7]. Since May, 2014, all school children seen in the clinic with AFI (defined as a subjective history of fever and/or fever on presentation in a child with no obvious source of infection) have been asked to provide a blood sample for viral screening [8–10]. All clinical data are collected and recorded by the clinic physician or nursing staff as part of routine clinical care, with data extracted from clinical charts for analysis.
The University of Florida IRB and the Haitian National IRB have approved all protocols, and written informed consent was obtained from parents or guardians of all study participants.
Whole blood was collected into yellow top (Acid Citrate Dextrose/ACD) tubes (Becton, Dickinson, and Company, Franklin Lakes, New Jersey), and an aliquot used to prepare blood smears for microscopic analyses for malaria parasites. To obtain plasma for virology analyses, a portion of the collected blood was centrifuged to pellet red and white blood cells, and the resulting plasma (ca. 650μL) transferred to sterile screw-top vials and stored at -80°C pending tests.
As there was a possibility that viruses such as yellow fever virus or EEEV might be present in samples, RNA extractions and virology work were performed in the Lednicky BSL3 laboratory at the University of Florida’s Emerging Pathogens Institute, Gainesville, FL. RT-PCR tests for the detection of chikungunya virus (CHIKV), dengue virus (DENV), and ZIKV-genomic RNAs (vRNAs) in the plasma were accomplished as previously described [10,11]. Briefly, vRNA was extracted from virions in the plasma using a QIAamp Viral RNA Mini Kit (Qiagen Inc., Valencia, CA), and the extracted vRNAs tested using previously described RT-PCR primers [12–14].
To explore the possibility that the aforementioned viruses were present at levels too low to detect by RT-PCR in samples negative for CHIKV, DENV, and ZIKV vRNAs, or that other viruses were the causative agents, aliquots of plasma were inoculated onto monolayers of LLC-MK2, MRC-5, and Vero E6 as previously outlined [10,11]. In samples from eight patients with AFI, batches of inoculated cells formed virus-induced cytopathic effects (CPE) within 6 to 22 days that were reminiscent of the CPE observed for alphaviruses such as CHIKV: the infected cells developed dark, granulated cytoplasms with inclusion bodies, became enlarged, then either detached from the growing surface or appeared to undergo apoptosis (Fig 2; virus strain list, and cell culture information, included as S1 Table). RT-PCR tests for the detection of CHIKV, DENV, and ZIKV were performed on vRNAs extracted from spent cell-media [10], and all were negative for the viruses. Therefore, they were next tested with universal primer systems for the detection and identification of both alpha- and flaviviruses [15]. A very weak alphavirus amplicon was generated, though the putative alphavirus amplicon did not correspond in size to the alphaviruses identified by de Morais Bronzoni et al. [6,15].
Suspecting MADV, primers used in our previous work were used to screen samples by RT-PCR, and specific amplicons formed by primer pairs were sequenced [6]. At the same time, aliquots of spent cell media from four cell cultures that displayed advanced CPE were treated with cyanase nuclease (RiboSolutions, Inc., Cedar Creek, Texas), and vRNA thereafter extracted from the treated material [10]. Synthesis of complementary DNA was achieved as previously described [9] using non-ribosomal hexamers to favor the reverse transcription of viral genomes over ribosomal RNA [16]. PCR was subsequently performed with random hexamers and One Taq DNA polymerase (New England Biolabs). Various prominent amplicons purified from a 2% agarose gel stained with ethidium bromide were sequenced. Both approaches revealed that the virus isolated was MADV.
Sanger sequencing was performed on vRNA from spent cell lysates from four patients to obtain the MADV consensus sequences using methods similar to a previously published genome walking approach using overlapping primers (S2 Table)[6,9,10]; GenBank numbers are MH359230, MH359231, MH359232, and MH359233 (S1 Table). Because primers described in our previous MADV article [6] were suboptimal, the primer list in S2 Table depicts primers that were purpose-designed for work with these strains.
All available MADV full genome sequences were downloaded from Genbank and codons aligned using MUSCLE [17]. Nucleotide substitution saturation and phylogenetic signal were assessed using DAMBE6 [18] and IQ-TREE [19] respectively (see S1 Fig). Maximum likelihood (ML) phylogeny was inferred using IQ-TREE based on the best-fitting model (GTR+F+G4) chosen according to Bayesian Information Criterion (BIC). Strong statistical support along the branches was defined as bootstrap > 90% based on 2,000 replicates of Ultrafast Bootstrap Approximation [20]. The ML tree was used to check for temporal signal with TempEst [21]. A time-scaled phylogeny for the MADV isolates was then inferred with BEAST [22] v.1.8.4. by using the HKY85 nucleotide substitution model [23], empirical base frequencies, and gamma distribution of site-specific rate heterogeneity. Strict and uncorrelated relaxed clocks, as well as constant size and Bayesian Skyline demographic priors were compared. The best-fitting model was chosen by calculating the Bayes Factor (BF) of marginal likelihood estimates (MLE) of different models, inferred with path sampling (PS) and stepping-stone sampling (SS) methods [22,24,25]. The strength of evidence against the null hypothesis (H0) in favor of the more complex model (HA), is evaluated according to the following guidelines: lnBF<2 no evidence; lnBF = 2–6—weak evidence; lnBF = 6–10—strong evidence, and lnBF>10 very strong evidence [25]. The best model for MADV isolates was strict molecular clock and Bayesian Skyline demographic prior (S3 Table). A Markov Chain Monte Carlo (MCMC) sampler was run for 200 million generations, sampling every 200,000, and proper mixing of the MCMC was confirmed when Effective Sampling Size (ESS) values for the parameter estimates were >200 using TRACER from the BEAST package. Maximum Clade Credibility (MCC) tree was extracted after 10% burn-in using Tree Annotator from the BEAST package.
From May, 2014, through June, 2016, 484 children seen at the Christianville School clinic were diagnosed as having AFI, and had a blood sample collected. From May, 2014, through February, 2015 (initial time period), 252 children with AFI were seen. As previously reported [8], confirmed laboratory diagnoses in this initial time period included CHIKV (31% of children), DENV1 (9%), DENV4 (13%), ZIKV (2%), and Mayaro virus (0.4%). No infections with MADV were identified.
From March, 2015, though June, 2016 (later time period), 232 children seen in the school clinic were diagnosed as having AFI and had plasma samples collected for analysis. In this later time period we identified eight AFI case patients who were infected with MADV. Cell cultures inoculated with plasma from the eight patients demonstrated typical alphavirus-induced CPE, as described above and shown in Fig 2. Spent culture media from the eight cultures displaying CPE were positive for MADV vRNA by RT-PCR, whereas mock-infected cells maintained in parallel were negative for MADV vRNA. Complete MADV consensus genome sequences were obtained when sequencing was performed on vRNA from spent cell lysates from four of the eight patients. Despite the very limited amounts of plasma available for the patients after the analyses described above, we were able to go back to the original samples and demonstrate positive RT-PCR signals for MADV in plasma from five of the eight patients; of these five patients, two had had fever when seen in the clinic, while three did not. We hypothesize that failure to identify RT-PCR signals in the other three samples (all of which were culture-positive for MADV) reflects very low quantities of the virus in the samples.
The first MADV infection occurred in an 8-year-old girl in April, 2015; she was one of seven children with AFI seen that month. The remaining seven cases were in 2016, including three in February (27% of 11 AFI cases that month); one in April (3% of 31 AFI cases); and 3 in May (4% of 75 AFI cases [May was the peak of the ZIKV epidemic in Haiti and at the school]). Six of the children were from school campus A, with one case from each of two of the other campuses. Mean age of case patients was 9 years (range 3–12 years); six were girls and two were boys.
When seen in the clinic, four children had temperatures above 37 degrees C; maximum recorded temperature in the clinic for a MADV case patient was 39 degrees C. The other four children were afebrile at the time of the clinic visit, but they or their parents reported subjective fever prior to coming to the clinic. Five reported cough, four reported headache, three reported nonspecific abdominal pain, and one reported myalgia; none reported arthralgias. When compared with previously reported symptom profiles of children with AFI in this cohort [8], symptoms most closely resembled those seen with laboratory-confirmed DENV infection. Arthralgias were significantly more common in patients with CHIKV infection than in those with MADV (64% in CHIKV vs. 0% in MADV, p<0001, Fishers Exact test, two tail).
One patient, who had a prior history of seizure activity, had an episode described as syncope followed by a short period of confusion and amnesia before being brought to the clinic. On exam, one child had a rash; one had conjunctivitis; and one was noted to have swollen tonsils. Physical exams were otherwise unremarkable. All children recovered without apparent sequelae, with six of the eight having remained in the school system through the time of preparation of this manuscript (2018), with continued follow-up by the clinic.
The topology of the ML (S2 Fig) and MCC (Fig 3) phylogenies were in agreement. As noted previously, EEEV isolates cluster within four lineages: Lineage I constitutes the North American EEEV strains, while MADV fall into lineages II, III, and IV (Figs 3 and S2). The new isolates from Haiti cluster within Lineage III, which comprises isolates from Central and South America, and separate the Central American isolates from the Southern American ones, forming a Central American-Caribbean monophyletic sub lineage. The time of the most recent common ancestor (tMRCA) for the sub lineage was 1939, with a 95% High Posterior Density (HPD) interval of 1931–1948. Within this new sub lineage the new MADV isolates from Haiti cluster close to Panama isolates collected in 2010. The tMRCA for the MADV Haitian cluster was December 2013, with a 95% HPD interval of October 2012- January 2015, which corresponds to time window for the recent introduction of MADV in the island, likely from Panama. The evolutionary rate estimated for MADV was a 1.2 × 10−4 nucleotide substitution rate per year, in agreement with previous estimates calculated for this virus [1].
Reports of human infection with MADV are rare, although cross-sectional serologic studies (using plaque-reduction neutralization tests for confirmation) in Panama and Peru have reported seropositivity rates in human populations of between 2 and 5% [2,3,5], consistent with low-level MADV endemicity. Exposure may be substantially higher in epidemic settings, and/or with concurrent equine or animal epizootics: in recent work on seropositivity in household contacts of MADV and Venezuelan equine encephalitis virus cases during the MADV epidemic/epizootic in Panama in 2010, 19.4% of household contacts were seropositive for MADV [26]. In that same study, it was also noted that seroprevalence was comparable in all age groups, as might be expected if the virus had been recently introduced into Panama [26]. The phylogenetic analysis of our Haitian strains is consistent with recent introduction of MADV into Haiti, while our report of isolation of MADV from a child in Venezuela documents ongoing transmission in that country, concurrent with a possible equine epizootic [6]. Taken together, these observations are consistent with ongoing transmission/emergence of MADV at multiple sites in the Caribbean and South and Central America.
While most MADV case reports have involved patients with encephalitis, it is likely that the majority of infections occur in persons who are asymptomatic or who have only relatively mild disease [2,3,26]. This concept is supported by the previously noted work from our group in Venezuela [6], with identification of MADV (from a clade linked with Columbian and Venezuelan strains, distinct from our Haitian and Panamanian strains [Fig 3]) in a 12 year-old girl with headache and fever, rash, and conjunctivitis, but no evidence of encephalitis. In our Haitian study we saw a similar pattern of symptoms and signs, but with only one child with a rash, and only one with conjunctivitis. Carrera et al, in their study in Panama [3], used a pre-determined case definition for MADV of fever and headache. Only four of the eight patients in our study reported headache, suggesting that the Carrera case definition was overly restrictive. Interestingly, the pattern of symptoms in Haitian matched most closely with that previously reported from children in the same clinic with laboratory-confirmed DENV infections [8]; as with the DENV patients, the MADV patients were distinguished from the CHIKV patients by the lack of arthralgias. Overall, however, clinical presentation (in the absence of meningeal symptoms and signs) would appear to provide little assistance in diagnosing MADV infection.
Culex (Melanoconion) pedroi has been identified as a primary enzootic vector for MADV in the Amazon Basin [27]. The virus was also recovered from Cx. (Melanoconion) taeniopus in an epidemic outbreak in Panama [28], and in vector competence studies Aedes fulvus and Psorophora albigenu and Ps. ferox have been shown to be susceptible to and capable of transmitting the virus [29]. Cx. pedroi has not been previously identified in Haiti, but there are four known species in the Melanoconion subgenus including Cx. atrutus, Cx. carcinophilus, Cx. erraticus, and Cx. pilosus that are present [30], together with Ps. ferox, which is known to be a very aggressive biter of humans. It remains to be determined if these native Melanoconion subgenus mosquito species and/or Ps. ferox serve as vectors for MADV in Haiti. In an extensive survey by Vittor and colleagues [2] of possible reservoir hosts in Panama, evidence of infection was only found in rat species, with the highest seroprevalence in the short-tailed cane rat (Zygodontomys brevivauda; 8.7% seroprevalence, with one animal viremic for MADV) and the black rat (Rattus rattus; 3.9% seroprevalence). While Vittor found no evidence of infection in birds [2], there are suggestions in earlier studies that birds and reptiles can also be infected [1].
In Haiti, no prior data are available on MADV in vectors, animal reservoirs, or humans. While we cannot exclude the possibility that MADV was present in Haiti before the current case cluster, our phylogenetic studies are strongly suggestive of recent introduction of the virus into Haiti from Panama. Recent work by our group [8–11,31,32], and others, has underscored the apparent ease with which virus strains move among Caribbean and South and Central American countries. The drivers for this strain movement are varied. For Mayaro virus, we have shown a correlation between recent circulation of strains in this region and increased immigrant flow from Haiti to Peru and Brazil after the 2010 earthquake–and the counter-movement of peace-keeping troops from Brazil into Haiti during this same time period [32]. MADV is a little more complicated, as questions remain as to whether humans are a dead-end host, or if they can directly contribute to movement of the virus from one location to another. Over the past decade, there has been substantial movement of refugees from Haiti to and through Panama, as well as movement of Haitian workers to Panama; these population shifts may have provided an opportunity for movement of MADV from one country to the other. There is also the possibility that movement of strains was a function of movement of animal reservoirs (such as the black rat) on ships or in or on shipping containers; “hitch-hiking” of infected mosquitoes on airplanes is also a possibility [33]. At this point we know too little about the transmission and ecology of the virus to be able to predict its ability to move into other parts of the northern Caribbean, or areas such as Florida where North American EEEV is already endemic. Under these circumstances, the initiation of ongoing surveillance for MADV in humans, animals, and mosquitos throughout this region is clearly of public health importance.
|
10.1371/journal.ppat.1005916 | The Tax-Inducible Actin-Bundling Protein Fascin Is Crucial for Release and Cell-to-Cell Transmission of Human T-Cell Leukemia Virus Type 1 (HTLV-1) | The delta-retrovirus Human T-cell leukemia virus type 1 (HTLV-1) preferentially infects CD4+ T-cells via cell-to-cell transmission. Viruses are transmitted by polarized budding and by transfer of viral biofilms at the virological synapse (VS). Formation of the VS requires the viral Tax protein and polarization of the host cytoskeleton, however, molecular mechanisms of HTLV-1 cell-to-cell transmission remain incompletely understood. Recently, we could show Tax-dependent upregulation of the actin-bundling protein Fascin (FSCN-1) in HTLV-1-infected T-cells. Here, we report that Fascin contributes to HTLV-1 transmission. Using single-cycle replication-dependent HTLV-1 reporter vectors, we found that repression of endogenous Fascin by short hairpin RNAs and by Fascin-specific nanobodies impaired gag p19 release and cell-to-cell transmission in 293T cells. In Jurkat T-cells, Tax-induced Fascin expression enhanced virus release and Fascin-dependently augmented cell-to-cell transmission to Raji/CD4+ B-cells. Repression of Fascin in HTLV-1-infected T-cells diminished virus release and gag p19 transfer to co-cultured T-cells. Spotting the mechanism, flow cytometry and automatic image analysis showed that Tax-induced T-cell conjugate formation occurred Fascin-independently. However, adhesion of HTLV-1-infected MT-2 cells in co-culture with Jurkat T-cells was reduced upon knockdown of Fascin, suggesting that Fascin contributes to dissemination of infected T-cells. Imaging of chronically infected MS-9 T-cells in co-culture with Jurkat T-cells revealed that Fascin’s localization at tight cell-cell contacts is accompanied by gag polarization suggesting that Fascin directly affects the distribution of gag to budding sites, and therefore, indirectly viral transmission. In detail, we found gag clusters that are interspersed with Fascin clusters, suggesting that Fascin makes room for gag in viral biofilms. Moreover, we observed short, Fascin-containing membrane extensions surrounding gag clusters and clutching uninfected T-cells. Finally, we detected Fascin and gag in long-distance cellular protrusions. Taken together, we show for the first time that HTLV-1 usurps the host cell factor Fascin to foster virus release and cell-to-cell transmission.
| Human T-cell leukemia virus type 1 (HTLV-1) is the only human retrovirus causing cancer and is transmitted via breast feeding, sexual intercourse, and cell-containing blood products. Efficient infection of CD4+ T-cells occurs via polarized budding of virions or via cell surface transfer of viral biofilms at a tight, specialized cell-cell contact, the virological synapse (VS). The viral protein Tax and polarization of the host cell cytoskeleton are crucial for formation of the VS, however, only little is known about the link between Tax and remodeling of the cytoskeleton to foster viral spread. The actin-bundling protein Fascin has evolved as a therapeutic target in several types of cancer. Here, we show that Fascin is also crucial for release and transmission of the tumorvirus HTLV-1. Since Fascin is a transcriptional target gene of Tax in T-cells, our work provides a link between Tax’s activity and virus transmission. Visualization of cell-cell contacts between infected and uninfected T-cells suggests a role of Fascin in viral transmission potentially by facilitating the transport of viral proteins to budding sites. Thus, Fascin is not only crucial for metastasis of tumors, but also for transmission of HTLV-1 and is a new cellular target to counteract HTLV-1.
| Human T-cell leukemia virus type 1 (HTLV-1), which infects approximately 5–10 million people worldwide [1], is the only human retrovirus causing cancer: adult T-cell leukemia/lymphoma (ATL), a fatal neoplasia of CD4+ T-cells [2–4]. Further, HTLV-1 is the causative agent of a neurodegenerative, inflammatory disease, HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [5,6]. Both diseases can develop as a consequence of prolonged viral persistence in T-cells after a clinical latency of decades in 1–5% (ATL) or 3–5% (HAM/TSP) of infected individuals [7,8].
Activated CD4+ T-cells are the main and preferential target for HTLV-1 infection, but the virus is also present in very low amounts in other cell types including CD8+ T-cells, monocytes, and dendritic cells (DC) [9]. After binding to its receptor, which is composed of the glucose transporter GLUT-1, neuropilin-1 (NRP-1) and heparan sulfate proteoglycans (HSPGs) [10–12], HTLV-1 integrates into the host cell genome. The virus is mainly maintained in its provirus form (9.1 kb), which is flanked by long terminal repeats (LTR) in both the 5’ and 3’ region. In addition to structural proteins and enzymes common for retroviruses, HTLV-1 encodes regulatory (Tax, Rex) and accessory (p12/p8, p13, p30, HBZ) proteins [13]. HTLV-1 replicates either by infecting new cells or by mitotic division and clonal expansion of infected T-cells [14–16].
Efficient infection of CD4+ T-cells requires cell-cell contacts and coordinated steps of the virus infectious cycle with events in the cell-cell adhesion process. Thus, transmission of HTLV-1 occurs via breast feeding, sexual intercourse, and cell-containing blood products [9,17]. Unlike human immunodeficiency virus (HIV) or murine leukemia virus (MLV), cell-free transmission of HTLV-1 to T-cells is inefficient. Therefore, only a limited amount of poorly infectious viral particles is produced from infected lymphocytes and free virions can hardly be detected in infected individuals [18–20]. Thus far, two types of cell-cell contacts have been described to be critical for HTLV-1 transmission, tight cell-cell contacts and cellular conduits [21,22]. For transmission at tight cell-cell contacts, two non-exclusive mechanisms of virus transmission at the virological synapse (VS), a virus-induced specialized cell-cell contact [23], have been proposed [17]: (1) polarized budding of HTLV-1 into synaptic clefts [21], and (2) cell surface transfer of viral biofilms [24]. The latter consist of extracellular, concentrated viral assemblies that are surrounded by components of the extracellular matrix and cellular lectins [24]. Beyond, transmission via biofilms seems to be a major route of transmission since removal of biofilms by heparin treatment impairs cell-to-cell transmission by 80% in vitro [24]. Independent of the route of HTLV-1 transmission, viral particles are thought to be transmitted in confined areas protected from the immune response of the host in vivo. Moreover, cytoskeletal remodeling and cell-cell contacts are a prerequisite for all routes of virus transmission [21,25]. Although it is known that the viral protein Tax and polarization of the host cell cytoskeleton are crucial for formation of the VS and for HTLV-1 transmission (for details see: [17,23]), only little is known about the link between Tax and remodeling of the cytoskeleton to foster viral spread.
The regulatory protein Tax is essential for viral replication due to strong enhancement of viral mRNA synthesis by transactivating the HTLV-1 LTR (U3R) promoter. Beyond, Tax is a potent transactivator of cellular transcription and important for initiating oncogenic transformation [13]. Tax is also critical for HTLV-1 transmission since Tax cooperates with intercellular adhesion molecule 1 (ICAM-1), thereby inducing polarization of the microtubule organizing center (MTOC) at the VS [26] and thus, enhancing HTLV-1 cell-to-cell transfer. Furthermore, Tax enhances both actin- and tubulin-dependent transmission of virus-like particles (VLPs; [25]). However, only few host cell factors with a role in Tax-induced virus transmission have been characterized. Among those is ICAM-1, which is induced by Tax and cooperates with Tax in VS formation [26,27]. The Tax-induced small GTP-binding protein GEM enhances cellular migration, conjugate formation, and thus, is required for viral transmission [28].
In our search of novel target genes of Tax with a putative role in virus transmission, we have previously identified the evolutionary conserved actin-bundling protein and tumor marker Fascin as a new host cell factor strongly induced by Tax [29]. Fascin cross-links filamentous actin and stabilizes cellular protrusions, filopodia, and invadopodia [30]. Recent work shows that Fascin also interacts with microtubules to regulate adhesion dynamics and cell migration [31]. Fascin has evolved as a therapeutic target in several types of cancer since Fascin expression is associated with metastasis in malignant tumors and it correlates with clinical aggressiveness of some tumors [30]. In hematopoietic cells, Fascin is expressed in mature DC where it is important for stability of dendrites and for formation of the immunological synapse [32], while no expression of Fascin can be detected in unstimulated human T-cells [33]. We found that expression of Fascin is a common feature of chronically HTLV-1-infected T-cell lines. Fascin colocalizes with actin in the cytoplasm and at the membrane of HTLV-1-infected cells. Furthermore, knockdown of Fascin reduces the invasive capacity of HTLV-1-infected ATL-derived T-cells into extracellular matrix [29]. Since expression of Tax is sufficient to induce expression of Fascin [29,34] and Tax enhances actin-dependent virus transmission [25], we now asked whether Fascin affects HTLV-1 cell-to-cell transfer.
Here, we report that Fascin is crucial for release and cell-to-cell transmission of HTLV-1 in different cell model systems. While T-cell conjugate formation is Fascin-independent, cell adhesion of infected cells in co-culture with uninfected cells is impaired upon repression of Fascin. Imaging of Fascin and the viral gag protein at cell-cell contacts suggests a role of Fascin in transmission potentially by redirecting viral proteins to budding sites. Thus, Fascin as a major contributor to HTLV-1 transmission provides a link between Tax’s activity and virus transmission.
293T cells (kindly provided by Ralph Grassmann (deceased), FAU, Erlangen, Germany) were cultured in DMEM containing 10% fetal calf serum (FCS), L-glutamine (0.35g/l) and penicillin/streptomycin (Pen/Strep; 0.12g/l each). For selection of stable 293T cells carrying shRNAs, 4μg/ml puromycin was added to the media. The CD4+ T-cell line Jurkat (ATCC, LGC Standards GmbH, Wesel, Germany) from acute lymphoblastic leukemia was cultured in RPMI 1640M, Panserin, 10% FCS, L-glutamine and Pen/Strep [35]. The human Epstein-Barr virus (EBV)-positive B-cell line Raji derived from Burkitt’s lymphoma containing the surface receptor CD4 (Raji/CD4+) was a kind gift from Vineet N. Kewal Ramani (NIH, Frederick, Maryland, USA) and was cultured in RPMI 1640M, Panserin, 10% FCS, L-glutamine and Pen/Strep containing 500μg/ml geneticin to ensure retainment of the CD4 receptor [36]. The HTLV-1 in vitro transformed CD4+ T-cell line MT-2 [3] and the ATL-derived CD4+ T-cell line HuT-102 [2,37] were kindly provided by Ralph Grassmann (deceased, FAU, Erlangen, Germany) and were cultured in RPMI 1640M, 10% FCS and Pen/Strep. The HTLV-1 in vitro transformed T-cell line MS-9 (containing a single, full-length provirus) [38] was a kind gift from Charles Bangham (Imperial College, London, UK) and was cultured in RPMI 1640M, Panserin, 20% FCS, Pen/Strep and 100U/ml interleukin 2 (IL-2). All cell lines were checked for integrity by DNA profiling of eight different and highly polymorphic short tandem repeat loci (DSMZ, Braunschweig, Germany).
In general, 107 Jurkat T-cells were transiently transfected by electroporation using the Gene Pulser X Electroporation System (BioRad, Munich, Germany) at 290V and 1500μF. Cells were transfected using a total of 50 or 100μg of DNA. 5x105 293T cells or stable 293T cell lines that carry shNonsense, shFascin5 or shFascin4 were seeded in 6-well plates 24h prior to transfection. Cells were transfected with GeneJuice reagent (Merck Millipore, Darmstadt, Germany) according to the manufacturer’s protocol using a total amount of 2μg DNA.
HuT-102 cells stably transduced with shRNAs targeting Fascin (shFascin5) or a control (shNonsense) were co-cultured with Jurkat T-cells that had been transfected 24h earlier with the luciferase reporter plasmid pGL3-U3R-Luc carrying the luc gene under control of the HTLV-1 core promoter U3R [44], or with the control plasmid pGL3-Basic (Promega, Mannheim, Germany). After 48h of co-culture at 37°C, luciferase reporter gene assays were performed. Relative light units (RLU) were normalized on protein content and on background activity of controls (pGL3-Basic). Values obtained in control cells were set as 100% and at least three independent experiments each performed in triplicate were executed.
Cells were washed once with PBS (without Ca2+ and Mg2+) and then lysed in 100μl lysis buffer (100mM Tris/HCl (pH 7.8), 1M dithiotreitol (DTT), 0.18mM DCTA, 0.2% Triton X-100, 20% glycerol). After shaking for 30min at 30°C, samples were centrifuged (14.000rpm, 15min, 4°C) and supernatants were kept. Luciferase activities were measured according to the manufacturer’s instructions (Orion luminometer) using assay buffer (100mM KPO4, 15mM MgSO4, 4mM ATP) and D-luciferin (0.26 mg/ml; Roche Diagnostics, Indianapolis, IN, USA) dissolved in assay buffer.
5x105 of the respective cells were seeded and incubated for 48h at 37°C. Cells were centrifuged (1200rpm, 5min, 25°C), pellets were used for western blot analysis, and supernatants of MT-2 cells or of co-cultures from experiments using the single-cycle replication-dependent reporter vectors (see Infection assays) were sterile filtrated, and virus release was measured using gag p19 ELISA according to the manufacturer’s protocol (ZeptoMetrix Corporation, Buffalo, NY, USA). MT-2 cells were either left untreated or treated with DMSO (solvent control), cytochalasin D or nocodazole (5μM each) 48h prior to harvest. Additionally, MT-2 cells that stably carry shRNAs (shNonsense or shFascin5) were analyzed. Data were obtained using Softmax Pro Version 5.3 software (MDS Analytical Technologies, Sunnyvale, California, USA). At least, four independent experiments were performed.
Cells were washed once with PBS and protein lysates were obtained by lysis of cells in 100μl lysis buffer (150mM NaCl, 10mM Tris/HCl (pH 7.0), 10mM EDTA, 1% Triton X-100, 2mM DTT and protease inhibitors leupeptin, aprotinin (20μg/ml each) and 1mM phenylmethylsulfonyl fluoride (PMSF; 1mM)). After repeated freeze-and-thaw cycles, lysates were centrifuged (14.000rpm, 15min, 4°C). For detection of Tax, samples were sonicated three times for 20sec before centrifugation. Equal amounts of protein (50μg) were denatured for 5min at 95°C in sodium dodecyl sulfate (SDS) loading dye (10mM Tris/HCl (pH 6.8), 10% glycerin, 2% SDS, 0.1% bromophenol blue, 5% β-mercaptoethanol). After SDS-PAGE and immunoblotting on nitrocellulose transfer membranes (Whatmann, Protran, Whatmann GmbH, Dassel, Germany), proteins were detected using the following antibodies: rabbit polyclonal antibodies anti-V5 (Sigma), mouse monoclonal antibodies anti-Fascin (55K-2; Dako Deutschland GmbH, Hamburg, Germany), anti-β-actin (ACTB; Sigma), anti-Hsp90 α/β (F-8; Santa Cruz Biotechnology, Heidelberg, Germany), anti-HTLV-1 gag p19 (ZeptoMetrix Corporation), and anti-GFP (Sigma), and mouse antibodies to Tax, which were derived from the hybridoma cell line 168B17-46-34 (provided by B. Langton through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH; [45]). Secondary antibodies conjugated with horseradish peroxidase (HRP; GE Healthcare, Little Chalfont, UK) were used. Peroxidase activity was detected by enhanced chemiluminescence (ECL; 98.9% ECL A, 1% ECL B, 0.031% H2O2) using a CCD camera (Kodak Image Station 4000MM Pro camera, Kodak or Fujifilm LAS-1000 Intelligent Dark Box; Fujifilm). One of at least three independent western blots per experiment is shown. Intensities of specific bands were quantitated using Advanced Image Data Analyser (AIDA Version 4.22.034, Raytest Isotopenmessgeräte GmbH, Straubenhardt, Germany), and values were normalized on those of the housekeeping gene Hsp90 α/β.
107 Jurkat T-cells were transfected with 50μg pEFTax or pEF. 5x105 293T cells were transfected with 2μg of pEFTax or pEF (see Transfections). 48h later, total cellular RNA was isolated from transfected Jurkat or 293T cells (RNA isolation Kit II, Macherey-Nagel, Düren, Germany) and reversely transcribed to cDNA using SuperScript II and random hexamer primers (both Life Technologies GmbH). 200ng of cDNA and SensiMix II Probe Kit (BioLine GmbH, Luckenwalde, Germany) were used according to the manufacturer’s instructions for quantitative real-time RT-PCR (qPCR) in an ABI Prism 7500 Sequence Analyzer (Applied Biosystems, Foster City, CA, USA). Primers and FAM (6-carboxyfluorescein) / TAMRA (tetramethylrhodamine)-labeled probes for detection of β-actin (ACTB) and Tax transcripts have been described before [46]. A TaqMan Gene Expression Assay (Hs00979631_g1; Applied Biosystems) was used for quantitation of Fascin transcripts. Expression levels were computed by interpolation from standard curves generated from plasmids carrying the respective target sequences and calculation of the mean of triplicated samples. Relative copy numbers (rcn) were determined by normalizing copy numbers on those of ß-actin (ACTB). At least, three independent experiments were performed.
Microsoft Office Excel software was used for statistical analysis using the t-test (unpaired). P<0.05 was considered to be significant.
To assess the role of Fascin on HTLV-1 transmission, we made use of a single-cycle replication-dependent reporter system that is transfected into donor cells and allows monitoring of reporter gene activity in newly infected target cells only [25]. Briefly, a virus packaging plasmid encoding all HTLV-1 genes (wildtype; wt) and a replication-dependent HTLV-1 reporter vector containing a CMV-promoter driven luciferase (luc) gene were co-transfected into 293T cells. Alternatively, an HTLV-1 packaging plasmid carrying a mutation in the envelope (env) gene, and a VSV-G-expression plasmid were co-transfected instead of wt. The luc gene is oriented in antisense and is interrupted by an intron oriented in sense, therefore translation of the reporter mRNA in transfected cells is precluded. The vector mRNA is spliced and packaged into VLPs. After infection and replication, a provirus that lacks the intron is generated, and reporter gene expression (luc activity) can be measured in the target cell [25]. We previously used this system to assess the role of cellular restriction factors on HTLV-1 [49]. To analyze whether Fascin is important for transmission of these HTLV-1 reporter vectors, stable 293T cells with a knockdown of Fascin were generated. For this purpose, cells were transfected with two different shRNA constructs carrying an IRES-EGFP expression cassette and shRNAs targeting Fascin (shFascin5, shFascin4; [29,42]) or a control (shNonsense), and cells were selected with puromycin. Flow cytometry monitoring GFP-expression revealed that approximately 90% of cells carried the shRNA-constructs (S2A Fig). Beyond, vitality of stable cell lines was unaffected by the presence of shRNAs as detected by live/dead staining (S2B Fig). Cell lines were transfected with single-cycle replication-dependent HTLV-1 reporter vectors (inluc), a viral packaging plasmid (Δenv or wt), and as indicated with VSV-G for pseudotyping (Fig 1A). sh293T cells (shNonsense) transfected with inluc and Δenv served as negative control (control, Fig 1B) for both wt env-carrying and VSV-G-pseudotyped viral particles. After 24h, media were changed, and after another 24h, cells were harvested to measure cell-to-cell transmission in luciferase assays (Fig 1B), virus release by gag p19 ELISA (Fig 1C) and protein expression by western blot analysis (Fig 1D).
Making use of single-cycle replication-dependent HTLV-1 reporter vectors revealed that stable repression of endogenous Fascin by shRNAs leads to a significant reduction of reporter gene activity (Fig 1B). While shFascin5 resulted in a strong reduction of reporter gene activity by more than 70% (Fig 1B) and, in parallel, of Fascin protein (Fig 1D), the influence of shFascin4 on reporter gene activity (Fig 1B; by 30%) and on Fascin protein expression (Fig 1D) was less pronounced. Overexpression of Tax (black bars) did not enhance transmission of VSV-G-pseudotyped HTLV-1 (Fig 1B, left part of left panel), confirming earlier observations [25]. However, overexpressed Tax enhanced cell-to-cell-transmission of HTLV-1 reporters packaged with wt env (Fig 1B, right part of left panel and enlargement in right panel) contrary to previous observations [25]. The latter suggests that Tax and wt env cooperate in cell-to-cell transmission. The relative infectivity of VSV-G pseudotyped reporter vectors was about 7-fold higher than that of wt env-pseudotyped reporter vectors (Fig 1B) and undetectable, if no envelope was added (control). However, independent of the envelope type used, repression of Fascin significantly reduced the relative infectivity of both wt env-carrying and VSV-G-pseudotyped reporter vectors. Moreover, independent of the experimental condition, Tax could not further induce expression of Fascin protein in 293T cells (Fig 1D). While Tax led to a robust induction of Fascin mRNA (S4A Fig) and protein (S4B Fig) in Jurkat T-cells confirming our previous work [29], Tax did not further modulate Fascin expression in 293T cells, which already exhibit high amounts of endogenous Fascin (S4A and S4B Fig). Yet, reporter gene activity as a measure of HTLV-1 cell-to-cell transmission was Fascin-dependent in presence of overexpressed Tax, too (Fig 1B), suggesting that Fascin is also important for HTLV-1 cell-to-cell transmission in cells which express high amounts of endogenous Fascin. To analyze whether Fascin also impairs virus release, the viral gag p19 protein was measured by ELISA in cells that had been transfected with HTLV-1 reporters packaged with wt env. Overexpression of Tax (Fig 1C, black bars) did not further enhance gag p19 levels in the supernatants of 293T cells and repression of Fascin led to approximately 40% reduction of virus release only when Tax was supplemented suggesting that Tax and Fascin cooperate in processes that are important for virus release. Since repression of Fascin led to a severe defect of cell-to-cell transmission also in absence of supplemented Tax (Fig 1B, grey bars), but not to a decrease in virus release (Fig 1C, grey bars), this suggests that Fascin’s role in cell-to-cell transmission dominates over its role on virus release in this experimental setup. However, results obtained by the reporter system only provide a signal upon productive infection of a target cell, while the gag p19 ELISA also quantifies non-infectious VLPs. To exclude that virus production in the cell is already impaired by repression of Fascin, western blots detecting gag were performed (Fig 1D). Overall levels of cell-associated gag p55 were comparable between different experimental conditions. Beyond, Fascin was strongly repressed in presence of shFascin5 and only moderately repressed in presence of shFascin4. We could also detect Tax expressed from the packaging plasmids as tiny band, and an increased expression of Tax upon supplementing a Tax expression plasmid. Taken together, our data indicate that Fascin is important for transmission of HTLV-1 reporter vectors independent of the envelope type in 293T cells.
To strengthen our results, we made use of Fascin-specific nanobodies that had been developed and characterized previously [43]. Briefly, nanobodies are antigen-binding domains of camelid heavy-chain antibodies. The employed Fascin-specific nanobodies contain a mitochondrial outer membrane (MOM) signal that leads to targeted subcellular delocalization of Fascin to the MOM [43]. Use of these nanobodies allowed us to trigger Fascin protein loss of function without changing its expression. Upon transfection of HTLV-1 reporter vectors and expression plasmids encoding Fascin-specific nanobodies into 293T cells, luciferase assays (Fig 2A), gag p19 ELISA (Fig 2B), western blot analysis (Fig 2C) and immunofluorescence stains followed by confocal laser scanning microscopy (Fig 2D–2F) were performed. We found that Fascin nanobody 5 (FASNb5) significantly reduced reporter gene activity in a dose-dependent manner compared to a control nanobody (GFPNb) and to FASNb2 (Fig 2A). In gag p19 ELISA (Fig 2B), we measured a dose-dependent decrease of released gag p19 into the supernatants in presence of FASNb5, suggesting that this nanobody also impairs release of HTLV-1. Expression of the nanobodies and the unaltered expression of Fascin were confirmed by western blot analysis (Fig 2C). Additionally, immunofluorescence was performed confirming earlier studies [43] showing co-localizations of V5-tagged and MOM-expressing nanobodies (GFPNb, FASNb2, FASNb5) with mitochondria (Fig 2D). Next, we checked whether Fascin-specific nanobodies lead to efficient delocalization of Fascin by staining V5-tagged nanobodies and Fascin. Immunofluorescence analysis revealed that FASNb5 lead to a more efficient delocalization of Fascin (90.2% of Fascin delocalized) compared to FASNb2 (71.1%; Fig 2E), which mirrors the different impact of FASNb5 and FASNb2 on cell-to-cell transmission (Fig 2B) and virus release (Fig 2C). However, since FASNb5 also impairs Fascin-mediated actin-bundling compared to FASNb2 [43], these data also suggest that Fascin’s actin-bundling activity could be required for transmission of HTLV-1. Contrary to delocalizing Fascin, FASNb5 did not delocalize gag to the mitochondria (Fig 2F), suggesting that Fascin and gag do not directly interact, or, if they interact, the interaction is not sustained during delocalization. Moreover, the impact of FASNb5 on virus release as measured by gag p19 ELISA may be indirect (Fig 2B), e.g. by impairing the transport of gag to budding sites or by impairing budding itself. Summed up, not only repression, but also delocalization of Fascin in the cell interferes with HTLV-1 cell-to-cell transmission.
Our results obtained thus far do not exclude that repression of Fascin impairs viral entry since we used a one-step transfection/infection co-culture system, where transfected cells produce VLPs that infect neighboring cells [25], which also harbor a knockdown of Fascin, or which could be impaired by Fascin-specific nanobodies. Further, since HTLV-1 predominantly infects CD4+ T-cells in vivo, we switched to a more physiological system and analyzed the role of Tax and Fascin on HTLV-1 transmission in CD4+ Jurkat T-cells in co-culture with Raji/CD4+ B-cells, a co-culture system that had been described earlier to allow monitoring of HTLV-1 transmission with single-cycle replication-dependent reporter vectors [25]. Upon co-transfection of Jurkat T-cells with HTLV-1 reporters (inluc), packaging plasmids (wt), Tax expression plasmids and shRNAs targeting Fascin (Fig 3A), media were changed at 24h, and Jurkat T-cells were co-cultured for another 48h with Raji/CD4+ B-cells. Measurement of luciferase reporter gene activity reflecting cell-to-cell transmission revealed that repression of Fascin did not affect basal HTLV-1 cell-to-cell transmission (Fig 3B, grey bars). However, upon overexpression of Tax (black bars), reporter gene activity significantly increased confirming earlier observations in this cell type [25]. Interestingly, repression of Fascin led to a reduction of Tax-induced reporter gene activity suggesting that Fascin is a major contributor of Tax-induced cell-to-cell transmission. To exclude and to confirm that the measured reporter gene activity is not due to cell-free virus transmission, we incubated Raji/CD4+ B-cells with supernatants of Jurkat T-cells that had been transfected with the reporter system, which did not result in detectable luciferase signals (S3 Fig; [25]). Measuring of gag p19 release by ELISA mirrored the results obtained by luciferase assays and showed that Tax-enhanced virus release in Jurkat T-cells occurs Fascin-dependently (Fig 3C). Knockdown of Fascin in presence of overexpressed Tax led to a reduction of gag p19 release nearly reaching those levels measured without supplemented Tax. Western blot analysis showed that, contrary to 293T cells (Fig 1D, S4 Fig), Tax is a potent inducer of Fascin transcript and protein expression in Jurkat T-cells (Fig 3D, S4 Fig) confirming our previous results [29,34]. Thus, too low levels of endogenous Fascin in Jurkat T-cells without overexpressed Tax (S4 Fig) could be a potential explanation for the Tax-dependency of the Fascin-effect in this cell type. Concomitant with our findings obtained in 293T cells (Fig 1D), western blot analysis revealed that the levels of cell-associated gag p55 were comparable between different experimental conditions also in Jurkat T-cells (Fig 3D). Contrary to 293T cells, Fascin was induced by Tax in Jurkat T-cells. Further, Fascin was strongly repressed in presence of shFascin5 and moderately repressed in presence of shFascin4. Tax expressed from the packaging plasmids could be detected as a tiny band, and an increased expression of Tax was detectable upon supplementing a Tax expression plasmid. Thus, Tax enhances virus release, and augments cell-to-cell transmission from Jurkat T-cells to Raji/CD4+ B-cells dependent on Fascin. To substantiate these findings, we tested Fascin-specific nanobodies instead of shRNAs in the Jurkat-Raji/CD4+ co-culture model. Compared to the control nanobody GFPNb and to FASNb2, FASNb5 led to a significant reduction (by 61%) of Tax-induced HTLV-1 reporter gene activity (Fig 3E). This suggests that delocalization of Fascin without changing its expression (Fig 3F) and inhibition of Fascin’s actin-bundling activity by FASNb5 [43] impair HTLV-1 cell-to-cell transmission in different cell types (Figs 2, 3E and 3F). Taken together, use of single-cycle replication-dependent HTLV-1 reporter vectors revealed that stable repression of endogenous Fascin (293T cells), or of Tax-induced Fascin (Jurkat T-cells) by shRNAs and inhibition of Fascin using specific nanobodies impair both gag p19 release and HTLV-1 cell-to-cell transmission.
Next, we asked whether Fascin contributes to HTLV-1 cell-to-cell transmission also in chronically HTLV-1-infected T-cells, which express high amounts of Fascin protein. For this purpose, ATL-derived HTLV-1-infected HuT-102 cells were stably transduced with lentiviral vectors expressing either a shRNA targeting Fascin (shFascin5) or a nonsense shRNA (shNonsense) [29]. According to a published protocol [50], HuT-102 cells were co-cultured with Jurkat T-cells that had been transfected with an HTLV-1-LTR (U3R)-dependent luc gene reporter system (pGL3-U3R; Fig 4A). Upon infection of Jurkat T-cells, the viral Tax protein should activate expression of the HTLV-1 U3R resulting in enhanced luciferase activity. After 24h of co-culture, luciferase activity was measured and normalized on protein content and on transactivation of a mock luciferase construct (Fig 4B). Transactivation of the reporter in Jurkat T-cells was diminished by more than 50% when Fascin was knocked down in the co-cultured HTLV-1-infected HuT-102 cells hinting at a role of Fascin for HTLV-1-mediated cell-to-cell transfer. In parallel, knockdown of Fascin in HuT-102 was verified in immunoblots and shown to be Fascin-specific since Tax protein and the housekeeping gene ß-actin (ACTB) were not affected by shFascin5 (Fig 4C). Further, we excluded detrimental effects of the shRNA on cell vitality by measuring apoptosis and cell death in stable cell lines compared to cells treated with 15μM etoposide, which is known to induce cell death (S5 Fig). Since results from co-culture assays may also reflect cell fusion events or the transfer of Tax-containing exosomes [51], we decided to measure HTLV-1 infection also directly by a flow cytometry based assay that allows monitoring of newly infected cells. For this purpose, we first stably transduced the chronically HTLV-1-infected T-cell line MT-2 cells with lentiviral vectors expressing either shFascin5 or shNonsense. According to an established protocol [28], MT-2 cells were then co-cultured with uninfected Jurkat T-cells for 1h and the number of newly infected Jurkat T-cells was detected by flow cytometry by measuring the amount of the viral matrix protein gag p19 in these cells (Fig 4D and 4E). For this purpose, co-cultures were permeabilized and stained with antibodies targeting HTLV-1 gag p19 and the IL-2 receptor alpha chain CD25, which is present on MT-2 cells, but not on Jurkat T-cells [52]. Flow cytometry revealed that repression of Fascin led to a significant reduction of newly infected, gag p19-positive Jurkat T-cells to 68% compared to the control (Fig 4E). Beyond, western blot analysis confirmed a robust reduction of Fascin protein in MT-2 cells carrying shFascin5 (Fig 4F), while Tax and ACTB were unaffected. Further, cell vitality was also unaffected by repression of Fascin as indicated by live/dead stainings (S5 Fig). We also measured release of gag p19 into culture supernatants and found that knockdown of Fascin not only reduced infection of co-cultured Jurkat T-cells, but also diminished the release of gag p19 (Fig 4G). Similar results were obtained with MT-2 cells treated with cytochalasin D or nocodazole (5μM each), which interfere with actin or tubulin polymerization, respectively (Fig 4H). Overall, our observations are in line with the data we obtained with the HTLV-1 reporter vectors (Figs 1–3) suggesting that, independent of the cell and test system used, repression of Fascin impairs release and cell-to-cell transmission of HTLV-1. Immunoblot analysis revealed that cell-associated gag and processing of the gag p55 precursor into gag p19 and gag p27 was unaffected by knockdown of Fascin (Fig 4I). However, treatment of MT-2 cells with the compounds cytochalasin D and nocodazole interfered with processing of gag p55 suggesting that chemical interference with the cytoskeleton acts differently on virus production than Fascin repression. Taken together, we found that Fascin is critical for release and cell-to-cell transmission of HTLV-1 reporter vectors, and for transactivation and infection of co-cultured T-cells indicating an important role of Fascin in HTLV-1 cell-to-cell transmission.
To shed light on the mechanism of Fascin’s role during HTLV-1 transmission, we asked whether Fascin enhances conjugate formation between infected and uninfected T-cells similar to the small GTP-binding protein GEM [28]. For this purpose, we performed a flow cytometry-based conjugate formation assay between Jurkat T-cells that had been transfected with Tax and Fascin-specific shRNAs as donor cells, and Raji/CD4+ B-cells as acceptor cells according to a previously described protocol [25]. Briefly, Jurkat T-cells were co-transfected with one of two different Tax-expression constructs (GFP-Tax; pEFTax) and one of two different shRNAs encoding IRES-EGFP and targeting Fascin (shFascin5, shFascin4) or a control (shNonsense). After 24h, cells were co-cultured with Raji/CD4+ B-cells for 1h and the percentage of conjugate formation between the two cell types was quantitated by flow cytometry (S6 Fig). Detection of GFP encoded by GFP-Tax and/or the shRNA constructs was used to differ between transfected (GFP+) and untransfected (GFP-) Jurkat T-cells. Conjugates of Jurkat T-cells (CD3+) with Raji/CD4+ B-cells (HLA-DR+) were identified in GFP+ (Tax-positive) and GFP- (Tax-negative) gates as double-positive signals (HLA-DR+CD3+) and normalized on the total number of Jurkat T-cells. While Tax enhanced conjugate formation between the two different cell types confirming earlier observations (Fig 5A; [25]), we found that Tax-induced cell aggregation was independent of Fascin (Fig 5B). Western Blot analysis confirmed the expression of GFP-Tax, Tax, and the functionality of the Fascin-specific shRNAs (Fig 5C).
To validate these findings also in chronically infected T-cells, we asked whether Fascin is important for conjugate formation between HTLV-1-infected MT-2 cells and Jurkat T-cells, or whether Fascin contributes to adhesion of HTLV-1-infected cells on different attachment factors. For this purpose, HTLV-1-infected MT-2 cells with repressed Fascin (shFascin5) and the respective controls (shNonsense, untreated) were co-cultured with Jurkat T-cells that had been pre-stained with the life cell dye Calcein-AM (green). After 1h of co-culture at 37°C, cells were spotted on glass slides either coated with poly-L-lysine or fibronectin (Fig 6A). Co-cultures were stained with antibodies targeting the viral matrix protein gag p19 (blue) to label HTLV-1-infected MT-2 cells and with antibodies targeting Fascin (red; Fig 6B).
Immunofluorescence revealed that gag was detectable in all experimental conditions, while expression of Fascin was repressed in MT-2 shFascin5 cells. Overlay of the respective channels and transmitted light showed that protrusive structures between chronically infected MT-2 cells and uninfected Jurkat T-cells could be detected (Fig 6Bd and 6Be). The length of the protrusions was approximately 8.35μm+/-2.05μm (on fibronectin) or 6.16μm+/-2.24μm (on poly-L-lysine). In most cells, Fascin and gag localized in close proximity (Fig 6Bi), and occasionally, both proteins co-localized (Fig 6Bd). This was further confirmed by determining fluorescence intensities of both Fascin and gag along arbitrary drawn ROIs (regions of interest; S7a–S7g Fig): (1) Fascin and gag localize in close proximity, but do not co-localize (ROI 1). (2) The parallel shape of Fascin and gag fluorescence intensities at ROI 2 suggests that both proteins co-localize. Since co-localization events were only found in < 5% of all MT-2 cells, our findings suggest rather a transient or indirect than a tight and direct interaction between Fascin and gag during the dynamic process of virus budding.
Next, automatic image analysis was performed to count the number of cell-cell contacts between infected MT-2 cells and uninfected Jurkat T-cells and to handle the large numbers of images and cells. An image processing algorithm was developed and applied that allowed for automatic quantitation of the respective cell types, and for counting of cell-cell contacts between infected and uninfected cells [47]. Automatic image analysis discriminated between Jurkat T-cells and MT-2 cells based on a cell segmentation approach using an active contour algorithm incorporating a priori shape information. The accuracy of cell segmentation determined on a single cell basis added up to aJ = 82.5% for Jurkat T-cells and aM = 77.8% for MT-2 cells. Correctly identified Jurkat cells were segmented with a hit quality hJ = 96.4% and correctly identified MT-2 cells with hM = 83.2%. In parallel, the micrograph images were checked manually to include cells into the evaluation that had been missed by the algorithm. Applying this algorithm, the following findings were obtained: (1) The number of cell-cell aggregates between HTLV-1-infected and uninfected T-cells is independent of the attachment factor and of Fascin (Fig 6C), confirming our results from the flow cytometry-based assay (Fig 5). (2) Adhesion of HTLV-1-infected MT-2 cells is significantly impaired upon knockdown of Fascin (Fig 6D), while cell vitality is unaffected (S5 Fig). Thus, Fascin seems to be important for proper attachment of MT-2 cells on fibronectin- and poly-L-lysine-coated matrices and could favor dissemination of infected cells in vivo. Taken together, although Fascin seems to be required for proper attachment, Fascin does not affect the quantity of cell-cell contacts to uninfected Jurkat cells.
Having found that knockdown of Fascin impairs release and cell-to-cell transmission of HTLV-1, we performed imaging analysis to shed light on the localization of Fascin and of the viral gag protein (Fig 7A). Most chronically infected T-cell lines harbor more than one copy of HTLV-1 provirus and produce large amounts of gag, which is unfavorable for imaging analysis. Therefore, we decided to analyze the chronically infected T-cell line MS-9, which harbors only one integrated provirus and thus, reasonable amounts of gag to perform imaging analysis [26] (Fig 7B). MS-9 cells (Fascin-positive) were co-cultured with Jurkat T-cells that had been pre-stained with the live cell dye Calcein-AM (green) and express only low amounts of endogenous Fascin. Cells were spotted on poly-L-lysine- and fibronectin-coated coverslips and incubated for 0, 30, or 60min at 37°C before fixation. Afterwards, cells were stained with antibodies targeting Fascin (red) and gag p19 (blue), and confocal laser scanning microscopy was performed (Fig 7A). Imaging revealed different patterns of Fascin localization at cell-cell contacts. First, we examined cells where the viral gag protein polarizes towards the uninfected target cell, suggesting the presence of the virological synapse (VS) [21] (Fig 7Aa–7Aj). At cell-cell contacts we found not only clusters of gag (Fig 7Ab; thin white arrows), which are reminiscent of viral biofilms [24,53], but also clusters of Fascin protein (Fig 7Ac). Concomitant with our previous findings in MT-2 cells (Fig 6B; S7 Fig), we could rarely detect co-localizations between Fascin and gag. However, polarized gag clusters were interspersed with Fascin clusters (Fig 7Ad) at cell-cell contacts, which was confirmed by analysis of fluorescence intensities across the cell-cell-contact region (S8 Fig). This suggests that Fascin makes room for gag clusters at the VS. Since large aggregates of HTLV-1 virions in the viral biofilm on the surface of infected cells are important for efficient infection of target cells [53], Fascin could be important for the transport of viral proteins to the budding site, and thus, foster HTLV-1 transmission. This idea is supported by the quantitative evaluation of T-cell conjugates between MS-9 and Jurkat T-cells: If Fascin is localized at the cell-cell contact region, the frequency of polarized gag (suggesting formation of the VS) is 79.5% or 43.8% on poly-L-lysine or fibronectin, respectively. In contrast, if Fascin is dispersed and not accumulated at the cell-cell contact, the frequency of gag polarization (at the VS) is much lower (1.1% on poly-L-lysine, 1.8% on fibronectin). Thus, this suggests a direct role of Fascin in the local distribution of gag to budding sites and an indirect effect on cell-to-cell transmission. Beyond, we observed short, Fascin-containing membrane extensions that clutched uninfected T-cells (Fig 7Ai; white-framed arrows) and made room for these gag clusters (Fig 7Ai; thin white arrows). Finally, we found long-distance connections (approximately 15.21±5.48 μm in length on poly-L-lysine and 24.14±1.29μm on fibronectin) between infected MS-9 cells and uninfected Jurkat T-cells (Fig 7Ak–7Av; thick white arrows). The frequency of protrusions emanating from infected MS-9 cells was low (approximately 3.28% of all MS-9 cells), however, protrusions were found independent of the time point of analysis (Fig 7Ao and 7Au). Interestingly, we found Fascin and gag p19 protein expression in 65.3% or 79.5% of all protrusions (on poly-L-lysine or on fibronectin, respectively). Among the Fascin-positive protrusions all except one were also stained positive for gag p19. As depicted in Fig 7A, Fascin and gag partially co-localize in these cellular protrusions (Fig 7At, 7Av and 7Aw), or the proteins are located in clusters in close proximity (Fig 7An and 7Ap) between infected MS-9 cells and newly infected Jurkat T-cells. Thus, formation of Fascin-containing protrusions could potentially account for the transfer of gag to target cells.
To summarize our data, Fig 8 gives an overview of our current findings and provides a model of Fascin’s role in HTLV-1 transmission. HTLV-1-infected T-cells express the transactivator Tax that upregulates Fascin expression via the NF-κB signaling pathway. Not only Tax-induced Fascin, but also endogenous Fascin seems to be required for virus release and cell-to-cell transmission. Beyond, adhesion of infected cells in co-culture with uninfected cells occurs Fascin-dependently, which may favor dissemination of infected cells in vivo. Functionally, Fascin clusters intersperse with gag clusters suggesting that Fascin makes room for gag clusters reminiscent of viral biofilms at the VS. Furthermore, short-distance Fascin-containing membrane extensions clutch uninfected T-cells, which could favor the transfer of viral material to target cells via budding of enveloped virions at tight cell-cell contacts at the VS. Additionally, Fascin localizes with gag in long-distance connections between chronically infected and newly infected T-cells. It is conceivable that Fascin is required for the proper organization of protrusive structures, which may account for budding of HTLV-1 at the tip of the protrusion towards the target cell via a putative “mini VS”, a structure that had been proposed earlier [22,54]. Overall, our data suggest that Fascin could be important for the transport of viral proteins to foster polarized budding, virus release and cell-to-cell transmission of HTLV-1. Thus, Fascin is an interesting novel target to inhibit HTLV-1 cell-to-cell transmission.
In this work we found that the actin-bundling protein Fascin is critical for HTLV-1 transmission. Fascin is known as a tumor marker, which is highly upregulated in many types of cancer and crucial for invasion and metastasis, potentially by stabilizing invasive structures [30]. We previously showed that Fascin is also important for invasive migration of virus-transformed lymphocytes [29,42]. Using different cell culture systems and infection models, we now found that repression of Fascin by shRNA or by Fascin-specific nanobodies severely impairs release and cell-to-cell transmission of the retrovirus HTLV-1 shedding new light on the function of Fascin.
To address the role of Fascin in HTLV-1 cell-to-cell transmission, we made use of a single-cycle replication dependent reporter system, which allows automatic quantitation of productive infection in newly infected target cells only [25]. Using this system, our results indicate that Fascin is a major contributor of HTLV-1 cell-to-cell transmission independent of the cell type and the envelope type tested. Despite recent work showing that the reporter system we used even underestimates cell-to-cell transmission events [55], we still see a significant reduction of reporter gene activity reflecting cell-to-cell transmission upon repression of endogenous or Tax-induced Fascin expression. However, new reporter vectors with improved splicing and packaging of the spliced reporter RNA might allow for better quantitating the role of Fascin on cell-to-cell transmission in lymphocytes and in primary cells in future work [55].
Our data suggest that both virus release and cell-to-cell transmission appear Fascin-dependent, while the amount of cell-associated virus (as reflected by western blots of gag) is not affected by repression of Fascin. However, our data also suggest that Fascin’s role in cell-to-cell transmission dominates over its role on virus release. Despite a significant impact on cell-to-cell transmission, virus release was not affected by Fascin-specific shRNAs in every experimental condition (Fig 1C). This may be due to the fact that ELISAs measuring gag p19 also quantify non-infectious VLPs, while reporter assays (Figs 1–3) and flow cytometry measuring gag p19 transfer (Fig 4) quantitate newly-infected cells only. Although the release of virions is impaired upon knockdown of Fascin, cell-free virions carrying the wildtype env of HTLV-1 are severely impaired in infecting target cells (S3 Fig; [25]). Thus, it is very likely that reduced infectivity of co-cultured target cells upon knockdown of Fascin results from cell-to-cell transmission, and not from infection with poorly infectious free viral particles. Thus, the impact of Fascin on direct cell-to-cell transmission could be underestimated. To confirm the relevance of Fascin for HTLV-1 transmission, we also investigated chronically infected T-cells. Both fusion-based assays and measuring of gag transfer to target cells confirmed the relevance of Fascin for HTLV-1 release and cell-to-cell transmission.
Use of Fascin-specific nanobodies that target Fascin to the mitochondrial outer membrane (MOM) confirmed a role of Fascin in gag p19 release and HTLV-1 cell-to-cell transmission. Nanobodies are new promising stable recombinant antigen-binding domains of camelid heavy-chain antibodies that had already been successfully used to prevent Fascin-dependent invasion and migration of cancer cells [43,56]. Thus, nanobodies trigger Fascin protein loss of function without changing its expression [43]. Since the potent nanobody FASNb5 not only delocalizes Fascin to the MOM more efficiently, but also impairs Fascin-mediated organization of actin-bundles [43,56], Fascin’s actin-bundling activity might be required for transmission of HTLV-1. For HTLV-1, the role of the actin cytoskeleton on virus transmission has not been analyzed in detail, however, it is known that polarization of the MTOC and transfer of HTLV-1 reporter vectors to target cells is impaired in presence of compounds interfering with actin polymerization [21,25,57]. We found that not only repression of the actin-bundling protein Fascin, but also interference with actin and tubulin polymerization led to reduced gag p19 levels in the supernatants of HTLV-1-infected T-cells. These observations are in contrast to HIV, where the budding process does not strictly rely on cytoskeleton remodeling. Although filamentous actin co-localizes with budding structures, inhibition of actin does not change localization of budding sites and packaging of actin and actin-binding proteins into virions seems to be a secondary consequence of the high abundance of these molecules at budding sites [58].
Yet, it is unknown, whether Fascin also contributes to release and cell-to-cell transmission of other viruses than HTLV-1. We found that not only Tax [29], but also latent membrane protein 1 (LMP1) of Epstein-Barr virus (EBV) is a potent inducer of Fascin [42]. Interestingly, LMP1-deleted EBV is severely impaired in virus release into culture supernatants, potentially due to a defect in particle transport [59]. Thus, LMP1-mediated induction of Fascin and its continuous expression suggest a role of Fascin in virus release also of EBV. This is further corroborated by the finding that cell-to-cell transmission of EBV to epithelial cells also depends on canonical NF-κB signaling [60], which is a prerequisite for efficient Fascin induction by LMP1 [42].
Although it is known that Tax is required for formation of the VS and efficient virus transmission [21], only little is known about host factors that are regulated by Tax to modulate virus transmission [17]. With regard to pathways important for viral transmission, Tax transcriptionally alters the expression of cell adhesion and surface molecules, leads to cytoskeletal remodeling and complexes with proteins involved in cytoskeleton structure and dynamics. These Tax-interacting proteins include α-internexin, cytokeratin, actin, gelsolin, annexin, γ-tubulin and small GTPases of the Rho family [61] Two of these Rho-GTPases, Rac-1 and Cdc42, complex with Tax and seem to be important for Tax-induced MTOC-polarization [57,62]. Thus, it is conceivable that Tax might connect Rho GTPases to their targets and affect cytoskeleton organization which could favor HTLV-1 transmission. Interestingly, Chevalier et al. found that GEM, which is an upstream negative regulator of ROCK-I Rho kinase, is induced by Tax [28]. GEM is a small GTP-binding protein and enhances cellular migration and conjugate formation between infected and uninfected T-cells. Knockdown of GEM in chronically infected T-cells reduces gag transfer to target cells showing that GEM is required for viral transmission [28]. It had been suggested earlier that not only GEM, but also Fascin and collapsin response mediator protein 2 (CRMP2), which is induced by Tax and important for migration [63], might contribute to HTLV-1 transmission [17,28]. We now show that this is true for Fascin, however, the mechanism differs from the one described for GEM. Both GEM and Fascin are important for HTLV-1 cell-to-cell transmission, whereas only GEM is required for T-cell conjugate formation between infected and uninfected T-cells [28]. Contrary, Fascin also impairs virus release, which seems to be unaffected by GEM.
Our findings that the adhesion of infected cells to different matrices is modulated by Fascin, in co-cultures with uninfected cells, could explain our previous observations where we found Fascin to be important for the invasion of ATL-derived cells through ECM and for the invasive migration of EBV-transformed and LMP-1-expressing lymphocytes [29,42]. These results are in line with recent data showing that Fascin forms a complex with focal adhesion kinase (FAK) and Src to control adhesion stability [31]
Contrary to the test systems used in our manuscript, free viral particles of HTLV-1 are hardly detectable in vivo [20]. Viruses are transmitted at tight cell-cell contacts or via cellular protrusions protected from the host’s immune response [21,22]. It is estimated that HTLV-1 buds into a synaptic cleft and is transferred to target cells [21]. Moreover, viruses are tethered to and embedded in extracellular assemblies, viral biofilms, and transmitted at virological synapses to target cells [24]. It is likely that immune pressure and specific signals from uninfected target cells play a role in preventing release of HTLV-1 in vivo. Thus, it remains to be determined how Fascin affects HTLV-1 transmission in natural infection.
Imaging revealed that Fascin clusters localize in close proximity to gag clusters at cell-cell-contacts, which are reminiscent of viral biofilms. Viral biofilms are carbohydrate-rich surface assemblies of viral particles which are composed of various components of the ECM and they account for the majority of HTLV-1 cell-to-cell transmission in vitro [24]. The localization of Fascin in close proximity to gag suggests that Fascin makes room for gag clusters at viral biofilms. Beyond, it is also conceivable that Fascin is required for formation, maintenance or tethering of viral biofilms, e.g. by redirecting the transport of viral and cellular proteins to budding sites via reorganization of the actin- or microtubuli-cytoskeleton [30,31]. Since Fascin is concentrated at cell-cell contacts, and localizes in close proximity to gag clusters, it is possible that Fascin may be packaged into HTLV-1 particles. We also observed short, Fascin-containing short membrane extensions clutching uninfected T-cells. These potentially support the transfer of virions to target cells, but presumably not due to enhanced conjugate formation, which remains unaffected by Fascin. Surprisingly, we observed Fascin and gag localization in long-distance protrusions between chronically infected and newly-infected T-cells. Long distance connections for the transfer of retroviruses or viral proteins have previously also been found in cells infected with MLV [64] or HIV [65]. For HTLV-1, the viral p8 protein was identified as inducer of cellular protrusions [22]. Therefore, it remains to be determined, whether p8-induced protrusions are Fascin-dependent, and whether viruses bud from these protrusions at a “mini VS” to target cells.
Taken together, our data suggest that Fascin could be important for the transport of viral proteins to budding sites, and thus, foster HTLV-1 transmission. However, the detailed mechanism of Fascin-dependent HTLV-1 transmission remains to be determined. Since repression of Fascin also reduces release of gag p19 into culture supernatants, it is conceivable that either the transport of viral proteins to the budding sites is impaired, or that viral particles are retained inside the infected cell, or at the viral biofilms. Since co-localization events between Fascin and gag were rare, our findings suggest a transient or indirect Fascin:gag interaction during the dynamic process of virus budding. Despite playing a crucial role in cell-to-cell transmission of HTLV-1, it is not settled yet whether Fascin is also essential for formation of the VS. Localization of Fascin at cell-cell contacts and its association with a high frequency of polarized gag suggests that Fascin is involved in recruiting gag to the VS, and, thus, indirectly affects cell-to-cell transmission. However, it is unclear whether gag protein could localize at the VS in the absence of Fascin. These experiments are not accomplishable with chronically infected MT-2 cells, which can be manipulated by knockdown strategies, since these cells carry several proviral copies- some of them defective [66]- and excessive amounts of cell-associated gag protein.
Fascin may also represent an interesting regulator of HTLV-1 cell-to-cell transfer in other cell types than infected T-cells. It is estimated that dendritic cells (DC) are the primary cells to be infected in vivo and that they play a pivotal role in transmitting the virus to CD4+ T-cells depending on cell-cell-contacts. Beyond, infection of DC may also be required for the establishment and maintenance of HTLV-1 infection in primate species [67]. Contrary to CD4+ T-cells, DCs are efficiently infected cell-free by highly concentrated viruses or by separated viral biofilms in vitro [68,69]. Since Fascin expression is selectively induced in mature DC [70], important for the stability of dendrites and for formation of the immunological synapse [32], future work should also investigate whether Fascin plays a role in dissemination of HTLV-1 from DC to T-cells.
For a long time, Fascin has been known as an actin-bundling protein only. However, Fascin exerts other functions independent of its role in actin-binding and -bundling. Recent findings have supported this notion showing that Fascin also interacts with microtubules [31]. In light of HTLV-1 transmission, which depends on polarization of the MTOC and on proper actin and tubulin function [21,25], our work identifying Fascin as a critical host factor in HTLV-1 transmission may provide a link between the activity of Tax and regulation of both the actin and microtubule cytoskeleton. Thus Fascin is a promising candidate to counteract HTLV-1 transmission.
|
10.1371/journal.pntd.0004368 | Neuromuscular Effects of Common Krait (Bungarus caeruleus) Envenoming in Sri Lanka | We aimed to investigate neurophysiological and clinical effects of common krait envenoming, including the time course and treatment response.
Patients with definite common krait (Bungarus caeruleus) bites were recruited from a Sri Lankan hospital. All patients had serial neurological examinations and stimulated concentric needle single-fibre electromyography (sfEMG) of orbicularis oculi in hospital at 6wk and 6–9mth post-bite.
There were 33 patients enrolled (median age 35y; 24 males). Eight did not develop neurotoxicity and had normal sfEMG. Eight had mild neurotoxicity with ptosis, normal sfEMG; six received antivenom and all recovered within 20–32h. Seventeen patients developed severe neurotoxicity with rapidly descending paralysis, from ptosis to complete ophthalmoplegia, facial, bulbar and neck weakness. All 17 received Indian polyvalent antivenom a median 3.5h post-bite (2.8–7.2h), which cleared unbound venom from blood. Despite this, the paralysis worsened requiring intubation and ventilation within 7h post-bite. sfEMG showed markedly increased jitter and neuromuscular blocks within 12h. sfEMG abnormalities gradually improved over 24h, corresponding with clinical recovery. Muscle recovery occurred in ascending order. Myotoxicity was not evident, clinically or biochemically, in any of the patients. Patients were extubated a median 96h post-bite (54–216h). On discharge, median 8 days (4–12days) post-bite, patients were clinically normal but had mild sfEMG abnormalities which persisted at 6wk post-bite. There were no clinical or neurophysiological abnormalities at 6–9mth.
Common krait envenoming causes rapid onset severe neuromuscular paralysis which takes days to recover clinically consistent with sfEMG. Subclinical neuromuscular dysfunction lasts weeks but was not permanent. Antivenom effectively cleared venom but did not prevent worsening or reverse neuromuscular paralysis.
| Common krait bites cause muscular paralysis due to the venom disrupting communication between the nerves and muscles. This becomes life-threatening for the patient if there is paralysis of the muscles used for breathing. We studied the severity of paralysis, long term effects and the value of antivenom treatment in authenticated Indian krait bite patients from Sri Lanka. In addition to standard treatment with antivenom, the patients had single-fibre electromyography done, a sensitive neurophysiological test that detects the abnormalities of communication between the nerves and muscles. Half of the patients had severe paralysis and required mechanical ventilation, and the remainder had mild or no effects. Antivenom was given to all patients with severe paralysis and most with mild effects. However, despite antivenom binding all free venom after it was administered, it did not prevent or reverse already developed paralysis. Clinically evident paralysis resolved after a few days, but the neurophysiological abnormalities lasted for weeks. No permanent neurological damages were noted at 6 to 9 months after the snake bite.
| Globally snake envenoming is a major cause of morbidity and mortality in the rural tropics [1]. Neurotoxicity causing paralysis is one of the major clinical syndromes of snake envenoming, and occurs mainly with elapids (Australasian elapids, American coral snakes, Asian kraits and some cobra species)[2,3].Envenoming may result in prolonged hospital stay if ventilatory support is available or significant mortality where such medical resources are not available. Despite the magnitude of the health impact of neurotoxic snake envenoming, it continues to be associated with several unresolved issues of clinical importance[2].
Envenoming due to krait (Genus: Bungarus) bites is a common, serious health issue in South and South-East Asia. Common krait (Bungarus caeruleus) is distributed throughout South Asia, and is responsible for large numbers of cases of severe neurotoxic envenoming each year[4]. It results in a descending flaccid paralysis progressing to life threatening respiratory paralysis unless mechanical ventilation is available[5–7]. Krait bites typically occur at night and are not painful, so many patients do not notice the bite and continue sleeping[7–10],which delays medical care.
Neuromuscular paralysis in krait envenoming is characterized by progressive descending paralysis. Krait venom contains β-bungarotoxins, which are presynaptic neurotoxins with phospholipase A2 activity and considered to be the major cause of paralysis[11]. The pre-synaptic action is irreversible and is the reason that once paralysis develops it is not reversed with antivenom[12].
The pathophysiology of neuromuscular paralysis in snake envenoming remains poorly understood. This is due to the lack of detailed clinical studies of authenticated bites that report the time course of paralysis, including the recovery of paralysis, and very few studies of neurophysiological function in snake envenoming[2]. There is also little evidence for the effectiveness of antivenom in reversing neuromuscular paralysis[13,14]. Neurophysiological testing could provide objective evidence of the progression and recovery of neurotoxicity, and the response to antivenom. Previous neurophysiological testing in krait bites with nerve conduction studies and repetitive nerve stimulation tests, provide some information on krait neurotoxicity[15]. Single fibre electro-myography (sfEMG) is a more sensitive test of neuromuscular function[16]. It has only been reported in one study of neuromuscular paralysis from Papuan taipan envenoming.[17]Better methods to measure neuromuscular dysfunction in neurotoxic envenoming will improve our understanding of neuromuscular paralysis and the clinical benefit of antivenom.
This study aimed to describe detailed neurological examinations in patients with krait envenoming, the use sfEMG to objectively measure the time related progression and recovery of neuromuscular paralysis and the relationship to serum venom concentrations.
Ethical approval was granted by the research ethics review committees of the University of Peradeniya, Sri Lanka, the Rajarata University of Sri Lanka, and Monash University, Australia. Written informed consent was sought prior to the recruitment of all patients. For patients aged 16 and 17 years consent was also obtained from the patient’s parent or guardian.
This study is part of a prospective study of all snakebites admitted from April to October 2014 to the Teaching Hospital, Anuradhapura, a tertiary care centre and the largest hospital in the north central province of Sri Lanka. The region has one of the highest incidences of snakebite in Sri Lanka.
Patients aged 16 years or over with definite common krait (B. caeruleus) bites were recruited. Cases were authenticated as common krait bites by either expert snake identification or detection of krait venom in patient serum. All patients presenting with live or dead specimens of the offending snakes (Fig 1A), had the snake identified by one author (AS) who is a herpetologist. Patients with clinical suspicion of a common krait bite had serum tested for krait venom by venom-specific enzyme-immunoassay.
All patients with suspected krait envenoming had a complete neurological examination on admission to hospital, then repeat examinations every 2h for the first 24h post-bite, and then every 4h for the remainder of their hospital stay. After discharge patients had a full neurological examination at six weeks and six months post-bite to detect residual neurological impairment. The Medical Research Council scale for muscle strength[18] was used wherever appropriate. For the assessment of respiratory muscles, tidal volume was measured using a spirometer. Ptosis was graded from grade I to III using a visual analogue scale, with complete ptosis being grade III. Weak or sluggish eye movements in one or all directions was considered to be partial ophthalmoplegia and absent eye movements in all directions, complete ophthalmoplegia. At each time point patients were specifically examined for features of autonomic neurotoxicity (heart rate, blood pressure, lacrimation, sweating and salivation), central effects (level of consciousness and occulocephalic reflexes) and myotoxic effects (muscle pain and tenderness, both local and general).
Clinical assessments were done by one author (AS) or medically qualified clinical research assistants. All assessments performed by clinical research assistants were reviewed by AS and approximately one third were reviewed by another medically trained author (SS). In addition to a neurological assessment, all patients had a full clinical examination on admission, at 12h and 24h post-bite, and then daily until discharged. Local effects included pain, swelling, paraesthesia or regional lymphadenopathy and non-specific systemic effects were defined as headache, nausea, vomiting or abdominal pain. All assessments were recorded using a pre-formatted clinical data form.
Antivenom administration was decided by the treating physician. All patients who received antivenom in this study received 20 vials of Indian polyvalent antivenom as the first dose from VINS Bioproducts (batch numbers: 01AS11119, 01AS11121, 01AS13100, 01AS14025, 01AS14026, 01AS14035). No patient in this study received a second antivenom dose. The antivenom infusion was ceased briefly for 5–10min in patients who developed anaphylaxis. Antivenom reactions were treated according to the attending physician with adrenaline, antihistamines and corticosteroids.
During review visits patients had a routine physical and neurological examination. They were also questioned specifically about the presence of neuromuscular effects that prevented them performing routine daily work, and about any recovery of local effects.
Patients were classified into three groups based on the presence and severity of neurotoxicity. The first group included patients who developed no clinical evidence of neurotoxicity. The second group was patients with mild neurotoxicity defined as the presence of one of the following clinical features; ptosis, ophthalmoplegia or facial muscle weakness, but not bulbar, respiratory or limb weakness. The third group was severe neurotoxicity defined as patients developing paralysis that involved bulbar and respiratory muscles requiring mechanical ventilation.
Stimulated sfEMG of the orbicularis oculi muscle was performed using disposable concentric needle electrodes (diameter: 0.3mm) and a portable Medelec Synergy N2 EMG system (Medelec Synergy, UK). All tests were performed by AS at the bedside during the hospital admission and in a separate examination room during review visits at 6 weeks and 6 months. In order to establish the baseline values for stimulated sfEMG jitter for normal Sri Lankan subjects, 29 healthy individuals had sfEMG performed under the same conditions. The normal upper limits of jitter for individual fibres and individual subjects were established as 30μs and 20.6μs respectively[19].
Stimulation of the suprazygomatic branches of the facial nerve was by monopolar needle electrode at a frequency of 10Hz, with the stimulus delivered as rectangular pulses of 0.1ms. Stimulation intensity was increased until the appearance of visually discernible twitches of the orbital part of the orbicularis oculi muscle. Stimulus strength did not exceed 2mA. All recordings were done after the stimulus intensity had reached supra-threshold for the fibre being examined. Recording and analysis of the sfEMG followed Kouyoumdjian and Stålberg[20] and the jitter of individual fibres was expressed as a mean consecutive difference (MCD). Neuromuscular block was where there was no transmission across the neuromuscular junction in fibres on the sfEMG recording.
Blood samples from patients with suspected krait bites were collected on admission and at 1, 4, 8, 12 and 24h post-bite, and daily thereafter. All samples were immediately centrifuged, and serum aliquoted and frozen at -80°C. Venom was quantified using a sandwich enzyme immunoassay as previously described[21–23]. In brief, rabbit IgG antibodies (provided by University of Rajarata) were bound to microplates as well as conjugated to biotin for the sandwich enzyme immunoassay, detecting with streptavidin-horseradish peroxidase. The lower limit of detection for the assay was 0.2ng/ml.
Creatine kinase (CK) activity was measured in the serum samples of patients obtained 24h post-bite, or a close to this time point as possible if the 24h sample was unavailable. Creatine kinase was measured using a commercially available assay kit using the CK-NAC Reageant (Creatine Kinase, activated by N-acetylcysteine, Thermo Scientific, Middletown, VA, USA).
Analysis of clinical and sfEMG data was done using Prism, version 6.05 (GraphPad Software, Inc.). Continuous variables were reported as medians, range and interquartile range (IQR). Jitter values and creatine kinase concentrations of different patient groups were analysed using one-way ANOVA followed by multiple comparison tests.
During the study period 773 snakebite patients were admitted and 38 of these were suspected common krait bites. Thirty one patients had the offending snake positively identified as a common krait and two others were positive for krait venom in their blood. The remaining five cases were only suspected common krait bites based on circumstances of the bite, but had no features of local or systemic envenoming, despite having fang marks. None of these patients had a previous history of neurological disorders or therapeutic agents known to cause neurotransmission abnormalities. Table 1 provides the demographic features of the 33 confirmed cases of common krait bites.
Eight patients who brought the offending snake and had evidence of a bite with fang marks, did not develop neurotoxic features during their hospital stay. None of these received antivenom (Table 2).
Eight patients had mild neurotoxicity and six of these received antivenom (Table 2). All these patients had partial ptosis on admission. In one patient, this progressed to complete ptosis with partial ophthalmoplegia and mild facial weakness within 12h of the bite, but 2h after antivenom was given. Two patients not receiving antivenom did not have any progression of neurotoxicity. All eight patients were fully recovered the next day and were discharged 20 to 32h post-admission.
Seventeen patients developed severe neurotoxicity with a range of neurotoxic clinical features on admission (Table 2). Three patients appeared to have altered consciousness and had no respiratory effort on admission with oxygen saturations < 50%. These three were immediately intubated and ventilated. All 17 patients were given antivenom a median of 3.5h post-bite (2.8–7.2h). Antivenom therapy did not stop progression of neurotoxic features in any patient and intubation and mechanical ventilation was required within 7h of the snake bite in all 17 patients. The clinical features of neurotoxicity progressed in a descending order from eyes to peripheral limbs (Fig 2A). Five patients developed complete limb paralysis with areflexia occurring within 8h of the bite (Figs 1C and 2A).
Neuromuscular function began to recover in all 17 patients on average 30h post-bite (22–51h). Muscle groups recovered in the reverse order to that in which paralysis developed. Fig 2B shows the time scale of the recovery of neurotoxic features. Patients were extubated a median of 96h post-bite (54–216 h).
One patient with severe paralysis and no motor function appeared to be in a deep coma with absent brainstem reflexes 12h post-bite. There was no recordable response on sfEMG done 10h and 16h post-bite. The patient had slight movement of the toes in response to a sternal rub 26h post-bite, but no occulocephalic reflex. The occulocephalic reflex recovered 6h later, followed by gradual ascending recovery of all motor function and sfEMG jitter.
All 33 patients were discharged with no clinically detectable neuromuscular paralysis. However, on discharge 23 patients had no pain, touch or temperature sensation in an area of 3–8cm diameter around the bite site. Of 23 patients who received antivenom, 19 developed an adverse reaction and six of these developed severe anaphylaxis with hypotension (systolic blood pressure <90mmHg) and tissue hypoxia (SpO2<92%) that required resuscitation. Of these, three were intubated for the antivenom reaction before they developed respiratory muscle paralysis due to envenoming.
At 6 weeks, six of eight patients with no acute neurotoxicity, seven of eight with mild neurotoxicity and 14 of 17 with severe neurotoxicity were reviewed. None had clinically detectable neuromuscular paralysis. Sixteen patients (five mild and 11 severe neurotoxicity) still had absent pain, touch and temperature sensation around the bite site (4–7cm diameter area).
At 6 to 9 months after the krait bite, six with no acute neurotoxicity, five with mild neurotoxicity and ten with severe neurotoxicity were reviewed and there was no evidence of clinical neuromuscular dysfunction. All sensory abnormalities around the bite site had resolved and the patients stated the feeling of numbness around the bite site disappeared 2 to 3 months post-bite.
sfEMG were undertaken in seven non-envenomed patients, seven with mild neurotoxicity and ten with severe neurotoxicity. On admission, all patients with no neurotoxicity or mild neurotoxicity (n = 14) had a median MCD in jitter on sfEMG similar to normal controls (Figs 3A and 4A). Two patients with mild neurotoxicity had neuromuscular blockade in 3% and 12% of fibres respectively. The other patients had no neuromuscular block. Jitter was not markedly increased and no blocks were seen in the no neurotoxicity and mild neurotoxicity groups 6–12h post-bite (n = 5, n = 5; Fig 4B). The median MCD was also not different from normal on discharge, at 6 weeks and at 6 to 9 months in all patients in these two groups.
All patients with severe neurotoxicity had a high median MCD in jitter compared to normal, with most having neuromuscular blockade on admission (Figs 3B and 4A). At 6–12h post-bite, jitter was markedly increased and there was increased neuromuscular block in recorded fibres (Figs 4B, 5A and 5B). In two patients who had severe neurotoxicity, sfEMG recordings showed no response during this period (9 and 10.5h post-bite respectively), indicating complete neuromuscular blockade. Eleven patients received atracurium for rapid sequence intubation. The post-intubation sfEMG for these patients was done a median of 5.8h (4.5–23.2h) after receiving the atracurium dose.
In all patients with severe neurotoxicity, sfEMG24–36h post-bite showed improved jitter (decreased median MCD) and decreased blocks compared to 6–12h post-bite (Fig 5A and 5B). On discharge, the median MCD in jitter on the sfEMG was still abnormally high in 11 patients. Of these, three had 3 to 10% fibres with blocks (10–12 days post-bite). At 6 weeks, seven of the fourteen patients reviewed still had high MCD jitter values (>22.6μs) and six of them had blocks in 3 to 13% of the recorded fibres. At 6 to 9 months, ten of these patients were reviewed and all had normal jitter with no blocks recorded.
Krait venom was not detectable in the eight non-envenomed patients. Of the six patients with mild neurotoxicity who received antivenom, pre-antivenom blood samples of four patients were available but no venom was detected in any. The two patients with mild neurotoxicity who did not receive antivenom had no detectable venom in blood. Nine of the 17 patients with severe neurotoxicity had pre-antivenom blood samples available, and krait venom was detected in eight (0.3 to 52.2ng/ml) (Fig 6). None of the post-antivenom samples had detectable free venom, indicating rapid and persistent binding of venom by antivenom (Fig 6).
Serum samples were available in 32 of the 33 study participants and had a median creatine kinase concentration of 43 U/L (8 to 274 U/L). There were no significant differences in the creatine kinase concentrations between the mild, moderate and severely envenomed groups.
In this cohort of patients with definite common krait envenoming, about half developed life-threatening neuromuscular paralysis that did not appear to be prevented by or respond to antivenom treatment. Patients who had sfEMG performed had increased jitter and increased neuromuscular block that correlated with the clinical severity. The neurophysiological abnormalities improved in line with clinical recovery but were still abnormal 6 weeks after the bite, despite the patients being clinically normal. The prolonged high jitter during the recovery phase may represent immaturity of the motor nerve terminals undergoing the re-innervation process. Excepting the three patients who were intubated due to antivenom reactions, all other patients were intubated due to bulbar weakness and/or respiratory paralysis, which developed within 7h of the bite. This demonstrates that severe neuromuscular paralysis develops rapidly. Based on this finding it appears that patients who do not develop bulbar weakness or respiratory paralysis within 12h of the bite, are highly unlikely to develop severe paralysis. These figures are largely in agreement with previous reports from Sri Lanka[7,10].
Although krait venoms contain both pre-synaptic neurotoxins (β-bungarotoxins) and post-synaptic neurotoxins (α-bungarotoxins), it is generally accepted that the pre-synaptic neurotoxins are more important in human envenoming[24,25]. Presynaptic neurotoxins cause irreversible injury. For example, the major pre-synaptic neurotoxin from the Chinese many banded krait (Bunragus multicinctus), β-bungarotoxin, causes depletion of synaptic vesicles followed by destruction of the motor nerve terminals in the isolated mouse phrenic nerve-hemi-diaphragm preparation and the soleus muscle of the mouse hind limbs[12]. Common krait venom contains β1–β5-caerulotoxins, a group of toxins similar to β-bungarotoxin, that are most likely responsible for the paralysis in common krait bites.[26]
Ultrastructural damage and functional injury caused by β-bungarotoxin to the motor nerve terminals recovers over about 7 days[24]. In our study, clinically detectable neurotoxicity resolved 4 to 12 days post-bite, but subclinical neuromuscular dysfunction remained for at least 6 weeks in some patients. This may be due to some motor nerve terminals taking longer to fully recover and re-innervate muscle, and the already re-innervated muscle fibres compensating for recovering fibres. sfEMG directly measures transmission across the neuromuscular junction by measuring security of transmission to individual muscle fibres, so is more sensitive than other neurophysiological investigations and likely to detect abnormal neurotransmission in recovering fibres. Therefore, sfEMG measurements are likely to give a more accurate measure of the recovery time of neuromuscular dysfunction in snake envenoming.
One previous study reported abnormalities in nerve conduction one year after the snake bite, particularly in patients presumed to be bitten by elapids[27]. However, this study by Bell et al. only reported nerve conduction studies one year post-bite, with no comparison at the time of the bite. In addition, these nerve conduction studies are not as sensitive as sfEMG, making it difficult to interpret their results. sfEMG recordings in our study demonstrate the complete recovery of neuromuscular function at 6 months, even in the patients who had severe neuromuscular dysfunction within 24h post-bite, suggesting the findings of Bell et al. one year post-bite are unlikely to be related to the snake bite.
In all 17 patients with severe neuromuscular paralysis, the peak or most severe effects, both clinical and neurophysiological, were observed during the first 24h after the bite. Patients then began to recover during day two, with considerable improvement in both neuromuscular jitter and neuromuscular block compared to day one. In contrast, animal experiments show that the initiation of re-innervation with motor nerve terminal sprouting occurs three to five days after venom inoculation [12,25]. The reason for the more rapid and marked improvement on the second day in humans is unclear. One possible reason is that those muscle fibres of the orbicularis oculi which are already denervated due to toxin induced injury by the second day, fail to produce any response with electrical stimulation. This means they are no longer part of the sampled fibres in the sfEMG recording. Motor nerve terminals that were less damaged are then the ones that make up the majority of the sampled fibres falsely improving the measurement. Although we are unable to exclude this on sfEMG, there was also clinical recovery in patients during the second day, based on the re-appearance of deep tendon reflexes and the plantar reflex. Further investigation is required to understand the pathophysiological basis of these observations in human recovery.
Serum venom concentrations depend on multiple factors including the venom dose delivered, the rate of venom absorption and therefore the time post-bite, individual patient factors that affect the pharmacokinetics of the venom (e.g. effect of patient size/weight, renal function) and the sensitivity of the assay. Kraits inject very small amounts of venom during their bite, explaining the low venom concentrations in our study and previous studies.[9,28] The eight patients who had no detectable venom in blood and had no features of envenoming, were ‘dry bites’, as previously seen among patients with common krait bites[7,9]. The absence of detectable venom in patients with mild neurotoxicity is more difficult to explain. The most likely explanation is that the assay is not sensitive enough to detect the small amount of such a potent elapid venom that can cause minor toxicity. A similar phenomena is seen with Australian brown snake (Pseudonaja spp) venom where < 0.2ng/mL can result in a mild coagulopathy [29]. An additional explanation is that this is due to late sampling times after the peak venom concentrations, when venom has distributed out of the central compartment. Either way these patients are likely to have had small venom doses from the bite. This is also the likely explanation for the two patients with mild neurotoxicity who did not receive antivenom and did not worsen. In the severe neurotoxicity group, there was also large variations in the venom concentrations (Fig 6).
Presynaptic neurotoxins cause irreversible nerve injury, so neurotoxicity is expected not to respond to antivenom once it has developed[24]. Despite most patients receiving early antivenom and antivenom rapidly clearing free venom in blood, the paralysis worsened and required mechanical ventilation in all 17 patients for several days. In the mildly neurotoxic patients one patient progressed despite antivenom and two patients who did not receive antivenom had similar outcomes to those receiving antivenom. Antivenom cannot reverse neuromuscular injury and recovery occurs through the natural nerve terminal repair[24,25]. These results demonstrate that Indian polyvalent antivenom is efficacious (binds venom) but is not effective for common krait envenoming in Sri Lanka, because of the irreversibility of the pre-synaptic neurotoxicity.
Antivenom was able to clear circulating free venom, so given early enough antivenom may still be beneficial in preventing progression of neuromuscular dysfunction. This has been demonstrated in studies of Papuan taipan bites where early antivenom (<6h post-bite) reduced the number of patients requiring intubation[17]. Unfortunately, the majority of patients (19/23) who received antivenom in our study developed acute adverse reactions, including some with life threatening anaphylaxis. Therefore, the safety and benefits of antivenom need to be carefully weighed up along with the clinical status of the patient, before deciding on antivenom therapy.
The majority of patients in this study reached a primary care centre early, but because of concerns about antivenom reactions, antivenom was not usually administered prior to transfer to the study hospital. If Indian polyvalent antivenom had a lower reaction rate, this would encourage primary care doctors to administer antivenom as early as possible, and before transferring them to tertiary care hospital. Such an approach would help prevent neurotoxicity in the majority of cases, without risk of life-threatening adverse reactions.
Although generalized myalgia and muscle tenderness were observed in some patients, the normal serum creatine kinase concentrations in patients is consistent with common krait envenoming not causing myotoxicity. Mildly elevated serum myoglobin levels were previously reported in one envenomed krait patient in Sri Lanka,[28] but serum myoglobin is not a very specific marker of muscle injury. Myotoxicity has been reported in envenoming by other krait species, including B. niger [30], B. multicinctus [31] and B. candidus [32]. However, in the study of B. candidus there were only mild elevations of creatine kinase, and the study of B. multicinctus only reports myalgia.
Coma has been previously reported in common krait envenoming [7,33]. In one study, two patients with deep coma were reported to have electroencephalogram abnormalities, abnormal brain stem visual and auditory evoked potentials, leading to the conclusion that krait venom can cause cortical and brain stem effects [33]. However peptide and protein toxins are unlikely to cross the blood brain barrier making this theoretically unlikely. In the present study, one patient with severe paralysis had deep coma, absent brainstem reflexes and no sfEMG recordings. Interestingly, there was a period of time when the patient had absent brain stem reflexes but some motor function, suggesting that the patient was more likely to have had severe paralysis mimicking coma, rather than coma itself. Similar observations have previously been made in snakebite patients in India [34–37]. The altered consciousness observed in three patients on admission was most likely due to hypoxia secondary to respiratory muscle paralysis, rather than any direct central effect of the venom.
sfEMG jitter results can be influenced by pre-existing medical conditions that affect the peripheral nervous system, such as myasthenia gravis, diabetes mellitus and leprosy. None of the patients in this study had a history of any of these conditions. Two-thirds of the patients were farmers who may have had pre-existing neurotransmission abnormalities secondary to chronic exposure to organophosphates. However, we did not see a difference in the jitter values of the present cohort of patients at 6 months compared to the normal subjects, so this is unlikely.
A limitation of the study was that sfEMG was only performed on the orbicularis oculi muscle. This was done because it is one of the muscles affected earliest in snake bite paralysis and it is convenient to access. The neuromuscular jitter and blocking correlated well with the clinical picture indicating that this muscle is likely to be representative of the neurophysiology of the neuromuscular paralysis in snake envenomed humans. Another limitation of the study was that the recovery of certain muscles, e.g. neck extensors, buccinator, bulbar muscles, could not be assessed while patients were intubated. In addition, while the patients were sedated, assessment of the power of voluntary contractions of muscles was not possible. Hence the exact sequence of muscle involvement, particularly during the recovery, could not be documented in some patients. Finally, it is important to consider the risk of sfEMG in patients with coagulopathy.
Our study highlights the usefulness of sfEMG as a biomarker, particularly as a research tool on snakebite neurotoxicity. The sfEMG and clinical findings suggest that recovery occurs more rapidly than expected, based on animal studies, and further work is required to explain this. The study confirmed that antivenom did not reverse neurotoxicity and that if antivenom is going to prevent neurotoxicity, it must be given much earlier, prior to development of any neurotoxic effects. A placebo controlled trial is required to determine if antivenom hastens the recovery in established neurotoxicity.
|
10.1371/journal.ppat.1007101 | An orphan kinesin controls trypanosome morphology transitions by targeting FLAM3 to the flagellum | Trypanosoma brucei undergoes life cycle form transitions from trypomastigotes to epimastigotes in the insect vector by re-positioning the mitochondrial genome and re-locating the flagellum and flagellum-associated cytoskeletal structures. The mechanism underlying these dramatic morphology transitions remains poorly understood. Here we report the regulatory role of the orphan kinesin KIN-E in controlling trypanosome morphology transitions. KIN-E localizes to the flagellum and is enriched at the flagellar tip, and this localization depends on the C-terminal m-calpain domain III-like domains. Depletion of KIN-E in the trypomastigote form of T. brucei causes major morphology changes and a gradual increase in the level of EP procyclin, generating epimastigote-like cells. Mechanistically, through its C-terminal importin α-like domain, KIN-E targets FLAM3, a flagellar protein involved in morphology transitions, to the flagellum to promote elongation of the flagellum attachment zone and positioning of the flagellum and flagellum-associated cytoskeletal structure, thereby maintaining trypomastigote cell morphology. Our findings suggest that morphology transitions in trypanosomes require KIN-E-mediated transport of FLAM3 to the flagellum.
| Trypanosoma brucei, the causative agent of sleeping sickness in humans and nagana in cattle in sub-Saharan Africa, has a complex life cycle by alternating between the tsetse fly vector and the mammalian hosts. In the gut of tsetse flies, trypanosomes undergo life cycle transitions from the trypomastigote form to the epimastigote form by re-positioning the mitochondrial genome and re-locating the flagellum and flagellum-associated cytoskeletal structures. Previous work demonstrated that elongation of the flagellum attachment zone plays an important role in controlling morphology transitions, but how it is regulated remains poorly understood. This work discovered that an orphan kinesin plays an essential role in regulating trypanosome morphology transitions. This novel kinesin localizes to the flagellum and targets FLAM3, one of the two flagellar proteins involved in morphology transitions, to the flagellum. This work suggests that trypanosome morphology transitions require kinesin-mediated transport of FLAM3 to the flagellum to promote the elongation of the flagellum attachment zone, thereby maintaining flagellum-cell body attachment and positioning the flagellum and flagellum-associated cytoskeletal structures to assume trypomastigote cell morphology.
| Trypanosomatids, a group of protozoan parasites consisting of Trypanosoma brucei, Trypanosoma cruzi, and Leishmania spp., transition to different developmental forms with distinct cell morphology during their life cycle between the insect vectors and the mammalian hosts. These life cycle forms are distinguished by the relative position of the kinetoplast, the cell’s mitochondrial genome, to the nucleus, the cellular location from which the flagellum emerges, and the length of the free, unattached flagellum [1]. Inside the proventriculus of the tsetse fly vector, T. brucei differentiates from trypomastigote form to epimastigote form, which then undergoes an asymmetrical cell division and further develops to metacyclic form, the mammal-infective form of the parasite, in the salivary gland [1]. Although the molecular mechanisms underlying the transitions between these life cycle forms in trypanosomatids remain poorly understood, several proteins, including some RNA-binding proteins and a few flagellum-associated cytoskeletal proteins, were recently found to be involved in life cycle transitions in T. brucei [2,3,4,5,6,7]. The involvement of RNA-binding proteins ALBA3/4 [3] and RBP6 [2] in trypanosome life cycle transitions suggests a posttranscriptional regulation scheme, but mechanistically how these proteins contribute to this process is still elusive. The involvement of two flagellum attachment zone (FAZ) proteins in the flagellum, ClpGM6 and FLAM3 [4,5], and two intracellular FAZ proteins, FAZ9 [6] and TbSAS-4 [7], in life cycle form transitions suggests that the morphology transitions require the modulation of flagellum-associated cytoskeletal structures mediated by these FAZ proteins.
Kinesins are evolutionarily conserved microtubule-based motor proteins performing crucial roles in regulating microtubule dynamics and intracellular transport [8]. T. brucei possesses an expanded repertoire of kinesin-like proteins, including 13 kinetoplastid-specific kinesins and 15 orphan kinesins, most of which are of unknown function [9]. Previous work on Aurora B kinase-associated proteins identified two orphan kinesins, KIN-A and KIN-B, as nucleus- and spindle-associated kinesin proteins required for spindle assembly and chromosome segregation in T. brucei [10]. Given the essential roles of KIN-A and KIN-B in mitosis, they may function to compensate for the absence of mitotic kinesin homologs, such as the spindle motor protein BimC, the central spindle kinesin MKLP1/Pavarotti/ZEN-4, or the chromokinesin KLP3A, in T. brucei. Other studies uncovered the requirement of two kinetoplastid-specific kinesins, KIN-C and KIN-D, for maintaining cell morphology by modulating the organization of the subpellicular microtubule corset [11,12,13]. These findings highlighted the diverse cellular functions of kinetoplastid-specific kinesins and orphan kinesins, and further motivated us to explore the function of other kinesins in T. brucei.
To understand the potential roles of other orphan kinesins in T. brucei, their subcellular localization and biological functions were investigated in the procyclic form of T. brucei. Here we report one of the orphan kinesins, which is encoded by Tb927.5.2410 and was named KIN-E, and its essential function in targeting FLAM3, a FAZ flagellum domain protein crucial for trypanosome morphology transitions [4,14], to the flagellum to promote FAZ elongation and organelle positioning, thus maintaining flagellum-cell body attachment. These findings identify a new regulator and its mechanistic role in controlling trypanosome life cycle form transitions.
In an attempt to understand the function of the orphan kinesins in T. brucei, we first determined the subcellular localization of the remaining 13 orphan kinesins, each of which was tagged with a triple HA epitope at their respective endogenous locus, by immunofluorescence microscopy. One of these orphan kinesins, which is encoded by Tb927.5.2410, was found to localize to the flagellum and was enriched at the flagellar tip, and thus was further characterized. We named it KIN-E, following previous nomenclature of the four kinetoplastid-specific kinesins and orphan kinesins (KIN-A to KIN-D) [10,11,12,13]. KIN-E contains an N-terminal kinesin motor domain (MD), which comprises a conserved nucleotide-binding domain, consisting of a highly conserved P-loop (phosphate-binding loop) motif and two conserved switch motifs (switch I and switch II), and a conserved microtubule-binding motif (Fig 1A). Swiss modeling [15] showed that KIN-E contains an unusual importin α-like domain, which is about half size of the yeast importin α protein (Fig 1A–1C), and two m-calpain domain III-like domains, which is characterized by an antiparallel β-sandwich of a pair of four β-sheets [16] (Fig 1A, 1D and 1E), in addition to a small coiled-coil motif at the C-terminus (Fig 1A). The two m-calpain domain III-like domains (abbreviated as mCL#1 and mCL#2) in KIN-E share similar folds with the domain III of the human m-calpain (Fig 1D and 1E), suggesting that mCL#1 and mCL#2 in KIN-E may possess similar functions. KIN-E appears to be different from many known kinesins that contain mostly coiled-coil motifs at their C-terminus [17].
The subcellular localization of KIN-E during the cell cycle of T. brucei was investigated by immunofluorescence microscopy. Endogenously 3HA-tagged KIN-E is enriched at the distal tips of both the new and old flagella throughout the cell cycle and also localizes along the entire length of the flagella at a lower level (Fig 2A). At the distal tip of the new flagellum, KIN-E partly overlaps with the flagella connector protein FC1 [6] (Fig 2B). To investigate the potential contribution of the importin α-like domain and the two m-calpain domain III-like domains to KIN-E localization, we ectopically expressed KIN-E mutants deleted of the importin α-like domain (KIN-E-ΔIMPα) or the two m-calpain domain III-like domains (KIN-E-ΔmCL) in the 29–13 cell line, and then examined the subcellular localization of these mutants by immunofluorescence microscopy. The KIN-E-ΔIMPα mutant, which lacks the importin α-like domain, is still localized to the flagellum and is enriched at the flagellar tip (Fig 2C, arrow), similar to the wild-type KIN-E (Fig 2C, arrow), suggesting that the importin α-like domain is not required for KIN-E localization. Intriguingly, the KIN-E-ΔmCL mutant is not localized at the flagellum and the flagellar tip, but instead at the posterior end of the cells (Fig 2C, arrowhead), indicating that the m-calpain domain III-like domains in KIN-E are required for targeting KIN-E to the flagellum. Given that kinesins are microtubule plus end-directed motor proteins, it is likely that the KIN-E-ΔmCL mutant is directed to the cell posterior, the plus ends of the cytoskeletal subpellicular microtubules in T. brucei.
We next investigated the function of KIN-E in the procyclic form of T. brucei by RNAi. Induction of RNAi by tetracycline causes a gradual decrease of KIN-E protein, which was endogenously tagged with a triple HA epitope, to the level of <10% of the control level after RNAi induction for 4 days and beyond (Fig 3A). Knockdown of KIN-E causes a severe growth defect (Fig 3B) and a drastic change in cell morphology, resulting in the production of epimastigote-like cells among almost 80% of the 1N1K cells after RNAi induction for 48 h (Fig 3C). Epimastigote-like morphology is characterized by a re-positioned kinetoplast, which is either juxtaposed or anterior to the nucleus, and a long unattached flagellum (Fig 3D). Scanning electron microscopy further confirmed the epimastigote-like morphology of the cells with a single flagellum (Fig 3E, panel b). Among the cells with two flagella, ~80% of them possess a long, unattached new flagellum and a normal-length old flagellum (Fig 3E, panle d), and the rest of them possess a long, unattached new flagellum and a long, unattached old flagellum (Fig 3E, panel f). The latter type of two-flagella cells are likely developed from the epimastigote-like one-flagellum cells following further cell cycle progression. While KIN-E RNAi induction for shorter times such as 24 and 48 hours causes major changes in cell morphology, RNAi induction for longer times results in the accumulation of multi-nucleated (>2) cells up to 75% of the total population after 7 days (Fig 3F), suggesting that cytokinesis is arrested after prolonged RNAi induction.
The generation of cells with epimastigote-like morphology by KIN-E RNAi raised the question of whether these epimastigote-like cells, other than morphology resemblance, also possess certain biological features of the epimastigote form. During the development of T. brucei from the trypomastigote form to the epimastigote form within the gut of tsetse flies, a prominent change is the alteration of the expression of EP and GPEET procyclins [18,19]. During early stages (day 3) of fly infection, only GPEET procyclin is expressed, but later in the infection (day 7), GPEET procyclin disappears and EP procyclin is expressed [19]. EP procyclin is also found in the epimastigote form of T. brucei [20]. To investigate the relative expression of EP procyclin after KIN-E RNAi, we carried out flow cytometry, which showed that the level of EP procyclin was gradually increased after KIN-E RNAi induction (Fig 3G). Quantitative western blotting showed that the level of EP procyclin was increased to ~3.5-fold of the control level after 3 days of RNAi (Fig 3H and 3I). These results suggest that KIN-E RNAi not only produced cells of epimastigote-like morphology but also altered certain biological features of the trypomastigote cells, such as the expression of EP procyclin, towards the epimastigote form. Since the anti-GPEET procyclin antibody obtained from Cedarlane Labs failed to detect GPEET procyclin, the expression of GPEET procyclin in KIN-E RNAi cells was unable to be investigated.
Given that the epimastigote-like 1N1K cells from KIN-E RNAi possess a re-positioned kinetoplast and a long, unattached flagellum, we examined whether KIN-E RNAi affected biogenesis and/or elongation of the FAZ and the flagellum. To this end, we immunostained the 1N1K cells with anti-CC2D antibody to label the FAZ and with L8C4 (anti-PFR2 antibody) to stain the paraflagellar rod in the flagellum (Fig 4A), and then measured the length of the FAZ and the flagellum (Fig 4B). Additionally, we also measured the length of the unattached flagellum and the cell body, the distance between kinetoplast and nucleus, and the distance between kinetoplast and cell posterior (Fig 4B). The results showed that the 1N1K cells from KIN-E RNAi possess a significantly shorter FAZ than the control 1N1K (an average length of 4.7 ± 0.3 μm vs 12.9 ± 0.2 μm, n = 104 for RNAi cells and n = 101 for control cells) (Fig 4A and 4B), and the unattached flagellum of the RNAi cells is more than four times longer than the unattached flagellum of the control cells (an average length of 11.9 ± 1.6 μm vs 2.5 ± 0.2 μm) (Fig 4A and 4B). The 1N1K cells from KIN-E RNAi also appear to be significantly smaller in size than the control 1N1K cells (an average cell body length of 13.7 ± 0.5 μm vs 18.6 ± 0.8 μm) (Fig 4A and 4B). Additionally, the average distance between the kinetoplast and the nucleus is significantly reduced from 3.7 ± 0.4 μm to 1.5 ± 0.1 μm upon KIN-E RNAi (Fig 4B) and consequently, the average distance from the kinetoplast to the cell posterior is increased from 4.9 ± 0.4 μm to 6.2 ± 0.2 μm (Fig 4B). These epimastigote-like 1N1K cells also contain re-positioned cytoskeletal structures, such as the flagellar basal body (Fig 4C), the bilobed structure, and the flagellar pocket (Fig 4D), all of which are anterior to the nucleus rather than posterior to the nucleus as in control cells (Fig 4C and 4D).
The morphological changes observed in 1N1K cells are also detected in 2N2K cells (Fig 5). The new FAZ of the RNAi cells is significantly shorter than that of the control cells (an average length of 3.1 ± 0.08 μm vs 9.6 ± 0.7 μm, n = 97 for RNAi cells and n = 103 for control cells), and the new, unattached flagellum of the RNAi cells is significantly longer than that of the control cells (an average length of 9.5 ± 0.2 μm vs 1.32 ± 0.07 μm) (Fig 5A and 5B). However, the old FAZ and the old flagellum are not affected (Fig 5A and 5B), indicating that RNAi of KIN-E only disrupts the elongation of the new FAZ. It should be noted that some 2N2K cells also possess a longer unattached old flagellum (Figs 5B and 3E), but these cells likely are developed from the 1N1K cells with a longer unattached flagellum. Another change observed in the 2N2K cells is the reduced inter-kinetoplast distance (an average length of 1.8 ± 0.1 μm in RNAi cells vs 5.02 ± 0.01 μm in control cells) and the increased distance between the posterior kinetoplast to the cell posterior (an average length of 6.7 ± 0.2 μm in RNAi cells vs 3.8 ± 0.1 μm in control cells) (Fig 5A and 5B), suggesting that migration of the posterior kinetoplast is defective. Moreover, there is also a defective migration of other cytoskeletal structures, such as the posterior basal body (Fig 5C), the posterior bilobe structure (Fig 5D), and the posterior flagellar pocket (Fig 5D). These results demonstrated that RNAi of KIN-E abolishes the migration of the posterior kinetoplast and its associated basal body, the posterior bilobe structure, and the posterior flagellar pocket towards the cell posterior. This likely is due to the defective elongation of the new FAZ (Fig 5A), as it was previously proposed that basal body positioning is controlled by FAZ elongation [21].
The requirement of the two unusual structural motifs for KIN-E function was investigated using the KIN-E RNAi complementation cell lines in which wild-type KIN-E or KIN-E mutants were ectopically expressed in KIN-E-3’UTR RNAi background. RNAi of KIN-E by targeting its 3’UTR and ectopic expression of KIN-E or its mutants were induced by incubating with tetracycline. Ectopically expressed KIN-E and KIN-E-ΔIMPα in KIN-E-3’UTR RNAi cells localized to the flagellar tip, whereas KIN-E-ΔmCL localized to the posterior tip of the cell (Fig 6A). Knockdown of endogenous KIN-E protein, which was tagged with an N-terminal PTP epitope, and ectopic overexpression of KIN-E and its mutants, which were tagged with a C-terminal triple HA epitope, were confirmed by western blotting (Fig 6B). RNAi of KIN-E by targeting the 3’UTR of KIN-E also causes a growth defect (Fig 6C) and produces epimastigote-like cells (Fig 6D), similar to the RNAi by targeting the coding region of KIN-E (Fig 3B and 3C). Ectopic expression of wild-type KIN-E in KIN-E-3’UTR RNAi background restores cell growth to the rate of non-induced control cells (Fig 6C) and restores trypomastigote morphology (Fig 6D), demonstrating the rescue of the RNAi cells by ectopically expressed KIN-E. Expression of KIN-E-ΔIMPα and KIN-E-ΔmCL, however, does not restore cell growth (Fig 6C) and still produces epimastigote-like cells (Fig 6D), indicating that the importin α-like domain and the m-calpain domain III-like domains are required for KIN-E function.
Previous studies showed that depletion of two flagellum FAZ domain proteins, ClpGM6 and FLAM3, causes major changes in cell morphology in the trypomastigote form of T. brucei [4,5], similar to the phenotypes caused by KIN-E RNAi (Figs 3–5), suggesting that KIN-E may function in the same pathway as ClpGM6 and FLAM3. To test whether KIN-E interacts with the two proteins, we carried out co-immunprecipitation experiments. Our numerous attempts to tag ClpGM6 with a triple HA epitope at either the N-terminus or the C-terminus failed due to unknown reasons; therefore, our efforts were focused on FLAM3. Immunoprecipitation of FLAM3, which was tagged with a C-terminal PTP epitope from its endogenous locus, is able to pull down KIN-E, which was endogenously tagged with a triple HA epitope in the same cell line (Fig 7A), indicating that KIN-E interacts with FLAM3 in vivo in T. brucei.
Given that KIN-E interacts with FLAM3 in vivo in T. brucei, we investigated their co-localization. FLAM3 localizes to the flagellum and the flagella connector, but its level is much reduced at the unattached flagella near the distal flagellar tips [4,14]. At the distal tip of the old flagellum, KIN-E does not co-localize with endogenously PTP-tagged FLAM3 (Fig 7B), in agreement with the lack of FLAM3 at the old flagellar tip in 1N2K and 2N2K cells [14]. At the flagella connector in 1N2K and 2N2K cells during late S phase to mitosis, however, KIN-E co-localizes with FLAM3 (Fig 7B). During cytokinesis when the flagella connector is dissolved, KIN-E at the tip of the new flagellum does not co-localize with FLAM3 (Fig 7B). It should be noted that KIN-E in the flagella, albeit at a lower level, also co-localize with FLAM3 except at the unattached flagella near the distal tips (Fig 7B).
We next asked whether FLAM3 is required for KIN-E localization. To this end, KIN-E was endogenously tagged with a C-terminal triple HA epitope in cells harboring the FLAM3 RNAi construct. Control and FLAM3 RNAi cells were then immunostained with anti-HA antibody to detect TbKIN-E-3HA and anti-CC2D antibody to label the FAZ. Efficient knockdown of FLAM3 was confirmed by western blotting (S1 Fig), and depletion of FLAM3 produces cells with epimastigote-like morphology (Fig 7C), similar to the previous report [4]. In these epimastigote-like cells, KIN-E is still detectable in the flagellum and enriched at the distal tip of the flagellum in all (>1,000) of the cells examined (Fig 7C, arrow), demonstrating that FLAM3 is not required for KIN-E localization.
Conversely, the effect of KIN-E depletion on FLAM3 localization was also examined. FLAM3 was endogenously tagged with a PTP epitope in cells harboring the KIN-E RNAi construct, and immunofluorescence microscopic analysis of control and KIN-E RNAi-induced cells showed that KIN-E RNAi disrupts the localization of FLAM3, causing its mis-localization to the cytosol (Fig 7D). The effect on the subcellular distribution of FLAM3 was further investigated by western blotting, which showed that FLAM3 is partly distributed to the soluble cytosolic fraction upon KIN-E RNAi for 24 h and then exclusively distributed to the cytosol after RNAi induction for 48 h and beyond (Fig 7E), in contrast to its exclusive distribution in the cytoskeletal pellet fraction in non-induced control cells (Fig 7E). Together, these results suggest that KIN-E is required for targeting FLAM3 to the flagellum.
The generation of KIN-E RNAi complementation cell lines allowed us to investigate which domain(s) in KIN-E is required for interaction with FLAM3. To this end, FLAM3 was endogenously tagged with an N-terminal PTP epitope in the RNAi complementation cell lines described above, and co-immunoprecipitation was carried out. The results showed that immunoprecipitation of FLAM3 pulls down wild-type KIN-E and KIN-E-ΔmCL, but not KIN-E-ΔIMPα (Fig 8A), indicating that the importin-α-like domain is required for interacting with FLAM3. We next carried out GST pull-down experiments using purified recombinant GST-fused importin-α-like domain and GST-fused m-calpain domain III-like domain as the baits, and the results showed that the importin-α-like domain, but not the m-calpain domain III-like domain pulls down FLAM3 (Fig 8B), demonstrating that the importin-α-like domain mediates the interaction between KIN-E and FLAM3.
We next examined the localization of FLAM3 in KIN-E mutants by immunofluorescence microscopy. In KIN-E-3’UTR RNAi cells expressing 3HA-tagged KIN-E, PTP-tagged FLAM3 localizes to the flagellum, as in the non-induced control cells (Fig 8C) and as reported previously [4,14]. However, in KIN-E-3’UTR RNAi cells expressing 3HA-tagged KIN-E-ΔIMPα, PTP-tagged FLAM3 is mis-localized to the cytosol (Fig 8C) despite that KIN-E-ΔIMPα still localizes to the flagellum and the flagellar tip (Fig 8C). This result is consistent with the observation that KIN-E-ΔIMPα does not interact with FLAM3 (Fig 8A). In KIN-E-3’UTR RNAi cells expressing KIN-E-ΔmCL, FLAM3 is detected at the posterior end of the cell, where it co-localizes with KIN-E-ΔmCL (Fig 8C). The re-distribution of FLAM3 from the flagellum to the posterior cell end likely is attributed to the re-distribution of KIN-E-ΔmCL, which is still capable of interacting with FLAM3 (Fig 8A).
The distribution of FLAM3 between cytosolic and cytoskeletal fractions in these RNAi complementation cell lines was assessed by western blotting. In non-induced control cells and KIN-E-3’UTR RNAi cells expressing wild-type KIN-E, FLAM3 is detected in the insoluble, cytoskeletal fraction (Fig 8D). However, in KIN-E-3’UTR RNAi cells expressing KIN-E-ΔIMPα, FLAM3 is gradually shifted to the soluble, cytosolic fraction upon tetracycline induction and is exclusively in the soluble fraction after induction for 72 h (Fig 8D), consistent with its subcellular localization to the cytosol (Fig 8C). In these cells, KIN-E-ΔIMPα is still detected in the cytoskeletal fraction (Fig 8D), similar to the wild-type KIN-E (Fig 8D), suggesting the dissociation of FLAM3 from KIN-E-ΔIMPα due to lack of interactions (Fig 8A). In KIN-E-3’UTR RNAi cells expressing KIN-E-ΔmCL, FLAM3 is detected in the soluble, cytosolic fraction, together with KIN-E-ΔmCL (Fig 8D), suggesting that at the posterior cell tip, the mis-localized FLAM3 and KIN-E-ΔmCL either do not associate with the microtubule cytoskeleton or associate with the cytoskeleton in a manner that cannot tolerate detergent treatment.
Morphology transitions during trypanosome life cycle development appear to involve the modulation of the length of the FAZ through a cohort of FAZ flagellum domain proteins, such as ClpGM6 [5] and FLAM3 [4,14], and some intracellular FAZ proteins, such as FAZ9 [6] and TbSAS-4 [7]. Here an orphan kinesin, KIN-E, was found to play an essential role in controlling morphology transitions in T. brucei (Fig 3). KIN-E is unusual in that it contains in its C-terminus an importin α-like domain and two m-calpain domain III-like domains (Fig 1), which have not been found in any other kinesins in any organisms. The importin α-like domain and the m-calpain domain III-like domains play distinct roles in regulating KIN-E function. The importin α-like domain is not required for KIN-E localization (Fig 6A), but it is essential for KIN-E function (Fig 6C) and for interacting with FLAM3 and transporting the latter to the flagellum (Fig 8C). The m-calpain domain III-like domains are not involved in binding to FLAM3 (Fig 8A and 8B), but are essential for KIN-E localization to the flagellum (Fig 8C) and for KIN-E function (Fig 6C). The domain III in calpain, a calcium-dependent cysteine protease in vertebrates, functions as a calcium-regulated phospholipid-binding domain [22]. Thus, the two m-calpain domain III-like domains in KIN-E may also be capable of binding to calcium and lipid. The biochemical function and the mechanistic role of the m-calpain domain III-like domains in mediating KIN-E localization require further investigation.
Most kinesins are microtubule plus end-directed motor proteins that transport cargos along the microtubules to their cellular destinations [8]. The enrichment of KIN-E at the distal tip of the flagellum (Fig 2), where the plus ends of the axonemal microtubules are located, and the localization of the KIN-E-ΔmCL mutant to the posterior tip of the cell (Fig 6A), where the plus ends of the subpellicular microtubules are located [23], suggest that KIN-E is plus end-directed and may transport cargos toward the distal tip of the flagellum along the axonemal microtubules. FLAM3 has been identified as one of the cargos that KIN-E transports to the flagellum. This important point was demonstrated by several lines of evidence. First, FLAM3 interacts with KIN-E in vivo and co-localizes with KIN-E at the flagella connector region and, to a less extent, along the most part of the flagellum (Fig 7A and 7B). Second, depletion of KIN-E disrupts FLAM3 localization, resulting in the distribution of FLAM3 to the cytosol (Fig 7D and 7E). Finally, in cells expressing the KIN-E-ΔmCL mutant that still interacts with FLAM3 (Fig 8A), FLAM3, together with KIN-E-ΔmCL, is re-directed to the posterior end of the cell (Fig 8C). In contrast, in cells expressing the KIN-E-ΔIMPα mutant that still localizes to the flagellar tip but does not interact with FLAM3, FLAM3 is not targeted to the flagellum (Fig 8C), providing another line of evidence to demonstrate that targeting of FLAM3 to the flagellum depends on interaction with KIN-E.
Proteins and vesicles that are transported into the flagellum need to pass through the transition zone, a structural junction between the basal body and the flagellar/ciliary axoneme [24]. At the proximal end of the transition zone, a terminal plate crosses the transition zone, and it contains pores for the passage of intraflagellar transport (IFT) trains, which deliver axonemal components to the flagellar tip [24]. At the distal end of the basal body, striated transitional fibers radiate to join the plasma membrane [25], and the junction between the transitional fibers and the plasma membrane serves as a docking site for IFT trains [26]. We speculate that KIN-E may carry FLAM3 and other unidentified cargos through the transition zone [27] to deliver cargos to the flagellum. Given the requirement of the m-calpain domain III-like domains for KIN-E localization (Figs 2C and 6A), we postulate that these domains may mediate the trafficking of KIN-E and its cargos to the transition zone or the loading of them onto the transitional fibers before passing through the terminal plate of the transition zone to enter the flagellum. Within the flagellum/cilium, the axonemal microtubule doublets function as double-track railways to transport cargos bi-directionally, with the B-microtubules transporting anterograde IFT trains and the A-microtubules transporting retrograde IFT trains [28]. In trypanosome flagellum, anterograde IFT trains also move on a restricted set of axonemal microtubules [29]. It is thus possible that KIN-E also travels along the B-microtubules of the axonemal microtubule doublets to deliver FLAM3 and other cargos to the distal tip of the flagellum for elongation of the FAZ.
KIN-E-mediated transport of FLAM3 likely is independent of the Kinesin-2-mediated IFT system in T. brucei flagellum [30], as IFT is essential for flagellum assembly [31], but KIN-E is not required for flagellum formation (Fig 5A and 5B). KIN-E also appears to act independently of the flagella connector [32] that functions at the membrane junction between the new and old flagella, despite it partly overlaps with the flagella connector 1 (FC1) protein (Fig 2B), as KIN-E is not identified by flagella connector protein 1 (FCP1) immunoprecipitation [33] and depletion of KIN-E does not disrupt the connection of the new flagellum to the old flagellum (Fig 5C and 5D).
RNAi of KIN-E, FLAM3, and ClpGM6 all produce epimastigote-like cells, but there appear to be significant differences in cell proliferation between ClpGM6 RNAi cells and KIN-E and FLAM3 RNAi cells. Knockdown of ClpGM6 does not affect cell proliferation [5], whereas depletion of KIN-E or FLAM3 inhibits cell proliferation and generates multi-nucleated cells (Fig 3F and [4,14]). Although the detailed mechanism underlying this distinction is unknown, the phenotypic difference may be attributed to their differential effect on the length of the FAZ. RNAi of ClpGM6 produces epimastigote-like cells that contain a FAZ of 5–10 μm in length [5], whereas RNAi of FLAM3 [4] and RNAi of KIN-E (Fig 4) both produce epimastigote cells that contain a FAZ of less than 3 μm in length. It was proposed that the critical minimum length of the FAZ is 3 μm, and the FAZ of less than 3 μm in length is unable to support cytokinesis [4]. In this regard, KIN-E and FLAM3 appear to function similarly in controlling morphology transitions and cell proliferation. Interestingly, previous work showed that ClpGM6 and FLAM3 are co-immunoprecipitated from trypanosome lysate and are interdependent in maintaining their protein levels [4]. The current work showed that KIN-E and FLAM3 interact in trypanosomes and FLAM3 localization depends on KIN-E, but not vice versa (Fig 7). Given the effect of KIN-E depletion on FLAM3 and the interdependence between FLAM3 and ClpGM6, it is possible that KIN-E RNAi may also affect ClpGM6 localization in an indirect manner through disrupting FLAM3 localization (if ClpGM6 is not a cargo for KIN-E). It would also be interesting to investigate whether ClpGM6 is similarly targeted to the flagellum by KIN-E.
The morphological differences between the trypomastigote form and the epimastigote form lie in the length of the FAZ and the position of the mitochondrial genome, the flagellar pocket, the flagellum, and flagellum-associated structures relative to the nucleus [1]. KIN-E-deficient cells possess characteristic features of the epimastigote form with a shorter FAZ and re-positioned organelles/cytoskeletal structures, including flagellum, flagellar pocket, basal body, bilobe, and kinetoplast, to the anterior of nucleus (Figs 3 and 4). The fact that depletion of KIN-E in trypomastigote cells produces epimastigote-like cells suggests that KIN-E functions to maintain trypomastigote cell morphology, likely by promoting FAZ elongation and positioning organelles and cytoskeletal structures. However, given that FAZ elongation appears to control basal body positioning [21], it is likely that the defects in organelle/cytoskeletal structure positioning caused by KIN-E depletion are attributed to defective FAZ elongation, which is attributed to the failure to target FLAM3, a known FAZ flagellum domain protein required for FAZ elongation and life cycle form morphology transitions [4,14], to the flagellum (Fig 7). The trypomastigote and epimastigote forms also differ in certain biological features, including the distinct expression levels of the procyclins. During the development of T. brucei from the trypomastigote form to the epimastigote form in the gut of tsetse flies, the level of GPEET procyclin gradually decreases, but the level of EP procyclin gradually increases [19]. Intriguingly, KIN-E RNAi cells also have higher levels of EP procyclin (Fig 3G–3I), suggesting that the epimastigote-like cells produced by depletion of KIN-E also possess certain biological features of the epimastigote form.
In summary, we identified a novel function of the orphan kinesin KIN-E in controlling morphology transitions in T. brucei and uncovered the mechanistic role of KIN-E in targeting the FAZ flagellum domain protein FLAM3 to the flagellum to promote FAZ elongation, thereby maintaining flagellum-cell body attachment and positioning the flagellum and flagellum-associated cytoskeletal structures to assume trypomastigote cell morphology.
The procyclic form of T. brucei strain 427 was grown at 27°C in SDM-79 medium containing 10% heat-inactivated fetal bovine serum. The procyclic form of T. brucei strain 29–13 [34] was cultured at 27°C in SDM-79 medium supplemented with 10% heat-inactivated fetal bovine serum (Atlanta Biologicals, Inc), 15 μg/ml G418, and 50 μg/ml hygromycin. Cells were sub-cultured by 1/10 dilution with fresh medium whenever the cell density reached 5×106/ml.
To generate the KIN-E RNAi cell line, a 623-bp DNA fragment (nucleotides 235–866) corresponding to the N-terminal coding region of KIN-E was cloned into the pZJM vector [35]. To generate the FLAM3 RNAi cell line, a 938-bp DNA fragment (nucleotides 1004–1941) corresponding to the middle portion of the coding region of FLAM3 was cloned into the pZJM vector. The same DNA fragment of FLAM3 gene was used for RNAi previously [4]. The pZJM-KIN-E and pZJM-FLAM3 plasmids were each linearized with NotI and transfected into the 29–13 strain by electroporation. Transfectants were selected with 2.5 μg/ml phleomycin in SDM-79 medium containing 15 μg/ml G418 and 50 μg/ml hygromycin, and then cloned by limiting dilution in a 96-well plate. RNAi was induced by incubating the cells with 1.0 μg/ml tetracycline. Growth of cells was monitored daily by counting the cells with a hemacytometer. Three independent clones were selected and analyzed, which all generated almost identical phenotypes. Only the results obtained from characterizing one clone was presented.
For RNAi complementation experiments, a new KIN-E RNAi cell line was generated by targeting against the 3’UTR of KIN-E. A 618-bp fragment from the 3’UTR of KIN-E gene, which does not overlap with the downstream gene, was cloned into the pZJM-PAC vector. The resulting construct was linearized with NotI, and transfected into the 29–13 cell line. Transfectants were selected under 1 μg/ml puromycin in addition to 15 μg/ml G418 and 50 μg/ml hygromycin B, and further cloned by limiting dilution in a 96-well plate.
Full-length KIN-E gene was cloned into pLew100-3HA-BLE vector, which bears the actin 3’-UTR. KIN-E gene sequence with the deletion of the importin α-like domain (a.a. 413–684) and KIN-E gene sequence with the deletion of the m-calpain domain III-like domains (a.a. 713–997) were each cloned into the pLew100-3HA-BLE vector for expression of 3HA-tagged KIN-E-ΔIMPα and KIN-E-ΔmCL, respectively. These plasmids were each linearized with NotI and transfected into the cell line containing the pZJM-TbKIN-E-3’UTR-PAC construct. Transfectants were selected under 2.5 μg/ml phleomycin in addition to 1 μg/ml puromycin, 15 μg/ml G418, and 50 μg/ml hygromycin B, and cloned by limiting dilution in a 96-well plate. These plasmids were also transfected into the 29–13 cell line for determining the localization of ectopically overexpressed proteins.
Epitope tagging of KIN-E, FLAM3 [14], and FC1 [6] from their respective endogenous locus was carried out using the PCR-based method [36]. To examine whether epitope-tagging of KIN-E disrupts its function, the second allele of KIN-E was knocked out by replacing the coding sequence with puromycin-resistance gene. The resulting cell line grows normally like the wild-type 427 cell line. Epitope tagging of FLAM3 and FC1 does not disrupt their function in previous reports [6,14]. For KIN-E and FLAM3 co-localization, KIN-E was tagged with a C-terminal triple HA epitope (neomycin resistance) and FLAM3 was tagged with a C-terminal PTP epitope (puromycin resistance) in the 427 cell line. For KIN-E and FC1 co-localization, KIN-E was tagged with a triple HA epitope (neomycin resistance) and FC1 was tagged with a PTP epitope (puromycin resistance) in the 427 cell line. Transfectants were selected under 40 μg/ml G418 and 1 μg/ml puromycin, and cloned by limiting dilution in a 96-well plate.
KIN-E was also tagged with a C-terminal triple HA epitope (puromycin resistance) in the 29–13 cell line containing the pZJM-KIN-E RNAi construct or the pZJM-FLAM3 RNAi construct. FLAM3 was tagged with a C-terminal PTP epitope (puromycin resistance) in the 29–13 cell line containing the pZJM-KIN-E RNAi construct. Transfectants were selected under 1 μg/ml puromycin in addition to 15 μg/ml G418 and 50 μg/ml hygromycin B, and further cloned by limiting dilution in a 96-well plate.
FLAM3 was also tagged with an N-terminal PTP epitope (blasticidin resistance) in the 29–13 cell line containing the pZJM-KIN-E-3’UTR-PAC RNAi construct, pZJM-FLAM3 RNAi cell line, and the pLew100-3HA-BLE construct for ectopic expression of 3HA-tagged wild-type KIN-E or various KIN-E mutants. Transfectants were selected under 10 μg/ml blasticidin in addition to 15 μg/ml G418, 50 μg/ml hygromycin B, 2.5 μg/ml phleomycin, and 1 μg/ml puromycin, and cloned by limiting dilution in a 96-well plate.
Flow cytometry analysis of the expression of EP procyclin in control and KIN-E RNAi cells was carried out according to the procedure described previously [37]. EP procyclin was detected using the monoclonal antibodies TRBP1/247 (Cedarlane Labs, Canada) [38], which was used at a dilution of 1:500, and the FITC-conjugated anti-mouse IgG (Sigma-Aldrich), which was used at a dilution of 1:400. Cells were analyzed on a fluorescence-activated cell sorter (Becton Dickinson & Co., Sunnyvale, CA).
For quantitative western blotting, an equal number (5×105) of control and KIN-E RNAi cells were lysed, and cell lysate was fractionated on SDS-PAGE, transferred onto a PVDF membrane, and immunoblotted with anti-EP procyclin monoclonal antibody TRBP1/247 (1:1,000 dilution, Cedarlane Labs) and anti-TbPSA6 polyclonal antibody (1: 2,000 dilution), which detects the alpha-6 subunit of the 26S proteasome [39], for 1 hr at room temperature. After washing three times with TBST, the membrane was incubated with FITC-conjugated anti-mouse IgG (1: 400 dilution, Sigma-Aldrich) and IRDye 680LT anti-rabbit IgG (1:2,500 dilution, Li-Cor Cooperate), and western blot signals were captured using the Bio-Rad ChemiDoc MP imaging system, which allows multiplex fluorescent western blot imaging and quantitative analysis of protein bands.
Co-immunoprecipitation was carried out according to our previous procedures [40]. Briefly, cells (5x107) were lysed by incubating with 1 ml immunoprecipitation buffer (25 mM Tris-HCl, pH7.6, 500 mM NaCl, 1 mM DTT, 1% NP-40, and protease inhibitor cocktail) for 30 min on ice. Cleared lysate was incubated with 50 μl settled IgG beads for 1 h at 4°C, and immunoprecipitates were washed six times with the immunoprecipitation buffer. Proteins bound to the IgG beads were eluted with 10% SDS, separated by SDS-PAGE, transferred onto a PVDF membrane, and immunoblotted with anti-HA antibody to detect 3HA-tagged KIN-E and its various mutants and with anti-Protein A antibody to detect PTP-tagged FLAM3. Cells expressing 3HA-tagged KIN-E alone and PTP-tagged FLAM3 alone were included as negative controls.
The DNA sequences encoding the importin-α-like domain (a.a. 413–684) and the m-calpain domain III-like domain (a.a. 713–997) of KIN-E gene were each cloned into the pGEX-4T-3 vector (Clontech). The resulting plasmids were transformed into E. coli BL21 strain. Expression of GST-fusion proteins was induced with 0.1 mM IPTG for 16 h at room temperature, and purified through glutathione sepharose beads. Purified GST fusion proteins bound on the beads were incubated at 4°C for 1 h with T. brucei lysate prepared by lysing T. brucei cells expressing PTP-FLAM3 in lysis buffer (25 mM Tris-HCl, pH7.6, 500 mM NaCl, 1 mM DTT, 1% NP-40, and protease inhibitor cocktail). The beads were washed six times with the lysis buffer, and bound proteins were eluted by boiling the beads in 1× SDS sampling buffer. Eluted proteins were separated on SDS-PAGE, transferred onto a PVDF membrane, and immunoblotted with anti-Protein A antibody to detect PTP-FLAM3. GST alone was used as the negative control. GST and GST-fusion proteins were stained by Coomassie Brilliant Blue dye.
Cytoskeleton of T. brucei cells were prepared by incubating cells with PEME buffer (0.1 mM PEPES, pH6.9, 2 mM EGTA, 1 mM MgSO4, 0.1 mM EDTA) containing 1% Nonidet P-40 at room temperature for 5 min [41]. Cells were then spun down at 13,000 rpm in a microcentrifuge to separate the soluble cytosolic fraction and the insoluble cytoskeletal fraction. The cytoskeletal (pellet) fraction was re-suspended in PBS of the equal volume of the previously added PEME buffer. Both soluble and pellet fractions were boiled for 5 min after adding an equal volume of SDS-PAGE sampling buffer. Samples were separated by SDS-PAGE, transferred onto PVDF membrane, and immunoblotted with anti-Protein A (anti-ProtA) antibody to detect PTP-tagged FLAM3, anti-HA antibody to detect 3HA-tagged KIN-E and its mutants. The same blot was also probed with anti-α-tubulin mAb (Sigma-Aldrich) as cytoskeleton marker and with anti-TbPSA6 pAb as the cytosol marker.
Cells were washed once with PBS, adhered to the coverslips for 30 min at room temperature, fixed with cold methanol (-20°C) for 30 min, and then rehydrated with PBS for 10 min at room temperature. Cells adhered on the coverslips were blocked with 3% BSA in PBS for 1 h at room temperature, and incubated with the primary antibody for 1 h at room temperature. The following primary antibodies were used: FITC-conjugated anti-HA monoclonal antibody for 3HA-tagged proteins (1:400 dilution, Sigma-Aldrich), anti-Protein A polyclonal antibody for PTP-tagged proteins (1:400 dilution, Sigma-Aldrich), anti-CC2D polyclonal antibody (1:1,000 dilution) [42] for the FAZ filament [43], 20H5 monoclonal antibody (1:400 dilution) for the bilobe [44], anti-TbSAS-6 antibody (1:400 dilution) for basal body [45], YL 1/2 monoclonal antibody (1: 2,000 dilution) for mature basal body [46], L8C4 (anti-PFR2) monoclonal antibody (1:50 dilution) for the flagellum [43], L3B2 (anti-FAZ1) monoclonal antibody (1: 50 dilution) for the FAZ filament [43], and anti-CRAM polyclonal antibody (1:400 dilution) for the flagellar pocket [47]. Subsequently, cells were washed three times with PBS, and then incubated with FITC-conjugated anti-mouse IgG (1:400 dilution, Sigma-Aldrich) or Cy3-conjugated anti-rabbit IgG (1:400 dilution, Sigma-Aldrich) for 1 h at room temperature. Cells on the coverslips were washed three times with PBS, mounted with DAPI-containing VectaShield mounting medium (Vector Labs), and imaged under an inverted fluorescence microscope (Olympus IX71) equipped with a cooled CCD camera (model Orca-ER, Hamamatsu) and a PlanApo N 60x1.42-NA lens. Images were acquired using the Slidebook 5 software.
Scanning electron microscopy was performed essentially as described in our previous publications [40,48]. Cells were settled onto coverslips and fixed with 2.5% (v/v) glutaraldehyde in PBS for 2 hours at room temperature. Cells were washed three times with PBS, and then dehydrated in alcohol. After critical point drying, samples were coated with a 8-nm metal film (Pt:Pd 80:20, Ted Pella Inc.) using a sputter-coater (Cressington Sputter Coater 208 HR, Ted Pella Inc.), and then imaged using Nova NanoSEM 230 (FEI). The scanning work distance was at 5 mm, and the accelerating high voltage was at 8 kV.
Statistical analysis was performed using the t-test in the Microsoft Excel software. Detailed n values for each panel in the figures were stated in the corresponding legends. For immunofluorescence microscopy, images were randomly taken and all cells in each image were counted.
|
10.1371/journal.pcbi.1005073 | Qualitative Dynamical Modelling Can Formally Explain Mesoderm Specification and Predict Novel Developmental Phenotypes | Given the complexity of developmental networks, it is often difficult to predict the effect of genetic perturbations, even within coding genes. Regulatory factors generally have pleiotropic effects, exhibit partially redundant roles, and regulate highly interconnected pathways with ample cross-talk. Here, we delineate a logical model encompassing 48 components and 82 regulatory interactions involved in mesoderm specification during Drosophila development, thereby providing a formal integration of all available genetic information from the literature. The four main tissues derived from mesoderm correspond to alternative stable states. We demonstrate that the model can predict known mutant phenotypes and use it to systematically predict the effects of over 300 new, often non-intuitive, loss- and gain-of-function mutations, and combinations thereof. We further validated several novel predictions experimentally, thereby demonstrating the robustness of model. Logical modelling can thus contribute to formally explain and predict regulatory outcomes underlying cell fate decisions.
| We delineate a logical model encompassing 48 components and 82 regulatory interactions controlling mesoderm specification during Drosophila development, thereby integrating all major genetic processes underlying the formation of four mesodermal tissues. The model is based on in vivo genetic data, partly confirmed by functional genomic data.
Model simulations qualitatively recapitulate the expression of the main lineage markers of each mesodermal derivative, from developmental stage 8 to 10, for the wild type case, as well as for over twenty reported mutant genotypes.
We further use this model to systematically predict the effects of over 300 loss- and gain-of-function mutations, and combinations thereof. By generating specific mutant combinations, we validated several novel predictions experimentally demonstrating the robustness of model.
This modelling study is the first to tackle the regulatory network controlling the specification of mesoderm during Drosophila development, and more broadly deals with one of the most comprehensive developmental networks that have been modelled to date.
| Functional genomic approaches (based on microarrays and next-generation sequencing) provide a powerful means to decipher the molecular mechanisms underlying the control of development and cell differentiation, as well as deregulations thereof associated with diseases such as cancer. Together with low-throughput experimental data, these high-throughput methods enable the delineation of large and sophisticate regulatory networks. Understanding and predicting the behaviour of such complex networks require the use of proper mathematical modelling frameworks. Various dynamical models have been proposed for a handful of relatively well known developmental processes, many using differential equations and referring to Drosophila development (see e.g. [1–5] and references therein). However, these modelling studies consider relatively limited numbers of regulatory components (at most a dozen) and require the quantitative determination of poorly documented parameters. In this context, formal qualitative modelling approaches constitute an interesting alternative, at least as a first step towards more quantitative modelling. In particular, logical (Boolean or multilevel) modelling has been applied to various regulatory and signalling networks of increasing sizes over the past decade (see e.g. [6–19] and references therein). But only few attempts were made to predict novel phenotypes, and therefore the full predictive value of the network and its usefulness to test hypotheses regarding novel genetic perturbations remain unclear. Here, we set out to decipher the network controlling the specification of mesoderm, one of the three germ-layers, into its four main derivatives, namely visceral muscle, heart, somatic muscle, and fat body, the primordia of which are iterated in segmentally repeated units along the anterior-posterior axis of the Drosophila embryo (Fig 1).
Mesoderm specification is induced by ectodermal signals such as Decapentaplegic (Dpp), which controls dorsal-ventral differentiation [20–25], Wingless (Wg), which is essential for dorsally located heart cell precursors and to the majority of somatic muscles that develop from more ventrally located cells [26–28], and Hedgehog (Hh), which specifies the visceral mesoderm dorsally and the fat body ventrally, itself characterised by the expression of Serpent (Srp) [29–31]. During embryonic stages 8–10, the mesoderm is thereby progressively specified into four different tissue primordia, each of which is characterised by the expression of specific lineage transcription factors (Fig 2) [27, 32].
Collating all phenotypic data from the literature into a mathematical model allows to formally assess the coherence between the current view of the network with individual published results on single or multiple mutant phenotypes. More specifically, we aim to further characterise the crucial regulatory components and interactions driving mesoderm specification. As we mostly rely on published qualitative molecular and genetic data, we use a flexible logical modelling framework and the software GINsim (cf. Material and methods), which enables the use of multilevel variables whenever justified, along with fully asynchronous updating. Systematic simulations of the resulting logical model were then performed to (i) assess the coherence and comprehensiveness of our representation of the underlying network, (ii) identify gaps in the current understanding and characterisation of mesoderm specification, and (iii) ultimately predict phenotypic outcomes of novel genetic perturbations. We demonstrate that the resulting logical model can recapitulate all known mutant phenotypes, therefore indicating that this formal representation of the network is sufficient and coherent to explain mesoderm cell fate decisions. By running simulations on over 300 genetic mutation combinations (many of which are double mutants with non-intuitive outcomes), the model could predict the phenotypic outcome for each novel mutant background, at least in terms of gene expression patterns, thereby providing new testable hypothesis that we experimentally confirmed. This approach thus provides developmental biologists with a very useful tool kit to test novel hypotheses, which are often very difficult to carry out experimentally. Moreover, the model provides novel insights into the underlying regulatory network driving these cell fate decisions.
To initiate this study, we performed an extensive analysis of all reported genetic and molecular data in the literature to identify the main regulatory components involved in Drosophila mesoderm specification, along with the known interactions between them. Indeed, dozens of articles extensively cover the genetic bases of the sub-division of Drosophila mesoderm (this is evident by the bibliographical entries linked to key regulatory components in the model file and model documentation provided as S1 Text). Cis-regulatory information is sometimes available, enabling us to infer direct interactions and epistatic relations. In particular, we relied on recent ChIP data reporting the in vivo occupancy of six key mesoderm transcription factors (Bagpipe (Bap), Biniou (Bin), Dorsocross 1, 2 and 3 (Doc), Myocyte Enhancer Factor 2 (Mef2), Tinman (Tin) and Twist (Twi)) [33–35] to assess direct interactions inferred from genetic experiments.
Encoded using the software GINsim (Computational and experimental procedures), the resulting regulatory graph (Fig 3) is provided in a computer readable format, along with extensive annotations (text and links to relevant literature and database entries, see S1 File).
This regulatory graph encompasses 48 nodes (including 12 input components, representing mainly ectodermal signals) and 82 regulatory interactions. In many cases, the definition of the logical rule associated with each node is straightforward (e.g. when a node is the target of a unique regulator). However, for more complex regulatory relationships, i.e. when multiple interactions converge on the same component, we examine the following scenarios: (i) Is the presence of an inhibitor sufficient to completely or partially block gene expression? (ii) Which activator(s) is (are) sufficient to drive the expression of the target gene? (iii) Can the activators do so in the presence of repressor(s)? After several iterations, we obtained a set of logical rules consistent with all available knowledge on the regulation of each gene in the network, which further enabled the recapitulation of all published phenotypes (see below), demonstrating the robustness of the model.
Before attempting to simulate the specification of the mesoderm into its four main presumptive tissues (visceral muscle (VM), heart (H), somatic muscle (SM), and the fat body (FB), we needed to specify the patterns of gene expression expected as a result of wild type development. Based on published data (mainly in-situ hybridization or immunostaining assays), we have derived the qualitative levels of expression of the 48 network components in each of the four presumptive territories (VM, H, SM and FB) from the literature (S1 Fig). Only subsets of these components are crucial in the specification of each of the four tissue subtypes. These tissue markers can be readily identified based on the phenotypes reported in loss-of-function mutant embryos, leading to severe defects in tissue formation, or following ectopic expression, often leading to specific tissue expansion. Embryos lacking Tin, Bap or Bin, for example, do not develop VM. Moreover, tin mutant embryos fail to develop H cells, and have severe defects in all tissues derived from the dorsal mesoderm [32, 36, 37]. Overall, ten network components play such dramatic roles in specific tissues (emphasised by bold contours for the corresponding coloured cells in the S1 Fig). Note that the ventral mesoderm territory that gives rise to both SM and FB is subdivided into regions that have low (yet significant, hence the use of the value 1) and high Twi expression. Indeed, the inhibition of Notch (N) combined with the presence of Wg and Daughterless (DA) activates Twi at a higher level (maximal level, i.e. value 2), thereby delimiting a region of high Twi expression [38, 39]. We systematically searched for relevant information and refined the logical rules until model behaviour was found fully consistent with all published data.
To ease simulations, our regulatory graph was reduced by hiding intermediate signalling components (components in grey in Fig 3, see also Material and methods). Provided that we do not delete any regulatory circuit, the resulting reduced model preserves the stable states of the system, which represent the different specification states (i.e. mesoderm derivatives for wild-type or mutant situations). To perform simulations, the initial values for each component must be specified, in particular for the signalling input components coming from the ectoderm. For each of the four presumptive tissue territories, we thus have a specific input combination (S1 Fig, left).
For the sake of simplicity, we set all internal nodes to zero for each wild-type initial state, with the notable exception of Twi, which was set to the value 1. For each territory, the target values at stage 10 were evaluated based on published data (S1 Fig, right). For example, in the region that will form VM, the initial state (stage 8) is characterised by the presence of Twi, which activates Tin [40] and Mef2 [41] expression. Bap is then activated by Tin at stage 9 [37], which is followed by the activation of Bin by Bap [42, 43] in late stage 9 embryos. Finally (stage 10), Bap is activated at its maximum level (value 3) by Cubitus Interruptus (Ci) and Engrailed (En) [27].
To recapitulate the formation of each mesodermal tissue derivative in the wild-type situation, we thus ran four different simulations using an asynchronous updating policy (Material and methods).
A detailed comparison of simulation results with experimental data led us to refine the logical rules, and sometimes even to consider additional regulators, until we converged on the regulatory graph shown in Fig 3, along with the rules listed in S1 Table (see also the S2 Text for more information about the delineation of the logical rules associated with Tin and Bap). Our final model qualitatively recapitulates all aspects of the major events in the specification of the four main domains of the mesoderm, from stage 8 to stage 10 (S2 Fig). In parallel, we also simulated the effects of genetic perturbations reported in the literature, the results of which led to some model adjustments. Iterating this procedure for all known mesodermal mutants led to a model that is robust and consistent with all relevant published data. The simulated phenotypes resulting from seven selected genetic perturbations are illustrated in Fig 4.
For example, the simulation of a wg loss-of- function (lof) gives rise to a loss of cardiac tissue, as observed experimentally [27, 44], while dpp lof gives rise to an extension of FB at the expenses of VM, mirroring previously reported experimental data [30, 32]. We can also simulate more complex genetic backgrounds. For example, a double gof of dpp and hh combined with a lof of wg leads to an expansion of VM in the entire mesoderm [27]. To date, our simulations recapitulate all mutants reported in the literature. Although expected, since this literature information was used as input to generate the model, these results demonstrate the coherence of the model, which was based on disjoint information generated from published studies from different labs, mostly based on single mutant phenotypes, with only a small number of documented multiple genetic perturbations. A study focused on heart or VM development, for example, often will not have examined markers for FB, yet the model can simulate the phenotypes in all four mesodermal domains.
Given the accuracy of our model to recapitulate all known published phenotypes, we reasoned that the model provides a very useful tool to perform systematic novel in silico perturbations at large scale. In fact, some of the results obtained with the simulations described above already correspond to new predictions, as biologists typically check only subsets of markers for each mutant studied (see in particular the tissue domains shown in yellow in Fig 4, which correspond to situations that have not been fully analysed experimentally, and for which we obtain combinations of markers associated with different tissues). Nevertheless, our aim here is to go beyond this and perform a more systematic assessment of the effects of combinations of two perturbations affecting different pathways and/or tissue markers. Single and multiple mutants can be readily defined using GINsim (cf. Material and methods), while they can often be very difficult, and sometimes impossible, to generate experimentally. To this end, we simulated the effects of single or double perturbations in each of the four tissue domains. The interpretation of the results generated is not trivial. To assess these results more efficiently, we used the expression of the key lineage transcription factors for each tissue as defining signature: for VM, expression of Tin (level 1), Bap (level 2), and Bin; for H, expression of Tin (level 2), Doc, and Pannier (Pnr); for FB, expression of Srp; and for SM, expression of Twi (level 2), Pox meso (Poxm), DSix4 and Zfh1. These tissue signatures were then matched against the stable states reached during model simulations, thereby automating the interpretation of the resulting phenotypes. Practical considerations (mutant strain availability) led us to consider fourteen components (Twi, Tin, Bap, Bin, Ci, Doc, Pan, Pnr, Slp, Srp, Mef2, Mad, Med, Nicd) for systematic single and pairwise combinations of loss- and gain-of function in silico perturbations.
The results of the 338 mutant simulations performed are displayed in a matrix form (Fig 5) and can be browsed in a convenient searchable web archive (S2 File). This format enables an easy comparison of the effects of different perturbations, which facilitates the detection of dominant or synergic effects of different perturbations.
For example, slp lof generally shows a loss of H tissue, a result similar to that obtained for wg lof [26, 27]. Although some of these mutants have been partly documented experimentally, most of the double perturbations listed in Fig 5 have not been fully experimentally assessed in all four-tissue domains.
To demonstrate the usefulness and accuracy of these predictions, we experimentally tested six genetic perturbations (two double mutants and the associated four single mutants), examining the effects within all four tissue domains (Fig 6 and S3 Fig). Model predictions for each of these mutants are highlighted in Fig 5.
To examine the phenotype of each tissue, Tin, Srp, Glycogen phosphorylase (GlyP) and Bin were used as markers for the development of H, FB, SM and VM, respectively. We first assessed our predictions for lof mutants of Medea (Med) and Sloppy-paired (Slp), and the double mutant. Medea is directly required for the induction of tinman (Tin) by Dpp via the tin-D dorsal mesoderm enhancer [23]. Once expressed, Tin and Med have a direct protein-protein interaction that is required for dorsal mesoderm specification [28]. As the heart specification requires both activation by Med-Tin and repression of the VM within the heart domain by Slp, we were interested to examine if loss of Med and Slp would completely abolished cardiac mesoderm specification and be sufficient to extend the VM territory. For Med lof, our model simulations predict a loss of H, which is indeed what we observed experimentally (Fig 6A). Although the expression status of some genes within the VM region is changed in the mutant, the VM develops largely unperturbed, as predicted. The simulation of slp lof also results in a loss of H, and two stable states within the SM, one leading to normal SM development, while the other state lacks some marker expression, and therefore should perturb SM development. When we examine slp lof mutant experimentally, we observe the predicted loss of H, while SM appears largely normal, indicating that the corresponding stable state is the correct outcome. Simulations of the double lof mutant give the combined phenotype of the two single mutants, which again qualitatively fits with experimental data, with the H even more severely affected in the double mutant (Fig 6A, in situ for tin expression). In contrast, the VM develops largely unperturbed, indicating that loss of heart, even the severe disruption seen in the double mutant, is not sufficient to lead to expansion of VM, in this genetic background.
We next tested a combination of two gof conditions, where it is not a priori obvious what the phenotypic consequence would be within the FB or SM domains. Slp is normally expressed in the H region, where it inhibits VM development through the direct inhibition of Bap expression [32, 45]. Doc is expressed within the Heart domain (segmentally repeated patches of cells within the dorsal mesoderm) at stage 10, where it is essential for heart development [46, 47]. For a gof of Doc, our simulations predicts normal H development, with minor perturbations of VM, FB and SM. Our experimental results largely confirm these predictions, with very minor perturbations on the development of each tissue (based on the expression of the corresponding tissue markers), despite the ubiquitous expression of Doc (Fig 6B). The simulation of a gof of Slp predicts a severe perturbation of VM (yellow cell), characterised by the lack of expression of the key cell markers Bap (level 1 instead of 3) and Bin, as expected [32, 45], along with a potential perturbation of FB (obtention of two stable states, both with Srp expression). When the two gof genotypes are combined, our model predicts normal H and SM specification, but a loss of VM and FB, which is exactly what we observe experimentally, as seen in the in situ shown in Fig 6B. These results therefore demonstrate that our qualitative model can correctly predict the interaction between two gof causing a severe loss of FB. The expansion of heart cells can be further explained by the ectopic expression of heart markers in our simulations.
The logical model presented here integrates all major genetic processes underlying the formation of four tissues during Drosophila mesoderm specification. The model is based on the integration of extensive analysis of in vivo experimental data, especially genetic data (patterns of gene expression and mutant phenotypes), partly confirmed by functional genomic data (ChIP data for transcription factor occupancy). These data were translated mathematically in terms of a regulatory graph and logical rules. The simulation of our model qualitatively recapitulates the expression of the main lineage markers of each region from developmental stage 8 to 10, for the wild type case, as well as for over twenty reported mutant genotypes.
This study is the first attempt to model the regulatory network controlling the specification of mesoderm during Drosophila development, and more broadly represents one of the most comprehensive developmental networks that have been modelled to date. Mesoderm specification has been extensively studied in many species, including the sea urchin [48]. Recently, the Davidson group developed a Boolean model that recapitulates the specification of the sea urchin endo-mesoderm in the wild-type case, as well as experimental data for three genetic perturbations [49]. The approach of Davidson's group converge with ours in the delineation of a reference network with reliable annotations, which then serve as a scaffold to define logical rules and perform simulations. Both approaches implement the crucial components and interactions, along with the dynamical unfolding of the corresponding developmental network in an intuitive manner. Importantly, we demonstrate that we can not only recapitulate the known mutant phenotypes, but also predict various novel phenotypes.
In the case of our study, several regulatory mechanisms were simplified, in particular regarding the signalling pathways involved. We have developed more complete models of most Drosophila signalling pathways [50], but we retained simpler implementations of these pathways to keep our mesoderm specification model computationally tractable.
A limitation of this study resides in the poor documentation of specific markers associated with each type of embryonic domain. In particular, our marker set is limited to Srp in the case of FB. Presumably, others regulatory factors must be implicated in the specification of this tissue, which remain to be discovered. This lack of information complicates the interpretation of mutant phenotypes. For example, it is known that Bap lof leads to the loss of VM, but we miss information about effects on other tissues. Although Bap is crucial for VM development, it is also expressed at later stages in H. At this point, we assume that H, SM and FB develop normally in Bap lof mutant, as no other experimental defect has been reported.
Finally, Boolean models of embryonic processes generally rely on qualitative expression data from in-situ hybridisation. Our discrete model (as the sea urchin model [49]) is therefore limited to qualitative results, such as the presence or absence of a given tissue in a given presumptive territory. Although we cannot reproduce quantitative data, such as an increase or a decrease of specific cell numbers, we can still recapitulate the presence of different cell types. Our logical model could further serve as a scaffold to build more quantitative models when more quantitative and systematic experimental datasets will become available. For now, the advantage of logical modelling is that models can be easily abstracted at a level subsuming missing data, which is less straightforward for more quantitative modelling frameworks, such as differential or stochastic equations. Given the complexity of embryonic development, the shear number of parameters involved and the high inter-connected nature of regulatory networks, logical modelling offers an accurate solution that can be applied to many systems with the amount of data that is available today.
We use the multilevel logical formalism, originally proposed by René Thomas [51], which has already been used to model various networks involved in the control of cell differentiation or proliferation (see e.g. [6, 8, 9, 11, 12, 19]. In short, both the structure of a logical model and its dynamics are represented in terms of graphs (in the sense of the graph theory), called regulatory graphs and state transition graphs, which are briefly described hereafter.
In a regulatory graph, the vertices (or nodes) represent regulatory genes or products (transcription factors, kinases, etc.). In many cases, these regulatory components can be satisfactorily represented by Boolean variables, which can take only two values, 0 or 1, corresponding to the absence or presence of the component, respectively. However, in some situations (e.g. the consideration of a morphogen), more qualitatively different levels may be required. The arcs (or arrows) connecting pairs of vertices represent regulatory interactions between components (e.g. transcriptional activations or inhibitions, phosphorylation, etc.). These arcs are usually associated with a plus (+) or minus (-) sign, denoting an activation or inhibition effect of the source node onto the target node, respectively. When the source of an arc is associated with a multilevel variable, a threshold (i.e. minimal level) must be specified. To complete this model description, logical rules (or logical parameters) are further defined to indicate how each component reacts to different combinations of regulatory interactions (S2 Text and S1 Table).
The simulation of a logical model can be represented by a state transition graph (STG), whose vertices represent logical states (i.e. a vector encompassing values for all components), whereas arcs represent transitions between states enabled by the corresponding regulatory graph and logical rules. In this work, we use an asynchronous updating mode, meaning that we consider all possible unitary transitions (affecting only one variable at a time) whenever there is a call to change some component value(s) at a given state. One recurrent problem with logical simulations (in particular when using asynchronous updating) is the potential combinatory explosion of the STG when dealing with large regulatory graphs. Consequently, it is often difficult to generate and analyse the STG for complex networks encompassing several dozens of components. However, using proper algorithms and software tools (see below), it is possible to characterise the asymptotical behaviour of the systems, which is of special interest for us here. Indeed, attractors, especially stable states (states with no successor), are usually associated with specific differentiated states. Logical models provide a realistic description of cellular events, as they are capable of reproducing time dependent processes in a qualitative manner (i.e. focusing on the sequential order of transitions).
The software GINsim (for Gene Interaction Network simulation) implements the logical formalism [52]. It allows the edition, analysis and simulation of regulatory graphs. Freely available (http://ginsim.org), GINsim supports the annotation of components and interactions with free text and URLs. Once a model is defined, the user can select a simulation mode and define a set of initial states. GINsim can then be used to compute state transition graphs and report the stable states. GINsim also enables the definition and the simulation of different types of mutants (loss-of-function, ectopic gene expression, and combinations thereof) by blocking the levels of expression of the corresponding variables in defined intervals. To further ease the analysis of multiple perturbations, we have written a set of scripts in python, which iteratively compute the behaviour of our mesoderm specification model for each region and mutant considered, process the results and generate a synthetic web page (cf. Results and S2 File).
To enable the dynamical analysis of comprehensive regulatory graphs, we take advantage of a novel reduction method implemented in GINsim. This functionality allows the user to select components of a regulatory graph to be made implicit. The software verifies that the proposed reduction does not fundamentally change the network topology (elimination of regulatory circuits) and update the logical rules for the components targeted by reduced nodes. The original and reduced networks have the same stable states (in terms of levels of common variables), while differences may appear as to their reachability [53].
The following Drosophila lines were used: UAS-Slp and UAS-Doc lines were kindly provided by M. Frasch (Doc line C2 [46]). We crossed both stocks with a marked double balancer to generate the homozygous stock x/y;UAS-Doc;UAS-Slp. Males from the UAS-Doc, UAS-Slp and UAS-Doc;UAS-Slp lines were crossed with females carrying a homozygous twist-GAL4 driver, kindly provided by Maria Leptin. Slp1 and Med1e loss-of-function mutations were obtained from the Bloomington stock centre (stock numbers 5349 and 9033), and crossed together to make the double loss-of-function stock, which were placed over lacZ-marked balancers.
Embryos were collected using standard procedures. Fluorescent in situ hybridisation was performed as described previously [54]. The following ESTs were used to generate anti-sense probes: RE01329 (tin), SD07261 (srp), and LD24485 (Glyp), while a full length cDNA was used for bin (gift from M. Frasch) and lacZ. The probes were detected with peroxidase-conjugated antibodies (Roche) and developed using the TSA system (Perkin Elmer). slp and Med mutant embryos were unambiguously identified based on the absence of lacZ expression from the balancer chromosome.
|
10.1371/journal.ppat.1000533 | Mobile Genetic Element-Encoded Cytolysin Connects Virulence to Methicillin Resistance in MRSA | Bacterial virulence and antibiotic resistance have a significant influence on disease severity and treatment options during bacterial infections. Frequently, the underlying genetic determinants are encoded on mobile genetic elements (MGEs). In the leading human pathogen Staphylococcus aureus, MGEs that contain antibiotic resistance genes commonly do not contain genes for virulence determinants. The phenol-soluble modulins (PSMs) are staphylococcal cytolytic toxins with a crucial role in immune evasion. While all known PSMs are core genome-encoded, we here describe a previously unidentified psm gene, psm-mec, within the staphylococcal methicillin resistance-encoding MGE SCCmec. PSM-mec was strongly expressed in many strains and showed the physico-chemical, pro-inflammatory, and cytolytic characteristics typical of PSMs. Notably, in an S. aureus strain with low production of core genome-encoded PSMs, expression of PSM-mec had a significant impact on immune evasion and disease. In addition to providing high-level resistance to methicillin, acquisition of SCCmec elements encoding PSM-mec by horizontal gene transfer may therefore contribute to staphylococcal virulence by substituting for the lack of expression of core genome-encoded PSMs. Thus, our study reveals a previously unknown role of methicillin resistance clusters in staphylococcal pathogenesis and shows that important virulence and antibiotic resistance determinants may be combined in staphylococcal MGEs.
| The extreme danger associated with Staphylococcus aureus infections is due to the combination of frequent antibiotic resistance, which prevents efficient treatment, with extraordinary virulence, which determines the severity of disease. S. aureus is known to exchange antibiotic resistance and virulence determinants between different strains, thereby spreading the capacity to cause serious infections in the S. aureus population. The genetic information for these determinants is usually found on so-called mobile genetic elements. It has been noted that such exchangeable elements carry genes for either virulence or antibiotic resistance, but not both. Here, we identified and characterized a potent toxin, whose gene is located within an element that encodes resistance to the important antibiotic methicillin. The toxin had strong capacity to kill human white and red blood cells and significantly affected the capacity of MRSA to cause disease. Our study shows that acquisition of methicillin resistance may be combined with gaining possession of potent toxins by a single event of genetic exchange, which likely represents an important feature accelerating the evolution of MRSA virulence.
| Staphylococci are ubiquitous colonizers of human epithelia and frequent opportunistic pathogens involved in nosocomial infections [1]. In addition, the most virulent species, Staphylococcus aureus, can cause severe disease such as septicemia, toxic shock syndrome, and endocarditis, in both hospital and community settings [2].
The severity of a S. aureus infection is to a large extent determined by the toxin repertoire of the infecting strain. For example, S. aureus may produce toxic shock syndrome toxin-1 and other superantigens (enterotoxins), leukocidins, α-toxin, and phenol-soluble modulins (PSMs) [3],[4]. Many of these molecules destroy immune cells, thereby contributing considerably to the immune evasion capacity of S. aureus. Some toxins, such as α-toxin and the PSMs, are encoded on the bacterial core genome. Strain-to-strain differences in the secretion of these toxins are mainly due to differential gene expression [5]–[7]. In contrast, many other toxin genes are located on mobile genetic elements (MGEs). While many S. aureus strains produce MGE-encoded toxins, production of a given MGE-encoded toxin is usually limited to a small number of strains and strain-specific [8].
Staphylococcal infections are further complicated by frequent and sometimes multiple antibiotic resistance [9]. After the wide distribution of penicillinase-resistant strains in the middle of the last century [10], methicillin became the antibiotic of first choice for S. aureus infections. However, methicillin-resistant S. aureus (MRSA) are now prevalent in hospitals [11]. In addition, community-associated MRSA strains have emerged more recently and are spreading globally [12]. Furthermore, methicillin resistance is also common in coagulase-negative staphylococci such as S. epidermidis (methicillin-resistant S. epidermidis, MRSE) [13].
Antibiotic resistance genes are often located on MGEs such as transposons, plasmids, or genomic islands [14]. Specifically, the staphylococcal cassette chromosome mec (SCCmec) carries the mecA gene responsible for resistance to methicillin. There are at least 4 main and several sub-types of SCCmec elements, ranging from 21 to 67 kb in size, which are characterized by the two essential mec and ccr gene complexes and accessory gene loci such as transposons [14]. Importantly, while toxins and other virulence determinants are often encoded on MGEs, they have not been found within SCCmec elements or on widespread staphylococcal plasmids. Thus, acquisition of antibiotic resistance determinants by horizontal gene transfer in staphylococci is usually not linked to that of virulence factors [14].
PSMs are small, amphipathic and α-helical peptide toxins that attract and activate neutrophils [7]. In addition, PSMs of the α-type have pronounced capacity to lyse neutrophils and other cell types. Highly virulent CA-MRSA produce large amounts of the strongly cytolytic PSMα peptides, which are encoded in the core-genome located psmα operon and represent the main toxins contributing to neutrophil lysis in these strains. Deletion mutants in the psmα operon have dramatically reduced capacity to cause skin infections and bacteremia, indicating a crucial role of these toxins in S. aureus pathogenesis [7].
Here we identified and characterized an α-type PSM peptide that has pro-inflammatory and cytolytic activity and an important role in S. aureus infection. In contrast to all PSMs found so far, the newly identified psm-mec gene is encoded within an SCCmec MGE rather than on the core genome, providing a molecular connection between virulence and antibiotic resistance in staphylococci.
PSM peptides share physico-chemical properties rather than amino acid sequence similarity [7],[15],[16]. Additionally, psm genes are shorter than most cutoff thresholds for gene annotation. Therefore, to identify and classify a peptide as a member of the PSM family, initial detection and characterization of the peptide by means such as reversed-phase HPLC/mass spectrometry (RP-HPLC/MS), subsequent identification of the encoding gene, and detection of PSM-typical features such as amphipathy and α-helicity are required. Commonly, a given species produces a characteristic pattern of PSM peptides due to the fact that psm genes are encoded on the core genome rather than on MGEs [7],[15],[16]. However, while screening collections of S. aureus and S. epidermidis strains [17]–[21], we found that some strains showed an additional peptide peak in the RP-HPLC profile within the elution range characteristic for PSMs (shown for one S. epidermidis and one S. aureus strain in Figure 1A). The molecular weight of the peptide, 2414.6 Da (Figure 1B), as calculated from the electrospray ionization (ESI) mass spectrum obtained by RP-HPLC/ESI-MS, was the same in all these strains.
In detail, we analyzed a collection representing a wide variety of S. aureus strains [22], which contained 34 strains, 11 of which were MRSA, 79 MRSA strains of pulsed-field types USA100, USA200, USA300, USA500, USA1000, and USA1100 from infection and carriage isolates from San Francisco [17],[21], 54 infectious S. epidermidis strains from Paris, 56% of which were MRSE [23], and 180 S. epidermidis strains from Norway, 29% of which were MRSE [20]. Furthermore, we analyzed an S. epidermidis strain collection from Shanghai [19] that included 51 colonizing strains (no MRSE) and 41 isolates from infection (29% MRSE). 10% of all analyzed MRSA and 68% of all analyzed MRSE strains produced the peptide, while it was never detected in methicillin-sensitive S. aureus (MSSA) or S. epidermidis (MSSE). Accordingly, in the Shanghai collection, all producing S. epidermidis strains were isolated from human infections, whereas the MSSE skin isolates (colonizers) never produced the peptide (Figure 2). In the San Francisco strain collection, the peptide was found in 5 of 14 infectious USA100 and USA200 isolates, but never in other pulsed-field types. These results indicated that peptide production is linked to specific SCCmec elements.
Next, we purified the peptide and determined the N-terminal sequence, which allowed identification of the peptide-encoding gene (Figure 3). Analysis of published staphylococcal genome sequences revealed presence of the gene in the type II SCCmec clusters of S. epidermidis strain RP62A [24], and S. aureus strains Mu50, N315 and Sanger 252 [25],[26] (Figure 3). In addition, a tblastn search (www.ncbi.nlm.nih.gov/blast.cgi) showed that the gene is present within SCCmec clusters of types II or III in a series of staphylococcal strains including strains of S. aureus, S. epidermidis, S. saprophyticus, S. pseudintermedius, and S. sciuri. Furthermore, we typed the analyzed MRSE producers from the Paris and Shanghai collections as predominantly of SCCmec type III (21/27) and the four MRSA producers from the Fitzgerald et al. collection [22] as SCCmec type II. Finally, we also detected the gene in MRSA strains from Canada and New York City (strains C10682 and BK20781, GenBank FJ390057 and FJ670542.1) that contain the novel SCCmec type VIII, which appears to have arisen from recombination between different SCCmec elements [27]. These results indicated that the gene is typically encoded in SCCmec elements, specifically in the J1 region that is common to SCCmec types II and III (Figure 3) [14]. Thus, we termed the novel PSM peptide PSM-mec owing to the fact that it is encoded within SCCmec clusters. Furthermore, presence of the gene in SCCmec types II and III is in accordance with the data obtained with different MRSA pulsed field types, particularly the absence from community-associated MRSA of pulsed-field type USA300, which contain SCCmec type IV [28].
Like all psm genes [7],[15],[16], the psm-mec gene contained only the DNA sequence encoding the final peptide product and no signal peptide. In addition, comparison of the theoretical mass of the translation product (2386.8 Da) with the detected mass of the secreted peptide indicated formylation of the N-terminal methionine (mass difference of 28 Da), which is common in bacterial proteins and found in all PSMs [7],[15],[16]. Analysis of secondary structure by circular dichroism (CD) (Figure 4A) and arrangement of the peptide sequence in an α-helical wheel (Figure 4B) revealed strong α-helicity and amphipathy, confirming that PSM-mec has characteristics typical of PSM peptides.
All known PSM peptides are under control of the agr quorum-sensing system [7],[29]. Growth phase-dependent production of PSM-mec followed the same pattern as observed for other PSMs (Figure 5A), suggesting quorum-sensing control. Furthermore, the agr-dysfunctional MRSA strains N315 and Mu50 have all psm genes including psm-mec [26], but do not produce the corresponding gene products (data not shown). Moreover, we never detected PSM-mec in strains without δ-toxin production, which is indicative of a defective agr system. These observations suggested that PSM-mec production is dependent on agr. To further evaluate this hypothesis, we applied cross-inhibiting S. epidermidis autoinducing peptide, an efficient and specific inhibitor of S. aureus agr [30], to cultures of PSM-mec producing S. aureus. This led to complete absence of all PSMs, including PSM-mec (Figure 5B). Thus, PSM-mec production is under control by the quorum-sensing system agr like other PSMs.
The psm genes are regulated by direct binding of the AgrA response regulator to the psm promoter regions, distinguishing psm from many other agr-regulated genes that are under control of the regulatory RNA, RNAIII [31]. Analysis of the psm-mec promoter regions in the published genomes did not reveal consensus binding sites for AgrA (data not shown). However, as AgrA binding sites are not always completely conserved [31], the exact mechanism of psm-mec regulation by agr remains to be evaluated.
Notably, in many strains PSM-mec was produced at high levels, approximately achieving or in one strain exceeding production levels of the otherwise most abundant PSM, δ-toxin (PSMγ) (Figure 1, 2, 5C). Furthermore, while PSM-mec production was usually correlated with that of other PSMs, some strains showed a different production pattern. Strain MSA890 for example had high relative production of PSM-mec compared to other PSMs (Figure 5C). Thus, the fact that PSM-mec production is not always entirely correlated with that of δ-toxin indicates regulatory influences in addition to agr, as previously shown for other PSMs [7],[31].
PSM peptides, particularly those of the α-type, cause chemotaxis, specific release of cytokines such as IL-8, and lysis of neutrophils and erythrocytes [7]. To analyze whether PSM-mec, which by its size and physico-chemical characteristics forms part of the PSM α-type family, has similar pro-inflammatory and lytic capacities, we first determined chemotaxis and calcium flux in human neutrophils. PSM-mec had lower chemotactic activities (Figure 6A) and elicited lower calcium flux (Figure 6B) than the most potent PSMα3, but in a range similar to that detected for other α-type PSMs and in general higher than that of β-type PSMs. Then, we determined the capacity of PSM-mec to activate human neutrophils by measuring surface exposure of gp91phox and CD11b (Figure 6C, 6D). Capacity of PSM-mec to activate human neutrophils was lower than that of the most potent PSMα3, but in the range of the other α-type PSMs and δ-toxin, and higher than that of β-type PSMs. The capacity of PSM-mec to elicit production of the cytokine IL-8 was somewhat higher than that detected for other α-type PSMs, but about in the same range (Figure 6E). Neutrophil lysis as likely the most crucial immune evasion property of PSMs was lower in PSM-mec than in other α-type PSMs. However, at 50 µg/ml, neutrophil lysis by PSM-mec obtained approximately the same level (Figure 6F) as observed previously for other α-type PSMs at 10 µg/ml [7]. Of note, these concentrations are typically achieved by many strains in vitro [7] (Figures 2, 5), indicating that the contribution of PSM-mec to overall cytolytic capacity of PSM-mec producing strains achieves that of PSMα peptides. Finally, lysis of sheep erythrocytes by PSM-mec was in an intermediate range compared to other PSMs (Figure 6G). Overall, these results demonstrate that PSM-mec has pro-inflammatory capacities similar to other α-type PSMs, although not as pronounced as for the most potent PSMα3.
In contrast to all other PSM peptides identified until now, PSM-mec has one cysteine residue, which in secreted peptides usually indicates dimerization. To evaluate whether PSM-mec is present as a dimer and needs dimerization for its biological function, we synthesized mutant peptides with alanine and serine substitutions for the cysteine residue at position 17 of PSM-mec (PSM-mec C17A, PSM-mec C17S) (Figure 3). We first used size exclusion chromatography (SEC)/ESI-MS of PSM-mec C17A, PSM-mec C17S, and unreduced and reduced versions of PSM-mec to investigate whether PSM-mec is present as a dimer in its natural form. All peptides eluted at the same retention time (Figure 7). Furthermore, masses of the isolated and reduced versions of PSM-mec were the same and no peaks were obtained when calculating extracted ion chromatograms (EICs) with the mass of the theoretical oxidized, dimeric PSM-mec (Figure 7). Finally, EICs of S. aureus and S. epidermidis supernatants only showed the monomeric PSM-mec. These results indicate that PSM-mec does not dimerize despite the single cysteine in its amino acid sequence.
We then analyzed whether the cysteine residue is important for the biological function of PSM-mec, focusing on IL-8 secretion and neutrophil lysis. Substitution with alanine or serine only led to slightly reduced capacity to elicit production of IL-8 and lyse human neutrophils. Neutrophil lysis and IL-8 secretion were impaired to a more pronounced extent when the cysteine residue was replaced by serine than alanine, which is likely due to the fact that the cysteine residue is placed within the hydrophobic side of the amphipathic α-helix (Figure 4) and the hydroxyl side chain of serine may interfere with the amphipathic arrangement of the PSM-mec α-helix. These results indicated that the role of the cysteine residue in PSM-mec is likely limited to contributing to the PSM-mec α-helical structure and the cysteine sulfhydryl group does not have an additional specific function such as in peptide dimerization.
PSM peptides have been suggested to impact biofilm formation based on their detergent-like structure that indicates surfactant capacities [32],[33]. To analyze whether PSM-mec influences biofilm development in S. aureus, we determined in vitro biofilm formation on microtiter plates. First, we added synthetic PSM-mec to the biofilm-positive, agr-negative strain SA113, which lacks PSM production (data not shown). We measured the impact of PSM-mec on biofilm formation directly on plastic and on fibrinogen-precoated plates, to mimic both possible mechanisms of attachment to indwelling medical devices [34] (Figure 8A). In both cases, there was reduced biofilm formation at intermediate PSM-mec concentrations (50 µg/ml), which corresponds to the range of PSM-mec production in bacterial culture filtrates. It is possible that the lack of biofilm-inhibiting activity at higher concentrations is due to peptide aggregation. Aggregation into micelle-like multi-molecular clusters in a concentration-dependent fashion has been described for the PSM δ-toxin [35].
To analyze the biological role of PSM-mec in vivo, we focused on S. aureus owing to its greater importance as a pathogen. We produced isogenic mutants by allelic replacement of the psm-mec gene in strains S. aureus Sanger 252 and the four MRSA strains from the analyzed S. aureus strain collection that showed PSM-mec production (MSA820, MSA890, MSA1601, MSA3407). Then, we compared the isogenic psm-mec deletion mutants with the corresponding wild-type strains. There were slight, yet significant influences on biofilm formation and intercellular aggregation in some strains (Figure 8B, 8C). Together, these results indicate that PSM-mec has a small concentration-dependent capacity to impact adhesion to surfaces, biofilm formation, and intercellular aggregation.
To investigate whether PSM-mec has a role in pathogenesis, we first analyzed neutrophil lysis caused by culture filtrates of the isogenic psm-mec deletion mutant strains compared to those of the corresponding wild-type strains. We detected significantly decreased capacity to lyse human neutrophils in the psm-mec deletion mutant of strain MSA890, but not in the other deletion strains (Figure 9A). Most likely, this is due to the fact that strain MSA890 produces considerably more relative amounts of PSM-mec, compared to core-genome encoded PSMs, than the other strains (Figure 5C). Addition of increasing concentrations of PSM-mec to culture filtrates of the MSA890 psm-mec deletion strain, up to 100% of that detected in the wild-type strain under corresponding growth conditions, completely restored the neutrophil-lytic capacity of the MSA890 wild-type strain (Figure 9B), ruling out the possibility that the observed phenotype was due to unintended second site mutations. Furthermore, pronounced synergistic hemolysis of strain MSA890, a phenotype caused by concerted activity of δ-toxin, other PSM or PSM-like peptides and α-toxin or β-toxin [36],[37], was considerably reduced by deleting the psm-mec gene in MSA890, whereas no marked reduction was detected in the other isogenic strain pairs (Figure 9C). These results indicate that PSM-mec production can substitute for the lack of cytolytic capacity in strains such as MSA890, in which expression of genome-encoded cytolytic PSMs is low. Notably, this includes lysis of human neutrophils as likely the most crucial function of PSMs in pathogenesis.
To analyze whether psm-mec impacts pathogenesis in important manifestations of S. aureus disease, we performed murine bacteremia and skin infection models. We have previously shown in these models that deletion of the strongly cytolytic α-type PSMs encoded on the psmα operon leads to greatly decreased potential of S. aureus to cause disease [7]. We selected the wild-type and psm-mec deletion mutant pairs of strains S. aureus MSA890 and Sanger 252, the latter as an example of the strains in which there was no change in cytolytic activity between psm-mec deletion and wild-type strains. With MSA890 and MSA890Δpsm-mec, we detected very significant differences in lesion size and weight loss in the skin infection model (Figure 10A–C) and in animal survival rates in the bacteremia model (Figure 10D). In contrast, there were no significant differences between strains S. aureus Sanger 252 and Sanger 252Δpsm-mec in the same models (data not shown). These results are in accordance with those achieved in the neutrophil lysis and hemolysis assays, indicating that the presence of PSM-mec may significantly impact S. aureus pathogenesis when PSM-mec levels exceed those of other cytolytic PSMs.
In this study, we identified a pro-inflammatory and cytolytic PSM peptide, PSM-mec, that is genetically linked to methicillin resistance, thus providing a molecular connection between two key traits determining the outcome of S. aureus disease. Our results indicate that acquisition of SCCmec elements encoding PSM-mec by horizontal gene transfer may significantly alter the capacity of S. aureus and potentially other staphylococci to cause disease, in addition to their established role in conferring resistance to methicillin. This represents a previously unknown example of toxin hitchhiking on staphylococcal mobile genetic elements primarily in charge of transferring antibiotic resistance.
While expression of the PSM-mec peptide did not significantly alter disease progression in a strain that produces high relative and absolute amounts of other cytolytic PSMs, PSM-mec had a very significant impact when other PSMs were only expressed at low levels, indicating that acquisition of psm-mec-encoding SCCmec elements may substitute for low level PSM expression in the recipient strain. Whereas PSM-mec expression is under control of agr as all other PSM peptides, additional regulatory factors are likely responsible for the high relative production of PSM-mec observed in strains such as S. aureus MSA890.
Using large strain collections, we did not detect PSM peptides other than PSM-mec and the previously described core genome-encoded PSMs. Therefore, PSM-mec is likely the only MGE-encoded staphylococcal PSM. The frequency of PSM-mec production was high in MRSE and considerable in MRSA of pulsed-field types USA100 and USA200, which represent the most common types associated with hospital infections [38].
Kaito et al. recently described an open reading frame called “fudoh” that is reportedly involved in colony spreading and virulence [39]. The 5′ end of fudoh overlaps to a large extent with the psm-mec gene, which is transcribed in opposite direction to fudoh (Figure 4). However, in contrast to psm-mec, there is no evidence for expression of fudoh. In addition, the fudoh gene does not have a Shine-Dalgarno sequence indicating a fudoh protein product is made. Possibly, the phenotypes attributed to fudoh by Kaito et al. [39] may thus have been due, at least in part, to interference with the psm-mec locus. Nevertheless, our results and those by Kaito et al. both emphasize the importance of the fudoh/psm-mec locus in S. aureus pathogenesis.
Finally, identification of PSM-mec may explain reports on a short-chain teichoic acid with pro-inflammatory properties described in S. epidermidis and termed “lipid S” [40]. The identification of “lipid S” as such was based on an electrospray mass spectrogram that showed exactly the same two m/z peaks as PSM-mec [40]. In contrast, it did not show the series of equidistant peaks commonly found for homopolymers such as teichoic acids. It is thus likely that the mass spectrograms leading to the description of “lipid S” were misinterpreted and the extracts contained PSM-mec. This would also explain a later report on pro-inflammatory capacities of “lipid S” [41].
In conclusion, our study shows that in contrast to previous belief, staphylococci may bundle resistance and virulence factors on mobile genetic elements, thus combining the transfer of two important determinants for causing human disease in one genetic exchange event.
All animals protocols were reviewed and approved by the Animal Use Committee at Rocky Mountain Laboratories, NIAID, NIH. Human neutrophils were obtained from healthy volunteers in accordance with protocols approved by the Institutional Review Board for Human Subjects, NIAID, and the University of Tübingen, Germany.
S. aureus and S. epidermidis genome sequencing strains (S. epidermidis RP62A and ATCC12228, S. aureus COL, Sanger 252, Sanger 476, N315, Mu50, USA300, and MW2) were acquired from the Network on Antimicrobial Resistance in S. aureus (NARSA). Other MRSE and MSSE S. epidermidis strains were from Shanghai (∼100 strains), Paris (∼70), and Norway (∼100) [18]–[20], and other S. aureus strains were from a San Francisco strain collection (∼80, all MRSA) in addition to those published by Fitzgerald et al. (∼35, MRSA and MSSA) [17],[21],[22]. All strains were grown in tryptic soy broth (TSB). When necessary during cloning of the allelic replacement plasmid, antibiotics were added at appropriate concentrations (ampicillin at 100 µg/ml for cloning in E.coli; chloramphenicol at 10 µg/ml for staphylococci). For strains for which information on methicillin resistance was not available from the literature, methicillin resistance was determined by plating on TSB agar containing 6 µg/ml oxacillin.
Allelic replacement of the psm-mec gene was performed using the procedure described by Bae and Schneewind [42] which allows for gene deletion without replacement by an antibiotic resistance cassette. Using this procedure, the psm-mec gene was deleted in its entirety. Briefly, 2 PCR fragments up- and downstream of psm-mec, introducing att1 and att2 recombination sites at the distal ends and an EcoRI site at the psm-mec ends were amplified from genomic DNA of S. aureus Sanger 252. Oligonucleotides used were PSMErev1 (caagacttgcattcaggctttcggtgaattctttc), PSMEatt1 (ggggacaagtttgtacaaaaaagcaggctggaagttttgtgctttataatgaacgggagcaagc), PSMErev2 (caccagtgaattccatatgcataccctctttc), and PSMEatt2 (ggggaccactttgtacaagaaagctgggtgtaccacctagcaaagttgcaaatttgac).
After digestion with EcoRI and ligation, the resulting fragment with flanking att1 and att2 sites was cloned into plasmid pKOR1 [42] using att recombination and a Clonase kit (Invitrogen). The resulting plasmid was electroporated in S. aureus RN4220, isolated from this strain and electroporated in the target strain. Afterwards, the allelic recombination procedure was performed as described [42]. Fidelity of gene deletion was determined by analytical PCR and RP-HPLC/ESI-MS. The PSM production phenotype of all deletion and wild-type strains was verified regularly and in all pre-cultures grown for key experiments using RP-HPLC/ESI-MS. This is important to rule out spontaneous mutation in the agr system, which happens frequently [43] and owing to agr control of all PSMs [7],[29],[31] may lead to strains completely devoid of PSM production.
Typing of S. epidermidis and S. aureus SCCmec was performed using the method by Kondo et al. [44].
Peptides were synthesized by commercial vendors with an N-terminal formyl methionine residue in each peptide. Peptide sequence fidelity was determined by the Peptide Synthesis Unit of the NIAID.
The structures of synthetic PSM peptides were analyzed by CD spectroscopy on a Jasco spectropolarimeter model J-720 instrument. Solutions of PSM peptides, at 1.0 mg/ml, were prepared in 50% trifluoroethanol. Measurements were performed in triplicate and the resulting scans were averaged, smoothed, and the buffer signal was subtracted.
RP-HPLC/ESI-MS was performed on an Agilent 1100 chromatography system coupled to a Trap SL mass spectrometer using a Zorbax SB-C8 2.3×30 mm column as described [29]. Quantification was performed by integration of the UV spectra, if peaks were well separated. Alternatively, quantification was based on extracted ion chromatograms using the most abundant peaks of the electrospray ion mass spectra of the respective PSM peptides, with calibration using synthetic peptides, as described [29]. SEC/ESI-MS was performed using the same equipment as RP-HPLC/ESI-MS with a Superdex Peptide HR 10/30 column (GE Healthcare) applying an isocratic gradient of 0.1% trifluoroacetic acid in 30% acetonitrile at 0.5 ml/min.
PSM-mec was purified from S. epidermidis RP62A stationary phase culture using the same procedure as used previously for the large-scale isolation of other PSMs [16]. Briefly, supernatant was precipitated using 10% ice-cold trichloroacetic acid. The pellet was dissolved in 100 mM Tris buffer pH 8.0 and taken to neutral pH with 6 N NaOH. Then, a 2-step reversed-phase chromatography protocol was used for purification as described [16]. For N-terminal sequencing at the Peptide Sequencing Unit of the NIAID, the N-terminal formyl group was removed by heating for 2 h at 55°C as described [45].
Semi-quantitative biofilm assays using polystyrene microtiter plates and safranin staining were performed as described [46]. To assess the impact of PSM-mec on biofilm formation, the peptide was added at the time of inoculation with the indicator strain SA113 from pre-cultures (1∶100) at different concentrations. For pre-coating with fibrinogen, a 25 mg/l fibrinogen solution in phosphate-buffered saline (PBS) was filter-sterilized and 100 µl solution were pipetted in each well. After 18 h at 4°C, wells were washed twice with PBS, blocked with 2% sterile bovine serum albumin (BSA) solution for 2 h at 37°C, and washed 4 times with PBS. Then the biofilm assay was performed as described [46].
PMNs were isolated from venous blood of healthy volunteers as described [47],[48].
Neutrophils were subjected to a brief hypotonic shock with pyrogen-free water (Sigma), washed, and suspended at 5×106 cells/ml in HBSS containing 0.05% human serum albumin (HSA) (CLB). Chemotaxis of neutrophils was determined by using fluorescently-labeled neutrophils that migrated through a membrane fitted into an insert of a 24-well microtiter plate transwell system (Costar) containing a prewetted 3-µm-pore-size polycarbonate filter as described [47]. For measurement of calcium ion fluxes, 5×106 neutrophils/ml were loaded with 2 µM Fluo-3-AM (Molecular Probes) in RPMI containing 0.05% HSA (RPMI-HSA) for 20 min at room temperature under agitation, washed twice with buffer, and resuspended in RPMI-HSA at 106 cells/ml. Calcium fluxes were analyzed with a FACScalibur (Becton Dickinson).
Priming of PMNs by synthetic PSMs was determined by increased surface expression of CD11b and gp91phox (granule exocytosis). PMNs were incubated with 10–10000 ng/ml PSMs in 96-well tissue culture plates at 37°C with rotation for 60 min. The assay was terminated by centrifuging cells at 4°C for 8 min at 350×g. Cells were washed twice in cold Dulbecco's phosphate-buffered saline and stained with and isotype control antibody (BD Biosciences) or those specific for CD11b (mAb 44, BD Biosciences) or gp91phox (mAb 7D5 [49]). Propidium iodide (0.5 µg/ml) was used to identify dead cells. PMNs were analyzed on a FACSCalibur flow cytometer (Becton Dickinson) and dead cells were excluded with a single gate. Percent positive neutrophils were determined with a marker defined by the boundary of the isotype-matched control antibody.
Lysis of PMNs by synthetic PSMs or clarified S. aureus culture media was determined essentially as described [48],[50]. Synthetic PSMs (10 or 50 µg/ml) were added to wells of a 96-well tissue culture plate containing 106 PMNs and plates were incubated at 37°C for up to 3 h. At the desired times, PMN lysis was determined by release of lactate dehydrogenase (LDH) (Cytotoxicity Detection Kit, Roche Applied Sciences). Alternatively, wild-type and isogenic mutant S. aureus strains were cultured for 24 h at 37°C in 50 ml TSB with shaking using a 100 ml flask. Bacteria were removed by centrifugation and culture media were sterilized by filtration and stored at −80°C in aliquots until used. Culture medium (diluted 1∶10) was mixed with human PMNs (106) and tested for its ability to cause PMN lysis.
Measurement of IL-8 production in human neutrophils was performed as described with a commercial ELISA assay kit (R&D systems) according to the manufacturer's instructions [7].
Hemolytic activity of PSM peptides was determined by incubating samples with a 2% (v/v) sheep red blood cells and incubation for 1 h at 37°C as described [7]. Hemolytic activity of S. aureus wild-type and psm-mec deletion strains was assessed by streaking on sheep blood agar plates.
Bacteremia and skin abscess models were performed as described [7]. Briefly, mice were between 4 and 6 weeks of age at the time of use. Mice were inoculated with S. aureus from mid-exponential growth phase (3 h) at ∼1×108 CFUs/100 µl (bacteremia model) or ∼1×107 CFUs/50 µl (abscess model) as described [48]. As strains showed different aggregation leading to different CFU, optical density was used to compare cell numbers and injection of equal cell numbers was verified by quantitative RT-PCR using the gyrB gene as described [51]. Control animals received sterile saline only.
For the bacteremia model, health and disease advancement of CD1 Swiss female mice were monitored every 3 h for the first 24 h, then every 8 h for up to 72 h. Animals were euthanized immediately if showing signs of respiratory distress, mobility loss, or inability to eat and drink. All surviving animals were euthanized at 72 hours.
For the abscess model, Crl: SKH1-hrBR hairless mice were examined for skin lesions and weight at 24-h intervals for a total of 14 days. Skin lesion dimensions were measured daily with a caliper. Length (L) and width (W) values were applied to calculate the area of lesions using the formula of L × W. All animals were euthanized after completion of the entire procedure.
Statistical analysis was performed using Student's t-tests for 2, or 1-way-ANOVA with Bonferroni post-tests for more than 2 groups, and Graph Pad Prism version 5 software.
|
10.1371/journal.pcbi.1003690 | Yeast Mating and Image-Based Quantification of Spatial Pattern Formation | Communication between cells is a ubiquitous feature of cell populations and is frequently realized by secretion and detection of signaling molecules. Direct visualization of the resulting complex gradients between secreting and receiving cells is often impossible due to the small size of diffusing molecules and because such visualization requires experimental perturbations such as attachment of fluorescent markers, which can change diffusion properties. We designed a method to estimate such extracellular concentration profiles in vivo by using spatiotemporal mathematical models derived from microscopic analysis. This method is applied to populations of thousands of haploid yeast cells during mating in order to quantify the extracellular distributions of the pheromone α-factor and the activity of the aspartyl protease Bar1. We demonstrate that Bar1 limits the range of the extracellular pheromone signal and is critical in establishing α-factor concentration gradients, which is crucial for effective mating. Moreover, haploid populations of wild type yeast cells, but not BAR1 deletion strains, create a pheromone pattern in which cells differentially grow and mate, with low pheromone regions where cells continue to bud and regions with higher pheromone levels and gradients where cells conjugate to form diploids. However, this effect seems to be exclusive to high-density cultures. Our results show a new role of Bar1 protease regulating the pheromone distribution within larger populations and not only locally inside an ascus or among few cells. As a consequence, wild type populations have not only higher mating efficiency, but also higher growth rates than mixed MATa bar1Δ/MATα cultures. We provide an explanation of how a rapidly diffusing molecule can be exploited by cells to provide spatial information that divides the population into different transcriptional programs and phenotypes.
| Haploid budding yeast cells cannot actively move to find a mating partner, like some flagellated bacteria do. Instead they must grow a so-called shmoo – a mating projection – precisely into the direction of a potential partner. They communicate with each other by releasing pheromones into their environment, which are sensed by cells of the opposite mating type. This serves the localization of nearby cells and initiates growth arrest and mating. Paradoxically, yeast cells also secrete the protease Bar1 that destroys pheromones. To visualize the resulting pheromone distribution and understand the effect on mating efficiency, we combined fluorescence imaging and mathematical modeling. We observed that the controlled destruction of pheromones by the yeast cells is beneficial to communication since it causes relatively higher pheromone concentrations in areas where cells are dense and vanishing pheromone concentrations elsewhere. This allows the population to maintain two different cellular behaviors at the same time, i.e. mating and continued growth, a behavior which disappears when we genetically delete the gene for the pheromone-destroying protein.
| Cells communicate with each other by detecting and responding to external cues and stimuli. In cellular systems one can find several examples where cells coordinate growth in specific regions by sensing and responding to small differences in the concentrations of substances that provide positional information [1], [2]. Such signaling among cells is a common feature of cell populations as well as multicellular organisms, where cells often operate close to the physical limit of gradient or concentration detection [3]–[5].
The pheromone response of budding yeast (Saccharomyces cerevisiae) is an example for a eukaryotic cell communication system. Yeast cells occur either in the haploid forms MATa and MATα, or as MATa/α diploid. Haploid and diploid cells are both able to replicate vegetatively. Mating of two haploid cells with opposite mating types yields diploid cells, while haploid cells are formed through spore formation in meiosis [6], [7].
Mating is initiated by the secretion of mating type-specific pheromones, called a-factor and α-factor, which are sensed by haploid cells of the opposite mating type and trigger the mating response [8], [9]. During the mating response, yeast cells arrest their cell cycle in G1 phase and elongate in the direction of the pheromone signal by forming directed mating projections called “shmoos” [6], [10]. Yeast can sense pheromone gradients as well as absolute concentration levels. Nevertheless, yeast cells are not capable of chemotaxis and, thus, mating requires the haploid cell types to signal their location particularly to nearby potential mating partners. One way for MATa cells to regulate the extracellular α-factor is the secretion of Bar1, an aspartyl protease, which degrades α-factor [11], [12]. This leads to the paradox that MATa cells degrade the signal they need to receive. Theoretical investigations hypothesized that the major role of Bar1 is to sharpen the pheromone gradient [13], [14].
However, direct visualization of the resulting spatial α-factor concentration profiles between secreting and receiving cells is impossible due to the small size of diffusing molecules and because such visualization requires experimental perturbations such as attachment of fluorescent markers, which change diffusion properties and activities. Therefore, reaction-diffusion (RD) models have been used to simulate the pheromone distribution on the basis of physical properties of the molecules [13]–[16], but neither the validity of the model predictions nor the effect of the pheromone distribution have been tested experimentally. Except of the publication of Jin et al. [15], where it has been suggested that Bar1 promotes avoidance of the same mating type and accurate gradient detection. Using a microfluidic device it had been shown how MATa cells avoid each other when exposed to an artificial unidirectional gradient, which was reproduced quite vividly by simulations of an RD model. However, the assumptions and choice of parameters were in contrast to other works [14], [16] (compare Text S1). In general, experimental validation of RD models is complicated, especially on the molecular level [17].
Moreover, recent theoretical findings support the theory, that secretion of Bar1 in the extracellular medium does not help to align gradients in the direction of the opposing mating type [16]. In summary, the role of Bar1 is still controversially discussed. Also, none of the models proposed so far has investigated interactions of more than a few cells or the four haploid spores in an ascus, even though mating occurs not only inside the ascus, but also in a cell population which was shown by new findings where a remarkably high outcrossing rate from asci was reported [18]. This indicates that mating yeast cells interact with quite a number of potential mating partners in a natural environment. Furthermore, a recent study has shown the potential of simple secrete and sense motifs to exhibit surprising effects on the population level [19].
Therefore, we designed a method to identify the most likely α-factor distribution within mixed haploid yeast populations of thousands of cells directly from confocal microscopic images with fluorescently tagged marker proteins. Here, an RD model is used to simulate interactions of a few hundred cells at the same time. In MATa cells, the protein Fus1, which is strongly expressed upon pheromone stimulation, is tagged with GFP to record pheromone pathway activation and serves as a proxy for Bar1 induction. Therefore, the experimentally observed pheromone activation level of each MATa cell is integrated into the model and compared to the experiments. MATα cells are modified to express mCherry from the TDH3 promoter to indicate their mating type and location. In a unique way we coupled physical RD models with experimental imaging in order to quantify the spatial distribution of extracellular α-factor. The use of simple marker constructs, that altered neither α-factor nor Bar1, served for minimal interference with the biological system. We used this approach to directly estimate the influence of Bar1 on the distribution of α-factor in a mixed yeast population and suggest a novel function of Bar1 to enable the coordination of mating and growth in a yeast population in vivo.
We combined image analysis with spatiotemporal mathematical modeling to determine spatial concentration distributions of Bar1 and of α-factor. Figure 1 introduces the concept of the approach:
In order to obtain in vivo conditions during microscopy, which were suited for the described methodology, confocal microscopic images were taken from synchronized haploid cells or from equally mixed haploid MATa/MATα cell populations. The cell culture samples were spun down with low g-force on glass bottom dishes in order to have immobile cells without using concanavalin A coated dishes (which we found to alter the population response). This protocol essentially yielded sedimented cells in the same way as they would be present in any laboratory or naturally occurring non-agitated medium, with the difference that sedimentation here was achieved under controlled conditions.
Location, shape, and mating type of the cells were extracted from out-of-focus images in the brightfield and mCherry channel (Figure S1 in Text S1). The cells' spatial arrangement on the images was transferred into locally refined triangular meshes for the model (Figure S2 in Text S1) and used to calculate the extracellular spatial distributions of Bar1 and α-factor.
In mathematical terms the problem (ii) is described as a pure extracellular reaction-diffusion process for α-factor and Bar1 with distinct boundary conditions (Figure S3 in Text S1). We formulated an RD model by the following set of partial differential equations:This system covers the extracellular dynamics of α-factor concentration, , and the activity of Bar1 protease, , over time t and position in space, . The equations quantify two types of processes: (1) diffusion of both α and B, where and are diffusion constants and Δ denotes the Laplacian, and (2) degradation of α by B. Boundary conditions at the cell surfaces define secretion of α-factor by MATα cells and induced secretion of Bar1 by MATa cells.
The diffusive flux of α-factor on the cell surface and exterior system boundaries is given by:Here, is a constant specifying the secretion of α-factor. The vector points towards the cell interior and, therefore, the diffusion flux at the boundary takes a negative value for the secretion of molecules and a positive sign for the absorption of molecules. The induction of Bar1-activity at each MATa cell is calculated from the average α-factor concentration at the surface of this cell, which is represented by a Hill-curve (compare Figure 1 E, for details see Figures S4 and S5 in Text S1):The expression is the average concentration of α-factor at the i-th MATa cell at time , which promotes the secretion of Bar1. We assume a zero Bar1 flux on the MATα cells and the system exterior:The remaining exercise was to identify the parameter values of the RD model. Diffusion constants for the proteins were directly calculated from protein properties (size and density). Thus, only two parameters had to be identified by parameter estimation: the activity of Bar1 and the secretion rate of α-factor, as described below.
To determine the activity of Bar1, we used a three-step procedure. First, we performed an initial calibration to quantify the α-factor concentration perceived by individual MATa cells. We used MATa cells carrying a BAR1 deletion (bar1Δ) and the pheromone response marker Fus1-GFP and synchronized them in G1 phase, where they are responsive to pheromone. Their response to varying concentrations of α-factor (artificially added, in the absence of MATα cells) was quantified as Fus1-GFP fluorescence intensity on microscopic images. Since bar1Δ strains do not secrete the protease, the local α-factor concentration was equal to the applied concentration. Fluorescence intensity of Fus1-GFP in correlation with α-factor concentration was recorded as a calibration curve (see Figure S4 in Text S1). The calibration curve was then applied to mixed haploid cultures to determine the perceived α-factor concentration for each MATa cell on the image (see Figure 1). In mathematical terms, the calibration yielded the boundary value of α-factor concentration at the surface of a MATa cell and a functional relation between this α-factor concentration and the induced Bar1 expression.
Second, the steady state activity of Bar1 was quantified by stimulating wild type MATa cells carrying the Fus1-GFP marker in G1 phase with given concentrations of α-factor. Since there were no MATα cells, the α-factor secretion rate was equal to zero at this stage. For the model, these data were used in a mathematical optimization to quantify the activity of Bar1 for wild type cells. In order to verify that Fus1 can in fact be used as a proxy for Bar1, we used a strain expressing qVenus fluorophore under the control of the Bar1 promoter (Bar1pr-qVenus, [20]) while maintaining wild type Bar1 activity (see Figure S5 in Text S1 and Materials and Methods). The induction occurred with the same kinetics and Hill-coefficients as the Fus1 induction, but with a slight delay of the Bar1 expression verifying our used induction kinetics for Bar1. The Bar1pr-qVenus construct was able to quantify the expression levels of Bar1, but not the Bar1-induced degradation rate of α-factor in the extracellular medium, which is why we preferred the use of the Fus1-GFP data.
Third, the α-factor secretion rate could be calculated by parameter optimization from images of mixed haploid yeast cultures containing MATa Fus1-GFP cells and MATα mCherry cells and the information we obtained for the Bar1 activity. Due to the lack of evidence for an extracellular protease activity in MATα cells, we neglected potential differences in a-factor induction of α-factor. In practice, the induction of α-factor secretion should be nearly homogeneous on a single image, but may vary for different images. We obtained secretion rates from significant fits (F-test p<0.05) of 9 images (see Text S1). The mean fitted value of the α-factor secretion rate of 865 molecules per second and MATα cell is in good agreement with recent experimental measurements (550 molecules per second and cell measured as basal secretion with 2.5 - 4-fold maximal induction) [21], [22]. The fully parameterized model could now be used to efficiently calculate the entire α-factor distribution on arbitrary images. We validated the obtained model and parameters by predicting the Fus1-GFP fluorescence on an image of a larger mixed yeast population not used for model fitting (F-test p<2.2e-6). The power of this combined imaging and modeling approach is illustrated in Figures 2 and 3 and in Movie S1, which shows a mixed haploid population during growth and mating next to the simulation of the distribution of α-factor on a computational grid generated directly from the corresponding microscopic image.
We observed large differences in the estimated local α-factor concentrations between wild type cell populations and cell populations with a bar1Δ background. Dense wild type cell populations showed a strongly localized α-factor distribution at sites of high MATα cell density, with α-factor concentration quickly declining with distance. Consequently, MATa cells far away from a cluster of MATα cells experienced significantly lower local α-factor concentrations than close-by cells, and hence were often non-permissive for induction of the pheromone response (Figure 3). Populations with bar1Δ background showed an almost uniform distribution of very high pheromone concentrations, resulting in global pathway activation as evidenced by high Fus1-GFP expression. Nevertheless, the global (over-) activation led to reduced mating events.
We wanted to see whether this behavior arises in general and independently of the exact spatial composition of the culture. Thus, we performed a computational study using randomly generated cell populations mimicking the ones observed microscopically with varying cell densities (Figure 4). Each virtual population was simulated both with wild type Bar1 secretion and in bar1Δ background. We tracked key parameters such as the average α-factor concentration, the pheromone gradients perceived by the individual MATa cells (calculated as the average difference in α-factor concentration a cell would sense at its shmoo tip and the opposing cell site), and the maximum information content of the α-factor distribution. Information (or Shannon entropy) quantifies the “surprisal” of a specific α-factor concentration, i.e. how likely an observed α-factor concentration is given an overall α-factor distribution obtained for many cell populations.
Virtual wild type populations exhibited a strong gain for the information content of the α-factor distribution as the population size increases (Figure 4A). This was accompanied by increasing α-factor gradients across MATa cells (Figure 4B). In contrast, populations not secreting Bar1 showed information contents close to zero (Figure 4A) as well as insignificant pheromone gradients (Figure 4B), both independently of population density. We noted that the overall pheromone concentration remained within a range of up to 20 nM in wild type, but in the mutant linearly increased with population density (see Figure S6 in Text S1). This observation indicates that the gradients and, thus, the reachability of nearby mating partners can only be detected faithfully in cell populations secreting Bar1, particularly in high cell densities.
Additionally, we simulated various scenarios where a high-density subpopulation was placed next to a low-density subpopulation (Figure 4C–D). Here, the wild type is capable of limiting the α-factor distribution to the corresponding subpopulation, leaving the low-density subpopulation unaffected by the high local α-factor concentration of the high-density subpopulation. The same also holds true for random cell distributions (see Figure S7 in Text S1). Again, this behavior was not observed in the absence of Bar1, showing that Bar1 activity restricts the distribution of α-factor. Hence, only subpopulations with high local cell densities and small intercellular distances, as required for successful mating, were exposed to α-factor concentrations permissive for mating.
We could validate this model prediction concerning the dependency of mating success on culture density experimentally by incubating mixed MATa Rpl9A-GFP/MATα mCherry populations in cell densities varying from 0.5–10 million cells/ml for various time-points up to 5 h. We incubated the cells in Petri dishes of 36 mm diameter to ensure that the experimental density in the resulting cell layer is in agreement with the simulated density (density and distance calculations can be found in Text S1). Subsequent cell counting during bead-normalized flow cytometry allowed us to quantify the absolute number of each cell type along with the number of diploid cells in each sample (Figure 5). Here we also observed a strong dependency of diploid formation on cell density where efficient mating was predominantly observed in cell concentrations higher than 5 million cells/ml, which coincided with cell distances permissive for mating (Figure 5A). At the same time we could observe a pronounced growth phase taking place in parallel with diploid formation (Figure 5B), giving support to the finding that there was indeed a variety of different phenotypes (mating and growing cells) occurring at the same time in the same culture. Figure 5C visualizes the cell densities used in the measurements for comparison and Figure S7 in Text S1 shows the simulated levels of α-factor under these conditions, with representation of α-factor distribution and Bar1 activity distribution for one selected cell density.
Our results suggested that Bar1 acts by restricting the activity of α-factor to sites, where successful mating is possible, leaving the remaining areas free for continued growth.
In order to test the validity of this prediction and its dependency on Bar1 we observed mating between MATa cells (here marked with Rpl9A-GFP) and MATα cells (marked with mCherry) and quantified their growth rates with FACS analysis (Figure 6) and OD measurements in wild type and in bar1Δ populations during incubation (Figure 7).
Mating rates were quantified by the rate of diploid formation. To measure the diploid formation rate, we used flow cytometry for mixed populations of MATa and MATα (where we took care to not destroy extracellular gradients before the actual measurement) to quantify the fractions of MATa, MATα and MATa/α diploids over time for a fixed number of cell counts (Figure 7A). We found no difference in the rate of diploid formation between wild type and bar1Δ cultures before completion of the first cell cycle (<120 min). This observation is in agreement with our results that positive effects on the perceived pheromone gradients require higher cell densities (Figure 4A,B). However, after passing the first cell cycle, the relative fraction of diploids is clearly larger in the wild type cultures than in the mutant, consistent with the general view that Bar1 activity helps to reveal the position of mating partners [3], [13], [14]
Looking at population growth during mating, we found strong differences between wild type and bar1Δ cultures (Figure 7B). For bar1Δ cultures, the global activation of the pheromone response in effectively all MATa cells of the population led to an almost complete loss of population growth (Figure 7B,C). This also caused a characteristic population phenotype with many pheromone-stimulated MATa cells being significantly larger than normal MATa cells and showing multiple mating projections (Figure 7E,F). This phenotype was never encountered in unperturbed wild type mixtures of MATa and MATα cells, but could be induced by swirling them rapidly to inhibit cell fusion (Figure S8 in Text S1). Thus, this phenotype appears associated with induction of pheromone response in vivo under conditions where a cell cycle arrest has been induced but successful mating is inhibited.
Wild type cultures exhibited significant growth on the population level despite the higher rate of diploid formation and a normal phenotype of MATa cells (Figure 7B,D–E).
The effect of Bar1 secretion on haploid growth rates was even more prominent when looking at the MATa/MATα ratio in the population (Figure 7C). There is no known secretion of an extracellular protease described for MATα cells equivalent to Bar1. Co-cultured wild type MATa cells strongly outperform MATα cells in growth during mating to an extent that within 5 hours MATa is the predominant haploid cell type in the population. This cannot be observed in bar1Δ background where the MATa/MATα ratio remains constant, presumably because both haploid cell types are equally inhibited in growth. In summary, secretion of Bar1 enables a high mating rate on a population level, but also strongly optimizes the population growth rate by avoiding unnecessary cell cycle arrest when mating is improbable (Figure 7F).
In order to further validate our findings about the role of Bar1 in the mating process, we mixed MATα cells with different ratios of wild type and bar1Δ MATa cells. We measured the amount of haploid and diploid cells after 4 h of incubation by FACS analysis (Figure 8). For labeling of MATα cells we again used mCherry, whereas MATa BAR1 wild type cells were labeled with Rpl9a-GFP and bar1Δ MATa cells with Rpl9a-TagBFP2. For equal ratios of wild type cells of both mating types at time 0 h we obtained a diploid fraction of 14% at time 4 h (leftmost columns). However, MATα cells assumed about 23% while MATa cells reached more than 62% of the total population, again supporting the observation that a part of the MATa population engages in mating and other cells continue to grow.
For equal ratios of wild type MATα and bar1Δ MATa at start, we obtained only 3.1% diploids and roughly equal ratios of the haploid cells after 4 h. This is in agreement with the view that essentially all cells stop growing and start to prepare for mating. Microscopic imaging confirms that all cells are shmooing under this condition (Figure 8B). When mixing 50% of MATα with different ratios of wild type bar1Δ MATa cells, we obtain diploid cells of both types at ratios as could roughly be expected from the mixes with either wild type or mutant MATa. Strikingly, for small ratios (5%) of wild type MATa cells at 0 h, we see that bar1Δ MATa mate more frequently (4.5% of total population) than in pure mutant mixes (3.1%), indicating that they profit from the Bar1 secreted by wild type MATa cells. Figure 8C shows a microscopic image of a mixture of 5% wild type MATa and 45% bar1Δ MATa cells together with 50% MATα at the beginning. Here, the bar1Δ MATa cells exhibit clearly lower levels of shmooing compared to the 50%/50% mix in Figure 8B. A few successful mating events leading to diploids are indicated by white arrows.
Again, the experiments mixing wild type cells with cells not secreting Bar1 – the bar1Δ or cheater cells – confirm the role of Bar1 which is even supportive for those cells not actively secreting it, but experiencing its effect on α-factor levels and gradients.
The combination of spatial modeling and a series of in vivo experiments employing images of mating cells allowed us to quantitatively describe the distribution of the pheromone α-factor in the intercellular space and the role of the protease Bar1 in shaping the pheromone pattern. We found that secretion of Bar1 is a highly cooperative mechanism. Haploid MATa cells secrete individually small quantities of Bar1 molecules. These few molecules are quickly distributed across the population by diffusion and generate Bar1 activity that strongly influences the diluted distribution of α-factor. For example, a low cell density of about 106 cells/ml induces a degradation rate of α-factor in the range of 10−3 molecules per second which corresponds to almost no degradation of α-factor, however, high cell densities can create substantial degradation by the Bar1 activity where single α-factor molecules are cleaved within a second after secretion (compare Text S1). Thus, individual MATa cells need to secrete only very small numbers of Bar1 protease. The global concentration of Bar1 appears to be fine-tuned to ensure a highly informative α-factor distribution. This effect is based on the interplay of the secretion of α-factor, its diffusion, the activation of Bar1 transcription and α-factor degradation by Bar1. Thus, we observed steep gradients in the distribution of α-factor only in high densities of MATa cells where the joint degradation by Bar1 limits α-factor diffusion sufficiently. Our conclusions differ from what has been shown in previous publications where the effect of Bar1 is demonstrated either completely theoretical for one or a few cells [13], [14], [16] or experimentally for an artificial setup in a microfluidic device without mating partner [15]. First, we observed little to no effect of Bar1 in a setting with only few cells since the Bar1 activity is insufficient to degrade α-factor before it diffuses a large distance. Our results also indicate that Bar1 rather acts on the level of subpopulations than on a few individual cells. Second, we also did not observe an effect reported earlier that Bar1 secretion leads to “self-avoidance” [15]. Third, we modeled and experimentally quantified the induction of Bar1 due to stimulation with α-factor (compared to constant Bar1 levels as in [13] or constant secretion as in [14], [15]).
Remarkably, the simple circuit of pheromone secretion and degradation by jointly secreted Bar1 is able to produce highly dynamic behavior in yeast populations. In general, the α-factor concentration profile recovers the regions where mating has a high chance of success, but quickly drops at all other regions, thus allowing cells to continue growth when there is only little chance for successful mating (see Figure 7E, F). This is further regulated by a stimulated Bar1 production, which depends on the extracellular α-factor concentration. Stimulation of Bar1 secretion might be a strategy to adapt the zone of influence of α-factor to varying cell densities and numbers of MATα cells in the population. Cells will react to high α-factor concentrations with strong secretion of Bar1, which culminates in a steady state permissive for efficient mating (also see Text S1 and Movie S1).
A lack of Bar1 in mixed haploid populations has crucial influence on the population phenotype since it leaves many cells in a prolonged cell cycle arrest in G1 phase along with activation of the pheromone response. Moreover, the temporally extended stimulation leads to larger cells with many mating projections. This result is further supported by the observation that MATa cells strongly outperform MATα cells in growth, an effect depending on Bar1. Thus, Bar1 is beneficial for growth as well as diploid formation because it enables continued growth for large parts of the population, but it also provides an enhanced ability to interpret the extracellular pheromone signal at sites where many cells cluster into locally dense subpopulations. While the gradient-enhancing effect of Bar1 has been reported before, we additionally connect it to the requirement of high local cell densities [13]. This indicates that the gradient enhancing ability of Bar1 has evolved to take place selectively in dense yeast populations and is not a treat of individual yeast cells in large volumes.
Our observations suggest that the yeast populations segregated into a spatial pattern with localized regions of diploid formation and other regions of continued growth. As a consequence, the entire population is divided into two different work programs. Both of those programs are performed in parallel and Bar1 is sufficient to induce this separation by forming a locally varying α-factor pattern. Depending on the overall cell density, the locations reserved for mating show high local concentrations of α-factor, whereas the locations reserved for growth show negligible concentrations (compare Figures 5 and 7). However, one may wonder how this cooperation arose during evolution since cheater MATa cells not producing Bar1 can also easily exploit it. A possible explanation lies within the spatial structure of the population, since the fitness benefit conferred by Bar1 is locally restricted. Haploids can only arise from a small set of initial spores, which form subpopulations by budding and mating type switching. As a consequence, cells profiting from the collective secretion of Bar1 are likely to be genetically related. Under these circumstances there is indeed evidence that cooperation can be conserved during evolution [23], [24]. Our experiments show that cheater cells not producing Bar1 can indeed profit from the Bar1 secreted by non-cheater wild type MATa cells in a mixed population (compare Figures 7 and 8). However, the non-cheater cells consistently outperform the cheater cells in mating as well as growth, indicating that the non-cheater cells maintain an advantage even in a mixed population of cheaters and non-cheaters.
The analyzed regulatory circuit of combined pheromone and protease secretion is not only observed in Saccharomyces cerevisiae, but is also found in other fungi [25]–[27]. Furthermore, a similar mechanism is known in Dictyostelium discoideum secreting phosphodiesterase (PDE) during detection of cAMP [28], [29]. Taking this into consideration, the described mechanism might be a general strategy to separate a cell population into subpopulations with different transcriptional programs.
Our methodology of quantifying the distribution of extracellular morphogens in the absence of direct measurement also has potential applications in other problems of cellular communication and pattern formation. A reduction from the computationally very expensive 3D problem (especially for parameter estimation) to an integrated 2D problem is feasible for any cells that sediment to the bottom of the containing volume under non-agitated conditions. However, this computational tool could be used to model the behavior of any culture in a non-moving liquid film such as on the surface of fruits or any controlled fermentation such as wine or beer production where the liquid is kept still for some time. This makes the method applicable for clinical research as well since biofilm formation involving quorum sensing is a major complication when fighting bacterial infections [30], [31]. Here, it might be helpful to quantify the distribution of quorum signals in order to find possible ways to optimally disrupt the system.
Wild type MATa reporter strains used in this study are Fus1-GFP and Rpl9A-GFP. They are based on BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) and part of the yeast GFP collection [32]. The MATα reporter strain expressing mCherry under control of the TDH3 promoter (MATα can1Δ STE2pr-SpHIS5 lyp1Δ::STE3pr-LEU2 his3Δ1 leu2Δ0 ura3Δ0 met15Δ hoΔ0::TDH3pr-mCherry-NATMX4) was a friendly gift of Alexander DeLuna [33]. As mutant MATa reporter strains we used two different strains: bar1Δ Fus1-GFP and bar1Δ Rpl9a-TagBFP2. The first mutant was created by deletion of the BAR1 gene in the Fus1-GFP strain mentioned above, the second was cloned by tagging of the Rpl9a Gene with Tag-BFP2 in the BY4741 bar1Δ strain. Rpl9a tagging was used because of the high expression level and also since it is not known to be involved in the mating process. Yeast strains were cultivated at 30°C in synthetic medium (0.17% yeast nitrogen base without amino acids, 0,5% ammonium sulfate, 2% glucose, 55 mg/l adenine, 55 mg/l L-thyrosine, 55 mg/l uracil, 20 mg/l L-arginine, 10 mg/l L-histidine, 60 mg/l L-isoleucine, 60 mg/l L-leucine, 40 mg/l L-lysine, 10 mg/l L-methionine, 60 mg/l L-phenylalanine, 50 mg/l L-threonine and 40 mg/l L-tryptophane). BAR1 deletions in MATa reporter strains were inserted by homologous integration of a URA3 cassette in the BAR1 locus (bar1Δ0::URA3). PCR amplification of the URA3 cassette from plasmid template pESC-Ura (Stratagene) was done by sequential amplification with the primer pairs 1/2 and 3/4 shown in Table 1.
This was followed by transformation and selection on agar plates with synthetic medium lacking uracil. Verification of the BAR1 deletion was done with a physiological assay based on growth inhibition by α-factor pheromone [34].
The bar1Δ Rpl9a-TagBFP2 reporter strain was cloned by PCR amplification of a TagBFP2 loxP-Ura3-loxP transformation cassette with primer pairs 5 and 6 from Table 2. As PCR template we used vector EKP232. EKP 232 was cloned by ligation of TagBFP2 into PstI site of pUG72 [35].
The qVenus expressing strain under control of the Bar1 promoter (Bar1pr-qVenus) was cloned by using the plasmid pSP 34 from Serge Pelet [20]. Bar1 promoter region [−500 bp] was amplified from genomic DNA (strain BY4741) by PCR using primer pair 9 and 10 as well as the first 51 bp of the BAR1 gene using primer pair 7 and 8 shown in Table 3. PCR product of promoter and gene were mixed and used as template for a fusion PCR with a PmlI restriction site between promoter and gene. Included in forward and reverse primer were restriction sites for SacI and PstI respectively. The fusion PCR product was ligated into SacI/PstI site of pSP 34 resulting in plasmid EKP252. Plasmid EKP252 was linearized using PmlI restriction enzyme and used for transformation and homologous integration into the BAR1 locus under preservation of the BAR1 gene. Positive clones were selected in minimal medium lacking leucine, α-factor induced qVenus expression was controlled microscopically, and wild type Bar1 activity was verified by a physiological assay based on growth inhibition by α-factor pheromone [32] and comparison with Bar1 wild type and deletion strains.
Microscopic images were acquired with an inverted FluoView 1000 microscope (Olympus, Tokio, Japan) equipped with a 60× (1.2 N.A) water-immersion objective and a climate chamber (Tokai Hit, Japan). GFP was excited with a 488 nm argon laser and mCherry with a 559 nm laser diode. Fluorescence emission was detected in the range 500–545 nm and 570–670 nm, respectively. The Bar1pr-qVenus construct was excited with 515 nm and detected between 530 nm and 630 nm, for Rpl9a-TagBFP2 we used 405 nm excitation and as detection range 425–475 nm. For mating experiments, MATa Fus1-GFP wild type and bar1Δ reporter strains as well as the MATα reporter strain (TDH3pr-mCherry), were cultivated to mid logarithmic phase and mixed equally. Mating was followed over indicated time periods microscopically while microscopic samples were kept in cultivation medium at 30°C.
Image acquisition for α-factor calibration curves was done with synchronized cultures of Fus1-GFP wild type and bar1Δ. Cultures were synchronized in G1 phase by elutriation with a Beckman Coulter JE-5.0 elutriation system. Synchronized cells were incubated with α-factor pheromone for 3 hours at 30°C. Afterwards cells were spinned down on the surface of a glass bottom dish (MatTek Corporation, Ashland, US) by centrifugation at 100× g using self-built accessories. For Fus1-GFP wild type, α-factor pheromone concentrations in the range between 0 µM–100 µM were used and for bar1Δ we used 0.1 nM - 1 µM. Mean fluorescence intensity of Fus1-GFP was analyzed as described in the Computational Techniques (see Text S1). For validating the employment of Fus1-GFP as proxy for the mating response pathway we used a strain expressing qVenus under control of the Bar1 promoter in BAR1 wild type background. Non-synchronized cells were incubated with α-factor pheromone as described for Fus1-GFP BAR1 wild type cells and analyzed in the same way. A comparison of the results is shown in Figure S5 in Text S1.
Growth of equally mixed MATa and MATα reporter strains, as well as a haploid control strain was analyzed by measuring optical density at 600 nm with a Photometer (Eppendorf Bio Photometer plus) and in parallel by analysis of cell number and cell size distribution with a cell counter (Casy Counter TTC, Schärfe System). Yeast cells were incubated in a water bath at 30°C without shaking. In time steps of 15 min samples were removed from the water bath, vortexed, appropriately diluted, and analyzed in duplicate. To quantify the amount of haploid MATa and MATα cells, of diploid cells or of cells within the mating process, we measured fluorescence intensities for GFP and mCherry of 10.000 living cells of each sample by FACS analysis taking advantage of the fluorescence of MATa Rpl9A-GFP and MATα mCherry in a BD FACS AriaII cell sorter (Becton Dickinson, Franklin Lakes, NJ), equipped with a 488 nm and a 561 nm laser with filter sets for GFP (525/50 BP, 505LP) and for mCherry (610/20BP, 600LB). Cultures were incubated in a water bath at 30°C without shaking. In 20 min time steps, duplicate samples were removed from the water bath, mixed vigorously, diluted in PBS and FACS analyzed. Gates for MATa, MATα and diploids were set by hand identifying the cell types as shown in Figure 6.
As proof of the model prediction we performed a mating experiment with different cell densities. MATα TDH3pr-mCherry and MATa RPL9a-GFP BAR1 wild type cells were grown in SD medium to mid log phase. Cells were diluted in SD medium and cell numbers were adjusted to 10·106 cells/ml by measuring the cell number with a CasyTTC cell counter. MATα and MATa cells were mixed 1∶1 and diluted in SD medium in following concentrations: 10·106, 5·106, 2.5·106, 1·106, 0.5·106 cells/ml. 2 ml aliquots of the diluted cultures were incubated at 30°C in Petri dishes with a diameter of 36 mm (Falcon), in order to get an average monolayer of cells after sedimentation. In time steps of 15 min one Petri dish of each cell dilution was removed from the incubator, cells were re-suspended in the medium by intensive pipetting and 400 µl of the cultures were mixed with 400 µl PBS supplemented with CaliBRITE APC Beads (BD Biosciences #340487). The samples were analyzed by FACS. APC Beads were recorded with 640 nm excitation and 670/41BP filter, MATa Rpl9a-GFP and MATα mCherry reporter strains as mentioned above. APC beads were gated and used as internal standard. In each sample the number of cells corresponding to a fixed number of 90000 APC beads was analyzed, giving not only the relative amount of haploids and mating events but also the growth behavior of the components of the mixed culture (results of the experiment are shown in Figure 5).
To analyze the influence of bar1Δ cheater cells in mating mixtures we used MATa Rpl9a-TagBFP2 bar1Δ reporter strain, together with the already introduced MATa and MATα reporter strains. The three strains were grown in SD media to mid logarithmic growth phase and diluted in SD media to 1·107 cells/ml. Several mating-with-cheaters-mixtures were prepared as shown in Table 4.
The images were analyzed with CellID [36] to extract mating type, fluorescence activity, as well as position, size and shape of the cells. These data were transferred to a computational domain (see Figures S2 and S3 in Text S1 for details). From this computational domain a triangular mesh was generated using Gmsh [37] that can be used by various RD toolboxes. Here, we used the open source Toolbox DUNE [38] to solve the stationary as well as time-dependent equations by a finite element method with high accuracy [39], [40]. In Supporting Text S2 we systematically compare 2D and 3D simulations to account for the fact that images are taken in 2D, while diffusion and mating happen in 3D (see Figures S9, S10, S11 in Text S2). We found that the maximum difference between 2D and 3D in the simulated α-factor distribution was below 5%. Therefore, for the parameter fit the 2D solution was used, which resulted in a major speed-up.
|
10.1371/journal.pgen.1005822 | Regulation of Stem Cell Proliferation and Cell Fate Specification by Wingless/Wnt Signaling Gradients Enriched at Adult Intestinal Compartment Boundaries | Intestinal stem cell (ISC) self-renewal and proliferation are directed by Wnt/β-catenin signaling in mammals, whereas aberrant Wnt pathway activation in ISCs triggers the development of human colorectal carcinoma. Herein, we have utilized the Drosophila midgut, a powerful model for ISC regulation, to elucidate the mechanisms by which Wingless (Wg)/Wnt regulates intestinal homeostasis and development. We provide evidence that the Wg signaling pathway, activation of which peaks at each of the major compartment boundaries of the adult intestine, has essential functions. Wg pathway activation in the intestinal epithelium is required not only to specify cell fate near compartment boundaries during development, but also to control ISC proliferation within compartments during homeostasis. Further, in contrast with the previous focus on Wg pathway activation within ISCs, we demonstrate that the primary mechanism by which Wg signaling regulates ISC proliferation during homeostasis is non-autonomous. Activation of the Wg pathway in absorptive enterocytes is required to suppress JAK-STAT signaling in neighboring ISCs, and thereby their proliferation. We conclude that Wg signaling gradients have essential roles during homeostasis and development of the adult intestine, non-autonomously controlling stem cell proliferation inside compartments, and autonomously specifying cell fate near compartment boundaries.
| The highly conserved Wingless/Wnt signal transduction pathway directs many cellular processes in metazoans and its deregulation underlies numerous human congenital diseases and cancers. Most notably, more than 80% of colon cancers arise from aberrant activation of the Wnt pathway. A better understanding of how Wnt signaling functions in the intestinal stem cells (ISCs) during homeostasis and in disease states is thus critical. The Drosophila digestive tract provides a powerful genetic model and an entry point to study these questions. Here, we find that the Wg ligand and pathway activation are enriched at Drosophila intestinal compartment boundaries and are essential for development and homeostasis of the adult gut. During homeostasis, Wg signaling in enterocytes is required to prevent the overproliferation of ISCs non-autonomously. In addition, during development, Wg signaling ensures proper cell fate specification near compartment boundaries. These findings provide insight into the mechanisms underlying the Wg-dependent regulation of adult intestinal function.
| The evolutionarily conserved Wnt/β-catenin signal transduction pathway regulates cell proliferation and tissue patterning in metazoans, and deregulation of this pathway is associated with numerous human diseases [1,2]. In the absence of Wnt/Wg exposure, the key transcriptional activator β-catenin/Armadillo (Arm) is targeted for proteasomal degradation by a “destruction complex” comprised of Axin, Adenomatous polyposis coli (Apc1 and Apc2) and glycogen synthase kinase 3 (GSK3)/shaggy (sgg). Binding of Wnt/Wg to the transmembrane co-receptors Frizzled (Fz and Dfz2) and low-density lipoprotein receptor-related protein 5/6 (LRP6)/Arrow (Arr) recruits the adaptor protein Dishevelled (Dvl/Dsh) and deactivates the destruction complex. Stabilized β-catenin subsequently translocates to the nucleus, and associates with the transcription factor T-cell factor/lymphoid enhancer factor (TCF/LEF) and the cofactors Pygopus (Pygo) and BCL9/Legless (Lgs) to activate target genes (S1A Fig) [3–7]. Notably, Wnt signaling is essential for self-renewal and proliferation of mammalian ISCs, whereas Wnt pathway hyperactivation triggers the development of the vast majority of colorectal carcinomas [2,8].
The Drosophila digestive tract, with its remarkable similarity to the mammalian intestine but simpler anatomy, is an ideal model for studying intestinal development, homeostasis, and disease [9–12]. Like its mammalian counterpart, the fly gut undergoes rapid turnover and is replenished by ISCs. The ISCs, which are distributed along the basement membrane of the gut epithelium, divide asymmetrically to give rise to enteroblasts and pre-enteroendocrine cells that differentiate into either absorptive enterocytes or secretory enteroendocrine cells, respectively [13–17]. Visceral muscles envelop this monolayer epithelium. Food is successively digested, absorbed and eliminated through the foregut, midgut and hindgut. The midgut can be further partitioned into fine-scale compartments of unique histological structure, gene expression, and physiological function [18–20], denoted as R1 (Region 1) through R5 (Fig 1A). Intriguingly, the peak expression of frizzled 3 (fz3), a direct target gene of the Wg pathway [21,22], occurs at four of these compartment boundaries (cardia-R1, R2-R3, R3-R4 and R5-hindgut) [18].
Wg signaling is required for intestinal regeneration after injury caused by toxins or bacterial infection, whereas the aberrant activation of signaling deregulates ISC proliferation [23–29]. The roles of Wg signaling during homeostasis have also been examined previously. An initial study indicated that wg expression in the muscle surrounding the midgut activates the pathway in midgut ISCs, and is required for their self-renewal and proliferation [30]. However, later studies challenged these conclusions [25,26]. First, ISC self-renewal was not affected upon loss of Apc [26]. Second, knockdown of wg from both muscle and epithelial sources, or in wgCX4 heterozygous mutants did not lead to significant loss of ISCs even after 30 days [25]. Third, genetic inactivation of core Wnt pathway components with null alleles resulted in only mild or no effects on ISC proliferation during homeostasis [25]. The only method that revealed strong effects of Wg signaling on ISC self-renewal [25,26,30] required ectopic expression of a dominant negative dTCF protein [4].
The recent discovery that Wg signaling is enriched specifically at compartment boundaries, as revealed by activation of the target gene fz3 [18], prompted us to reexamine the source and roles of Wg during adult intestinal homeostasis and development. Here, we identify several novel sources of Wg in the intestinal epithelium and in the surrounding visceral muscle, which all peak at compartment boundaries. We confirm that Wg pathway activation also peaks at these boundaries, but also find that low-level Wg signaling is present throughout compartments, where it is essential for maintenance of homeostasis. Further, in contrast with the prior focus on Wg signaling in midgut stem cells, our findings reveal that enterocytes, and not ISCs, are the primary site of Wg pathway activation during homeostasis. Wg signaling in enterocytes non-autonomously regulates JAK-STAT signaling in neighboring ISCs, and thereby prevents ISC overproliferation during homeostasis. Further, we demonstrate that Wg signaling is required for proper cell fate specification near compartment boundaries during development. We conclude that gradients of Wg signaling that peak at adult intestinal compartment boundaries are essential to control stem cell proliferation during homeostasis and to specify cell fate during development.
To identify regions in the adult intestine that express wg, we examined a wg-lacZ enhancer trap [31](Fig 1B). Consistent with prior findings, four rows of wg-expressing cells were detected in the surrounding circular visceral muscles throughout the entire length of the midgut [30](Figs 1D and S1C). Furthermore, at the foregut-midgut boundary, wg was expressed in the anterior cardiac epithelium, whereas at the midgut-hindgut boundary, wg was expressed in the spindle cells of the hindgut proliferation zone (HPZ)[23,32–35] (Figs 1C, 1E and S1B).
Unexpectedly, we also identified four novel sources of Wg at intestinal compartment boundaries (Figs 1F–1I” and S1D–S1D”). First, we observed high-level wg expression in a band of approximately 18 visceral muscle cells in the middle of the midgut, adjacent to the anterior border of the copper cell region (CCR, marked by Cut and α-Spectrin [24]) and the R2-R3 compartment boundary (Fig 1B and 1F-1G”). Second, we identified a zone of wg expression in the visceral muscle at the posterior terminal midgut, immediately anterior to the R5-HPZ border [23,34] (Fig 1H–1H”). Third, we detected high-level wg expression in the intestinal epithelium underneath this wg-enriched muscle segment, which was present not only at the R5-HPZ boundary, but also extended anteriorly into the posterior midgut (Fig 1I–1I”). Lastly, a transverse stripe of wg was present within the rectal epithelium (S1D-S1D” Fig) [36]. Together, these results indicated that wg is not expressed uniformly along the fly gut, but is enriched at intestinal compartment boundaries. Notably, this expression pattern was largely retained in aged flies, indicating that the enrichment of wg expression at major compartment boundaries was stable over the course of adulthood (S1E–S1E” Fig).
These unprecedented observations of boundary-enriched wg expression were recapitulated using a wg:mcherry knock-in line [37] (S2 Fig), and when lacZ expression was driven by a wg:Gal4 knock-in line [37] (S3 Fig). Of note, expression of wg>lacZ culminated at all major compartment boundaries (S3 Fig) and the epithelial sources were detected in enterocytes (S4 Fig). In summary, the newly discovered sources in the epithelium and muscle, coupled with previous findings, revealed marked enrichment of wg expression at adult intestinal compartment boundaries that correlate with major sphincters and tissue-organizing centers.
Having established the sources of Wg that are present in the epithelium and visceral muscle of the adult midgut, we sought to identify the regions where the Wg pathway is activated during homeostasis. We confirmed that a direct target gene of Wg signaling, frizzled 3 (fz3) [21,22,38], was expressed in five gradients in the adult gut, each of which peaked at intestinal compartment boundaries, as described previously [18] (Fig 2A–2D and 2E–2F). We also detected strong fz3-RFP signal at the anterior border of the rectal papillae (S5A and S5A” Fig). To validate these conclusions, we analyzed a reporter for another direct Wg pathway target gene, naked (nkd) [39]. As observed for fz3-RFP, nkd-lacZ was also expressed highly at intestinal compartment boundaries, though in a narrower pattern (Fig 2A and 2B’–2F). Further, nkd-lacZ was also expressed in the rectum (S5A’ and S5A” Fig), and, as observed for wg, in four rows in the visceral muscle surrounding the intestine (Fig 2A).
Notably, we also detected expression of both fz3-RFP and nkd-lacZ at low levels within posterior midgut compartments (Fig 2A and 2E–2F). To determine whether this low-level expression was dependent on Wg signaling, we inactivated the pathway by generating MARCM clones [40,41] of null alleles of the Wg pathway components pygo and dsh [6,42] alongside wild-type control clones. The expression of fz3-RFP and nkd-lacZ were eliminated in mutant clones, but not in wild-type clones, indicating that the Wg pathway is indeed activated at low levels in midgut compartments (Fig 2G–2J’). Together, these findings indicated that the Wg pathway is activated at high levels at the boundaries between functionally distinct intestinal compartments, but also activated at low levels within these compartments (Fig 2K).
Recent studies have indicated that the larval intestine is also compartmentalized [18,19,43]. We therefore sought to determine whether the boundary enrichment of wg and Wg pathway activity are also present in the larval counterpart, and thus examined the expression pattern of wg-lacZ, wg>lacZ and fz3-RFP in 3rd instar larval guts. Remarkably, as in adults, wg expression that emanates from muscle and epithelial sources (S6 Fig) was enriched at compartment boundaries of the larval intestine. Further, fz3-RFP expression also culminated at these boundaries (S7 Fig). These findings suggested that even though the larval and adult intestines are derived independently and face different functional demands, the patterns of boundary-enriched Wg pathway activation are largely shared. To further determine whether Wg signaling was required for fz3-RFP expression in the larval gut, we induced either null mutant clones of Wg pathway components, or wild-type control clones, during larval development. Inactivation of Wg signaling resulted in specific loss of fz3-RFP within AMPs at compartment boundaries (S8A–S8D”‘ Fig). Moreover, inactivation of the Apc-Axin destruction complex resulted in the ectopic expression of fz3-RFP in AMPs but not enterocytes [44] (S8E–S8E”‘ Fig). We conclude that in the larval midgut, the Wg pathway is activated exclusively in progenitor cells, and this activation is restricted to compartment boundaries.
To test whether stem cells are also the primary sites of Wg pathway activation in the adult intestinal epithelium, we first identified the cell types in which the two Wg pathway reporters, fz3-RFP and nkd-lacZ, were expressed (Fig 3A–3B”‘). We found that nkd-lacZ was expressed primarily in enterocytes (demarcated by their large polyploid nuclei) and also in a subpopulation of enteroendocrine cells (small diploid cells that are Escargot negative and Prospero positive [13]) (Fig 3A–3A”‘), whereas fz3-RFP was detected not only in enterocytes, but also in progenitor cells (small diploid Escargot positive cells [13]) (Fig 3B–3B”‘). Thus, unexpectedly, despite the overlapping boundary-enriched pattern, the expression of the two Wg pathway reporters was partially distinct with regard to individual cell types.
We therefore sought to determine whether expression of the reporters was dependent on Wg signaling by generating marked null mutant clones of four Wg pathway components: Dsh, Pygo, Arr [45] and the functionally redundant Fz and Dfz2 [46], alongside wild-type controls. Nkd-lacZ and fz3-RFP expression were not affected in wild-type control clones (Figs 2G–2H’, 3C–3C”‘ and 3E–3E”‘). In contrast, both inside compartments and at compartment boundaries, the inactivation of Wg signaling in the mutant clones eliminated expression of both nkd-lacZ (Figs 2J–2J’, 3D–3D”‘, S9A–S9A”‘ and S10A–S10A”‘) and fz3-RFP (Figs 2I–2I’, 3F–3G”‘, S9B–S9E”‘ and S10B–S10C”‘) in enterocytes, revealing dependence on Wg pathway activation in this cell type. Unexpectedly, in mutant clones, there was no decrease in the level of nkd-lacZ expression in the subpopulation of enteroendocrine cells (Figs 3D–3D”‘ and S10A–S10A”‘), indicating that the nkd-lacZ reporter expression in enteroendocrine cells was independent of Wg pathway activation. Further, there was no decrease in the level of fz3-RFP expression in the vast majority of progenitor cells (Figs 3F–3F”‘, S9B–S9D”‘ and S10B–S10C”‘), suggesting that the fz3-RFP reporter expression in nearly all progenitors was independent of Wg pathway activation. The only exception was inside R5, where both enterocytes and progenitors were responsive to Wg exposure, as indicated by loss of fz3-RFP expression in the mutant clones of Wg pathway components in this subregion (Figs 3G–3G”‘ and S9E-S9E”‘). We conclude that, in contrast with the larval gut, the primary site of Wg pathway activation during adult homeostasis is in enterocytes, and not ISCs, throughout the majority of the midgut.
The pattern of wg expression cannot account for the preferential activation of signaling specifically in enterocytes. Therefore, we sought to determine whether intrinsic differences in the distinct intestinal cell types could explain their differential response to Wg exposure. We examined the response of the different cell types to ectopic pathway activation by analyzing the expression patterns of fz3-RFP in Apc1 null mutant flies or in Apc2 Apc1 double mutant clones [47], in which the Wg pathway is constitutively activated due to loss of destruction complex activity. Compared with wild-type, fz3-RFP expression was markedly increased in Apc1 mutant guts and in Apc2 Apc1 double mutant clones (Figs 4A–4C”‘ and S11). This aberrant Wg pathway activation was not restricted to enterocytes, but instead observed in all cell types in the intestinal epithelium. These findings indicated that the Wg pathway components acting downstream of the Apc-Axin destruction complex are present in all cell types in the intestinal epithelium, and thus do not underlie their differential responsiveness to Wg stimulation.
To determine whether Wg pathway components that act upstream of the destruction complex are functional in all cell types of the intestinal epithelium, we expressed wg throughout the muscle using a temperature sensitive dMef2-Gal4 driver [48]. As compared with controls (Fig 4D), overexpressing wg in the muscle markedly increased the fz3-RFP signal throughout the entire length of the intestine, and most pronouncedly in the anterior midgut (Fig 4D’). These results indicated that Wg originating from the visceral muscle is sufficient to activate signaling in the intestinal epithelium. Notably, this ectopic activation of Wg signaling was observed in all intestinal cell types (Fig 4E–4E”‘). Thus, all cell types in the intestinal epithelium express the pathway components necessary for transduction of the Wg signal, and are thus capable of responding to Wg exposure. Therefore, the activation of signaling primarily in enterocytes during homeostasis may reflect inherent differences in the threshold for pathway activation (see discussion).
Having discovered that enterocytes are the primary sites of Wg pathway activation in the homeostatic adult gut, we sought to determine whether this signaling was required to maintain homeostasis. We inactivated the Wg pathway by inducing arr, pygo or dsh null mutant clones during early adulthood and examined the effects 5 to 7 days later. We found that wild-type control clones were comprised primarily of 1 to 2 cells (Fig 5A), and were surrounded by regularly spaced progenitor cells. In contrast, a higher percentage of multi-cellular clones of Wg pathway mutants were detected (Fig 5A). Furthermore, an increased number of wild-type progenitors (esg-lacZ marked cells or small Armstrong Prosperoneg cells) were present in clusters adjacent to the mutant clones, whereas progenitor cells located farther away from the clones exhibited normal spacing and number (Figs 5B and S12A–S12B”). Further, Dl-lacZ and Su(H)-lacZ expression revealed that these clusters contained an increased number of both ISCs and EBs (Fig 5C–5F). We further sought to determine whether the aberrantly increased number of progenitors adjacent to Wg pathway mutant clones resulted from their overproliferation. To test this, we compared the mitotic index in posterior midguts bearing either wild-type or Wg pathway mutant clones. Indeed, many more phospho-histone H3 (pH3) positive cells were observed in posterior midguts containing pygo or dsh clones by comparison with controls (Figs 5G and S12C and S12D). Thus, non-autonomous ISC overproliferation was detected. There remained the possibility that the normal process of ISC migration following mitosis was also disrupted, and contributed to their aberrant clustering. This non-autonomous overproliferation defect was observed only when the Wg pathway was inactivated during adulthood, but not prior to eclosion (S10 Fig). Together, these findings indicate that Wg signaling prevents the non-autonomous overproliferation of neighboring ISCs during adult homeostasis.
To test the possibility that Wg signaling in enterocytes non-autonomously regulates the proliferation of neighboring progenitor cells, we disrupted signaling by knocking down Wg pathway components using RNA-mediated interference (RNAi), or by expressing dominant-negative Legless (Lgs17E) [5] or dominant-negative TCF (dTCFΔN) in enterocytes or progenitor cells with the cell type-specific MyoIA or esg drivers, respectively [13,49,50]. Inhibition of Wg signaling in enterocytes resulted in ISC overproliferation, as revealed by the aberrantly increased number of pH3+ cells (Fig 5H) and the presence of progenitor cells that were grouped in clusters (Figs 5I–5L and S13A-S13F). By contrast, when the same components were knocked down in progenitor cells, the phenotype was either absent or very weak (S13J–S13L Fig). In addition, ISC overproliferation was observed only when Wg signaling was disrupted during adulthood, but not prior to eclosion (S13G–S13I and S13M–S13O Fig), consistent with our analysis of Wg pathway mutant clones. Together, these findings indicated that the activation of Wg signaling in enterocytes prevents the non-autonomous overproliferation of neighboring ISCs during adult homeostasis.
We further sought to determine the mechanism by which loss of Wg signaling in enterocytes induces the non-autonomous overproliferation of nearby ISCs. We postulated that following Wg pathway inactivation, cytokines or ligands secreted from enterocytes could activate signaling, and thereby the proliferation of neighboring ISCs. This hypothesis is consistent with previous findings, which revealed that ligands from several different signal transduction pathways are released in this manner, including Unpaired (Upd2 and Upd3) from the JAK-STAT pathway, Krn from the EGF pathway and Dpp from the TGF-β pathway [49–54]. Therefore, we compared the levels of transcripts encoding these ligands in control intestines with those in which signaling had been disrupted in enterocytes by RNAi-mediated knockdown of Wg pathway components or by expression of dominant negative Lgs or dominant negative TCF. We observed a marked increase in the expression of upd2 and upd3, but none of the other ligands tested (Figs 6A and S14A).
Based on these results, we sought to determine whether the JAK-STAT pathway was aberrantly activated in ISCs following disruption of Wg signaling in enterocytes. Indeed, strong activation of stat-GFP expression, a JAK-STAT pathway reporter, was observed in clusters of wild-type progenitor cells near pygo null mutant clones (Fig 6C–6C’). In contrast, in cells farther away from mutant clones, or in cells adjacent to wild-type clones, stat-GFP was present at basal levels (Fig 6B–6C’). Consistent with this observation, strong induction of Socs36e, a direct target gene of the JAK-STAT pathway, was also detected following RNAi-mediated knockdown of Wg pathway components or overexpression of dominant negative Lgs (S14B Fig). To determine whether JAK-STAT pathway activation mediates the non-autonomous effects of Wg pathway mutant enterocytes, we used RNAi-mediated knockdown to reduce upd2 and upd3 expression in enterocytes in which Wg signaling was disrupted concomitantly (Fig 6D–6E). Notably, the aberrant increase in pH3+ cells and the abnormal clustering of Deltapos cells were both suppressed (Fig 6D–6E). We conclude that JAK-STAT pathway activation is required for the non-autonomous overproliferation of ISCs that results from inhibition of Wg signaling in enterocytes.
The JAK-STAT pathway is activated in response to various challenges in the midgut, including infection, apoptosis, and stress to promote rapid proliferation [50,55]. We sought to determine whether Wg pathway inhibition in enterocytes induces either a stress response and/or apoptosis, and secondarily results in JAK-STAT pathway activation and a proliferative response in neighboring ISCs. Therefore, we examined the activation of the two major stress-responsive signaling pathways, JNK (c-Jun N-terminal kinase) and Nrf2 (Nuclear factor 2) [56–58], through analysis of their respective target genes: puc and keap1 [59–61]. Importantly, we found that neither of these pathways was activated following Wg signaling disruption (S14B Fig). Furthermore, enterocytes in which Wg signaling was inhibited using mutant clones or RNAi-mediated knockdown did not exhibit any hallmarks of apoptosis, including nuclear fragmentation, detachment from the epithelium, or caspase activation (S14C–S14E’ Fig). These findings provided evidence that the induction of JAK-STAT pathway and ISC overproliferation is more likely a direct consequence of Wg pathway inactivation in these enterocytes. Together, we conclude that the maintenance of intestinal homeostasis requires activation of Wg signaling in enterocytes to prevent the non-autonomous activation of JAK-STAT signaling, and thereby the aberrant overproliferation of neighboring ISCs.
The non-autonomous effect on ISC proliferation described above was observed within intestinal compartments. We also sought to determine the function of high-level Wg pathway activation at compartment boundaries and focused our analysis on the R5-HPZ border, which partitions the posterior terminal midgut (R5) from the anterior hindgut [23,34]. This boundary is distinguished by the juxtaposition of two distinct epithelial cell populations that are derived from distinct origins and differ with respect to cell size, nuclear size and cell adhesion (Fig 7A–7B’) [34,62]. To examine the roles of Wg signaling at the midgut-hindgut boundary, we generated mutant clones of the Wg pathway components arr, dsh, pygo and the functionally redundant fz and Dfz2 in larvae and analyzed adult guts shortly after eclosion. When wild-type clones crossed the midgut-hindgut boundary or were confined within the R5 region, normal cell morphology and cell-cell junctions were observed, and a discrete border between R5 and the HPZ was clearly demarcated, with high levels of Fas3 restricted to the hindgut (Fig 7C–7D”‘). In contrast, two distinct phenotypes were observed in Wg pathway mutant clones near the R5-HPZ boundary: the mutant epithelial cells either formed tightly-packed swirls or displayed markedly larger nuclear and cell size by comparison with their wild-type neighbors.
The first class of mutant clones at the R5-HPZ boundary was comprised of masses of tightly-spaced cells of aberrant nuclear and cell size that were arranged in a spiral pattern, which were readily distinguishable from the surrounding loosely-ordered wild-type epithelium (Figs 7E–7I”‘ and S15B–S15E”‘). This “tightly packed” phenotype occurred with high penetrance (S15A Fig) and was often very severe. Of note, many of the tightly packed mutant clones extended outside the gut (Figs 7F–7G”‘, 7I–7I”‘, S15B and S15E–S15E”‘), and formed a hollow mass with a haze of DAPI staining at its center (Figs 7F and S15B). Importantly, despite their location at the terminal midgut, these mutant clones expressed Fas3 at high levels, a characteristic that is normally restricted to the hindgut cells (Figs 7F–7G”‘ and S15D–S15D”‘). These observations suggested that the “tightly-packed” Wg pathway mutant clones failed to adopt a proper midgut fate, and instead displayed characteristics of the hindgut. To further test this conclusion, we analyzed the mutant clones for cell-specific markers. We found that the vast majority of mutant clones lacked Deltapos ISCs and Prosperopos enteroendocrine cells, and were also negative for Pdm-1, a marker for differentiated enterocytes (Fig 7H–7I”‘) [26,63]. In rare instances, Deltapos or Prosperopos cells were found at the clone periphery. In addition, no Cut was detected in the mutant clones, ruling out the possibility of misadoption of the renalcyte fate (S15E–S15E”‘ Fig)[64,65]. Together, these findings indicated that Wg signaling at the R5-HPZ boundary is critical for proper fate specification of posterior terminal midgut cells.
The second major defect that we observed in Wg pathway mutant clones at the midgut-hindgut boundary were abnormally large cells with large nuclei as compared with their wild-type neighbors, which we termed “large cell” clones (Fig 8A–8A”‘). These large cell clones were found mainly in the midgut, but in some cases, they intercalated within the tightly-spaced rows of anterior hindgut cells in the HPZ (S16A–S16A”‘ Fig). In sharp contrast with the “tightly-packed” clones, the cells within the large cell clones had low Fas3 levels and remained contiguous with the epithelial lining of the gut lumen (S16B–S16B”‘ Fig). Thus, the “large cell” clones and the “tightly-packed” clones represented two distinct classes. Previous studies have shown that ISCs or EEs that underwent excessive replication and cell growth without cell division or differentiation displayed abnormally large nuclear and cell size [66,67]. To determine whether the “large cell” clones caused by inhibition of Wg signaling resulted from a similar mechanism, we stained these clones with midgut markers. Of note, Deltapos ISCs and Prosperopos EEs of normal size and ploidy were present inside these “large cell” clones; however, the cells of abnormally large size were negative for Delta and Prospero (Figs 8B–8B”‘ and S16C–S16C”‘). Further, many of the large cells were also devoid of Pdm-1 staining (Fig 8C–8C”‘), the loss of which correlated with the severity of the phenotype. Together, these findings indicated that the abnormally large cells had not adopted any of the known terminal cell fates of the midgut. Therefore, the “large cell” phenotype resulting from inactivation of Wg signaling was likely due to a defect in cell fate specification during midgut development. We conclude that Wg pathway activation ensures proper cell fate specification at compartment boundaries during the development of the adult intestine (Fig 8D).
The aberrant cell fate specification upon Wg pathway inhibition was not only observed at the midgut-hindgut boundary, but also at the foregut-midgut boundary, a site of high-level Wg pathway activity where the cardia forms [33,35] (Fig 2B and 2B’). Importantly, when wild-type clones crossed the cardia, neither nuclear morphology nor cell-cell junctions were affected (S17A–S17A”‘ Fig). In contrast, several defects were observed in Wg pathway mutant clones located in the cardia. Specifically, normal cell alignment was disrupted, cells of abnormal size were detected, and further, mutant cells extended outside the cardia (S17B–S17D”‘ Fig). Fz3-RFP was lost inside these mutant clones, confirming disruption of Wg pathway activity (S17B–S17B”“Fig). Together, these observations provided evidence that Wg signaling directs proper cell fate specification at two major intestinal boundaries.
Despite the critical processes mediated by Wnt signaling in mammalian intestinal homeostasis, the functions of this pathway in maintaining homeostasis in the adult Drosophila intestine have remained uncertain [25,26,30]. Here, motivated by the recent discovery of graded activation of the Wg pathway at adult intestinal compartment boundaries [18], we examined the sources and function of Wg signaling during homeostasis and development. Our findings demonstrate that Wg pathway activation is essential for the non-autonomous control of ISC proliferation within compartments during homeostasis and also for specifying cell fate near compartment boundaries during the development of the adult gut.
The adult fly gut is subdivided into distinct compartments that have unique function, histological structure, and gene expression repertoire [18–20]. Recently, the graded expression of fz3, a known target gene of the Wg pathway, was found to peak at compartment boundaries [18]. This boundary-enriched activation of the Wg pathway could not be fully explained by the previously identified patterns of wg expression in the intestine [30,34,35]. In this study, we uncovered several novel regions of wg expression both in the gut epithelium and within the surrounding muscle. Together with previous reports, our new findings indicate that Wg is markedly enriched around intestinal compartment boundaries that coincide with prominent sphincters and tissue-organizing centers. An analogous enrichment of wg expression at major compartment boundaries was observed in the larval intestine. How these wg-enriched zones are established and maintained awaits further investigation. The Wnt pathway is known to be one of the key regulators of antero-posterior patterning and regional specificity in the development of vertebrate gastrointestinal tract [68,69], suggesting that the requirement for Wg signaling at intestinal compartment boundaries in flies might be evolutionarily conserved.
Previous studies of Wg signaling in intestinal homeostasis were focused on transduction of the pathway in ISCs [25,26,30,70]. Here, however, starting with Wg signaling reporters that enabled cell-type specific analysis and coupling these findings with functional studies, we have discovered that the primary site of Wg pathway activation during adult homeostasis is in enterocytes and not ISCs. Two factors may account for the discrepancy between our results and previous reports. First, the primary conclusions regarding strong effects on ISC self-renewal in previous studies were based mainly on a dominant negative TCF (dnTCF) [25,26,30], which is a truncated and overexpressed protein [4]. Indeed, the strong negative effects on ISC proliferation resulting from dnTCF expression were not recapitulated by null alleles of the essential components Frizzled or Pygo [25], indicating that dnTCF exhibited some phenotypes that were not present upon complete inactivation of Wg pathway, and thus some dnTCF effects do not represent the physiological roles of Wg signaling in intestinal homeostasis. For this reason, we based our studies on null alleles, and tested our conclusions by analyzing multiple essential Wg pathway components. Second, the initial study investigating the roles of Wg signaling on ISC renewal examined mutant clones 30 days after clone induction [30]. Our clonal analysis was performed at much earlier timepoints (5–7 days ACI). Therefore, the differing results may indicate that Wg signaling has biphasic roles in adult gut homeostasis.
Of note, we found that all cell types in the adult intestinal epithelium have the capability to respond to Wg exposure. Therefore, the responsivity of different intestinal cell types to Wg stimulation may reflect inherent differences in their threshold for pathway activation. We postulate that the threshold for Wg pathway activation is higher in ISCs than enterocytes, and therefore signaling is activated in ISCs only under hyperactivated contexts, as in Apc mutants [26–29], or in response to high levels of wg that are expressed following intestinal injury [25]. Nonetheless, ISCs at the R5-HPZ boundary are an exceptional population in which the Wg pathway is active during homeostasis. The R5-HPZ compartment boundary is a unique region at which three distinct sources of the Wg ligand converge, and exhibits the highest level of Wg pathway activation in the gut (our observations and [18]). Therefore, the level of wg in this region may surpass the threshold required for activation of signaling in ISCs.
We have found that when Wg signaling is inactivated in enterocytes, the JAK-STAT pathway is aberrantly induced in neighboring wild-type ISCs, and drives their non-autonomous proliferation (Fig 8D). In wild-type flies, the JAK-STAT pathway is activated in response to various challenges such as infection, apoptosis, and stress to promote rapid ISC proliferation [50,55]. However, we found that neither of the two major stress response pathways in the fly gut, JNK and Nrf2, nor apoptosis, was induced following Wg pathway inhibition in enterocytes. Therefore, the non-autonomous ISC overproliferation is not a secondary consequence of stress or cell death in neighboring enterocytes. Transcription factors including TCF that are critical for maintenance of gut regionalization in the adulthood can also regulate homeostasis [18]. Therefore, inactivation of Wg signaling during adult homeostasis could potentially disrupt gut regionalization and thus trigger JAK-STAT pathway activation. Alternatively, Wg pathway activation might be required for the tight control of JAK-STAT signaling during homeostasis. As a critical barrier to toxins and infection, intestines are subjected to constant injury and activate JAK-STAT signaling as a compensatory response. We postulate that Wg signaling is required to prevent the inappropriate activation of this critical response during homeostasis and that this “brake” must be shut off or bypassed during regeneration following injury.
Wg pathway activation has known critical roles in cell sorting and patterning at several distinct compartment boundaries in metazoans, including at fly embryonic parasegmental boundaries, the larval wing disc dorsal-ventral boundary, and vertebrate rhombomere boundaries [71–74]. Here, we have found that Wg pathway activation may serve a similar tissue-organizing role at intestinal compartment boundaries, where it is required for proper cell fate specification and lineage separation. We focused on the R5-HPZ boundary, a site of high-level Wg pathway activity, where cells of completely distinct functional and morphological properties are separated [34]. During development of the adult intestine, the R5 compartment is formed by the anterior migration of HPZ cells from the hindgut to the midgut accompanied by their re-specification as midgut cells, and the posterior migration of midgut AMPs [62]. Despite this striking bi-directional movement, the precise coordination of cells in this region ensures proper fate specification and demarcation by a sharp border.
Here, we discovered that disruption of Wg signaling at the R5-HPZ boundary results in two distinct defects: the formation of either “tightly packed” groups of cells or abnormally “large” cells (Fig 8D). Both of these defects affect proper cell fate specification and boundary maintenance, but are disparate in nature. First, the tightly packed cells segregate away from their wild-type neighbors, and cluster together to form a hollow spherical mass that can grow outside the midgut. Intriguingly, despite their location in the midgut, these “tightly packed” cells lack cell-type specific midgut markers, and express Fas3 at high levels that are normally found in the hindgut cells. Based on these observations, we speculate that these “tightly packed” clones are hindgut cells that fail to adopt midgut cell fate following migration. Unlike the “tightly packed” cells, the abnormally “large” cells that also result from inactivation of Wg signaling are likely midgut-derived, as they express Fas3 at low levels and are contiguous with midgut epithelium. Previous reports revealed that defects in mitosis result in dysfunctional ISCs that undergo replication without division [66,67], which in principle could have been the cause of the aberrantly large cells we observed. Similarly, mitotic deregulation in EEs also results in an abnormal increase in ploidy [67]. However, the “large cells” that resulted from Wg pathway inhibition are devoid of normal ISC and EE markers. Further, almost all of the abnormally large cells do not express markers characteristic of differentiated enteroctyes. Importantly, these “large cell” clones are multi-cellular, and include ISCs and EEs of normal ploidy, size, and cell-type specific markers. These characteristics distinguish the phenotype of the Wg pathway mutant clones from the previously reported large cells that result from defective mitosis [66,67] and suggest that these large cells, like the “tightly-packed” cells, likely arise from deregulated cell fate specification and have likely adopted an intermediate or novel cell fate.
Crosses were performed at 25°C except those under control of gal4/gal80ts driver, which were set up at room temperature. To test the effects specifically during adulthood, progeny were collected within 2 days after eclosion and shifted to 29°C for another 7–10 days. To determine the potential function during development, 2nd instar larvae were switched to 29°C and dissected right after eclosion. For the rpr-induced cell death experiments, flies overexpressing rpr, as well as control flies, were shifted to 29°C 3 days after eclosion for 40 hours before examination.
Primary antibodies used for immunostaining were mouse anti-β-gal (Promega) 1:500, rabbit anti-β-gal 1:5000 (MP Biomedicals), mouse anti-Arm 1:20 ([83]; Developmental Studies Hybridoma Bank, DSHB), mouse anti-delta 1:100 ([78]; DSHB), mouse anti-Prospero 1:20 ([84]; DSHB), mouse anti-Cut 1:20 ([85]; DSHB), mouse anti-α-Spectrin 1:20 ([86]; DSHB), mouse anti-Fas3 1:20 ([87]; DSHB), rabbit anti-Pdm-1 1:200 [63], rabbit anti phosphor-histone H3 (Ser10) 1:1000 (Millipore), rabbit anti-DsRed 1:500 (Clontech), chicken anti-GFP 1:10000 (Abcam), rabbit anti-GFP 1:500 (Life Technologies), Alexa Fluor 555 phalloidin 1:500 (Life Technologies), Alexa Fluor 488 phalloidin 1:500 (Life Technologies), rabbit anti-cleaved Drosophila Dcp-1 (Asp216) 1:100 (Cell Signaling) and DAPI 1:100 (Sigma). Secondary antibodies used were goat or donkey Alexa Fluor 488 or 555 conjugates at 1:400 (Life Technologies), and goat or donkey Cy5 conjugates at 1:200 (Life Technologies/Jackson Immunochemicals). Confocal images were captured on a Nikon A1RSi confocal microscope and processed with Adobe Photoshop software.
Adult fly guts were dissected in PBS, fixed in 4% paraformaldehyde for 45 mins to 1hr and washed with PBS+0.1% Triton X-100. Specifically, adult wg-lacZ guts were fixed in sodium cacodylate buffer [48] for 25 mins, and for Delta antibody staining, guts of desired genotypes were fixed in sodium cacodylate buffer for 20 mins. After blocking with PBS+0.1% Tween-20+10% BSA for 1h at room temperature, the samples were incubated with primary antibody (diluted in PBS+0.5% Triton X-100) at 4°C overnight. Secondary antibody incubation was carried out at room temperature for 2 hrs. The samples were subsequently stained with DAPI (2 μg/ml) and mounted in Prolong Gold Antifade Reagent (Life Technologies). Larval guts were immunostained in the same way except that the wandering 3rd instar larvae were dissected and fixed for only 20 mins. More than 20 guts of desired genotypes were examined unless specified and the representative image is shown. The non-autonomous ISC overproliferation defects were observed in both anterior and posterior midguts, and the specific subregions shown in the figures are indicated in the figure legends.
Quantification of fz3-RFP and nkd-lacZ expression level in posterior midgut was performed by NIS-Elements software. Stat-GFP intensity was analyzed via Imaris software (Bitplane). For ISC quantification, flies were stained with anti-Delta and anti-Prospero antibodies. 60x images of posterior midgut region (R4-R5) were obtained and the total number of Delta positive cells in the field was counted. T-tests were performed using Prism (GraphPad software, USA).
MARCM clones were generated as described [40,41]. To induce clones during development, 1st and 2nd instar larvae were heat shocked in a 37°C water bath for 2–3 hrs, except for crosses driven by one MARCM 82B driver [79], which were induced during larval-pupal transition. For specific MARCM lines, the heat-shock was repeated the next day. To generate MARCM clones during adulthood, progeny of the desired genotype were collected within 2 days after eclosion and were heat-shocked in a 37°C water bath. The length of the heat shock varied with the MARCM line, from one 30-min heat shock to four 90-min heat shocks over 2 days. The adult clones were examined 7–10 days later. For larval clones, 1st and 2nd instar larvae were heat-shocked for 2–3 hrs and guts of wandering 3rd instar larvae were examined.
15–20 fly guts of proper genotype and age were dissected and homogenized in Trizol (Invitrogen). Total RNA was extracted according to the protocol from the Drosophila Genomics Resource Center or using the RNA miniprep kit (Zymo research). The RNA was subsequently treated with RQ1 DNase (Promega). 1 μg of RNA was reverse transcribed using p(dT)15 primers (Roche) and M-MLV reverse transcriptase (Invitrogen). Expression level of candidate ligands was quantified using the StepOne Real-time PCR system (Life Technologies) with SYBR green (Life Technologies/Biorad) and the results were presented as mean fold change with standard deviation.
The primers for rpl32 (internal control), upd3 and Socs36e were adapted from [28], whereas primers for krn were adapted from [88]. The other primer sequences were: dpp F: 5’- TCT GCT GAC CAA GTC GG -3’, dpp R: 5’- GCG GGA ATG CTC TTC AC -3’, upd2 F: 5’- TGG TAT TCG CTC ATC GTG A -3’, upd2 R: 5’- GGC AAA TCA GAG ATC CCG -3’. puc F: 5’- CAC ATC AGA ACA TCA AGC AGT AC –3’, puc R: 5’-GTA GGC GAT GGC AAT GG -3’ and Keap1 F: 5’-TAC AAG AGT CCA GCG ATC CA –3’ and keap1 R: 5’-GTC ACC GAA ACA TGG CGT-3’.
|
10.1371/journal.pbio.0050310 | Population Genomics: Whole-Genome Analysis of Polymorphism and Divergence in Drosophila simulans | The population genetic perspective is that the processes shaping genomic variation can be revealed only through simultaneous investigation of sequence polymorphism and divergence within and between closely related species. Here we present a population genetic analysis of Drosophila simulans based on whole-genome shotgun sequencing of multiple inbred lines and comparison of the resulting data to genome assemblies of the closely related species, D. melanogaster and D. yakuba. We discovered previously unknown, large-scale fluctuations of polymorphism and divergence along chromosome arms, and significantly less polymorphism and faster divergence on the X chromosome. We generated a comprehensive list of functional elements in the D. simulans genome influenced by adaptive evolution. Finally, we characterized genomic patterns of base composition for coding and noncoding sequence. These results suggest several new hypotheses regarding the genetic and biological mechanisms controlling polymorphism and divergence across the Drosophila genome, and provide a rich resource for the investigation of adaptive evolution and functional variation in D. simulans.
| Population genomics, the study of genome-wide patterns of sequence variation within and between closely related species, can provide a comprehensive view of the relative importance of mutation, recombination, natural selection, and genetic drift in evolution. It can also provide fundamental insights into the biological attributes of organisms that are specifically shaped by adaptive evolution. One approach for generating population genomic datasets is to align DNA sequences from whole-genome shotgun projects to a standard reference sequence. We used this approach to carry out whole-genome analysis of polymorphism and divergence in Drosophila simulans, a close relative of the model system, D. melanogaster. We find that polymorphism and divergence fluctuate on a large scale across the genome and that these fluctuations are probably explained by natural selection rather than by variation in mutation rates. Our analysis suggests that adaptive protein evolution is common and is often related to biological processes that may be associated with gene expression, chromosome biology, and reproduction. The approaches presented here will have broad applicability to future analysis of population genomic variation in other systems, including humans.
| Given the long history of Drosophila as a central model system in evolutionary genetics beginning with the origins of empirical population genetics in the 1930s, it is unsurprising that Drosophila data have inspired the development of methods to test population genetic theories using DNA variation within and between closely related species [1–4]. These methods rest on the supposition of the neutral theory of molecular evolution that polymorphism and divergence are manifestations of mutation and genetic drift of neutral variants at different time scales [5]. Under neutrality, polymorphism is a “snapshot” of variation, some of which ultimately contributes to species divergence as a result of fixation by genetic drift. Natural selection, however, may cause functionally important variants to rapidly increase or decrease in frequency, resulting in patterns of polymorphism and divergence that deviate from neutral expectations [1,2,6]. A powerful aspect of inferring evolutionary mechanism in this population genetic context is that selection on sequence variants with miniscule fitness effects, which would be difficult or impossible to study in nature or in the laboratory but are evolutionarily important, may cause detectable deviations from neutral predictions. Another notable aspect of these population genetic approaches is that they facilitate inferences about recent selection—which may be manifest as reduced polymorphism or elevated linkage disequilibrium—or about selection that has occurred in the distant past—which may be manifest as unexpectedly high levels of divergence. The application of these conceptual advances to the study of variation in closely related species has resulted in several fundamental advances in our understanding of the relative contributions of mutation, genetic drift, recombination, and natural selection to sequence variation. However, it is also clear that our genomic understanding of population genetics has been hobbled by fragmentary and nonrandom population genetic sampling of genomes. Thus, the full value of genome annotation has not yet been applied to the study of population genetic mechanisms.
Combining whole-genome studies of genetic variation within and between closely related species (i.e., population genomics) with high-quality genome annotation offers several major advantages. For example, we have known for more than a decade that regions of the genome experiencing reduced crossing over in Drosophila tend to show reduced levels of polymorphism yet normal levels of divergence between species [7–10]. This pattern can only result from natural selection reducing levels of polymorphism at linked neutral sites, because it violates the neutral theory prediction of a strong positive correlation between polymorphism and divergence [5]. However, we have no general genomic description of the physical scale of variation in polymorphism and divergence in Drosophila and how such variation might be related to variation in mutation rates, recombination rates, gene density, natural selection, or other factors. Similarly, although several Drosophila genes have been targets of molecular population genetic analysis, in many cases, these genes were not randomly chosen but were targeted because of their putative association with phenotypes thought to have a history of adaptive evolution [11,12]. Such biased data make it difficult to estimate the proportion of proteins diverging under adaptive evolution. In a similar vein, the unique power of molecular population genetic analysis, when used in concert with genome annotation, could fundamentally alter our notions about phenotypic divergence due to natural selection. This is because our current understanding of phenotypic divergence and its causes is based on a small and necessarily highly biased description of phenotypic variation. Alternatively, a comprehensive genomic investigation of adaptive divergence could use genome annotations to reveal large numbers of new biological processes previously unsuspected of having diverged under selection. Here we present a population genomic analysis of D. simulans. D. simulans and D. melanogaster are closely related and split from the outgroup species, D. yakuba, several million years ago [13–15]. The vast majority of D. simulans and D. yakuba euchromatic DNA is readily aligned to D. melanogaster, which permits direct use of D. melanogaster annotation for investigation of polymorphism and divergence and allows reliable inference of D. simulans–D. melanogaster ancestral states over much of the genome. Our analysis uses a draft version of a D. yakuba genome assembly (aligned to the D. melanogaster reference sequence) and a set of light-coverage, whole-genome shotgun data from multiple inbred lines of D. simulans, which were syntenically aligned to the D. melanogaster reference sequence.
Seven lines of D. simulans and one line of D. yakuba were sequenced at the Washington University Genome Sequencing Center (the white paper can be found at http://www.genome.gov/11008080). The D. simulans lines were selected to capture variation in populations from putatively ancestral geographic regions [16], recent cosmopolitan populations, and strains encompassing the three highly diverged mitochondrial haplotypes previously described for the species [17]. These strains have been deposited at the Tucson Drosophila Stock Center (http://stockcenter.arl.arizona.edu). A total of 2,424,141 D. simulans traces and 2,245,197 D. yakuba traces from this project have been deposited in the National Center for Biotechnology Information (NCBI) trace archive. D. simulans syntenic assemblies were created by aligning trimmed, uniquely mapped sequence traces from each D. simulans strain to the euchromatic D. melanogaster reference sequence (v4). Two strains from the same population, sim4 and sim6, were unintentionally mixed prior to library construction; reads from these strains were combined to generate a single, deeper, syntenic assembly (see Materials and Methods), which is referred to as SIM4/6. The other strains investigated are referred to as C167.4, MD106TS, MD199S, NC48S, and w501. Thus, six (rather than seven) D. simulans syntenic assemblies are the objects of analysis. Details on the fly strains and procedures used to create these assemblies, including the use of sequence quality scores, can be found in Materials and Methods. The coverages (in Mbp) for C167.4, MD106TS, MD199S, NC48S, SIM4/6, and w501, are 56.9, 56.3, 63.4, 42.6, 89.8, and 84.8, respectively. A D. yakuba strain Tai18E2 whole-genome shotgun assembly (v2.0; http://genome.wustl.edu/) generated by the Parallel Contig Assembly Program (PCAP) [18] was aligned to the D. melanogaster reference sequence (Materials and Methods). The main use of the D. yakuba assembly was to infer states of the D. simulans–D. melanogaster ancestor. For many analyses, we used divergence estimates for the D. simulans lineage or the D. melanogaster lineage (from the inferred D. simulans–D. melanogaster ancestor) rather than the pairwise (i.e., unpolarized) divergence between these species. These lineage-specific estimates are often referred to as “D. simulans divergence,” “D. melanogaster divergence,” or “polarized divergence.”
A total of 393,951,345 D. simulans base pairs and 102,574,197 D. yakuba base pairs were syntenically aligned to the D. melanogaster reference sequence. Several tens of kilobases of repeat-rich sequences near the telomeres and centromeres of each chromosome arm were excluded from our analyses (Materials and Methods). D. simulans genes were conservatively filtered for analysis based on conserved physical organization and reading frame with respect to the D. melanogaster reference sequence gene models (Materials and Methods). We took this conservative approach so as to retain only the highest quality D. simulans data for most inferences. The number of D. simulans genes remaining after filtering was 11,466. Ninety-eight percent of coding sequence (CDS) nucleotides from this gene set are covered by at least one D. simulans allele. The average number of lines sequenced per aligned D. simulans base was 3.90. For several analyses in which heterozygosity and divergence per site were estimated, we further filtered the data so as to retain only genes or functional elements (e.g., untranslated regions [UTRs]) for which the total number of bases sequenced across all lines exceeded an arbitrary threshold (see Materials and Methods). The numbers of genes for which we estimated coding region expected heterozygosity, unpolarized divergence, and polarized divergence were 11,403, 11,439, and 10,150, respectively. Coverage on the X chromosome was slightly lower than autosomal coverage, which is consistent with less X chromosome DNA than autosomal DNA in mixed-sex DNA preps. Variable coverage required analysis of individual coverage classes (n = 1–6) for a given region or feature, followed by estimation and inference weighted by coverage (Materials and Methods). The D. simulans syntenic alignments are available at http://www.dpgp.org/. An alternative D. simulans “mosaic” assembly, which is available at http://www.genome.wustl.edu/, was created independently of the D. melanogaster reference sequence.
One of the main goals of large-scale investigations of sequence divergence is to characterize the many biological factors influencing variation in substitution rates throughout the genome. Most analyses of Drosophila data focus on variation in functional constraints or directional selection as the main cause of heterogeneity in substitution rates across genes or functional elements [20,21]. However, the available data have been too sparse to detect any patterns of increasing or decreasing divergence along chromosome arms.
Centromere proximal regions tend to be more divergent than distal regions (Figure 1, Figure S4, and Table S5). This pattern is more consistent for D. simulans than for D. melanogaster. Proximal euchromatic regions tend to have lower inferred ancestral GC content compared with distal regions of chromosome arms (Figure S4 and Table S5), which is consistent with the observation that D. simulans divergence was negatively correlated with inferred ancestral GC content (Materials and Methods) (50-kb windows, Spearman's ρ = −0.23, p = 1.4 × 10−26) [30]. The correlation between ancestral GC content and divergence was much weaker and only marginally significant for D. melanogaster (Spearman's ρ = −0.05, p = 0.03). However, while chromosomal gradients of divergence were observed for most chromosome arms (Figure S4 and Table S5), inferred ancestral GC content tends to show a less-consistent pattern. For example, some arms showed a more U-shaped distribution, with euchromatic regions near centromeres and telomeres tending to have higher estimated ancestral GC content (Figure S5). More proximal and distal regions also tend to have reduced crossing-over [39], which is consistent with the observation that inferred ancestral GC content is negatively correlated with cM/kb (Materials and Methods) on the X chromosome (Spearman's ρ = −0.33, p = 0.0002) [59], the only chromosome arm for which we investigated correlates of recombination rate variation (see below).
The neutral model of evolution predicts that gradients of divergence along chromosome arms are explained by gradients of functional constraint or mutation rates. For example, higher divergence in regions near centromeres could be explained if such regions harbor a lower density of functional elements (e.g., genes). However, with the exception of chromosome arm 2L (Spearman's ρ = −0.19, p = 6 × 10−5), variation in coding sequence density (CDS bases per 50-kb window) showed no significant chromosomal proximal–distal trend, suggesting that variation in constraint that is associated with coding density plays, at best, a small part in explaining chromosomal gradients of divergence. More generally, the expectation of a negative correlation between coding density and nucleotide divergence in D. simulans was not met. This seemingly counterintuitive result probably reflects the fact that exons constitute a relatively small fraction of the genome and were not dramatically less diverged (0.016) compared with intergenic DNA (0.027).
If proximal–distal gradients of decreasing divergence along chromosome arms result from variation in mutation rates, then the neutral theory predicts that we should observe similar gradients of polymorphism. This is the case for some chromosome arms but not others (Figure 1 and Table S5), after regions of reduced πnt in the most distal/proximal regions are excluded (Materials and Methods; this result is robust to variation in the extent of proximal and distal chromosomal regions removed from the analysis). Thus, variable neutral mutation rates alone is an insufficient explanation for the overall genomic patterns of variation. Below we address the possibility that recombination rate variation contributes to variation in D. simulans πnt and divergence across chromosome arms.
There was considerable variance of polymorphism and divergence across chromosome arms, even when regions of severely reduced heterozygosity near centromeres and telomeres were excluded. Figure 1 clearly shows that variance in polymorphism and divergence is not randomly arranged, but rather appears to be spatially structured on the scale of several tens of kilobases. These qualitative visual assessments were supported by significant statistical autocorrelations (Materials and Methods) for nucleotide heterozygosity and divergence across all chromosome arms (Table S6) [60]. Furthermore, the strength of this autocorrelation appeared to differ across arms, because X and 3L show evidence of stronger correlations over longer distances (Figure 1). The strength of autocorrelation is consistently higher for heterozygosity than for divergence.
Under the neutral theory, fluctuations in polymorphism and divergence could be the result of variation in gene density, with windows that have more exons per kb showing lower polymorphism and divergence. This expectation was not met. Indeed, for 50-kb autosome windows (but not X-linked windows), divergence is positively correlated with coding density (Spearman's ρ = 0.12, p < 0.0001). This is consistent with an important role of directional selection on coding sequence to genome divergence, a point we will revisit in several analyses below. In contrast to the positive correlation between coding density and divergence, we found a negative correlation between coding density and D. simulans πnt (autosome Spearman's ρ = −0.10, p < 0.0001; X Spearman's ρ = 0.29, p < 0.0001). Overall, the contrasting correlations between coding density and polymorphism versus divergence suggest that directional selection in exon-rich regions generates greater divergence and reduced polymorphism due to hitchhiking effects [3,6,61].
The analyses presented above, especially for the X chromosome data, strongly suggest that hitchhiking effects contribute to shaping patterns of polymorphism across the D. simulans genome. To provide a more quantitative assessment of the physical extent, magnitude, and biological basis of these hitchhiking effects, we carried out a genomic analysis of polymorphism and divergence in the context of the Hudson-Kreitman-Aguade (HKA) test [2] (Materials and Methods). The analysis should be thought of as a method for identifying unusual genomic regions rather than as a formal test of a specific model, since our data violate the assumptions of the simple neutral model (neutral alleles sampled from a single, equilibrium, panmictic population). The results (Figure 1, Datasets S6, S16–S20) statistically support our earlier contention and previous reports [7,8,10,34,36], that Drosophila chromosomes show greatly decreased polymorphism, relative to divergence, in both telomere- and centromere-proximal regions. The fact that corrected X chromosome heterozygosity was not significantly different from autosomal heterozygosity, although X chromosome divergence was significantly higher than autosomal divergence, supports a role for hitchhiking effects reducing nucleotide variation on the X chromosome.
Our previously mentioned result, that coding density is positively correlated with divergence and negatively correlated with polymorphism, suggested that hitchhiking effects of directional selection are more common in exonic sequence. The HKA-like analysis supports this contention. We identified regions of the genome that had either two or more consecutive, nonoverlapping 10-kb windows with p < 1 × 10−6 or four such windows with p < 0.01. The number of coding nucleotides per 10 kb in these “hitchhiking windows” (n = 378 windows, mean coding density = 2,980 bp) was much higher than coding density in other windows (n = 9,329, mean coding density = 1,860 bp) (Mann-Whitney U, p < 0.0001).
An alternative hypothesis for the strong correlation between recombination and polymorphism and the high density of coding sequence in regions showing reduced heterozygosity-to-divergence ratios is background selection, a phenomenon whereby the removal of deleterious mutations reduces polymorphism at linked sites [1]. To address this possibility, we calculated Fay and Wu's H [56] for 10-kb windows across the genome using only sites with a coverage of five alleles and windows not located in extended regions of reduced heterozygosity near the distal and proximal ends of chromosome arms (Materials and Methods). Hitchhiking effects of beneficial mutations are expected to cause an excess of high-frequency derived alleles (and a more-negative H statistic) relative to neutral theory predictions, while background selection predicts no such excess [1,72]. We compared the average H statistic for regions of the genome showing four or more consecutive 10-kb windows with an HKA-like test of p < 0.01 versus 10-kb windows from the rest of the genome. For each chromosome arm, the H statistic was significantly more negative in windows showing a reduced heterozyogsity-to-divergence ratio (Mann Whitney U, p < 10−4 for each arm), which strongly supports the proposition that hitchhiking effects of beneficial variants is a major cause of the fluctuations in heterozygosity across the genome. Note, however, that this analysis does not rule out a contribution of background selection [1].
Several factors can generate lineage differences in divergence. For example, higher divergence in a lineage (relative to the lineage of its sister species) could be due to higher mutation rates, shorter generation times, or stronger directional selection. Investigating which classes of mutations or functional elements tend to show different levels of divergence in two lineages can inform our understanding of the causes of rate variation.
Previously collected data from coding regions suggest that D. melanogaster evolves faster than D. simulans [89,90]. We found a similar pattern in that dN and dS are greater in D. melanogaster (median = 0.0045 and 0.0688) than in D. simulans (median = 0.0036 and 0.0507) (Table 1 and S3). This pattern has been interpreted as reflecting the reduced efficacy of selection against slightly deleterious variants in D. melanogaster, supposedly resulting from its smaller effective population size relative to D. simulans [89]. However, a different pattern is observed on a genome-wide scale, as median D. simulans divergence (50-kb windows; 0.025), though only slightly greater than D. melanogaster (50-kb windows; 0.022), is consistently greater across a large proportion of windows (Wilcoxon sign rank test, p = 1.8 × 10−275). We consider the genomic faster D. simulans finding as provisional due the potential biases associated with D. melanogaster-centric alignments. For example, genomic regions that are evolving quickly only in D. melanogaster may drop out of the D. melanogaster–D. yakuba alignment, whereas regions evolving quickly only in D. simulans may be retained because of the relatively short D. melanogaster–D. simulans branch. Analysis of rate variation across site types (Table 1 and Table S3) reveals a more complex pattern. For example, D. simulans shows greater divergence than D. melanogaster for intergenic, intron, and 3′ UTR sites, whereas D. melanogaster shows greater divergence than D. simulans for 5′ UTRs, nonsynonymous sites, and synonymous sites.
A decades-old issue in population genetics is the extent to which directional selection determines protein divergence. Several analytic strategies for investigating the prevalence of adaptive protein divergence between closely related species have been proposed (reviewed in [91]). Here we focused on two approaches. First, we used comparisons of synonymous and nonsynonymous polymorphic and fixed variants in individual genes to test the neutral model. Second, we identified proteins that show very different divergence estimates in D. simulans versus D. melanogaster.
The same logic originally proposed in the MK test using nonsynonymous and synonymous variation can be extended to any setting in which variant types can be categorized, a priori. We tested variation in individual noncoding elements (introns, UTRs, and intergenic sequences) relative to variation at tightly linked synonymous sites (Materials and Methods) using the same criteria described for the MK tests; we present only polarized analyses (Datasets S2–S5). The proportion of tests (Materials and Methods) that rejected (p < 0.05) the null model for 5′ UTR, 3′ UTR, intron, and intergenic sites are 0.13, 0.13, 0.12, and 0.17, respectively. However, unlike the case for the nonsynonymous versus synonymous polarized MK tests, of which only 6% of the significant tests deviated in the direction of excess polymorphism (relative to synonymous sites), a much greater proportion of noncoding MK tests deviated in this direction—0.13, 0.24, 0.28, and 0.28 for 5′ UTR, 3′ UTR, intron, and intergenic regions, respectively. Thus, the proportion of noncoding elements showing evidence of adaptive evolution for 5′ UTR, 3′ UTR, intron, and intergenic sites is 0.12, 0.10, 0.08, and 0.12, respectively, which is similar to the proportion of coding sequences inferred (by polarized MK tests) to be under direction selection (0.14). It would be tempting to conclude from this result that intergenic variants are as likely to be under directional selection as nonsynonymous variants. However, such an interpretation ignores the fact that the number of variants per element for each MK test is much greater for intergenic sequence (median = 87) compared to the numbers for coding regions (median = 42), 5′ UTRs (median = 34), 3′ UTRs (median = 35), or introns (median = 64). Thus, there is more power to reject the neutral model for intergenic sequence and introns than for exonic sequence. The fact that MK p-values are significantly negatively correlated with the total number of observations per test is consistent with this explanation. There was no evidence of different proportions of significant versus nonsignificant tests for X-linked versus autosomal elements.
Tables S22–S24 report data from the ten most highly significant MK tests (average coverage > 2) indicative of directional selection on 5′ UTRs, 3′ UTRs, and intron sequences, respectively. Among the most unusual 5′UTRs are those associated with genes coding for proteins associated with the cytoskeleton or the chromosome, categories that also appeared as unusual in the MK tests on protein variation. Two of the top-ten 3′ UTRs are associated with the SAGA complex, a multi-subunit transcription factor involved in recruitment of RNA Pol II to the chromosome [111]. Among the extreme introns, two are from genes coding for components of the ABC transporter complex and two are from genes coding for centrosomal proteins, again pointing to the unusual evolution of genes associated with the cytoskeleton and chromosome structure and movement. As previously noted, a large number of significant UTRs deviate in the direction of excess polymorphism (relative to synonymous mutations). Given the potential importance of the UTRs in regulating transcript abundance and localization, translational control, and as targets of regulatory microRNAs [112], such UTRs could be attractive candidates for functional investigation. Contingency tests of significant versus nonsignificant MK test for amino acids versus each of the noncoding elements yielded p-values of 0.65, 0.04, and 0.07 for 5′ UTRs, 3′ UTRs, and introns, respectively. Thus, there is weak evidence that genes under directional selection on amino acid sequences tend to have 3′ UTRs and introns influenced by directional selection as well.
Up to this point, our analyses have investigated various attributes of polymorphism and divergence based on windows or genes. An alternative approach for understanding the causes of variation and divergence is to analyze polymorphism and divergence across site types. Table 2 shows whole-genome counts of polymorphic and polarized fixed variants for UTRs, synonymous sites, nonsynonymous sites, introns, and intergenic sites. We also provide data for polarized, synonymous preferred or unpreferred variants. Almost all preferred versus unpreferred codons in D. melanogaster end in GC versus AT, respectively [113]; thus, preferred versus unpreferred codons can be thought of as GC-ending versus AT-ending codons.
Nonsynonymous sites showed the smallest ratio of polymorphic-to-fixed variants, which is consistent with the MK tests and supports the idea that such sites are the most likely to be under directional selection. Nonsynonymous polymorphisms also occur at slightly lower frequency than do noncoding variants (Table S25). Synonymous sites have the highest ratio of polymorphic-to-fixed variants, which supports the previously documented elevated ratio of polymorphic-to-fixed unpreferred synonymous variants in D. simulans [89]. The confidence intervals of the ratio of polymorphic-to-fixed variants among site types are nonoverlapping with the exception of intron and intergenic sites. If preferred synonymous mutations are, on average, beneficial [89,114], then the smaller polymorphic-to-fixed ratio for nonsynonymous and UTR variants versus preferred variants implies that a large proportion of new nonsynonymous and UTR mutations are beneficial. Using similar reasoning, the data in Table 2 suggest that directional selection plays a larger role in nonsynonymous and UTR divergence compared to intron and intergenic divergence [20,115,116]. These conclusions are consistent with estimates of α [11,117], the proportion of sites fixing under directional selection (assuming that synonymous sites are neutral and at equilibrium) for different site types.
Determining the relative contributions of various mutational and population genetic processes to base composition variation and inferring the biological basis of selection on base composition remain difficult problems. Much of the previously published data on base composition variation in D. simulans have been from synonymous sites [55,89,90,118]. Several lines of evidence [55,89,90,113,118] suggest that on average, preferred codons have higher fitness than unpreferred codons, with variation in codon usage being maintained by AT-biased mutation, weak selection against unpreferred codons, and genetic drift [23,114]. However, the possibilities of nonequilibrium mutational processes and/or natural selection favoring different base composition in different lineages have also been addressed [119]. The D. simulans population genomic data allow for a thorough investigation of the population genetics and evolution of base composition for both coding and noncoding DNA [59,120]. The analyses discussed below use parsimony to polarize polymorphic and fixed variants. Complete genomic and gene-based data are available as Datasets S9 and S10.
The genomic analysis of polymorphism and divergence based on alignments to a reference sequence is poised to become a central component of biological research. Here we have demonstrated that such analysis can be based on high-quality whole-genome syntenic assemblies from light shotgun sequence data; accounting for variable coverage and data quality is fundamentally important. Several, noteworthy new results have been reported here. First, our genome-level investigation of adaptive protein evolution has revealed a large number of proteins and biological processes that have experienced directional selection, setting the stage for a general analysis of functional protein divergence under selection in Drosophila. Second, we identified several UTRs, introns, and intergenic sequences showing the signature of adaptive evolution. The functional biology of such noncoding elements and their connections to adaptive protein and gene expression evolution is open to investigation. Third, D. simulans populations exhibit large-scale chromosomal patterning of polymorphism and divergence that is poorly explained by current genome annotations. Variation in recombination rates across chromosomes may contribute to these patterns. Fourth, the population genetics of the X chromosome differs in several ways from that of the autosomes. It evolves faster, harbors less polymorphism, and shows a different spatial scale of variation of polymorphism and divergence compared to the autosomes. Finally, base composition is evolving in both coding and noncoding sequences, for reasons that are as of yet unclear. This project is, in many ways, a first step toward population genomics in general, and in the Drosophila model specifically. For example, the average number of alleles sampled per base is too small for investigating many interesting properties of variation. Some genomic regions have been excluded due to low coverage, their repetitive nature, or very high divergence from D. melanogaster. Many aspects of biological annotation have not been investigated here, and many new Drosophila annotations will be produced in the near future as comparative and functional annotations of the D. melanogaster genome move forward. Nevertheless, the data are a stunningly rich source of material for functional and population genetic investigation of D. simulans polymorphism and divergence. It will be interesting to compare the processes determining polymorphism and divergence in D. simulans to those controlling variation in D. melanogaster (http://www.dpgp.org) and in other species, such as humans. Such comparisons are likely to result in new insights into the genetic, biological, and population genetic factors responsible for similarities and differences among species in the genomic distribution of sequence variation.
D. simulans 4 (males and females). This strain was established by ten generations of sibling mating from a single, inseminated female collected by D. Begun in the Wolfskill orchard, Winters, California, USA, summer 1995.
D. simulans 6 (males and females). This strain was established by ten generations of sibling mating from a single, inseminated female collected by D. Begun in the Wolfskill orchard, Winters, California, summer 1995.
D. simulans w501(males and females). This strain carries a white (eye color) mutation. It has been in culture since the mid 20th century, likely descended from a female collected in North America. The strain used for sequencing was sib-mated for nine generations by D. Barbash at UC Davis. Libraries for sequencing were prepared from DNA isolated from embryos.
D. simulans MD106TS (males and females). This strain was descended from a single, inseminated female collected by J. W. O. Ballard in Ansirabe, Madagascar on 19 March 1998. It has the siII mitochondrial genotype, and was cured of Wolbachia by tetracycline. The strain was sib-mated for five generations in the Ballard lab, followed by an additional five generations of sib-mating by D. Begun.
D. simulans MD199S (females). This strain was descended from a single, inseminated female collected by J. W. O. Ballard in Joffreville, Madagascar on 28 March 1998. It has the siIII mitochondrial genotype, and probably lost Wolbachia infection. The strain was sib-mated for five generations in the Ballard lab, followed by an additional five generations of sib-mating by D. Begun. All-female DNA was made to assist in assembly of the Y chromosome by comparison to mixed-sex libraries of other lines.
D. simulans NC48S (males and females). This strain was descended from a collection by F. Baba-Aissa in Noumea, New Caledonia in 1991. It has the siI mitochondrial genotype, and was sib-mated for five generations in the Ballard lab, followed by an additional five generations of sib-mating by D. Begun.
D. simulans C167.4 (males and females). This strain was descended from a collection in Nanyuki, Kenya. It is unusual in that can produce fertile females when hybridized to D. melanogaster. The line used for genome project was obtained from the Ashburner laboratory via D. Barbash, and was subjected to a total of 13 generations of sib- mating.
D. yakuba Tai18E2 (males and females). This strain derives from a single inseminated female captured in 1983 by D. Lachaise in the Taï rainforest, on the border of Liberia and Ivory Coast. This line was sib-mated for ten generations by A. Llopart and J. Coyne. Inspection of 21 salivary gland polytene chromosomes showed no chromosomal rearrangements segregating within the strain. Therefore, Tai18E2 appears homokaryotypic for the standard arrangement in all chromosome arms, save 2R, which is homokaryotypic for 2Rn.
DNA preparations for sequencing all lines except w501 and Tai18E2 were made from adults. Drosophila nuclei were isolated following Bingham et al. [121]. For all lines except w501 and Tai18E2, DNA was isolated by phenol/chloroform extraction of nuclei followed by ethanol precipitation. For lines w501 and Tai18E2, embryos were collected using standard procedures [122] followed by DNA isolation on CsCl gradients [121].
Sequence data were obtained from paired-end plasmids and fosmids (Table S32) using standard Washington University Genome Sequencing Center laboratory protocols (http://genome.wustl.edu). A highly automated production pipeline using a 384-well format ensured the integrity of the paired-end data.
We determined the nucleotide-level accuracy by reviewing the quality values of the D. yakuba consensus assembly and by comparison to manually edited D. yakuba sequence. More than 97% of the D. yakuba genome sequence had quality scores of at least 40 (Q40), corresponding to an error rate of ≤10−4. Further, we extracted reads from two local fosmid-sized regions (68 kb, defined by fosmid-end sequence pairs, one on chromosome X and one on chromosome 3L) of the assembly and reassembled using Phrap (P. Green, unpublished data). The resulting “fosmids” were manually reviewed and edited. Comparison of the sequence to these manually edited regions revealed a high-quality discrepancy rate of 2 × 10−4 substitutions and 1 × 10−4 insertion/deletion errors, consistent with the above estimates based on consensus base quality. We also found no evidence of misassembly when comparing the WGS assembly to these projects.
Repetitive content was estimated both in D. yakuba and D. melanogaster using RECON to generate the repeat families and RepeatMasker to then identify those repeats in the genomes. The D. yakuba genome was ∼27% repetitive overall (of which ∼2.5% is simple sequence repeats/low complexity sequence) and 8% in the euchromatic portion of the genome. The D. melanogaster genome was ∼11% repetitive overall (of which 2.3% is simple sequence repeats/low complexity sequence) and ∼7% in the euchromatic portion of the genome.
The first step in creating D. yakuba chromosomal fasta files was to align the D. yakuba WGS assembly data against the D. melanogaster genome. D. yakuba supercontigs were artificially broken into 1,000-bp fragments and aligned against the D. melanogaster genome using BLAT [123]. An alignment was defined as “unique” if its best scoring match had a score of at least twice that of its next best scoring alignment. Of the 139.5 Mb of D. yakuba supercontigs that uniquely aligned to the D. melanogaster genome (4.2 Mb of which aligned uniquely to D. melanogaster unlocalized sequence, chrU), only 16 supercontigs totaling 15.1 Mb contained unique assignments to more than one chromosome arm. Eleven of these involved alignments to heterochromatin where only less than ∼5% of the supercontig aligned uniquely to the D. melanogaster genome. These contigs were assigned to either chrU or the heterochromatic portion of the chromosome for cases where the contig aligned to both the heterochromatic and nonheterochromatic portion of the same chromosome. One 200-kb contig had only 6.2 kb that uniquely mapped to the D. melanogaster genome, 3.8 kb mapping to chr2R, and 2.4 kb mapping to chrX. This supercontig was assigned to chrU. The remaining four supercontigs were alignments to chromosome arms 2L and 2R, the location of a known pericentric inversion between D. melanogaster and D. yakuba.
The D. yakuba contigs were initially ordered by their position along the assigned D. melanogaster chromosomes. Because there are rearrangements in D. yakuba as compared to D. melanogaster, we allowed one portion of a D. yakuba supercontig to align to one region of a chromosome and the remaining portion to align elsewhere along that chromosome. For example, four supercontigs aligned to both chromosome arms 2L and 2R. However, these 2L/2R cross-overs and other interspecific nonlinearities are expected given the known chromosome inversions [124] between D. yakuba and D. melanogaster. This initial ordering for 2L, 2R, 3L, 3R, and X was used as the starting point for manually introducing inversions in the D. melanogaster-ordered D. yakuba supercontigs. The goal was to minimize the total number of inversions required to “rejoin” all D. yakuba supercontigs previously assigned to distant chromosomal regions based on D. melanogaster alignments (L. Hillier, unpublished data). Inversions were only introduced between contigs and not within contigs. Using this process, we created the final chromosomal D. yakuba sequence.
Sequence data were obtained from paired-end plasmids from the various D. simulans strains using standard laboratory protocols (http://genome.wustl.edu). A genomic assembly was also created. We began by generating an ∼4× WGS assembly of D. simulans w501 using PCAP [18]. The w501contigs were initially anchored, ordered, and oriented by alignment with the D. melanogaster genome in a manner similar to that described above for alignments between the D. yakuba and D. melanogaster genome. The assembly was then examined for places where the w501 assembly suggested inversions with respect to the D. melanogaster assembly. One major inversion was found, confirming the already-documented inversion found by [124]. Six other D. simulans lines (C167.4, MD106TS, MD199S, NC48S, SIM4, and SIM6) were also assembled using PCAP with ∼1× coverage. Using the 4× WGS assembly of the D. simulans w501 genome as a scaffold, contigs and unplaced reads from the 1× assemblies of the other individual D. simulans lines were used to cover gaps in the w501 assembly where possible. Thus, the resulting assembly is a mosaic containing the w501 contigs as the primary scaffolding, with contigs and unplaced reads from the other lines filling gaps in the w501 assembly (L. Hillier, unpublished data). The D. simulans w501 whole-genome shotgun assembly can be accessed at GenBank.
The goal was to align unique D. melanogaster reference sequence assembly v4 to orthologous D. simulans sequence. The D. melanogaster genome was preprocessed to soft mask all 24mers that were not unique, as such sequences were not expected to have a discriminating effect during mapping of D. simulans reads. Transposable elements in the reference sequence were also masked.
The D. simulans WGS reads were quality trimmed prior to assembly based on their phred-score derived error probability. These error probabilities were used to trim the read back to the longest contiguous interval with an average probability of error less than 0.005. Each end was then examined and trimmed until its terminal 10 bp had an average probability of error less than 0.005. If the read was less than 50 bp after this process, it was discarded. These criteria resulted in 164,480 discarded reads from a total 2,424,141 reads. See Table S33 for read and trim statistics.
A dynamic programming algorithm was used to create a maximum-likelihood description of the evolutionary path between sequences from the two species with respect to the standard alignment model, which was extended to incorporate the possibility of sequencing error. To improve the accuracy of the alignments, optimal parameters were estimated with respect to the overall expected evolutionary distance between the two species. This was done from a first-pass assembly using the method described in [129]. Because dynamic programming is not feasible on a genomic scale, we determined candidate locations for each read using the MegaBLAST (http://www.ncbi.nlm.nih.gov/BLAST/docs/megablast.html) algorithm. A read was then realigned to each candidate location as a single contiguous alignment using a derivative of the Smith-Waterman algorithm, which was adapted to incorporate the expected divergence and the error probabilities provided by Phred quality scores. Alignments were ranked by score. Reads were considered unambiguously mapped if their alignment covered at least 90% of the sequence and showed more than two high-quality differences between the putative best orthologous location and a possible secondary candidate location. Reads were incorporated into the assembly on a clone-by-clone basis only if both mate-pairs were unambiguously mapped with the proper orientation and appropriate distance from each other.
For each D. simulans line, the aligned reads were coalesced into syntenic contigs using their overlap with respect to the D. melanogaster genome. Note that “overhanging” or unaligned sequence that may represent transposable elements, other repetitive sequence, or highly diverged sequence, was not considered. This “master–slave” multiple alignment contains reads that are aligned “optimally” with respect to the D. melanogaster reference sequence. However, this does not ensure that the reads are optimally aligned with respect to each other. For instance, small, identical insertion or deletion variants may not be mapped to precisely identical locations in all D. simulans reads. To address this problem, the D. melanogaster reference sequence was set aside, and the method of Anson and Meyers [125] was used to optimize the alignment of each component read of each D. simulans line with respect to a D. simulans–only consensus sequence. This method, which minimizes the sum of differences between each of the reads and the consensus sequence, belongs to the class of expectation maximization algorithms [125]. The round robin, align-and-update algorithm converges on a consensus sequence and alignment that most parsimoniously describe the differences between each read and the consensus. This has the effect of coalescing deletions and aligning insertions. The end result of the assembly is a multi-tiered alignment with associated quality scores for (i) the trimmed reads, (ii) the assembled sequences within lines, and (iii) a species consensus sequence, all aligned to the D. melanogaster reference sequence. A reference sequence was produced for each D. simulans line by concatenating the syntenically assembled contigs that were padded with respect to the D. melanogaster reference sequence. The result is a set of D. simulans genomes onto which D. melanogaster annotation can be directly mapped.
Nine regions, including coding and noncoding DNA, were chosen to cover a range of polymorphism levels as predicted by an early version of the syntenic assembly. These regions were amplified from lines C167.4, MD106TS, NC48S, and w501, and sequenced at UNC-Chapel Hill High-Throughput Sequencing Facility. Sequences were assembled using Consed; a minimum quality score of 30 was required. Approximately 27,500 bp were sequenced per line. The per-base discrepancy between these sequences and the current syntenic assembly (insertions, deletions, and masked bases omitted) was estimated as 0.00043.
An orthology map (with respect to the D. melanogaster reference sequence) of D. yakuba assembly (v1.0) was generated by the Mercator program (http://rd.plos.org/pbio.0050310a). The MAVID [126] aligner was run on each orthologous segment in the map. MAVID uses protein-coding hits reported by Mercator to anchor its alignment of each segment. It recursively finds additional anchors and then runs the Needleman-Wunsch algorithm in between the anchors to obtain a single, global alignment of the entire orthologous segment.
These regions were filtered based on manual examination of the density of annotated repetitive sequence in the centromere and telomere proximal regions of the five large arms. The transition from the “typical” euchromatic density of large repeats to the typical “beta heterochromatic” pattern is obvious. The “euchromatic/heterochromatic boundaries” were drawn roughly at the edges of the first annotated gene within each euchromatic arm.
The following regions were excluded from analysis: (i) X 1 to 171944 AND 19740624 to END; (ii) 2L 1 to 82455 AND 21183096 to END; (iii) 2R 1 to 2014072 AND 20572063 to END; (iv) 3L 1 to 158639 AND 22436835 to END; and (v) 3R 1 to 478547 AND 27670982 to END.
The sequence for each line is derived from the multiple alignment of reads to the D. melanogaster reference assembly (v4). For each line and each column (nucleotide position) corresponding to a D. melanogaster base, a likelihood model was used to determine the quality score for each of the four bases. The quality score was calculated as −10log(1 – probability base is correct). To compute the probability a base call is correct, we assume that each read is an observation of a random variable with equal likelihoods for all four bases with some probability of error. From the definition of a phred score, the probability of error for a particular observed call is: 10(phred score/–10 ). We assumed that a base in error is equally likely to be any one of the three other bases. Then, for a given position A, Bayes theorem implies the probability (Pr) that the call is correct is
Where Pr[A] = 1/4, Pr[Observations|A is correct] = likelihood of A observations being correct and non-A observations being incorrect, and Pr[Observations] = likelihood of seeing observed values given frequency and error rates.
Quality scores were truncated at 90. The sequences for each line were investigated for regions containing unusually high densities of high-quality discrepancies, which are due to residual heterozygosity, duplication, and erroneous sequence. These regions were filtered from subsequent analysis (see below). For each line, the support for each alternative (A, G, C, and T) at each aligned base was the sum of the qualities, with the highest quality base assigned as the base for that line/position. Implicit in this approach is that a base is called only if the highest quality base has a quality score that is 30 or more greater than that of the next highest quality base. The combined SIM4/SIM6 consensus was also treated in this manner.
Residual heterozygosity within lines or duplications present in D. simulans but not D. melanogaster can lead to regions with excess high-quality discrepancies between reads within lines. We refer to these as single-nucleotide discrepancies. We derived a distribution of the number of discrepancies per site over each chromosome for each D. simulans line. We based the distributions on counts of within-line discrepancies per site in 500-bp windows that had 250-bp overlap. We took the conservative approach of filtering windows in all the lines that fell into the top 0.5% of the distribution in any single line. In other words, a window with a high-quality discrepancy in one line was filtered from the entire dataset, even if the other lines had no discrepancy. Overall, 334,500 base pairs were filtered from the genome. The number of sites filtered for each chromosome arm were 39 kb for 2L, 86.5 kb for 2R, 58 kb for 3L, 73 kb for 3R, and 78 kb for X.
One large inversion on chromosome arm 3R distinguishes the two species. Phylogenetic analysis of the cytogenetic data suggested that the inversion fixed in the D. melanogaster lineage [39]. Thus, D. simulans 3R is rearranged with respect to the D. melanogaster reference sequence. We used D. melanogaster/D. simulans alignments provided by the UC Santa Cruz Genome Browser to locate the putative breakpoints of the inversion as Chr3R: 3874907 and 17560827.
All features were defined in the D. melanogaster v4.2 annotation (http://flybase.org). For each gene, the longest isoform (i.e., the isoform the with greatest number of codons) was chosen for analyses. Exons that were not part of the longest isoform were excluded from all feature-based analyses, but were included in window analyses. The analyzed introns correspond to these longest isoforms; all introns were included in window analyses. Intronic sequences within annotated UTRs or that overlapped any coding sequence were excluded. UTRs investigated for this paper were restricted to those inferred from “Gold Collection” genes with completely sequenced cDNAs (http://www.fruitfly.org/EST/gold_collection.shtml). All annotated CDS sequences were used regardless of the associated empirical support. Intergenic regions were defined as noncoding segments between annotated genic regions (UTRs, coding sequence, and noncoding RNAs) regardless of strand. Defined intergenic regions from v4.2 annotation were checked against all known coding and UTR coordinates; any nucleotides that overlapped a genic region were removed from the intergenic set before analysis.
We established a conservative gene set for analyses (base composition analyses excepted) by including only genes for which the start codon (ATG or otherwise), splice junctions (canonical or otherwise), and termination codon position agreed with the D. melanogaster reference sequence. We took the conservative approach of excluding from the gene-based analysis any gene for which any of the six D. simulans gene models disagreed with the D. melanogaster gene model.
Long insertions and deletions (indels) are difficult to identify using only aligned reads. As indel length increases, the likelihood that indels are missed increases because they are either too long or occur near the end of a read, which compromises alignment. Furthermore, indel error probabilities are difficult to estimate. These considerations led us to restrict our analysis to indels of 10 bp or less and to restrict our analysis of divergence to the D. simulans versus D. melanogaster comparison. Variants were classified as insertions or deletions relative to the D. melanogaster reference sequence. The quality score for an insertion was the average quality score of sequence in that insertion; the quality score for a deletion was the minimum of qualities of the two flanking nucleotides. Qualities were determined this way to provide a metric of overall sequence quality in the region of a putative indel, thereby allowing a quantitatively defined cutoff for inferring indel variants; only indels of high quality (over phred 40) were considered in the analysis.
Light, variable coverage of each line requires that statistical estimation and inference account for coverage variation. When appropriate (e.g., contingency tables of frequency variation), counts of variants within a coverage category were used. In other estimation and inference settings, the familiar estimators were applied to each coverage class and then averaged, weighting by the proportion of total covered base pairs in the window or other feature.
Heterozygosity. The expected nucleotide, insertion, and deletion heterozyogsity was estimated as the average pairwise differences between D. simulans alleles as follows:
πi is the coverage-weighted average expected heterozygosity of nucleotide variants (i = nt), deletions (i =Δ) or insertions (i = ▿) per base pair. “Expected heterozygosity” assumes the six sequenced genomes were drawn from a single, randomly mating population. Variable coverage over sites led us to extend the typical calculation of expected heterozygosity [127,128] to the following:
where nc is the number of aligned base pairs in the genomic region (e.g., gene feature or window) with sequencing coverage c. kcj is the number of sites in this region with coverage c at which the derived state (nt, ▵, or ▿) occurs in j out of the c sequences. This estimator was used for 10-kb windows, 50-kb windows, 30-kb sliding windows (10-kb increments), 150-kb sliding windows (10-kb increments), and 210-kb windows (10-kb increments), including all windows for which coverage was >200 bp. Expected heterozygosity was also estimated for genomic features (exons, introns, UTRs, and intergenic sequence) that had a minimum size and coverage [i.e., n(n – 1) × s ≥ 100, where n = average number of alleles sampled and s = number of sites]. For coding regions, the numbers of silent and replacement sites were counted using the method of Nei and Gojobori [129]. The pathway between two codons was calculated as the average number of silent and replacement changes from all possible paths between the pair.
The variance of pairwise differences in sliding windows (150-kb windows, 10-kb increments) was used as a method of summarizing the magnitude of linkage disequilibrium across the D. simulans genome. For each window, we calculated coverage weighted variance of the expected heterozygosity (see above) for all pairs of alleles.
Divergence. Unpolarized (i.e., pairwise) divergence between D. simulans and D. melanogaster was estimated for 10-kb windows, 50-kb windows, 30-kb sliding windows (10-kb increments), 150-kb sliding windows (10-kb increments), 210-kb windows (10-kb increments), and genomic feature that had a minimum number of nucleotides represented [i.e., n × s > 100, with n and s as above in calculations of π. Unpolarized divergence was calculated as the average pairwise divergence at each site, which was then summed over sites and divided by the total number of sites. A Jukes-Cantor [130] correction was applied to account for multiple hits. For coding regions, the numbers of silent and replacement sites were counted using the method of Nei and Gojobori [129]. The pathway between two codons was calculated as the average number of silent and replacement changes from all possible paths between the pair. Estimates of unpolarized divergence over chromosome arms were calculated for each feature with averages weighted by the number of sites per feature.
Lineage-specific divergence was estimated by maximum likelihood using PAML v3.14 [131] and was reported as a weighted average over each line with greater than 50 aligned sites in the segment being analyzed. Maximum likelihood estimates of divergence were calculated over 10-kb windows, 50-kb windows, 30-kb sliding windows (10-kb increments), 150-kb sliding windows (10-kb increments), 210-kb windows (10-kb increments), and gene features (exons, introns, and UTRs). PAML was run in batch mode using a BioPerl wrapper [132]. For noncoding regions and windows, we used baseml with HKY as the model of evolution to account for transition/transversion bias and unequal base frequencies [133]; for coding regions, we used codeml with codon frequencies estimated from the data.
Insertion and deletion divergence was calculated as divi, the coverage-weighted average divergence of deletions (i = ▵) or insertions (i = ▿) per base pair.
where nc is the number of aligned base pairs in the genomic region (e.g., gene feature or window) with sequencing coverage c. kcj is the number of sites in this region with coverage c at which the derived state with respect to the D. melanogaster reference sequence (▵ or ▿) occurs in j out of the c sequences.
Unpolarized MK tests [4] used D. simulans polymorphism data and the D. melanogaster reference sequence for counting fixed differences. Polarized MK tests used D. yakuba to infer the D. simulans/D. melanogaster ancestral state. For both polarized and unpolarized analyses, we took the conservative approach of retaining for analysis only codons for which there were no more than two alternative states. For cases in which two alternative codons differed at more than one position, we used the pathway between codons that minimized the number of nonsynonymous substitutions. This is conservative with respect to the alternative hypothesis of adaptive evolution. Polymorphic codons at which one of the D. simulans codons was not identical to the D. melanogaster codon were not included. To reduce multiple testing problems, we filtered the data to retain for further analysis only genes that exceeded a minimum number of observations; we required that each row and column in the 2 × 2 table (two variant types and polymorphic versus fixed) sum to six or greater. Statistical significance of 2 × 2 contingency tables was determined by Fisher's Exact test. MK tests were also carried out for introns and Gold Collection UTRs by comparing synonymous variants in the associated genes with variants in these functional elements. For intergenic MK tests, we used synonymous variants from genes within 5 kb of the 5′ and/or 3′boundary of the intergenic segment. For some analyses, we restricted our attention to MK tests that rejected the null in the direction of adaptive evolution. This categorization was determined following Rand and Kann [134].
Polarized 2 × 2 contingency tables were used to calculate α, which under some circumstances can be thought of as an estimate of the proportion of variants fixing under selection [11]. Bootstrap confidence intervals of α and of the ratio of polymorphic-to-fixed variants for each functional element (Table 2) were estimated in R using bias correction and acceleration [135].
Our approach takes overall rate variation among lineages into account when generating expected numbers of substitutions under the null model and allows for different rates of evolution among chromosome arms (e.g., a faster-X effect). For example, the number of substitutions for all X-linked 50-kb windows was estimated using PAML (baseml), allowing different rates for each lineage. All D. simulans lines were used, with the estimated substitution D. simulans rate for each window being the coverage-weighted average. This generated an empirically determined branch length of the X chromosome for the average over each of the D. simulans lines (from all three way comparisons with D. melanogaster and D. yakuba) weighted by the number of bases covered. We carried out a relative rate test for windows or features in D. simulans and D. melanogaster by generating the expected number of substitutions for each window/feature/lineage based on the branch length of the entire chromosome in each lineage (PAML) and the coverage of the window/feature in question in each lineage. We then calculated the deviation from the expected number of substitutions as (observed – expected substitutions)2/expected substitutions for any window/feature/lineage.
For each GO term associated with at least five MK tests, we calculated the proportion of significant (p < 0.05) tests. We then randomly selected n p-values from the set of all MK p-values, where n is the number of tests in the ontology category. We repeated this procedure 10,000 times to get the empirical distribution of the proportion significant p-values for each GO term.
The relative rate χ2 for dN was calculated for each gene as described above. Genes showing a significant (p < 0.05) acceleration of dN in the D. simulans lineage were identified as described in the previous section. The probabilities of observing as many, or more, significant relative rate χ2 tests for dN were determined by permutation as described in the previous section.
We retrieved ontology terms associated with genes that fell under windows of interest in linked selection analyses. Then, for each term, we divided the number of instances that the term was represented in the windows of interest by the total number of genes in the genome that are associated with the ontology term. This gave us a proportional representation of each GO term in windows of interest. We compared this proportion for each GO term with the empirical distribution of proportions derived from permuted datasets. For each permuted dataset, we randomly picked a nonoverlapping set of windows that were the same size in numbers of base pairs as the observed windows. Each window was guaranteed to contain at least one gene, given that windows of interest have higher-than-average gene density. We then retrieved the ontology terms associated with the genes under the random set of windows. We next calculated the proportion of each term as described above for the observed windows. We repeated this procedure 1,000 times to obtain an empirical distribution of proportions of each term in random windows. The proportion of each GO term in the original observed windows of interest was compared to this empirical distribution to obtain a probability of observing that proportion of each term in windows of interest.
We wanted to know whether ontology terms were clustered in the genome. We tested whether each ontology term was significantly clustered by calculating the coefficient of variation based on occurrence in 1-Mb, nonoverlapping windows and compared that to the coefficient of variation from permuted datasets in which we randomized the locations of genes on each chromosome arm.
Genes were assigned to expression categories, with the goal of determining whether certain categories had a greater proportion of significant MK tests for adaptive protein divergence than expected by chance. Two types of data, expressed sequence tag (EST) collections and microarray experiments, were used. Genes associated with EST collections from D. melanogaster (head, ovary, and testis from Flybase and spermatheca from Swanson et al [136]) were assigned to that tissue expression category. Female-mating responsive genes were those defined by microarray experiments [137]. Male- and female-biased genes were defined based on microarray experiments of Parisi et al. [138] and Arbeitman et al. [139]. Male- and female-biased genes from Parisi et al. [138] were obtained directly from their Tables S41 and S42. Arbeitman et al. [139] measured expression over the D. melanogaster life cycle for adult males and females. We averaged expression for each gene over the time points taken for each stage. For example, there were 30 time point measurements during the egg stage; we used the average expression over those 30 time points. We repeated this for larvae, metamorph, adult female, and adult male stages. Each gene was provisionally designated as having biased expression for the stage with the maximum average expression, which we will call the biased stage. For each gene, we calculated the average difference between the biased stage expression value and the other stage expression values. This generated a distribution of differences for each comparison of stages. A gene was finally determined to have biased expression if the average difference between the biased stage and the other stages fell into top half for that stage distribution. This procedure resulted in 592, 374, 223, 466, and 296 stage-biased genes for egg, larvae, metamorph, adult male, and adult female stages, respectively. We calculated the proportion of genes in a group (e.g., male-biased) that had significant MK tests (p < 0.05). We used permutation testing to determine whether the proportion of significant polarized MK tests deviating in the direction of adaptive protein evolution exceeded the 95% tail of the empirical distribution, based on 10,000 datasets of randomly selected MK tests, sampled without replacement.
We tested whether pairs of proteins that interact with one another were more likely to show evidence of adaptive protein divergence than random pairs of proteins with no evidence of interaction. Data were from Giot et al. [140]. We considered pairs of genes to have a significant interaction if the probability of interaction was greater than 0.5. We calculated the proportion of interacting pairs where both members had significant evidence of adaptive evolution (MK p-values < 0.05). We compared this proportion to the distribution of proportion generated from permuted datasets generated by randomly drawing pairs of genes without replacement from the Giot et al. [140] dataset.
Hudson, Kreitman, and Aquadé [2] proposed a test of the neutral theory of molecular evolution in which the numbers of polymorphic and (fixed) divergent sites are contrasted between two independent loci (genomic regions). The distribution of a χ2-like test statistic can be determined by simulation for any assumed values of recombination within each locus. However, given the small sample size here and the genomic scale of the data, we used an analogous statistic for polymorphisms and fixations on the D. simulans lineage in various sizes of sliding windows, combined over coverage classes. First, the average proportion of segregating sites in D. simulans and parsimony-inferred fixed differences for each chromosome arm in D. simulans was determined for each coverage class over a range of sliding window sizes (10 kbp to 510 kbp). The test statistic is a simple two-cell χ2, in which the observed numbers (summed over coverage classes) of segregating and fixed sites are contrasted with their expected numbers (summed over coverage classes, the chromosome arm average for each coverage class times the total numbers of segregating and fixed sites in that class). Only sites for which unambiguous, parsimony-inferred D. simulans/D. melanogaster ancestral states could be determined were included in the analysis. In a number of figures, χ[–log10(p)] is plotted; –log10 of p, critical value for this χ2, was given the sign of the difference, observed numbers of segregating site – expect number of segregating sites. As expected (Figure 1), there is a clear tendency for the level of polymorphism (both πnt and proportion of segregating sites) to decline proximal to the telomeres and centromeres. Therefore, the test statistics discussed in this section were determined by generating expected values as described above, but only including the “central euchromatic” regions. These were defined as the regions bounded by the first and last position on each chromosomes arm for which the proportion of segregating sites was greater than or equal to the chromosome arm average in a 510-kbp window. While this makes deviations in the centromere and telomere proximal regions appear greater, it removes the obvious bias toward positive deviations (i.e., excess polymorphism) that would be created by including large genomic regions known to show reduced polymorphism when generating expectations. Minimum values for the expected numbers of segregating and fixed sites were one (unless otherwise indicated). Windows with coverage <200 bp were dropped (unless otherwise indicated).
Expected nucleotide heterozygosity and polarized divergence were calculated for 10-kb and 50-kb nonoverlapping windows spanning each chromosome arm as described above. For each arm, autocorrelation between successive windows was calculated as:
where there are n windows along an arm, and xt represents the value of nucleotide heterozyogsity or divergence for the t-th window. Significance of r for all arms for both polymorphism and divergence was calculated by permutation. All calculations were carried out in R (http://www.r-project.org).
We set out to find putative selective sweeps that occurred concomitantly with migration by D. simulans out of Africa/Madagascar. We expect the signature of these sweeps to be low variation in New World (NW) lines, defined here as w501 and SIM4/6, compared to Old World (OW) lines, defined here as C167.4, MD199S, and MD106TS. The method described here addresses the issue of autocorrelated loci. Our approach was to simulate datasets with the same degree of autocorrelation as that of the observed data, and to determine whether there are longer runs of windows with disproportionately low NW π in the actual data than one would expect by chance. All data manipulation, calculations, and simulations were carried out in R using functions available within the “base” and “stats” packages. Mean nucleotide diversity (π) from 10-kb nonoverlapping windows throughout the genome from the two NW and three OW lines were used. Adjacent 10-kb windows were averaged (weighted by coverage) to obtain 20-kb windows. Remaining windows for which no estimate of π was available were conservatively estimated by interpolation. There were no gaps in the 20-kb window data longer than three consecutive windows in either population. For each window, the ratio of NW π:OW π (π NW:πOW) was measured. Maximum likelihood estimates of first-order coefficients of autocorrelation for each of the chromosome arms were found (all were significant).
Monte Carlo simulations of the ratio πNW:πOW were performed according to the following procedure. We first randomly sampled ratios of π NW: πOW from the data with replacement for each arm separately; this ensures that our simulated data has the same mean and variance as the actual data. A first-order autoregressive filter was then applied to the randomly sampled data using the estimate of autocorrelation for the given chromosomal arm, according to the following relationship:
where parameter μ is the mean of the sampled data, ρ is the autocorrelation, Xi – 1 is previous value in the series, and Xi is the original sampled measure for the ith window. This filter imposes the observed autocorrelation onto the sampled data to mimic the observed autocorrelation, resulting in a new value, Xi*, for each window. Variance and estimated first-order autocorrelation of the simulations were similar to those of the empirical data without altering this procedure.
A lower threshold of π NW: πOW, below which 5% of the empirical data points reside, was determined. Significance of runs of windows below this threshold was determined by comparison to the distribution of the run lengths in 10,000 Monte Carlo simulation runs for each chromosome arm, performed as described above. P-values for each arm were corrected for multiple comparisons conservatively via Bonferroni correction (Dunn-Sidak corrections did not result in an increased number of significant sweeps).
Parsimony was used to infer D. simulans/D. melanogaster ancestral states; D. yakuba was the outgroup. Only codons with one synonymous variant among the three species were included in these analyses. The preferred codon set was defined following Akashi [113]. For some analyses, preferred and unpreferred substitution rates were determined by dividing the number of substitutions of each type by the number of ancestral codons of the appropriate ancestral state (unpreferred ancestors for the preferred substitution rate and preferred ancestors for the unpreferred substitution rate), all inferred by parsimony. In principle, excess unpreferred polymorphisms at synonymous sites could erroneously lead one to infer directional selection on other sites. However, the ratio of preferred-to-unpreferred polymorphisms is not significantly different (pooled across genes or gene-by-gene) for UTRs that had significant versus nonsignificant MK tests in contrasts of synonymous and UTR sites. For introns that showed a significant MK test versus synonymous sites, there was a slightly larger ratio of unpreferred-to-preferred polymorphisms compared to the ratio for introns that were not significant. However, this was seen only in the pooled analysis and not in the gene-by-gene analysis. We found that significant intron and UTR MK tests had more similar coverages (e.g., 5′ UTR versus synonymous) compared to tests that were not significant, suggesting that the large number of significant noncoding versus synonymous tests cannot be explained by relatively small coverage differences across site-types. Overall, these data suggest that most of the highly significant MK tests of noncoding DNA are not explained by excess unpreferred polymorphisms or coverage variation.
Base composition analyses on noncoding DNA were carried out in a similar fashion, with parsimony being used to infer the D. simulans/D. melanogaster ancestor. Only unambiguous parsimony-inferred sites were used in these analyses.
All X-linked genes for which Flybase reported genetic and physical locations (first nucleotide of the gene in Flybase annotation of D. melanogaster v4.2) were used. Genetic and physical distances were determined for 12-gene intervals, sliding one gene at a time; estimates of cM/kb per interval were used as estimates of recombination rate per physical length. Mean physical and genetic distances per interval were 1.55 Mb and 5 cM, respectively. Two intervals with negative estimates of cM/kb, indicative of discordant genetic and physical data were removed, leaving estimates of cM/kb for 150 intervals. The physical location of the interval was defined as the midpoint between physical locations of the first and last gene. For analyses investigating correlations of 50-kb windows of polymorphism and divergence with crossing-over, midpoints were rounded to the nearest 50,000. If multiple intervals were rounded to the same number, the distal interval was used in the analyses.
Cloned elements. The “hanging ends” of well-mapped plasmid clones that were not fully aligned to D. melanogaster were examined by BLAST for extensive (100 bp or greater), high-quality (90% or greater) sequence similarity to known transposable elements of D. melanogaster (v 9.2, http://www.fruitfly.org/p_disrupt/TE.html). The coordinates are slightly rounded to facilitate finding duplicates slightly off in alignment.
Clustered elements. This analysis used plasmid clones for which only one mate pair mapped uniquely and unambiguously to the genome according to the method described previously. The other mate pair was compared to the D. melanogaster transposable element database v9.2. If the read mapped uniquely and unambiguously to a transposable element (90% coverage, 90% identity, at least two high quality differences to a secondary candidate), a transposable element was considered as mapped to the general genomic location of its mate pair. The estimated location begins at the end of the mate pair read and ends 10 kb away in the appropriate direction determined by the direction of the alignment. Transposable elements from the same family located within 5 kb of each other in the same D. simulans line were considered the same element, and therefore, were clustered.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank/) accession number for D. yakuba is AAEU01000000 (version 1) and for the D. simulans w501 whole-genome shotgun assembly is TBS-AAEU01000000 (version 1).
|
10.1371/journal.pgen.1003320 | Conditional Inactivation of the DNA Damage Response Gene Hus1 in Mouse Testis Reveals Separable Roles for Components of the RAD9-RAD1-HUS1 Complex in Meiotic Chromosome Maintenance | The RAD9-RAD1-HUS1 (9-1-1) complex is a heterotrimeric PCNA-like clamp that responds to DNA damage in somatic cells by promoting DNA repair as well as ATR-dependent DNA damage checkpoint signaling. In yeast, worms, and flies, the 9-1-1 complex is also required for meiotic checkpoint function and efficient completion of meiotic recombination; however, since Rad9, Rad1, and Hus1 are essential genes in mammals, little is known about their functions in mammalian germ cells. In this study, we assessed the meiotic functions of 9-1-1 by analyzing mice with germ cell-specific deletion of Hus1 as well as by examining the localization of RAD9 and RAD1 on meiotic chromosomes during prophase I. Hus1 loss in testicular germ cells resulted in meiotic defects, germ cell depletion, and severely compromised fertility. Hus1-deficient primary spermatocytes exhibited persistent autosomal γH2AX and RAD51 staining indicative of unrepaired meiotic DSBs, synapsis defects, an extended XY body domain often encompassing partial or whole autosomes, and an increase in structural chromosome abnormalities such as end-to-end X chromosome-autosome fusions and ruptures in the synaptonemal complex. Most of these aberrations persisted in diplotene-stage spermatocytes. Consistent with a role for the 9-1-1 complex in meiotic DSB repair, RAD9 localized to punctate, RAD51-containing foci on meiotic chromosomes in a Hus1-dependent manner. Interestingly, RAD1 had a broader distribution that only partially overlapped with RAD9, and localization of both RAD1 and the ATR activator TOPBP1 to the XY body and to unsynapsed autosomes was intact in Hus1 conditional knockouts. We conclude that mammalian HUS1 acts as a component of the canonical 9-1-1 complex during meiotic prophase I to promote DSB repair and further propose that RAD1 and TOPBP1 respond to unsynapsed chromatin through an alternative mechanism that does not require RAD9 or HUS1.
| Meiosis is a specialized cell division process in which germ cells undergo two cell divisions to produce haploid progeny. Two processes, genetic recombination and chromosome pairing/synapsis, are critical for successful meiosis and the production of gametes with high chromosomal integrity. The RAD9-RAD1-HUS1 (9-1-1) complex has been proposed to play critical roles in recombination as well as in the checkpoint-dependent monitoring of chromosomal synapsis by facilitating activation of the ATR checkpoint kinase. Our data indicate that HUS1 is required for normal germ cell development and fertility, for efficient completion of a subset of meiotic DNA recombination events, and for proper exclusion of the non-sex chromosomes from a specialized, repressive chromatin domain containing the X and Y chromosomes. However, HUS1 is not required for the meiotic functions of ATR in responding to chromosome synapsis defects. Furthermore, RAD1 localizes to sites along asynapsed chromosomes that lack detectable RAD9, and does so in the absence of Hus1, implicating RAD1 in a novel response to unsynapsed chromatin that is independent of the canonical 9-1-1 complex. Since mice lacking Hus1 in germ cells exhibit chromosomal abnormalities and severely reduced fertility, this work has broad implications for the maintenance of genome stability in the germline and for human reproductive health.
| The requirement for effective genome maintenance is particularly notable in germ cells, which must transmit high quality DNA to future generations. Therefore, germ cells must employ DNA damage response mechanisms that are at least as stringent as those present in somatic cells. Meiosis, which includes the intentional generation and subsequent repair of DNA double-strand breaks (DSBs), involves a variety of DNA repair mechanisms as well as cell cycle checkpoints that monitor chromosomal integrity; however, much remains unknown about how these mechanisms operate in mammalian cells. In this study, we investigated how an essential DNA repair and DNA damage checkpoint complex, the RAD9-RAD1-HUS1 (9-1-1) complex, functions to maintain genome integrity in the germline.
The 9-1-1 complex is a heterotrimeric ring that shares extensive structural similarity with PCNA, the sliding clamp that functions in DNA replication and repair [1]. In mammalian somatic cells, the best characterized role of 9-1-1 is in activation of ATR pathway DNA damage checkpoint signaling following replication-associated DNA damage [2]. In response to stalled replication forks or lesions involving single-stranded DNA, the 9-1-1 complex is loaded onto 5′ recessed ends by the RAD17-RFC clamp loader complex. The ATR kinase is independently loaded onto RPA-coated single-stranded DNA through the interaction of ATR-interacting protein (ATRIP) with both ATR and RPA [3]. The phosphorylated RAD9 C-terminus physically recruits the ATR activator TOPBP1, which then stimulates the kinase activity of ATR [reviewed in 4]–[6]. Once active, ATR phosphorylates downstream effectors such as CHK1 that promote cell cycle arrest, DNA repair, or apoptosis.
In addition to its known checkpoint signaling function, the 9-1-1 clamp physically interacts with a variety of DNA repair proteins, indicating that, like PCNA, 9-1-1 may also function as a scaffold for recruiting DNA repair proteins to damage sites. Indeed, 9-1-1 has been shown to interact with RAD51 in human cells [7], and 9-1-1 complexes from yeast, mouse, and human physically interact with, and in some cases, stimulate the activity of, translesion polymerases [8] as well as base excision repair factors, including DNA Polymerase β [9], [10], FEN1 [11], [12], APE1 [10], DNA ligase I [13], and the NEIL1, TDG, OGG1, and MutY glycosylases [14]–[18]. Other 9-1-1 physical interactors are as varied as HDAC1 histone deacetylase [19], WRN helicase [20], mismatch repair factors MSH2, MSH3, MSH6, and MLH1 [21], [22], and the ATR co-activator RHINO [23]. That 9-1-1 interacts with and stimulates numerous DNA repair factors points to a direct role of 9-1-1 in facilitating DNA repair apart from its ATR-dependent checkpoint signaling activities, though in vivo evidence for this proposed function for the mammalian 9-1-1 complex remains limited.
During meiosis, homologous chromosomes pair, synapse, and undergo recombination. The ATR signaling pathway has been proposed to monitor both meiotic chromosomal synapsis and recombination as part of the pachytene DNA damage checkpoint that is believed to prevent cells with unrepaired breaks or synapsis defects from progressing beyond meiotic prophase I. ATR and its activator TOPBP1 localize to unsynapsed autosome cores in zygotene spermatocytes and to the sex body domain during pachytene [24]–[26], where the mostly unsynapsed X and Y chromosomes in males are packaged into a unique chromatin territory [27], [28]. Phosphorylation of histone H2AX to γH2AX, which marks the entire sex body domain, is thought to be ATR-dependent, whereas H2AX phosphorylation at DSB sites on autosomes during leptotene and early zygotene stages appears to depend primarily on another checkpoint kinase, ATM [26], [29]. In the sex body, ATR promotes transcriptional silencing of the X and Y chromosomes in a process termed meiotic sex chromosome inactivation (MSCI) [30], which is required for male fertility [31]. ATR and TOPBP1 also promote meiotic silencing of unsynapsed chromatin (MSUC), a process fundamentally similar to MSCI that functions at sites of autosomal asynapsis during pachytene [24], [25], [30], [32]. MSUC and MSCI require BRCA1 loading at asynapsed chromatin, and both BRCA1 and the HORMA domain-containing protein HORMAD2 must be present at asynapsed axes in order for ATR to be recruited [26], [33]. While TOPBP1 localizes to the sex body and sites of failed autosomal synapsis and is a known ATR activator, it is unclear whether it is strictly required for MSCI and MSUC, and also whether its recruitment is dependent upon the 9-1-1 complex, as during somatic responses to replication stress. The canonical model for mammalian ATR activation predicts a requirement for 9-1-1 in meiotic ATR activation, perhaps after HORMAD2 or BRCA1 loading, although this has remained untested until this study.
Studies in several organisms indicate that the 9-1-1 complex plays critical roles during meiosis, not only for activation of the pachytene meiotic checkpoint in response to DNA damage or synapsis defects [34]–[36], but also in DSB repair. Yeast rad17 (mRad1) mutants exhibit persistent RAD51 foci [37], in addition to increased rates of ectopic recombination and altered crossover frequencies during meiosis [38], and DDC1 (mRAD9) colocalizes with RAD51 [35]. In Drosophila, meiotic DSBs are not repaired efficiently in the absence of Hus1 [39]. Additionally, Spo11-deficient mouse mutants, which lack meiotic DSBs, have reduced Hus1 expression, suggesting that 9-1-1 functions depend upon DSB formation [40]. The same is true for expression of Mre11 and Brca2, which are known to be involved in DSB repair, further supporting the idea that 9-1-1 may function at meiotic DSBs.
The existing evidence from lower eukaryotes, together with reported physical interactions with mammalian DNA repair and meiotic proteins, suggests a possible critical role for the 9-1-1 complex during mammalian meiosis, either through activation of ATR-dependent checkpoint functions or direct roles in DNA repair at damage sites. Although mouse RAD1 has been determined to localize to both synapsed and unsynapsed meiotic chromosomes [41], little is known about the requirements for 9-1-1 complex components in mammalian germ cells because mutation of Rad9, Rad1, or Hus1 results in embryonic lethality [42]–[45]. We addressed this by producing mice with germ cell-specific Hus1 deletion. Two conditional knockout models were generated, utilizing Stra8-Cre expressed in spermatogonia and Spo11-Cre expressed in spermatocytes, and in both cases, Hus1 inactivation in testicular germ cells resulted in persistent meiotic DNA damage, chromosomal defects, and germ cell depletion. Meiotic silencing, on the other hand, appeared unaffected, indicating that HUS1 is dispensable for at least some meiotic ATR functions, including meiotic sex chromosome inactivation. Intriguingly, Hus1 inactivation resulted in complete loss of RAD9 but not RAD1 foci from meiotic chromosomes, suggesting that RAD1 has additional meiotic functions, likely related to the monitoring of unsynapsed chromatin based on its localization pattern. Together, these data indicate that efficient DSB repair, proper maintenance of the sex body domain, and preservation of chromosome integrity during meiosis require HUS1, and additionally highlight novel roles for 9-1-1 components outside of the conventional heterotrimeric complex.
Several lines of evidence from non-vertebrate organisms suggest that the mammalian 9-1-1 complex, the components of which are highly expressed in mouse and human testis, is likely to be critical for normal germ cell development and completion of meiosis. In order to assess the meiotic functions of the 9-1-1 complex, we generated mice in which the Hus1 gene was conditionally inactivated specifically in testicular germ cells. We combined a conditional floxed Hus1 allele (Figure S1A; [46]) with two Cre-expressing lines, Stra8-Cre and Spo11-Cre. Stra8-Cre mice begin to express the CRE recombinase in spermatogonia [47], which undergo several rounds of mitotic division prior to meiosis, whereas CRE expression in Spo11-Cre mice begins in spermatocytes that have initiated meiosis (Figures S1 and S2; Text S1). We focused our analysis on mice made with Stra8-Cre and used the Spo11-Cre line to confirm our findings and to assess whether defects we observed originated during pre-meiotic or meiotic processes. Hus1 deletion in the testis was confirmed by Southern blot detection of Hus1Δ2,3, the null allele produced by CRE-mediated recombination (Figures S1C, S1D and S2D). As shown in Figure 1A and Figure S3B, Western blotting of whole testis lysates further confirmed that Hus1 gene deletion using Stra8-Cre resulted in a drastic reduction in HUS1 protein level in the testis. Despite high genomic Hus1 deletion efficiency, the reduction in HUS1 protein levels was more subtle in Spo11-Cre Hus1 conditional knockout (CKO) mice, consistent with expectation that CRE expression from the Spo11 promoter would result in HUS1 loss in a more restricted subset of cells and with delayed kinetics relative to that in mice with Stra8-Cre (Figure S3C). Hus1 loss also led to a significant reduction in total RAD9 and RAD1 protein levels in testes from Stra8-Cre Hus1 CKO animals (Figure S3A, S3B), indicating that the entire 9-1-1 complex was destabilized in the absence of HUS1.
In both the Stra8-Cre and Spo11-Cre Hus1 CKO models, Hus1 loss resulted in reduced adult testis size (Figure 1B, 1E). As expected, the phenotype was more severe in Stra8-Cre Hus1 CKOs with earlier Hus1 loss than Spo11-Cre Hus1 CKOs. Testis weights were significantly reduced in Stra8-Cre Hus1 CKO males as early as postnatal day 17 (Figure 1C), raising the possibility of roles for Hus1 during early meiotic and/or pre-meiotic stages, including during spermatogonial DNA replication given the role of 9-1-1 during S-phase in somatic cells [2], [48]–[50]. Spo11-Cre Hus1 CKO males, on the other hand, exhibited normal testis sizes at 17 and 28 days but significantly reduced testis weight in adults (Figure 1F; data not shown), consistent with meiotic defects beginning at the spermatocyte stage when the Spo11 promoter is active. Due at least in part to greater fecundity in the FVB strain background, experimental mice were produced more readily using Stra8-Cre (FVB) compared to Spo11-Cre (129/B6) mice, and we therefore focused our further analysis of meiotic chromosome stability on the Stra8-Cre Hus1 CKO model. As described below, our data indicate that the majority of meiotic phenotypes associated with Hus1 inactivation are similar in both models.
The reduction in testis size in Stra8-Cre and Spo11-Cre Hus1 conditional knockout animals was accompanied by significant reductions in the production of spermatozoa. The number of sperm in the caudal epididymis of 12-week old Stra8-Cre or Spo11-Cre conditional Hus1 knockout males was reduced by approximately 10-fold and 2-fold, respectively (Figure 1D, 1G). To test sperm function in these mice, we mated Stra8-Cre Hus1 conditional knockout males to wild-type female mice. While control matings produced 18 pregnancies and 154 viable pups, 49 matings with 5 different Stra8-Cre Hus1 CKO males produced only 2 pregnancies and no viable pups (Table 1). Overall, we conclude that HUS1 is required for normal spermatogenesis and fertility.
To determine the underlying cause of the fertility defects in Hus1-deficient mice, we analyzed testes from Stra8-Cre Hus1 CKO and control animals. As shown in Figure 2A, Stra8-Cre Hus1 CKO adult testes exhibited a marked decrease in tubule size and cellularity, with many tubules lacking spermatogenic cells within the lumen, and an abundance of both pyknotic nuclei as well as multinucleate spermatid giant cells. We additionally performed immunohistochemical staining with GCNA1 antibody to detect germ cells in testis sections. Stra8-Cre Hus1 CKO testes exhibited significant loss of germ cells, including spermatogonia (Figure 2B). To further assess germ cell loss, we utilized TUNEL staining to detect fragmented DNA in Stra8-Cre Hus1 CKO nuclei. Hus1 mutant testes exhibited a significant increase in TUNEL-positive nuclei at both 17 days (Figure S4) and 12 weeks (Figure 2C–2E). Interestingly, a significant proportion of TUNEL-positive nuclei appeared to have progressed beyond pachytene stage to the first meiotic division (arrows, Figure 2E, right panel). 21% of all TUNEL-positive tubules contained apoptotic metaphase cells, although the majority of TUNEL-positive tubules (75%) contained apoptotic cells of a morphology consistent with zygotene/pachytene spermatocytes (Figure 2E, arrows, center panel). Spo11-Cre Hus1 CKO animals, in which Hus1 inactivation occurred later in germ cell development, also exhibited a decrease in testis cellularity and an increase in both multinucleate spermatid giant cells and TUNEL-positive cells (Figure S5). Altogether, these data indicate a requirement for Hus1 and the 9-1-1 complex for germ cell maintenance and development.
Because the 9-1-1 complex protects chromosomal integrity in somatic cells, we hypothesized that germ cell loss in Hus1 CKOs might be due to DNA damage accumulation. We prepared meiotic chromosome spreads and assessed the localization of γH2AX, the phosphorylated form of the histone variant H2AX, by indirect immunofluorescence. γH2AX marks DSBs as well as unsynapsed chromatin, and is phosphorylated by the ATM and ATR kinases [26], [29]. We found that γH2AX staining during leptotene and zygotene, when meiotic DSBs are generated and processed, was similar in Hus1 CKOs and controls. Furthermore, γH2AX localized normally to the sex body domain in pachytene Hus1 CKO spermatocytes (Figure 3), suggesting that ATR-dependent H2AX phosphorylation is independent of HUS1 as is also true during responses to replication stress [51]. However, in Hus1 mutant spermatocytes, the sex body domain marked by γH2AX was enlarged and extended, and often encompassed parts of or entire autosomes (Figure 3B, 3C, 3F). 56% of mutant pachytene spermatocytes showed some type of sex body abnormality (compared to 13% of control nuclei), with 39% of Stra8-Cre Hus1 CKO nuclei having sex body extensions or protrusions and 33% with autosome inclusion (compared to 4% and 9%, respectively, in control spermatocytes; Table 2). Similar phenotypes were also observed in Spo11-Cre Hus1 CKO spermatocytes (Table 2). Autosomes included in the sex body domain often appeared synapsed (based on chromosome number and SYCP3 staining intensity), though asynapsed chromosomes and parts of chromosomes were also included in the sex body, as shown in Figure 3F. Additionally, a substantial number of meiotic nuclei harbored apparent end-to-end fusions between the X chromosome and an autosome (Figure 3E). In such cases, the X and Y chromosomes were synapsed at the pseudoautosomal region (PAR) and were contained within the sex body domain containing γH2AX, while the autosome remained outside of the sex body domain and was devoid of γH2AX. The frequency of X-autosome fusions was significantly increased in Stra8-Cre Hus1 CKO pachytene spermatocytes, from 2% in controls to 12% in CKO mice (p<0.001; Table 2).
Extended sex body domains marked by γH2AX persisted in diplotene-stage Stra8-Cre Hus1 CKOs (25% of Hus1 CKO nuclei versus 6% of controls; Table 2), indicating that these cells did not arrest at the pachytene/diplotene transition despite chromatin abnormalities. Additionally, the sex body domain in both pachytene and diplotene spermatocytes was located at the nuclear periphery less often in both Stra8-Cre and Spo11-Cre Hus1 CKOs than in controls (Table 2), indicating a perturbation in the normal compartmentalization or localization of the sex chromatin. Although we cannot entirely rule out the possibility that the 9-1-1 complex directly promotes sex body maintenance through stimulation of ATR activity or other roles, we favor the idea that defects in maintenance of sex body integrity following Hus1 loss could instead be related to aberrant responses to meiotic DSBs. In cases where autosomes were asynapsed, clouds of γH2AX surrounded the unsynapsed regions (Figure 3D), supporting the idea that ATR-mediated responses to unsynapsed chromatin remain intact despite Hus1 loss. Consistent with the idea that expanded sex body domains marked by γH2AX reflect upregulation of ATR kinase activity, we also observed that Hus1 loss was associated with increased levels of both total and phosphorylated CHK1 (Figure 3G and Figure S3), the latter being an established indicator of ATR activity in somatic cells [52]. Increased phosphorylated CHK1 was also observed in extracts prepared from Spo11-deficient testes (Figure 3G), which have severe defects in chromosome synapsis.
A significant percentage of pachytene Stra8-Cre Hus1 CKO nuclei exhibited γH2AX staining on fully synapsed autosomes (39% versus 15%, p<0.001; Table 2), generally in a burst-like pattern perpendicular to the synaptonemal complex, indicative of unrepaired DSBs (Figure 3B). This eruption-like pattern of γH2AX staining on Hus1 CKO cores is consistent with previously described γH2AX L-foci [53], which are proposed to be sites of delayed or unregulated DSB repair events, and have been observed in several other meiotic mutants with break repair defects [54]–[56]. A similar pattern of γH2AX staining was observed in Spo11-Cre- and Stra8-Cre-driven Hus1 CKO mice (Table 2), suggesting that many of the persistent breaks originate during meiosis and are not due exclusively to aberrant pre-meiotic replication in the absence of Hus1. The autosomal γH2AX staining in Hus1 CKOs persisted into diplotene (Figure 3C), with 35% of Hus1 mutant nuclei containing bursts of autosomal γH2AX at this stage compared to 4% of control nuclei (Table 2). Persistence of these γH2AX foci into diplotene indicates that homologous chromosomes began to desynapse despite incomplete DSB repair, and that these aberrant spermatocytes did not arrest in the pachytene stage in the absence of Hus1. Consistent with the possibility of impaired DSB-induced DNA damage checkpoint signaling following Hus1 loss, we observed that the phosphorylated form of the DSB-responsive checkpoint kinase CHK2, which was evident in extracts from control testes, was absent in those from adult Hus1 CKO mice (Figure 3H and Figure S3B).
Since increased numbers of γH2AX foci were present on Hus1 CKO meiotic chromosomes, we further investigated whether the RAD51 strand exchange protein accumulated normally on Hus1-deficient chromosomes. During the early stages of meiotic DSB repair in zygotene and early pachytene spermatocytes, RAD51 properly localized to punctate structures along meiotic chromosome cores in the absence of Hus1 (Figure 4A). By late pachytene, when most RAD51 foci are normally lost from autosomal and XY chromosome cores, all Hus1 CKO spermatocytes exhibited persistent localization of RAD51 to the autosomes as well as the X chromosome, indicative of a DSB repair defect (Figure 4C). These foci persisted into diplotene as well (Figure 4D), demonstrating that nuclei with unrepaired DSBs were also able to progress beyond the pachytene stage. RAD51 focus numbers were quantified at each of these stages (Figure 4F–4I), and the number of foci in Hus1 CKO nuclei was significantly increased in late pachytene and diplotene. While the number of persistent RAD51 foci per nucleus was modest, the total number of nuclei with at least one RAD51 focus was high in Hus1 CKOs, at 100% of Hus1 CKO nuclei versus 52% of controls in late pachytene, and 73% of Hus1 CKOs versus 17% of controls in diplotene (Figure 4). Further, the reduction in the frequency of cells with RAD51 foci from 100% to 73% of Hus1 CKO nuclei from late pachytene to diplotene could indicate that Hus1 mutants can repair DSBs, albeit inefficiently in some cases, such that repair is significantly delayed. We conclude that HUS1 is critical for efficient completion of DNA repair at a subset of meiotic DSBs.
To determine whether Hus1 loss alters later stages in DSB repair such as crossover formation, we assessed MLH1 localization in pachytene spermatocytes. The mismatch repair complex MLH1-MLH3 is required for the formation of Class I crossover events, which account for approximately 90% of all meiotic crossovers in mice [57], [58]. Mouse and human RAD9 proteins have been shown to interact with MLH1 [21], although it is unclear whether this interaction is relevant for meiotic crossover recombination or whether it is limited primarily to post-replicative DNA mismatch repair. As shown in Figure S6A, the total number of MLH1 foci per nucleus was not significantly different in Hus1 CKO cells relative to controls, with one to two MLH1 foci per homolog pair. Hus1 CKOs exhibited a mean of 25.4±0.3 MLH1 foci per nucleus, compared to 25.9±0.3 in controls (Figure S6B), suggesting that Hus1 loss does not significantly impair crossover formation. However, a general impairment of meiotic DSB repair in the absence of functional 9-1-1 complex might manifest as only a modest reduction in the number of MLH1 foci per nucleus, as 10% of crossovers are not dependent on MLH1-MLH3 and the majority of meiotic DSBs are repaired as noncrossovers [59]. To determine whether the persistent RAD51 foci observed in Hus1 CKOs were primarily associated with noncrossover or crossover events, we assessed the relative localization of RAD51 and MLH1. As shown in Figure S6C, the majority of persistent RAD51 foci in late pachytene were exclusive of MLH1, though an occasional single RAD51 focus localized at or adjacent to an MLH1 focus. These data indicate that while some meiotic DSBs are not repaired efficiently in the absence of Hus1, these unrepaired breaks consist primarily of MLH1-independent repair events (most likely noncrossovers) but can include MLH1-dependent crossovers.
Since a significant proportion of TUNEL-positive Hus1 CKO germ cells appeared to be in meiotic metaphase, it remained possible that Hus1 CKOs might be deficient in crossover formation, despite the normal complement of MLH1 foci we observed. Thus, we prepared meiotic diakinesis chromosome spreads from control and Hus1 CKO testes to assess chromosome integrity at the end of prophase I, when the SC has disintegrated and chiasmata are first evident. As shown in Figure S6D–S6E, the majority of nuclei had 20 bivalents, indicating that each homolog pair received at least the one obligate crossover necessary for proper chromosome segregation. Taken together, these results indicate that while HUS1 is critical for repair of a subset of meiotic DSBs, it appears to be dispensable for crossover formation.
To investigate additional causes for germ cell loss in Hus1 CKOs, we assessed the localization of various meiotic markers on chromosome spreads from control and Stra8-Cre Hus1 CKO males. Meiotic chromosome synapsis appeared normal in most Hus1 mutant nuclei, with colocalization of the synaptonemal complex proteins SYCP1 and SYCP3 in pachytene spermatocytes and the presence of 19 pairs of fully synapsed homologs in addition to the X and Y chromosomes paired at the pseudoautosomal region. However, as shown in Figure 5B, a significant number of Stra8-Cre Hus1 CKO nuclei exhibited synapsis defects, usually involving the X chromosome (14% of Hus1 CKOs versus 3% of controls; N = 161 and 185, respectively; p<0.001). Notably, Hus1 CKOs also frequently displayed ruptures in the synaptonemal complex during diplotene (36% of Stra8-Cre Hus1 CKOs and 33% of Spo11-Cre Hus1 CKOs compared to 2–6% of controls; Table 2). In these cases, there were clear discontinuities in synaptonemal complex protein staining and the ends of the broken SC were spatially separated (Figure 5C, arrows), indicating either a defect in SC integrity or breakage of chromosomal DNA in the absence of HUS1.
During meiosis, TOPBP1 normally colocalizes with ATR at sites of unsynapsed chromatin, including unsynapsed autosomes in early spermatocytes as well as the XY chromatin throughout prophase I [24], [60]. In mammalian somatic cells, the best characterized function of 9-1-1 is to recruit TOPBP1 to sites of DNA damage, where it physically interacts with and stimulates the kinase activity of ATR [61]–[63]. Therefore, we next tested whether Hus1 loss affected the meiotic localization of TOPBP1. In contrast to the 9-1-1-dependent mechanism elucidated in somatic cells, TOPBP1 localization to the asynapsed sex chromatin was unperturbed in Hus1 CKO mice (Figure 6). Additionally, TOPBP1 localized to autosomes that resided within the sex body domain (Figure 6B, asterisk), and to perpendicular foci on some autosomes in Stra8-Cre Hus1 CKOs (Figure 6C) similar to the γH2AX eruptions described above, indicating that HUS1 loss does not preclude assembly of TOPBP1 on chromatin and results in TOPBP1 localization to abnormal autosomes as well as to unsynapsed sex chromatin.
Because the downstream consequence of ATR action at asynapsed meiotic chromatin is meiotic silencing [30], [32], we also assessed the localization of RNA Polymerase II, which indicates sites of active transcription. Consistent with the apparently normal TOPBP1 localization following Hus1 deletion in germ cells, RNA Pol II immunofluorescence staining revealed no difference in transcriptional silencing of the XY chromatin, as indicated by a similar lack of RNA Pol II signal in the sex body domain of cells from both Hus1 CKO and control mice (Figure 6D, 6E). Together, these results indicate that despite the absence of the HUS1 component of the 9-1-1 complex in Hus1 CKO spermatocytes, TOPBP1- and ATR-dependent responses to unsynapsed chromatin remained intact. These findings raise the possibility that ATR activation in response to at least some signals during mammalian meiosis, such as unsynapsed chromatin, may occur through a distinct mechanism independent of the canonical 9-1-1 complex.
Since Hus1 disruption resulted in chromosomal defects without affecting TOPBP1 localization, we next analyzed RAD9 localization to meiotic chromosomes to gain insights into the normal functions of 9-1-1. As shown in Figure 7A–7C, RAD9 localized to meiotic chromosome cores during prophase I, with many foci on both unsynapsed and synapsed chromosomes during zygotene stage, fewer autosomal foci and prominent X chromosome foci in early pachytene, and even fewer foci by mid- to late-pachytene stage when most staining was confined to bright foci along the X chromosome core. By diplotene stage, RAD9 foci were not detectable on meiotic chromosomes. Thus, the RAD9 subunit of 9-1-1 is appropriately positioned for a potential role in genome maintenance during meiosis.
The pattern and timing of the RAD9 signal on meiotic chromosomes was reminiscent of that of the DSB repair factor RAD51, particularly with respect to its presence at late-persisting breaks along the X chromosome core. To determine whether RAD9 colocalized with RAD51, we simultaneously stained meiotic chromosomes for RAD9 and RAD51. In pachytene-like nuclei, the majority of RAD9 foci colocalized with RAD51 foci, most notably along what is presumably the X chromosome axial element (Figure 7D). Thus, RAD9 appears to localize to DSBs undergoing homologous recombinational repair. To further assess RAD9 function in DSB repair, we also analyzed RAD9 localization in Spo11−/− mutants [64], which lack meiotic DSBs, and Dmc1−/− mutants [65], which harbor persistent unrepaired DSBs. As shown in Figure S7, RAD9 staining was much reduced in the absence of meiotic DSBs and elevated in the presence of increased DSBs. Thus, RAD9, and likely its partner HUS1 (see below), function together at meiotic DSBs at a stage overlapping with RAD51/DMC1.
The RAD1 subunit of 9-1-1 was previously determined to localize to meiotic chromosomes, where RAD1 foci outnumber RAD51 foci and nearly continuously coat pachytene chromosome cores, especially along the unsynapsed X and Y [41]. To reconcile this staining with our observations of fewer, more punctate RAD9 foci, we assessed the colocalization of RAD1 and RAD9 subunits in control nuclei. These experiments revealed that RAD1 had a broader distribution than RAD9, with RAD1 foci present along the entire unsynapsed regions of the X and Y while RAD9 was detected primarily in bright foci along the X. Overall, only 23% of RAD1 foci colocalized with RAD9 (16% of autosomal and 46% of XY RAD1 foci colocalized with RAD9). A larger percentage (64%) of RAD9 foci colocalized with RAD1, particularly on the sex chromosomes (52% of autosomal and 86% of XY RAD9 foci colocalized with RAD1). We next assessed both RAD1 and RAD9 localization on abnormal chromosomes in Slx4mut/mut and Msh4−/− mutant spermatocytes, which exhibit asynapsis as well as inclusion of autosomes within the sex body domain [55], [66]. Consistent with the co-staining results, RAD1 localized along synapsed autosomal cores and to a greater extent along asynapsed autosomal and XY cores (Figure 8B, 8D). RAD9 on the other hand localized in a more punctate pattern along chromosome cores in the sex body domain and on asynapsed autosomes (Figure 8C, 8E). Together, these results indicate that RAD1 and RAD9 partially colocalize along abnormal chromosomes and the sex chromosome cores, and that a subset of RAD1-containing sites lack RAD9.
Prompted by the differences in RAD1 and RAD9 localization described above, we next assessed localization of these 9-1-1 subunits in Hus1 CKO spermatocytes. Since the chromatin association and nuclear localization of the 9-1-1 complex in somatic cells requires all three subunits [48], we expected RAD9 and RAD1 to be depleted from chromosomes in the absence of HUS1. However, the more extensive RAD1 staining observed in normal and aberrant spermatocytes suggested a possible separation of functions for the individual subunits. Remarkably, bright RAD1 staining of XY cores as well as punctate staining of autosomes persisted upon Hus1 deletion (Figure 9A). By contrast, RAD9 foci were undetectable in nuclei from Hus1 CKO mice (Figure 9B). Whereas RAD9 staining was readily detected in all control nuclei, approximately 99% of Stra8-Cre Hus1 CKO pachytene nuclei lacked RAD9 foci. The continued presence of RAD1 on meiotic chromosomes following Hus1 loss despite the significant reduction in total RAD1 levels (Figure S3) may indicate that a fraction of RAD1 protein participates in chromatin-associated complexes that remain stable independently of HUS1, while the remaining pool of RAD1 is unstable in the absence of HUS1.
The occasional occurrence of asynapsed homologs and the presence of autosomes within the sex body domain of pachytene Hus1 CKO cells additionally allowed for evaluation of the dependency on Hus1 for RAD9 and RAD1 recruitment to abnormal structures during meiosis. Similar to our observation that the localization of TOPBP1 was unperturbed in Hus1 CKOs, RAD1 clearly localized to autosomes within the vicinity of the sex body domain (Figure 9C) and to asynapsed autosomes (Figure 9D) in the absence of Hus1. The localization of RAD9 on the other hand appeared strictly Hus1-dependent, as RAD9 was not detected on asynapsed chromosomes or sex chromosome cores in Hus1 CKOs (Figure 9E), although it did localize to such structures in Hus1-expressing cells (Figure 7 and Figure 8). Together with the results described above, these data indicate that RAD9 and RAD1 have overlapping Hus1-dependent functions that intersect with RAD51, and that RAD1 has additional meiotic functions that may involve TOPBP1 but are independent of its traditional binding partners, HUS1 and RAD9.
In order to examine the role of the 9-1-1 complex during mammalian meiosis, we analyzed mice with germ cell-specific Hus1 deletion and evaluated the localization of the RAD9 and RAD1 subunits to normal and aberrant meiotic chromosomes. Most notably, we report that the RAD9 and RAD1 subunits of the 9-1-1 complex exhibit overlapping yet distinct localization patterns along mammalian meiotic chromosomes and are differentially affected by loss of the HUS1 subunit. RAD1 forms more foci along chromosome cores than RAD9 (this study) or RAD51 [41], and we show that it preferentially localizes to the axial elements of asynapsed chromosomes or autosomes included within the sex body domain in a manner that is independent of Hus1. Previous electron microscopy analysis indicated that while RAD1 localizes to meiotic chromosomes cores, it does not colocalize with DMC1 [41], which is consistent with our finding of RAD1 at asynaptic sites. We show here that only a subset of RAD1 foci colocalize with RAD9, raising the possibility that RAD1 is present both at DSBs, as part of the canonical 9-1-1 complex, and at other chromosomal sites.
We propose a model in which meiotic genome maintenance involves several distinct checkpoint clamp complexes, some of which function in DSB repair, and others of which carry out separate functions, possibly including ATR activation through TOPBP1 in response to asynapsis. Of particular note are the previous findings that the mouse and human genomes contain paralogs of RAD9 and HUS1, termed RAD9B and HUS1B, respectively, that are highly expressed in the testis [67], [68]. Given that RAD1 localized to some chromosomal sites that lacked RAD9 and did so following genetic ablation of Hus1, it is tempting to speculate that RAD9B and HUS1B associate with RAD1 to form an alternative 9-1-1-like heterotrimer, although it remains possible that RAD1 localizes to chromosomes on its own or in conjunction with other cofactors. In this regard, it is worth noting that a substantial portion of RAD1 protein in cultured human cells exists in monomeric form, independent of the heterotrimeric 9-1-1 complex [69]. Overall, our results are consistent with formation of the canonical 9-1-1 complex composed of RAD9, RAD1, and HUS1 at meiotic DSBs, as well as with RAD1 assembling on chromosomes independently of HUS1 and RAD9 in patterns overlapping with TOPBP1 and ATR.
Despite a well-established role for 9-1-1 in ATR activation in somatic cells, we observed unperturbed localization of the 9-1-1-interacting, ATR-activating TOPBP1 protein and similarly no change in RNA Polymerase II exclusion from the XY body domain in Hus1 CKO spermatocytes. These findings indicate that ATR-dependent responses to unsynapsed chromatin, including meiotic silencing via MSUC and MSCI, do not require HUS1, which is in distinct contrast to the canonical model of mammalian ATR activation in which 9-1-1-mediated TOPBP1 recruitment to damaged DNA allows TOPBP1 to contact and activate ATR. Since RAD1 localization to asynapsed meiotic cores was also unperturbed in the absence of Hus1, it is possible that a RAD1-containing complex functions along with TOPBP1 to activate ATR at such sites, in a manner that may be at least in part distinct from established mechanisms for 9-1-1 loading and function at DNA damage sites. The meiosis-specific HORMAD1 and HORMAD2 proteins are required for ATR recruitment to sites of unsynapsed chromatin [33], [70], and together with BRCA1, which is also required for ATR recruitment [26], these proteins could provide an alternative mechanism for ATR activation that is independent of 9-1-1 and typical substrates for 9-1-1 loading. We favor the idea that alternative complexes involving RAD1 and possibly RAD9B and HUS1B might engage in a 9-1-1-like interaction with TOPBP1. It remains to be determined whether the recruitment of such a complex to asynaptic sites is facilitated by particular DNA structures or binding partners that would not attract conventional 9-1-1 complexes.
Conditional Hus1 deletion in germ cells resulted in numerous chromosomal defects, including asynapsis, unrepaired meiotic DSBs, X-autosome end-to-end fusions, autosomes incorporated into the sex body domain (which presumably are inappropriately transcriptionally silenced), and SC ruptures. We propose that some of these defects (asynapsed chromosomes, X-autosome fusions, and autosomes incorporated into the sex body domain) lead to loss of cells during the pachytene stage in Hus1 CKOs, consistent with an absence of cells with such abnormalities in diplotene (Table 2), whereas other defects (unrepaired DSBs, extended sex body domains, and SC ruptures; Table 2) either may not be actively monitored or may normally require HUS1 for checkpoint-dependent clearance and therefore persist beyond pachytene into diplotene in Stra8-Cre Hus1 CKOs. The latter set of phenotypes was less prominent in Spo11-Cre Hus1 CKO mice, with fewer defective cells persisting into diplotene, suggesting that HUS1-dependent checkpoint monitoring may be an earlier function that is less affected by mid-prophase I loss of Hus1. Notably, SC ruptures were still prominent in diplotene cells of both Stra8-Cre and Spo11-Cre Hus1 CKOs, affecting one-third of the diplotene cell population. Thus, HUS1 is required for some aspect of chromosome integrity that affects the SC, the failure of which may contribute to the high level of metaphase germ cell loss. This SC rupture phenotype (Figure 5C; Table 2) is particularly striking and may be a key driver of the fertility defect observed following Hus1 loss, as no significant change was observed in the number of MLH1 foci in Hus1 mutant pachytene spermatocytes or in the number of bivalents seen at diakinesis (Figure S6). The molecular cause of the SC rupture defect remains unresolved, although it could be related to incomplete repair of meiotic DSBs.
It is still unclear why Hus1 CKO cells have a small number of persistent DSBs, and why many of these are associated with death at metaphase I. Perhaps the 4–8 remaining unrepaired breaks are not sufficient to trigger the pachytene checkpoint (or cannot, due to a requirement for HUS1), but lead to a defect that is recognized at metaphase by the spindle checkpoint. It is possible that the persistent breaks are not simple D-loop intermediates containing RAD51 and γH2AX, but are complex multi-chromatid intermediates, such as those seen in Blm/Sgs1 helicase mutants [71], [72]. The symmetrical RAD51 foci seen on either side of the chromosome cores in 9% of diplotene nuclei from Hus1 CKOs (see Figure 4, asterisks) would support this, pointing to RAD51/DMC1 presence on more than one homolog during homolog separation. Clearly, the extent of the DNA repair defect in Hus1 CKO mice is less severe than that observed in other meiotic mutants. For instance, Trip13 mutant spermatocytes have been reported to exhibit between 99 and 138 RAD51 foci during pachytene compared to 11–18 foci in controls [54], [73], whereas we observed an average of 25 RAD51 foci in early to mid-pachytene Hus1 CKOs (compared to 22 foci in controls) and 9 RAD51 foci in late pachytene (compared to 2 foci/nucleus in controls). These results are consistent with the idea that HUS1 has a relatively late and restricted role in DSB repair, perhaps related to the completion of a subset of late recombination events.
The persistent γH2AX and RAD51 foci on Hus1-deficient meiotic chromosomes as well as the colocalization of RAD9 with RAD51 on normal chromosomes suggest that the 9-1-1 complex is critical for the efficient repair of a subset of meiotic DSBs. Several lines of evidence in somatic cells also indicate that 9-1-1 functions directly in homologous recombination (HR) repair of DSBs. In human cells, 9-1-1 is reported to physically interact with RAD51, and Rad9 knockdown results in reduced HR [7]. In a separate system using human cells with conditional Rad9 repression, RAD9 enhances survival and DNA repair in response to ionizing radiation [74], and similarly, mouse Rad9−/− ES cells are sensitive to γ-irradiation [43]. Reducing Hus1 expression in mouse cells via siRNA also decreases the efficiency of HR repair [75]. In contrast to the situation in yeast where 9-1-1 is proposed to be important for the loading or assembly of RAD51 complexes onto meiotic chromosomes [37], we propose that mammalian 9-1-1 may be important for later steps of meiotic recombination, since we did not observe delayed loading of RAD51, and RAD51 foci persisted in the absence of Hus1. Among the known 9-1-1 binding partners are DNA ligase I [13] and DNA Polymerase β [9], [10], the latter of which is known to play critical roles during mammalian meiosis [76], [77], raising the possibility that 9-1-1 may recruit Pol β and DNA ligase to recombination intermediates to complete repair. Alternatively, interactions between the 9-1-1 complex and translesion synthesis polymerases [8], which have been implicated in HR in somatic cells [78]–[80], could promote the extension of the 3′ ends subsequent to RAD51-mediated strand exchange. The requirement for 9-1-1 could be accentuated due to unique features of genome maintenance in meiotic cells. For instance, non-homologous end joining, a major mechanism for DSB repair in somatic cells, is suppressed during meiosis [29], [81]. In addition, the unique structure of SPO11-induced meiotic DSBs may create a greater demand for 9-1-1 complex-mediated repair functions than a typical mitotic DSB. The repair of most meiotic DSBs occurred normally in the absence of HUS1, suggesting that the 9-1-1 complex may be necessary to deal with only a subset of breaks, perhaps ones that prove difficult to repair because of the sequence context, chromatin structure, or other factors.
The production of haploid gametes during meiosis clearly raises challenges for genome maintenance, many of which are distinct from those in somatic cells. Based on the findings reported here, we propose that mammalian 9-1-1 components have acquired specialized roles during meiosis, with the canonical RAD9-RAD1-HUS1 complex functioning at DSBs and an alternative RAD1-containing complex functioning at sites of asynapsis. The mouse models described here represent powerful systems to elucidate how the mammalian 9-1-1 complex promotes meiotic chromosome integrity, in some cases distinct from the well-established roles of 9-1-1 in TOPBP1- and ATR-dependent checkpoint signaling, and highlight the intriguing possibility of alternative checkpoint clamps functioning in various capacities in the mammalian germline.
All animals used in this study were handled in accordance with federal and institutional guidelines, under a protocol approved by the Cornell University Institutional Animal Care and Use Committee (IACUC).
Mice harboring two conditional Hus1 alleles (Hus1flox/flox) were crossed to Cre-positive mice harboring one null Hus1 allele (Stra8-Cre+ Hus1+/Δ1 or Spo11-Cre+ Hus1+/Δ1) to generate experimental germ cell-specific Hus1 conditional knockout mice (Cre+ Hus1flox/Δ1) as well as littermate control animals (Cre+ Hus1+/flox and Cre- Hus1flox/Δ1), as shown in Figure S1A. Conditional and null Hus1 alleles were described previously [46]. Stra8-Cre transgenic FVB mice were kindly provided by Bob Braun (The Jackson Laboratory; [47]). Spo11-Cre mice were generated as described in Figure S2 and in Text S1. Msh4−/− mice were kindly provided by Winfried Edelmann (Albert Einstein College of Medicine; [66]), and Dmc1−/− [65] and Spo11−/− [64] mutant mice were generously provided by John Schimenti (Cornell University). Slx4mut/mut mice were derived from the previously reported Btbd12tm1a(EUCOMM)Wtsi strain and carried an intact β-geo cassette and germline Slx4 exon 3 deletion [55]. For fertility testing, 8- to 12-week old Stra8-Cre Hus1 CKO and control males were singly housed with wild-type 129 or FVB females. Copulatory plugs were monitored daily, and plugged females were removed to separate cages and monitored for pregnancy. Viable pups were counted on the first day of life.
Flash frozen testes from 17-day old, 20-day old, or adult animals (12–14 weeks, unless labeled otherwise) were homogenized in RIPA buffer supplemented with protease inhibitors and sodium orthovanadate using a Tissuelyzer, sonicated at 24–27W twice for 2 minutes each, then cleared by centrifugation. Antibodies included rabbit polyclonal anti-HUS1 HM199 (rabbit polyclonal antiserum generated against purified recombinant GST-tagged full-length mouse HUS1 protein), anti-phosphoCHK1 (Ser345; Cell Signaling #2341), anti-CHK1 (Santa Cruz), anti-CHK2 (clone 7, Millipore #05-649), anti-RAD9 HM456 (rabbit polyclonal antiserum generated against purified recombinant HIS-tagged full-length mouse RAD9 protein), affinity purified rabbit anti-RAD1 [41], and anti-β-actin (Sigma).
Testes were fixed overnight at 4°C in Bouin's fixative (for H&E and GCNA1 staining) or at room temperature in 10% neutral-buffered formalin (for TUNEL staining), embedded in wax, and sectioned at 5 µm. Immunohistochemical staining for germ cell nuclear antigen [GCNA1; 82] was performed using anti-GCNA1 antibody provided by George Enders. TUNEL assays were performed using the Apoptag kit (Millipore) as per the manufacturer's instructions. TUNEL data were quantified by counting the number of TUNEL-positive cells per tubule in at least 50 tubules from the testes of at least 3 different mice of each genotype, and differences between controls and Hus1 CKOs were analyzed statistically via Student's t-test.
Both caudal epididymides from each 12-week old mouse were minced with fine forceps in a petri dish with 37°C PBS, incubated, and fixed in 10% neutral-buffered formalin (1∶25 dilution). Sperm counts shown in Figure 1 are the mean of 6 to 9 mice per group ± SEM, analyzed statistically using a Student's t-test.
Surface-spread nuclei were prepared from 12-week old male mice as described previously [58], with the exception of Dmc1−/− and control littermate samples which were prepared similarly from flash-frozen testes. Briefly, tubules were incubated in hypotonic extraction buffer on ice for 1 hour, minced in 100 mM sucrose, and spread on slides dipped in 1% PFA with 0.15% Triton X-100. Slides were incubated in a humid chamber for 2.5 hours, dried, and washed in PBS and water containing Photoflo (Kodak). Immunofluorescence staining was performed following blocking in 10% goat or donkey serum and 3% BSA, with primary antibodies incubated overnight at room temperature and secondary antibodies incubated at 37°C for one hour in the dark. Slides were mounted with coverslips using homemade anti-fade mounting medium (2.3% DABCO, 20 mM Tris pH 8.0, 8 µg DAPI in 90% glycerol).
Primary antibodies used for immunofluorescence staining included those recognizing: γH2AX (1∶5000; Upstate/Millipore), SYCP1 (1∶500), SYCP3 (1∶5000; [83]), RAD9 HM456 (1∶600; see above), RAD1 (1∶17; [41]), TOPBP1 (1∶500; [84]), RAD51 (1∶500; Oncogene Research Products/EMD Biosciences), RNA Polymerase II (1∶500; Millipore), and MLH1 (1∶50; BD Pharmingen). For co-staining of RAD9 and RAD1, we additionally used affinity purified sheep anti-RAD1 (sheep polyclonal antiserum generated against purified recombinant 6X HIS full-length human RAD1) generated in the Freire laboratory, and for co-staining of RAD9 with RAD51, we used mouse monoclonal anti-RAD51 (Abcam). For MLH1/SYCP3/RAD51 co-staining, anti-SYCP3 antibody was used at 1∶50,000 as described Lipkin et al. [85]. Human CREST serum was used to detect centromeres in some experiments as previously described [86]. Secondary antibodies were used at 1∶1000 dilution and included goat anti-mouse Alexafluor 488, goat anti-rabbit Alexafluor 555, donkey anti-sheep Alexafluor 488, and donkey anti-rabbit Alexafluor 555 (Invitrogen). Microscopy and imaging was performed as described previously [56].
For quantification of phenotypes shown in Table 2, a “grossly normal” nucleus was defined as one with normal synapsis and γH2AX staining confined to the sex body, and an “abnormal sex body” was defined as one with extended γH2AX signal, inclusion of whole or partial autosomes, and/or the presence of an apparent X chromosome-autosome end-to-end fusion. P-values for differences in γH2AX staining (Table 2), sex body abnormalities (Table 2), and synapsis defects (text) were calculated using a z-test of sample proportions, comparing Hus1 CKOs to control animals. Diakinesis chromosome spreads (Figure S6D) were generated as described previously [56] from two independent 4- to 5-week old mice per genotype.
|
10.1371/journal.pcbi.1005082 | Leukocyte Motility Models Assessed through Simulation and Multi-objective Optimization-Based Model Selection | The advent of two-photon microscopy now reveals unprecedented, detailed spatio-temporal data on cellular motility and interactions in vivo. Understanding cellular motility patterns is key to gaining insight into the development and possible manipulation of the immune response. Computational simulation has become an established technique for understanding immune processes and evaluating hypotheses in the context of experimental data, and there is clear scope to integrate microscopy-informed motility dynamics. However, determining which motility model best reflects in vivo motility is non-trivial: 3D motility is an intricate process requiring several metrics to characterize. This complicates model selection and parameterization, which must be performed against several metrics simultaneously. Here we evaluate Brownian motion, Lévy walk and several correlated random walks (CRWs) against the motility dynamics of neutrophils and lymph node T cells under inflammatory conditions by simultaneously considering cellular translational and turn speeds, and meandering indices. Heterogeneous cells exhibiting a continuum of inherent translational speeds and directionalities comprise both datasets, a feature significantly improving capture of in vivo motility when simulated as a CRW. Furthermore, translational and turn speeds are inversely correlated, and the corresponding CRW simulation again improves capture of our in vivo data, albeit to a lesser extent. In contrast, Brownian motion poorly reflects our data. Lévy walk is competitive in capturing some aspects of neutrophil motility, but T cell directional persistence only, therein highlighting the importance of evaluating models against several motility metrics simultaneously. This we achieve through novel application of multi-objective optimization, wherein each model is independently implemented and then parameterized to identify optimal trade-offs in performance against each metric. The resultant Pareto fronts of optimal solutions are directly contrasted to identify models best capturing in vivo dynamics, a technique that can aid model selection more generally. Our technique robustly determines our cell populations’ motility strategies, and paves the way for simulations that incorporate accurate immune cell motility dynamics.
| Advances in imaging technology allow investigators to monitor the movements and interactions of immune cells in a live animal, processes essential to understanding and manipulating how an immune response is generated. T cells in the brains of Toxoplasma gondii-infected mice have previously been described as performing a Lévy walk, an optimal strategy for locating sparsely, randomly distributed targets. Determining which motility model best characterizes a population of cells is problematic; multiple metrics are required to specify as intricate and nuanced a process as 3D motility, and the tools to evaluate model-parameter combinations have been lacking. We have developed a novel framework to perform this model evaluation through simulation, a popular tool for exploring complex immune system phenomena. We find that Lévy walk offers an inferior capture of our data to another class of motility model, the correlated random walk, and this determination was possible because we are able to explicitly evaluate several motility metrics simultaneously. Further, we find evidence that leukocytes differ in their inherent translational and rotational speeds. These findings facilitate more accurate immune system simulations aimed at unravelling the processes underpinning this critical biological function.
| Cellular motility and interactions underlie many processes in the immune response, including lymphocyte recirculation through blood and lymphoid organs, their interactions with cells presenting specific antigen, and relocation to the specific tissues where they engage in protective immunity [1]. In the last decade, two-photon microscopy has provided unprecedented insight into how immune cells move and interact in vivo [1, 2]. Parallel to this, computational modeling and simulation techniques have been applied to exploring hypotheses of immune system function [3, 4], even simulating the effects of interventions [5, 6].
Agent-based simulations (ABS), wherein individual immune cells are simulated as discrete entities with their own state in a spatially explicit environment, have found widespread application in immunology, with far-ranging applications including: understanding granuloma development [7], Payers patch development [8], the search efficiency of lymphocytes in the lymph node [9, 10], the establishment and subsequent recovery from autoimmune disease [5], and the mechanisms underlying cancer [11]. There is clear scope to combine detailed spatio-temporal two-photon microscopy data with spatially-explicit agent-based simulation to further understanding of how cellular motility integrates with other immune processes to impact health.
An established body of research in ecology has demonstrated, however, the complexities of determining which models of motility best characterize a given dataset. In the Lévy walk model, an agent’s motility is described by a sequence of randomly oriented straight line movements drawn from a power-law, long-tailed distribution [12]. Hence, agent motilities are characterized by many relatively short movements punctuated by rare, very long movements. A diverse range of organisms have been described as exhibiting Lévy walk motility, including bacteria, honey bees, fruit flies, albatrosses, spider monkeys, and sharks [13, 14]. T cells in the brains of Toxoplasma gondii-infected mice have also been shown to perform a Lévy walk [15]. Interest in the Lévy walk is in part due to theoretical work demonstrating it an optimal strategy for finding sparsely, randomly distributed targets [16, 17]. However, subsequent work has cast doubt on Lévy walk’s apparent pervasiveness in nature, owing to methodological discrepancies in its identification [18, 19].
The spatial motility of agents in both two- and three-dimensions is an intricate and nuanced phenomenon that cannot be well specified using only one metric. The mean squared displacement over time metric is frequently used to differentiate Lévy walk and Brownian motion characteristics in a dataset, yet models differing in key aspects of motility can produce identical measures [20, 21], e.g., slow moving directionally persistent cells, or fast moving less-directional cells. To accurately simulate the motility dynamics of a biological dataset requires an appropriate motility model assigned appropriate parameter values, and evaluating putative parameter values requires simultaneous consideration of several complementary motility metrics.
Here we evaluate several random walk models’, including Brownian motion, Lévy walk, and several correlated random walks, capacities to capture the motility dynamics of lymph node T cells responding to inflammation and neutrophils responding to sterile laser injury of the ear pinnae. Each model is independently simulated, and those model parameter values that best align simulation and in vivo motility dynamics are determined through novel application of a multi-objective optimization (MOO) algorithm: NSGA-II [22]. Parameter estimation is performed through simultaneous consideration of three metrics of cell population motility: the distributions of translational and turn speeds observed across the population, and the distribution of meandering indices. The differences between simulation and in vivo distributions generated under each metric form objectives for the MOO algorithm. The resulting Pareto fronts generated under each model, representing parameter values delivering optimal trade-offs in performance against each metric, are contrasted to ascertain which model best captures the biology.
Our random walk models are designed following a detailed analysis of which statistical distributions best fit a cellular population’s translational and turn speed data. Such assessment is complicated by inherent biases in imaging experiments, wherein fast moving and directionally persistent cells rapidly leave the imaging volume. Hence, slower, less directional cells are over-represented in in vivo datasets. It is unclear whether cells observed to differ in directional persistence and translational speed are a result of these biases, or whether these observations represent fundamental differences in cellular motilities. Our novel analytical approach fits a given statistical distribution to a population’s pooled translational (or turn) speeds, whilst segregating observations drawn from the distribution into groups that correspond to tracks in the in vivo dataset. This segregation reproduces the imaging experiment biases, therein discounting their confounding influence on the analysis. We find that cells comprising our in vivo datasets are genuinely heterogeneous, differing in their inherent translational speed and directionality. This finding could reflect intrinsic cellular characteristics, or may arise as features of the environment through which they migrate. In subsequent analysis, we find that translational and turn speeds in both in vivo populations are significantly negatively correlated, indicating that cells do not simultaneously perform very fast translational movements and turns. To investigate the significance of these two observations on leukocyte motility we designed four correlated random walk models that differentially include (or exclude) each. We then simulate each to evaluate the integrative impact of these features on overall motility dynamics. We determine that Brownian motion poorly reflects both our datasets. Lévy walk competitively captures directional persistence, but performs poorly on translational and turn speed metrics, underscoring the value of considering several motility metrics simultaneously. Interestingly, for neutrophils Lévy walk provides the most even balance of metric trade-offs of any model examined. Both T cell and neutrophil motility dynamics were better captured by CRWs simulating cells as heterogeneous, rather than homogeneous, populations. Capture of T cell dynamics was further improved by negatively correlating simulated cell translational and turn speeds, however this was not as evident for neutrophil data.
We have provided here evidence, for the first time, that cells within both T cell and neutrophil populations exhibit a continuum of inherent directionalities and translational speeds. Further, we have shown that cells do not simultaneously perform very fast translational and turn movements. We have developed a novel framework to fit statistical distributions to cell translation and turn speeds whilst accounting for experimental bias. Thereafter, the manner in which these two components of motility combine to impact overall spatial exploration is analysed through a novel coupling of 3D agent-based simulation with multi-objective optimization. This latter framework for the first time calibrates and assesses putative motility models through simultaneous consideration of several motility metrics, accounting for trade-offs in performance against each. These frameworks provide the means to robustly analyse and accurately reproduce cellular motility patterns, as they explicitly reflect the constraints of in vivo data.
Our analysis and reproduction of leukocyte motility is performed in two stages. First, we analyse a given dataset’s cellular translation and turn speed dynamics separately. This stage does not attempt to reproduce cellular motility, which is performed later. Instead, it determines the extent to which observed heterogeneity in a cellular population, evidenced through tracks differing substantially in their median translation and turn speeds, is explained by imaging experiment bias, and which statistical distributions best fit this data. In the second stage we construct random walk models based on these distributions, and assess their capacity to reproduce leukocyte motility dynamics through agent-based simulation. In vivo data were obtained through two-photon microscopy of mouse lymphoid T cells in explanted lymph nodes in response to challenge and neutrophils in the mouse ear following sterile injury. The motility dynamics of our leukocyte datasets are characterised in S1 and S2 Figs.
We hypothesized that our T cell and neutrophil cellular populations were statistically heterogeneous, comprising cells differing in their inherent directionalities and translational speeds. Accordingly, we observed varying median track translational and turn speeds within both cellular populations, Fig 1A and 1B. These distributions could reflect genuinely heterogeneous features between cells, or could represent statistical sampling artifacts arising from finite cellular observation durations within a finite spatial volume. We quantified this experimental bias, Fig 1C, S3 and S1F Figs. Median track translation speed was strongly negatively correlated with the number of times the cell was observed in the imaging volume, and median track turn speed was strongly positively correlated with number of observations. Together these data indicate that fast cells moving in a highly directional fashion quickly left the imaging volume.
We sought to establish whether the perceived heterogeneous cellular characteristics (Fig 1A and 1B) represent a genuinely heterogeneous population, or arise from experimental bias, and which statistical distributions best describe these data. We devised a novel statistical approach to address this question (S4 Fig, Methods and S1 Algorithm), wherein observations are drawn from given statistical distribution and grouped. The groups reflect the structure in which the translational (or turn) speeds observed in a cellular population come from specific tracks. Hence, we could analyse all observations as one pooled dataset or extract the median values across the groups. This structure exactly matches that of the in vivo dataset being analysed, with the number of groups matching the number of tracks, and the number of observations within each group matching that of each track. Further, we impose similar correlations between the number of observations in each group and the median observation value, therein reflecting the experimental biases present in the in vivo dataset. This is done by establishing the maximum number of observations of any track in the in vivo dataset, and initially populating each group with the same number of observations. Thereafter we iterate through each track in the in vivo dataset, and select a group from which to discard data such that track and group share the same number of observations, and the correlations between median observation data and number of observations are similar.
This procedure is used to assess how well a given statistical distribution captures cellular translational (or turn) speed data, despite the experimental biases inherent in obtaining it. A successful capture must reproduce both the distribution of all translational (or turn) speeds pooled from all tracks (S1A and S1B Fig), and how these are allocated into tracks yielding the distribution of median track characteristics (Fig 1A and 1B). We assess a variety of statistical distributions, depicted in S5 Fig, including uniform, Lévy and Gaussian; these are termed ‘homogeneous’ as the same parameterized distribution is used to populate all groups with observations. We also assess a ‘heterogeneous’ Gaussian, wherein each group is populated by a bespoke Gaussian sub-distribution; hence, these groups are statistically heterogeneous, each is composed of observations drawn from a (potentially) unique Gaussian.
A given statistical distribution is first fitted to the in vivo dataset’s pooled translational speeds (or turn speeds, S1A and S1B Fig respectively), pooling all groups’ observations when performing the fitting. This is done 5 independent times for statistical rigour. We use each fitted solution to generate 100 datasets using the procedure outlined above, giving 500 datasets in total. For each of these, we contrast the groups’ median observation values with the tracks’ median translational (or turn) speed values using the Kolmogorov-Smirnov (KS) statistic. This yields 500 KS values for each statistical distribution we examine. The statistical distribution yielding lowest KS values best reflects the in vivo translational (or turn) speed dynamics; these are graphed as cumulative distribution functions in Fig 1E to 1H, explored below. Cellular turn dynamics are analysed using the same procedure, but additionally accounting for the maximum discernible rotational velocities for each track as determined by imaging experiment temporal resolution (Methods, S1 Algorithm and S1 Table).
T cell and neutrophil translational dynamics are better captured with a statistically heterogeneous Gaussian distribution than a homogeneous Gaussian distribution. When fitting distribution parameters against pooled in vivo translational speed data both statistical distributions performed well, Fig 1D, S6 and S7 Figs: KS values differentiating modeled and in vivo pooled translational speed data were low. However, median track translational speed data were better captured by the heterogeneous Gaussian distribution, Fig 1E and 1F, and S8 Fig. We also evaluated the capacity for Lévy distributions, the foundation of the Lévy walk, to reproduce in vivo translational dynamics. The Lévy distribution was competitive with the Gaussian distributions in capturing pooled translational speed data, Fig 1D, but was inferior in its capture of median track translational speed data, Fig 1E and 1F and S8 Fig.
We determined that homogeneous and heterogeneous Gaussian distributions both accurately capture pooled turn speed data, Fig 1D, S9 and S10 Figs, but the heterogeneous Gaussian proved superior in reproducing in vivo median track turn speed distributions, Fig 1G and 1H and S8 Fig. We additionally evaluated a uniform distribution’s capture of turn speed dnymaics, which corresponds with Brownian motion and Lévy walk motility models where successive trajectories are uncorrelated. We determined that the uniform distribution provided a competitive reflection of in vivo pooled turn speed distributions, but was the worst of the three models in reproducing median track turn speed dynamics.
We hypothesized that owing to physical constraints on rates of cytoskeletal remodeling cells are unable to perform both very fast translational movements and turns simultaneously. We confirmed this in both our datasets, Fig 2. The Spearman’s correlation coefficient between cell turn speed and the median of the translational speeds recorded immediately before and after the turn was -0.29 and -0.27 for T cell and neutrophil datasets respectively.
Collectively these data suggest that cells in both our T cell and neutrophil datasets are statistically heterogeneous: the distributions of varying median track translational and turn speeds reflect inherent differences in cellular speed and directionality, rather than sampling artifacts. Further, they suggest a trade-off between fast translational movement and large directional alterations. We next sought to investigate the significance of these observations by designing several correlated random walk models around the statistical distributions explored here, and evaluating their capture of our leukocytes’ spatial exploration through simulation.
Through agent-based simulation we have assessed the ability of six random walk models to reproduce the motility dynamics of our T cell and neutrophil datasets (full details in Methods): Brownian motion, Lévy walk and four correlated random walks (CRW). Table 1 details how these random walk models are designed around the statistical distributions explored in the previous section (see S5 Fig). The HomoCRW and IHomoCRW both represent cellular populations as statistically homogeneous: all cells draw translational speeds from the same homogeneous Gaussian distribution, and similarly for turn speeds. The HeteroCRW and IHeteroCRW models instead define bespoke, potentially unique distributions for each individual cell, rendering them statistically heterogeneous in inherent translational speed and directionality. The IHomoCRW and IHeteroCRW models impose an inverse correlation between translational and turn speeds, whereas HomoCRW and HeteroCRW do not.
Each model was independently implemented in a 3D simulation, and subsequent calibration identified parameter values that align simulation with in vivo motility dynamics. Calibration was performed through multi-objective optimization [22], therein simultaneously considering several metrics (‘objectives’) of cellular motility. A multi-objective approach is necessary as no single metric can fully specify the complexities of 3D motility. Three objectives are employed, each quantifying a specific difference between the motility profiles of the target in vivo dataset and a given model-parameter set simulation dataset respectively. A motility profile constitutes: the distribution of translational speeds observed across all cells at all time points pooled together; similarly for turn speeds; and the distribution of cell meandering indices, defined as a cell’s net displacement divided by its total distance traveled. These distributions are contrasted using the Kolmogorov-Smirnov (KS) statistic, therein forming the three calibration objectives. The meandering index was selected over alternatives such as mean squared displacement (MSD) for it’s ability to capture a distribution of heterogeneous cellular directionalities, which MSD does not; this choice is revisited in the Discussion.
A calibration exercise yields a three-dimensional Pareto front comprising sets of putative model parameter values (‘solutions’, S11 Fig). These solutions are Pareto-equivalent: no solutions offer an improvement in any objective without a worsening in another. We evaluate which models best reproduce in vivo motility by contrasting their respective Pareto fronts through three complementary analyses (additional details in Methods). Firstly, the proportion of each Pareto front that is non-dominated by each of the others is ascertained (S11 Fig); a solution is dominated if there exists another with at least equal performance on all objectives and superior performance on at least one. Secondly, the best (lowest) 30 Λ values of each front are contrasted; a low Λ value reflects a solution with low mean and variance in its 3 objective KS scores. The best 30 Λ values represent those in the centre of the front, providing good performances on all objectives simultaneously. Lastly, the distribution of KS scores represented across each Pareto front for each objective are directly contrasted. Each model is independently calibrated three times against each in vivo dataset, the best solutions of which form a Pareto front for subsequent evaluation. Calibration is performed using NSGA-II [22] for 40 generations, comprising between 20 and 100 candidates per generation, a reflection of the number of model parameters and hence the difficulty of the calibration exercise (see Methods). The performances of the very best solutions found for each motility model, that with the lowest Λ value, against in vivo data are shown in S12 to S23 Figs.
The Pareto fronts of best calibrated solutions for Lévy walk outperform those of Brownian motion. All Lévy walk solutions are non-dominated by Brownian motion solutions in both datasets, Table 2. For the T cell dataset all Brownian motion solutions are dominated by those of Lévy walk, and for neutrophil data only 7% of solutions are non-dominated. These patterns are reflected in the superior Λ values that Lévy walk solutions’ offer over those of Brownian motion (Fig 3). Brownian motion constitutes a particularly poor reflection of our T cell data, with only 7 tracks in the best Λ value solution remaining after applying a 27μm net displacement filter (applied to reflect in vivo data preprocessing to remove anomalous imaging artifacts, see Methods), S12 Fig.
Brownian motion and Lévy walk are inferior to all the CRW models in capturing T cell motility; ≥99% of their solutions are dominated in all cases (Table 2), and they provides the poorest Λ values found in any model (Fig 3).
For the neutrophil dataset, Brownian motion is again suboptimal compared to all CRWs, in terms of both Λ values and non-domination. In contrast to T cell capture, however, Lévy walk solutions are not universally dominated by those of the CRWs, with as much as 72% of the Lévy walk Pareto front being non-dominated by that of HomoCRW. The HeteroCRW and IHeteroCRW models fare better, with >76% of their Pareto fronts non-dominated by that of Lévy walk, versus 32% and 45% vice versa. In terms of Λ value performance Lévy walk completely dominates all other models, Fig 3. This is somewhat surprising given the competitive non-domination performances, and likely reflects how performances against each objective, explored below, are balanced in Lévy walk solutions; solutions with equal KS measures on each objective score lower Λ values (Methods).
Lévy walk accurately and competitively captures the directional persistence of in vivo data, but Brownian motion does not. Brownian motion delivers a narrow distribution of meandering index KS values, far inferior to other models’ values (Figs 4 and 5 for T cells and neutrophils respectively). Lévy walk mirrors the performance of the best CRW model in capturing neutrophil meandering indices, Fig 5, and is statistically indistinguishable from all CRWs in capturing T cell meandering indices except the IHeteroCRW which offers the best performance, Fig 4.
The meandering index reflects the interplay between cellular translation and orientation adjustments, and it is notable that Lévy walk does not perform particularly well in either of these, despite offering such competitive performance in capturing meandering indices. For turn speed KS values, Lévy walk and Brownian motion represent the poorest fits. Lévy walk offers a poor fit to T cell translational speeds, but exhibits a strangely narrow distribution of KS values for neutrophils; as shown in Fig 5, Lévy walk’s worst translational KS values are better than the worst of the CRWs, however its best are worse than those of CRWs. Brownian motion poorly reflects the in vivo translational speed dynamics of both datasets.
The previous section’s exploration of leukocyte heterogeneity found the Lévy distribution to poorly fit to leukocyte median translational speed data, and similarly, the uniform distribution to poorly fit median turn speed data; it is through these distributions that Lévy walk translational and reorientation adjustments are drawn. We analysed the present Lévy walk simulation’s capture of leukocyte heterogeneity, S24 Fig. Overall, Lévy walk performance is better than might be expected given the previous section’s results. It offers competitive or improved capture when contrasted with the HomoCRW and IHomoCRWs, but generally inferior to the HeteroCRW and IHeteroCRW, with the exception of median neutrophil turn speed data.
In summary, Brownian motion is universally poor in capturing both T cell and neutrophil motility. Lévy walk is similarly poor in capturing T cell performance, but for neutrophils the situation is more complex. Though uncompetitive in turn speed capture, moderately so in translational speed capture, and being largely Pareto-dominated by the IHeteroCRW’s Pareto front, Lévy walk offers by far the best performance on Λ values. We theorise that there exists a portion of the neutrophil Lévy walk Pareto front comprising solutions with similar, low mean objective KS values, as this would yield low Λ values despite not dominating in any particular objective alone.
We find that CRW models accommodating heterogeneous characteristics between cells better reflect in vivo data (Table 2). HeteroCRW Pareto fronts for both datasets are almost entirely non-dominated by the HomoCRW Pareto fronts, versus <7% vice versa. A similar trend is found when comparing the inverse CRW class of models, where IHeteroCRW solutions were largely non-dominated by IHomoCRW models. Here, however, the IHomoCRW was was 42% non-dominated on the neutrophil dataset, substantially higher than the 5% of the T cell dataset.
The superiority of heterogeneous over homogeneous CRW models was also reflected through the best 30 Λ value distributions, Fig 3. We find no overlap in Λ value distributions between HeteroCRW and HomoCRW models on either dataset, with the former providing superior values. On the T cell dataset IHeteroCRW is similarly superior, however neutrophil dataset IHeteroCRW and IHomoCRW Λ distributions overlap, with the former providing marginally superior values. To provide an intuition into the magnitude of the separation between heterogeneous and homogeneous CRW models’ Pareto fronts, S25 and S26 Figs provide three-dimensional plots of Pareto front solutions against each objective KS score. The separation between Pareto fronts is particularly large for HeteroCRW and HomoCRW on the neutrophil dataset, and IHeteroCRW and IHomoCRW on the T cell dataset.
The better fit of heterogeneous over homogeneous models of cellular populations is reflected in performance on each objective (Figs 4 and 5). HeteroCRW yields a distribution of KS values statistically significantly lower than HomoCRW models for all three measures of cellular motility on both datasets, with the exception of T cell meandering indices where no statistically significant difference is observed (Fig 4). We similarly find IHeteroCRW to yield statistically significantly lower values than IHomoCRW on both datasets, with the exception of neutrophil meandering indices where IHomoCRW provides lower values (Fig 5).
Our simulation studies further support our determination that our in vivo cellular populations are statistically heterogeneous, and that observed distributions of median track translational and turn speeds (Fig 1A and 1B) are not sampling artifacts. Our simulations impose the same experimental constraints as are present in vivo: finite observations of cells within a finite imaging volume. Despite not being used as criteria for model calibration, the heterogeneous CRW models best captured median track translation and turn characteristics, with the exception of neutrophil median track turn speeds (S24 Fig).
Analysis of both in vivo datasets revealed significant negative correlations between cellular translation and turn speeds (Fig 2). This correlation could impact cellular directionality, and hence meandering indexes, as subsequent fast translational movements would be directionally persistent. As such, we examined whether the inverse CRW formulation, which imposes this quality on simulated cells (see S27 Fig), better reflects the in vivo data than the standard formulation.
Inverse CRW models better capture T cell motility dynamics than the standard formulations, but the difference is moderate. 85% of IHomoCRW solutions are non-dominated by HomoCRW models, in contrast to 27% vice versa (Table 2). A larger disparity is found for IHeteroCRW and HeteroCRW models, with values of 95% and 10% respectively. Fig 3 reveals, however, that the magnitude of this dominance is marginal: the Kolmogorov-Smirnov statistic reveals a difference of 0.7 between HomoCRW and and IHomoCRW Λ value distributions, and no statistically significant difference between HeteroCRW and IHeteroCRW models. Given the Pareto-dominance of IHeteroCRW over HeteroCRW models this Λ value finding is surprising, and suggests that the dominance occurs on the periphery of the Pareto fronts; Λ values focus on the centre only. Corresponding analysis of model performances’ on each objective highlights translational speeds as the objective where IHeteroCRW outperforms HeteroCRW; no significant difference is observed on other objectives (Fig 4). IHomoCRW’s capture of T cell turn speeds is significantly better than HomoCRW’s, but we find no other statistically significant differences across objectives. The lack of statistically significant differences between inverse and standard model formulations on meandering index performance is surprising, given that it was specifically this objective that the inverse formulation was hypothesized to offer improvement in. Rather, the inverse formulation facilitates performance improvement on other objectives whilst maintaining a similar meandering index profile. S28 and S29 Figs provide three-dimensional plots of Pareto front solutions against each objective KS scores for standard versus inverse model formulations.
IHomoCRW better captures neutrophil dynamics than HomoCRW, but this finding does not extend to heterogeneous CRW formulations. 98% of IHomoCRW solutions are non-dominated by the HomoCRW model, and 9% vice versa. Conversely, IHeteroCRW and HeteroCRW are largely Pareto-equivalent, where 67% of HeteroCRW solutions are non-dominated in contrast to only 50% of IHeteroCRW models. These findings are supported by Λ value distributions, where IHomoCRW yields substantially better values then HomoCRW, yet no significant difference is found between IHeteroCRW and HeteroCRW models (Fig 3). We find that the IHomoCRW models offers significantly better meandering index and translational speed values than the HomoCRW model, Fig 5. HeteroCRW provides superior translational speeds to IHeteroCRW, but otherwise these two models are statistically indistinguishable.
The advent of two-photon in vivo cellular imaging techniques facilitates detailed examination of cellular motility and interaction. The resultant data permits identification of cellular motility strategies, which can be incorporated into broader immune simulations to understand the development and potential manipulation of the immune response. Determining which motility model best fits a biological dataset requires simultaneous consideration of several metrics of motility; three dimensional motility is too intricate a phenomenon to be fully specified in only one metric. Here we have evaluated the capacity of six random walk models, including Brownian motion, Lévy walk and four correlated random walks, to reproduce the motility dynamics of lymph node T cells and neutrophil datasets under inflammatory conditions. Our evaluation is made possible through the development of a novel simulation calibration methodology, where multi-objective optimization identifies parameter values that provide optimal trade-offs for a given model against several metrics of motility.
We found that Lévy walk, an optimal strategy for finding sparsely randomly distributed targets [16, 17], and identified as the motility pattern of CD8+ T cells in Toxoplasma gondii-infected mouse brains [15], does not optimally capture our T cell motility dataset. Its performance in capturing neutrophil motility was competitive with other models’, performing well in some motility measures but poorly in others. We attribute our finding to the simultaneous consideration of multiple motility metrics. Lévy walk’s best meandering index performance is competitive with other models’, and as such optimization on that metric alone might highlight Lévy walk as an optimal model (Figs 4 and 5). It is only when performance against this metric balanced with others that Lévy walk’s quality of capture diminishes. It is the micro-level details of leukocyte motility that Lévy walk fails to capture, given their straight-line directional persistence punctuated by uniformly random reorientations of direction. This is supported by our fitting of statistical distributions to cell translational and turn speed data, where Lévy and distributions and uniform distributions (corresponding with uncorrelated cellular trajectories) poorly captured the data. We do not discount the possibility that hybrid strategies, where micro-level correlated random walks are subject to macro-level directional persistence captured by Lévy walks [16], might better reflect in vivo data.
We determined our T cell and neutrophil populations to be statistically heterogeneous in their inherent translational speed and directional persistence. We devised a novel approach for fitting statistical distributions to either translational or turn speed data whilst accounting for imaging experiment bias. Our approach ruled out the possibility that these heterogeneous qualities arise as sampling artifacts from observing cells for finite durations within a finite imaging volume; cells that simply happen to be moving fast in a persistent direction as they crossed the imaging volume would give the illusion of being statistically distinct from cells that happened to be moving slowly with little directional persistence at time of observation. We quantified this bias, and found strong negative correlations between median cell track translational speed and observational duration in both datasets. Likewise, we found strong positive correlations between median turn speeds and observational duration. Our statistical distribution fitting approach uses a given distribution to reproduce in vivo data, capturing the same number of tracks, the same observations per track, and imposing similar correlations between median track feature and number of observations. A heterogeneous statistical distribution, wherein each track’s data is generated from a bespoke, potentially unique, Gaussian distribution best reflected our in vivo data in all cases. Homogeneous distributions, wherein the same parameterized distribution was used to model all tracks’ data could not reproduce the heterogeneity observed in vivo, despite accounting for the experimental biases.
We confirmed the significance of this cellular heterogeneity through agent-based simulation, which, rather than separately exploring translation and turn dynamics, integrates them to produce 3D tracks. CRW models representing a continuum of heterogeneous qualities within a cellular population proved superior to treating all cells as statistically equivalent. This finding supports the conclusion that leukocytes differ in their inherent rotational and translational speeds. We discount the alternative conclusion that these more complex models capture nuances (rather than general qualities) of the training data set, as levels of over-fitting were monitored and deemed acceptable (see Methods). The large sizes of our datasets, 751 T cells & 1017 neutrophils (see Methods), further suggest that these heterogeneous qualities do not result from small sample sizes. Banigan et al. first described a heterogeneous population of CD8+ T cells in uninflamed lymph nodes, characterizing them as two distinct homogeneous sub-populations, 30% of which perform Brownian motion and the remainder a persistent random walk, all of them drawing velocities from the same distribution [21]. In contrast, here we identified an entire continuum of inherent cellular translation and turn characteristics, in both neutrophils in the mouse ear pinnae, and lymph node T cells, both under inflammatory conditions.
Analysis of both our T cell and neutrophil datasets revealed strong inverse correlations between cell translational and turn speeds: cells do not simultaneously perform fast translational movements and large reorientations. This has been shown previously for neutrophils [23], but we are unaware of any such finding in T cells. We again used simulation to evaluate the impact of this characteristic on overall motility, devising CRWs that impose this negative correlation (‘inverse’ CRW) and contrasting their capture of in vivo dynamics with those that do not. We found inverse CRWs to better capture T cell data than standard formulations, in particular improving capture of translational speeds when coupled heterogeneous qualities. In neutrophil data, an inverse homogeneous CRW substantially improves upon standard homogeneous CRW performance, yet inverse and standard heterogeneous CRW models are indistinguishable. This finding could originate from constraints on the cytoskeleton remodeling processes [24]. Alternatively, cellular dynamics can be explained through the configuration of obstacles in the environment [25]; our findings might represent features of the environment rather than the cell, where cells must slow in order to move around an obstacle. We conclude that the inverse heterogeneous CRW models best capture leukocyte motility: their corresponding Pareto fronts are non-dominated by any other model (Table 2), with one exception where IHeteroCRW and HeteroCRW were indistinguishable.
Previous lymphocyte modeling efforts have incorporated explicit cellular arrest phases between periods of fixed speed, straight-line motility [15, 26]. Our in vivo datasets do not record cells as being stationary, or moving in straight lines (S1A and S1B Fig). As such, we have explored CRW models that explicitly capture distributions of translational and turn speeds. Other work has focused on modeling lymphocytes as point-processes confined to the lymph node reticular network [27], explicitly modeling cellular morphology [25, 28], and conceptualizing cell trajectories as features of environmental obstacles [25]. The possibility of calibrating the configuration of an environment by proxy of the resultant cellular motility is intriguing. Our multi-objective optimization framework is independent of the motility paradigm and could be more broadly applied in these contexts.
We opted to employ three objectives in our multi-objective approach, based on the pooled translational speeds of all cells across all time points into a single distribution, similarly for turn speeds, and track meandering indices. We consider this the minimum required to accurately specify motility, capturing how cells move translationally through space, how subsequent trajectories are correlated, and how these two aspects integrate to define overall spatial coverage. Multi-objective optimisation can accommodate more objectives, and hence additional motility metrics could be incorporated (or substituted). In particular, we believe there is merit in studying how recent, more sophisticated motility metrics might be incorporated into our framework [20, 21]. It is practical, rather than technical, considerations that limit the number of objectives one can use: in our experience the number of Pareto front members tends to increase with each additional objective, and more objectives constitute a more complex problem which can require greater computational effort to solve to a similar extent (e.g., as measured through objective KS values). Convention in multi-objective optimisation dictates that one choose objectives which are not correlated with one another; to do so increases the complexity of the optimisation problem whilst providing little benefit in capturing better quality solutions. Candidates for additional objectives might include the median track translational or turn speed distributions, however we note that for our favoured motility model, the inverse heterogeneous CRW (IHeteroCRW), these characteristics are well captured despite not being explicit criteria in model calibration (S17B, S17C, S23B and S23C Figs).
We consider it essential to include an objective capturing how translational and turn characteristics integrate to dictate spatial coverage. In this regard we employ the meandering index, but possible alternatives include mean squared displacement (MSD) or spatial volume explored. Each of these introduces some bias, and hence the decision is somewhat arbitrary. For instance, meandering indices tend towards 1 for short tracks; S2A and S2B Fig quantify this for our in vivo datasets. We note, however, that our simulation approach imposes the same experimental constraints as exist in vivo, and all our data are processed through the exact same analytical pipeline (see Methods). As such the same biases arise in all our data, providing a fair comparison between in vivo and simulation experiments. It is notable that similar correlations and scatter plots occur between track duration and meandering index for our simulation and in vivo datasets (T cells: S2A, S12H to S17H Figs; neutrophils: S1C Fig) MSD has been used extensively to discriminate between motility models, however, in addition to the known issues with this metric [20, 21], our characterisation of statistically heterogeneous populations prompted our choice of the meandering index, which neatly captures the distribution of cellular directional persistencies and which the MSD does not. We have performed a pilot study substituting MSD in place of the meandering index, calibrating the IHeteroCRW model against neutrophil data (details in Methods). Capture of pooled translational and turn speed data formed the remaining two objectives. S30 Fig contrasts IHeteroCRW’s capture of neutrophil motility under each calibration scenario. As to be expected, the meandering index and MSD metrics were best aligned when used directly as a calibration objective. Pooled translational speed data was best captured using the meandering index, and turn data capture was statistically indistinguishable. Interestingly, median track translational speeds were better captured using the meandering index, and median track turn speeds through MSD (neither were used in driving calibration). The best single solution arising from the MSD calibration is shown in S31 Fig, and can be contrasted with that of meandering index calibration, S23 Fig. Again, the results are remarkably similar, with the exception that using the meandering index better captures median track translational speeds and correctly captures the in vivo MSD, whilst calibrating with MSD poorly captures in vivo meandering indices. The similarities in these data support our belief that both meandering index and MSD capture similar aspects of motility when coupled with metrics of pooled translational and turn speed data, as in the current context. Banigan et al. have proposed metrics capturing displacement probability densities, and displacement autocorrelations for given time intervals [21]. We consider these metrics more statistically robust than either the meandering index or MSD, and have calculated displacement autocorrelation measures for our leukocyte and modeled datasets (S32 and S33 Figs). However, given our focus on cellular heterogeneity, captured in both the data spread at each time interval and how individual cells perform across intervals, it is not clear how to integrate such high dimensional data into an objective to be used in the present calibration framework. This we highlight as meritorious further work. We note that the IHeteroCRW model generally deemed superior by our present methodology also provides a close qualitative alignment with in vivo displacement autocorrelation data.
Our novel method for contrasting putative models, and therein parameterizing them, has a valuable role to play in the development of biological simulations. The construction of simulations which demonstrably capture biological systems has received recent attention [29]. This resulted in a process through which assumptions underpinning the abstraction of key biological components and processes into a conceptual model and thereafter a software implementation are explicitly captured [30]. A complementary technique, borrowing from safety critical systems engineering, decomposes a claim such as “This simulation is an adequate representation of the biology” into sub-claims against which evidence is cited [31]. Additionally, statistical analyses quantifying the impact of biological uncertainty on simulation results by highlighting critical parameters and pathways have been developed [32, 33]. Together these techniques support the development and interpretation of biologically meaningful simulations. The novel multi-objective optimization approach developed here is complementary in helping select between competing abstractions of the biology by providing numerical evidence of improved capture. Whilst there exist established model selection techniques such as the Akaike Information Criterion and Schwarz criterion [34], it is unclear how to apply them over multiple metrics of biological capture, as in the present case. A strength of both the Akaike Information Criterion and the Schwarz criterion is their consideration of model parameter number when determining the most appropriate model. This feature is currently lacking from our multi-objective approach, and we see value in further work investigating how to reconcile these approaches. In the context of our present simulation work, the model with the most parameters (inverse heterogeneous CRW) yielded either the outright or joint best capture of the biology. We note, however, that this model’s parameters and the features they represent are not arbitrary, but are instead biological driven: they were found to be present in both our in vivo datasets.
Simulation parameterization presents another challenge in biological simulation. The required biological data do not always exist as the corresponding experiments either have not or cannot be performed, and simulation’s abstractive nature complicates their adoption. Existing parameterization approaches include exhaustive search of all possible parameter value combinations [35, 36], maximum-likelihood estimation [15], various forms of regression [37], and genetic algorithms [38]. These techniques do not always scale to simulations with many parameters, and none accommodate the simultaneous consideration of several metrics of simulation’s capture of the biology as our present MOO-based approach does.
We have developed and demonstrated a technology that more robustly determines which motility strategies best characterize a given biological dataset. Furthermore, it can implicitly embed these motility dynamics in a simulation, therein enabling more accurate simulations of immune response development. The intricate and nuanced motility patterns that our method reproduces are important, as it is at this scale that two nearby cells either contact or not, and these interactions can have a profound downstream influence on the immune response. Our approach can be used to characterise and quantify, in detail, how various factors impact and manipulate cellular motility, such as was done through inhibition of LFA-1 affinity and avidity regulation in T cells [39].
All procedures involving mice were reviewed and approved by the Garvan/St Vincents Animal Ethics Committee (AEC). The AEC fulfills all the requirements of the National Health and Medical Research Council (NHMRC) and the NSW State Government of Australia.
Neutrophil data was obtained using in vivo two-photon microscopy of ear pinnae in anesthetized C57/BL6 mice. Neutrophils were recruited in response to sterile needle injury and neutrophil migration was recorded and analyzed following the induction of a small sterile laser injury as was described previously [40]. Neutrophils were visualized with the aid of Lysozyme M fluorescent reporter.
The analysis of lymphocyte motility fluorescent lymphoid cells were adoptively transferred and cell migration was visualized 24 hours later in explanted cervical lymph nodes perfused with warmed and oxygenated medium. Inflammation was induced using either S. aureus bioparticles or ovalbumin in Sigma adjuvant.
Two-photon imaging was performed using an upright Zeiss 7MP two-photon microscope (Carl Zeiss) with a W Plan-Apochromat 20′/1.0 DIC (UV) Vis-IR water immersion objective. High repetition rate femtosecond pulsed excitation was provided by a Chameleon Vision II Ti:Sa laser (Coherent Scientific) with 690-1064nm tuning range. We acquired 3μm z-steps at 512×512 pixels and resolution 0.83μm/pixel at a frame rate of 10 fps and dwell time of 1.27μs/pixel using bidirectional scanning. Neutrophil dataset z-depths were 180μm, and T cell dataset z-depths ranged from 150 to 220μm. Both datasets were cropped using Imaris software to correct for tissue drift as needed.
Raw image files were processed using Imaris (Bitplane) software. A Gaussian filter was applied to reduce background noise. Tracking was performed using Imaris spot detection function to locate the centroid of cells and x,y and z coordinates of each spot were exported together with track ID and time interval information.
The T cell calibration data is pooled from 9 individual imaging datasets, comprising a total of 751 cells tracked for a total of 20424 spots, yielding a mean of 27 spots per track. The neutrophil dataset comprises data pooled from 6 individual imaging datasets, totaling 1017 cells encompassing 24619 spots, a mean of 24 spots per track. The T cell experiments were conducted for around 30 min with time-series data recorded every 35 seconds, and for 45 min with time samples every 45 seconds for neutrophil data; exact figures are given in S1 Table.
Several statistical distributions, graphically depicted in S5 Fig, were independently fitted to a given dataset: either cellular translational or turn speed data. This was performed for both our T cell and neutrophil datasets independently of one another. A graphical overview of our method is given in S4 Fig.
We obtain a Lévy distributed random variable as follows, adapted from [41]:
L ( α , β ) = β sin ( α X ) cos ( X ) 1 / α cos ( ( 1 - α ) X ) Y ( 1 - α ) / α (1)
Where random variable X has uniform density on the interval [−π/2, π/2]; Y has unit exponential density, generated as Y = − lnZ where Z is uniformly distributed over [0, 1]; and β is a scaling factor. L is symmetrical around 0 and hence we take the absolute value, represented as |x|.
A ‘homogeneous Gaussian’ distributed variable G(μ, σ2) has mean μ and standard deviation σ2. It is homogeneous in that the same parameterized Gaussian is used to represent all cells’ translational (or turn) values. In contrast, a ‘heterogeneous Gaussian’ distribution comprises a bespoke Gaussian G i ( μ i , σ i 2 ) for each cell i in the dataset. The mean μi and standard deviation σ i 2 of Gi are themselves drawn from Gaussian distributions; this is done once at Gi’s creation, and the values are maintained throughout Gi’s use thereafter. Hence, a heterogeneous Gaussian is formulated as G i ( μ i = G ( μ M , σ M 2 ) , σ i 2 = G ( μ S , σ S 2 ) ), and has parameters μM, σ M 2, μS and σ S 2.
U(λ) represents a uniformly distributed random variable over the range (0, λ].
The parameters describing each statistical distribution are shown in Table 3:
To evaluate the capacity for a given statistical distribution, D, to reproduce an in vivo dataset’s translational data we create an artificial dataset of similar structure. Values are drawn from D, and allocated into groups. There is one group for each track in the in vivo dataset, and initially each group contains as many observations drawn from D as the maximum number of observations found in any in vivo track. S1 Algorithm, in the supplementary data, discards observations from each group such that the number of observations in each group exactly matches the number of observations in a specific in vivo track. The observations to be discarded from each group are chosen such that the correlation between the number of observations in groups and the median observation values of those groups align with the correlations found for in vivo tracks. In this manner, the artificial dataset generated by D reflects the experimental bias inherent in the in vivo data. The pooled observational data, and the median observation values amongst the groups are then extracted, and contrasted with in vivo translation or turn data being analysed as follows.
Let T represent the target data, be it either translational or turn speed data from one of our datasets, to which a given statistical distribution is to be fitted. First D is fitted against the pooled data T, that is, all the translation/turn observations pooled into one distribution. Fitting is performed using the python scipy.optimize.minimize method, using the ‘Powell’ solver, on the basis of minimizing the Kolmogorov-Smirnov (KS) statistic between pooled T data and pooled data generated using D in S1 Algorithm. This is performed 5 independent times, the results of which are shown in S6, S7, S9 and S10 Figs. Upon the conclusion of each fitting exercise, 100 further datasets are generated using the fitted D. We quantify how well each dataset captures the median track data in T using the KS statistic, yielding a total of 500 KS values for each D. Contrasting these 500 KS values reveals which statistical distribution best captures T, with low values indicating a better capture. The best alignment for each model on each in vivo dataset is shown in S8 Fig.
We highlight that this procedure does not attempt to reproduce cellular motility in space, which is an emergent product of how translational and turn movements are integrated. Rather, it determines which distributions best capture translational and turn data independently of one another, and assess whether cells are heterogeneous in these characteristics. We design several random walk models based the distributions investigated here, and assess their capture of cellular motility in space through 3D agent-based simulation, as detailed in the Sections that follow.
The six random walk models explored in this paper are detailed below. The models are constructed around the statistical distributions described above, and illustrated in S5 Fig. Table 1 summarizes which statistical distributions are employed in each random walk model, and how. The random walk models are simulated over time, and as we adopt the notion Dt to indicate a value drawn from randomly distributed variable D at time t.
The random walk models are implemented in a discretized time, three dimensional continuous space agent-based simulation wherein cells are implemented spheres that cannot overlap. Only cells residing within a 412×412×100μm volume are tracked, replicating in vivo experimental conditions. T cell simulation state was updated and recorded for downstream analysis every 30s, and simulation were executed for 30min of simulated time. The corresponding neutrophil figures are 45s and 50min. These values were selected to broadly mirror in vivo experiments, as described in S1 Table. Note that Lévy walk simulation states were updated every 3s instead, owing to the variable cell run-durations of this model, however simulation state was still recorded every 30s and 45s as with other models.
Both simulated and in vivo data undergo the same motility analysis, based on time series tracked cell spatial locations sampled every Δt seconds. For each time point ti, the vector describing the movement of a cell to its current location is calculated, and termed di. The displacement and translational speed over vector di are calculated. A cell’s turn speed at time ti is calculated as the angle between vectors di+1 and di divided by Δt.
The largest measurable turn angle is 180°, and conversion into turn speeds (°/min) depends on the time step. Simulation time steps, 30s for T cells and 45s for neutrophils, correspond with maximum turn speeds of 360 and 240°/min respectively. These figures match the maximum discernible turn speeds for the in vivo datasets. However, the maximum discernible turn speed for each experiment within a dataset will differ with the time step (see S1 Table), and this could represent an artifact for our calibration experiments. Given the majority of recorded turn speeds lie well below the maximum values (S1B Fig) we believe the influence of this discrepancy on calibration experiments to be minor, however we acknowledge its existence.
A cell’s meandering index is defined as the net displacement from its first to last observed locations divided by its total distance traveled. This yields a value between 0 and 1, respectively indicating the extremes of a cell finishing where it started or traveling in a straight line. Cells with total displacements <27μm are excluded from the analysis to avoid artifacts introduced by the sessile contaminating cell types such as dendritic cells, or cells that are dead or dying. This same displacement threshold is also applied to simulation data to ensure fair comparisons. The figure of 27μm was derived empirically using Imaris software, and represents an optimal trade-off for removing unwanted artifacts whilst minimizing the exclusion of motile T cells and neutrophils.
The motility profile for a dataset, which typically constitutes several replicate experiments, comprises the following metrics. All translational speeds for all cells are pooled together to form one distribution. A similar pool of all cell turn speeds is constructed. All cell meandering indexes are pooled together into one distribution. Only these three metrics are used in simulation-based motility model calibration and evaluation.
The following additional metrics are also derived, but not used in calibration or evaluation. We construct distributions of median track translation and turn speeds. Mean squared displacement (MSD) over time interval plots are produced. Displacement data for a given time interval is extracted from anywhere in the time-series, i.e., time intervals are not absolute from time zero. Time intervals of 0 to 25% of the maximum track length are investigated. Slopes for MSD plots are calculated using linear regression. Displacement autocorrelation was calculated as in [21].
Each of the six models is implemented in simulation in turn, and then independently calibrated against each of the in vivo datasets. Calibration is performed using NSGA-II [22], a multi-objective optimization algorithm based on a genetic algorithm that uses Pareto fronts to track candidate solutions representing the best trade-offs found to date with respect to each objective. NSGA-II is an elitist algorithm, meaning that a subsequent generation’s population is composed of the best solutions found to date: the solutions comprising the Pareto front. If the Pareto front comprises more members than the population size, a subset composed of those Pareto members having the largest fitness differences between their immediate neighbours summed for all objectives is selected, a strategy intended to promote full coverage of the Pareto front. If the Pareto front comprises fewer members than the population size then members of the next front (those dominated by only one other solution) are selected in the same manner, and so on until the entire population has been selected. New solutions are generated through blended crossover of their two parents, coupled with Gaussian mutation using the standard normal distribution. These evolutionary operators correspond to the Inspyred python package implementation of NSGA-II. For further details on NSGA-II we refer to the reader to [22].
Candidate solutions represent putative model parameters. Evaluation of a solution entails executing ten replicate simulations with the parameters it represents, and generating a motility profile from the pooled results. This motility profile is contrasted with that of the in vivo dataset: the Kolmogorov-Smirnov (KS) difference between the motility profiles’ distributions of cell translation (S1A Fig) and turn speeds (S1B Fig), and meandering indices (S1C Fig) together form three objectives. A perfect simulation representation of an in vivo data set would yield a KS value of 0 for each objective. In reality, no random walk model, by virtue of being an abstract model, will likely achieve this. Instead, some disparity in at least one metric will exist. The use of Pareto fronts accommodates trade-offs between metrics; two solutions are Pareto-equivalent if neither provides better alignments with in vivo data across all measures.
An individual calibration is performed for a maximum of 40 generations of the genetic algorithm, for all models. Calibration is terminated before 40 generations only if over-fitting, as described below, is detected. The number of candidates in each generation is scaled with the number of model parameters, thereby reflecting the complexity of the problem, as shown in Table 4:
We avoid over-fitting models, wherein calibrated solutions represent the nuanced stochastic-sampling-derived features of the data rather than its general qualities, by dividing in vivo datasets into training (70% of cell tracks) and validation sets (30%), as is standard machine learning practice [37]. Each putative model parameter set is independently evaluated against both training and validation datasets, and two Pareto fronts, representing the best solutions found with respect to each, are maintained throughout calibration. Progression of candidate solutions through subsequent generations is determined through performance against the training dataset alone. The over-fittedness of the population is defined as the proportion of training dataset Pareto front solutions that are not also members of the validation dataset Pareto front. Calibration is stopped when either the maximum number of generations have been run, or the over-fitted metric >0.8.
The model assessments reported here are made on the basis of validation dataset Pareto front solutions. We note that in no cases were any calibration efforts terminated prematurely on the basis of over-fitting, but over-fitted scores of around 0.6 were not uncommon.
Calibration produces a Pareto front comprising those parameter values yielding the best reflections of the in vivo dataset. By contrasting Pareto fronts produced by two different models, that which is most capable of reproducing the motility of in vivo cells is ascertained. For a given model and in vivo dataset (T cell or neutrophil), calibration is performed three times. One overarching Pareto front is then generated from the best solutions generated under each exercise, and is used in model evaluation.
Three complementary analyses are performed when contrasting two models. First, the proportion of each models’ front that is non-Pareto-dominated by the other is calculated. If two models are exactly equal in their capture of the biology across all objectives, then these values should be 100% for each. If the two values are equal, but not 100%, then the models are still considered equal reflections of the biology overall, but they differ in how well they reflect particular objectives. Pareto front sizes are reported alongside these proportions, to highlight where high or low values simply reflect fronts containing few or many solutions.
Second, we contrast the best (lowest) 30 Λ values found within a Pareto front using the Kolmogorov-Smirnov statistic (Fig 3). The Λ function, defined below for a candidate m, delivers low values to solutions having low mean objective KS values with small variance. Hence, it selects those solutions that perform well, and equally well, on all objectives.
Λ ( m ) = α · K S ¯ ( m ) 2 + ∑ o ∈ Ω K S o ( c ) - K S ¯ ( m ) 2 (8) K S ¯ ( m ) represents the mean objective KS score for member m, Ω represents the set of objectives and KSo(m) represents the KS scores for member m against objective o. The coefficient α can be used to prioritize mean or variance terms, a problem specific decision; a value of α = 1 is used throughout this manuscript. S35 Fig depicts how Λ values vary in a hypothetical scenario comprising two objectives, under different values of α.
Lastly, the distribution of scores for each objective generated under each Pareto front are contrasted, thereby highlighting how well each model captures each motility characteristic. These are shown in Figs 4 and 5. The distributions are plotted on the left of these figures and are statistically contrasted using the Kolmogorov-Smirnov statistic, the values of which are given in the tables on the right of these figures.
Experiments where the meandering index was replaced with mean squared displacement (MSD) as an objective for multi-objective optimisation used the same experimental setup as reported above. The MSD calibration objective operates by taking the absolute difference between the MSD linear regression slopes generated for candidate solution and neutrophil dataset as reported above. Two remaining calibration objectives are constructed from KS statistics applied to pooled translational and turn speed data, as reported above. Calibration was performed three independent time using 100 candidates for 40 generations, with an overfitting termination threshold of 0.8.
The best solution from the MSD-based calibration exercise, reported in S31 Fig, is that with the lowest sum of objective values. The Λ function described above is inappropriate in this context, as the MSD objective is not based on the KS statistic. Hence, is it nonsensical to take their mean value.
The 3-dimensional continuous space simulation is written in Java, using the MASON simulation framework library [42]. We use the Inspyred implementation of NSGA-II, written in Python, to perform calibration. Kolmogorov-Smirnov statistics, and their associated p-values, are determined using Python’s scipy.stats.ks_2samp module. The statistical modeling of cellular translation and turn speed dynamics was performed using python, and its numpy and scipy packages. The 3D agent-based simulation and multi-objective optimisation software we developed for this manuscript is distributed under version 3 of the GNU General Public License in the S1 Software ZIP file (the third party libraries we employ will need to be acquired separately from their respective sources for licensing reasons).
|
10.1371/journal.ppat.1001180 | Evolution of Linked Avirulence Effectors in Leptosphaeria maculans Is Affected by Genomic Environment and Exposure to Resistance Genes in Host Plants | Brassica napus (canola) cultivars and isolates of the blackleg fungus, Leptosphaeria maculans interact in a ‘gene for gene’ manner whereby plant resistance (R) genes are complementary to pathogen avirulence (Avr) genes. Avirulence genes encode proteins that belong to a class of pathogen molecules known as effectors, which includes small secreted proteins that play a role in disease. In Australia in 2003 canola cultivars with the Rlm1 resistance gene suffered a breakdown of disease resistance, resulting in severe yield losses. This was associated with a large increase in the frequency of virulence alleles of the complementary avirulence gene, AvrLm1, in fungal populations. Surprisingly, the frequency of virulence alleles of AvrLm6 (complementary to Rlm6) also increased dramatically, even though the cultivars did not contain Rlm6. In the L. maculans genome, AvrLm1 and AvrLm6 are linked along with five other genes in a region interspersed with transposable elements that have been degenerated by Repeat-Induced Point (RIP) mutations. Analyses of 295 Australian isolates showed deletions, RIP mutations and/or non-RIP derived amino acid substitutions in the predicted proteins encoded by these seven genes. The degree of RIP mutations within single copy sequences in this region was proportional to their proximity to the degenerated transposable elements. The RIP alleles were monophyletic and were present only in isolates collected after resistance conferred by Rlm1 broke down, whereas deletion alleles belonged to several polyphyletic lineages and were present before and after the resistance breakdown. Thus, genomic environment and exposure to resistance genes in B. napus has affected the evolution of these linked avirulence genes in L. maculans.
| The fungus Leptosphaeria maculans causes blackleg, the major disease of canola worldwide. Populations of this fungus rapidly adapt to selection pressures such as the extensive sowing of canola with particular disease resistance genes. This can lead to a breakdown of resistance and severe economic losses. We describe mutations in key fungal genes involved in the interaction with canola, and we report the first large scale study of evolutionary processes affecting such genes in any fungal plant pathogen. We relate these changes to the genomic environment of these genes and to the breakdown of disease resistance in canola.
| The fungus Leptosphaeria maculans causes blackleg (phoma stem canker) and is the major disease of Brassica napus (canola) worldwide [1]. The major source of inoculum is wind-borne ascospores, which are released from sexual fruiting bodies on infected stubble (crop residue) of previous crops and can be transmitted several kilometres. This fungus has a ‘gene for gene’ interaction with its host (canola) such that pathogen avirulence alleles render the pathogen unable to attack host genotypes with the corresponding resistance genes [2].
Twelve genes conferring resistance to L. maculans (Rlm1-9, LepR1-3) have been identified from Brassica species [3], [4]. Nine of these genes have been mapped but none have yet been cloned [5], [6]. Of the corresponding avirulence genes in L. maculans, seven have been mapped to two gene clusters, AvrLm1-2-6 and AvrLm3-4-7-9, located on separate L. maculans chromosomes [7]. Three genes, AvrLm1, AvrLm6 and AvrLm4-7, have been cloned and characterised [8], [9], [10]. Avirulence proteins belong to a class of molecules called effectors. Effectors are small molecules or proteins produced by the pathogen that alter host-cell structure and function, facilitate infection (for example, toxins) and/or induce defence responses (for example, avirulence proteins), and are generally essential for disease progression [11]. Many effectors are small, secreted proteins (SSPs), which are cysteine-rich, share no sequence similarity with genes from other species, are often highly polymorphic between isolates of a single species and are expressed highly in planta [12]. The AvrLm1, AvrLm4-7 and AvrLm6 proteins of L. maculans are effector molecules with an avirulence function. AvrLm6 and AvrLm4-7 encode SSPs with six and eight cysteine residues, respectively, whilst the AvrLm1 protein, which is also a SSP, has only one cysteine [8], [9], [10].
Leptosphaeria maculans undergoes sexual recombination prolifically and populations can rapidly adapt to selection pressures imposed by the host such as exposure to resistance conferred by single or major genes. This situation increases the frequency of virulent isolates and can cause resistance to break down, often resulting in severe yield losses [13]. Three examples of this breakdown are discussed below, the most dramatic one occurring in Australia in 2003.
Prior to 2000 the Australian canola industry relied on ‘polygenic’ cultivars with multiple resistance genes. Yield losses were generally low. In 2000, cultivars with major gene resistance, termed ‘sylvestris’ resistance, were released commercially and grown extensively in some areas of Australia. These cultivars were derived from a synthetic B. napus line produced by crossing the two progenitor species, B. oleracea subsp. alboglabra and an accession of B. rapa subsp. sylvestris that had a high level of resistance to L. maculans [14]. For several years these cultivars showed little or no disease but in 2003, resistance failed, resulting in up to 90% yield losses in the Eyre Peninsula, South Australia, costing the industry between $5–10 million AUD [15], [16]. These cultivars contained resistance genes Rlm1 and RlmS, suggesting that both sources of resistance failed simultaneously [17]. After 2004, these cultivars were withdrawn from sale although they were still grown in yield trial sites around Australia. A similar but less dramatic situation occurred in France when resistance conferred by Rlm1 was rendered ineffective within five years of commercial release of Rlm1-containing cultivars [13], [18]. Another breakdown of resistance was observed in field trial experiments in France that had been designed to assess the durability of a resistance gene Rlm6. In these experiments B. napus lines containing Rlm6 were sown into L. maculans-infected stubble of a Rlm6–containing line over a four year period [19]. After three years of this contrived selection, the frequency of virulence in fungal populations towards Rlm6 was so high that this resistance was rendered ineffective and the lines suffered extremely high levels of disease.
The three avirulence genes, AvrLm1, AvrLm6 and AvrLm4-7, cloned from L. maculans are located within AT-rich, gene-poor regions that are riddled with degenerated copies of transposable elements [8], [9], [10]. These transposable elements appear to have been inactivated via repeat-induced point (RIP) mutations [20]. RIP is an ascomycete-specific process that alters the sequence of multicopy DNA. Nucleotide changes CpA to TpA and TpG to TpA are conferred during meiosis, often generating stop codons, thereby inactivating genes [21]. Additionally, RIP mutations have been inferred bioinformatically in various transposable elements throughout the L. maculans genome and in AvrLm6 [22]. AvrLm1 and AvrLm6 are genetically linked and different types of mutations leading to virulence have been reported. The entire AvrLm1 locus was deleted in 285 of 290 (98%) isolates that were virulent towards Rlm1 [20]. Fudal et al. characterised the AvrLm6 locus in a different set of 105 isolates, most of which were cultured from Rlm6–containing lines during the field trial in France described above [22], [23]. Deletions and RIP were responsible for virulence in 45 (66%) and 17 (24%) isolates, respectively, that were virulent towards the Rlm6 gene [22].
In this paper we relate changes in the types and frequencies of mutations in genes including AvrLm1 and AvrLm6 to the selection pressure imposed by extensive regional sowing of B. napus cultivars with sylvestris resistance.
In preliminary experiments to see if the breakdown of ‘sylvestris resistance’ in B. napus seen in the field [15] correlated with changes in frequency of virulence towards Rlm1 in individual L. maculans isolates, 11 isolates collected prior to the breakdown (before 2004) and 12 isolates collected after the breakdown (2004 onwards) were screened for virulence on B. napus cultivars Q2 (with Rlm3) and Columbus (Rlm1, Rlm3). Cotyledons of 14 day old plants were inoculated with individual isolates and symptoms were scored 17 days later. All isolates were virulent on the susceptible control, cv. Q2. Ten of the 11 isolates collected prior to 2004 were avirulent on the Rlm1-containing cultivar, Columbus. However, seven of ten isolates collected from 2004 onwards were virulent towards Rlm1 (Table 1). A subset of these isolates was inoculated onto the B. napus cultivar Aurea (Rlm6). Of seven of the 11 isolates collected prior to 2004, only one was virulent towards Rlm6. However, of 12 isolates collected from 2004 onwards, eight were virulent towards Rlm6 (Table 1). These results suggested that there was a significant change in frequencies of virulent alleles of AvrLm1 and AvrLm6 associated with breakdown of ‘sylvestris resistance’. These isolates were then genotyped at the AvrLm1, AvrLm6 and mating type loci using PCR-based screening [20], [22], [24] and as expected, the virulence phenotypes corresponded with avrLm1 and /or avrLm6 genotypes (Table 1).
Changes in allele frequencies were then examined in a total of 295 Australian isolates. Of these 137 were collected between 1987 and 2003, prior to the breakdown of sylvestris resistance, whilst the remaining 158 isolates were collected between 2004 and 2008, after the resistance breakdown (Table S1). These isolates were collected from stubble of a range of canola cultivars with different resistance genes. One third of the isolates had a deletion of AvrLm1 (Table 2), whilst 63% had the allele of isolate v23.1.3, whose genome has been sequenced. Alleles of this isolate hereafter are referred to as wild type alleles (e.g. AvrLm1-0). As expected, isolates with AvrLm1-0 were avirulent towards Rlm1 (Tables 1 and 2). The remaining eight isolates comprised four alleles with coding sequence changes conferring non-synonymous substitutions (I125K and/or H155Y). Isolates harbouring these alleles were avirulent on the Rlm1- containing B. napus line (Table 1). Thirteen alleles of AvrLm6 were detected (Table 2). AvrLm6 was deleted in 20% of isolates, thus conferring virulence towards Rlm6, whilst 24% had the allele of the sequenced isolate and conferred avirulence towards Rlm6 (Tables 1 and 2). Other isolates virulent towards the Rlm6-containing cultivar had alleles with stop codons conferred by base changes reminiscent of RIP mutation. Accordingly, allele sequences were analysed by RIPCAL, a software tool that visualises the physical distribution of RIP mutation and reports a RIP dominance score indicating the degree of RIP in each sequence. Sequences with RIP dominance scores >1 are highly RIP-affected having a high proportion of CpA to TpA, or TpG to TpC RIP mutations [25]. RIPCAL analysis did not detect RIP in any AvrLm1 alleles, but detected RIP in seven AvrLm6 alleles. These RIP mutations resulted in numerous non-synonymous changes as well as premature stop codons (between 4 and 6), which were within all RIP-affected alleles (Table 2). Southern hybridisation results suggest that there is only a single copy of the AvrLm6 locus within isolates harbouring the RIP alleles (Figure S1). The remaining alleles of AvrLm6 (AvrLm6-1, -2,-3 and -4) harboured single or few nucleotide changes leading to synonymous or amino acid substitutions (G123C, K127E, F54L) compared to isolate v23.1.3 (AvrLm6-0). These amino acid substitutions were generated via non-RIP like mutations (henceforth referred to as non-RIP amino acid substitutions). Isolates harbouring AvrLm6-1, AvrLm6-3 or AvrLm6-4 alleles were avirulent towards cv. Aurea, whilst the four isolates harbouring the AvrLm6-2 allele (G123C) were virulent (Table 1).
No RIP-affected alleles were present in isolates collected prior to the breakdown of sylvestris resistance. However, the 158 isolates collected after the breakdown had seven AvrLm6 RIP alleles (frequency of 8.9%) and there was a very large increase in the frequency of deletion alleles of AvrLm1 (22.6 to 41.8%) and AvrLm6 (4.4 to 38.7). This was a nine-fold increase in frequency of avrLm6 (Table S2). Thirty four (11.5%) isolates had deletions of both AvrLm1 and AvrLm6 and only one of these isolates was collected prior to the breakdown. All 295 isolates were grouped into four genotypic classes (AvrLm1, AvrLm6; AvrLm1, avrLm6; avrLm1, Avrlm6; avrLm1, avrlm6). The frequency of AvrLm1, AvrLm6 isolates was 73.7% prior to the breakdown, but decreased to 37.3% afterwards (Table 3). Conversely, the frequency of avrLm1, avrlm6 isolates was only 0.8% prior to the breakdown, but increased to 28.5% afterwards.
Nine of the 14 isolates harbouring RIP alleles (64%) had a deletion allele at the AvrLm1 locus and all these isolates were cultured from stubble of cultivars with sylvestris resistance. Additionally, all four isolates harbouring the virulent AvrLm6-2 allele (G123C) had a deletion allele at AvrLm1 (Table S3). When the isolates cultured from 2004 onwards were categorised in terms of the stubble from which they were derived, all those cultured from ‘sylvestris stubble’ had the avrLm1 allele (Table 3) as expected, due to the presence of Rlm1 in these cultivars [17]. Conversely, only 15% of isolates cultured from stubble of polygenic cultivars, which do not have Rlm1 nor Rlm6, had the avrLm1 allele. The frequency of avrLm6 was 38.7% in isolates cultured from ‘polygenic’ stubble, compared to 68.7% of isolates cultured from ‘sylvestris’ stubble (Table 3). For all comparisons of isolates, there were no significant differences in allele frequencies at the mating type locus. Since the mating-type locus is not under selection pressure, a 1∶1 ratio of each allele suggests that sampling of isolates has been random (data not shown).
Because of the marked difference in the AvrLm1 and AvrLm6 allele frequencies before and after the breakdown of sylvestris resistance, the region flanking these genes was characterised to identify any features that might have influenced allele frequency. A 520 kb AT-rich genomic region bordered by AvrLm1 and LmCys2 (Figure 1A) was examined. Part of this region has been described previously in isolate v23.1.3 [8] and includes two additional genes encoding SSPs, LmCys1 and LmCys2. Three other genes, LmTrans, LmGT and LmMFS were present; LmCys1, LmTrans, LmGT, LmMFS and LmCys2 have been reported previously [8], [9] but sequence data are only in the form of BAC clones. The features of the proteins are listed below and in Table S4.
The predicted LmCys1 protein (220 aa) contained eight cysteine residues and its single match was to a ‘secreted in xylem’ Six1 effector (also known as Avr3) of Fusarium oxysporum (26% identity, 40% similarity, accession number CAE55870.1) [26]. The predicted LmCys2 protein (247 aa), also containing eight cysteine residues, had no matches within the NCBI database. Both LmCys1 and LmCys2 were predicted to be secreted with signal peptides of 19 and 18 aa, respectively. To determine whether LmCys1 and LmCys2 were expressed highly in planta, which would be expected of genes encoding effector-like proteins, quantitative reverse transcriptase PCR (RT-PCR) analyses were performed. Transcript levels of LmCys1 and LmCys2 in planta were six times higher than those of actin at seven days after inoculation of cotyledons of a susceptible B. napus cultivar (Figure 2). LmCys1 and LmCys2 were expressed at 0.1 and 0.01 times, respectively, that of actin in seven day in vitro cultures. Similar results were obtained with a second isolate (data not shown). This extremely high level of expression in planta compared to in vitro is similar to that seen for AvrLm1 and AvrLm6 (Figure 2) [8]. These characteristics (small cysteine-rich secreted proteins with no or few matches in databases and high in planta expression) strongly suggested that LmCys1 and LmCys2, like AvrLm1 and AvrLm6, encode effector-like proteins.
The LmTrans protein (373 aa) contained a DDE superfamily endonuclease domain predicted to be involved in efficient DNA transposition, and its best match was a putative transposase from Stagonospora nodorum (61% identity, 68% similarity, accession number CAD32687.1). The predicted LmGT protein (414 aa) was a putative glycosyltransferase with best matches to hypothetical protein PTRG_04076 of Pyrenophora tritici-repentis (89% identity, 95% similarity, EDU46914.1). The LmMFS protein (591 aa) belonged to the Major Facilitator Superfamily (MFS). This protein had best matches to putative protein SNOG_04897 from S. nodorum (74% identity, 82% similarity, EAT87288.2).
Since LmCys1 and LmCys2 had effector-like properties and thus putative roles in the plant-fungal interaction, these genes were sequenced in the 295 isolates. Five and two alleles were detected for LmCys1 and LmCys2, respectively (Table 1), including the wild type alleles obtained from published BAC sequences. RIPCAL analysis showed that only one isolate, which was collected after the breakdown of sylvestris resistance had a RIP allele of LmCys1, and there were no deletion alleles, whereas there were no RIP alleles of LmCys2, but two isolates collected before breakdown of sylvestris resistance had a deletion of LmCys2 (Table 1). Overall, the frequencies of individual alleles ranged from 99% for LmCys2-0 to 0.3% for AvrLm1-4. The 295 isolates comprised 34 haplotypes based on alleles of AvrLm1, AvrLm6, LmCys1 and LmCys2 (Table S3).
Four non-coding, non-repetitive regions (Figure 1A) ranging in size between 247 and 657 bp were analysed, to see whether single copy non-coding regions, like the single copy genes, were affected by RIP mutation. These regions and the three other genes LmTrans, LmGT and LmMFS in this region (Figure 1A) were sequenced in a subset of 84 of the 295 isolates, which included isolates representing all 34 haplotypes described above (Table S3). Three, four, 14 and two alleles were detected for NC1, NC2, NC3 and NC4, respectively, whilst two, one, and three alleles were detected for LmTrans, LmGT and LmMFS, respectively (Table 4 and Table S5). The mutations giving rise to the alleles of NC1, NC2 and LmMFS were single non-RIP-like nucleotide substitutions (both synonymous and non-synonymous), and single base pair deletions. No polymorphisms were detected in LmGT and no RIP alleles were detected for NC1, NC2, LmMFS or LmGT. A single isolate had RIP alleles for NC4 and LmTrans in addition to NC3, AvrLm6 and LmCys1. The NC3 region had an extremely high frequency of RIP mutation and the ten RIP alleles were all associated with AvrLm6 RIP alleles. Based on all seven genes and four non-coding regions, 51 haplotypes were identified among the 84 isolates (Table S6).
The distribution and degree of RIP mutation across each allele of each gene was determined. This was represented as a ratio of the number of mutated CpA and TpG sites relative to number of potential RIP sites in isolate v23.1.3 in a 100 bp rolling window. The seven RIP alleles of AvrLm6 had the highest frequency of RIP towards the 3′ end (Figure 3A). The frequency of RIP was higher within the 3′ UTR and the exons, than in the introns and 5′ UTR (Figure 3B). The ten RIP alleles of NC3 and single RIP allele of LmCys1 showed the highest frequency of RIP at their 5′ ends, whilst RIP mutations were evenly distributed throughout the single RIP allele of LmTrans (Figure S2), the latter gene being the furthest 3′ characterised single copy sequence affected by RIP mutation within the AvrLm1-LmCys2 genomic region (Figure 3C). The NC3-Avrlm6 and LmCys1-NC4-LmTrans single copy regions in which RIP mutations were detected, were separated by 37 kb of repetitive elements (Figure 1B). Both these single copy regions were closer to upstream repetitive elements (<342 bp) than to downstream repetitive elements (>1 kb) (Figure 1B). The degree of RIP within these single copy regions was proportional to their proximity to repetitive elements, and thus a gradient of RIP mutations was apparent in a 5′ to 3′ direction (Figures 1B and 3B).
Since amongst these single copy regions the degree of RIP mutations was highest in NC3, a repeat region (620 bp) directly upstream was analysed to see if it was severely RIP-affected and thus might act as a point of ‘leakage’ of RIP into NC3 and AvrLm6 (Figure 1B). BLAST analysis of the genomic sequence of isolate v23.1.3 revealed 293 copies of this repeat. RIPCAL analysis showed that the repeat directly upstream of NC3 was amongst the most highly RIP-affected copy of that repeat in the genome.
The mutations of AvrLm1, AvrLm6, LmCys1 and LmCys2 (deletions or RIP mutations including stop codons) would be expected to lead to lack of transcription of these alleles. This hypothesis was assessed in a range of isolates seven days after inoculation of cotyledons of B. napus cv. Beacon, which all the isolates could attack. Primers designed to amplify 500–700 bp products within the coding region of these genes were used in end-point RT-PCR (Table 5). Isolates with either AvrLm1-0 or AvrLm1-1 allele had an AvrLm1 transcript of the appropriate size. As expected, isolates with the deletion allele had no transcript. The AvrLm6-0, AvrLm6-1 or AvrLm6-2 alleles were expressed, whilst the RIP alleles, AvrLm6-6, AvrLm6-7, AvrLm6-8 or AvrLm6-9 were not. As expected, no expression of AvrLm6 or of LmCys2 was detected in isolates with the deletion alleles of these genes. Isolates with LmCys1-0, LmCys1-1 or LmCys1-2 alleles had an LmCys1 transcript, whilst the isolate with the RIP allele, LmCys1-4, did not. Actin transcripts were detected in all isolates.
To determine the rates of mutations, genes were analysed by phylogeny-based likelihood ratio tests (LRT) implemented in the program HyPhy [27]. These tests suggested that nucleotides of AvrLm1 and LmMFS mutated at a constant rate, i.e. mutations did not deviate significantly (p = 0.544) from a clock-like rate of evolution. In contrast, an accelerated mutation rate was detected for AvrLm6 and LmCys1 loci (P<0.001) compared to expectations under constant (i.e. clock-like) evolution. However, when RIP alleles were excluded, mutations evolved at a clock-like rate (Table 6). Genetic divergence between haplotypes was calculated using Molecular Evolutionary Genetics Analysis (MEGA) software [28]. Relative rates of sequence evolution of non-RIP alleles were much lower than those of RIP alleles (Table 6). To determine whether the seven proteins were undergoing positive selection, the rates of non-synonymous and synonymous substitutions were compared. All RIP alleles were excluded from this analysis since they contained multiple stop codons. Evolution of codon changes within AvrLm1 and LmCys1 was best explained by a model of positive selection, as shown by a likelihood ratio test implemented using two complementary approaches, the sitewise likelihood-ratio (SLR) and phylogenetic analysis by maximum likelihood (PAML) methods (Table 7). This interpretation was supported by the finding of positive selection at a single codon site in AvrLm1 and at three sites in LmCys1 (Table 7). Analysis of the AvrLm6 protein using the SLR approach suggested a single amino acid (codon 54) may be undergoing positive selection; however, this finding was not supported by the PAML analysis. Similar analyses could not be performed on LmCys2, LmGT or LmTrans as these genes had fewer than three alleles.
Two hypotheses for the evolution of RIP alleles, monophyly (having a single origin or having evolved only once) and polyphyly (multiple origins or having evolved several times independently) were tested by comparing tree topologies using the Kishino–Hasegawa (KH) test [29]. For both the Maximum Parsimony (MP) and Maximum Likelihood (ML) approach, trees based on the assumption of a monophyletic origin of RIP alleles performed significantly better than those based on polyphyletic origin (Table 8). The phylogenetic relationship of all detected haplotypes is depicted in Figure 4. The RIP and non-RIP associated alleles form two distinct clusters, supporting the hypothesis of a single origin of RIP alleles. In contrast, haplotypes associated with gene deletions are associated with multiple clades of the tree and have probably arisen several times.
Breakdown of disease resistance has been observed in other plant fungal pathogen systems where fungal populations evolve rapidly (for review see [30]). In the canola- L. maculans interaction described here, strong selection pressure was exerted on the AvrLm1 locus, due to extensive sowing of sylvestris cultivars with Rlm1, which was consistent with the finding of a rapid increase in the frequency of isolates with virulent (avrLm1) alleles after the breakdown of resistance. Surprisingly the frequency of isolates virulent (with the avrLm6 allele) towards Rlm6 increased, although cultivars with polygenic or with sylvestris resistance have not been shown to contain Rlm6 [17], [31]. The linkage and genomic location of AvrLm1 and AvrLm6 may have led to a selective sweep whereby selection at AvrLm1 affected the frequency of avrLm6 alleles through ‘hitchhiking’. Thus strong selection imposed by wide-spread deployment of a plant resistance gene that favors a complementary effector allele in a pathogen could affect evolution of closely-linked effector genes. It is intriguing that in France recurrent sowing of Rlm6-containing cultivars in localised field trials led to an increase in avrLm6 isolates and a corresponding increase in isolates avirulent at the AvrLm1 locus [19]. This difference may be due to the different targets of selection pressure - Rlm6 in France and Rlm1 in our study.
This situation whereby selection pressure on one gene affected allele frequencies of another may be partly due to the presence of these two effector genes in a repeat-rich region, where there is a low recombination frequency. Effectors in other fungi are present in repeat-rich regions. In the rice blast fungus, Magnaporthe oryzae, avirulence gene Avr-Pita is located 48 bp from telomere repeats whilst Avr1-CO39 is associated with transposable elements [32], [33]. An avirulence gene (SIX1) of Fusarium oxysporum f.sp. lycopersici is flanked by transposable elements [34] and other effectors are localised on a single, transposon-rich chromosome [35]. Toxin-encoding genes, Tox3 and ToxA, are located next to repetitive elements in Stagonospora nodorum [36], [37], [38] and effectors are in repeat-rich regions of the genome of the oomycete, Phytophthora infestans [39]. In the sequenced isolate of L. maculans, AvrLm1 and AvrLm6 are located amongst multiple copies of long terminal repeat retrotransposons, namely Pholy, Olly, Polly and Rolly, which are generally incomplete. Furthermore the distribution and number of these elements within this genomic location varies considerably between isolates [20], [22]. The presence of effector genes within such regions is suggested to promote their adaptation and diversification when exposed to strong selection pressure [40]. Rep and Kistler have speculated that the presence of highly repetitive regions containing transposons, may promote mutation of resident effector genes [41].
The genes and non-coding regions undergoing RIP within the AvrLm1- LmCys2 gene region are single copy and therefore are not expected to be targeted by RIP. Two explanations for the presence of RIP mutations in these genes are as follows. Firstly, if these genes are the result of an ancestral duplication, RIP mutation may be acting directly on them. However, Fudal et al showed that no closely related paralogs of AvrLm6 exist in the genome of isolate v23.1.3 suggesting that RIP would not be targeting these sequences due to a duplication event [22]. Alternatively, the AvrLm6 locus may be completely or partially duplicated in isolates where RIP was detected. However, southern hybridisations suggest only a single copy of the Avrlm6 locus in isolates where RIP was detected, which does not support the possibility of RIP being targeted to this locus. A more likely explanation is that RIP mutations ‘leak’ from adjacent repetitive sequences. As RIP mutation is traditionally observed to be restricted to repetitive regions and not single copy regions, leakage of RIP mutation might occur within a relatively short distance of a RIP-affected repeat, as suggested by Fudal et al [22]. Indeed, this has been reported in N. crassa whereby leakage of RIP was detected in single copy sequences at least 930 bp from the boundary of neighbouring duplicated sequences [42]. This is consistent with our finding of a high frequency of RIP mutations in single copy regions of L. maculans with the degree of RIP mutation being proportional to the proximity of flanking repetitive elements. The potential ‘leakage’ of RIP mutations into closely linked effector genes highlights the power of this process to lead to major evolutionary changes to genes such as effectors that play an important role in the lifestyle of an organism.
All haplotypes associated with RIP alleles of AvrLm6 and LmCys1 appeared to have a single origin indicating that the event that led to the RIP occurred only once. Furthermore, haplotypes associated with the virulent AvrLm6-2 allele (cysteine substitution) arose as a single clade. These single origins suggest that events leading to both the RIP mutations and non-RIP amino acid substitutions are much rarer than those leading to the deletion mutations. The RIP event occurred after the breakdown of sylvestris resistance, as isolates with RIP mutations were not detected before this time. This rapid appearance of RIP mutations is not unprecedented; such mutations have been shown to occur after one generation in L. maculans. When transformants with several tandemly linked copies of a hygromycin resistance gene were crossed with a wild type strain, none of the progeny were resistant to hygromycin due to the presence of multiple stop codons generated by RIP mutations in the hygromycin resistance gene [43].
Mutations associated with resistance to azole fungicides in Mycosphaerella graminicola also are derived from a single origin. Resistance emerged only once following strong selection due to widespread use of azole fungicides [44]. In both the M. graminicola and L. maculans examples, a similar phylogeny was found despite differences in origin and type of evolutionary pressure. In L.maculans haplotypes associated with deletion alleles conferring virulence towards AvrLm1 and AvrLm6 appeared to have a polyphyletic origin. Isolates with these deletions were detected prior to 2004 when the resistance breakdown occurred, albeit at a much lower frequency than afterwards. Haplotypes associated with deletion of both AvrLm1 and AvrLm6 might have been derived directly from the ancestral wild type rather than via the deletion of one gene followed by that of the second.
Deletions, RIP mutations and non-RIP amino acid substitutions conferred virulence at the AvrLm6 locus, whilst only deletions were responsible for virulence at the AvrLm1 locus. Similar types of mutations were detected in French populations of L. maculans isolates [22]. The finding of a virulence allele of AvrLm6 arising from a non-synonymous, non-RIP like mutation, of a glycine to cysteine substitution, was intriguing. In other avirulence proteins the loss rather than the gain of cysteine through non-synonymous substitutions confers virulence. For example, in the AVR4 protein of Cladosporum fulvum, loss of cysteine residues renders the isolate virulent towards tomato cultivars with the corresponding Cf-4 resistance gene [45]. The AvrLm6-2 allele of L. maculans gives rise to a protein with seven rather than six cysteine residues in the protein encoded by AvrLm6-0. This latter protein is proposed to have two disulphide bridges between C109 and C130, and C103 and C122 [8] on the basis of the SCRATCH disulphide bond prediction program [46]. The program predicts that the presence of the additional cysteine (C123) in the AvrLm6-2 allele would result in a third disulphide bridge, between C26 and C123.
As well as the mechanisms leading to inactivation of alleles described above, some of the proteins were undergoing non-RIP amino acid substitutions which did not lead to a change in phenotype. Some of these mutations, in AvrLm1 and LmCys1, were the results of positive selection, which favours new mutations that confer a fitness advantage and thus lead to an increase in gene diversity [12], [47]. Positive selection has been detected in pathogen effector genes including the avirulence gene that encodes NIP in Rhynchosporium secalis [47] and in genes encoding host-specific toxins such as S. nodorum ToxA [36], [48]. In contrast to AvrLm1, AvrLm6 and LmCys1, the remaining genes in the 520 kb AT- rich region, including LmCys2 showed very little variation. Despite positive selection driving amino acid substitutions within some of the effector-like proteins, deletion and RIP mutations are by far the major mechanisms leading to virulence at the AvrLm1 and AvrLm6 loci.
Stubble of cultivars of B. napus and B. juncea infested with L. maculans was collected each year from 1997 to 2008 from 25 locations across Australia (Table S1). For instance, stubble from a crop sown in 2003 was collected from the field in 2004 and isolates were then cultured from it. Cultivars with ‘polygenic’ resistance (Beacon, Dunkeld, Emblem, Grace, Hyden, Jade, Pinnacle, Skipton, Pinnacle and Tornado TT) had one or more Rlm genes, but none had Rlm1 nor Rlm6. The identity of resistance genes in some of these cultivars has been reported [44]. The category of ‘sylvestris resistance’ refers to cultivars (Surpass 400, Surpass 501TT, Surpass 603CL, 45Y77 and 46Y78) with resistance derived from B. rapa spp. sylvestris [14] have Rlm1 and RlmS [17]. The category of ‘juncea’ resistance refers to cultivars and lines of B. juncea (cv. Dune and lines JC05002, JC05006 and JC05007). Stubble of the latter two categories was not collected prior to 2004. From 2004 onwards although cultivars with sylvestris resistance were withdrawn from sale, these lines were grown in yield trials across Australia, and stubble was collected from them. Isolates (287) were cultured from individual ascospores discharged from stubble collected the previous year as described previously [15]. In addition, eight Australian isolates collected in 1987 and 1988 were analysed. All isolates were maintained on 10% Campbell's V8 juice agar.
Virulence of a subset of isolates was tested on three B. napus and one B. juncea cultivars. The B. napus cv. Beacon and cv. Q2 are susceptible controls that all isolates could attack. Cultivar Columbus contains Rlm1 and Rlm3 and B. juncea cv. Aurea contains Rlm5 and Rlm6 [3]. No Rlm1-only or Rlm6-only cultivars were available. Cotyledons of 14-day old seedlings were wounded and inoculated with conidia of individual isolates representing different haplotypes for AvrLm1 and AvrLm6. Symptoms were assessed at 10, 14 and 17 days post-inoculation (dpi) and pathogenicity scores determined at 17 dpi by scoring lesions on a scale from 0 (no darkening around wounds) to 9 (large grey-green lesions with profuse sporulation). Mean pathogenicity scores (determined from 40 inoculation sites) ≤3.9 were assigned as an avirulent phenotype whilst scores ≥4.0 were assigned as a virulent phenotype [17].
Non-coding, non-repetitive regions in the AvrLm1-LmCys2 genomic region and genes 3′ of AvrLm6 (Figure 1) were identified using published information [22] and by BLAST searches. Primers were designed upstream and downstream of start and stop codons to allow analysis of the sequences of entire open reading frames. For transcriptional analyses, primers were designed to amplify a 500–700 bp region of the coding sequence, flanking an intron where possible. All primers were designed using the program Primer3 [49] (Table S7). Primers to amplify the mating type locus and actin have been described previously [8], [24].
The sequence information for all genes has been deposited in GenBank with the following accession numbers; AvrLm1, AM084345 [1], AvrLm6, AM259336 [2], LmCys1, GU332625, LmTrans, GU332626, LmGT, GU332627, LmMFS, GU332628 and LmCys2, GU332629.
Genomic DNA was isolated from mycelia. Conditions for all PCR experiments were 95°C for 3 min; 35 cycles of 95°C for 30 sec, 59°C for 30 sec and 72°C for 1 min; 72°C for 6 min. PCR products were purified using QIAquick PCR purification kit (Qiagen) and sequenced using BigDyeTM terminator cycling conditions. Sequences were analysed using Sequencher v 4.0.5. Deletion genotypes were assigned if no band was produced following amplification with the AvrLm1, AvrLm6 or LmCys2-specific primers. Amplification of the mating type locus was a positive control for DNA quality. In a subset of eight isolates, deletion alleles were confirmed by Southern analysis of genomic DNA that had been digested with HindIII and hybridised with the appropriate probe (Figure S1). Additionally, PCR screens were performed on genomic DNA from two (for LmCys2) to 25 (for AvrLm6) isolates using multiple primer sets that amplify specific regions of the AvrLm1, AvrLm6 and LmCys2 gene regions. These amplifications confirmed that the entire locus was deleted in all isolates tested (data not shown).
Allele sequences were analysed by RIPCAL for the presence of RIP mutations [25]. RIPCAL generates a RIP dominance score, which is the frequency of the dominant dinucleotide RIP mutation (in this case CpA→TpA) relative to the sum of the alternative mutations (CpC→TpC, CpG→TpG and CpT→TpT). Sequences with RIP dominance scores >1 are considered to be highly RIP-affected. The ‘model’ sequence used for all RIPCAL analyses was the ‘wild type’ allele (designated with a -0 suffix) from the isolate (v23.1.3) whose genome has been sequenced. The spatial distribution of RIP was assessed for each gene and four non-coding regions by comparing the ratio of mutated CpA or TpG sites detected by RIPCAL, relative to the number of available CpA and TpG nucleotides present within the wild type allele over a 100 bp rolling window.
Ten infected cotyledons of B. napus cv. Beacon were harvested at 7 dpi. Necrotic tissue surrounding the inoculation wounds of each cotyledon was harvested using a hole punch (diameter 8 mm). Total RNA was purified from this tissue using the RNeasy Plant Mini Kit (Qiagen) and was treated with DNaseI (Invitrogen) before cDNA was synthesized using a first strand cDNA synthesis kit and an oligo-dT primer. End point RT-PCR was used to assess expression of AvrLm1, AvrLm6, LmCys1, LmCys2 and actin.
Quantitative RT-PCR was used to determine levels of expression of AvrLm1, AvrLm6, LmCys1 and LmCys2 in planta and in vitro culture. RNA was prepared from seven day old cultures of isolates with the wild type alleles of these genes, which were growing in 10% V8 juice. Additionally RNA was prepared from cotyledons of B. napus cv. Beacon 7 dpi infected with isolates with the wild type alleles of these genes. Total RNA and cDNA synthesis was performed as described above. Controls lacking reverse transcriptase were included. Quantitative RT-PCR was performed using Rotor-Gene 3000 equipment (Corbett Research, Australia) and QuantiTect SYBR Green PCR kit (QIAgen). A standard curve of amplification efficiency of each gene was generated from purified RT-PCR products [50]. Diluted RT product (1 μl) was added to 19 μl of PCR mix and subjected to 40 cycles of PCR (30 s at 94°C, then 60°C and then 72°C). All samples were analysed in triplicate. The amplified product was detected every cycle at the end of the 72°C step. Melt curve analysis after the cycling confirmed the absence of non-specific products in the reaction. The fluorescence threshold (Ct) values were determined for standards and samples using the Rotor-Gene 5 software. Ct values were exported to Microsoft Excel and analysed [51]. Actin was used as a reference gene.
Deviation from a constant rate of molecular evolution within the data sets (i.e. a “molecular clock”) was assessed using the phylogeny-based likelihood ratio test (LRT) implemented in the program HyPhy [27]. To estimate the contribution of the RIP alleles, likelihoods were calculated both for the total data sets and for data sets excluding RIP alleles. MEGA was also used to infer relative rates of sequence evolution by calculating means of genetic distances (Kimura-2-Parameter) between haplotypes.
Evidence for non-neutral evolution was assessed using two complementary approaches by comparing the rate of non-synonymous substitutions with the rate of synonymous substitutions (dN/dS = ω). Firstly, the analysis was based on the “sitewise likelihood-ratio” method as implemented in the SLR software package [52]. The test consists of performing a likelihood-ratio test on a site-wise basis, testing the null model (neutrality, ω = 1) against an alternative model ω≠1 (i.e. purifying selection ω<1; positive selection ω>1). Secondly, dN/dS = ω was tested using a phylogenetic analysis based on maximum likelihood as implemented in the PAML software package [53]. Two codon substitution models were compared via likelihood ratio tests (LRT). The comparison included the likelihood estimates of the neutral null model (M7) and the alternative model of positive selection (M8). RIP alleles were excluded from these analyses since such alleles encode sequences with stop codons.
To test different hypotheses of emergence of haplotypes associated with RIP alleles (Table S6), tree topologies using concatenated DNA sequences of all the genes (AvrLm1, AvrLm6, LmCys1, LmTrans, LmGT, LmMFS and LmCys2) and non-coding, non-repetitive regions (NC1-4) were generated and compared using the Kishino–Hasegawa (KH) test [29] as implemented in PAUP* 4.0b 10. Since the RIP mechanism produces the same mutations at specific sites, it is likely that formerly unrelated nucleotide sequences converge, leading to the false impression of similarity due to common descent. To avoid this bias, all CpA to TpA and TpG to TpA nucleotide changes were removed from the data set prior to inferring the phylogenetic relationships of haplotypes. Two alternative hypotheses were then compared; (i) haplotypes containing RIP alleles were monophyletic, i.e. they emerged only once. Trees representing this hypothesis were “constrained” by restricting RIP alleles to cluster only amongst each other (ii) haplotypes containing RIP alleles were polyphyletic, i.e. they emerged several times independently. Trees representing this alternative hypothesis were “unconstrained”, i.e. the pairing of particular alleles in the topology was not restricted. One thousand trees were generated representing each hypothesis, and the probabilities (P) of obtaining better trees were assessed using two-tailed tests, the full optimization criterion and 1000 bootstrap replicates. The KH test was conducted for trees constructed under the maximum likelihood and the maximum parsimony criterion.
The phylogenetic relationship among isolates based on the concatenated DNA sequences of all genes and non-coding, non-repetitive regions was assessed by PAUP* using maximum likelihood. Tree searches were conducted with the “fast-stepwise-addition” option and 1000 bootstrap replicates to assess statistical significance of nodes. The GTR-model with estimated substitution-rate matrix was used to evaluate molecular rate constancy.
|
10.1371/journal.ppat.1000396 | Naturally Occurring Lipid A Mutants in Neisseria meningitidis from Patients with Invasive Meningococcal Disease Are Associated with Reduced Coagulopathy | Neisseria meningitidis is a major cause of bacterial meningitis and sepsis worldwide. Lipopolysaccharide (LPS), a major component of the Gram-negative bacterial outer membrane, is sensed by mammalian cells through Toll-like receptor 4 (TLR4), resulting in activation of proinflammatory cytokine pathways. TLR4 recognizes the lipid A moiety of the LPS molecule, and the chemical composition of the lipid A determines how well it is recognized by TLR4. N. meningitidis has been reported to produce lipid A with six acyl chains, the optimal number for TLR4 recognition. Indeed, meningococcal sepsis is generally seen as the prototypical endotoxin-mediated disease. In the present study, we screened meningococcal disease isolates from 464 patients for their ability to induce cytokine production in vitro. We found that around 9% of them were dramatically less potent than wild-type strains. Analysis of the lipid A of several of the low-activity strains by mass spectrometry revealed they were penta-acylated, suggesting a mutation in the lpxL1 or lpxL2 genes required for addition of secondary acyl chains. Sequencing of these genes showed that all the low activity strains had mutations that inactivated the lpxL1 gene. In order to see whether lpxL1 mutants might give a different clinical picture, we investigated the clinical correlate of these mutations in a prospective nationwide observational cohort study of adults with meningococcal meningitis. Patients infected with an lpxL1 mutant presented significantly less frequently with rash and had higher thrombocyte counts, consistent with reduced cytokine induction and less activation of tissue-factor mediated coagulopathy. In conclusion, here we report for the first time that a surprisingly large fraction of meningococcal clinical isolates have LPS with underacylated lipid A due to mutations in the lpxL1 gene. The resulting low-activity LPS may have an important role in virulence by aiding the bacteria to evade the innate immune system. Our results provide the first example of a specific mutation in N. meningitidis that can be correlated with the clinical course of meningococcal disease.
| Neisseria meningitidis is a Gram-negative bacterium that can cause the life-threatening diseases meningitis and sepsis. Most of the symptoms seen in these diseases are the result of excessive stimulation of the immune system by lipopolysaccharide (LPS), a major component of the outer membrane of Gram-negative bacteria. The structure of the membrane-anchoring part of LPS, named lipid A, determines how well it is recognized by TLR4, the receptor which triggers this response. N. meningitidis has been shown previously to make a highly active form of LPS with six fatty acyl chains in its lipid A moiety. We now show that a surprisingly large fraction of meningococcal patient isolates have a mutation in one of their lipid A biosynthesis genes, which makes their LPS underacylated and which strongly decreases their capability to activate the immune system. Presumably, selection of these mutants is promoted because the reduced activation aids the bacteria to escape elimination within the human host. We further demonstrate that meningitis patients infected with these LPS mutants have less severe symptoms than patients infected with wild-type strains. This is the first instance where a specific bacterial mutation can be correlated with the clinical course of meningococcal disease.
| Neisseria meningitidis is a major cause of bacterial meningitis and sepsis worldwide [1]. While it is a frequent commensal of the human upper respiratory tract, in some individuals the bacterium spreads to the bloodstream causing meningitis and/or sepsis, serious conditions with high morbidity and mortality. As in all Gram-negative bacteria, lipopolysaccharide (LPS) is a major component of the outer membrane of N. meningitidis. It is now well established that LPS is sensed by mammalian cells through Toll-like receptor 4 (TLR4), in combination with coreceptors MD-2 and CD14 [2]. Activation of this complex leads to recruitment of the adapters MyD88, Mal, TRIF, and TRAM to the cytoplasmic domain of TLR4 [3]. These adapters initiate signal transduction pathways that lead to induction of innate immunity. These pathways are classified in a so called “MyD88-dependent” pathway involving MyD88 and Mal, and a “MyD88-independent” pathway involving TRIF and TRAM. Hallmarks of MyD88-dependent and MyD88-independent signaling are induction of pro-inflammatory cytokines and type I IFN respectively. While the response to LPS can be beneficial to the host by containing a beginning infection, it can also be detrimental when excessive stimulation occurs through growth of large numbers of bacteria in the bloodstream as happens during sepsis [2],[4],[5].
TLR4 recognizes the lipid A moiety of the LPS molecule [2]. The chemical composition of the lipid A determines how well it is recognized by TLR4 and consequently it determines the biological activity of the LPS. N. meningitidis has been reported to produce lipid A with six acyl chains, the optimal number for TLR4 recognition [6]. Indeed purified LPS of this bacterium is highly active and plasma concentrations of LPS in patients with meningococcal disease correlate strongly with mortality risk [7]. LPS is also important in the activation of the coagulation system through upregulation of tissue factor. Excessive activation of the coagulation system can lead to disseminated intravascular coagulation (DIC), the most feared complication of invasive meningococcal disease [1]. DIC is clinically characterized by hypotension, petechial rash, and depletion of thrombocytes and coagulation factors.
Uniquely among Gram-negative bacteria, N. meningitidis can grow without LPS, as was shown by us when we constructed a mutant with an inactivated lpxA gene, required for the first step in LPS biosynthesis [8]. In addition, we have previously shown that insertional inactivation of the lpxL1 or lpxL2 genes required for addition of secondary acyl chains leads to reduced biological activity of meningococcal LPS [9],[10]. The possibility that such mutations might also occur naturally was suggested to us by a report showing that the group Y strain HF13 was defective in signaling through the MyD88-independent pathway and TLR4 [11].
Here we report that strain HF13 has penta-acylated lipid A due to a mutation in its lpxL1 gene. Screening of a selection of clinical isolates revealed lpxL1 mutations in approximately 13% of meningococcal disease isolates of all major serogroups and clonal complexes. Several different kinds of mutations were found. We also found evidence for on-and-off switching of lpxL1 in vivo in humans. Importantly, patients with meningococcal meningitis that were infected with an lpxL1 mutant strain had less severe systemic inflammation and reduced coagulopathy.
Mogensen et al. demonstrated that the serogroup Y strain HF13 is defective in TLR4 activation and initiation of MyD88-independent signaling [11]. Reduced biological activity of meningococcal LPS is associated with altered lipid A structure [9],[10]. Therefore, the lipid A structure of strain HF13 was assessed by mass spectrometry (Figure 1A). The spectrum shows major peaks that correspond with lipid A with only five acyl chains. One of the two secondary C12 acyl chains is absent, but the spectrum is not conclusive on which one, since the C12 acyl chains have the same mass. This result implies that in strain HF13 either lpxL1 or lpxL2 is inactive, as we previously found that the addition of the secondary C12 acyl chains to lipid A requires active lpxL1 and lpxL2 [9]. Sequence analyses of both genes showed a normal lpxL2 sequence, but the lpxL1 sequence contained one adenosine deletion in a poly adenosine tract, leading to a frameshift and a premature stop of the translated protein (Figure 2, Table 1).
The inactivated lpxL1 gene in strain HF13 results in a penta-acylated lipid A lacking the secondary acyl chain at the 2′-position in lipid A, while N. meningitidis typically has a hexa-acylated lipid A (Figure 1B). These results provide an explanation for the inability of strain HF13 to activate TLR4 and to initiate MyD88-independent signaling.
To evaluate the distribution of lpxL1 mutations among meningococcal isolates from patients, we initially screened a panel of 56 serogroup Y meningococcal isolates for their capacity to induce the MyD88-independent cytokine IP-10 in the mouse macrophage cell line J774A.1 (Figure S1). As controls, strain H44/76 and HF13 were included. Of 56 serogroup Y isolates, eight strains induced like HF13 little or no IP-10. Sequence analyses of lpxL1 of these isolates revealed that they all had mutations in lpxL1, resulting in an inactive gene. Five strains had one adenosine deletion in a poly A tract just like strain HF13 (type V mutation, Figure 2, Table 1), two strains had a deletion of ten nucleotides (type VI mutation), and one strain had an insertion of the insertion element IS1301 (Type I mutation).
These results prompted us to investigate the distribution of lpxL1 mutations among meningococci of the major serogroups and clonal complexes. Previously, we have shown that at higher dilutions an lpxL1 mutant induces less pro-inflammatory cytokines than wild-type N. meningitidis [9],[10]. To identify meningococcal isolates with mutations in lpxL1, isolates were tested on their capacity to induce IL-6 in the human monocytic cell line Mono Mac 6 (MM6). Of 114 isolates, representing all major serogroups and clonal complexes, 13 were found to induce low amounts of IL-6 (Figure S2). Sequence analyses of lpxL1 showed that 12 isolates had a mutation in lpxL1, rendering the gene inactive (Figure 2, Table 1). Of these strains, 10 had an insertion or deletion in a polyadenosine or polyguanosine tract (type III, IV and V mutations); six of these had the same mutation as found in the majority of mutant serogroup Y strains. One strain had a type VI mutation, like in the two aforementioned serogroup Y strains. One strain had a deletion of the C-terminal part of the gene (type VII mutation). The remaining strain (2020799) had apparently no mutation in lpxL1 that would lead to its inactivation. However, closer examination of its putative amino acid sequence showed that one amino acid was altered at a position conserved in all known lpxL1 homologues. Therefore, the LpxL1 protein of this strain is probably nonfunctional. Indeed, we confirmed that strain 2020799 had penta-acylated lipid A by mass spectrometry (data not shown). As a control, also lpxL1 of 34 strains that induced a normal level of IL-6 was sequenced. As expected, these strains had no mutations in lpxL1 (data not shown). Together, seven unique lpxL1 mutations were found among this panel of different serogroups and different clonal complexes, indicating that inactivation of lpxL1 must have occurred multiple times independently. The results show that lpxL1 mutations are not associated with serogroup or clonal complexes and occur also among the serogroup B and C strains, which are prevalent among isolates from patients with meningococcal disease in Europe.
Most of the identified lpxL1 mutations were in nucleotide repeats of adenosines and guanosines, the type III, IV and V mutations (Figure 2, Table 1). These sequences are prone to cause slippage of the DNA polymerase during DNA replication, leading to reversible frameshift mutations. This slipped-strand mispairing is the most common mechanism of translational phase variation, the process of random and reversible on-and-off switching of a gene. Phase variation creates a phenotypically diverse population, allowing the bacterium to adapt to different microenvironments within the human host. To investigate whether N. meningitidis can switch lpxL1 on-and-off we screened a panel of strains obtained from different anatomical locations within individual patients: isolates from the blood and/or cerebrospinal fluid (CSF) as well as from the throat and/or nose of 40 patients were used. The MM6 cell line was stimulated with these strains and IL-6 production was measured with ELISA. Three strains induced low levels of IL-6 compared to wild-type N. meningitidis (Figure S3). These isolates were from three different patients. Two strains were isolated from the cerebrospinal fluid and one strain was isolated from the throat. The other isolates of these patients induced normal levels of IL-6. The lpxL1 genes of all isolates of these three patients were sequenced and found to be mutated in the isolates that induced low IL-6, but not in the isolates that induced normal IL-6 (Figure 2, Table 1). Two strains had a type IV mutation, which potentially is reversible. The third strain had a point mutation leading to substitution of a conserved amino acid. These results suggest that in the host the expression status of lpxL1 of meningococci is subject to phase variation.
The identified lpxL1 mutations occurred in strains of widely varying genetic background, and it is therefore conceivable that other factors besides altered LPS contribute to their reduced cytokine induction. To investigate this, titrations of four of the spontaneous lpxL1 mutants were compared in their capacity to induce cytokines in MM6 cells with titrations of our previously constructed lpxL1 knockout mutant and its parent strain H44/76, as well as the completely LPS-deficient strain pLAK33 (Figure 3). Clearly, the LPS-deficient strain pLAK33 is much less potent in inducing IL-6 than the wild-type strain H44/76. IL-6 induction by the constructed lpxL1 mutant is similar to that by pLAK33 and the four lpxL1 mutants isolated from patients.
To demonstrate that the lpxL1 mutants induced less cytokines than wild-type strains because their LPS is less well recognized by the LPS receptor complex, titrations of a similar panel of strains was used to stimulate HEK293 cells transfected with human TLR4, MD-2, and CD14. Activation of the receptor complex was assessed by measuring IL-8 production (Figure 4). Wild-type strain H44/76 was much more efficient in TLR4 activation than the mutants. All lpxL1 mutants, either constructed or isolated from patients, showed a similar decrease in IL-8 induction, while the LPS-deficient pLAK33 cells were even less active. Together, these results demonstrate that the lpxL1 mutants activate human TLR4 less efficiently, and this is the sole reason for their reduced biological activity.
We have shown that lpxL1 mutants induce less cytokines in human and murine cell lines. However, these in vitro models do not necessarily represent the situation in vivo and do not take into account the genetic diversity of the human population. To mimic a systemic meningococcal infection more closely, also human peripheral blood mononuclear cells (PBMCs) of several donors were stimulated with titrations of a selection of N. meningitidis strains. After stimulation, concentrations of IL-6, TNF-α, and IL-1β were determined in the supernatant (Figure 5). These pro-inflammatory cytokines are known to mediate the toxic effects of LPS [2]. In all donors, wild-type strain H44/76 induced much more IL-6, TNF-α, and IL-1β than the mutants. Overall, the constructed and spontaneous lpxL1 mutants showed a similar reduction in cytokine induction.
We next explored whether infection with lpxL1-mutant meningococcal strains was associated with a particular clinical phenotype. The meningococcal isolates from 254 patients from a prospective nationwide observational cohort study of 696 adults with community-acquired bacterial meningitis in the Netherlands (period, 1998–2002) [12],[13] were analyzed for their ability to induce IL-6. Of the 254 isolates, 172 (68%) were of serogroup B, 78 (31%) of serogroup C, 3 (1%) of serogroup Y, and one (<1%) of serogroup W135. Multilocus sequence typing showed 91 unique sequence types. The most prevalent clonal complexes were cc41/44 (41%), cc11 (24%), and cc32 (16%) [13].
MM6 cells were stimulated with these strains and IL-6 induction was assessed (Figure S4). The isolates of 17 patients (7%) showed a decreased IL-6 induction and sequencing revealed mutations in lpxL1 in all but one (Figure 2, Table 1). Twelve isolates had a type III, IV or V mutation. Three strains had a point mutation leading to substitution of an essential amino acid, and one strain had an IS1655 insertion. In one strain (992008) we were unable to identify a mutation in lpxL1 that could lead to gene inactivation or inactive gene product. Further analyses with mass spectrometry to determine the mass of its lipid A and silver staining of a Tricine-SDS-PAGE gel to analyze the size and quantity of its LPS, demonstrated that LPS was not detectable in this strain (results not shown). The responsible mutation remains to be identified. There were no overall differences in lpxL1 mutation frequency between serogroups (P = 0.85) and clonal complexes (P = 0.56).
Next, we correlated results of the mutation analysis with clinical data (Table 2) [12],[13]. Patients infected with lpxL1 mutant strains tended to be younger (P = 0.053) and to present less frequently with fever (P = 0.057). None of the patients infected with an lpxL1 mutant strain presented with hypotension and these patients had correspondingly lower levels of serum creatinine. They were less likely to present with rash compared with those infected with wild-type meningococci (5/16 [31%] vs. 157/236 (67%); P = 0.006; Figure 6) and had higher platelet counts (P = 0.005). Rash was strongly related with lower platelet counts (P<0.0001). To investigate the possibility that the clinical differences found between the two patient groups were confounded by the different ages of the patient groups, a multivariate analysis adjusting for age was performed. The difference in platelet count (P = 0.003) and rash (P = 0.004) remained statistically significant after adjusting for age. Subgroup analysis of clonal complex 41/44 showed similar results. The differences in platelet count (P = 0.007), rash (P = 0.006), and age (P = 0.053) between patients infected by mutant and wild-type strains were also present in the subgroup of clonal complex 41/44.
None of the patients infected with lpxL1 mutant strains developed septic shock during clinical course, while 13% of the wild-type-infected patients did. One patient infected with an lpxL1 mutant strain died of respiratory failure after multiple seizures. By contrast, sepsis was the leading cause of death among patients infected with wild-type meningococci (14 of 16 fatalities, 88%).
Thus, the lpxL1 mutation occurs frequently among meningococci causing meningitis. Patients infected by mutant strains have a clinical phenotype consistent with less systemic inflammation and reduced activation of the coagulant system.
Overall, we screened meningococcal isolates of 464 different patients and identified 40 strains with an lpxL1 mutation. An additional low-activity strain had no lpxL1 mutation but appeared to be LPS-deficient; the responsible mutation is currently under investigation. Thus, 8.6% of patients were infected with an lpxL1 mutant, which is surprisingly common. There are several lines of evidence making it very likely that lpxL1 mutants arise spontaneously in the host instead of introduction of an lpxL1 mutation into one or several clones and subsequent spreading among the meningococcal population. Firstly, lpxL1 mutations exist in isolates of many serogroups and clonal complexes. Secondly, we identified 12 unique mutations in lpxL1. Thirdly, most lpxL1 mutations (71%) are due to frameshifts in homopolymeric nucleotide tracts, making phase variation likely. Finally, we found evidence for switching from wild-type to lpxL1 mutant in vivo in patients for which multiple isolates were available.
A picture emerges of N. meningitidis modulating its lipid A structure under selective pressure. Under some conditions, hexa-acyl lipid A has to be beneficial to compensate for enhanced recognition by the innate immune system. Lipid A with six acyl chains can protect bacteria from the antibacterial molecules in mucosal secretions, consistent with the observation that many bacteria inhabiting the respiratory tract and gut still produce hexa-acyl LPS [14]. Chronic inflammation of these environments due to LPS stimulation is probably prevented because epithelial cells express low levels of either TLR4, MD-2, or CD14 at the mucosal surface. On the other hand, the submucosal spaces are normally sterile and the defense cells present there, such as macrophages, dendritic cells, and neutrophils, express all the components of the LPS receptor complex and can therefore respond potently after an encounter with a Gram-negative bacterium [14]. Perhaps for this reason most species of Gram-negative bacteria with hexa-acyl lipid A that inhabit the mucosal surfaces rarely become invasive. On the other hand, many Gram-negative pathogens that cause systemic infection do not produce hexa-acyl lipid A. Most of these bacteria have other habitats than the mucosa and enter the body via nonmucosal routes [6]. A good example is the plague bacillus Yersinia pestis. At mammalian body temperature Y. pestis normally produces tetra-acyl LPS that is poorly recognized by TLR4. Interestingly, a modified strain that produced hexa-acyl LPS at 37°C was no longer virulent in wild-type mice but fully virulent in TLR4-deficient mice, demonstrating the importance of evasion of TLR4 activation for this bacterium [15]. N. meningitidis seems to be one of the exceptions to the general rule that Gram-negative bacteria with hexa-acyl lipid A do not cause systemic disease. However, our observation that a proportion of clinical isolates have penta-acylated LPS suggests that evasion of TLR4 activation might aid the bacterium to circumvent host defences after crossing the nasopharyngeal epithelium. The hypothesis that TLR4 plays an important role in the prevention of meningococcal disease corroborates with the finding that subjects with rare TLR4 mutations have an increased risk for developing the disease [16]. If the assumption is correct that hexa-acyl LPS gives the bacterium an advantage on mucosal surfaces and that non hexa-acyl LPS is better for bacteria in submucosal spaces, one would expect that the frequency of lpxL1 mutants is lower in meningococcal isolates from the respiratory tract compared to meningococcal isolates from the cerebrospinal fluid or blood.
Mogensen et al. showed that strain HF13 is specifically defective in activation of the MyD88-independent pathway, but not in inducing the MyD88-dependent pathway [11]. However, we demonstrate that strain HF13 and other lpxL1 mutants are also defective in inducing the MyD88-dependent cytokines IL-6, TNF-α, and IL-1β. Our experiments indicate that lpxL1 mutants or purified lpxL1 LPS compared to wild-type controls are not specifically deficient in inducing the MyD88-dependent vs. independent pathway. This apparent discrepancy might be explained by the dose of bacteria used. If cells are stimulated with a high dose of bacteria the difference between lpxL1 mutant and wild-type is only detectable for the MyD88-independent pathway. This is because LPS is the only bacterial component capable of inducing the MyD88-independent pathway, while many other bacterial components can induce the MyD88-dependent pathway (e.g. TLR2 ligands). When cells are stimulated with lower doses of bacteria the difference in induction of the MyD88-dependent pathway becomes apparent, because LPS is by far the most active component of the bacterium and the other non-TLR4 ligands that can activate the MyD88-dependent pathway are diluted too far to be still active.
The relatively high frequency of phase variation raises the question whether the lpxL1 mutations might have arisen in vitro after isolation from the patient. Previously, we have performed extensive research on the phase variation of porA in N. meningitidis. In this gene, homopolymeric nucleotide tracts are found in the promoter (polyguanidine) and in the coding region (polyadenine). The frequencies by which these sequences vary in length are 10−3 [17],[18]. Others showed phase variation of capsule expression caused by insertion of IS1301 in the siaA gene with a frequency of phase variation of 9×10−4 [19],[20]. In vitro selection of porA phase variants and siaA phase variants have not been reported. Meningococcal isolates received by the Netherlands Reference Laboratory for Bacterial Meningitis (NRLBM) are low passages (up to 2 passages). We sequenced the lpxL1 gene of 20 individual colonies of a culture of a mutant isolate (971859 I) and of 25 individual colonies of a culture of isolate 971859 III and found in each instance the same sequence, i.e. 20 mutant sequences and 25 wild-type sequences, respectively. Therefore, we estimate the frequency of phase switching to be less than 2.2×10−2. In addition, we sequenced lpxL1 of DNA extracted from a swap taken from 4 different quadrants of another culture plate of isolate 971859 III. All 4 lpxL1 sequences were homogeneous and identical. Thus we are confident that the discovered lpxL1 mutations are not caused by in vitro phase variation.
Infection with lpxL1-mutant meningococcal strains is associated with a particular clinical phenotype, which consisted of less systemic inflammation and reduced activation of the coagulant system, reflected in less fever, higher serum platelet counts, and lower numbers with rash. Moreover, our in vitro data have shown that lpxL1 mutants induce much less pro-inflammatory cytokines than wild-type strains. The coagulation system is activated through upregulation of tissue factor [1]. It has been demonstrated that LPS upregulates tissue factor on monocytes and endothelial cells [21]–[23]. Furthermore, in particular the pro-inflammatory cytokine IL-6 appears to mediate in vivo expression of tissue factor [24],[25]. Finally, IL-1β and TNF-α inhibit anticoagulant pathways by downregulating thrombomodulin at the endothelial surface and by increasing plasminogen activator inhibitor type-1 (PAI-1) [26],[27]. Thus, our finding that patients infected with an lpxL1 mutant show less activation of the coagulation system is consistent with our results that show that lpxL1 LPS is less potent and that lpxL1 mutants induce less pro-inflammatory cytokines.
Remarkably, the lpxL1 mutants induced the same degree of CSF leukocytosis as wild-type strains. There are several explanations for “normal” CSF white cell counts in patients infected by mutant strains. Patients in the cohort all had positive CSF cultures; almost all had clinical signs of meningitis and CSF leukocytosis. Likely, leukocytosis is not only mediated by lipid A, but also by other microbial constituents.
It should be noted that not all groups of patients were included in our analysis of clinical patient data. The study only included adults with meningitis. Patients younger than 16 years or patients with sepsis only were not included. Therefore, our results are potentially biased by excluding these patient groups. Patients with meningitis often have a less severe form of the disease, as reflected by the overall low mortality of 8% in our study. However, patients with sepsis have very serious symptoms resulting from high concentrations of bacteria in the circulation. Mortality rates in these patients can be as high as 50%. Also, patients younger than 16 years are an import group, because rates for meningococcal disease are highest for young children [1]. It would be interesting to see whether lpxL1 mutants also exist in these patients groups, and if so, if these patients have a different clinical course compared to patients infected with a wild-type strain. These additional data are needed to fully understand the impact of lpxL1 mutations on meningococcal disease.
Meningococcal sepsis is generally seen as the prototypical endotoxin-mediated disease. Here we report for the first time that meningococcal lipid A mutants which are defective in TLR4 activation occur naturally. Their frequency is unexpectedly high, suggesting an important role in virulence for the resulting low-activity LPS. Our results suggest that in most cases this mutation has occurred through phase variation, and may give the bacteria an advantage because they are less well sensed by the innate immune system. Patients infected with these mutant strains endure milder symptoms with less systemic inflammation and reduced activation of the coagulant system, showing that our findings are clinically relevant. Importantly, these results with lpxL1 also provide the first example of a specific bacterial mutation which can be associated with the clinical course of meningococcal disease. More generally, it shows how there can be an underestimated heterogeneity in the TLR4-activating capacity of pathogenic bacteria.
This observational study with anonymous patient data was carried out in accordance with the Dutch privacy legislation. Written informed consent to use data made anonymous was obtained from the patient (if possible) or from the patient's legal representative.
Strain HF13 was a kind gift from M. Kilian. The constructed lpxA and lpxL1 mutants were generated in the H44/76 strain as previously described [8],[9]. All other strains were selected from the collection of the Netherlands Reference Laboratory for Bacterial Meningitis. Details about year of isolation, serogroup, genotype and anatomical site of isolation are presented in Table S1. Meningococci were cultured in GC broth or on GC plates (Difco laboratories) supplemented with 1% (vol/vol) Vitox (Oxoid) at 37°C in humified atmosphere of 5% CO2 [18]. Bacteria were suspended in PBS and the A620 was determined. The bacteria were heat inactivated at 56°C for 30 min. Serogrouping were performed as described elsewhere [12]. MLST was performed as described by Maiden et al [28].
Bacteria were grown as described above and suspended in isobutyric acid-ammonium hydroxide 1 M (5∶3, v/v). Lipid A was extracted as described previously [29] with slight modifications. The lipid A structure was analyzed by nanoelectrospray tandem mass spectrometry (MS/MS) on a Finnigan LCQ in the negative (MS) or positive (MS/MS) ion mode [30].
DNA was extracted from boiled cultures of N. meningitidis. Sequencing of lpxL1 was carried out using primers 344-2 and 670-1 (Table S2) and BigDyeTerminator chemistry (Applied Biosystems) according to the instructions of the manufacturer. The primers used to obtain sequences upstream and downstream of lpxL1 are presented in Table S2. Sequence traces were obtained with ABI Big-dyes and an ABI 3730 sequencer.
PBMC from HLA-oligotyped donors after leukapheresis were isolated by centrifugation of buffy coat cells on Ficoll-Hypaque (Pfizer) and were used after cryopreservation. For experiments and/or maintenance, the human monocyte cell line Mono-mac-6 (MM6), the mouse macrophage cell line J774A.1, and PBMCs were suspended in IMDM (Gibco BRL) supplemented with 100 units/ml penicillin, 100 µg/ml streptomycin, 300 µg/ml l-glutamine (Gibco BRL), and 10% heat-inactivated fetal calf serum (FCS) (Gibco BRL). For experiments and maintenance of HEK-293 cells stably transfected with human TLR4A, MD-2, and CD14 (Invivogen), DMEM (Gibco BRL) was used, supplemented with 10% FCS, 10 µg/ml blasticidin (Invivogen), and 50 µg/ml Hygromycin B (Invivogen).
Depending on the experiment either J774A.1, MM6, PBMCs, or HEK-293 hTLR4/MD-2/CD14 cells were used. Different plates and quantities of cells were used: 1.106 cells in 1 ml medium per well in 12-well plates, 9.104–5.105 cells in 250–1000 µl medium per well in 24-well plates, and 1.105–3.105 cells in 200–300 µl medium per well in 96-well plates. Cells were stimulated with bacteria and incubated o/n at 37°C in a humidified atmosphere containing 5% CO2. Cytokine concentrations in the culture supernatants were quantified with ELISA. Mouse IP-10 was determined with mouse IP-10 ELISA kit (R&D systems) and human IL-6, TNF-α, IL-1β, and IL-8 with PeliPairTM reagent sets (Sanquin).
The Dutch Meningitis Cohort Study included 258 patients with meningococcal meningitis; from 254 patients the bacterial strain was stored in the Netherlands Reference Laboratory for Bacterial Meningitis [13]. Inclusion and exclusion criteria have been described extensively elsewhere [12]. In summary, eligible patients were older than 16 years, had bacterial meningitis confirmed by culture of cerebrospinal fluid (CSF), and were listed in the database of the Netherlands Reference Laboratory for Bacterial Meningitis from October 1998 to April 2002. This laboratory receives CSF isolates from about 85% of all patients with bacterial meningitis in the Netherlands. The treating physician was contacted, and informed consent was obtained from all participating patients or their legally authorized representatives. This observational study with anonymous patient data was carried out in accordance with the Dutch privacy legislation. Patients underwent a neurologic examination at discharge, and outcome was graded with the Glasgow Outcome Scale. This measurement scale is well validated with scores varying from 1 (indicating death) to 5 (good recovery). A favourable outcome was defined as a score of 5, and an unfavourable outcome as a score of 1–4. Focal neurologic deficits were defined as focal cerebral deficits (aphasia, monoparesis, or hemiparesis) or cranial nerve palsies. Serogrouping, MLST, and susceptibility testing of meningococcal isolates were performed by the Netherlands Reference Laboratory for Bacterial Meningitis.
The Mann-Whitney U test was used to identify differences between groups in continuous variables, and dichotomous variables were compared by the chi-square or Fisher exact test. All statistical tests were 2-tailed, and a p value less than 0.05 was regarded as significant.
Please see Table S3 for accession numbers.
|
10.1371/journal.ppat.1003667 | Assisted Evolution Enables HIV-1 to Overcome a High TRIM5α-Imposed Genetic Barrier to Rhesus Macaque Tropism | Diversification of antiretroviral factors during host evolution has erected formidable barriers to cross-species retrovirus transmission. This phenomenon likely protects humans from infection by many modern retroviruses, but it has also impaired the development of primate models of HIV-1 infection. Indeed, rhesus macaques are resistant to HIV-1, in part due to restriction imposed by the TRIM5α protein (rhTRIM5α). Initially, we attempted to derive rhTRIM5α-resistant HIV-1 strains using two strategies. First, HIV-1 was passaged in engineered human cells expressing rhTRIM5α. Second, a library of randomly mutagenized capsid protein (CA) sequences was screened for mutations that reduced rhTRIM5α sensitivity. Both approaches identified several individual mutations in CA that reduced rhTRIM5α sensitivity. However, neither approach yielded mutants that were fully resistant, perhaps because the locations of the mutations suggested that TRIM5α recognizes multiple determinants on the capsid surface. Moreover, even though additive effects of various CA mutations on HIV-1 resistance to rhTRIM5α were observed, combinations that gave full resistance were highly detrimental to fitness. Therefore, we employed an ‘assisted evolution’ approach in which individual CA mutations that reduced rhTRIM5α sensitivity without fitness penalties were randomly assorted in a library of viral clones containing synthetic CA sequences. Subsequent passage of the viral library in rhTRIM5α-expressing cells resulted in the selection of individual viral species that were fully fit and resistant to rhTRIM5α. These viruses encoded combinations of five mutations in CA that conferred complete or near complete resistance to the disruptive effects of rhTRIM5α on incoming viral cores, by abolishing recognition of the viral capsid. Importantly, HIV-1 variants encoding these CA substitutions and SIVmac239 Vif replicated efficiently in primary rhesus macaque lymphocytes. These findings demonstrate that rhTRIM5α is difficult to but not impossible to evade, and doing so should facilitate the development of primate models of HIV-1 infection.
| Retroviruses such as HIV-1 often exhibit limited capacity to infect species other than their natural hosts. This phenomenon is partly due to the existence of antiviral proteins that protect against infection by viruses that have not adapted to a particular species. For example, the resistance of rhesus macaques, the monkey species most commonly used in medical research, to HIV-1 infection is partly attributable to the vulnerability of HIV-1 to TRIM5α. Rhesus macaque TRIM5α (rhTRIM5α) blocks HIV-1 infection by recognition of the viral capsid following its entry into the cell, and it has proven difficult to derive HIV-1 strains that are resistant to rhTRIM5α. However, by devising an ‘assisted evolution’ approach, we identified particular combinations of mutations that render HIV-1 resistant to rhTRIM5α. These mutations enable HIV-1 to evade rhTRIM5α by abolishing recognition of the capsid. Notably, introduction of rhTRIM5α-resistant capsids into an HIV-1 that was also engineered to avoid the rhesus macaque APOBEC3 antiviral proteins, allowed efficient HIV-1 replication in rhesus macaque lymphocytes. These discoveries have the potential to advance the development of rhesus macaque models of HIV-1 infection.
| The narrow species tropism of HIV-1 is, in part, caused by species-specific variation in restriction factors that inhibit retroviral infection. This fact has important corollaries, one of which is that humans are likely protected from infection by many retroviruses. Conversely, many animal species commonly used in biomedical research cannot be infected by HIV-1, imposing severe limitations on the development of non-human primate models of HIV-1 infection and pathogenesis [1]. One antiretroviral protein that limits HIV-1 tropism is TRIM5α, a restriction factor that was initially identified in a screen of rhesus macaque (rh, M. mulatta) genes for inhibitors of HIV-1 infection [2], but in fact can have broad antiretroviral activity [3]–[8]. As a member of a large family of tri-partite motif (TRIM) containing proteins, TRIM5α contains N-terminal RING, B-box, and coiled-coil domains [9]. TRIM5α, like other restriction factors such as tetherin and APOBEC3 proteins, inhibits HIV-1 in a species-specific manner [10]–[17]. In particular, HIV-1 is vulnerable to restriction by rhTRIM5α but is resistant to restriction by human TRIM5α due to sequence differences in their C-terminal PRY/SPRY domains [18]–[23]. TRIM5α functions by targeting the incoming viral capsid within minutes of viral entry into the cell cytoplasm [24]. Viral capsids are likely directly bound by the PRY/SPRY domain during restriction [25]–[28] leading to inactivation of the viral core, followed by dissolution and, in some cases, degradation of viral core components [27], [29], [30].
In several instances, restriction factors target generic viral or cellular features that are required by retroviruses (such as lipid membranes, dNTP pools, and RNA) in a manner that is not directly affected by viral RNA or protein sequence. These nonspecific mechanisms by which restriction factors act makes viral escape difficult, and the major way that retroviruses evade such inhibitors is through the acquisition and adaptation of specific antagonists [31]. For example, the viral proteins Vpu, Nef, Vif and Vpx are able to rescue HIV-1 or SIV from restriction by human tetherin, APOBEC3 and SamHD1 proteins by targeting the restriction factors for removal from their sites of action, or for degradation [10], [11], [13]–[15], [32]–[38]. In contrast, TRIM5α proteins are unusual among the known antiretroviral restriction factors, because they bind to a specific viral protein (CA). While some TRIM5α proteins, particular those from old world monkeys have broad antiretroviral activity, and can inhibit retroviruses with widely divergent CA sequences [3], [6]–[8], the requirement that a specific protein be recognized affords viruses the opportunity to evade restriction by evolving a CA protein that cannot be recognized. Indeed, transplantation of SIVmac239 CA in the context of a chimeric HIV-1 confers rhTRIM5α resistance [8], [39]. Similarly, sensitivity or resistance of murine leukemia virus to human TRIM5α can be acquired by mutations in CA [5], [6], [40], [41]. Amino acids in CA that confer TRIM5α sensitivity are generally exposed on the surface of the viral core [42], [43] and are therefore accessible for binding to TRIM5α following viral entry into the cell.
While it should be possible for HIV-1 to evade rhTRIM5α via changes in CA sequence, it has proven difficult to isolate mutants of HIV-1 that have this property. A few studies have described mutations in HIV-1 CA that confer modestly reduced sensitivity to macaque (M. mulatta or M.fascicularis) TRIM5α proteins [44], [45]. Recently one study showed that replacing the entire predicted surface of HIV-1 CA with SIVmac239 sequences (twenty five amino acid changes) conferred insensitivity to rhTRIM5α, but at the cost of an ∼15-fold reduction in single-cycle replication fitness [46]. As yet, no study has reported the identification of mutants that give full or nearly full resistance to rhTRIM5α while retaining viral fitness. A possible reason for this is that some monkey TRIM5α proteins (including rhTRIM5α) are capable of recognizing retroviral CA proteins (e.g. HIV-1 and MLV) of very different sequence [3]–[8]. The molecular basis for this broad antiretroviral specificity, and consequently the difficulty with which rhTRIM5α is evaded, may be that rhTRIM5α interacts with multiple determinants on the 3-dimensional HIV-1 CA protein structure [46], [47]. It is also noteworthy that HIV-1 CA is the most genetically fragile (intolerant of amino acid substitution) protein for which robustness/fragility has been quantitatively measured [48]. Together, these factors may impose a high barrier to the evolution of fit, rhTRIM5α-resistant HIV-1 strains, an in a broader sense may be why TRIM5α proteins have been selected as antiviral proteins during host evolution. If it were straightforward for retroviruses to acquire resistance to TRIM5α, then little selective advantage would be conferred on hosts that employ TRIM5α as an antiretroviral protein.
An appreciation of how and why it is difficult for HIV-1 to acquire resistance to rhTRIM5α is important for a complete understanding of factors limiting transmission of primate lentiviruses among divergent primate species. Additionally, overcoming restriction by rhTRIM5α is required to enable HIV-1 replication in rhesus macaque cells and, thus, for the generation of an animal model that is based on HIV-1 infection of rhesus macaques [39]. Successful simian tropic (st) HIV-1 infection of pig-tailed macaques (M. nemestrina) [49], which express a TRIM5-cyclophilin fusion (TRIM-Cyp) that is unable to restrict HIV-1 [50]–[53], demonstrates that the HIV-1 host-range can include macaques when APOBEC3- and TRIM5-imposed restrictions are absent.
In the work presented here, we employed various strategies to identify a number of point mutations in HIV-1 CA that can confer partial resistance to restriction by rhTRIM5α. By combining assortments of these mutations, in an assisted evolution approach, and selecting for viruses that could replicate well in cells expressing rhTRIM5α, we generated mutant HIV-1 capsids that exhibit near-complete resistance to rhTRIM5α and retain fitness. Mutations that conferred resistance to rhTRIM5α were distributed in several different locations over the exterior surface of the HIV-1 capsid, suggesting that rhTRIM5α recognizes several different determinants on HIV-1 CA. The mutations that caused resistance also abolished the ability of rhTRIM5α to cause disintegration of HIV-1 core components, as well as the ability of HIV-1 to saturate rhTRIM5α, suggesting that they exert their effect by preventing rhTRIM5α binding to CA. Notably, when incorporated into an stHIV-1 construct that encodes SIVmac239 Vif, rhTRIM5α-resistant CA sequences enabled efficient stHIV-1 replication in primary rhesus macaque lymphocytes.
To derive HIV-1 CA proteins that confer resistance to rhTRIM5α we first employed an adaptation approach in which HIV-1 was passaged in human cells expressing rhTRIM5α. Initially, a cloned recombinant virus, termed NHG, that contains portions of HIV-1NL4-3 and HIV-1HXB2, and encodes GFP in place of the Nef gene, was grown in MT2 cells to generate a virus stock with genetic diversity. After fifteen days of replication, cell-free supernatant was used to infect four different clonal cell lines stably expressing rhTRIM5α (two derived from MT2 cells and two derived from MT4 cells) or cells transduced with an empty vector (Figure 1 A–D). The rhTRIM5α variant used belongs to a class of rhTRIM5α variants that encode TFP at amino acids 239–241 and restrict HIV-1 more potently than other variants [54], [55]. The first cycle of HIV-1 replication in these four rhTRIM5α-expressing lines was restricted by 50- to 100-fold relative to that in the empty vector control cells, confirming the potent anti-HIV-1 activity of rhTRIM5α therein (Figure 1 A–D). Infection was monitored by visual inspection of cytopathic effects and by measuring the fraction of GFP-positive cells in each culture at 1 to 3 day intervals. Because MT2 and MT4 cells are exceptionally permissive to HIV-1 replication (they were selected for this experiment for that reason), HIV-1 was able to spread through the four cultures and infect most cells after ∼8 days, despite potent inhibition by rhTRIM5α (Figure 1 A–D). After the majority of cells became GFP-positive and cytopathic effects had become abundant in the cultures, cell free supernatants were used to inoculate fresh rhTRIM5α-expressing cells, and this process was repeated for 10 passages in each of the four cell lines. The time taken for the majority of the cells to become GFP-positive during each passage appeared to decrease as the number of passages increased (Figure 1 A–D), implying adaptation of the virus to the rhTRIM5α expressing cells. At each passage, we measured the infectious HIV-1 titer in the four culture supernatants on both rhTRIM5α expressing and empty vector expressing cell lines. These analyses indicated that there was a progressive decrease in the sensitivity of the virus population to rhTRIM5α as the number of passages on rhTRIM5α expressing cells increased (Figure 1E–H). However, even the passaged virus stock retained at least some degree of sensitivity to rhTRIM5α, and each of the four adapted viruses was restricted by ∼5- to 10-fold after 6 to 8 passages. Sensitivity to rhTRIM5α did not appear to decrease with further passage (Figure 1E–H).
PCR amplification and bulk sequencing of viral DNA present in the four rhTRIM5α expressing cell cultures after various numbers of passages revealed that four CA mutations (V86E, I91N, I91T, and G116E) were present at a sufficiently high frequency to be detected in bulk sequences (Table 1). These mutations occurred at three different codons, all of which specified amino acids that are exposed on the presumptive outer face of the viral core (Figure 2A). Specifically, amino acids V86 and I91 are both within the cyclophilin-binding loop and G116 is on an outer turn of helix six (Figure 2A). Strikingly, each of the four mutations occurred in each of the four parallel virus cultures. In some cases, the mutant codons were present as mixtures with the WT codons, while in others WT codons became undetectable. In three virus cultures, all four amino acid changes were present in viral populations sampled after ten passages, at which point the WT I91 codon was no longer detected. Conversely, in one viral lineage, mutations at I91 were detected at passage 4, but then disappeared from the culture, and were undetectable after the tenth passage, at which point only V86E and G116E mutations were present as mixtures with the WT codon at both positions. Notably, amino acid substitutions V86M (V86E is described here) and G116E have previously been shown to decrease sensitivity to rhTRIM5α and crab-eating macaque TRIM5α, respectively [44], [45].
To determine whether and how mutations that arose in CA during passage decreased sensitivity to rhTRIM5α, PCR products amplified from viral cultures were cloned and inserted into the parental pNHG proviral plasmid. The resulting clones included four single CA mutants; V86E, I91N, I91T and G116E, as well as double-mutants I91N/G116E and I91T/G116E. Certain combinations of mutations were not recovered as individual species from the four viral cultures, and so site-directed mutagenesis was used to construct all other possible double and triple mutant combinations of V86E, I91T, I91N and G116E (see below). To test the resulting viruses for sensitivity to rhTRIM5α, single-cycle infection assays were employed, using the MT2 cell lines expressing either rhTRIM5α or an empty vector as targets (Figure 2B).
All four of the single amino acid mutants (V86E, I91N, I91T and G116E) exhibited decreased rhTRIM5α sensitivity. In particular, substitutions V86E and I91T each reduced rhTRIM5α sensitivity from ∼60-fold inhibition to ∼15-fold (Figure 2B). The I91N/G116E and I91T/G116E double mutants exhibited even lower rhTRIM5α sensitivity and were inhibited by less than 10-fold. Notably, the aforementioned mutants all occurred in individual species that were present in the viral culture experiment and exhibited little or no decrease in infectivity on the vector control cell line, suggesting that these individual and combined mutations did not incur major fitness costs. The data obtained with these reconstructed viral clones correlated quite well with the results obtained with each uncloned viral stock harvested after 10 passages in MT2-rhTRIM5α (Figure '1E–H). Specifically, virus populations harvested from MT4-rhTRIM5α#29 carried V86E and G116E, but not I91T or I91N, and this stock was restricted by 16-fold by rhTRIM5α. Conversely the other three uncloned, adapted viruses all carried I91T, I91N, and G116E and were restricted by only 5–10 fold.
We next reconstructed the viral mutants that contained combinations of the aforementioned substitutions that were not detected as individual species in the passage experiment, namely the V86E/I91N, V86E/I91T and V86E/G116E, double mutants and the V86E/I91N/G116E and V86E/I91T/G116E triple mutants. Strikingly, some of these mutants appeared completely, or nearly completely resistant to rhTRIM5α restriction in MT2 cells (Figure 2B). However, those mutants also exhibited severely reduced titers on the control vector expressing MT2 cell line, indicating that large fitness costs were imposed by combinations of mutations that gave rhTRIM5α resistance. The V86E/I91N and V86E/I91T double mutants also had apparent fitness defects that accompanied improvements in resistance to rhTRIM5α. These fitness costs were also observed in the context of spreading replication assays (Figure S1). Specifically, while the partially rhTRIM5α-resistant I91N/G116E and I91T/G116E double mutants were able to replicate as rapidly as WT on control vector expressing MT2 cells, the other combinations of mutations, particularly the V86E/I91N/G116E and V86E/I91T/G116E triple mutants, resulted in attenuated replication (Figure S1). Surprisingly, replication of these triple mutants was inhibited by rhTRIM5α during spreading replication assays (Figure S1), even though they appeared nearly completely resistant in single-cycle infection assays (Figure 2), suggesting the possibility that poor fitness might amplify the effect of residual restriction by rhTRIM5α during spreading replication assays. Notably, the attenuated double and triple mutants exhibited low titer in hamster pgsA cells, but were not sensitive to restriction by human TRIM5α expressed therein (Figure S2). This finding indicates that the apparent fitness defects observed in human cell lines were not due to acquisition of sensitivity to endogenous human TRIM5α, but instead were TRIM5α-independent deficits that coincided with decreases in sensitivity to rhTRIM5α. It therefore appeared that, in spite of the potential for HIV-1 CA to acquire near complete rhTRIM5α resistance via two or three amino acid substitutions, deleterious effects on fitness prevented some combinations of mutations from emerging as individual species during adaptation on rhTRIM5α expressing cell lines.
All four of the capsid amino acid changes that arose during evolution in the rhTRIM5α cell lines were the result of single nucleotide substitutions. We considered the possibility that other amino acid substitutions at the same positions in CA would have conferred a greater degree of rhTRIM5α resistance, but simply did not arise because of their requirement for two or three nucleotide changes in a single codon. To address this possibility, we conducted a vertical mutagenesis experiment, in which we made every possible amino acid substitution at positions 86, 91, and 116 in CA. Thereafter, we screened each of the 57 single amino acid substitutions for rhTRIM5α sensitivity (Figure S3A, B, C). A variety of mutations at positions 86, 91, and 116 decreased TRIM5α sensitivity. For example, at position 91 there were six different amino acid substitutions (I91A, I91P, I91Q, I91Y, I91D, and I91E) that were phenotypically approximately equivalent to I91N and I91T. There were also several amino acid changes at position 91 that conferred a greater degree of TRIM5α resistance than I91N or I91T, but all of them caused decreases in viral titer. For example, the I91G substitution gave near complete rhTRIM5α resistance, but this mutation caused a 70-fold loss in viral titer on empty vector expressing cells (Figure S3B). Several different substitutions at positions 86 and 116 also conferred partial resistance to rhTRIM5α, but often these effects were accompanied by reduced viral titer. At position 86, most substitutions caused reduced rhTRIM5α sensitivity and a glutamine substitution caused the largest increase in viral titer on MT2- rhTRIM5α (Figure S3A). However, when V86Q was combined with substitutions at I91 and/or G116 it did not exhibit higher viral titers on MT2- rhTRIM5α cells than the double mutants that emerged during the viral evolution experiments described in Figures 1 and 2 (unpublished observations). At position 116, none of the mutants exhibited higher titer on MT2-rhTRIM5α than did G116E (Figure S3C). Thus it appeared that the adaptation experiment selected near-optimal amino acids at positions 86, 91 and 116, but was not capable of selecting CA mutants with combinations of mutations that conferred the desired property.
Because our selection experiments were done in human cells, it was possible that endogenous human TRIM5α could have limited the spectrum of rhTRIM5α-resistance mutations that arose at these positions. Indeed, mutations at positions 86, 91, and 116 have been associated with sensitivity to human TRIM5α in the context of cytotoxic T lymphocyte escape [56]. Therefore, we screened the vertical mutant collection for sensitivity to human TRIM5α (Figure S4 A, B, C). Although some sensitivity to human TRIM5α was observed in some mutants, these effects were minor: while N-MLV was restricted by 160-fold by huTRIM5α, the degree to which some HIV-1 CA mutants were restricted was usually not more than 2- to 3-fold. One mutant, V86W, was restricted 6-fold by huTRIM5α. Therefore, it seems unlikely that endogenous TRIM5α would have profoundly limited evolution at these three positions during the adaptation experiment.
In a second independent strategy to identify mutations in HIV-1 CA that might confer rhTRIM5α resistance, we screened 91 infectious mutants from a library of NHG clones encoding PCR-mutagenized CA proteins (see Materials and Methods). MT2 cells expressing either the empty vector or rhTRIM5α were infected in a single cycle assay with a single dose of each clonal mutant virus and the ‘fold restriction’ by rhTRIM5α was calculated for those CA mutants that yielded measurable infectivity (>0.1% infected cells) on MT2-rhTRIM5α cells (Figure 3A). The parental virus, NHG was restricted by over 100-fold, as expected, while 22 CA mutant clones were restricted by less than 50-fold. The infectious titer of these 22 mutants was measured on vector expressing and rhTRIM5α expressing cells to confirm their decreased rhTRIM5α sensitivity (Figure 3B). Although there were a variety of mutations that decreased apparent rhTRIM5α sensitivity, many also decreased viral titer on vector-expressing cells. For example, N57S was restricted by only 8-fold by rhTRIM5α, but its infectious titer on vector control cells was reduced 30-fold as compared to the parental NHG virus. Therefore, we selected only mutants that conferred increased titer on rhTRIM5α-expressing cells as candidates that had the potential to contribute to a CA that was both fit and resistant to rhTRIM5α. The infectivity of these mutants on control vector expressing cells was equivalent to or only marginally decreased compared to the parental virus.
Interestingly, two such mutations, V86E and I91T, had also emerged in the viral evolution experiments presented in Figures 1 and 2. Additionally, we selected M10L, M10V, V83M, H87Q, R100S, and A105T as candidate mutations for further analysis because they also caused an increase in infectivity on MT2-rhTRIM5α cells. Of note, each of these amino acid substitutions occurred at positions that are predicted to be exposed to the cytoplasm after viral entry (Figure 3C).
While the above experiments demonstrated that a variety of mutations could confer partial resistance to rhTRIM5α, none of the aforementioned single amino acid CA mutants generated fully resistant viruses that did not incur a large fitness cost. However, the I91N/G116E and I91T/G116E double mutants demonstrated the potential for the additive effect of multiple mutations in the acquisition of TRIM5α resistance, without fitness penalties (Figure 2). Conversely, for other double mutants, particularly the combination of V86E and G116E, mutations that conferred partial rhTRIM5α resistance and had little fitness cost on their own, combined to generate rhTRIM5α resistant viruses at the expense of large fitness costs (Figure 2). Thus, it was possible that combinations of mutations that we had identified might generate fully rhTRIM5α-resistant viruses, but it was unpredictable as to whether and which combinations of mutations would allow the maintenance of high viral fitness.
Therefore, in an attempt to derive combination mutant CA proteins that were both fully fit and rhTRIM5α resistant, we adopted an ‘assisted evolution’ approach and generated a viral population that contained all possible combinations of the aforementioned non-deleterious mutations that conferred partial TRIM5α resistance (Figure 4A). Overlapping oligonucleotides that contained degenerate nucleotides encoding mixtures of WT and mutant amino acids were used to assemble a synthetic CA library pool containing random assortments of the mutations M10L/V, L83M, V86E, H87Q, I91N/T, R100S, A105T, and G116E. The synthetic CA assortment was used to generate a proviral plasmid library whose theoretical complexity was 576. Proviral plasmid DNA was isolated from 2.5×103 pooled bacterial colonies and transfected as a mixture, yielding a viral population that should contain every possible combination of the above mutations.
This viral population was evolved in MT2-rhTRIM5α through a series of 15 passages (Figure 4B). Additionally, the final four passages were also carried out in CEM-rhTRIM5α cells. The rationale for this was that CEM cells are less permissive than MT2, and we reasoned that this might impose a more stringent requirement for rhTRIM5α resistance and high replicative fitness on the viral population. Unfortunately, beginning at about passage 10, the maximum number of GFP-positive cells began to decrease with each subsequent passage (Figure 4B), presumably as a result of inactivating mutations in the GFP reporter gene in NHG. Nonetheless, visual inspection of cytopathic effects suggested that virus replication accelerated during the serial passage experiment.
We conducted PCR amplification and bulk sequencing of viral DNA from infected cells at passage 4, 9, 11 and 15 (Table 2). This analysis revealed that some mutations were purged from the viral population at varying rates while others apparently became fixed. Additionally, new mutations arose during passage. Specifically, V86E, R100S, and A105T were rapidly purged (by passage 4) while L83M and H87Q were detected at passage 4, 9,and 11, but lost by passage 15. Conversely, mutations M10L, I91N, and G116E became dominant in the viral population, with I91N and G116E rising to dominance rapidly and M10L becoming dominant in MT2 cells (but not CEM cells) between passages 11 and 15. Two additional mutations, namely A92E and M96I, that were not deliberately included in the starting assortment, arose spontaneously and became dominant in the viral population in MT2-rhTRIM5α cells. Curiously, these cyclophilin-binding loop mutations arose at about the same time as two other cyclophilin-binding loop mutations (L83M and H87Q) were lost from the population, suggesting the possibility that these mutations were not compatible with each other. It is also possible that A92E and M96I have the same biological effect as L83M or H87Q but at a lower fitness cost. The apparent lack of heterogeneity at the aforementioned positions in the bulk PCR product suggested that a single CA species containing five mutations (M10L, I91N, A92E, M96I and G116E) had come to dominate the population in MT2-rhTRIM5α cells (Table 2). In CEM-rhTRIM5α cells, the results were slightly different in that a single species did not dominate the population at passage 15. Rather M10L and M96I were present as mixtures with the WT codons, and a third spontaneous mutation, L52I, was identified, again as a mixture with the WT codon.
Sequences encoding CA were cloned from the viral DNA present in both MT2-TRIM5α and CEM-TRIM5α cell lines at passage 15. One clone, containing the substitutions M10L/I91N/A92E/M96I/G116E (LNEIE), corresponded to the dominant species in MT2-rhTRIM5α cells and the other, L52I/I91N/A92E/M96I/G116E (INEIE), corresponded to a species that was co-dominant with LNEIE in CEM-rhTRIM5α cells. With the exception of L52I, all of the aforementioned mutations that became fixed, or arose during the passage experiment, were at residues that are predicted to be exposed on the surface of the HIV-1 capsid (Figure 5A).
We next compared the properties of the species that arose via assisted evolution from the assorted CA mutant pool (LNEIE and INEIE, Figure 5A), with the best performing virus species (in terms of fitness and rhTRIM5α resistance) that arose spontaneously during the initial passage in rhTRIM5α expressing cell lines, namely I91N/G116E (NE) (Figure 5B). Additionally, because L83M was among the best performing mutations in the random mutagenesis screen (Figure 3B) and may not have had the opportunity to coexist with the late appearing and proximal A92E and M96I mutations, we also used site-directed mutagenesis to introduce L83M into LNEIE to generate a virus carrying six mutations M10L/L83M/I91N/A92E/M96I/G116E (LMNEIE, Figure 5B).
While the I91N/G116E (NE) double-mutant was restricted in MT2-rhTRIM5α by ∼5-fold, LNEIE, INEIE and LMNEIE were completely or nearly completely resistant to inhibition by rhTRIM5α in single cycle infection assays, similar to a virus expressing the SIVmac239 CA (Figure 5C). Moreover in spreading replication assays, LNEIE, LMNEIE, and INEIE replicated with equivalent kinetics in rhTRIM5α-expressing MT2 cells and empty vector expressing MT2 cells (Figure S5). Moreover, the titers and replication kinetics of these viruses on non-restricting MT2 cells were indistinguishable from that of NHG (Figures 5C and S5). Thus, an assisted evolution approach, in which mutations that individually had modest effects on rhTRIM5α were randomly combined and subjected to selection in cell culture, enabled the acquisition of near complete rhTRIM5α resistance, with apparent retention of viral fitness in this cell line.
In some human cells lines, the A92E mutation, which was present in the LNEIE, LMNEIE, and INEIE mutants, has been shown to cause HIV-1 infection to become inhibited by cyclophilin A (CypA). Consequently, infection by A92E CA mutant viruses is increased in the presence of cyclosporin A (CsA), a drug that disrupts the CA-CypA interaction [57]–[59]. To test whether the CsA-dependence observed in A92E mutants is also present in the context of LNEIE, LMNEIE, or INEIE, VSV-G pseudotyped WT and mutant viruses were titrated on HeLa cells in the presence or absence of CsA (Figure S6). As has previously been shown to be the case in HeLa cells [57], the presence of CsA increased infection by the A92E mutant by >10-fold. In contrast, the enhancing effect of CsA on infection was minor in the case of LNEIE (2-fold) and absent in the case of NHG, LMNEIE, and INEIE.
To further demonstrate resistance of the aforementioned CA mutant viruses to rhTRIM5α, we tested whether viruses carrying the CA mutations were able to avoid the disruptive effects of rhTRIM5α on incoming viral cores [27], [29], [30], and thereby complete reverse transcription. To accomplish this, we employed our recently described assay in which the integrity of a subviral complex containing CA, integrase (IN) and viral nucleic acids is monitored in infected cells shortly after infection [30]. Specifically, VSV-G pseudotyped viruses encoding either WT or mutant CA proteins were generated, and equivalent amounts (Figure 6A) were applied to pgsA-745 or pgsA-rhTRIM5α cells, in a synchronized infection protocol. Two hours later, cytoplasmic extracts were fractionated and the presence of CA and IN protein, as well as viral cDNA was measured in each fraction. As expected, in the case of WT HIV-1, a dense complex containing CA and IN as well as viral DNA was readily detected in infected pgsA cells (Figure 6B). This complex appeared largely absent in identically infected pgsA-rhTRIM5α cells, consistent with the notion that rhTRIM5α disrupts incoming HIV-1 cores. In contrast, the dense complex containing CA and IN, as well as viral DNA, was not disrupted by rhTRIM5α during infection with viruses encoding the LNEIE, LMNEIE, INEIE, or NE CA mutants (Figure 6C–F) and only minor differences in the levels of CA, IN, and viral DNA in the dense complex were observed. These data suggest that the subviral complexes generated by LNEIE, LMNEIE, INEIE, and NE CA mutant viruses are preserved in the presence of rhTRIM5α and thus that the cores of these viruses are largely resistant to the biochemical effects of rhTRIM5α. Although the NE CA mutant retained residual sensitivity to rhTRIM5α in MT2-rhTRIM5α cells (Figure 5C), all four CA mutants (NE, LNEIE, LMNEIE, and INEIE) were able to infect pgsA-rhTRIM5α cells at similar levels (Figure S7). The minor differences among the four mutants in pgsA-rhTRIM5α cells were, apparently, insufficient to be evident in the biochemical assay of rhTRIM5α restriction (Figure 6C–F).
In principle, the above HIV-1 mutants may have acquired resistance to rhTRIM5α by avoiding recognition by rhTRIM5α or, less likely, by acquiring the ability to infect cells despite recognition by rhTRIM5α. To distinguish between these possibilities we performed an ‘abrogation of restriction’ assay. TRIM5α proteins can, in general, be saturated by high amounts of incoming capsids, thereby enabling infection by viruses that would otherwise be restricted [7], [60]. The ability of a given viral capsid to abrogate TRIM5α mediated restriction is thus taken as a surrogate of its ability to be recognized by and bind to that TRIM5α protein. Therefore, we determined whether HIV-1 particles carrying the rhTRIM5α-resistant CA proteins were capable of abrogating rhTRIM5α activity in the rhesus macaque cell line, FRhK-4. VSV-G enveloped HIV-1 particles carrying a minimal HIV-1 genome (lacking GFP) and GagPol proteins that included either WT or the aforementioned rhTRIM5α-resistant CA sequences were first normalized according to their infectious titer on TZM indicator cells. FRhK-4 cells were challenged with increasing amounts of these viral particles along with a fixed amount of a virus containing WT HIV-1 GagPol protein and a genome that encoded a GFP reporter. As expected, infecting FRhK-4 cells in the presence of abrogating HIV-1 virions encoding a WT CA protein dramatically increased the titer of a WT GFP reporter virus (Figure 7A). Notably however, HIV-1 virions encoding rhTRIM5α-resistant CA mutants (NE, LNEIE, INEIE and LMNEIE) failed to abrogate restriction (Figure 7A). the residual sensitivity to rhTRIM5α that was observed for the NE CA mutant in in MT2-rhTRIM5α cells, was barely evident in FRhK4 cells, and all four CA mutants (NE, LNEIE, INEIE and LMNEIE) exhibited only minor variation in infectivity therein (Figure S8). The lack of abrogation activity displayed by the CA mutants strongly suggests that the mechanism by which the CA mutations conferred rhTRIM5α resistance was through loss of specific recognition by rhTRIM5α. Consistent with this conclusion, each of the rhTRIM5α-resistant viruses was inhibited by owl monkey TRIMCyp (Figure 7B), indicating that they were intrinsically sensitive to a TRIM5 protein that is able to bind to these capsids.
Along with APOBEC3 proteins, TRIM5α imposes a major block to HIV-1 replication in rhesus macaque cells [39]. To determine whether HIV-1 mutations that conferred resistance to rhTRIM5α expressed in human cell lines enabled replication in primary rhesus macaque cells, the TRIM5α-resistant CA sequences NE, LNEIE, INEIE and LMNEIE were inserted into a chimeric HIV-1 containing SIVmac239 Vif, named stHIV-1 [49]. Virus stocks bearing either WT, NE, LNEIE, INEIE, LMNEIE or SIVmac239 (stHIV-SCA) [39] CA sequences were normalized for reverse transcriptase content and used to challenge peripheral blood mononuclear cells (PBMC) from two rhesus macaque donors (Figure 8). As expected, stHIV-1 carrying the WT HIV-1 CA sequence failed to replicate, while stHIV-SCA carrying the SIVmac239 CA initiated a spreading infection. Notably, all four rhTRIM5α-resistant CA sequences enabled efficient stHIV-1 replication in rhesus macaque PBMC. In one donor, the rhTRIMα-resistant CA mutants replicated similarly to stHIV-SCA while in a second, apparently less permissive donor, all of the CA-mutant HIV-1 strains outperformed stHIV-SCA (Figure 8). The LNEIE mutant appeared to perform marginally better than the other CA mutants, and better than stHIV-SCA in PBMC from both donors. Genotyping revealed that both donors were heterozygous for rhTRIM5α alleles. The first donor carried alleles 4 and 5, while the second carried alleles 3 and 4 as (following the nomenclature described by Newman et al and Wilson et al [54], [55]). Alleles 4 and 5 belong to a class that encodes a glutamine at residue 339 and allele 3 belongs to a class that encodes TFP at the same position. These two classes of rhTRIMα variants have been shown to differ in their restriction specificity and potency [54], [55]. Importantly, the ability of the HIV-1 mutants described here to replicate in PBMC from both animals suggests that they have acquired resistance to restriction by both classes of rhTRIM5α alleles.
In this study, we identified a number of mutations in HIV-1 CA that individually could reduce the sensitivity of the incoming capsid to restriction by rhTRIM5α. When present in the right combination, collections of these mutations could confer near complete resistance to rhTRIM5α, sometimes without a fitness cost. Indeed, mutations in capsid were necessary and sufficient for HIV-1 to evade restriction by rhTRIM5α, consistent with the notion that the antiviral activity of TRIM5 depends on specific capsid recognition [5], [8], [39], [40], [47]. The rhTRIM5α resistant CA sequences abolished the ability of rhTRIM5α to disrupt incoming HIV-1 cores, enabling reverse transcription and the formation of a complex containing CA, IN and viral DNA, which would normally be blocked by rhTRIM5α [30]. Ultimately, these mutations enabled uninhibited infection of human cell lines stably expressing rhTRIM5α, which ordinarily exhibit >50-fold reduced susceptibility to WT HIV-1 infection.
With one exception, the amino acid substitutions that were found in the rhTRIM5α-resistant CA sequences encoded amino acids that are exposed on the presumptive exterior surface of the capsid lattice. However, L52I, (which occurred in only one of two cell lines in which the assorted CA mutant pool was evolved) is buried in the interior of the CA protein structure [43]. It is possible that L52I contributes to resistance by shifting the conformation of the capsid surface. Alternatively, it may marginally stabilize or destabilize the viral core, thereby affecting rhTRIM5α action. It is also possible that L52I, or indeed other mutations described herein, arose as a compensatory mutation to maintain high viral fitness while not directly contributing to rhTRIM5α resistance. The contribution of the L52I mutation in the interior of the CA structure notwithstanding, the primary mechanism by which the HIV-1 CA mutants acquired resistance to rhTRIM5α appeared to be through loss of rhTRIM5α recognition, rather than resistance to the effects of TRIM5α after recognition of the incoming capsid. This conclusion is based on the findings that (i) rhTRIM5α-resistant mutant capsids were unable to saturate rhTRIM5α and thereby facilitate WT HIV-1 infectivity in rhesus macaque cells, suggesting that they were not recognized by rhTRIM5α and (ii) the rhTRIM5α-resistant capsids retained full sensitivity to another TRIM5 protein (omkTRIMCyp) with a different CA binding specificity. Neither of these results would be expected if resistance were acquired via a mechanism in which the viral capsid retained rhTRIM5α binding, but acquired the ability to resist its antiviral effects.
Our initial failure to derive HIV-1 variants with complete resistance to rhTRIM5α through replication in cell lines expressing this inhibitor, or through random mutagenesis, underscores the difficulty in deriving HIV-1 strains with this property. Indeed, our eventual success required the combined application of the two different approaches, and then further evolution during selection from an assortment of mutations identified by each strategy. A comparison of the results obtained from the initial adaptation (Figure 2) and random mutagenesis (Figure 3) approaches suggests that each had distinct advantages and disadvantages. Despite the fact that four parallel cultures were initiated, the adaptation approach clearly did not produce a wide variety of solutions, and the same four mutations (V86E, I91N, I91T, and G116E) arose in four independent cultures. No combination of these four mutations gave fully fit rhTRIM5α-resistant CA sequences. It is possible that this result is defined by a potentially limited complexity of the viral population to which selection pressure was applied, and that rather than selection of several independent solutions, a small number of initially dominant genotypes persisted. However, another explanation might be that mutations that reduce HIV-1 sensitivity to rhTRIM5α without a fitness cost are few in number. Consistent with this latter interpretation is the fact that 2 of the 3 positions that we found to be mutated in rhTRIM5α-selected viruses have also been identified in similar, but completely independent, selection experiments in other laboratories, in different cell lines [44], [45]. Those mutations that confer rhTRIM5α resistance but are marginally deleterious to virus replication might not be selected during the adaptation approach, even though they could contribute rhTRIM5α evasion if their accompanying fitness costs were alleviated by compensating mutations. Screening a random library of virus clones might be a better way to identify such mutants. Indeed, the random mutagenesis screen identified a larger number of mutants that conferred reduced rhTRIM5α sensitivity, even though many of these did have associated fitness defects. Notably, V86E and I91T emerged from both adaptation and the random mutant library screening approaches. Interestingly, however, the random mutagenesis screen identified several mutations that did not arise during adaptation yet exhibited reduced sensitivity to rhTRIM5α without an obvious fitness cost (e.g. V83M and M10L). Nevertheless, a drawback of random mutant screens is that only a limited number of mutants can be individually tested. Indeed, two amino acid substitutions emerged during in vitro evolution that were not represented in the random mutant library (G116E and I91N). Additionally, because the experiments were done in human cells, a potential limitation of both the random mutant library screening and the in vitro evolution strategies was the possibility that some rhTRIM5α-resistant mutants could be missed if they simultaneously caused gain of sensitivity to endogenous human TRIM5α.
Overall, however, the application of both approaches and assortment of the resulting mutants in an assisted evolution approach led to derivation of fit, rhTRIM5α-resistant CA sequences. Even then, further de novo mutations of the assorted variant pool was required to generate the optimally resistant CA sequences. One possible reason for the eventual success of our approach is that the second round of selection was performed using a population of CA sequences that was highly enriched for mutations conferring partial TRIM5α resistance. This population contained individual mutant assortants that are highly unlikely to have occurred by chance through the standard approaches to viral evolution that were attempted initially.
Clearly, each individual mutation identified by either adaptation or random mutant screening approaches enabled only a partial evasion of rhTRIM5α (Figures 2 and 3). These mutations were distributed over several determinants on the surface of the HIV-1 capsid, as has been found in previous studies of retrovirus sensitivity to TRIM5α [40], [41], [44]–[47]. Indeed, although the cyclophilin-binding loop was featured prominently as a site at which mutations conferring decreased HIV-1 sensitivity to rhTRIM5α occurred (positions L83, V86, I91, A92, M96), other determinants included the N-terminal β-hairpin (M10) and helix 6 (G116). One reasonable interpretation of these data is that several different sites on the capsid exterior contribute to the binding interaction with TRIM5α, and that mutations at any one of these sites, reduces, but does not eliminate interaction. Interestingly, a comparison of the distribution of mutations in MLV CA and HIV-1 CA that arose during the selection of rhTRIM5α-resistant viral variants [47] reveals striking similarity (Figure S9A and S9B). A key difference, however, is that single amino acid substitutions in MLV CA conferred near complete resistance to rhTRIM5α [47], while multiple substitutions were required in HIV-1 to achieve the same effect. Perhaps the viral challenges to which rhTRIM5α has been subjected to during its evolutionary history have shaped it in such a way that it is a more robust inhibitor of lentiviruses than gammaretroviruses.
Because mutations in the cyclophilin-binding loop of HIV-1 CA have previously been shown to reduce the contribution of CypA to rhTRIM5α activity [61], [62], it is possible that perturbation of CypA binding may have contributed to the acquisition rhTRIM5α resistance described herein. Although retention of sensitivity to restriction by TRIM-Cyp suggests that CypA binding has been retained in the mutants described herein (Figure 7B), it remains possible that the role of CypA in rhTRIM5α activity was perturbed. Indeed previous work has suggested that a V86M CA mutation, while not preventing CypA binding, eliminates the contribution of CypA to the restriction of HIV-1 by huTRIM5α mutants [63]. It therefore remains unclear whether cyclophilin-binding loop mutations in rhTRIM5α-resistant capsids have altered the involvement of CypA in restriction or simply decrease the binding of rhTRIM5α to HIV-1, independent of CypA.
Although HIV-1 CA mutations could be combined to give near complete escape from restriction, some combinations, particularly those derived from the first round of selection, were apparently deleterious to viral fitness. The need to alter multiple determinants, coupled with the inherent genetic fragility of HIV-1 CA [48] likely underlies the difficulty in generating combinations of CA mutations that confer rhTRIM5α resistance while maintaining high fitness. It is possible that the LNEIE and INEIE mutants, which were both fit and resistant to restriction by rhTRIM5α, include compensatory changes that alleviate fitness costs that accompanied escape from rhTRIM5α restriction. Our group has recently demonstrated that about 70% of randomly introduced single amino acid substitutions, and nearly all two-amino acid substitutions in CA are lethal to HIV-1 [48]. This lack of robustness in CA is also evident in MLV, as the N-terminal domain of MLV CA, unlike other regions of Gag, is intolerant of small insertions [64]. In the case of HIV-1, most lethal CA mutations were shown to disrupt normal particle formation [48]. However, interactions with multiple host factors, or structural requirements that dictate proper disassembly may also decrease the genetic robustness of CA. Thus, mutants containing combinations of changes required to both escape rhTRIM5α and retain fitness should be very rare, and this may be one reason why evolution has selected CA as a target for host antiviral proteins. Notably, the incomplete escape that was evident in viral populations derived from initial selection experiments included a double amino acid mutant (NE), which was both fit and incompletely resistant to restriction in MT2-rhTRIM5α cells, as well as mixtures of viruses containing WT and mutant codons at V86 and G116. Although a V86E/G116E double mutant clearly had the opportunity to dominate in these cultures, it exhibited a clear loss of fitness, probably explaining why it did not. While it is formally possible that the fitness defect in V86E/G116E and other unfit rhTRIM5α-resistant mutants is species-specific and would not exist in monkey cells, we consider this unlikely, given that low infectiousness was also observed in hamster cells (Figure S2).
Importantly, mutations conferring resistance in rhTRIM5α-expressing cell lines allowed HIV-1 encoding SIVmac239 Vif to replicate in PBMC from rhesus macaques. Indeed, an HIV-1 encoding the LNEIE mutant CA protein replicated as well as or better than a virus encoding an SIVmac239 CA protein, suggesting that TRIM5α activity was entirely bypassed. Therefore, the HIV-1 CA mutations identified here enable HIV-1 to overcome the species-specific tropism barrier imposed by rhTRIM5α. Thus, although the barrier to the evolution of fit, rhTRIM5α-resistant HIV-1 CA proteins is apparently high, it is not insurmountable. Moreover, although the two amino acid NE mutant was inferior to the 5 and 6 amino acid CA mutants, in that it retained some degree of sensitivity to rhTRIM5α in human cell lines, the NE mutant replicated well in rhesus PBMC, and had little if any ability to saturate rhTRM5α in the FRhK cell line. Thus, one caveat that should be attached to the above discussion is that the level of rhTRIM5α expression in the engineered human cell lines in which our selection experiments were performed likely exceeds the endogenous levels present in rhesus macaque cells. The number of mutations required to operationally escape TRIM5α inhibition in a natural setting may therefore be fewer than is suggested by our studies. In any case, the CA mutants identified herein should facilitate the development of rhesus macaque-based non-human primate models of HIV-1 infection.
Primate adherent cells (293T and FRhK-4) and suspension cells (MT2, MT4, CEM, and rhesus macaque PBMC) were cultured in DMEM and RPMI, respectively. CHO K1-derived pgsA-745 and the previously described pgsA-rhTRIM5α and pgsA-omkTRIMCyp cell lines [30] were maintained in Ham's F12 media with 1 mM L-glutamine. All growth media was supplemented with 10% fetal bovine serum and gentamicin.
Human T-cell lines (MT2, MT4 and CEM) expressing rhTRIM5α were prepared by transduction with a murine leukemia virus (MLV) vector containing rhTRIM5α cDNA, inserted 5′ to an internal ribosomal entry site and a blasticidin resistance gene. The rhTRIM5α cDNA was previously cloned from FRhK-4 cells [3] and the sequence matches allele 3, as described by Lim et al [65]. This rhTRIM5α allele belongs to a class of rhTRIM5α alleles that encode TFP at amino acid residues 239–241 and restrict HIV-1 more potently than other alleles [54], [55]. To generate the cell lines, 293T cells were cotransfected using polyethyleneimine (PolySciences) as previously described [7] with 100 ng of a vesicular stomatitis virus G protein (VSV-G) expression plasmid, 500 ng of a MLV GagPol expression plasmid, and 500 ng of the rhTRIM5α-expressing MLV vector. The resulting virus was used to transduce MT2, MT4, or CEM cells. Cells were selected with 10 µg/mL blasticidin and single-cell clones expressing rhTRIM5α were derived by limiting dilution.
The replication competent HIV-1/GFP used in viral evolution experiments and for measurements of rhTRIM5α sensitivity was produced by transfection of an HIV-1NL4-3/HIV-1HXB2 recombinant proviral plasmid that encoded GFP in place of Nef (pNHG, GenBank accession number: JQ585717) [66]. Infectious titer was determined by inoculating MT2 cells (2×104 per well) in 96-well plates with 50 µL of serially diluted supernatant. After allowing infection to proceed overnight, further replication was blocked by the addition of 100 µg/mL dextran sulfate. Two days after infection, cells were fixed in 2% paraformaldehyde and the number of infected cells enumerated by FACS.
Viruses used to saturate rhTRIM5α in FRhK-4 cells were prepared by cotransfection of 293T cells with pCRV1-HIV-1NL4-3 GagPol (WT or rhTRIM5α-resistant mutant), the HIV-1 based vector pV1, which lacks a reporter gene, and a VSV-G expression plasmid at a 5∶5∶1 ratio, respectively [16]. Titers were determined using TZM-bl indicator cells, infected colonies were stained and counted using X-Gal two days post-infection. The reporter virus used to test for rhTRIM5α saturation in FRhK-4 cells, the viruses used for the biochemical analysis of core components, and the viruses used to infect pgsA-745 cell lines were produced in the same manner, except an HIV-1 based vector encoding GFP (pCSGW) [67] was used instead of pV1. VSV-G enveloped N-MLV was also produced in the same manner, except N-MLV GagPol expression plasmid and an MLV based vector encoding GFP (pCNCG) were used [68].
Prior to inoculation of MT2 and MT4 cells expressing rhTRIM5α, a cloned pNHG-derived virus that lacked Vpr was allowed to replicate in MT2 cells to generate sequence diversity. Specifically, 5×105 MT2 cells were infected with NHG at an MOI of 0.001. To maintain a viable cell population (the culture would have otherwise died due to cytopathic effects) and expand the viral population, uninfected MT2 cells were added as follows: 5×105 cells at 4 dpi, 106 cells at 6 dpi and 8 dpi, 3×106 cells at 10 dpi, and 4×106 cells at 13 dpi. After 15 days of replication, cell-free supernatant was harvested and titrated on MT2 cells. Thereafter, 1.25×106 IU were used to infect two cultures of 5×106 MT2-rhTRIM5α cells, two cultures of MT4-rhTRIM5α cells, and one culture each of MT2 and MT4 cells transduced with empty vector. Cells were diluted to 35 mL after an overnight infection in 5 mL. To measure the initial infection, dextran sulfate was added to a 100 µL aliquot of cells one day after infection, which were then fixed the following day. To monitor the spreading infection, 100 µL aliquots of cells were fixed periodically in 2% paraformaldehyde and the fraction of infected (GFP positive) cells determined by FACS analysis. After cytopathic effects became abundant in the culture, and/or the percentage of infected cells had peaked, 2 mL cell-free supernatant was used to infect 5×106 fresh cells as before. However, beginning with the fifth passage in MT2- rhTRIM5α cell lines and the fourth passage in MT4- rhTRIM5α cell lines 5 mL of supernatant was used to initiate new passages.
To recover CA sequences from viruses replicating in rhTRIM5α expressing cells, DNA was isolated from infected cells using the Qiagen DNeasy Blood and Tissue Mini Kit and sequences encoding CA were amplified using PCR and a sense primer, 5′-AAGGAAGCCTTAGATAAGATAGAGG-3′ along with an antisense primer 5′- TTGCCTTTCTGTATCATTATGGTAG -3′. Products were directly sequenced, without cloning, using the sense primer. To generate proviral clones containing adapted CA sequences, PCR amplification was done using primers 5′-GCACAGCAAGCGGCCGCTGACACAGG-3′ and 5′-GCCAAAACGCGTGCTTTATGGCCGGG-3′ to add NotI and MluI restriction sites for insertion into a pNHG derivative that was engineered (by silent mutagenesis) to contain NotI and MluI restriction sites flanking CA (GenBank accession number JQ686832). Additional combinations of mutations were introduced into CA by PCR-based mutagenesis methods as previously described [69]. For follow-up studies of CA mutants, virus stocks were generated by transfection of 293T cells with 1 µg proviral plasmid/well in a 24-well plate. Reverse transcriptase in culture supernatants was quantified by a previously described assay based on real-time PCR [70], and infectious titers were determined as described above. Some CA mutant viruses were also used in spreading replication experiments, in which 2×105 MT2-rhTRIM5α cells or empty vector transduced MT2 cells in 12-well plates were infected at an MOI of 0.002. Culture volumes were kept at 1 mL throughout the experiment. One day after infection, a 100 µL aliquot of cells was treated with 100 µg/mL dextran sulfate and fixed the following day to measure the initial infection. Beginning two days after infection, 100 µL aliquots of cells were fixed daily for quantitation of GFP positive cells by FACS analysis.
The construction and analysis of a library of pNHG clones encoding PCR-mutagenized CA sequences has been previously described [48]. Briefly, the mutagenized CA library was generated using the Genemorph II random mutagenesis kit (Agilent) using the following oligos 5′-GTA AGA AAA AGG CAC AGC AAG CGG CCG CTG -3′ and 5′- CTT GGC TCA TTG CTT CAG CCA AAA CGC GTG-3′. The PCR template was a pNHG derivative (JQ686832) that contained silently mutated sequences to generate NotI and MluI sites flanking CA-encoding sequences (pNHGCapNM). The PCR product was cloned using a TOPO TA Cloning Kit and plasmid DNA was extracted from approximately ∼1×104 pooled, insert-positive colonies. After sequencing the amplicon from 20 clones to obtain a preliminary estimate of the mutagenesis frequency, this pooled plasmid DNA was digested using NotI and MluI and the CA-library insert subcloned into pNHGCapNM. Proviral plasmid DNA was extracted from individual cultures derived from 1056 colonies and subjected to sequencing and further analysis. Ninety-one of the resulting proviral plasmids that yielded infectious virions were screened for mutations that decrease sensitivity to restriction by rhTRIM5α. Specifically, viral stocks containing WT NHG or each of the 91 CA mutants were prepared by transfecting 2.5×104 293T cells/well with 100 ng of pNHG and polyethylenimine in a 96-well plate. Thereafter, MT2-rhTRIM5α cells or MT2 cells transduced with empty vector were infected with 2 µL filtered virus in a 96-well plate at 2×104 cells/well. Virus replication was limited to a single cycle by adding 100 µg/mL dextran sulfate after an overnight infection. Two days after infection, cells were fixed in 2% paraformaldehyde and subjected to FACS analysis. Stocks of 22 mutants that infected at least 0.1% MT2-rhTRIM5α cells and were restricted by less than 50-fold (see results) were prepared again by transfecting 2×105 293T cells with 1 µg pNHG in a 24-well plate. Again, single-cycle infections were done in MT2-rhTRIM5α cells or MT2 cells transduced with empty vector, using 50 µL of serially diluted supernatant.
To construct a library of CA sequences containing random assortments of CA mutations (M10L/V, L83M, V86E, H87Q, I91N/T, R100S, A105T, and G116E), ten overlapping oligonucleotides, four containing degenerate nucleotides, were used in a PCR-based gene synthesis to produce a 400 base pair DNA encoding the 5′ portion of CA that included all of the mutated positions. The heterogeneous gene synthesis product was designed to encode WT or mutant amino acids at a 1∶1 ratio, or a 1∶1∶1 ratio in cases where two mutations were included at a single position. Specifically, a 50 µL PCR reaction contained 3 pmol of each oligonucleotide and gene synthesis was carried out for 20 cycles using Phinzymes Phusion (1 min. 98°C, 1 min. 50°C, and 2 min. 72°C). Thereafter, 2 µL of crude gene synthesis product was used as template in a 50 µL PCR reaction with 200 nM external primers (1 min. 98°C, 1 min. 55°C, and 2 min. 72°C) and the resulting product was gel-purified. The 5′ end of the gene synthesis product was designed to contain a Not I restriction site while the 3′ end was designed to overlap with a second amplicon that was used to extend the synthetic CA sequence to include the 3′ portion of CA and an MluI restriction site. The two overlapping DNAs were used together as templates for a final PCR, which produced a DNA that was gel purified, and inserted into pNHG. Proviral plasmid DNA was purified from 2.5×103 pooled bacterial colonies and 5 µg of this DNA was used to transfect 293T cells. The resulting virus was used to infect 5×106 MT2-rhTRIM5α cells at an MOI of 0.005. Then, 1 mL of cell-free virus was used in a second passage onto MT2-rhTRIM5α cells and 100 µL was used in all subsequent passages. Following 11 passages in MT2-rhTRIM5α cells, 2 mL of cell-free supernatant was used for serial passaging in CEM-rhTRIM5α cells.
Biochemical analysis of retroviral cores in infected cells was done as previously described [30]. Briefly, pgsA745 cells or its derivative stably expressing rhTRIM5α were plated in 10-cm dishes one-day before infection. Cell culture supernatants containing VSV-G pseudotyped HIV-1NL4-3 GagPol/CSGW viruses were filtered and treated with RNase free DNaseI (Roche) at a concentration of 1 unit/ml for 1 hour at 37 °C in the presence of 6 mM MgCl2. Cells were washed with ice-cold phosphate-buffered saline (PBS) and 7 ml of chilled virus (with 20 mM HEPES) was added to the cells at an MOI of ∼0.01. After allowing virus to bind to cells for 30 minutes at 4°C, the inoculum was removed and cells were washed three times with PBS. Cells were then incubated at 37°C for 2 hours in complete cell culture media. Cells were collected in PBS-EDTA, pelleted, and resuspended in 1 ml of hypotonic buffer (10 mM Tris-Cl pH 8.0, 10 mM KCl, 1 mM EDTA) supplemented with complete protease inhibitors (Roche) and SuperaseIN (Life Technologies). After incubation on ice for 15 minutes, the cell suspension was dounce homogenized for 50 strokes, using pestle B. Nuclear material was pelleted by centrifugation at 1000×g for 5 minutes and post-nuclear supernatant was layered on top of a 10–50% (w/v) linear sucrose gradient prepared in 1X STE buffer (100 mM NaCl, 10 mM Tris-Cl pH 8.0, 1 mM EDTA). Samples were ultracentrifuged using a SW50.1 rotor at 30000 rpm for 1 hour. Ten 500 µl fractions from the top of the gradient were collected, and proteins and DNA in each fraction were analyzed as previously described [30].
Rhesus macaque PBMC were stimulated with 3 µg/mL staphylococcus enterotoxin B (SEB) for three days and 5% IL-2 (Hemogen) throughout the experiment. Following SEB stimulation, PBMC were infected with stHIV-1, which is a chimeric HIV-1NL4-3 –derived virus in which the Env protein is derived from SHIV/KB9 and the Vif protein is replaced by that of SIVmac239. Additionally, stHIV-1/S-CA encodes an SIVmac239 CA sequence [39], [49]. Mutant HIV-1NL4-3 CA sequences were transferred from NHG to stHIV-1, using BssHII and AgeI restriction sites. Virus stocks were normalized for reverse transcriptase activity and 1×105 PBMC were infected in a 96-well V-bottom plate with inocula containing 20 pg reverse transcriptase. Following overnight infection, cells were washed three times and supernatant from the third wash was frozen for RT quantitation (this value is plotted as 1 day after infection). After washing, cells were transferred to a 96-well U-bottom pate. At days 2, 3, 5, 7, 9, 11, and 14 after infection 80% of the cell supernatant was frozen for RT quantitation and replaced with fresh medium. For rhTRIMα genotyping, DNA was isolated from PBMC from both donors using the QIAGEN DNeasy Blood & Tissue Mini Kit. RhTRIMα exon 8 was amplified by PCR using sense primer 5′- TTGATGTGACACTGGCTCCAAACAAC-3′ and antisense primer 5′-TGGGTAAAGCGGCCGCCAGAGCTTG-3′. The PCR products were cloned using the Invitrogen Zero Blunt TOPO PCR Cloning Kit.
|
10.1371/journal.pntd.0000533 | Anemia of Inflammation Is Related to Cognitive Impairment among Children in Leyte, The Philippines | Many studies have addressed the relationship between iron deficiency anemia (IDA) and cognitive impairment, but none have evaluated the role of non-iron deficiency anemia (NIDA). One of the main causes of NIDA in developing countries is AI, largely due to infectious diseases, whereby iron is shunted away from bio-available forms to storage forms, making it less accessible for use by host tissues. The objective of this study was to determine the effect of NIDA, due largely to AI in this context, on cognitive function after adjustment for potential confounders.
This cross-sectional study was conducted in Leyte, The Philippines among 322 children ages 7–18 years. Blood samples were collected and analyzed at the time of cognition testing. Three stool samples were collected and evaluated by the Kato Katz method for quantitative assessment for Schistosoma japonicum and geo-helminth infection. Socio-economic status (SES) was evaluated by survey. Linear regression models were used to quantify the adjusted relationship between performance in different cognitive domains and both IDA and NIDA.
After adjusting for age, sex, SES and nutritional status, children in the NIDA had lower scores on the PNIT (P = <0.05) and the WRAML memory domain (P<0.05) compared to children in the non-anemic group. Children in the IDA had lower performance on the PNIT compared to the non-anemic group after controlling for potential confounders (P<0.05).
NIDA, predominantly due to AI in this context, was related to lower performance on two tests of cognitive function. This is likely due to decreased delivery of iron to host tissues in this context, including the CNS.
| Past studies have demonstrated that iron deficiency anemia is related to deficits in cognitive fucntioning in children, and treating iron deficiency anemia with iron supplementation can improve cognition. Anemia of inflammation is another type of anemia caused by many diseases of lesser-developed countries including bacterial and parasitic infections. Anemia of inflammation is characterized by disordered iron metabolism, such that iron is sequestered in storage forms, preventing its use from tissues that require it. We hypothesized that decreased iron delivery to the brain in the context of anemia of inflammation might lead to decreased cognitive performance. This study found that children with anemia of inflammation had decreased cognitive performance in specific domains, compared to subjects with no anemia. True total body iron deficiency anemia was related to lower performance in the same domains. The only treatment option for anemia of inflammation is treatment of the underlying disease. Iron supplementation will not prevent cognitive deficits in children with anemia of inflammation. Interventions aimed towards maximizing the cognitive development of children in lesser-developed countries will need to focus on the prevention and treatment of bacterial and parasitic infections.
| Anemia of inflammation (AI), is the second most prevalent type of anemia following iron deficiency anemia (IDA) [1],[2]. The most common conditions associated with AI include bacterial, viral, and parasitic infections, cancers, and autoimmune diseases [1]. AI leads to alterations in iron metabolism likely mediated by elevated levels of specific cytokines in response to the aforementioned disease processes. Hepcidin, an acute phase protein induced by interleukin-6, has been proposed as the link between inflammatory disorders and AI [3],[4]. Its actions include both blocking absorption of iron from the gut and its release from reticuloendotheial macrophages. In previous work in this study population we have identified AI as the predominant cause of S. japonicum-related anemia [5],[6].
Children and adolescents with IDA perform less well on specific cognitive tests than those without IDA [7]. Further, treatment of IDA has been demonstrated to improve cognitive performance among children and adults [8],[9],[10]. The cognitive deficits seem to be mediated by low iron levels as opposed to anemia itself. For example, a randomized controlled trial conducted among non-anemic iron deficient adolescent females demonstrated improved verbal learning and memory with iron supplementation [11], suggesting cognitive deficits are most related to decreased delivery of iron to the central nervous system (CNS).
In the context of AI, the serum concentration of iron and the iron saturation of transferrin are also decreased leading to decreased iron delivery to host tissues that necessitate it [1]. One well known consequence is iron deficient erythropoiesis, whereby anemia results despite normal total body supplies of iron, as iron is shunted away from bio-available forms, into storage forms. We hypothesized that AI might be associated with decreased cognitive function in domains sensitive to CNS iron status, due to decreased bio-availability of iron. Iron is required for many essential brain functions including myelination and synthesis of the neurotransmitters serotonin and dopamine [7]. It is also possible, as demonstrated in recent studies, that pro-inflammatory cytokines made in the context of parasitic diseases, which cause AI, might have direct effects on cognitive processing [12]. No studies have assessed the relationship between AI, pro-inflammatory cytokines elaborated in this context, and cognitive function in children.
The main objective of this study was to assess the relationship between NIDA, largely due to AI in our population, and cognitive function after adjustment for potential confounding factors. The secondary objectives were to 1) assess the relationship between IDA and cognitive domains not previously studied, with more careful control for confounders and 2) explore mechanisms through which helminth infections could mediate cognitive impairment through established relationships with both AI [5],[6] and pro-inflammatory cytokines [13],[14],[15]. We hypothesized that anemia of inflammation would lead to lower cognitive performance in domains sensitive to brain iron content and that this would be more important than serum levels of pro-inflammatory cytokines, which have been demonstrated to lead to impaired cognitive performance.
This study was conducted in Macanip, a rural rice farming village in Leyte, The Philippines, where S. japonicum and geo-helminth infections are prevalent. There is no malaria in this study area. Three to six months before the start of the study, a census was completed for the entire village. Study staff went door to door accompanied by a resident of the village, mapping each household with global positing satellite devices. The census enumerated number of individuals in the household, their gender, and dates of birth. At that time, all individuals in the household ages 8–30 were asked to provide informed consent for the stool screening procedures for a study investigating immune correlates of re-infection. The latter, longitudinal study only enrolled infected subjects. For this cross-sectional study, we used the initial census data to a) identify infected subjects ages 7–18 who were eligible for the longitudinal study and b) identify uninfected subjects ages 7–18 who were only eligible for this cross-sectional study. Exclusion criteria included pregnancy or lactation or the presence of a serious chronic disease determined by history, physical examination, or laboratory findings. An attempt was made to recruit all age eligible subjects (N = 394), however, 14 could not be re-located or scheduled for assessments, leaving 380 enrolled subjects. A separate consent process was used for the activities related to this study, which included the cognition testing for all subjects and a single blood test for the uninfected subjects, who otherwise would not have had this blood sample taken as part of their participation in the longitudinal study. Cognition testing occurred between 9/7/2002–10/8/2002 for all subjects. Treatment for infected subjects as well as the blood sample and all other assessments were taken between 10/28/02–11/6/02.
For all study activities, written informed consent was obtained from the parents of all study subjects and assent was provided by all children over the age of 8 who could understand the assent form and process. The study was approved by the Brown University and Philippines Research Institute of Tropical Medicine Institutional Review Boards.
The choice of cognitive tests employed was based on 1) known relationships between iron status and both memory (WRAML verbal memory and verbal fluency) and concept recognition (PNIT) domains [16],[17] and 2) known relationships between helminth infections and cognitive domains (verbal fluency and WRAML Learning) [18],[19].
The cognitive tests were translated and adapted for use in The Philippines. They were pilot tested among Filipino children (N = 51) from other villages near the study area. Reliability testing for each cognitive test included joint inter-rater and test-retest reliability with a six-week interval between tests. Cronbach's alpha coefficient was used to assess to the degree of internal consistency among the three subscales that constitute each of the learning and memory domains of the Wide Range Assessment of Memory and Learning (WRAML).
A complete blood count was obtained using a hematology analyzer (Serono Baker Diagnostics) on venous blood samples. Serum ferritin (SF), serum transferrin receptor (sTfR), C-reactive protein (CRP), and serum cytokines (interleukin-6 (IL-6), interferon gamma (IFN-γ), tumor necrosis factor alpha (TNF-α) were analyzed using a multiplex bead-based platform (BioRad Hercules, CA) as described previously [14]. SF is a measure of stored iron. sTfR is a circulating form of the transferrin protein receptor, derived mostly from red blood cell precursors. sTfR is increased when red blood cells are iron deficient, as the receptor is upregulated to increase iron uptake, and with expanded red blood cell in under various clinical settings [22].
Parasite egg counts were determined by examination of consecutive stool specimens obtained from each study participant on separate days. We requested three stool specimens from each subject, but individuals were eligible if they provided one or two specimens. All subjects provided at least one sample and 82.8% and 64.8% provided two or three samples, respectively. Each stool specimen was evaluated in duplicate for S. japonicum, Ascaris lumbricoides, Trichuris trichiura, and hookworm eggs by the Kato Katz method within 24 hours of collection. The mean eggs per gram of stool was used as the quantitative measure of infection status for each worm based on the average egg count for each duplicate specimen. Then, the mean egg count was taken across the stool samples (1–3), including samples with zero epg. World Health Organization criteria were used to classify each infection as uninfected or low, moderate, or high intensity [23].
Anemia was defined on the basis of age- and sex-specific hemoglobin cut-off values recommended by the WHO: hemoglobin <11.5 g/dL for children aged <12 years, hemoglobin <12.0 g/dL for males aged 12–14 years and females ≥12 years, and hemoglobin <13.0 g/dL for males aged ≥15 years [27]. Iron deficiency anemia (IDA) was defined as the presence of anemia and serum ferritin (SF) <12 ng/ml in children younger than 15 years and females of all ages, and SF<18 ng/ml in males age 15 years and older. Non-iron deficiency anemia (NIDA) was defined as the presence of anemia and SF≥12 ng/ml in children younger than 15 years and females of all ages and SF≥18 ng/ml in males age 15 years and older. SF best demonstrates reticuloendothelial iron status [2]. A limitation of using SF to determine iron status is its “false” elevation in the context of acute inflammation [28]. For this reason, we used an alternative SF cutoff of 30 ng/ml to define iron deficiency in individuals with concurrent inflammation (CRP level of >8.2 µg/ml) [27]. This definition is conservative in that individuals with ongoing inflammation and iron deficiency are less likely to be misclassified as iron-replete.
In addition, we sought to rule out other causes of NIDA. To this end, we evaluated mean corpuscular volume (MCV) with a cut-off of 100 fL to define macrocytosis. We defined the presence of hyperbilirubinemia, a marker for hemolysis, as indirect bilirubin >1 mg/dL. Splenomegaly was defined as spleen size >2 standard deviations (SD) above the reference mean of a healthy Chinese population [29].
Separate linear regression models were made for each cognitive test to quantify the relationship between performance on cognitive tests and both NIDA and IDA. Children with no anemia were the reference group for all analyses. Analyses were done in SAS software version 9.0 (SAS Institute, Cary, NC). Variables were evaluated as confounders of the relationship between anemia type and cognitive ability if they were independently associated with both anemia and cognitive performance in bivariate analyses. Potential confounders included sex, age, SES, and nutritional status. In our previous work, we found that specific helminth infections were related to cognitive deficits [18], and to AI [5],[6] therefore, the presence of these infections were added to final models for two reasons: 1) to assess whether helminth infection status confounded the relationship between cognition and anemia type such that the true independent predictor was helminth infection and 2) to assess the mechanistic role of AI in mediating the relationship between helminth infections and cognitive impairment. We evaluated whether the beta coefficient for the anemia status covariate changed by more than 10%, suggesting either confounding or inclusion of two variables in a causal pathway. Finally, given pro-inflammatory cytokines are related to both specific deficits in cognitive functioning and NIDA, we evaluated the direct effect of TNF-α, IFN-γ and IL-6 on cognitive outcomes [12],[30],[31], without anemia status in the model.
A significant proportion of the variance of our outcome measures was attributable to clustering within household. Therefore, multi-level statistical analyses were used to adjust for clustering at the household level. Specifically, multivariate random intercept models were implemented using Proc Mixed(with household as random effect and a compound symmetry correlation matrix. Least square (LS) mean values represent the mean adjusted for confounders in multivariable models.
Of the 380 children enrolled in the study, 44 did not complete cognition testing and 14 subjects had incomplete data with respect to anemia status, leaving an effective study sample of 322 children.
Table 1 presents the results of the reliability characteristics for the four cognitive tests. Inter-rater reliability (IRR) tests conducted among 51 subjects ages 8–16 demonstrated good consistency between raters of children's test scores (IRR≥0.98 for all tests). The test-retest analyses among 36 subjects showed good to excellent reliability (0.61–0.89). The degree of consistency among the subscales of the WRAML was fair-good. In addition, we assessed the validity of the tests in our study sample by comparing cognitive test scores to known factors related to cognitive performance such socio-economic status and found significant correlations (rho = 0.216–0.293, all P<0.001).
We first evaluated the role of other potential causes of NIDA (Table 2). We found that no subjects in the NIDA group were macrocytic, making deficiency of folate and vitamin B12 unlikely causes of anemia. In the NIDA group, only one subject had elevated indirect bilirubin (1.07 mg/dL) and one subject had splenomegaly.
We then evaluated our definitions of IDA and NIDA and found these were supported by sTfR relationships among our study subjects. (Table 2) Levels of sTfR in AI are not significantly different from sTfR levels in those with no anemia, as sTfR expression is downregulated by inflammatory cytokines [1]. In our study sample, the IDA group had the highest mean level of sTfR compared to both NIDA and non-anemic (P<0.01).
Other characteristics of the study population at baseline and by anemia status are presented in Table 2. The prevalence of NIDA was 17.4%, and the prevalence of IDA was 17.7%. Children with IDA were significantly more likely to have a low SES compared to those with no anemia (P<0.001). Children with IDA or NIDA had significantly higher prevalences of S. japonicum and hookworm infections compared with non-anemic children.
Figure 1 shows the mean cognitive scores across anemia groups adjusted for age, sex, SES, and BMI z-score.
This study presents strong evidence supporting the existence of a relationship between NIDA and performance on verbal memory and a test that captures abstract thinking after controlling for important confounding variables. To our knowledge, this is the first study to both investigate and demonstrate an association between NIDA and cognitive function. Other work with this cohort had established that the main cause of NIDA in this population is AI, rather than hemolysis or other micro-nutrient deficiencies.[5],[6]. Implicating AI as a cause of cognitive deficits heightens its public health importance given its association with many infectious and non-infectious diseases of developing countries including HIV [32],[33] and malaria [34].
In addition to our results for NIDA, we also provide evidence of a significant association between IDA and performance on an intelligence test after rigorously adjusting for potential confounders. Though IDA was associated with lower scores on all four of the cognition tests before adjusting for confounders, the relationship between IDA and cognition remained for just one of the four tests after controlling for confounders, highlighting the correlations among risk factors for both iron deficiency and cognitive function and the potential for confounding without adjustment [35].
Studies examining cognitive deficits in the context of other exposures have found similar differences in outcomes for the tests we used. Compared to a group of children with no anemia, we found differences in the WRAML verbal memory index of approximately 4 points on a scaled score. To place this in perspective, children with attention deficit hyperactivity disorder (ADHD) had verbal memory index scores that were 6.8 points lower than children without ADHD [36]. Among Filipino children who were normal birthweght and breast fed for 12–18 months versus less than six months, adjusted scores on the PNIT were 1.6 points higher at age 8.5 [37].
In our previous work, S. japonicum infection, which has been demonstrated to cause NIDA, was related to decreased performance on the learning domain of the WRAML [18], whereas NIDA was not related to decreased learning performance scores. Thus S. japonicum's effect on learning may occur through mechanisms other than NIDA. We also evaluated the role of A. lumbricoides because of its relationship to the WRAML verbal memory index, however, A. lumbricoides neither confounded nor acted as the distal determinant in a causal pathway linking this helminth to NIDA and ultimately decreased cognitive performance, suggesting other mechanisms are likely.
Inflammatory cytokines are involved in the development of AI. These cytokines, particularly TNF-α, IFN-γ and IL-6, are also known to induce a syndrome termed “sickness behavior,” characterized by fatigue, impaired sleep, and cognitive dysfunction [31],[38]. Most studies suggest these cytokines exert these effects by entering the brain through the blood-brain barrier and modifying inflammatory responses [39],[40]. Given the possibility that these cytokines could mediate the cognitive deficits in the context of AI through direct effects on the CNS, we evaluated their independent relationship with cognitive domains. The absence of a relationship between pro-inflammatory cytokines and cognition suggests that they are unlikely to be the major factor mediating cognitive impairment in this setting. Thus, it seems that AI-related decreases in iron bio-availability to the CNS, are the primary cause of cognitive impairment in this setting.
Based on both human and animal models, it is well accepted that IDA is related to cognitive impairment, most likely due to decreases in brain iron content. Iron is actively transported into the brain by transferrin receptors and is used for myelination and neurotransmitter synthesis. Iron is also necessary for brain-energy metabolism [41]. In this cohort, it is likely that iron's effect on neurotransmitter synthesis is of greater importance given most myelination occurs during late gestation until 24 months of age. We examined whether children who had normal total body levels of iron, but evidence of decreased iron transport to, and utilization by, host tissues as evidenced by anemia, might not perform as well on tests of cognitive function due to CNS iron deficiency. It is not surprising that, in the context of AI, where iron delivery to host tissues is already sufficiently decreased as to cause anemia, that other host tissues that are dependent on iron would be affected. Also in support of this conclusion is the fact that the same domains of cognitive function were adversely affected by NIDA and IDA, as one would expect if the proximal cause of lower performance is decreased delivery of iron to the CNS. Of note, memory function, as captured by the WRAML verbal memory index, has been demonstrated in many studies to be sensitive to iron status [10],[11].
Study limitations include the cross-sectional design limiting causal inferences and the possibility for residual confounding based on measured and unmeasured covariates. It is possible that children with AI had other infectious or non-infectious diseases that might be related both to AI and cognitive impairment. This is unlikely given children were screened for the presence of significant diseases. Further, the prevalence of HIV in this community is extremely low [42] and malaria is not endemic. S. japonicum infection was not related to the same cognitive deficits as AI. It is possible that there is variability in immune responses to S. japonicum and the many parasitic and other infectious diseases in this setting, such that children with more exuberant pro-inflammatory responses may experience greater AI, which may not be simply related to intensity of infection. It is also possible that our cognition tests, such as the WRAML, may not capture learning and memory abilities in the same manner as when used in US populations. This would likely lead to measurement error, however, rather than introduce bias. In addition, it is likely that some of our subjects actually had both iron deficiency anemia and AI, with some overlap in definitions possible. Current definitions preclude allowing for this combination, though it likely occurs frequently in developing countries. Further, it is possible that the group with NIDA had other causes of anemia, other than AI. We evaluated many other potential causes of NIDA, but cannot rule out other causes including genetic disorders such as thalassemias and G6PD deficiency or vitamin A deficiency. Though these entities may cause NIDA, there are less established mechanisms through which they would lead to cognitive impairment, as opposed to through alterations in iron metabolism in the context of AI as proposed.
Though IDA and AI may both cause cognitive deficits through decreased iron bio-availability to the CNS, the etiology and treatment differ substantially. Iron supplementation is generally provided for treatment of IDA. Iron supplementation in the context of AI, however, has minimal if any benefit, given AI leads to decreased iron absorption and movement of iron into storage forms. The costs and benefits of iron supplementation must be carefully weighed in this context, particularly given recent concerns that it may increase risk of malaria morbidity [43],[44]. The therapeutic approach of choice for AI is treatment of the underlying condition [1].
This study suggests that NIDA, largely due to AI, is associated with cognitive deficits in children. AI, caused by many diseases of lesser-developed countries, may further limit children's ability to take advantage of limited educational opportunities, and can only be addressed by treatment of underlying diseases.
|
10.1371/journal.pntd.0001705 | Living Invisible: HTLV-1-Infected Persons and the Lack of Care in Public Health | Human T-cell lymphotropic virus type 1 (HTLV-1) infection is intractable and endemic in many countries. Although a few individuals have severe symptoms, most patients remain asymptomatic throughout their lives and their infections may be unknown to many health professionals. HTLV-1 can be considered a neglected public health problem and there are not many studies specifically on patients' needs and emotional experiences.
To better understand how women and men living with HTLV-1 experience the disease and what issues exist in their healthcare processes.
A qualitative study using participant observation and life story interview methods was conducted with 13 symptomatic and asymptomatic patients, at the outpatient clinic of the Emilio Ribas Infectious Diseases Institute, in Sao Paulo, Brazil.
The interviewees stated that HTLV-1 is a largely unknown infection to society and health professionals. Counseling is rare, but when it occurs, focuses on the low probability of developing HTLV-1 related diseases without adequately addressing the risk of infection transmission or reproductive decisions. The diagnosis of HTLV-1 can remain a stigmatized secret as patients deny their situations. As a consequence, the disease remains invisible and there are potentially negative implications for patient self-care and the identification of infected relatives. This perception seems to be shared by some health professionals who do not appear to understand the importance of preventing new infections.
Patients and medical staff referred that the main focus was the illness risk, but not the identification of infected relatives to prevent new infections. This biomedical model of care makes prevention difficult, contributes to the lack of care in public health for HTLV-1, and further perpetuates the infection among populations. Thus, HTLV-1 patients experience an “invisibility” of their complex demands and feel that their rights as citizens are ignored.
| Human T-cell lymphotropic virus type 1 (HTLV-1) infection is commonly confounded with Human Immunodeficiency Virus (HIV) infection and it is unknown to many health professionals. It is endemic in many countries and there is no effective treatment available. Although a few individuals have severe symptoms, most patients remain asymptomatic throughout their lives. Further, HTLV-1 is considered a neglected public health problem and limited studies cover specific patients' needs and emotional experiences. To better understand how women and men living with HTLV-1 experience the disease and what issues exist in their healthcare processes, we conducted a qualitative study of both symptomatic and asymptomatic patients at an outpatient clinic at the Emílio Ribas Infectious Diseases Institute in São Paulo, Brazil. We found that the main focus of health staff was on illness risk, but not identifying infected relatives and preventing new infections. This point of view, ultimately neglected patients' complex demands, and overshadows the prevention of new infections and contributes to the lack of care in public health for HTLV-1 infected subjects. Furthermore, this perpetuates the infection among these populations and the patients experience an “invisibility” of their specific needs, such as reproductive rights and feel that their rights as citizens are ignored.
| The human T-cell lymphotropic type 1 (HTLV-1) [1] infection causes a life-long infection for infected subjects. HTLV-1 is endemic in various parts of the world, including Japan and countries in Africa, the Caribbean and South America [2]. It has been estimated that over 10 million individuals are infected with HTLV-1 [3], but most infected persons are asymptomatic and probably are not aware of their serological status. In addition, asymptomatic individuals may still infect sexual partners or offspring.
The most common diseases associated with HTLV-1 infection are adult T-cell leukemia/lymphoma (ATL) and HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [4]. No accurate case statistics exist for ATL or HAM/TSP because these diseases are not reportable by the World Health Organization (WHO). Despite several publication reviews on pathogenesis or molecular biology of HTLV-1 [5], [6], few studies have addressed treatments for HTLV-1 or the psychological issues related to having the disease. Therefore, HTLV-1 infection is considered, and was recently reported, as a neglected disease [7].
Brazil has received worldwide recognition for the country's public health policies, especially with respect to how the nation has addressed the pandemic of HIV/AIDS [8]. Such success against HIV/AIDS was possible only because Brazil's constitution recognizes and guarantees healthcare as a right of every citizen and the country's public health ministry provides a multidisciplinary prevention program and free medication for HIV/AIDS. [8]. However, such public health policies have not been applied to HTLV-1 infection, which is not even listed as an infectious disease that should be addressed by public health action. Thus, HIV/AIDS may overshadow the problem of HTLV-1 in Brazil, and consequently, HTLV-1 has become essentially unknown to most of health professionals and should be considered a public health problem in Brazil [9].
Few HTLV-1 epidemiological studies have been conducted in the general population, but the average prevalence of HTLV-1/2 among donor blood banks is 0.46% nationwide [10] and one few based on study population showed prevalence of 1.7% in Salvador, with 40000 individuals are estimated to be infected by HTLV-1 [11]. The highest prevalence of HTLV-1 has been recorded in São Luís city in Maranhão, with 10.0/1000 and Salvador in Bahia with 9.4/1000 [11]. However, other cities may present lower prevalence despite geographical proximity, which indicates heterogeneous distribution and cluster pattern [12].
In 1993, it was recommended by ministry of health the mandatory tests for hepatitis C and HTLV-1/2 in blood banks. In 2002, Brazil also instituted a policy which provided free testing of HIV/AIDS and syphilis during prenatal care visits as an approach to preventing mother-to-child-transmission (PMTCT). However, the same policy was not provided for HTLV-1 infection and the Brazilian Ministry of Health only published the Guide Recommendations on HTLV Management in 2003, without update [13].
From a public health perspective, prevention of HTLV-1 is crucial, especially because HTLV-1 is a life-long infection and is currently incurable and untreatable. The health field recognizes that disease prevention requires strategies beyond strictly medical approaches. Thus, the aim of this study was to identify perceptions of HTLV-1 by women and men living with this infection, and to better understand their families, subjective issues, needs and how they cope with health care.
The present study was conducted by using participant observation methods [14] in the HTLV outpatient clinic at the Emílio Ribas Infectious Diseases Institute in São Paulo, Brazil from June 2006 to April 2008. This is centenary national public referral hospital specializing in the diagnosis and treatment of individuals with infectious diseases.
Using a thematic script and in parallel an in-depth interviews with 13 individuals (11 women and two men) diagnosed with HTLV-1 infection and without co-infections (see Table 1). Subjects were selected and interviewed during the regular visit to the HTLV clinic. This convenience sample tried to cover the most part of situations such as gender, presence of symptoms, serological couple status and so on [14].The questions of the thematic script were about the subjects history of life, their knowledge and experiences related to the HTLV-1 infection, related diseases and family perception of HTLV-1 serological status disclosure.
Data were analyzed by using life narratives from each subject and constructed content categories from these narratives [15]. The reported names of participants are fictitious in order preserve their privacy and identities.
The study was approved by the Ethical Research Board Committee - CAPPesq (n. 297/06) and the Emílio Ribas Institute of Infectious Diseases Ethical Committee (n. 34/06). The volunteers signed an informed consent form that faithfully follows the resolution of the national Ministry of Health (CNS 196/96) and the Declaration of Helsinki.
While HTLV-1 infection is frequently “invisible” in a literal way, this article aimed to highlight the disease's symbolic invisibility from patients' perspectives and life experiences, as well as patients' relations with healthcare providers.
The HTLV-1 diagnosis is frequently a doubts and fear moment in primary medical care settings. While recalling their original diagnosis, subjects reported an initial shock followed by some feelings of denial (e.g., “it is not true”) or hopefulness for a “cure.” In addition, many participants thought that their illness was HIV infection, which has been similarly reported by Guiltinan (1998) [16]. This erroneous observation was also found in some health professionals speeches. Subsequently, this confusion illustrated the lack of knowledge about HTLV-1 among health professionals and society in general. Furthermore, subjects reported insecurity and doubts about the abilities of their primary healthcare providers; however, patients indicated that these problems were better addressed when they were referred to specialized care.
Interestingly, many subjects reported that “it was better not to know about HTLV” and “not to think about HTLV”. These observations indicated the use of denial as a defense mechanism. Subjects also reported fear of disclosure and that their HTLV-1 diagnoses would be disrupted family relations. Consequently, this perception leads subjects to hide their HTLV-1 diagnosis from everyone, as a “secret.” Such difficulties have also been observed in people living with HIV/AIDS [17], especially as related to their fears of disease stigma [18]. Subjects noted “justification[s]” for keeping their disease status secret, reasoning that “family members are healthy, symptomless and therefore there is no need worry” and they did not want to cause emotional suffering for relatives. These statements illustrated elements of a singular logic underlying the desire for HTLV-1 concealment among family members.
Regarding this “secret,” the thoughts and perceptions between subjects and health providers seem connected. Subjects and providers understood the absence of HTLV-1 symptoms as “normal”, in consequence HTLV-1 infection remained “invisible” because they are focused on risk of sickness. This observation has important implications for understanding the lack of diagnosis among partners and family members.
Except for Maria, all subjects indicated that someone in their family circle (i.e., sexual partners, relatives or children) should be tested for HTLV-1 infection but had not been tested yet. This situation illustrated how poor HTLV-1 disclosure among family members has negative consequences. In Maria's case the lack of disclosure was the possible cause of her infection, because Maria's husband, Walter, was infected by his mother, and hidden his HTLV-1 infection from her. Her husband, Walter, did not reveal his diagnosis to his wife because he thought they were both healthy. Although his wife was pregnant (30 weeks), Walter only sought medical attention when he began to have symptoms of HAM/TSP. Only on this time, his health team requested to his pregnant woman to be tested for HTLV-1 infection.
The symptomatic condition further expanded upon these issues, as patients experience changes in their daily activities, self-image and social lives. The progressive disability associated with symptomatic HTLV-1 infection limited these patients' lives, as illustrated by the sentences below.
The suffering and mobility limitations constitute “narcissistic injuries” [19], which is related to the self-steam in deeper emotional level or in unconscious influence. This situation generated intense anxiety, defense mechanisms and required coping strategies among these patients. In this context, it is understandable that persons living with HTLV-1 had difficulties, which may explain reluctance to share their experiences with their family.
Subjects report that their perceptions and experiences changed when their family members are symptomatic. HTLV-1 becomes present in subjects' lives through the symptoms of their family, and the need to care for these members may trigger additional emotional coping mechanisms.
In cases where subjects know that they were infected by their parents, HLTV-1 infection is viewed as a type of “family heritage” that the subject must “carry for life.” This knowledge may be very painful emotionally, as demonstrated by Alba and Walter:
This lack of disclosure to partners and family on HLTV-1 diagnosis could be explained by poor information regarding infection transmission; however, the speeches revealed an appropriate understanding of transmission modes. On the other hand, some subjects had doubts how they became infected. These subjects were not willing to search for this information, which illustrates their difficulties in dealing with the HLTV-1 diagnoses.
Invisibility among persons living with HTLV-1 may have meaning, especially because patients have reported feeling stigmatized and socially disadvantaged (e.g., “discarded” or “limited”) since receiving their diagnoses [17]. As patients became aware of their HTLV-1 infection, they were forced to reorganize their priorities. Reproductive decisions and sexuality were especially pronounced, as Alba's speech illustrated.
According to subjects, primary health teams are generally unaware of HTLV-1 and did not investigate or recognize the health problems associated with the infection and disease. Subjects reported a poor access to pre- and post-test counseling for HTLV-1 infection. In fact, an author showed that a person can be tested and remained without access to specific information and follow up [16]. Even when subjects had some guidance in counseling, health professionals focused on low risk of developing symptoms and failed to emphasize the risk of transmission to partners or to provide orientation regarding reproductive decisions and maternal-infant transmission risks.
Most interviewers reported they were the only sources of information about HTLV for the health teams, which created uncertainty concerning the teams' medical knowledge. This finding leads to a paradoxical question on healthcare for HTLV: how can a doctor inform and treat patients when himself does not know HTLV? Maria and Maria Rita's speeches, who were both pregnant, illustrated this paradox.
The “invisibility” of HTLV-1 infection should not be a problem for symptomatic patients such as HAM/TSP and ATL cases. However, this study revealed that symptomatic patients had to make “pilgrimages” to several specialists as they searched for a diagnosis. In fact, HAM/TSP patients experienced an average of eight years seeking a diagnosis [20], extending up to 11 years in some cases [21]. Being symptomatic, or considered a “patient,” should generate the search to identify the cause of symptoms. For HAM/TSP diagnosis, the HCP must have the knowledge and make the differential diagnosis, which firstly requires HTLV-1 serology. Unfortunately, HTLV-1 serology is not always available, even in institutions which are specialized in treating mobility problems [21].
Interviewers and medical staff were concerned on the risk of disease development, which is considered “low” for both parties. Healthcare providers and HTLV-infected subjects frequently do not realize that asymptomatic individuals may transmit the virus to partners or offspring. Consequently, preventive processes may become ineffective.
In conclusion, there appeared to be an underlying logic for the “invisibility” in people living with HTLV-1. Even when individuals had access to necessary information regarding the disease, they frequently denied the situation. In consequence, there were implications for self-care and understanding the need for testing family members for HTLV-1 infection.
As people living with HTLV-1 struggle to cope with this health problem, it is possible to observe that these difficulties were logically articulated through the hegemonic healthcare approach. This hegemonic conception of health care is referred to major paradigm in health focused at the biomedical approach. Nowadays, it has been questioned by the researchers at the public health field to contrast this model to the bio-psych-social model which aims the health care in the entire complexity. The biomedical perspective does not encourage prevention activities and health promotion [22]. The current hegemonic context, a true “poor-care” exists toward people living with HTLV-1, further contributing to the perpetuation of this neglected disease by society in endemic areas.
Beyond the “invisibility” of HTLV-1, there is an “invisibility” of the subjects living with the infection or disease, their individual difficulties and idiosyncrasies [23], and their rights as citizens. In order to bring HTLV-1 from “invisibility” to “visibility,” the paradigm of standard healthcare should focus on effective preventive practices, which may consider patients' complex demands. Reports from subjects also indicated the need for studies on sexuality and reproductive decisions among people living with HTLV-1 infection, as these issues had been shown to directly influence prevention efforts.
|
10.1371/journal.pgen.1007178 | Axin phosphorylation in both Wnt-off and Wnt-on states requires the tumor suppressor APC | The aberrant activation of Wnt signal transduction initiates the development of 90% of colorectal cancers, the majority of which arise from inactivation of the tumor suppressor Adenomatous polyposis coli (APC). In the classical model for Wnt signaling, the primary role of APC is to act, together with the concentration-limiting scaffold protein Axin, in a “destruction complex” that directs the phosphorylation and consequent proteasomal degradation of the transcriptional activator β-catenin, thereby preventing signaling in the Wnt-off state. Following Wnt stimulation, Axin is recruited to a multiprotein “signalosome” required for pathway activation. Whereas it is well-documented that APC is essential in the destruction complex, APC’s role in this complex remains elusive. Here, we demonstrate in Drosophila that Axin exists in two distinct phosphorylation states in Wnt-off and Wnt-on conditions, respectively, that underlie its roles in the destruction complex and signalosome. These two Axin phosphorylation states are catalyzed by glycogen synthase kinase 3 (GSK3), and unexpectedly, completely dependent on APC in both unstimulated and Wnt-stimulated conditions. In a major revision of the classical model, we show that APC is essential not only in the destruction complex, but also for the rapid transition in Axin that occurs after Wnt stimulation and Axin’s subsequent association with the Wnt co-receptor LRP6/Arrow, one of the earliest steps in pathway activation. We propose that this novel requirement for APC in Axin regulation through phosphorylation both prevents signaling in the Wnt-off state and promotes signaling immediately following Wnt stimulation.
| The Wnt signal transduction pathway directs fundamental cellular processes during development and in homeostasis. Wnt signaling is deregulated in 90% of colorectal cancers, most of which are triggered by inactivation of the tumor suppressor Adenomatous polyposis coli (APC). In the classical model, APC’s sole role in Wnt signaling is to target the transcriptional coactivator β-catenin for phosphorylation and subsequent degradation, and thereby to inhibit signaling in the unstimulated state. However, the mechanisms by which APC functions remain unknown. Herein, we provide evidence in Drosophila that supports a major role for APC in the direct regulation of the scaffold protein Axin in both Wnt-on and Wnt-off conditions. Our results indicate that APC promotes Axin phosphorylation, which is required not only to inhibit signaling in the unstimulated state, but also to activate signaling following Wnt stimulation. These unanticipated findings support a more active and multifaceted role for APC in Wnt signaling than previously known, and force revision of the current model for APC function.
| The Wnt/β-catenin signal transduction pathway orchestrates fundamental cellular processes during development and in adult homeostasis [1–4]. Wnt signaling is aberrantly activated in many human cancers, including nearly all colorectal cancers, most of which are triggered by inactivation of the tumor suppressor Adenomatous polyposis coli (APC) [2,3]. In the Wnt-off state, the transcriptional activator β-catenin (Armadillo in Drosophila) is targeted for proteasomal degradation by a multiprotein “destruction complex” that includes APC, the scaffold protein Axin, and the kinases glycogen synthase kinase 3 (GSK3) and casein kinase 1α (CK1α) [5–8]. The binding of Wnt ligands to their transmembrane co-receptors, Frizzled (Fz) and low-density lipoprotein receptor protein 5/6 (herein LRP6, Arrow in Drosophila), induces both rapid phosphorylation of the cytoplasmic tail of LRP6 [9–11] and the dissociation of Axin from the destruction complex [12–19]. Specifically, LRP6 phosphorylation creates binding sites for Axin, promoting its rapid recruitment and the assembly of a membrane-associated multiprotein “signalosome” that includes GSK3 and the cytoplasmic protein Dishevelled (Dsh) [9–11,16,20–22]. Signalosome formation inhibits destruction complex activity, leading to β-catenin stabilization, nuclear translocation, and the consequent transcriptional regulation of Wnt target genes [23–25].
Axin is a concentration-limiting scaffold that promotes assembly of both the destruction complex in the Wnt-off state and the signalosome following Wnt stimulation [11,19,26–29]. Axin is phosphorylated under basal conditions and dephosphorylated after Wnt stimulation [12,18,19,30,31]. Both GSK3 and CK1 catalyze Axin phosphorylation [27,30–32]. GSK3-mediated phosphorylation of Axin enhances its interaction with other components of the destruction complex in the Wnt-off state [19,27,30] and promotes Axin’s rapid association with phospho-LRP6 following Wnt stimulation [19]. This association of phospho-Axin with phospho-LRP6 triggers Axin dephosphorylation, inducing a conformational change that inhibits Axin’s interaction with both the destruction complex and the signalosome [12,19,33]. Several hours after Wnt exposure, Axin is degraded [13,20,23,31,34–37]. In addition to phosphorylation, ADP-ribosylation catalyzed by the enzyme Tankyrase (Tnks) also regulates Axin stability and activity. Tnks-mediated ADP-ribosylation targets Axin for ubiquitin-dependent proteasomal degradation in the Wnt-off state [38] and promotes pathway activation following Wnt exposure [37]. ADP-ribosylated Axin accumulates rapidly in response to Wnt stimulation, facilitating the interaction between Axin and phospho-LRP6/Arrow [37].
In the classical model, the major role of APC is to promote the phosphorylation and consequent proteolysis of β-catenin, thereby inhibiting signaling in the Wnt-off state. Whereas APC’s essential role in destruction complex activity is well documented, the mechanism by which APC promotes this activity remains unclear. APC was initially thought to act as a scaffold for the destruction complex, but after further assessment of protein interactions within the complex, this scaffold role was attributed to Axin [5,27,39–41]. Subsequently, APC was thought to promote destruction complex activity by preventing β-catenin dephosphorylation [42]; however, the level of β-catenin phosphorylation remains high in multiple different colon cancer cell lines in which APC is inactivated by mutation [43]. Furthermore, loss-of-function studies in Drosophila demonstrated that APC is required not only to inhibit signaling in the Wnt-off state but also to promote signaling following Wnt stimulation in multiple different tissues, suggesting that the roles of APC are broader than proposed in the classical model [44].
Herein, we present three lines of evidence in Drosophila that support a major role for APC in the regulation of Axin in both Wnt-on and Wnt-off conditions. First, we identify two distinct states of Axin phosphorylation in unstimulated and Wnt-stimulated conditions, respectively: Axin is “fully” phosphorylated by GSK3 in the Wnt-off state and partially phosphorylated following Wnt stimulation. Second, we demonstrate that APC is essential for this GSK3-catalyzed phosphorylation of Axin in both the Wnt-off and Wnt-on states. Third, we find that following Wnt stimulation, the rapid transition in Axin activity and its association with activated phospho-LRP6/Arrow are dependent on APC. Together, these findings reveal a novel role for APC in Axin regulation by phosphorylation that not only prevents signaling in the Wnt-off state, but also activates signaling following Wnt stimulation.
We sought to identify the mechanisms that rapidly regulate Axin during the transition in its roles from the destruction complex to the signalosome after Wnt exposure. We utilized Drosophila embryos, in which Wingless (Wg) is expressed in 14 segmental ectodermal stripes at three hours of development, providing an in vivo model for studying the immediate signaling events triggered by Wingless exposure [45,46]. The initial response of Axin to Wingless exposure is evident in a previously described in vivo system in which a maternal α-tubulin enhancer (mat-Gal4) [47] expresses Axin-V5 within two-fold of endogenous Axin levels [37]. This analysis revealed that the previously known degradation of Axin that occurs in Wingless/Wnt-responding cells several hours after stimulation [13,20,23,31,34–37] is preceded by an earlier phase that occurs within 30 minutes of Wingless exposure, during which Axin accumulates rapidly in Wingless-responding cells, which also have increased levels of Armadillo/β-catenin [37]. Evidence supporting an increase in endogenous Axin levels following Wingless stimulation was provided by biochemical analyses of lysates from embryos collected at stages that immediately precede or follow the onset of Wingless expression [37].
To identify the domains required for Axin regulation in response to Wingless stimulation, we performed a structure-function analysis in this in vivo model. We generated Axin transgenes with deletions in domains required for Axin’s interaction with other Wingless pathway components: the ADP-ribose polymerase Tankyrase (AxinΔTBD-V5), the tumor suppressor Apc (AxinΔRGS-V5), the transcriptional activator Armadillo (AxinΔARM-V5), the phosphatase PP2 (AxinΔPP2-V5), and the signalosome component Dishevelled (AxinΔDIX-V5) (S1 Table). These Axin transgenes were integrated at the same genomic site to permit their direct comparison in the absence of transcriptional position effects.
Next, we examined the effects of the Axin deletions during both the initial stage, when Axin accumulates rapidly in Wingless-responding cells (Stage 9, 40 minutes after the onset of Wingless expression in segmental stripes), and two hours later, when Axin is degraded in the same cells (Stage 10, 120 minutes after the onset of Wingless). As reported previously [37], full-length Axin-V5 was distributed uniformly throughout the ectoderm prior to Wingless expression (S2A–S2C Fig); however, by 30 minutes after Wingless exposure, Axin-V5 accumulated in wide segmental stripes in cells responding to Wingless stimulation (Fig 1A–1C). In contrast, by 120 minutes after Wingless exposure, Axin was degraded in these Wingless-responding cells, as described previously (S1D–S1F Fig) [34,37]. Consistent with previous findings [37], deletion of the Tankyrase binding domain in Axin (AxinΔTBD-V5) resulted in aberrant Axin stabilization in all ectodermal cells and loss of Axin’s initial accumulation in stripes (Fig 1D–1F), whereas the subsequent Wingless-dependent proteolysis of Axin occurred normally (compare S2A–S2C Fig and S1D–S1F Fig). Similarly, we found that deletion of the Apc binding domain (AxinΔRGS-V5), the PP2 binding domain (AxinΔPP2-V5), and the Dishevelled binding domain (AxinΔDIX-V5) each resulted in loss of the early accumulation of Axin in Wingless-responding cells (Fig 1G–1I, 1M–1O and 1P–1R respectively and S1 Table), but did not inhibit the subsequent Wingless-dependent Axin degradation (S2D–S2F, S2J–S2L, S2M–S2O Fig respectively). In contrast, the Armadillo binding domain (AxinΔARM-V5) was neither required for the initial accumulation of Axin in segmental stripes nor for its subsequent Wingless-dependent degradation (Fig 1J–1L, S2G–S2I Fig and S1 Table). Taken together, these results indicate that the TBD, RGS, PP2 and DIX domains are necessary, whereas the Armadillo binding domain is dispensable, for the rapid accumulation of Axin following Wingless stimulation in vivo.
As Tnks-catalyzed ADP-ribosylation promotes the rapid accumulation of Axin in stripes of embryonic cells responding to Wingless exposure [37], we hypothesized that the loss of Axin stripes in some Axin deletion mutants resulted from decreased ADP-ribosylation. To test this hypothesis in vivo, we examined Axin ADP-ribosylation in lysates from larvae expressing either full length Axin-V5 or the Axin-V5 deletions. To detect ADP-ribosylated Axin, which is present at low levels in vivo, we performed pulldowns with the Trp-Trp-Glu (WWE) domain from the E3 ubiquitin ligase RNF146/Iduna, which targets Tnks substrates for proteasomal degradation [48,49]. The WWE domain of RNF146 interacts directly with the poly-ADP in Tnks substrates, and thus permits the sensitive detection of ADP-ribosylated Axin in GST-WWE pull downs (pd) [49,50]. Previous work verified the specificity of this assay, as the WWE pull down of ADP-ribosylated Axin is abrogated in Tnks null mutants [51]. As expected, ADP-ribosylated Axin was readily detected in GST-WWE pulldowns from lysates of larvae expressing full-length Axin-V5, but not in those expressing AxinΔTBD-V5, in which the Tnks binding domain had been deleted ([37] and Fig 2A). Furthermore, analysis of lysates from larvae expressing AxinΔRGS-V5 or AxinΔARM-V5 revealed that both the Apc binding domain and the Armadillo binding domain were dispensable for Axin ADP-ribosylation (Fig 2B and 2C and S1 Table). In contrast, the PP2 and Dishevelled binding domains were required (Fig 2D and S1 Table). These findings indicate that the three Axin deletions that are not ADP-ribosylated (AxinΔTBD-V5, AxinΔPP2-V5, and AxinΔDIX-V5) do not accumulate in stripes following Wingless exposure, and instead are aberrantly stabilized in all ectodermal cells (S1 Table). In contrast, AxinΔARM-V5, which retains ADP-ribosylation, also retains the ability to rapidly accumulate in stripes in Wingless-responding cells (S1 Table). These data suggest that ADP-ribosylation is indeed required for the rapid accumulation of Axin in Wingless-responding cells. Unexpectedly however, AxinΔRGS-V5, another deletion that blocks the accumulation of Axin in Wingless-responding cells, did not inhibit ADP-ribosylation (Fig 2B, S1 Table). Together, these results provide evidence that ADP-ribosylation is required but not sufficient for the rapid regulation of Axin in cells responding to Wingless stimulation.
The accumulation of Axin in stripes in Wingless-responding cells correlates temporally with its role in the activation of Wingless signaling [37]. As the Apc binding domain of Axin is essential for accumulation of Axin in these stripes (Fig 1G–1I), we hypothesized that Apc may play a novel role in the rapid regulation of Axin following Wingless exposure. To test this hypothesis, we examined the role of Apc in the formation of the Axin-V5 stripes. For that purpose, we used mutant embryos in which Apc2 is eliminated both maternally and zygotically and Apc1 is reduced zygotically (see Methods). In these Apc mutants, the early Axin stripes did not form and Axin was aberrantly stabilized in all ectodermal cells (Fig 3G–3I), in sharp contrast with wild-type embryos (Fig 3A–3F). These results suggest that Axin stabilization in Wingless-responding cells may result from inhibition of Apc-mediated Axin degradation. However, the subsequent Wingless-dependent proteolysis of Axin was not affected by Apc depletion (S3 Fig). Thus both Apc depletion and deletion of the Apc-binding domain of Axin prevent formation of the initial Axin stripes in Wingless-responding cells (Fig 3 and Fig 1), but have no effect on the later Wingless-dependent proteolysis of Axin (S3 Fig and S2 Fig). Together, these findings suggest that Apc, and its interaction with Axin are essential for the rapid regulation of Axin in Wingless-responding cells, but dispensable for the Axin degradation that occurs hours after Wingless stimulation.
Apc is known to negatively regulate the basal levels of Axin in vivo; in the absence of Wingless stimulation, either inactivation of Apc or deletion of the Apc binding domain of Axin stabilizes Axin [29,44,52]. Therefore, we postulated that the elevated Axin levels resulting from Apc inactivation might inhibit the rapid regulation of Axin following Wingless exposure. We therefore tested whether an increase in Axin, to levels higher than those resulting from deletion of the Apc binding domain, blocks Axin regulation in response to Wingless exposure. We utilized transgenic flies in which the Axin-V5 transgene is integrated at a genomic site (attP40) that is known to result in higher expression levels than the original integration site (attP33) [37,53]. Indeed, immunoblotting of embryonic lysates revealed that the basal levels of Axin from Axin-V5 integrated at the attP40 site (attP40 Axin-V5) were higher than the basal Axin levels in embryos expressing either full-length Axin-V5, AxinΔRGS-V5, AxinΔTBD-V5, or AxinΔDIX-V5 integrated at the attP33 site (Fig 4A and 4B). Despite the elevated Axin levels in attP40 Axin-V5 embryos, Axin stripes formed rapidly in Wingless-responding cells, and were indistinguishable from observations in embryos expressing attP33 Axin-V5 (compare Fig 4C–4E and Fig 1A–1C, S4 Fig and S1D–S1F Fig) [37]. These findings suggest loss of the rapid Wingless-dependent regulation of Axin is not merely a consequence of increased basal Axin levels. Instead, these data provide additional evidence that Apc, and its interaction with Axin, are important for the rapid regulation of Axin that follows Wingless stimulation [44,52]. We speculate that the stabilization of Axin in Wingless-responding cells results from inhibition of Apc-dependent Axin degradation.
We sought to further investigate the mechanism underlying this novel role for Apc in the rapid transition of Axin in Wingless-responding cells. However, elucidation of Axin regulation under physiological conditions has been hindered by challenges in detecting the low levels of endogenous Axin. To overcome this obstacle, we capitalized on antibodies capable of detecting endogenous Drosophila Axin [54]. In lysates from both third instar larvae and Drosophila S2R+ cells, we observed several forms of Axin with distinct mobility in SDS-PAGE (Fig 5A and 5B). Axin is known to be highly regulated by phosphorylation, which is critical for its function [12,19]. To test whether the distinct forms of Axin we observed in SDS-PAGE resulted from differential phosphorylation, we treated lysates from either larvae or Drosophila S2R+ cells with lambda protein phosphatase (λ-pp). Immunoblots with Axin antibody revealed a downward shift of Axin bands after treatment with λ-pp (Fig 5A and 5B). Axin migrated as two bands after λ-pp treatment, which may reflect its known post-translational modifications other than phosphorylation [38,55–57]. These results reveal that the Axin antibody detects the presence of distinct phosphorylated forms of endogenous Axin.
Supporting previous findings that Wnt exposure induces dephosphorylation of mammalian Axin [12,31], a downward shift in the mobility of Drosophila Axin was observed following exposure of S2R+ cells to Wingless conditioned medium for one hour (Wg CM) (Fig 5C and 5D). Activation of the pathway by Wingless exposure was confirmed by the accumulation of phospho-Arrow (Fig 5D). To determine whether this shift in Axin mobility was indeed dependent on activation of the Wingless pathway, we used RNA-mediated interference to knock down Dishevelled (Dsh), an essential signalosome component [22,58–60]. dsRNA targeting of Dsh resulted in a marked decrease in its endogenous levels, and also abrogated Arrow phosphorylation following Wingless exposure (Fig 5D). Furthermore, in contrast with controls, Dsh knockdown abrogated the mobility shift in Axin following Wingless exposure (Fig 5D). To rule out off-target effects of the RNAi-mediated Dsh knockdown, we repeated this experiment with an independently derived dsRNA that targets Dsh, and again observed inhibition of the shift in Axin’s mobility following Wingless treatment (Fig 5D). To confirm that the downward shift in Axin induced by Wingless exposure resulted from its dephosphorylation, we tested the effect of phosphatase treatment. Treatment of S2R+ cell lysates with λ-pp resulted in no further shift in Axin in the presence of Wg CM, as compared to the absence of Wg CM (compare lanes 2 and 4 in Fig 5C), strongly suggesting that the Wingless-dependent mobility shift in Axin resulted from dephosphoryation. Together, these findings confirm that pathway activation is essential for Axin dephosphorylation following Wingless exposure.
In the classical model, Axin exists in a phosphorylated state under basal conditions and in an unphosphorylated state following Wnt stimulation. Our Axin antibody allowed us to test whether Wingless exposure indeed induces the complete dephosphorylation of endogenous Axin. To examine the state of Axin phosphorylation following Wingless stimulation, we treated S2R+ cells with Wg CM and subjected the cell lysates to λ-pp treatment. Shifts in Axin mobility in SDS-PAGE confirmed that Wingless stimulation induced Axin dephosphorylation, but also revealed that Axin was further dephosphorylated after treatment with λ-pp (compare lanes 3 and 4 in Fig 5C). Thus, importantly, our Axin antibodies detected both “fully” phosphorylated Axin that is present solely in the unstimulated state, and partially dephosphorylated forms of Axin that accumulate within an hour of Wnt stimulation, as revealed by shifts in migration in SDS-PAGE. Therefore, these results expand the classical model for Axin regulation, revealing that endogenous Axin is present in at least two distinct phosphoforms that are dependent on the state of pathway activation: in the unstimulated state Axin is fully phosphorylated, whereas Wingless stimulation induces the partial, rather than complete dephosphorylation of Axin.
Mammalian Axin is phosphorylated by both CK1 and GSK3 [27,30–32]. To determine whether either one or both of these kinases catalyze the phosphorylation of Drosophila Axin detected by our Axin antibody, we used RNAi to knock down either GSK3 (also known as Zeste-white 3 or Shaggy in Drosophila) or CK1α in S2R+ cells. To test whether the effectiveness of GSK3 and CK1α knockdown, we examined the levels of Armadillo (Arm)/β-catenin, a known substrate of both kinases for which phosphorylation results in consequent targeting for proteasomal degradation [5,6,61,62]. We found that basal Arm levels increased upon knockdown of either GSK3 or CK1α, confirming their effective knockdown (Fig 5E). GSK3 knockdown resulted in a downward shift in the migration of all Axin phosphoforms under basal conditions, which was similar to the shifts in Axin mobility observed following treatment with λ-pp (compare Fig 5B and 5E). In contrast, despite the effective knockdown of CK1α, the migration of Axin phosphoforms in SDS-PAGE was largely unchanged (Fig 5F). However, other CK1 isoforms may contribute to the Axin phosphorylation states recognized by our antibody. These findings indicate that the Axin antibody recognizes shifts in Axin mobility resulting primarily from phosphorylation by GSK3, but not CK1α.
We sought to determine if the phosphorylation of Axin that persists following Wingless stimulation requires GSK3. To examine this, we compared Axin phosphorylation in S2R+ cells treated with Wg CM and either with or without RNAi-mediated GSK3 knockdown. Immunoblot of cell lysates with Axin antibody confirmed that Wingless stimulation resulted in a partial dephosphorylation of Axin (Fig 5F). In contrast, GSK3 knockdown resulted in the complete dephosphorylation of Axin, as revealed by further shifts in Axin mobility, which were unchanged following treatment with Wg CM (Fig 5F). These results are consistent with previous findings regarding mammalian Axin, which revealed that Wnt stimulation induces Axin dephosphorylation specifically at GSK3-catalyzed phosphosites [19,30]. These results suggest that Axin phosphorylation by GSK3 is retained at specific sites but lost at others following Wingless stimulation, although conclusive testing of this hypothesis will require the generation of Drosophila Axin antibodies directed at specific Axin phosphosites. Importantly, our findings expand the classical model for Wnt signaling, as they indicate that GSK3-catalyzed Axin phosphorylation is only partially, rather than completely inhibited following Wingless stimulation. Taken together, our findings demonstrate that our Axin antibody recognizes several phosphoforms of endogenous Axin that are dependent on both GSK3 and the state of Wingless pathway activation: “fully” phosphorylated forms of Axin are present in the Wnt-off state, whereas partially phosphorylated forms of Axin are generated during the response to Wingless stimulation.
As Apc is essential for the rapid regulation of Axin following Wingless stimulation in vivo (Fig 3), and as the regulation of Axin phosphorylation underlies Axin’s transition in response to Wnt exposure, we tested whether Apc is important for Axin phosphorylation by subjecting Drosophila S2R+ cells to RNAi-mediated depletion of Apc (Apc1 and Apc2). Unexpectedly, Apc knockdown or GSK3 knockdown had the same effect on Axin: all forms of phosphorylated Axin detected by our Axin antibody were eliminated (Fig 6A). Furthermore, treatment of Apc-depleted cells with Wg CM did not result in further dephosphorylation of Axin, consistent with the findings observed with GSK3 depletion (Fig 6B and Fig 5F). Thus, unexpectedly, these results suggest that Apc is important for GSK3-mediated Axin phosphorylation in both the Wnt-off and Wnt-on states.
To further test this conclusion, we investigated whether the Apc binding domain of Axin is required for Axin phosphorylation. We expressed either Axin-V5 or AxinΔRGS-V5 in wing imaginal discs of third instar larvae, and analyzed the lysates by immunoblots with V5 antibody. This analysis revealed two Axin bands of distinct mobility (Fig 6C and S5 Fig). To determine whether differential phosphorylation resulted in the differences in the mobility of these two bands, we treated larval lysates with λ-pp. Phosphatase treatment resulted in a downward shift in the mobility of the upper band, confirming that the V5 antibody detected both a phosphorylated and dephosphorylated form of Axin-V5 (S5 Fig). Furthermore, deletion of the Apc binding domain of Axin (AxinΔRGS-V5) resulted in loss of the phosphorylated form of Axin-V5 (Fig 6C). These results suggest that the interaction between Axin and Apc promotes Axin phosphorylation in vivo. Taken together with the RNAi-mediated Apc depletion experiments, these findings indicate that Apc is required for the GSK3-catalyzed phosphorylation of Axin.
GSK3-catalyzed Axin phosphorylation and Tnks-catalyzed Axin ADP-ribosylation promote the interaction between Axin and phospho-LRP6 during the initial activation of Wnt signaling, and subsequently, Axin is dephosphorylated [19,37]. As our studies revealed that Apc is required for both the GSK3-mediated phosphorylation of Axin (Fig 6A) and for the rapid regulation of Axin in Wingless-responding cells (Fig 1G–1I and Fig 3G–3I), we hypothesized that Apc may therefore promote the association of Axin with phospho-LRP6/Arrow. We tested this hypothesis using two experimental approaches: co-immunoprecipitation and WWE pull down. First, we transfected Drosophila S2R+ cells with Axin-V5 or AxinΔRGS-V5 (in which the Apc binding domain is deleted), treated these cells with Wg CM, and subjected the cell lysates to immunoprecipitation with V5 antibody. As expected, deletion of the Apc binding domain of Axin resulted in diminished interaction between Axin and Apc (Fig 7A). Furthermore, deletion of the Apc binding domain of Axin also significantly diminished the interaction between Axin and phosphorylated Arrow following Wg exposure (Fig 7A and 7B). Of note, the interaction of Axin with LRP6/Arrow was mapped previously to regions far from the Axin RGS domain [23]. These results suggest that Apc promotes the interaction between Axin and phosphorylated Arrow.
Second, we further tested whether Apc promotes the interaction between Axin and phospho-Arrow using the WWE pull-down assay. ADP-ribosylated Axin is known to accumulate rapidly following Wnt stimulation, which facilitates the interaction between Axin and phospho-LRP6/Arrow [37]. Not only ADP-ribosylated Axin, but also phospho-Arrow is pulled down by WWE; thus the GST-WWE pull down assay permits sensitive detection of the interaction between Axin and Arrow, even at the low levels of endogenous Axin [37]. We tested whether Apc promotes the interaction between Axin and Arrow by depleting Apc using RNAi-mediated knockdown. We treated S2R+ cells with Wg CM and observed the accumulation of phospho-Arrow, indicating robust pathway activation (Fig 7D, left panel). As reported previously, WWE pull downs revealed that Wingless stimulation resulted in increased levels of ADP-ribosylated Axin, and that phospho-Arrow was simultaneously pulled down as a result of its interaction with ADP-ribosylated Axin (Fig 7D, right panel) [37]. Furthermore, by comparison with control, the interaction between ADP-ribosylated Axin and phospho-Arrow, as revealed by the amount of phospho-Arrow pulled down, was diminished significantly by depletion of Apc (Fig 7C and 7D, right panel). These results provide additional evidence supporting the hypothesis that Apc promotes the interaction between Axin and phospho-Arrow following Wingless stimulation.
Finally, this assay allowed us to also test the alternative model that Apc promotes the interaction between Axin and Arrow by enhancing Axin ADP-ribosylation. Since ADP-ribosylation promotes both the transition in Axin activity following Wingless stimulation and the interaction of Axin with phospho-Arrow [37], we examined the effect of Apc knockdown on Axin ADP-ribosylation using WWE pull downs. We found that the levels of ADP-ribosylated Axin pulled down by WWE following Wingless stimulation were unchanged by Apc knockdown (Fig 7D, right panel). Of note, neither Apc (Fig 7D) nor the Apc binding domain in Axin (Fig 2B) is essential for Axin ADP-ribosylation. Therefore Apc function has no effect on the levels of ADP-ribosylated Axin following Wingless stimulation, but is important for the robust interaction between Axin and phospho-Arrow. These findings suggest that the regulation of Axin by Apc following Wingless exposure is not mediated through ADP-ribosylation, and instead that the role of Apc in Axin phosphorylation or perhaps additional Apc-dependent mechanisms are likely critical for this process.
The mechanisms by which APC regulates the Wnt pathway have remained enigmatic despite intense investigation. Here, we provide evidence in a Drosophila model that APC is essential for the GSK3-catalyzed phosphorylation of Axin in both the Wnt-off and Wnt-on states. APC is also critical for the rapid reprogramming of Axin that follows Wnt exposure. These findings were enabled by three recently developed experimental approaches that have deepened our understanding of both Axin regulation and the essential role of APC in this process.
First, an in vivo model in Drosophila embryos has allowed unprecedented analysis of both the immediate regulation of Axin in cells responding to Wingless stimulation, as well as its subsequent regulation hours later [37]. The rapid accumulation of Axin in embryonic ectodermal cells exposed to endogenous Wingless is an in vivo hallmark for the initial activation of Wingless signaling as it requires Wingless activity, occurs rapidly after Wingless stimulation, and correlates with the timing of Wingless-induced Axin dephosphorylation. Using this model, we found that Apc and the Apc binding domain of Axin are essential for Axin’s rapid regulation in Wingless-responding cells, but dispensable for the subsequent Axin proteolysis that occurs hours later. These findings indicate, unexpectedly, that Apc acts not only with Axin in the destruction complex to inhibit signaling, but is also required for the rapid transition in Axin that occurs after Wingless exposure.
Second, a new antibody has enabled detection of Drosophila Axin at endogenous levels in immunoblots [54], allowing analysis of Axin regulation both before and just after Wingless exposure without the need for overexpression. With this antibody, we discovered that distinct phosphorylated forms of endogenous Axin exist in the Wnt-off and Wnt-on states, respectively. GSK3 is required for Axin phosphorylation in the Wnt-off state, in agreement with previous studies [12,31]. Furthermore, we also discovered that GSK3 is required for Axin phosphorylation that is present not only in the Wnt-off, but also in the Wnt-on state. Moreover, we discovered that the dephosphorylation of Axin that is induced following Wnt stimulation is partial rather than complete. Previous work revealed that Axin phosphorylation is retained during its initial association with LRP6 following Wnt exposure, and Axin is subsequently dephosphorylated, which prevents its interaction with both the signalosome and the destruction complex [19]. Thus the partially dephosphophorylated Axin we detected in the Wnt-on state may be the form that associates with LRP6. Importantly, we found that not only GSK3, but also Apc is essential for Axin phosphorylation in both the Wnt-off and Wnt-on states. Depletion of either GSK3 or Apc had the same effect as phosphatase treatment on Axin: both the fully and partially phosphorylated forms of Axin detected by the Axin antibody were eliminated (Fig 5B, 5C, 5E and 5F and Fig 6A and 6B). These findings suggest that APC-dependent phosphorylation of Axin is important for Axin regulation during the activation of signaling following Wingless exposure and further support previous studies demonstrating that GSK3-mediated phosphorylation of Axin is important for the association between Axin and phosphorylated LRP6 [19]. The mechanism by which APC promotes the GSK3-dependent phosphorylation of Axin requires further investigation, but may involve previously proposed functions for APC in regulating protein phosphorylation, GSK3 activity, Axin multimerization, or Axin membrane association [18,51,63–65]; alternatively, APC may prevent Axin dephosphorylation by a phosphatase such as PP1.
Third, a recently developed WWE pull-down assay [49] that isolates ADP-ribosylated Axin, which is present at low levels, has allowed detection of the association between endogenous Axin and phospho-LRP6/Arrow following Wnt stimulation [37]. Prior to the use of this assay, detection of the association of Axin with LRP6 at endogenous levels was limited to its recovery in sucrose density gradients [22,66]. As we discovered that Apc is required for the rapid transition of Axin following Wingless exposure (Fig 3), we used both WWE pull-downs and co-immunoprecipitation to test whether Apc promotes the association between Axin and phosphorylated Arrow, which is among the earliest events triggered by Wingless stimulation [20]. Both Apc depletion and deletion of the Apc binding domain in Axin significantly diminished the association between Axin and phosphorylated Arrow following Wingless stimulation (Fig 7). Previous work indicated that Wnt stimulation reduces the affinity of Axin for the destruction complex through Axin dephosphorylation [12,18,19]. However, other studies suggest that Axin and APC remain bound during signalosome formation following Wnt stimulation [3,29,36]. Our observation that APC enhances the association of Axin with Arrow/LRP6 does not discount the possibility that the APC-Axin interaction remains intact in the signalosome after Wingless stimulation; however, we have not been able to detect this interaction through co-immunoprecipitation studies. Furthermore, APC may promote the association of Axin with LRP6 through its roles in Axin phosphorylation, as described here, multimerization [64], or as yet unidentified mechanisms.
Although the requirement for APC in the destruction complex is well established, the mechanism by which APC promotes destruction complex activity remains unknown. Our results reveal a novel role for APC in promoting the GSK3-catalyzed phosphorylation of Axin in the Wnt-off state. We hypothesize that the physical interaction between APC and Axin facilitates the phosphorylation of Axin by GSK3 within the destruction complex. Previous studies indicated that APC promotes GSK3 activity [18], and that the phosphorylation of Axin promotes stability of the destruction complex [31]. Our findings are consistent with these previous results, and also indicate that APC’s role as a negative regulator of the pathway in the Wnt-off state requires its essential function in GSK3-mediated Axin phosphorylation.
Taken together, our findings have uncovered novel roles for APC in the regulation of Axin phosphorylation and in the initiation of Wingless signaling. Therefore, we propose the following three revisions to the classical model for Wnt signaling (Fig 7E). First, a major role of APC is to regulate the key scaffold protein Axin in both unstimulated and Wnt-stimulated states. Second, APC is essential for the rapid transition in Axin that occurs concomitantly with pathway activation after Wnt stimulation. Third, APC promotes the GSK3-catalyzed phosphorylation of Axin that is necessary in both Wnt-off and Wnt-on states. Therefore, our findings suggest that a key function of APC is to regulate Axin’s essential roles in both the destruction complex and the signalosome (Fig 7E). We speculate that through this novel mechanism, APC both prevents the aberrant activation of signaling in the unstimulated state and promotes physiological signaling following Wnt stimulation, supporting previous loss-of-function studies that revealed essential roles for APC in both the inhibition and the activation of Wnt signaling in vivo [44].
To generate the pUASTattB-AxinΔRGS-V5 transgene, residues T-54 through Y-168 were deleted by PCR-based mutagenesis of pUASTattB-Axin-V5 [37]. The resulting AxinΔRGS-V5 fragment was digested with KpnI and XbaI, and then inserted into the pUASTattB vector at the KpnI and XbaI sites. Transgenic flies were generated using site-specific integration at the attP33 site using phiC31-based integration [67].
Transgenes pUASTattB-AxinΔArm-V5, pUASTattB-AxinΔPP2-V5 and pUASTattB-AxinΔDIX-V5 were generated similarly, with deleted residues indicated in S1 Table.
Other stocks: UASTattB-Axin-V5 at the attP33 and attP40 sites [37], UASTattB-AxinΔTBD [37], C765-Gal4 (Bloomington Drosophila Stock Center, BDSC), Apc1Q8 [68] and Apc219.3 [44]. The maternal α4-Gal4:VP16 driver (mat-Gal4; line 67) contains the maternal tubulin promoter from αTub67C and the 3' UTR from αTub84B [47,69]. All crosses were performed at 25°C.
The primary antibodies used for immunostaining were mouse anti-V5 (1:5000; Invitrogen), mouse anti-Wingless (1:200, 4D4 concentrated antibody, Developmental Studies Hybridoma Bank, DSHB). The secondary antibodies used for immunostaining were goat or donkey Alexa Fluor 488 or 555 conjugates (1:400; Invitrogen). The primary antibodies used for immunoblotting were mouse anti-V5 (1:5000, Invitrogen), guinea pig anti-Axin (1:1000, [54]), rabbit anti-Kinesin Heavy Chain (1:10000, Cytoskeleton), mouse anti-Arm (1:100, N2 7A1, DSHB), mouse anti-alpha-Tubulin (1:10000, DM1A, Sigma), rabbit anti-alpha-Tubulin (1:10000, Sigma), rabbit anti-Gluthathione-S-Transferase (1:10000, Invitrogen), guinea pig anti-Apc2 (GP10) 1:5000 [44]; rabbit anti-phospho-LRP6 [Thr1572] (1:1000, Millipore), rat anti-Dishevelled (1:1000, [70]), and guinea pig anti-Arrow (1:1000, [71]). The secondary antibodies used for immunoblotting were: goat anti-rabbit HRP conjugate (1:10000, Biorad), goat anti-mouse HRP conjugate (1:10000, Biorad), and goat anti-guinea pig HRP conjugate (1:10000, Jackson ImmunoResearch).
For immunostaining, embryos were fixed in 4% formaldehyde, and rehydrated in PBT (phosphate buffered saline [PBS], 0.1% Tween-20, and 1% bovine serum albumin [BSA]). Following incubation for one hour in blocking solution (PBS, 0.1% Tween-20, 10% BSA), embryos were incubated overnight at 4°C with primary antibodies in PBT. After washing with PTw (PBS, 0.1% Tween-20), embryos were incubated with secondary antibodies for one hour at room temperature. Embryos were then washed with PTw and mounted in Prolong Gold (Invitrogen). Fluorescent images were obtained on a Nikon NIS confocal microscope and a Zeiss Axioskop 2 plus fluorescence microscope, and processed and assembled using Adobe Photoshop CS5 and Adobe Illustrator CS5.
For immunoblots, third instar larvae were dissected in cold PBS to remove salivary glands, fat body, gut, and carcass. After removal of PBS, 4X Laemmli loading buffer supplemented with 1M DTT was added and the lysates were vortexed briefly. For embryonic lysates, embryos were lysed in lysis buffer (50mM Tris-HCl [pH 8.0], 100 mM NaCl, 1% NP-40, 10% glycerol, 1.5 mM EDTA [pH 8.0]), supplemented with phosphatase and protease inhibitor cocktail (1:100, Thermo Scientific) and 1μM of the poly(ADP-ribose) glycohydrolase inhibitor ADP-HPD (Enzo Life Sciences). For S2R+ cell lysates used in immunoblots, cells were washed with cold PBS and lysed in 4X Laemmli buffer supplemented with 1M DTT. All the lysates were incubated for 5 minutes at 100°C before SDS-PAGE analysis. Quantification of immunoblots was performed with ImageJ (Wayne Rasband, National Institutes of Health).
For WWE pull downs (pd), GST-WWE beads were generated as described previously [49]. S2R+ cells were treated as indicated, then washed once with cold 1X PBS and lysed in RIPA buffer (50mM Tris [pH 8.5], 300 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS) supplemented with 1uM ADP-HPD and protease and phosphatase inhibitor cocktail (1:100). Lysates were incubated with GST-WWE beads overnight at 4°C. Following incubation, beads were washed four times in wash buffer (50mM Tris-HCl [pH 8.0], 150mM NaCl, 1% NP-40, 10% glycerol, 1.5mM EDTA [pH 8.0]) supplemented with 1uM ADP-HPD and protease and phosphatase inhibitor cocktail (1:100). Bound materials were eluted with 4X sample buffer and resolved by SDS-PAGE, transferred to nitrocellulose membranes and blotted with the indicated antibodies.
S2R+ cells and S2TubWg cells were obtained from the Drosophila Genomics Resource Center. Cells were maintained a 25°C in Schneider’s Drosophila medium + L-glutamine (Gibco) supplemented with 10% (V/V) fetal bovine serum (FBS, Gibco) and 0.1 mg/mL penicillin/streptomycin (Invitrogen) (Complete medium). Cells were transiently transfected using calcium-phosphate DNA precipitation [72].
Plasmids used for transfection of Drosophila S2R+ cells were pAc5.1-Axin-V5 and pAc5.1-AxinΔRGS-V5. To generate the pAc5.1-AxinΔRGS-V5 plasmid, a fragment encoding AxinΔRGS-V5 from pUASTattB-AxinΔRGS-V5 was digested using KpnI and XbaI. The resulting fragment was inserted into the pAc5.1 vector (Invitrogen) at the KpnI and XbaI sites.
To collect Wingless conditioned medium (Wg CM), S2TubWg cells (Drosophila Genomics Resource Center) were grown to confluence, then split 1:3 and incubated at 25°C for 72 hours. Cells were then resuspended in the media and centrifuged at 1000 x rpm for 5 minutes at room temperature; the supernatant was centrifuged again at 5000 x rpm for 5 minutes at room temperature. The resulting supernatant contained the Wingless conditioned medium, which was stored at 4°C. To treat cells with Wg CM, cells were washed 1X with serum-free, antibiotic-free Schneider’s medium; Wg CM or complete medium (CTR) was added and cells were incubated at 25°C for 1 hour.
The generation of double-stranded RNAs (dsRNAs) and dsRNA-mediated knockdown were performed as described previously [73]. Briefly, DNA templates of 200–900 nucleotides in length targeting dsh, CK1α, GSK3, Apc1, Apc2 and white (negative control) were generated by PCR from genomic DNA extracted from S2R+ cells. PCR templates contained T7 promoter sequences on both ends. The DNA templates were amplified using the following primer pairs:
white: forward 5’-T7- ACCTGTGGACGCCAAGG-3’ and reverse 5’-T7- AAAAGAAGTCGACGGCTTC-3’ (sequence from [49]).
dishevelled 39144: forward 5’-T7- TCTGGTGAAGATCCCCATTC-3’ and reverse 5’-T7-CATGCCCAATTCACACTCAC -3’ (sequence from Drosophila RNAi Screening Center).
dishevelled 19803: forward 5’-T7- GCGCCCAGCATGTCG and reverse 5’-T7- AACGATCTCCTCGAGGTTA-3’ (sequence from Drosophila RNAi Screening Center).
CK1α: forward 5’-T7- CACCCTGGTCATGGACC-3’ and reverse 5-T7- TCGAAGCGCAGGCTACG-3’ (sequence from [74]).
GSK3 23946: forward 5’-T7- AGCTACGCATGGAGGGTAA-3’ and reverse 5’-T7-TTACCAGATCCGGGTCCAC-3’ (sequence from Drosophila RNAi Screening Center).
GSK3 40670: forward 5’-T7- CGAGCCGAATGTATCGTAT-3’ and reverse 5’-T7-TTCTGCCATGGATGACTCTT-3’ (sequence from Drosophila RNAi Screening Center).
Apc1: forward 5’-T7-ACCATTCGTAGCTACTGCACCGAA-3’ and reverse 5’-T7-ATTGATGGCTATTGGCTGCGAGGA-3’.
Apc2: forward 5’-T7- GTCCACAATAATCCGGA-3’ and reverse 5’-T7-GTATTGCTGGTCCTCGGGACA-3’.
dsRNAs were transcribed from PCR generated templates using the T7 Megascript kit (Ambion) according to manufacturer’s instructions. For RNAi-mediated knockdown, S2R+ cells were plated in 10 cm2 plates with 2.5 mL of serum-free, antibiotic-free Schneider’s medium + L-glutamine. 25 μg of each dsRNA was added to the medium and cells were incubated with gentle rotation at room temperature for 1 hour. Following incubation, 2.5 mL of complete medium were added and cells were incubated at 25°C. After 24 hours, medium was removed from the cells. This procedure was repeated once every 24 hours for a total of 96 hours. For GSK3 knockdown, an equivalent amount of GSK3 23946 and GSK3 23946 dsRNA (25 μg of each) were mixed and added to the medium, 50 μg of white or CK1 dsRNA were used in the same experiments.
S2R+ cells or third instar larvae were lysed in lysis buffer (1% NP-40, 150 mM NaCl, 50mM Tris-HCl, 50 mM NaF) supplemented with protease inhibitor cocktail (Protease Arrest, GBiosciences). Lysates were treated with λ protein phosphatase for 30 minutes according to manufacturer’s instructions (NEB).
For immunoprecipitation experiments, S2R+ cells were harvested 48 hours after transfection, washed with 1X PBS, then lysed in lysis buffer (50mM Tris-HCl [pH 8.0], 100mM NaCl, 1% NP-40, 10% glycerol, 1.5mM EDTA [pH 8.0]) supplemented with 1uM ADP-HPD (Enzo Life Sciences) and phosphatase and protease inhibitor cocktail (1:100, Thermo Scientific). Lysates were incubated with mouse anti-V5 antibody (Invitrogen) overnight at 4°C, followed by addition of protein A/G-sepharose beads (Santa Cruz) for 1 hour at 4°C. Beads were washed three times with wash buffer (50mM Tris-HCl [pH 8.0], 150mM NaCl, 1% NP-40, 10% glycerol, 1.5mM EDTA [pH 8.0]) supplemented with 1uM ADP-HPD and phosphatase and protease inhibitor cocktail (1:100), and boiled with 4X sample buffer supplemented with 1M DTT. Samples were resolved by SDS-PAGE and immunoblotted with the indicated antibodies.
Student’s t-test with Welch’s correction was performed using Prism (GraphPad Software Inc., CA, USA) to compare two groups for all data sets. P values are provided in the figure legends.
|
10.1371/journal.pntd.0004498 | Successful Control of Ebola Virus Disease: Analysis of Service Based Data from Rural Sierra Leone | The scale and geographical distribution of the current outbreak in West Africa raised doubts as to the effectiveness of established methods of control. Ebola Virus Disease (EVD) was first detected in Sierra Leone in May 2014 in Kailahun district. Despite high case numbers elsewhere in the country, transmission was eliminated in the district by December 2014. We describe interventions underpinning successful EVD control in Kailahun and implications for EVD control in other areas.
Internal service data and published reports from response agencies were analysed to describe the structure and type of response activities, EVD case numbers and epidemic characteristics. This included daily national situation reports and District-level data and reports of the Sierra Leone Ministry of Health and Sanitation, and Médecins Sans Frontières (MSF) patient data and internal epidemiological reports. We used EVD case definitions provided by the World Health Organisation over the course of the outbreak. Characteristics assessed included level of response activities and epidemiological features such as reported exposure (funeral-related or not), time interval between onset of illness and admission to the EVD Management Centre (EMC), work-related exposures (health worker or not) and mortality. We compared these characteristics between two time periods—June to July (the early period of response), and August to December (when coverage and quality of response had improved). A stochastic model was used to predict case numbers per generation with different numbers of beds and a varying percentage of community cases detected.
There were 652 probable/confirmed EVD cases from June-December 2014 in Kailahun. An EMC providing patient care opened in June. By August 2014 an integrated detection, treatment, and prevention strategy was in place across the district catchment zone. From June-July to August-December 2014 surveillance and contact tracing staff increased from 1.0 to 8.8 per confirmed EVD case, EMC capacity increased from 32 to 100 beds, the number of burial teams doubled, and health promotion activities increased in coverage. These improvements in response were associated with the following changes between the same periods: the proportion of confirmed/probable cases admitted to the EMC increased from 35% to 83% (χ2 p-value<0·001), the proportion of confirmed patients admitted to the EMC <3 days of symptom onset increased from 19% to 37% (χ2 p-value <0·001), and reported funeral contact in those admitted decreased from 33% to 16% (χ2 p-value <0·001). Mathematical modelling confirmed the importance of both patient management capacity and surveillance and contact tracing for EVD control.
Our findings demonstrate that control of EVD can be achieved using established interventions based on identification and appropriate management of those who are at risk of and develop EVD, including in the context of ongoing transmission in surrounding regions. Key attributes in achieving control were sufficient patient care capacity (including admission to specialist facilities of suspect and probable cases for assessment), integrated with adequate staffing and resourcing of community-based case detection and prevention activities. The response structure and coverage targets we present are of value in informing effective control in current and future EVD outbreaks.
| Ebola Virus Disease (EVD) is a severe illness that is usually spread from person to person through caring for someone who is sick, or if they die, contact with their body during their funeral. The recent EVD outbreak in West Africa caused illness and death in many thousands in Guinea, Sierra Leone and Liberia. It has been the largest and most difficult to control of any EVD outbreak in history, and this led to doubts as to the effectiveness of established control measures. Our study describes the successful control of EVD in a rural district of Sierra Leone. As in previous outbreaks, we found that control was achieved by working with communities to identify people who may have been exposed to EVD; if they then became sick, their early admission for testing and care to specialised centres that have equipment and procedures to prevent EVD passing on to staff or other patients, and safe burial of those who die of EVD by trained workers with appropriate protective equipment. We describe the resources and response structure needed to implement such measures effectively, information that will assist in controlling future outbreaks.
| Ebola Virus Disease (EVD) is a severe viral illness of humans, with recorded case fatality rates of up to 90% in past outbreaks [1]. The current EVD outbreak, caused by the Zaire strain of Ebolavirus, is the first recorded in West Africa. It commenced in Guinea in December 2013, then spread to neighbouring Liberia and Sierra Leone, and has exceeded all previous EVD outbreaks collectively in terms of number of cases and geographical distribution. Sierra Leone has reported the highest number of cases [2], with EVD overwhelming existing health services and response structures in many areas of the country [3]. Prior to the EVD outbreak, Sierra Leone already faced considerable challenges. A decade long major civil war from 1991–2002 has contributed to 40% of the population living in extreme poverty [4]. With 0·03 physicians per 1,000 population, its health system struggles to provide even basic services, resulting in high levels of morbidity and mortality due to endemic diseases such as malaria [5].
Transmission of EVD is primarily through close, unprotected contact with the body fluids of patients in the late stage of illness or following their death [6, 7]. These transmission characteristics facilitate and necessitate identification of contacts at risk of infection, their early admission to safe patient care facilities if illness occurs, and safe burial procedures for those that die. Such measures have successfully controlled EVD in the other settings [8–10], and are key to controlling the current outbreak. However, the structure and type of response activities leading to successful EVD control have not been well documented in the context of an outbreak of the scale and geographical distribution of the current outbreak in West Africa. The aim of this paper was to describe EVD transmission and the implementation of successful control measures in Kailahun District, Sierra Leone, and to consider implications for EVD control more broadly.
Kailahun District (population of 350,000–450,000 [11] in an area of 4,859 km2) borders the region of Guinea in which the West African EVD outbreak commenced. Close ties between communities on both sides of this border facilitated the movement of people, and in this case, of disease. The first confirmed cases of EVD in the district were reported in late May 2014, and transmission peaked in August 2014 [12]. Following 42 continuous days without a confirmed case, Kailahun District was declared Ebola-free on 22 January 2015 [13].
We identified published and internal data from agencies involved in the EVD response in Kailahun district from June 2014 to June 2015. This included daily national situation reports and District-level data and reports of the Sierra Leone Ministry of Health and Sanitation, and Médecins Sans Frontières (MSF) patient data and internal epidemiological reports. We used EVD case definitions provided by the World Health Organisation over the course of the outbreak [2]. These data were used to map numbers of confirmed and probable EVD cases and response activities within Kailahun District. Data from the MSF Kailahun Ebola Management Centre (EMC) were analysed to assess epidemiological characteristics of confirmed EVD patients admitted to care. Characteristics included reported exposure (funeral-related or not), time interval between onset of illness and admission to the EMC, work-related exposures (health worker or not) and mortality. We then compared these characteristics between patients admitted during two time periods; June to July (the early period of response) compared with August to December (when coverage and quality of response had improved). This analysis relates to EVD transmission within Kailahun district. EVD patients were at times referred and brought by ambulance to Kailahun EMC from holding centres in other districts without adequate EMC capacity. Numbers admitted each day were dependant on EMC bed availability in Kailahun. These external referrals were excluded from this analysis. This study involved analysis of routinely collected surveillance and program data during the response to EVD in Kailahun district. Such analyses were conducted by program staff responsible for routine data management in response agencies, and results are presented in aggregate, anonymised form. This paper presents an analysis of routinely collected program data, an essential component of ongoing evaluation and quality improvement of clinical services in all settings. In keeping with international standards [14], in order to facilitate such essential quality assurance processes while at the same time safeguarding patient confidentiality, the MSF Ethics Review Board has determined requirements for exempting specific proposals addressing secondary analyses of routinely collected patient data from the requirement for formal human research ethical review [15, 16]. Specific written or verbal consent was therefore not sought from patients whose data contributed to this analysis. Data were analysed by staff responsible for entering and analysing data for routine patient care and program monitoring purposes and aggregate results are reported in this paper. Because of ethical restrictions, individual-level data cannot be shared publicly. The data are available under the terms of MSF's data sharing policy, found at http://fieldresearch.msf.org/msf/handle/10144/306501. Applications for access to MSF Datasets offered on the online catalogue should be submitted to MSF via [email protected].
We used mathematical modelling to evaluate the impact on EVD transmission of increased EMC capacity and improved case detection using a stochastic discrete-time model based around a local contact population (see full description in supporting material). As our aim was not to estimate parameters for Kailahun or to predict future case numbers, but to compare the broad effect of control measures, we have adopted a relatively simple model with a small number of parameters. The model is stochastic to allow for elimination of disease under successful control measures, and includes susceptible depletion within the local contact population. We consider the relative effect of increasing bed numbers under three surveillance response scenarios: Scenario A assumed surveillance activities detected 35% of EVD cases, Scenario B that 83% of cases were detected; the levels in these scenarios were based on analysis of program data (Table 1). Under Scenario C, we assumed 83% case detection, along with early admission and therefore less transmission from known contacts once symptomatic [7]. The model started with 100 cases, consistent with EVD case numbers in Kailahun over a generation of transmission (approximately 2–3 weeks) in June 2014. We assumed that patients not admitted to an EMC had a reproduction number of 1·7 [17], while patients that were admitted had a 50% reduction in their reproduction number to 0·85 [18]. Additionally, with earlier case detection and admission (Scenario C), we assumed the reproduction number decreased to 0·50 in secondary cases within each identified chain of transmission. We assumed an initial contact population for each case of 30 people and that 90% of transmission events occurred within this contact population. We conducted sensitivity analyses around all parameters to ensure that broad findings were not sensitive to these assumptions (see Supporting Material).
The chiefdom of Kissi Teng in Kailahun district closely borders the epicentre of the EVD outbreak in Guinea (Fig 1). EVD transmission was first identified in Kailahun in late May 2014, followed by a rapid increase in cases in June related to the funeral of a traditional healer in contact with EVD patients from Guinea [12]. 365 Ebola-related deaths have now been epidemiologically linked to this single funeral. The outbreak then spread from the Kissi Teng chiefdom to other chiefdoms throughout Kailahun district (Fig 1). The local Ministry of Health and Sanitation reported 652 EVD cases in district residents from June to December 2014, including 565 laboratory confirmed and 87 probable cases.
Response activities were initiated in May 2014, and early challenges included limited or no availability of most response activities, and community mistrust and fear affecting compliance with even the few available measures. By August 2014, safe patient care and burial services, surveillance, contact tracing and health promotion were in place and had increased in coverage (Fig 2). Contact tracers were community health workers (CHW) who prior to the EVD outbreak had been working with the District Health Management Team ensuring that essential services reach remote hard-to-reach communities where health centres are not present. They were selected by community stakeholders based on specific criteria, which included: resident in the community they represent; over 18 years of age; able to read and write; young and energetic; has interest in and willing to volunteer; and has moral standing and is well- respected in the community. A training package and standard operating procedures on contact tracing were developed by the Ministry of Health and Sanitation (MOHS) and training conducted based on these materials. Substantial increases in contact tracing staff and surveillance supervisors over time meant coverage of these measures increased from 1·0 surveillance personnel per confirmed case detected in June 2014 to 8·8 per confirmed case detected in September 2014. However, even with this increase, surveillance staff reported working long hours in order to ensure appropriate follow up of all reported cases and contacts. Quarantine measures restricting movement of household members of confirmed cases commenced in September 2014, and infection prevention and control activities in general health services commenced in October 2014.
The only EMC that operated in Kailahun opened on 26th June 2014. Information on the structure of clinical services within the EMC has been published previously [19]. From the outset all suspect, probable or confirmed cases could be referred to the EMC for epidemiological and clinical screening, laboratory testing and admission when warranted. Patients self-presented directly to the EMC, or were referred from general health services or from EVD holding facilities. From 36 beds in June 2014, EMC capacity expanded periodically, to 80 beds by early August and 100 by September in order to ensure that suspect, probable or confirmed cases of EVD were not refused or delayed admission once identified within the district. From September 2014 onwards laboratory results were generally available within 24 hours of sample collection for those admitted to the EMC.
Community-based health promotion activities constituted a core component of response. These health promotion activities were linked to patient care and surveillance through the EMC health promotion team. This team worked closely with patients and their families during admission to build understanding of the disease and trust in response measures. Accompanying patients who recovered and were discharged home was an important aspect of this team’s role, providing an opportunity to address stigma and enabling survivors to act as advocates within their community for early admission to care. The MSF health promotion team recruited and provided training and support to community health promotion volunteers, aiming to have one in each village in the district. These community health promotors (CHPs) were persons able to read and write in English and respected in the community. Their main role was to conduct health promotion activities in the village focussed on behaviour change related to risk factors for EVD transmission (caring for sick people and burial practices), prevention, health seeking behaviour and the services provided by the EMC. Specific practices in each village around these issues were also documented, and any rumours related to EVD that may influence community behaviour were addressed. CHP’s also reported cases and worked closely with contact tracers. This cooperation improved as the number of contact tracers increased.
CHP supervision and training was provided at primary healthcare unit (PHU) catchment area level with the support and agreement of the officer in charge of the PHU and village chiefs. The MSF health promotion team conducted CHP supervision and ongoing training, prioritising villages reporting cases for daily visits by both the EMC health promotion team and CHPs. Villages without current cases were visited either weekly or biweekly by supervisors and more frequently by CHPs. CHPs also met at PHU level weekly to discuss problems and solutions among themselves, and conducted discussions with community leaders, traditional healers and birth attendants, other groups and the general community. The number of CHP’s working in Kailahun increased to more than 600 by August 2014 and eventually to almost 1000 (one in each village).
Initially there was limited interaction at district level between response agencies, and between these agencies and affected communities. However, by September 2014 coordination across response agencies and health system levels was in place at district level, led by the surveillance staff of the MOHS District Health Management Team. Information exchange between district staff and community-level activities was directed through surveillance officers based in chiefdoms throughout the district. Surveillance staff also coordinated food and water distribution and hygiene activities in quarantined households assisted by contact tracing CHWs. Chiefdom leaders were members of the Ebola Taskforce. Local health facilities such as PHUs were also part of this network through their health officers.
From its opening until January 2015, there were a total of 614 admissions to the EMC from Kailahun District, of which 354 (58%) tested positive for Ebola virus. The proportion of confirmed or probable cases of EVD detected in the district who were admitted to the EMC increased from 35% in June-July to 83% in August-December (p<0·001) (Table 1). The overall case fatality for EVD admissions was 58%, with no significant variation over time (case fatality was 60% in June-July and 58% in August-December 2014 (p = 0·72).
The proportion of confirmed cases from the District admitted to the EMC less than three days following onset of symptoms increased significantly (p<0·001), from 19% in June-July to 37% in August-December 2014 (Table 1). The proportion of confirmed cases reporting funeral contact decreased from 33% to 16% (p<0·001) over this period. Amongst Kailahun EMC admissions, 18 were health staff, constituting 4% of EVD cases reported in the district from June to July 2014, and 6% of those from August to December (p = 0·35). Exposures reported by health staff who were infected included work in general health facilities, work in the EMC, unpaid care of family members and relatives and community-based private health service provision. Individual health workers with EVD generally reported multiple exposures, both health facility and community-based.
Table 2 gives the results of mathematical transmission modelling to compare the relative effects of increased EMC capacity and increased surveillance under three scenarios, showing the mean number of cases in each generation, together with cumulative case numbers and the median reproduction number (R). Estimates are based on 5,000 simulations of the stochastic model, and 95% intervals about these estimates are provided in the Supplementary Material. Outbreak control is achieved when the median reproduction number declines to less than 1. We see that control cannot be achieved without sufficient bed capacity in the EMC: all three scenarios fail to reduce the reproduction number below 1 when there are 25 or 50 beds. However, increasing bed capacity alone is insufficient; under low levels of case detection and referral (Scenario A), increasing capacity from 75 to 100 beds has no effect on the outbreak because cases are not detected to fill these beds. Increasing the proportion of cases detected from 35% (Scenario A) to 83% (Scenario B) is required to bring the outbreak under control, once there are sufficient beds. Earlier isolation of epidemiologically linked cases through contact tracing (Scenario C) can further reduce case numbers. Sensitivity analyses around our choice of reproduction numbers confirmed the broad findings that both bed capacity and integrated surveillance are required to achieve control (see Supplementary Material). Findings under an analogous deterministic model were broadly similar to the mean of the stochastic model, although the deterministic model was unable to capture fade-out of disease.
During the 2014–15 Ebola epidemic, a comprehensive response against Ebola was established in Kailahun district, Sierra Leone. The majority of EVD cases detected in the district were admitted to the EMC, and this proportion increased significantly over time. There was also a statistically significant increase in the proportion of patients admitted within 3 days of disease onset and in the proportion reporting funeral contact compared to the early period of the outbreak response. This suggests that once a comprehensive response was in place, the majority of EVD patients resident in the district were identified, safely cared for when symptomatic through admission to an EMC, and if they died, received a safe burial; thus decreasing the risk of EVD transmission to their family and community. A recent paper has published similar findings in relation to reduction in funeral attendance [20]. Our findings build on this by identifying interventions associated with such improvements in epidemiological risk factors for transmission.
Although challenging to implement, adequate EMC bed capacity, as was called for early in the West African outbreak [21], is vital to EVD control. This capacity must be available promptly, and must be accessible to patients with suspect, probable or confirmed EVD in order for early case detection to have an impact on transmission in the community and general health services through safe care of those who are ill. Once established, rapid availability of laboratory diagnostic results supported appropriate patient management and contact tracing activities.
A second key finding is that surveillance and contact tracing are essential to increasing numbers admitted to care and in reducing delay to admission. We found that 8·8 personnel per confirmed case, while imposing a heavy workload on staff, achieved adequate surveillance and contact tracing coverage in a rural, geographically dispersed population where individuals were known to each other at community level. Applicability of this coverage level in urban areas would be dependent on population density, catchment size and social connections within the community.
The value of the mathematical modelling is that it allows us to assess the effect of increased bed numbers and improved surveillance separately, while our data (Fig 2) reflects a simultaneous increase in bed numbers and epidemiologists conducting contact tracing. Nevertheless, our modelling finding that contact tracing is needed in addition to bed numbers is supported by the trend in case numbers, which began to decline in August following an increase in contact tracing.
Community engagement is crucial if surveillance and contact tracing activities are to achieve high levels of early case detection and appropriate management [22]. Unless communities understand the value of response measures, lack of awareness and fear will prevent compliance. In Kailahun, health promotion staff worked with community leaders, with patients and their families, with survivors, and with surveillance and contact tracing staff to build trust and convey appropriate messages around EVD control. Increased community awareness and understanding is likely to have been important in improving early admission to care and reduced funeral contact. Health promotion activities also covered other behaviour modification messages such as avoiding contact with the sick and regular handwashing, and this may also have contributed to decrease in community transmission.
The clearly defined catchment zone of Kailahun district was of a size that allowed it to function as a single health service area, enabling close links between individuals involved in response activities within and across different health system and administrative levels and with communities. These links enabled treatment, detection, and prevention activities to be integrated into a combined response strategy. This may be one factor which contributed to successful control in Kailahun, but is challenging to replicate in other settings such as large urban centres. The capital of Sierra Leone, Freetown, is such a setting. The Freetown area reported its first case of confirmed EVD on 23 June 2014, and case numbers peaked at 200–250 confirmed EVD cases per week in November 2014 [23]. Cases in this area began decreasing in December 2014, associated with an increase in EMC capacity, but contact tracing and case detection remained poor [12]. A program commenced in January 2015 in Freetown to support community-level response activities such as surveillance and contact tracing, with specific agencies each responsible for supporting a defined zone, and there was a further decrease in transmission from January 2015 onwards following implementation of these measures [23]. Based on our findings, we support this as a promising strategy for replicating the structure of response in Kailahun within a much larger urban setting.
Mathematical modelling of intervention scenarios assessed the importance of both EMC bed capacity and surveillance and control activities to detect cases early in their infectious period. There has been considerable modelling work done to predict future Ebola case numbers [17, 24, 25]. The aim of the modelling work presented here was to assess the relative importance of these components of Ebola response. The results were highly robust to model assumptions, and confirmed the epidemiological findings of this paper that both components are needed to control EVD.
Mortality for those admitted to the EMC remained similar throughout the outbreak in Kailahun. Data from past outbreaks and from other areas of Sierra Leone show that the average time from disease onset to death is 4 days, and the majority of those who die do so within the first 10 days of disease onset [17, 26]. As response improved, the proportion of patients admitted within 3 days of symptom onset to the Kailahun EMC almost doubled. As mortality is highest in the first week of illness, we expect that an increase in the proportion of patients admitted early in their illness would be associated with increased overall mortality in those admitted to the EMC. Such an increase may have hidden any reduction in mortality in EMC patients due to factors such as earlier access to or improvements in care.
Health workers comprised 5·1% of EVD cases detected in Kailahun. This is consistent with findings from the country as a whole, with 5·2% of cases reported up to November 2014 in health workers [27]. Despite an overall decrease in case numbers over these periods, the proportion of cases who were health care workers did not decrease over time. It indicates that health staff continued to be at risk of EVD even in the latter stages of the outbreak. Health workers reported a range of health facility and community-based exposures. There was also limited focus in Kailahun on infection prevention and control activities in general health facilities until late in the outbreak, with the first training for health staff held in the district in October 2014. Even at that time, there continued to be very limited availability of chlorine, gloves, masks and water at the level of PHUs. This was also the case in Sierra Leone more generally [28, 29]. Delays in implementing infection control in general health services are likely to have contributed to health worker infections in Kailahun, and must be a focus of current and future EVD response for general health facilities as well as EMC’s.
It is clear from our results that all components of response are important and must be funded appropriately in order to ensure timely and adequate coverage of affected areas. Our findings suggest that in Kailahun there were delays in the implementation and scale-up of control activities such as recruitment of adequate contact tracing and health promotion staff and infection prevention and control in general health services. Recent commentary has identified delays in the availability and disbursement of funding during this current outbreak [30]. The contribution of such delays to the scale of the outbreak is an important area for future research and analysis.
Limitations to this study include that we based our conclusions on cases detected and response activities within a single district. Although we believe that the response measures implemented in Kailahun contributed to control of EVD transmission, there may have been other contributing factors we are unaware of. Data on epidemiological characteristics such as date of onset and type of exposure were from patients confirmed with EVD in the Kailahun EMC facility, and therefore may not reflect the situation in the general community. The quality of data collection is likely to have improved over time as training and staffing increased, particularly for data from surveillance and contact tracing activities. Any bias due to improved case detection would result in an apparent increase in cases over time, suggesting the decrease in cases seen is a reliable finding. Data from the EMC also varied in quality. Variables such as age and sex were recorded consistently. However, completion of data on variables such as reported date of onset and source of exposure varied over time, with no consistent pattern. These variables were possibly affected by staff workload, but this affected all admissions during a particular period and therefore is unlikely to result in bias. Data on behaviour and contact history were based on patient report, and it is possible that public messaging encouraging early presentation and avoiding funeral contact may have influenced reporting rather than actual behaviour. Our mathematical model was designed for a small population centre where most transmission occurs within local population groups; transmission patterns likely differ somewhat in larger population centres such as Freetown.
Conclusion: Our study describes EVD transmission and the implementation of successful control in a rural district of Sierra Leone in a period when other areas of the country and the region were experiencing high case numbers and limited control. The key factor in achieving control appears to be admission of the majority of EVD cases to an appropriate care facility early in their illness. This requires sufficient EMC bed capacity (including the capacity to admit both suspect and probable cases for assessment) integrated with adequate community-based case detection and prevention activities. Information we present on the structure, type and level of intervention associated with achieving control in Kailahun may inform effective EVD response in future outbreaks.
|
10.1371/journal.pgen.1005391 | Emergence, Retention and Selection: A Trilogy of Origination for Functional De Novo Proteins from Ancestral LncRNAs in Primates | While some human-specific protein-coding genes have been proposed to originate from ancestral lncRNAs, the transition process remains poorly understood. Here we identified 64 hominoid-specific de novo genes and report a mechanism for the origination of functional de novo proteins from ancestral lncRNAs with precise splicing structures and specific tissue expression profiles. Whole-genome sequencing of dozens of rhesus macaque animals revealed that these lncRNAs are generally not more selectively constrained than other lncRNA loci. The existence of these newly-originated de novo proteins is also not beyond anticipation under neutral expectation, as they generally have longer theoretical lifespan than their current age, due to their GC-rich sequence property enabling stable ORFs with lower chance of non-sense mutations. Interestingly, although the emergence and retention of these de novo genes are likely driven by neutral forces, population genetics study in 67 human individuals and 82 macaque animals revealed signatures of purifying selection on these genes specifically in human population, indicating a proportion of these newly-originated proteins are already functional in human. We thus propose a mechanism for creation of functional de novo proteins from ancestral lncRNAs during the primate evolution, which may contribute to human-specific genetic novelties by taking advantage of existed genomic contexts.
| Although gene duplication has been believed as a predominant mechanism for creating new genes, recent reports suggested that new proteins could evolve “de novo” from non-coding DNA regions. These de novo genes are also named as “motherless” genes due to their lack of ancestral proteins as precursors, while recently we and others found that lncRNAs may represent an intermediate stage of their origination. To further elucidate this lncRNA-protein transition process, here we identified 64 hominoid-specific de novo genes and report a new mechanism for the origination of functional de novo proteins from ancestral non-coding transcripts: These non-coding “precursors” are generally not more selectively constrained than other lncRNA loci; and the existence of these de novo proteins is not beyond anticipation under neutral expectation; however, population genetics study in 67 human individuals and 82 macaque animals revealed signatures of purifying selection on these genes specifically in human population, indicating a proportion of these newly-originated proteins are already functional in human. We thus propose a mechanism for creation of functional de novo proteins from ancestral lncRNAs during the primate evolution.
| Although it is a generally accepted notion that gene duplication is the major way to create new genes [1–3], numerous cases have been reported in recent years demonstrating in multiple different species the creation of new proteins out of ancestral non-coding DNAs [4–18]. Recent studies further suggest that this de novo mechanism for gene origination may account for a significant proportion of new genes [2,14] and contribute to lineage-specific genetic novelties [18–20].
Currently, several comparative transcriptome studies have proposed that a proportion of de novo genes may originate from ancestral long non-coding RNAs (lncRNAs) [7,16,21], while the evolutionary mechanism underlying this lncRNA-protein transition remains elusive. First, it is unknown whether the differences in functional significance of lncRNAs, or some other sequence features, could explain the biased origination process of de novo genes from a specific subset of lncRNAs. Specifically, given that ancestral lncRNAs have precise splicing structures and tissue expression profiles similar to those of de novo proteins in human [16], it is unclear whether they have already obtained certain biological functions on the RNA level: one reason for us to hypothesize that functional non-coding genes may be favorable precursors is because they might survive longer during evolution, providing a wider time window for the emergence and stabilization of ORF, assuming that the emergence of the protein coding part does not interfere with the original function.
Second, it is unclear whether the human de novo genes have gained functional significance. Although it has been an established notion that de novo protein-coding genes could have important functions in Drosophila [17,18,22], the functional significance of de novo genes in hominoid lineage is still controversial–given the smaller effective population size in hominoids [23], the detection of these genes may be largely due to the weaker selection for removing the translational noises. Actually, for the dozens of human-specific de novo genes identified, only a few genes were linked to human diseases and regulations by circumstantial evidence [11,24–26]. While functional studies with transgenic monkeys could potentially characterize the functions of these hominoid-specific proteins, it is still technically challenging and could not provide a global view of the extent to which these genes are functional. Alternatively, a comparative population genetics approach, i.e. characterizing polymorphisms in the gene locus and comparing the pattern to that of the orthologous region in a closely related species, could provide evolutionary clues to the functional significance of the de novo genes.
We thus performed a population genetics study in human and rhesus macaque to interrogate the origin and functional significance of these newly-originated de novo protein-coding genes. We noted that these proteins in human seem to have originated from ancestral GC-rich lncRNAs. Although these lncRNAs generally are not more selectively constrained than other lncRNA loci, and the existence of these newly-originated proteins is not beyond anticipation under neutral expectation, our results showed that at least a proportion of these de novo proteins should have acquired protein-level functions, based on the signatures of purifying selection detected specifically in human populations. We thus propose a mechanism for creation of functional de novo proteins from ancestral lncRNAs during the primate evolution.
To interrogate the genesis and functional implications of de novo proteins in primates, we firstly performed a comprehensive survey for newly-originated de novo protein-coding genes in the hominoid lineage. We devised a genome-wide pipeline integrating ab initio identifications [16] and meta-analysis of public datasets [9–11,13,16] (Fig 1A; Materials and Methods). Briefly, we first inferred the locus ages on the basis of the syntenic genomic alignment generated by UCSC, and only retained human genes with high-quality alignments in the out-group species (Discussion). With this approach, the potential bias in de novo gene identification introduced by blast-like alignments is well controlled [27]. Then, for each locus, the existence of the ORF in multiple out-group species was inferred separately (Materials and Methods). Candidate de novo genes were then identified based on age assignments of ORFs, by summing up the information on the presence and absence of orthologous ORFs in vertebrate phylogeny with the principle of parsimony [2,16]. We further performed sequence similarity study to analyze these candidates against all annotated human proteins, further verifying that they originated through de novo evolution, rather than other mechanisms such as gene duplication (Materials and Methods).
The resulting 56 candidates, together with 99 literature-documented primate-specific de novo genes [9–11,13,16], were then subjected to additional inclusion criteria (Materials and Methods). Consequently, only genes with 1) reliable evidence for transcriptional and translational activities in human (S1 Fig; Materials and Methods), and 2) detectable common ancestral “disablers”, disrupting the ORFs in all out-group species at the same sequence position [9,11], as indication for newly-created but not old dying genes, were included (Fig 1A; Materials and Methods).
In total, 64 protein-coding genes were identified with recent origination in the hominoid lineage through de novo evolution (Fig 1B; Tables 1 and S1), with 43 encoding human-specific proteins (Class I, the younger proteins), and another 21 encoding similar proteins in human and chimpanzee but not in rhesus macaque (Class II, the older proteins).
The transcript structure and expression of these genes at the transcriptional or translational levels in human are strongly supported by public genomics data. The transcriptional structure for all of these genes were supported by full-length mRNA or spliced EST evidence (S2 Table), with 88% of the splicing junctions also supported by short RNA-Seq reads (Fig 1E; Materials and Methods); the full-length transcript structure for 17 of these genes were also verified by the Iso-Seq data, generated recently through the PacBio transcriptome sequencing (S2 Table; Materials and Methods). In addition, the protein expressions for all of these genes were supported by large-scale mass spectrometry studies in human (Tables 1 and S2; Materials and Methods).
To infer the transcriptional capacity of the 64 de novo genes in the common ancestor of human and closely related species, we performed cross-species transcriptome analysis in human, chimpanzee and rhesus macaque. First, we found that 83.9% of the 64 genes, and 92.9% of the 43 human-specific genes transcribed in at least one tissue in rhesus macaque or chimpanzee as lncRNAs (Fig 1C), with the expression levels significantly higher than the background expression levels (S1 Fig; Materials and Methods). Second, the expression levels of the genic regions relative to upstream and downstream regions were comparable among the three species (Fig 1D), and the majority of human splicing junctions were also detectable in chimpanzee and rhesus macaque orthologous regions (Fig 1E). Third, the non-coding orthologs of human de novo genes in rhesus macaque and chimpanzee also show tissue expression profiles similar to human (Fig 1F). The inter-species similarity of tissue expression profiles was further supported by clustering analysis, with the same tissue types from different species clustered together (S2 Fig and S3 Table).
By the parsimony principle, we conclude that the transcription structure and expression profile of these de novo genes had been shaped in the common ancestor prior to the acquisition of coding potential in the human lineage. It is thus interesting to investigate whether these lncRNA precursors with precise splicing structures and tissue expression profiles have already obtained certain biological functions on the RNA level, and may thus represent favorable precursors for new de novo proteins.
Because lncRNAs could have a variety of functions, not all of which can be easily assayed, as an alternative, we sought an evolutionary approach by quantifying the level of selective constraints in the orthologous lncRNA loci of these de novo genes in rhesus macaque as a proxy for determining the functional status in the ancestor. This assumes that the selective constraints on these loci have remained unchanged in the macaque lineage since it had a common ancestor with human.
We first compiled the whole lncRNAome in rhesus macaque using a similar strategy as described previously [31], on the basis of strand-specific RNA-Seq in ten tissues from the same macaque animal [16,32] (Fig 2A; Materials and Methods). A total of 5,641 lncRNA transcripts were assembled, verifying known features of transcripts such as the tight association with epigenetic markers and CpG islands [33,34] (Figs 2B and 2C and S3). Moreover, as positive control, we compiled a list of 89 non-coding genes in rhesus macaque. These non-coding genes are reportedly functional in human, as supported by experimental evidence [35]. Given the existence of the similar lncRNA transcripts in rhesus macaque, we assumed that these macaque lncRNA transcripts may also have functions and are under similar selective constraints as annotated non-coding genes in human (Materials and Methods).
To quantify the level of selective constraints, we then performed whole genome sequencing of 24 independent macaque animals and generated 23.7 billion paired-end reads with high quality, yielding high sequencing coverage of the macaque genome (ranging from 26- to 70-fold). On the basis of these sequencing data, as well as seven public datasets for macaque genomes [36–38], we profiled 54,079,575 single nucleotide polymorphic sites across the macaque genome, yielding a considerable number of polymorphisms located on the lncRNA loci (S4 Table; Materials and Methods).
On the basis of the polymorphism data in the population of 31 unrelated macaque animals, we measured the level of polymorphisms in the subset of macaque lncRNA loci, whose orthologous regions in human were de novo genes, and compared that to the same measures in all macaque lncRNAs, as well as the list of 89 established non-coding genes as a reference [35] (Materials and Methods).
As expected, we found that the list of 89 established non-coding genes are selectively constrained in rhesus macaque on the basis of the significantly decreased nucleotide diversity (π), compared with the synonymous sites of known macaque coding genes as a neutral control (Wilcoxon one tail test, p-value = 0.008, Fig 2D). The selection on these non-coding genes seems to be moderate, as compared with that of the non-synonymous sites of known macaque coding genes as a benchmark (Fig 2D). In contrast to this small repertoire of non-coding genes with functions, it seems that lncRNA transcripts are in general not selectively constrained in rhesus macaque, with the nucleotide diversity comparable with that of the synonymous sites across the macaque genome as a neutral control (Wilcoxon test, p-value = 0.917, Fig 2D). In addition, the orthologous loci for the lncRNA precursors of human de novo genes are not subjected to strong selective constraints as those non-coding genes, with the population genetics feature indistinguishable from that of the synonymous sites (Wilcoxon test, p-value = 0.570; Fig 2D), as well as that of the whole lncRNA pool (Wilcoxon test, p-value = 0.449; Fig 2D; Materials and Methods). In conclusion, we didn't find evidence for higher selective constraints for the orthologous lncRNA loci for human de novo gene precursors. Hence, it seems likely that the ancestor of de novo genes may not be particularly distinct in terms of functional importance before the proteins arise.
Given that the lncRNA precursors for human de novo genes did not display particularly distinct functional status, it is interesting to investigate whether other features, such as sequence features, may explain why de novo genes originate from some lncRNAs but not the others. In addition, given the smaller effective population size in hominoids, the detection of de novo proteins might arise from translational noise that is not acted upon or not yet removed by purifying selection, rather than being positively selected for due to their newly-acquired protein-level functions. We thus performed comprehensive sequence analysis of these de novo genes to investigate whether any sequence features could underlie the biased origination process of de novo genes from a subset of lncRNAs, and whether the existence of these de novo proteins is beyond anticipation in terms of their theoretical lifespan.
Interestingly, when analyzing the sequence features of these orthologous lncRNA precursors, we found that they have significantly higher GC contents in comparison with other lncRNAs and non-genic regions (Fig 3A and 3B; Wilcoxon one-tailed tests, p-value<1.0e-7 for both comparisons). The ORFs of de novo genes derived from GC-rich lncRNAs were also observed to have significantly higher GC content when compared with functional proteins in RefSeq (Fig 3B and 3C; Median of GC content = 0.57 vs. 0.53; Wilcoxon one-tailed test, p-value = 6.6e-6).
Further investigation revealed that such a GC-rich property endows these newly-originated ORFs with longer theoretical lifespan, even longer than their current age. As the stop codons are AT-rich, the higher GC content usually supports relatively stable ORF compared with GC-poor sequences [39,40] (S4 Fig). As expected, compared with RefSeq proteins, these newly originated ORFs have less content of fragile codons–codons convertible to stop codon by a single point-mutation, and are thus less susceptible to non-sense mutations (Fig 3D; Wilcoxon one-tailed test, p-value = 0.002). Accordingly, we found that these ORFs have long half-life time under neutrality (Materials and Methods), even significantly longer than other functional proteins in RefSeq (Fig 3E; Wilcoxon one-tailed test, p-value<2.2e-16). Especially, compared with the younger de novo genes, the older de novo genes have higher GC content (Fig 3C), less content of fragile codons (Fig 3D) and longer half-life time (Fig 3E). Overall, the theoretical lifespan of these newly-originated proteins is generally longer than their current age (Figs 1B and 3E), thus indicating that the existence of these de novo proteins is not beyond anticipation even under neutral expectation (Fig 3F).
Overall, the de novo gene repertoire we identified in the hominoid lineage actually represents a snapshot for the steady-state representation of a dynamic turnover process of ORFs. The detection of these GC-rich de novo proteins with stable ORFs, together with the previous reports that many de novo genes have stable expression profiles possibly by sharing the transcriptional context with nearby protein-coding genes through cis-natural antisense or bi-directional promoters [16,41], seems to favor the notion that a significant portion of the turnover is probably driven by genetic drift and those GC-rich "survivors" with long ORF lifespan and stable expression profile were retained and detected during a birth-and-death process.
What we found above suggest that the emergence and retention of de novo genes are likely under neutral forces. However, considering these GC-rich “survivors” have been exposed to natural selection for relatively long time due to their theoretically longer lifespan, it is interesting to investigate whether some of these newly originated ORFs have been maintained by selective constraint in the current population, due to their newly-acquired protein-level functions. We thus performed population genetics study in human and rhesus macaque populations to assess whether selective constraints are applied to these ORF regions in human populations but not their non-coding counterparts in rhesus macaque.
We first profiled a set of polymorphism sites in human populations, by re-analyzing whole genome sequencing data in 67 individuals from different sub-populations (Fig 4A and S5 Table; Materials and Methods). We expect that if the de novo genes encode functional proteins and are maintained by purifying selection, the polymorphism level for exonic regions of these genes will be lower than intronic regions. The polymorphism level for non-synonymous sites should also be significantly lower than that of synonymous sites, as the former will be under much stronger selection. Moreover, we expect to find a difference in the frequency spectra at nonsynonymous vs. synonymous sites, resulting in a skew towards low frequency variants compared to the latter. These are indeed what we found: 1) The θw and π measures were significantly lower in the exonic region of the de novo genes compared to the intronic region of the same locus (Monte Carlo p-values<1e-4; Figs 5A and S5; Materials and Methods). In addition, the UTR regions of these de novo genes showed θw and π measures that are lower than the intronic regions, while slightly higher than CDS regions (Figs 5A and S5). 2) Compared with synonymous sites, the nucleotide diversity for non-synonymous sites was significantly lower (Wilcoxon one-tail test, p-value = 0.019; S6 Fig). Accordingly, the ratio of the nucleotide diversity for non-synonymous sites to synonymous sites was generally smaller than 1 (Fig 5B). 3) The frequency spectrum of the derived alleles had an excess of low-frequency variants at the non-synonymous sites in the de novo genes compared to that at the synonymous sites (Fig 5C), which is similar to known protein-coding genes (Fig 5E). As a control, we classified mutations in human lncRNAs into synonymous or non-synonymous sites within the longest pseudo-ORFs, and didn’t observe any difference in their respective frequency spectrum (Fig 5F; Materials and Methods).
Accordingly, as a negative control, we performed population genetics study on macaque orthologous regions (without coding potential) of human de novo genes, in a population of 82 unrelated rhesus macaque animals (Materials and Methods). Custom library with >135,000 120-bp DNA oligos were designed to capture the macaque orthologous regions. Ultra-deep sequencing was then performed (Materials and Methods) and 222 million 150-bp paired-end reads were generated and uniquely located on the macaque genome (NCBI SRA accession number: SRP052932; S6 Table). The average effective coverage of the targeting regions reached to 94% in each sample (Figs 4B and 4C and S7), and a total of 10,162 single nucleotide polymorphisms were identified across the target genomic regions with high sensitivity and specificity, as verified by a follow-up whole genome sequencing with 30× coverage in one of these macaque animals (Fig 4D and 4E; Materials and Methods).
On the basis of the polymorphism data of the macaque orthologous regions, we found that both the θw and π measures were uniform across the length of these regions, in contrast to the clear differences observed in human (Figs 5A and S5). We further classified the polymorphism sites on macaque lncRNAs into synonymous or non-synonymous sites, according to codon-level alignments between human de novo proteins and their orthologous lncRNAs in rhesus macaque. No significant difference was detected for the nucleotide diversity of pseudo-non-synonymous and pseudo-synonymous sites in rhesus macaque (Wilcoxon one-tail test, p-value = 0.607; S6 Fig), with the ratio of the nucleotide diversity between these two groups comparable to 1 (Fig 5B). In addition, the resulting frequency spectrum of derived alleles at the pseudo-non-synonymous sites is indistinguishable from that at the pseudo-synonymous sites (Fig 5D). The population genetics analyses thus suggest that these newly-originated de novo genes have gained new functions specifically in human.
Taken together, although the de novo proteins seem to emerge from lncRNA precursors with no bias towards those functionally-constrained lncRNAs, and their existence is not beyond the anticipation under neutral expectation, at least a proportion of these proteins should have acquired protein-level functions specifically in human, as revealed by the species-specific signatures of purifying selection on these newly-originated de novo genes. We thus depicted a new mechanism for the origination of functional proteins from ancestral non-coding transcripts with precise splice structures and specific tissue expression profiles during the primate evolution.
Although our current study identified a list of 64 hominoid-specific de novo genes, more proteins are expected to originate through this de novo mechanism [42]. Several types of de novo genes might be underrepresented due to the computational pipelines currently used to identify these genes: 1) Genes with shorter ORF. Automatic gene annotation pipeline typically neglect genes with short ORFs by using arbitrary criteria to define the minimal ORF length. Consequently, considering de novo proteins generally have short ORFs [43–45], a large proportion of de novo proteins with short ORFs were removed. 2) Genes without orthologous regions in out-group species were not included, due to the requirements of the high-quality alignments for accurate age assignments of ORFs in vertebrate phylogeny, as well as of the detection of common ancestral “disablers” as indication for newly-created rather than old dying genes. Although such a design effectively lowered the false-positives and potential bias introduced by blast-like alignments in de novo gene identification [27], some false-negatives could still result from our stringent criteria. 3) Genes with non-stable accession numbers. Considering the difficulties in defining the coding potential, the Ensembl accession numbers assigned to de novo genes are typically not stable. Although our study has combined multiple versions of Ensembl databases to identify de novo genes, some genes may still be overlooked. 4) Genes with low expression. De novo genes typically express in low abundance [13,16,18]. As reliable evidence for transcriptional and translational expression is needed to define a de novo gene, especially considering the relatively low sensitivity of mass spectrometry technology to reliably detect peptides of low abundance, these genes may be missed. 5) Additionally, the current identification pipeline by comparative genomics approaches typically focuses on in-group ORFs that are missed in other out-group species. In such an occasion, a considerable proportion of de novo genes originated through lineage-specific expression of pre-existing ORFs might be neglected [17,18].
Overall, although these 64 genes may not fully recapitulate the true repertoire of de novo genes in the hominoid lineage, they should constitute a representative group for further analysis and elucidation of the evolving process of de novo genes from precursor lncRNAs.
Although de novo genes are also regarded as “motherless” genes due to their lack of ancestral protein-coding genes as precursors, we and others have found that at least a proportion of lncRNAs might represent an intermediate stage of their origination, narrowing the gap between non-coding DNA and protein-coding genes. Such an origination process may take advantage of existed genomic contexts. For example, these lncRNA “precursors” usually share the transcriptional context with the nearby protein-coding genes through cis-natural antisense or bi-directional promoters [16,41]. These lncRNAs with stable expression profiles, although not more selectively constrained according to our population genetics study, may then lay the foundation for the emergence of new de novo genes. In addition, the GC-rich sequence property of these lncRNAs further supports stable ORFs of the newly-originated proteins (Fig 3). Overall, these genomic features provide a theoretically favorable foundation for the birth of some functional proteins–a notion well supported by our population genetics data, which revealed that some of these loci already encode human-specific functional proteins (Fig 5).
Although such an origination process is plausible, currently both the definitions of de novo genes and lncRNAs are depending on some arbitrary criteria [9,13,31]. Additional lines of evidence are thus needed to fully support the mechanism through which de novo genes come from the lncRNA pools during the primate evolution. For example, since it is still technically challenging to fully annotate proteomes based on mass-spec studies across tissues, development stages and species, it is inadequate to directly identify human de novo genes on the basis of the presence or absence of peptides across different species. Alternatively, conceptual translation of ORFs between species is still the main strategy in the field to infer the existence of these de novo proteins in different out-group species [9,13]. In this context, although the orthologs of these human de novo genes could be defined as “lncRNAs” in chimpanzee and rhesus macaque by the current criteria, they may actually encode smaller version of these de novo proteins in out-group species.
We thus performed cross-species analyses to test this “functional ORF expansion model”. Briefly, if the functional proteins were indeed absent from out-group species, we would expect similar substitution rates between non-synonymous and synonymous sites when performing comparative genomics analysis. When aligning the truncated forms of the human de novo proteins in non-human primates, we found that the merged dN/dS ratio does not deviate significantly from 1 (dN/dS = 0.90). In line with this finding, our population genetics study on macaque orthologous regions of human de novo genes also indicated that these macaque orthologs may not encode similar functional proteins as in human (Fig 5). Even considering these population genetics evidence, we still could not fully exclude the possibility that some of these so-called “lncRNAs” might actually encode fast-evolving or smaller version of the protein in out-group species. As such, these proteins might be under weak selection, and the signals for selective constraints could not be detected based on the population size of this study. Future mass-spec studies with high sensitivity may aid in clarifying these issues.
Protein-coding genes have typically higher GC content than non-coding regions. It has been proposed that the increased GC content in genic regions could be maintained by natural selection, such as the GC preference on the wobble sites potentially shaped by adaptive evolution for the stability of mRNA secondary structure or the efficient protein synthesis [46].
However, here we found that other gene-associated genomic regions are also GC-rich, such as the intronic regions (Fig 3B). Theoretically, considering the models for new gene origination, each protein-coding gene could be traced to an ancient origination event from non-coding DNA. There is thus a formal but as yet unexplored possibility that the biased inheritance from GC-rich lncRNAs could be another major factor underpinning the different extents of GC content between coding regions and genomic background. As we provided an evolutionary and functional connection between protein-coding genes and non-coding DNA regions in the hominoid lineage, we formally tested this “GC-rich inheritance model”. Correspondingly, we found that GC-rich lncRNAs are favorable precursors for new proteins. More importantly, the GC-rich features could be detected in all genomic regions associated with these newly originated de novo genes, even for the wobble sites with established GC preference, as well as their lncRNA precursors, which also resembled those of the well-known protein-coding genes (Figs 3B and S8). Besides being a consequence of adaptive evolution after the acquirements of the ORFs, the GC-rich feature of the protein-coding genes may also inherit from the ancestor lncRNAs, thus complementing previous theory on GC-rich feature for protein-coding genes [46–49].
Rhesus macaque samples were obtained and manipulated from the internationally-accredited (Association for Assessment and Accreditation of Laboratory Animal Care, AAALAC) animal facility of the Institute of Molecular Medicine in Peking University. The present study was approved by the Institutional Animal Care and Use Committee of Peking University.
De novo protein-coding genes in the hominoid lineage were identified using a genome-wide pipeline integrating ab initio identifications and meta-analysis of public datasets. On the basis of Ensembl gene annotations (v68), de novo genes were identified using a similar pipeline as we published previously [16]. Briefly, 1) we inferred the locus ages on the basis of the syntenic genomic alignment generated by UCSC, and only human genes assigned with specific locus age were retained; 2) for locus with high-quality alignment (coverage >70% and identity >50%) in the out-group species, the existence of the ORF in multiple out-groups (chimpanzee, orangutan, rhesus macaque, mouse, guinea pig, dog, hedgehog and armadillo) was inferred separately by Exonerate [50], and if the sequence in particular out-group encoded at least one frame-disrupting indel or premature stop codon, with the subsequent maximum continuous ORF shorter than 70% of the human ORF length (a cutoff based on the previous practice in this field [9,13,16]), the ORF was regarded as non-existent in this out-group; 3) we inferred the origination timing of ORFs for these de novo genes by summing up information of their presence or absence in multiple out-group species, along with the phylogenetic tree with the principle of parsimony, and subsequently retained only genes originated in the hominoid lineage; 4) sequence alignments were then performed against all human proteins (BLAST e-value cutoff of 10−6) to ensure these new genes originated through de novo evolution other than gene duplications. Finally, 56 protein-coding genes were identified as candidate de novo genes in the hominoid lineage.
The resulted 56 candidate genes, together with 99 literature-documenting primate-specific de novo genes [9–11,13,16], were then subject to two additional inclusion criteria. First, genome-wide expression filters were introduced to ensure these genes had convincing evidence for transcriptional and translational expression in human. Public RNA-Seq data in 17 human tissues (Human BodyMap 2.0 data from Illumina and data from references [51,52]) were integrated and analyzed to estimate the gene expression level of each gene, according to a standardized pipeline [53]. To distinguish true transcription signals from the background expression, we first estimated the RPKM values for the genomic background represented by 10,000 randomly-selected intergenic regions. The expression levels of intergenic regions were significantly lower than 0.2 RPKM in all tissues (S1A Fig; Monte Carlo p-values ranging from 0.002 to 0.028). Therefore, a more conservative PRKM cutoff of 0.5 was arbitrarily set to confirm the transcriptional expression of these de novo genes in human. Two candidates (ENSG00000205056 and ENSG00000198547) with low RPKM scores were also included due to their reliable experimental evidence for transcriptional expression [9,11]. Peptide evidences from large-scale mass spectrometry studies were then extracted from PRIDE [28], PeptideAtlas [54], ProteomicsDB [55] and Human Proteome Map [29]. A peptide was considered to support the protein expression of a de novo gene only if 1) when performing BLAT similarity searches against all human proteins (Ensembl v68, BLAT settings-t = prot-q = prot-stepSize = 5), its whole sequence exactly match the CDS region of the de novo gene, with the second-best hit in the proteome (if existing) including at least one mismatch; and 2) when performing BLAT similarity searches against the human genome, its whole sequence identically and exclusively match the CDS region of the de novo gene (hg19, BLAT settings-stepSize = 5-stepSize = 5-t = dnax-q = prot). Only genes with 1) RNA-Seq RPKM >0.5 in at least one of the 17 human tissues, and 2) at least one convincing item of peptide evidence in support, were retained (Fig 1A). Second, to verify that these genes are newly-originated rather than old dying genes, we manually checked the corresponding ORF regions in multiple out-group species (chimpanzee, orangutan, rhesus macaque, mouse, guinea pig, dog, hedgehog and armadillo), and only genes with common ancestral disablers shared by multiple out-group species were retained (Fig 1A). Here, a common ancestral disabler refers to a mutation disrupting the ORF in multiple out-group species at the same sequence position [9,11]. In such scenario, the mutation is more likely to be of an ancestral status according to the parsimony principle, thus indicating the gene is newly-originated rather than old dying. Totally, a list of 64 genes was identified to originate recently in the hominoid lineage through de novo evolution (Fig 1B and Tables 1, S1 and S2).
We also studied the characteristics of these de novo genes across primate species in the context of new genomics technologies. According to computational pipelines described previously [9,13,16], mRNA and EST data from UCSC Genome Browser, RNA-Seq data archived in RhesusBase [53,56], as well as single-molecule long-read sequencing data on human transcriptome [57] were downloaded and analyzed to investigate the transcriptional structure of these de novo genes in human. On the basis of public RNA-Seq data in human, chimpanzee and rhesus macaque [16,51,52], comparative transcriptome studies were then performed to compare the transcription level, splicing structure and tissue expression profiles of these de novo genes with their non-coding orthologs in chimpanzee and rhesus macaque, according to a pipeline previously described by us [16]. Specially, an RPKM cutoff of 0.2 was set to distinguish convincing transcription and transcriptional noise as described above (S1B and S1C Fig).
Strand-specific, Poly(A)-positive RNA-Seq data in ten tissues (adipose, prefrontal cortex, cerebellum, heart, kidney, liver, lung, muscle, spleen, testis) of the same macaque animal were used to assemble the lncRNAome in rhesus macaque [16,32], following a computational pipeline as described previously [31]. Briefly, RNA-Seq reads of each macaque tissue were aligned separately to the macaque genome (rheMac2) with Tophat (v2.0.6) [58]. Transcriptome assembly was then performed with both Cufflinks (v2.0.2) and Scripture (VPaperR3) [59,60], and redundant transcripts were merged with Cuffcompare (v2.0.2) [60] after boundary correction. To control for false-positives, only long, multi-exonic transcripts (>200 bp) with supportive evidences in ≥2 tissues or by both assemblers were retained [31,61]. To evaluate the performance of this transcriptome assembly, Cuffcompare (v2.0.2) was also introduced to compare the assembled transcripts with multi-exonic protein-coding genes as annotated in RefSeq. Finally, a total of 90,322 multi-exonic transcripts were assembled, which represents transcript structures reconstructing for 95% known multi-exonic protein-coding genes, suggesting the feasibility of this assembly strategy (S3 Fig).
Several stringent criteria, such as proteome annotation- and comparative genomics-based filtering procedures, were incorporated to exclude protein-coding transcripts. Briefly, 1) transcripts with ≥1 splice junction overlapped with known protein-coding genes annotated in either Ensembl or RefSeq were discarded; 2) PhyloCSF was applied to score the coding potential of these candidates (multiple sequence alignments of 9 mammalian genomes,—frames = 3—orf = StopStop3) [62] and only transcripts with PhyloCSFscore <65 were retained, corresponding to a false negative rate of 1% and a false positive rate of 5% on the basis of RefSeq annotation. 3) the nucleotide sequences or 3-frame stop-to-stop translation products were subjected to Blastx, Blastp and HMMER searching against all human proteins or known protein domains (Pfam-A, Pfam-B) [63], and transcripts with significant hits (e-value ≤10−4) were discarded; 4) transcripts with putative ORFs ≥100aa longer were also discarded. The strategy had a good performance in distinguishing lncRNAs from protein-coding transcripts (S3 Fig) and a total of 5,641 lncRNA transcripts were assembled. On the other hand, non-coding genes in human were searched and downloaded from lncRNAdb (http://www.lncrnadb.org/) [35] and the macaque orthologs of these functional human lncRNAs were then retrieved by liftOver.
For human de novo genes and their orthologs in chimpanzee and rhesus macaque, sequences of different genomic regions were retrieved and the GC contents were calculated by a customized Perl script (https://github.com/Jia-Yu-Chen). Following a previous study [40], we also calculated for each de novo gene the proportion of fragile codons that could become stop codons by single mutation. We further investigated whether the existence of these de novo proteins is beyond anticipation in terms of their theoretical lifespan. Briefly, an ORF would eventually be interrupted if it does not experience any functional constraint, and the rate of ORF interruption under neutrality is largely determined by point mutation rate and insertion/deletion mutation rate. We thus estimated the interruption rate in terms of half-life time of ORF (t1/2) according to the computational simulation method developed by Zhang and Webb [64] (Fig 3E). The half-life time of a given ORF is the time required for an ORF to be interrupted in one-half of 20,000 simulation replicates, with the rates of point and insertion/deletion mutations being set as 1.25 and 0.1 per site per billion years, respectively, as previously defined [64]. Given this half-life time (t1/2), the minimal probability of an ORF remaining intact today under neutrality was then determined by the following equation, by assuming that the lineage-specific gene emerged right after the species divergence from the most recent common ancestor (T) (Fig 3F).
We firstly profiled the polymorphism data in human populations, by re-analyzing whole genome sequencing data in 67 individuals from different sub-populations and archived with high sequencing coverage by the 1000 Genomes Project (S5 Table). Briefly, deep sequencing reads were mapped to the human genome (hg19) using BWA [65], and the polymorphism sites of each sample were identified and evaluated according to the standard GATK pipeline with UnifiedGenotyper (V2.7–4) [66]. After stringent filtering strategies to remove false-positives in variant calling, 18,186,523 highly reliable single nucleotide variants were identified across the human genome, with 85.4% supported also by the 1000 Genomes Project (Fig 4A).
Accordingly, we profiled the distributions of polymorphic sites in rhesus macaque populations as a reference. For each human de novo gene, we performed targeted capture and ultra-deep sequencing of the macaque orthologous regions (and 1kb flanking regions) in a population of 82 unrelated male animals (S6 Table). Briefly, custom library with >135,000 120-bp DNA oligo probes were designed by Agilent SureSelect XT Target Enrichment System (Agilent Technologies, Inc., Santa Clara, USA), with a 3-folds tiling coverage, to capture the targeted regions in rhesus macaque. Genomic DNA from the macaque animals was isolated from 200–500μl whole blood using the QIAamp DNA Blood Mini Kit (Qiagen, Venlo, Netherlands) and 3μg DNA of each animal was sheared to fragments with a peak at 150–200bp using Covaris S220. Then, the adaptor-ligated libraries were amplified, purified and hybridized with SureSelect Capture Library according to the manufacturer’s instructions. After 16h hybridization at 65°C, the captured targets were pulled down by Dynabeads MyOne Streptavidine T1 (Life Technologies, Ltd., Carlsbad, USA) and amplified for the library preparation, which were then sequenced on Illumina Miseq system with 151-bp paired-end read mode. Totally 222 million 150-bp paired-end reads were generated and uniquely located on the macaque genome (S6 Table). The average effective coverage of the targeted regions reached to 94% in each sample (Fig 4B), and 86% of whole orthologous region (or 95% of CDS regions) were sequenced with coverage of ≥30 in all of the 82 macaque animals (Figs 4C and S6). A total of 10,162 highly-reliable single nucleotide polymorphisms were then identified, according to the standard GATK pipeline with UnifiedGenotyper (V2.7–4) [66].
To evaluate whether allele dropout or other false-positives introduced by target capture may compromise our approach, we further performed whole genome sequencing with 30× coverage in one of these macaque animals (Animal ID: 920653) for evaluation (Fig 4B). Genomic DNA was obtained for the library preparation of whole genome re-sequencing, deep sequencing was performed on a HiSeq 2000 Sequencing System with a 151×2 paired-end read mode, and single nucleotide polymorphisms were identified according to the standard GATK pipeline. Finally, 96.2% polymorphic sites identified in the targeted sequencing were verified by the whole genome sequencing, with 99.5% showing the same genotype (Fig 4D and 4E).
Whole genome sequencing data from 24 macaque animals generated previously in our lab, as well as seven animals published previously [36–38], were also analyzed to profile a genome-wide polymorphism dataset across the macaque genome, according to the pipeline as described above. All deep sequencing data in this study are available at NCBI SRA under accession numbers SRP052932.
On the basis of the polymorphism data from the population of 31 macaque animals, we measured the nucleotide diversity (π) for the orthologous lncRNA loci of human de novo genes, all lncRNAs, and a list of 89 functional non-coding genes in rhesus macaque. Non-synonymous and synonymous sites of macaque protein-coding genes as annotated by RefSeq were used as benchmarks for the extent of the selective constraints (Fig 2D). Wilcoxon test was performed to test whether the nucleotide diversity between two groups are significantly different, with a p-value cutoff of 0.05 (Fig 2D).
On the basis of the polymorphism data obtained by analyzing the whole genome sequencing data of 67 human individuals and the targeted sequencing data of 82 macaque animals, we estimated the polymorphism levels (θw and π) for different genomic regions (exon, intron, CDS and UTR) of the de novo genes in human and their orthologs in rhesus macaque (Figs 5A and S5). We further performed statistical tests to determine whether the different polymorphism levels between exonic and intronic regions of human de novo genes are statistically significant, with the background estimated by 10,000 times of Monte Carlo simulations, assuming the polymorphic sites were randomly distributed in exonic and intronic regions of human de novo genes. Considering that the average nucleotide diversity in rhesus macaque is higher [67], if the exonic regions are more selectively constrained than intronic regions, we should have greater statistical power to detect the difference. The observation of a comparable nucleotide diversity between macaque exonic and intronic regions then indicates that these macaque orthologs of human de novo genes may not encode similar functional proteins as in human (Figs 5A and S5).
The ratio of the nucleotide diversity between non-synonymous sites to synonymous sites was also determined for each de novo gene, as well as its non-coding ortholog in rhesus macaque (Fig 5B), in which the pseudo-non-synonymous and pseudo-synonymous sites in macaque orthologs were determined by codon-level alignment with human de novo proteins. For each polymorphic site, the derived allele was defined by the EPO pipeline [68,69]. The frequency spectra of derived alleles were then estimated, with 1,000 times of bootstrap performed to estimate the confidence intervals of the proportions of polymorphism sites (Fig 5C–5F). Similar analyses were performed for known protein-coding genes as annotated by RefSeq, as well as human lncRNAs as annotated by GENCODE (v19) as controls.
|
10.1371/journal.pbio.1000341 | Reinforcement of Gametic Isolation in Drosophila | Reinforcement, a process by which natural selection increases reproductive isolation between populations, has been suggested to be an important force in the formation of new species. However, all existing cases of reinforcement involve an increase in mate discrimination between species. Here, I report the first case of reinforcement of postmating prezygotic isolation (i.e., barriers that act after mating but before fertilization) in animals. On the slopes of the African island of São Tomé, Drosophila yakuba and its endemic sister species D. santomea hybridize within a well-demarcated hybrid zone. I find that D. yakuba females from within this zone, but not from outside it, show an increase in gametic isolation from males of D. santomea, an apparent result of natural selection acting to reduce maladaptive hybridization between species. To determine whether such a barrier could evolve under laboratory conditions, I exposed D. yakuba lines derived from allopatric populations to experimental sympatry with D. santomea, and found that both behavioral and gametic isolation become stronger after only four generations. Reinforcement thus appears to be the best explanation for the heightened gametic isolation seen in sympatry. This appears to be the first example in animals in which natural selection has promoted the evolution of stronger interspecific genetic barriers that act after mating but before fertilization. This suggests that many other genetic barriers between species have been increased by natural selection but have been overlooked because they are difficult to study.
| What stops newly formed species from interbreeding? Answering this question is fundamental to our understanding of speciation. One mechanism is that where such would-be species meet, the barriers against interbreeding are reinforced by natural selection (e.g., leading to greater mate discrimination). On the slopes of the African island of São Tomé, Drosophila yakuba and its sister species D. santomea hybridize within a well-demarcated hybrid zone. I found that D. yakuba females from within this zone, but not from outside it, show an increase in gametic isolation from males of D. santomea, such that before fertilization, the females deplete sperm from D. santomea males faster than from conspecific males. Consequently, there are fewer progeny produced from interspecific matings. By experimentally evolving the populations, I also show that such postmating isolation can rapidly evolve. Natural selection, therefore, has promoted the evolution of stronger interspecific genetic barriers that act after mating but before fertilization. D. santomea and D. yakuba, then, appear to represent an example of reinforcement for a postmating-prezygotic trait in an organism that has internal fertilization. This work shows that reinforcement of barriers other than sexual and other forms of premating isolation is possible. This also suggests that there are many “cryptic” barriers to gene flow that might be increased by natural selection in areas where species overlap and hybridize.
| The evolutionary process of “reinforcement,” often suggested as an important component of speciation, involves the strengthening by natural selection of prezygotic isolation between closely related taxa in response to maladaptive hybridization [1]–[4]. Reinforcement has often been inferred from a pattern of “reproductive character displacement,” in which individuals of different species are more behaviorally isolated if they come from the area where two species overlap (sympatric) than from areas outside each other's range (allopatric; [1]–[8]). Reinforcing selection, however, need not be limited to increasing premating isolation: other reproductive barriers that act after mating, such as gametic isolation, can also be reinforced [9]–[14]. Lorch and Servedio [15], for example, proposed that a species preference for fertilizing the gametes of conspecific versus heterospecific individuals could evolve through a reinforcement-like process, depending on the nature of selection against heterospecific matings. Here, I report the first, to my knowledge, apparent case of reinforcement in the wild of a genetic barrier—reduced production of hybrid eggs—that acts after mating but before fertilization; and I also demonstrate that the evolution of this form of gametic isolation can occur in the laboratory.
I looked for evidence of reinforcement in postmating-prezygotic isolating mechanisms in two African species of Drosophila in the melanogaster subgroup: D. yakuba and D. santomea. D. yakuba is widespread throughout sub-Saharan Africa and has extended its range to neighboring islands, including the Gulf of Guinea islands in the eastern Atlantic Ocean [16]. D. santomea, the closest relative of D. yakuba, is endemic to São Tomé, a small (860 km2) volcanic island 255 km west of Gabón. Molecular data show that D. yakuba and D. santomea diverged about 400,000 y ago [17],[18]. On the extinct volcano of Pico de São Tomé, D. yakuba occurs at elevations below 1,450 m, and is also common in the lowlands, villages, and plantations. In contrast, D. santomea occupies the mist forests at elevations between 1,153 and 1,800 m [16]–[19]. These species are unique within Drosophila in showing a well-demarcated hybrid zone.
Previous studies uncovered at least 11 distinct reproductive barriers that act over the entire life cycle, ranging from habitat isolation to hybrid dysfunction, although no single barrier completely impedes gene flow [20]–[24]. Five known barriers are of the postmating-prezygotic form [22],[24], including both competitive (conspecific sperm precedence [CSP]) and noncompetitive mechanisms (lower production of eggs after heterospecific matings). The yakuba–santomea species pair is ideal for studying reinforcement because it meets the requirements that 1) mating and introgression occur between the species in nature (as observed in the hybrid zone between yakuba and santomea) and 2) that hybridization be costly (all male hybrids are sterile). Previous studies of these species have failed to find evidence of reinforcement in premating barriers [20], but there was no search for reinforcement in postmating-prezygotic barriers.
Here, I report that reinforcement for a form of postmating-prezygotic isolation—gametic isolation—has apparently evolved in natural populations of D. yakuba sympatric with the sister species D. santomea. I demonstrate a clear fitness advantage for those individuals who have increased gametic isolation, and this advantage apparently leads to a remarkably rapid evolution of gametic isolation in laboratory populations. This appears to be the first known example in animals of the evolutionary increase of interspecific genetic barriers that act after mating but before fertilization.
Figure 1 and Table S1 describe the collection sites of the stocks used in this study. I used isofemale lines to study the among-line component of variation in gametic isolation. (The among-population variation was not evaluated because it is not possible to sample multiple populations from the area of sympatry.) Females of each line from each species were mated to both conspecific and heterospecific males to estimate egg number produced by each type of cross. I collected virgin males and females under CO2 anesthesia and kept them for 3 d in single-sex groups of 20 flies. On day 4, I aspirated flies into fresh food-containing vials, with one female and one male per vial. All copulations were watched to ensure that they were not abnormally short. To prevent females from remating, males were removed from a vial by aspiration after mating. After 1 h, I ended the observations and discarded females who did not mate. Each mated female was allowed to oviposit for 24 h in a vial, after which I counted the total number of eggs laid and transferred the female to a fresh vial. The counting was repeated daily for 10 d. In interspecific crosses, the reduced number of offspring constitutes a noncompetitive form of gametic reproductive isolation, as each female carries sperm from only one male [22],[24]. Twelve females were scored for each cross.
One way to measure the efficiency of sperm storage or survival is to estimate the proportion of eggs laid every day that hatch, following the decline in this statistic over time [24]. To this end, I used six D. yakuba lines (three allopatric and three sympatric) and three D. santomea lines, measuring the decline of egg hatchability for all the possible D. yakuba × D. santomea crosses. For each cross, I produced 100 inseminated females, divided into five subgroups of 20 females. Each subgroup was transferred without anesthesia to colored medium. Eggs were collected every 24 h, and the hatchability of each batch was measured daily for 10 d.
Heterogeneity in hatchability among crosses was analyzed by fitting a minimal random linear mixed model (LMM) [25] to the hatchability of eggs laid each day. I analyzed five main effects—geographic origin of female (sympatric vs. allopatric populations), geographic origin of male, female line nested within geographic origin, male line nested within geographic origin, and days after mating—as well as all interactions between these factors. The effect due to differences between groups of females was considered random. I analyzed the data following the maximum-likelihood model simplification approach of Crawley [26],[27], in which the full model containing all factors and interactions was fitted and then simplified by a series of stepwise deletions, starting with the highest-order interaction and progressing to lower-order interaction terms and then to main effects. The critical probabilities for retaining factors and determining whether effects or interactions were significant were 5% for main effects, 1% for two-way interactions, and 0.5% for three-way interactions [28]. To assess whether slopes (i.e., the rate at which hatchability decays along time) differ between matings, I formulated two models that differed in the assumptions about these slopes and compared the models using a likelihood ratio test (LRT). Model 1 was a full factorial analysis (i.e., different slopes and intercepts for each of the possible crosses), whereas Model 2 assumed different intercepts but identical slopes (rate of decline of fecundity). Finally, to determine whether different treatments produced differences in initial hatchability (as a proxy for the amount of sperm transferred during heterospecific matings), I analyzed hatchability data from the first day using a one-way ANOVA, with hatchability as the response and two fixed effects (origin of the female, and female line nested within origin).
To study whether an initial interspecific mating had any effects on the fertility of D. yakuba females after a second conspecific mating, I scored the egg production of heterospecifically mated females. After 4 d, interspecifically mated females were remated to D. yakuba males (from the same line than the female), and I scored the number of eggs laid during the subsequent 10 d. Eggs were counted every 24 h over the entire 14-d period. This analysis used 12 lines (six allopatric and six sympatric), with 25 individuals scored per line.
I analyzed differences in overall fecundity between the different crosses by fitting a nested ANOVA to the total number of eggs laid per female (the sum of heterospecific and conspecific eggs), with geographic origin of the female and female line (nested within origin) as fixed effects and variation among females within line as a random effect. To determine whether the proportion of conspecific eggs (relative to the total number of eggs) differed between sympatric and allopatric lines, I followed the same procedure used to analyze total fertility, but fitted the model to the number of conspecific eggs laid (i.e., eggs laid after the second mating).
To test whether natural selection on gametic isolation could have been responsible for the observed reproductive character displacement in natural populations, I kept seven populations of D. yakuba, originally derived from allopatric populations (Table S2), in experimental sympatry with D. santomea for ten generations, following the design of Koopman [29] and Higgie et al. [30]. These conditions were created by maintaining four bottles per population, with each bottle containing 50 D. yakuba females, 50 D. yakuba males, 50 D. santomea females, and 50 D. santomea males. Since D. yakuba always outcompetes D. santomea under these conditions [24], I added D. santomea females and males to the experimental sympatry bottles each generation to maintain a constant ratio of the two species. To set up each successive generation, I collected 50 flies of each sex of D. yakuba (easily identifiable by pigmentation) as virgins from the experimental bottles and transferred them to a new bottle. To reconstitute the sympatry conditions, 50 D. santomea flies of each sex (collected as virgins from stock bottles) were added to the bottle. This procedure was followed for ten generations. Control populations of D. yakuba were maintained for each population (four replicates) at the same density (100 flies per bottle) but without adding D. santomea. The maintenance conditions and population size of D. yakuba were the same between experimental sympatry bottles and control groups. The strength of sexual and gametic isolation was measured every two generations using methods described previously [22],[31],[32].
Finally, I set up an internal control to make sure that elimination of hybrids was complete, i.e., there was no gene flow between the two species in the experimental bottles. Taking into account the complete sterility of F1 hybrid males, who lack motile sperm [17],[21], I collected D. yakuba females from each experimental sympatry bottle and mated them to D. santomea males to produce 100 F1 heterospecific males (♀ D. yakuba × ♂ D. santomea) every other generation. These F1 males were scored for sperm motility. The idea behind this test was that if motile sperm were seen, it meant that there had been gene flow between species (i.e., not all the hybrids were killed when setting up a new generation), and the bottle was discarded.
The results from this experiment were analyzed using a paired t-test to compare the values of gametic and sexual isolation (transformed with arcsine) between experimental populations that were exposed to D. santomea and the unexposed control populations.
The test for reinforcement of gametic isolation involved mating D. yakuba and D. santomea females from either sympatric or allopatric populations to males of the other species and scoring the number of eggs produced by a single female—an index of the strength of noncompetitive gametic isolation—during the first 10 d after each mating (Figure 2). In D. santomea, I detected no heterogeneity in egg production when females were mated to D. yakuba males (gametic isolation) among lines (LMM, F1,13 = 1.644, p = 0.222, Figure 2A). In contrast, D. yakuba females from sympatric lines yield significantly fewer progeny than those from allopatric females when both were mated to D. santomea males, even when allopatric females were derived from populations close to the hybrid zone on São Tomé. D. yakuba females, therefore, show the pattern predicted by reinforcement of gametic isolation (LMM, F1,20 = 42.56, p<0.0001, Figure 2B). This suggests that in D. yakuba, increased gametic isolation has evolved as a response to the sympatric presence of the sister species. The results with synthetic lines (genetically heterogeneous strains of each species created by combining virgin males and females from several isofemale lines from the same location) were similar (unpublished data).
To determine whether the reduced number of hybrid eggs laid by D. yakuba females from sympatric populations was caused by a female trait, a male trait or the interaction of both, I randomly selected six D. yakuba lines (three allopatric and three sympatric) and six D. santomea lines, mated the D. yakuba females to D. santomea males in all the possible combinations, and performed the egg-counting protocol described above. The data were analyzed with a LMM with four fixed effects: female origin, female line (nested within female origin), and male line (nested within male origin), as well as all interactions between these factors. The minimal linear model for this design showed that there is a high degree of heterogeneity (F35,396 = 18.08, p<10–15) in the number of eggs produced. The results indicate that the among-line heterogeneity is explained by origin of the female (whether a population was allopatric or sympatric to D. santomea in the field, LMM, F1,4 = 124.818, p = 0.0004). The male origin effect was not significant, suggesting that the genotype of the male does not have an effect on female fertility (F1,4 = 1.822, p = 0.2484). More important, the interaction between female origin and male origin was not significant (F16,396 = 1.86, p = 0.023), demonstrating that the heterogeneity in fecundity (and therefore, the observed reproductive character displacement) is a characteristic that depends primarily on the genotype of the female, regardless of the genotype of the D. santomea male involved in the heterospecific cross. This kind of reinforcement is expected to be due to changes in females, because they suffer more than do males from interspecific mating [2].
I estimated how long a female could retain and use viable sperm when she was mated to a heterospecific versus a conspecific male. The aim of this test was to determine whether the rate at which a D. yakuba female lost D. santomea sperm—either by depletion or sperm death—differed between allopatric and sympatric D. yakuba lines. Figure 3 shows that heterospecific sperm loss (the most likely cause of noncompetitive gametic isolation) is more pronounced in sympatric than in allopatric lines. This conclusion rests on two results of this analysis. First, the initial hatchability of eggs did not differ between allopatric and sympatric lines (LMM; female origin: F1,4 = 0.585, p = 0.4869). There was no heterogeneity between the intercepts of these crosses, suggesting no substantive difference in number of sperm transferred. Moreover, the decline in egg hatchability over time (slope) was significantly heterogeneous (Model 1 vs. Model 2: LRT = 12.086, p = 5×10−4). This shows that interspecific sperm stored after crosses involving sympatric lines was either retained for a shorter time or became inviable more quickly than in crosses involving allopatric lines. The more rapid loss (or death) of heterospecific sperm in sympatric females is consistent with the observation that sympatric females produce fewer progeny after heterospecific crosses compared to allopatric females. Apart from noncompetitive gametic isolation, no other reproductive barrier shows the signature of reinforcement (Figures S1 and S2, Tables S4, S5, S6, and S7).
Although the evolution of behavioral isolation is clearly advantageous in a hybrid zone when hybrids are semisterile or partly inviable, the benefits of increasing postmating-prezygotic isolation are not so obvious [2],[12],[13]. One possibility is that eliminating heterospecific sperm more quickly allows a female to remate with males of her own species, increasing her chances of passing her genes to the next generation. In such a case, alleles fostering quicker elimination of heterospecific sperm could be selectively advantageous. To check this possibility, I measured the reproductive capacity of D. yakuba females from both allopatric and sympatric populations that had been initially mated to heterospecific males. Four days after this first mating, these females were remated to conspecific males, and I counted the number of eggs produced every day for the next 10 d.
Data from this experiment give two kinds of support for the idea that natural selection might have increased the gametic isolation of sympatric D. yakuba females in nature. First, D. yakuba females from sympatric populations remated more quickly to conspecific males than did sympatric females (LMM on arcsine of the remating probability, F1,12 = 13.4295, p = 0.0032; Figure 4A). Given the CSP that acts in D. yakuba (in double conspecific/heterospecific matings, regardless of mating order, conspecific sperm are used in fertilization much more often than heterospecific sperm, [22]), this faster mating would markedly reduce the proportion of hybrid progeny produced, decreasing the cost of maladaptive hybridization. Second, D. yakuba sympatric females mated to a conspecific male for a second time produced more conspecific progeny than did allopatric females (Figure 4B). Since the total number of eggs produced did not differ between allopatric and sympatric D. yakuba females after two matings (LMM, F1,8 = 0.0031, p = 0.957), the stronger gametic isolation of sympatric females reduces the production of hybrid progeny and increases the number of (more fit) conspecific progeny that they can produce (LMM on number of eggs laid after conspecific mating: F1,8 = 9.726, p = 0.0143). Taken together, these results show that increased gametic isolation can provide a fitness advantage to D. yakuba females who are sympatric with D. santomea.
To establish whether natural selection would increase gametic isolation in the laboratory when species were given the opportunity to hybridize, I exposed seven distinct allopatric lines of D. yakuba (each collected in different years and geographic localities) to experimental sympatry with D. santomea for ten generations. If maladaptive hybridization promotes the evolution of postmating-prezygotic isolation, and there is genetic variance for the character, we might be able to observe such isolation evolving in the experimental sympatry lines. It is important to note, however, that in this study, hybrids are rendered completely inviable, whereas a few viable hybrids have been found in the wild. Although these species do mate in the wild, female hybrids have never been found, male hybrids are completely sterile, and most hybrids have been F1 individuals, with only 4% of them being from backcrosses [19].
D. yakuba females exposed to experimental sympatry evolved substantial gametic isolation within four generations, whereas unexposed D. yakuba populations showed no change in isolation over time (Figure 5A). This difference was highly significant (paired t-test: t5 = 4.32, p = 0.0076). I also observed a substantial increase in sexual isolation between D. yakuba females and D. santomea males in sympatric, but not in the unexposed control populations (paired t-test: t5 = 4.85, p = 0.0047; Figure 5B). This is surprising in view of the lack of evidence for reinforcement of sexual isolation of these species in nature [20]. None of the other isolating barriers examined (copulation latency and duration) changed over time (Figures S3 and S4).
Although the experimental-sympatry study demonstrates the evolution of reproductive character displacement rather than reinforcement per se, for several reasons, these results increase the likelihood that increased gametic isolation in sympatry did result from reinforcement: i) gametic isolation is a heritable trait and responds to selection, ii) increased gametic isolation similar to that seen in nature is caused by the presence of D. santomea, and iii) the genetic variability required for sexual and gametic isolation to evolve is present in allopatric populations. Additionally, since D. santomea were added to the experimental sympatry bottles each generation, I did not examine the possibility of reinforcement of gametic (or sexual isolation) in that species.
Increased reproductive isolation in sympatry can be generated through a variety of processes. Although reinforcement is the most commonly invoked explanation for reproductive character displacement, other processes—such as ecological character displacement and differential extinction or differential fusion— can generate the same pattern [2],[33]–[36]. Two results, however, suggest that ecological character displacement is an unlikely explanation for the observed pattern. First, to control for this possibility, I included several allopatric lines of D. yakuba collected from higher elevations off of São Tomé (e.g., Mount Cameroon, Pico Basile, and Nairobi, Table S1), which thus lived at elevations similar to the D. yakuba lines derived from the hybrid zone (Table S1). The aim of this test was to examine the possibility that the observed reproductive character displacement was a byproduct of adaptation to high elevation alone (allopatric lines collected at high elevations are represented by bars K–M, Figure 2B). These allopatric, high-elevation lines of D. yakuba did not, however, show elevated gametic isolation. Moreover, the results from the experimental sympatry experiment show that reproductive character displacement occurs if D. yakuba is exposed to D. santomea and when there is strong selection against the hybrids, even when the “ecology” is that of a food-filled milk bottle in the laboratory.
The second possibility is that the observed range of gametic isolation reflects the results of a deme-sorting process involving differential extinction (or differential fusion) of populations based upon levels of reproductive isolation. Under this scenario, only those populations that have a high, pre-existing level of reproductive isolation will be able to colonize and persist in a region where a potentially interbreeding sister species is present. This hypothesis yields a clear prediction: if biased extinction is the driving force behind the observed differences in levels of reproductive isolation, then non-allopatric populations should show a distribution of reproductive-isolation values lying within the range of phenotypic values seen in allopatric populations [31],[32]. The data suggest that this is not a likely explanation for the elevated gametic isolation seen in sympatric D. yakuba lines: values of gametic isolation in sympatric lines are not a subset of that of the distribution of values of allopatric individuals (Figure 6, Figure S5, ANOVA on pooled individual values with resampling of cells: F1,310 = 341.93, p<1×10−4).
Two further lines of evidence render differential fusion/extinction an unlikely explanation. First, differential fusion predicts that premating, postmating-prezygotic, and postzygotic isolation would be stronger in sympatry than allopatry [2],[31]. Tables S4, S5, S6, and S7 show that it is not the case: the only reproductive isolating barrier appears to be strengthened in sympatry is gametic isolation. Second, to explain the existence of substantial differences in gametic isolation before secondary contact, differential fusion/deme selection would require very low levels of gene flow between populations [2],[34]. Previous studies have shown that this is not the case for D. yakuba, which exhibits very little population structure [17].
All these considerations render alternative possibilities, such as ecological character displacement and differential extinction/fusion, unlikely. I suggest that reinforcement is the most likely cause of the reproductive character displacement observed in D. yakuba populations that are sympatric to D. santomea.
Four conditions must be met before one can conclude that reinforcement is the cause of a pattern of reproductive character displacement between two species [33],[34]. First, gene flow, either current or recent, has to occur between them. Second, there must be, or have been, natural selection against maladaptive hybridization. Third, the trait causing reproductive isolation must be heritable and capable of responding to selection. Finally, one must rule out alternative explanations such as ecological character displacement. The work described above fulfills these requirements, suggesting that reinforcement for postmating prezygotic isolation has indeed evolved in populations of D. yakuba that are sympatric with D. santomea.
Some cases of reproductive character displacement of gametic isolation have been reported previously. Geyer and Palumbi [37] describe reproductive character displacement in the sequence of proteins involved in gametic interactions in sympatric populations of the sea urchin Echinometra oblonga. A similar example occurs in abalone and mussel species, which show a strong signature of positive selection in proteins involved in sperm–egg interaction, especially in sympatric species [38]–[42]. In all these cases, selection for local gamete coevolution (as a result of interactions between sympatric species) seems to be the driving force of speciation; however, the authors do not describe higher gametic isolation between sympatric than between allopatric populations of the same species pair, so it is possible that these patterns reflect processes other than reinforcement (e.g., differential fusion; [2],[34]).
D. santomea and D. yakuba, then, appear to represent the first example of reinforcement for a postmating-prezygotic trait in an organism that has internal fertilization. In this particular case, reinforcement operates when several reproductive barriers are already strong. Also, the major selection pressure seems to be direct—on the number of offspring produced—rather than indirect—on the fitness of hybrid offspring. The reason why only gametic isolation, and not sexual isolation, is reinforced in natural populations of D. yakuba remains an unanswered question, especially given that behavioral isolation mechanisms occurring earlier in the life history can more effectively reduce the costs of hybridization [2],[43],[44]. There are two explanations for why D. yakuba females show reinforced gametic isolation but no reinforced behavioral isolation. CSP reduces the cost of heterospecific matings for females, and thus reduces the likelihood reinforcement of sexual isolation [43]. It is possible that CSP reduces the likelihood of reinforcement of behavioral but not of gametic isolation; however, this seems unlikely given that CSP reduces the costs of hybridization as a whole, and its effects should reduce the likelihood of reinforcement of all mechanisms of reproductive isolation. A second possibility is that if behavioral isolation is not an effective isolating mechanism in nature, then gametic isolation can play a very prominent role on reproductive isolation, as occurs in free-spawning marine invertebrates. Again, previous inventories of reproductive isolation between D. yakuba and D. santomea and the low frequency of hybrids in nature (Ipsi = 0.54 for no choice experiments; [20],[21],[23],[45]) render this explanation as unlikely.
For sexual isolation, it has been predicted that reinforcement should be stronger in the rarer species, as rarity increases the probability of mating with the wrong species [2]–[7] and thus selection to avoid maladaptive hybridization stronger. Previous studies have demonstrated that in the hybrid zone D. yakuba is indeed rarer than D. santomea [19]. Although the reproductive mechanism that is reinforced in this case is not sexual but gametic isolation, our results do comply with this prediction.
Finally, I show that gametic isolation (and not only sexual isolation) can evolve under laboratory conditions—and can do so very quickly if natural selection is strong. These results, together with some previous examples [29],[30] in which artificial sympatry promoted the evolution of reproductive character displacement, demonstrate that prezygotic isolation (both premating and postmating-prezygotic) can evolve quickly given the strong selection regime and the presence of genetic variation. Whether reinforcement would evolve if gene flow was permitted and the selection regime was weaker is an unanswered question that I am currently investigating.
To date, the study of postmating-prezygotic barriers in speciation has focused largely on documenting their existence. The processes and mechanisms that generate such reproductive mechanisms are, understandably, less well understood than those that generate premating isolation [13],[46],[47]. Previous studies have shown that postmating-prezygotic characters can evolve rapidly and that such evolution can be the result of differences in the coevolutionary trajectory between males and females among populations or species [48],[49]. Postmating-prezygotic isolation can also evolve as a byproduct of ecological divergence and be heavily influenced by the ecology of a species [36],[48]–[51].
This work shows that reinforcement of barriers other than sexual and other forms of premating isolation is possible. This suggests that there are many “cryptic” barriers to gene flow that might be increased by natural selection in areas where species overlap and hybridize.
|
10.1371/journal.pbio.2004356 | Antibiotic combination efficacy (ACE) networks for a Pseudomonas aeruginosa model | The spread of antibiotic resistance is always a consequence of evolutionary processes. The consideration of evolution is thus key to the development of sustainable therapy. Two main factors were recently proposed to enhance long-term effectiveness of drug combinations: evolved collateral sensitivities between the drugs in a pair and antagonistic drug interactions. We systematically assessed these factors by performing over 1,600 evolution experiments with the opportunistic nosocomial pathogen Pseudomonas aeruginosa in single- and multidrug environments. Based on the growth dynamics during these experiments, we reconstructed antibiotic combination efficacy (ACE) networks as a new tool for characterizing the ability of the tested drug combinations to constrain bacterial survival as well as drug resistance evolution across time. Subsequent statistical analysis of the influence of the factors on ACE network characteristics revealed that (i) synergistic drug interactions increased the likelihood of bacterial population extinction—irrespective of whether combinations were compared at the same level of inhibition or not—while (ii) the potential for evolved collateral sensitivities between 2 drugs accounted for a reduction in bacterial adaptation rates. In sum, our systematic experimental analysis allowed us to pinpoint 2 complementary determinants of combination efficacy and to identify specific drug pairs with high ACE scores. Our findings can guide attempts to further improve the sustainability of antibiotic therapy by simultaneously reducing pathogen load and resistance evolution.
| Bacterial infections are commonly treated with a combination of antibiotic drugs. However, not all combinations are equally effective, and success is variable. One reason for this variation is that we usually do not know to what extent bacteria are able to adapt to different types of drug combinations. If they can and do adapt, then antibiotic resistance can spread, potentially aggravating the current antibiotic crisis. In the current study, we therefore asked whether combination therapy can be improved by considering the evolutionary potential of the bacteria. To address this question, we systematically assessed the efficacy of antibiotic combinations using controlled laboratory evolution experiments with the opportunistic human pathogen Pseudomonas aeruginosa as a model. We found that 2 factors consistently increase treatment efficacy. First, synergism between the combined drugs (i.e., the 2 drugs enhance each other’s effects) increases the rate of bacterial population extinction and thus clearance rate. Second, evolved trade-offs such as collateral sensitivity (i.e., evolution of resistance to one drug increases susceptibility to the other drug) limit the ability of bacteria to adapt to the antibiotic pair. Our findings may help to optimize combination therapy by focusing on drug pairs that interact synergistically and also lead to evolved collateral sensitivities.
| The rise of antibiotic resistance is reducing the arsenal of available drugs to treat bacterial infections [1–3]. Some infections are already nearly untreatable because the infecting pathogens are resistant to virtually all available drugs [4,5]. The identification and establishment of new antibiotics has become a major focus of national and international health programs, and substantial investments have been directed towards drug discovery, for example, by the United States and the European Union [6–10]. Yet even if these attempts succeeded and dozens of novel compounds became available tomorrow, the antibiotic crisis would not subside. The evolution of resistance is inevitable, and new drugs will be incapacitated within short time periods [2,3]. So how can we hamper this evolutionary march towards resistance? To some extent, we cannot escape the open-ended arms race between compound discovery and resistance evolution. Nevertheless, we may still use evolutionary thinking to enhance treatment efficacy and sustainability [11]. Combination therapy, the simultaneous deployment of 2 or more drugs, is commonly proposed [12]. Indeed, WHO has endorsed it as the first-line strategy to treat diseases such as tuberculosis, malaria, or HIV [13–15]. However, the nature of the drug combination is crucial for treatment success because initially effective combinations may maximize selection for antibiotic resistance [16,17].
The approach of experimental evolution has proven highly informative on exploring the dynamics that shape the emergence and spread of drug resistance [11,18]. Using this approach, drug pairs were previously suggested to be most effective at limiting bacterial adaptation if (i) antimicrobials display collateral sensitivity, such that bacteria that evolve resistance to one of the compounds immediately suffer exacerbated suppression by the other [19–22], or (ii) antibiotics interact antagonistically, such that they inhibit each other’s effect [16,23,24]. A mathematical model indicated that the latter empirical findings may not be generally applicable but depend on the exact conditions during evolution [25]. In particular, synergistic drug pairs generally favor bacterial clearance but only sometimes low adaptation rates. The strong reduction in population size by synergistic drugs decreases the likelihood of resistance mutations emerging and increases the chances of population extinction. However, these effects only correlate with low adaptation rates when resource competition is weak. When resource competition is high, resistance mutations have a strong selective advantage and may spread rapidly through the population due to competitive release. Under these conditions, antagonistic rather than synergistic drugs are most efficient in reducing adaptation rates [25]. To date, few experimental data are available to explore these particular model predictions—and, moreover, test the role of evolutionary trade-offs, such as the evolved collateral sensitivities—on bacterial adaptation in multidrug environments.
In the current study, we performed a systematic analysis using an experimental evolution approach and the gram-negative opportunistic human pathogen Pseudomonas aeruginosa as a model. We evaluated 38 drug pairs for their ability to effectively constrain bacterial adaptation in multidrug environments and calculated 2 antibiotic combination efficacy (ACE) networks based on either the rate of adaptation or bacterial clearance (i.e., frequency of population extinction). These measures provide complementary information on treatment efficacy. First, population extinction represents the ultimate aim of any antibiotic intervention; its frequency is a highly informative indicator of treatment efficacy under our specific experimental conditions, in which antibiotics are always applied at sublethal doses. Second, for the surviving populations, we further evaluated increases in growth rates as a measure of the bacteria’s adaptive potential in antibiotic environments [16]. We subsequently employed complementary statistical approaches, including an integrative Bayesian network (BN) analysis, to disentangle the relative impacts of drug interaction type and evolved collateral effects between individual drugs on the characteristics of the inferred ACE networks. For selected drug pairs, we additionally explored to what extent adaptation to the combinations is driven by the single-component drugs or by initial drug inhibitory levels.
Antibiotic interactions are defined as synergistic, additive, or antagonistic when the drug pair has a stronger, equivalent, or weaker inhibitory effect on bacterial growth than the corresponding single drugs (i.e., monotherapies), respectively. Here, we determined this interaction quantitatively using an estimator denoted α [17]. This estimator is obtained from a quadratic regression applied to growth measurements as a function of different drug proportions of 2 drugs. The concentration of each of the single drugs is chosen to fall onto the line of equal dose, in our case defined to inhibit 75% of growth (i.e., inhibitory concentration [IC] 75; Fig 1A, S1 Fig and Table 1). The estimator α describes the shape of the resulting response in growth whereby positive values indicate synergism and negative values antagonism (Fig 1B). This approach has two advantages: first, it provides a statistical framework for testing the significance of positive or negative α; and second, its inference is less laborious than alternative procedures, thus facilitating characterization of a larger number of drug interactions. Even though the approach was carefully evaluated previously [17], we specifically validated its suitability for our model system. We compared the inferred α values for 8 selected combinations (S2A Fig and S3 Fig) to the corresponding results obtained with one of the commonly used alternative methods, based on Bliss independence and the checkerboard approach (S1 Data for a key to all datasets and S2 Data), as previously described for Escherichia coli [16,26]. This comparison demonstrated that α correlates significantly with the degree of synergy (S), irrespective of whether S is calculated from the average of all viable concentrations across a grid defined by the 2 drugs (ABij = Ai + Bj; S2B Fig) or from combinations for which the 2 individual drugs had the same level of inhibition (ABij for which IC50[Ai] = IC50[Bj], S2C Fig). We thus conclude that the α estimator provides an informative, quantitative indicator of a 2-drug interaction.
We subsequently evaluated the interactions among 12 different antibiotics representing 5 classes (Table 1). We chose these drugs as representatives of the main classes of antibiotics, which are commonly used in combination to treat P. aeruginosa and to which most clinical P. aeruginosa strains are still susceptible [27–29]. Even though this choice could have introduced a bias in the overall pattern of inferred interaction types, these should nevertheless be representative of the clinically applied drug combinations. We characterized drug interactions for almost all of the possible combinations, resulting in a total of 52 measures that we summarized in an interaction network (Fig 1C, S3 Fig, S1 Table, S3 Data). Overall, synergistic combinations were more common than other interaction types (synergistic = 24/52; additive = 14/52; and antagonistic = 14/52). Combinations between cell wall inhibitors (β-lactams) and aminoglycosides most often produced synergisms, whereas those including ciprofloxacin (CIP) had exclusively antagonistic effects (Fig 1C).
We used evolution experiments to assess ACE, which is the ability of drug combinations to constrain bacterial adaptation either through population extinction or, in the case of surviving populations, reduced adaptation rates. Based on the inferred drug interactions and the previously obtained frequencies of collateral sensitivity between 8 of the considered antibiotics (Fig 2) [30], we selected 38 drug pairs covering all different types of drug interactions and collateral effects.
Based on this choice of drugs, we evolved a total of 1,672 populations through serial transfers into fresh media containing the respective antibiotics using a transfer period of 12 h and a total of 10 transfers (total duration of 120 h; Fig 3A and S4 Data). We assessed bacterial adaptive potential by integrating quantitative growth measurements taken in 15-min intervals from each evolving population (a total of 783,464 measurements for all treatments and populations; for a validation of our optical density (OD) measures as a proxy for bacterial growth, see Materials and methods and S4 Fig). For each population in a growth season, we then calculated the growth rate r during the exponential phase (Fig 3B). Following previous work [16], we defined the rate of adaptation as the change in growth rate over time for each evolving population (Fig 3C; for a validation of using growth characteristics as a proxy of evolutionary adaptation, see Materials and methods and S5 Fig). For subsequent analysis, we focused on the results of the 50:50 drug proportion (S6 Fig) and the single-drug treatments (S7 Fig).
We reconstructed the 2 ACE networks based either on adaptation rates of the surviving populations (Fig 4A) or on population extinctions (Fig 4B). Below, we first describe the patterns seen in the ACE networks, while their statistical analysis is explained in the next section. In all cases but one (for carbenicillin [CAR] plus gentamicin [GEN], all populations went extinct), adaptation to the combination treatment was possible. However, the rates of adaptation varied substantially across the different drug combinations, with lower rates of adaptation (below the 50th quantile) predominantly, but not exclusively, seen among antagonistic combinations that included CIP (Fig 4A; S8 Fig and S9 Fig show separate ACE networks for each drug interaction type and the 2 types of evolved collateral effects, respectively). Several synergistic drug pairs, combining an aminoglycoside with either a penicillin or carbapenem, led to similarly low rates of adaptation (below the 50th quantile, S8 Fig). Moreover, almost all cases of collateral sensitivity included in this study were associated with reduced adaptation rates (S9 Fig). This was not the case for combinations with cross-resistance. Furthermore, when estimating clearance efficacy, we found that extinctions almost exclusively occurred with the synergistic combinations (Fig 4B, S8 Fig). The synergistic combinations that did select for lower rates of adaptation did not necessarily have higher rates of extinction and vice versa (populations surviving synergistic combinations were not necessarily adapting more slowly; see azlocillin [AZL] plus streptomycin [STR], cefsulodin [CEF] plus CAR, or ticarcillin [TIC] plus GEN; S8 Fig).
We next performed 2 types of statistical analyses to assess to what extent the overall characteristics of the 2 ACE networks are determined by the 2 considered predictors of combination efficacy: interaction type inferred from α (Fig 1C) and collateral sensitivity profiles previously obtained from experimentally evolved resistant populations of P. aeruginosa (Fig 2, [30]). We first used a BN approach to assess the relationships among the considered variables (i.e., adaptation rate, extinction frequency, drug interaction, and frequency of collateral resistances [FCR]). The BN approach is based on a constraint-based interleaved incremental association algorithm [31–33] to dissect the relationships between our variables (see Materials and methods for details). The results are summarized in the BN (Fig 5A), in which nodes represent the different variables and arrows indicate the inferred dependencies. The BN analysis revealed that the type of antibiotic interaction strongly influenced the proportion of extinction, but not the rate of adaptation. Instead, the rate of adaptation was found to depend solely on the frequency of collateral sensitivities. No other dependency was inferred by the analysis.
Based on the BN structure, we calculated the conditional probabilities for the inferred dependencies between the frequencies of collateral sensitivity and the rates of adaptation as well as for the proportion of extinction and drug interaction type. In particular, we used the different types of evolved collateral effects (i.e., partial collateral sensitivity, partial cross-resistance, and cross-resistance; none of the combinations evaluated during evolution had complete collateral sensitivity between their components, as shown in Fig 2) and calculated the conditional probability of obtaining the distribution of observed adaptation rates across 5 equal quantile bins (Fig 5B, top panel). Similarly, given the different drug interaction types (synergism, additivity, and antagonism), we calculated the conditional probabilities of different extinction frequencies across 5 equal quantile bins (Fig 5B, bottom panel). These 2 additional analyses describe more clearly the inferred dependencies within the BN. Antibiotic combinations for which at least half of the populations had collateral sensitivity against one or both of the individual drug components (i.e., partial collateral sensitivity; purple bars in Fig 5B, top panel) have a higher probability of selecting for low but not high rates of adaptation. Conversely, combinations with partial or complete cross-resistance (green bars in Fig 5B, top panel) have a higher probability of producing the top scores of inferred adaptation rates. In addition, high probabilities of extinction are associated with synergistic and additive combinations, whereas the reverse is found for antagonistic drug pairs (Fig 5B, bottom panel).
We further validated the inferred dependencies between variables using partial correlation analysis, following the approach previously established for a similar analysis of combination efficacy in E. coli [16]. This approach allowed us to control for drug pair membership using the average rate of adaptation towards the corresponding single drugs of a particular combination as a covariate (Materials and methods). Statistical significance was subsequently inferred using a permutation test [16]. This analysis revealed a significant correlation between the FCR and the rate of adaptation (ρs = 0.52, P = 0.038) and between the proportion of extinction and the drug interaction type α (ρs = 0.51, P = 0.043), but not between the FCR and the proportion of extinction (ρs = 0.39, P = 0.146) or the drug interaction α and the rate of adaptation (ρs = 0.3, P = 0.262). This analysis, based on a distinct statistical approach, thereby corroborated the findings of the BN analysis. We conclude that synergistic drug interactions enhance bacterial clearance, whereas collateral sensitivity limits the adaptive potential of the bacteria.
We next assessed whether the ability of bacteria to adapt to the combination is mainly driven by adaptation to only one of the drugs rather than dependent on a unique property of the antibiotic pair. For our dataset, we related the inferred rates of adaptation in the combination treatments to those inferred for the corresponding single-drug environments (S10 Fig). We first compared the 2 corresponding monotherapies of a given drug pair and defined the drug leading to lower rates of adaptation as the stronger component (i.e., higher ability to minimize resistance evolution) and the other as the weaker component (i.e., lower ability to minimize resistance evolution). Thereafter, we calculated the relative rate of adaptation of the combination by standardizing it against either the stronger or the weaker component of the pair. The resulting ACE networks are shown in Fig 6A and 6B, respectively. Interestingly, the original ACE network for adaptation rates (Fig 4A) is more similar to that standardized by the weaker but not the stronger component drug (Fig 6; S2 Table). This suggests that the characteristics of the original ACE network (Fig 4A), and thus the efficacy of drug combinations to reduce adaptation rates, is primarily driven by adaptation to the stronger component, which—if accounted for by the standardizing scheme—removes important properties of the network (see as prominent examples the disappearance of the strong reduction in adaptation rate for doripenem [DOR] plus TIC, or DOR plus PIT [piperacillin + tazobactam]; Fig 4A and Fig 6A).
We further evaluated influence of the component drugs by repetition of the BN analysis. We found that the dependency observed between the FCR and the rates of adaptation of the combinations disappeared when the latter is weighted by the stronger but not the weaker component drug (Fig 6C). At the same time, the dependency between drug interaction and extinction frequency remained, while no additional relationship was revealed. Similar results were obtained when we repeated the correlation analysis with standardized adaptation rates. The originally identified correlation between the FCR and the rate of adaptation was no longer significant when the latter was standardized by adaptation to the stronger component drug (ρs = 0.33, P = 0.21), yet it still showed a statistical trend when we standardized by the weaker component drug (ρs = 0.45, P = 0.078). In these 2 analyses, drug interaction did not correlate significantly with the weighted adaptation rates (ρs < 0.47, P > 0.09). These results consistently indicate that adaptation to the stronger component drug influences adaptation to the combination and that this is dependent on the evolved collateral effects.
We next performed a separate evolution experiment with 4 selected combinations to assess to what extent the inherently different starting levels of inhibition—imposed by each type of interaction during the first season of growth (Fig 1B and S3 Fig)—influenced both the number of extinctions and adaptation rates. We performed this evolution experiment with 4 selected combinations with different interaction profiles: 2 interacting synergistically (GEN plus CAR and STR plus PIT) and 2 antagonistically (GEN plus CIP and Tobramycin [TOB] plus CIP). For these combinations, we varied the initial inhibition level of the combination across 8 steps, ranging from IC50 to >IC90. Populations were serially transferred into fresh media as explained before (S5 Data; and for the obtained changes in growth rate r, see S11 Fig).
This separate evolution experiment revealed that initial inhibitory levels of the tested combinations are significantly related to the rates of adaptation, irrespective of combination identity or drug interaction type (GLM, F1,336 = 37.735, P < 0.001; Fig 7A and S3 Table). In particular, increasing levels of inhibition are generally associated with higher rates of adaptation, suggesting that strong inhibition increases selection for an adaptive response [34,35]. At higher levels of inhibition, the synergistic and antagonistic combinations produce clearly distinct responses, especially regarding population extinction. Here, the 2 synergistic pairs are associated with a significant increase in the number of extinct populations (logistic regression, F12,336 = 21.15, P < 0.001; Fig 7B and S4 Table), while antagonistic combinations produced almost no extinction at all. Moreover, at the very high initial inhibitory levels, antagonistic pairs showed a sudden drop in adaptation rates (Fig 7A), as expected from previous work [16,24]. A similarly strong reduction is not observed for the synergistic combinations, possibly owing to the fact that only few populations survived and could be used to infer adaptation rates.
Taken together, the results from this separate evolution experiment suggest that the generally higher inhibition levels of the synergistic pairs in our main evolution experiment could potentially have contributed to higher adaptation rates for this type of combination (even though these were not found to be significantly increased compared to those for other interaction types; see above). This seems less likely the case for extinction events, which are generally more frequent in treatments with synergistic rather than antagonistic combinations, irrespective of the initial inhibition level.
Our study provides a systematic experimental analysis of the efficacy of antibiotic combination therapy in the opportunistic human pathogen P. aeruginosa. Based on evolution experiments with 38 distinct combinations, ACE networks were reconstructed for 2 complementary measures of treatment efficacy: the frequency of population extinctions and the reduction in adaptation rates. Subsequent statistical analyses identified the likely ACE determinants: Synergistic drug interactions enhanced the frequency of extinction, even at the same inhibitory level as antagonistic interactions, while reduced adaptation rates depended on the evolved collateral sensitivities among the drugs. The latter effect is likely driven by adaptation to the stronger component drug in a pair. Consequently, our findings suggest that treatment efficacy against P. aeruginosa can be optimized by drug combinations, which interact synergistically to increase bacterial clearance and which can evolve collateral sensitivity to each other to slow down the rate of adaptation.
The use of BN analysis enhanced dissection of the determinants of ACE. The BN approach has been widely applied across different fields of biology in recent years but not yet in studies on antibiotic resistance evolution [33,36–39]. Its accessible graphical output and the underlying probabilistic theory facilitate the inference of causal relationships between different variables [31,32]. It further offers estimation of conditional probabilities that reflect the strength of the inferred dependencies; a strategy well suited for the stochastic nature of biological systems and their measurements [40]. The latter is important for the analysis of antibiotic resistance evolution, for which we are mainly interested in anticipating bacterial adaptation based on distinct drug properties or deployment strategies [11,12,41–43]. The suitability of the BN approach for analysis of drug resistance evolution was corroborated with a previously established statistical approach, based on partial correlation analysis [16], which identified a significant relationship for the same pairs of variables.
Our analyses consistently revealed that synergistic drug interactions are an important ACE determinant, especially in terms of bacterial clearance (Fig 4A). The particular importance of bacterial elimination as a component of treatment efficacy was previously considered in a mathematical model [25] but has not yet been evaluated empirically. The previous model assessed the effect of antibiotic interactions on treatment efficacy [25] by modifying a previous infection model based on data from mice infected with P. aeruginosa [44]. The model is related to the design of our main evolution experiment in that the concentration of a particular drug in a combination is standardized by its inhibitory effect in monotherapy. The model predicted contrasting treatment outcomes for synergistic combinations: On the one hand, synergism enhances extinction, most likely because it strongly reduces population size, thereby decreasing the likelihood of new resistance mutations arising. On the other hand, if resistance emerges, synergism increases the selective advantage of the resistant mutants through competitive release, enhancing bacterial adaptation [25]. Our experimental results are consistent with both alternatives. Although synergism mainly favored bacterial extinction (Figs 5–7), it was in several cases associated with low adaptation rates (Fig 4A). However, in our study, the effect of drug interaction on adaptation rate was always insignificant, irrespective of the analytical approach.
Interestingly, we found higher population extinction for synergistic rather than antagonistic combinations also at low initial inhibitory concentrations (Fig 7). This finding cannot have resulted from the stronger reduction in population size (i.e., inhibitory levels were the same for the 2 interaction types) but must have depended on other properties of the synergistic drug pairs. A likely explanation may be found in the mechanism underlying synergism, which can rely on increased membrane permeability induced by one of the drugs, subsequently enhancing cellular uptake of the second drug [45]. Such mechanisms may have a cumulative effect across time [45] and/or may generally be difficult to counter. This, in turn, limits the number of suitable resistance mutations and ultimately increases the likelihood of extinction. A detailed exploration of this effect clearly warrants further research.
Our experiments further identified the potential to evolve collateral sensitivity as a key determinant of low adaptation rates. This result is generally consistent with previous work on E. coli and Staphylococcus aureus [46,47], although this is the first time it has been shown for P. aeruginosa. Adaptation rates are thus significantly influenced by evolutionary trade-offs, whereby adaptation to one of the drugs of a pair constrains adaptation to the other. Our findings and those of colleagues [46,47] thereby highlight that such trade-offs may not only improve treatment when drugs are applied sequentially, as originally proposed for evolved collateral sensitivities in E. coli (i.e., collateral sensitivity cycling; [20–22]). Instead, they can also optimize combination therapy. Our analysis further revealed that the involved dynamics are likely driven by adaptation to the stronger component drug of a pair (Fig 6). This suggests that, if adaptation to the stronger component comes with a higher likelihood of collateral sensitivity to the second drug, adaptation to the combination is systematically slowed down, as, for example, for CIP plus STR or CIP plus CAR (Fig 2, Fig 4A). In contrast, when adaptation to the stronger drug is more likely to cause cross-resistance, then this can enhance adaptation to the combination, as seen for GEN plus STR or CAR plus CEF (Fig 2, Fig 4A). The further exploration of these trade-offs represents a promising avenue to improve treatment efficacy.
Our finding of the high clearance efficacy of synergistic combinations shows some consistency with clinical practice. For P. aeruginosa, we predominantly observed drug synergism between β-lactams and aminoglycosides (Fig 1C). These 2 antibiotic classes are also most commonly used in combination therapy against this pathogen [29,48,49]. Our results empirically confirm the potency of the β-lactam–aminoglycoside combinations, especially penicillin–aminoglycoside pairs, in causing higher numbers of extinct replicate populations (Fig 4B and S8 Fig). In some cases, the populations surviving these specific combinations also adapted more slowly (e.g., STR plus PIT or TIC plus TOB in Fig 4A and 4B, and S8 Fig). Furthermore, the effectiveness of these combinations may not only be caused by drug synergism but additionally by reciprocal collateral sensitivity that can evolve among these pairs [30]. Our systematic analysis performed under controlled laboratory conditions thus provides empirical support for the often experience-driven choice in clinical treatment. In the future, the clinical applicability of our results should be further explored. For example, we identified high clearance efficacy of certain combinations of penicillins and cephalosporins (Fig 4B) or low adaptation rates if fluoroquinolones (e.g., CIP) were combined with aminoglycosides or penicillins (Fig 4A). It would be of particular interest to corroborate these patterns for clinical isolates in laboratory experiments or under clinical conditions.
In summary, our systematic analysis of antibiotic combinations identified the role of drug interactions and evolved collateral effects in determining 2 complementary properties of treatment efficacy. The comprehensive dataset collected in our study may serve as a useful reference for further exploration of effective therapy, including more detailed statistical analyses such as those that use the potency of pairwise interactions to estimate higher-order drug effects [50,51]. Our approach and the specific results obtained may, moreover, help to improve the design of medical treatment with the 2-fold aim of minimizing pathogen burden and reducing resistance evolution. A similar combined assessment of the efficacy of drug interaction and evolved collateral effects may not only be applicable to other pathogens and infectious diseases. It could similarly help to improve cancer therapy, as previously evaluated for selected cancer types and drug interactions [52–55].
All experiments were conducted with P. aeruginosa PA14. Cells were grown at 37 °C in sterile M9 minimal medium supplemented with 0.2% glucose and 0.1% casamino acids. All antibiotics were prepared according to the manufacturer’s instructions and filter sterilized before each experiment (Table 1). All experiments were carried out in randomized 96-well plates shaken and incubated at 37 °C in BioTek Eon plate readers, which were also used for regular measurement of ODs in 15-min intervals. Randomization schemes of plates for each experiment were different from each other. All analyses were performed using the R platform (version 3.3.2) unless specified otherwise [56].
We tested 14 different concentrations of each drug in order to establish dose-response relationships after 12 h of incubation. For all concentrations, a 1- to 2-ml 10× stock was prepared and then diluted in a randomized 96-well plate with 6 replicates per concentration, resulting in 90 replicates per antibiotic and 1,080 for all treatments. Ten microliters of an isogenic bacterial population of PA14 were added to a final volume of 100 μl, equivalent to 104 to 105 CFU/ml initial population size. In addition, 2 types of controls were included: one without antibiotic and a second one without both antibiotic and bacteria, each also replicated 6 times. We used a logistic regression to analyze the dose-response relationship of each drug using the package “drc” in R [57]. The obtained models (S1 Fig) allowed accurate calculation of different levels of inhibitory concentrations for each drug, including the minimum inhibitory concentration (MIC; here defined as the concentration inhibiting >90% of growth).
To measure the type of interaction using the checkerboard approach, we considered 9 concentrations of each antibiotic in a pair, including a no-drug control, and distributed them randomly across a 96-well plate. Each pair was evaluated twice. Plates were incubated at 37 °C for 12 h with constant shaking and regular OD measurements taken every 15 min. We then calculated the growth rate r for each individual well and combination by fitting a linear regression of growth over time during the exponential phase. Exponential phase was generally observed during 195 to 360 min of each season.
We subsequently determined the degree of synergy of any drug pair AB using the Bliss independence method described previously [16]:
S=(rA0/r00)(r0B/r00)−(rAB/r00),
such that rA0 represents the growth rate at a given concentration of drug A in the absence of B, and vice versa for r0B. r00 is the growth rate of the no-drug control, and rAB is the growth rate at any concentration in which drugs A and B are found together. The degree of synergy S was only calculated for drug combinations that had growth rates larger than 0. Positive values indicate synergism, whereas negative ones denote antagonism.
To classify the interaction between 2 drugs, we considered an environment in which each drug separately inhibits 75% ± 10% of bacterial growth (IC75). For each combination, we evaluated 11 treatments: 9 different proportions of a given pair of antibiotics, a control of uninhibited growth, and a control with only M9 medium. Nine replicates for all treatments were considered, except for the M9 control that consisted of only 6 wells. This resulted in 81 replicates per drug combination and 4,212 for all 52 antibiotic pairs. OD measurements were taken every 15 min for 12 h, resulting in a total of 48 data points per individual replicate and 202,176 for all combinations and replicates.
To determine whether interactions were antagonistic, synergistic, or additive, we used a t test on the second-order term (α) of a quadratic regression of our data, as established previously [17]. The α parameter expresses convexity or concavity of observed bacterial-density data in the model q(θ) = αθ2 + βθ + γ, such that θ represents any drug proportion between any drugs A and B (Fig 1B). Positive values of α indicate synergy and negative values antagonism.
We considered our previously published data on the evolved collateral effects of highly resistant populations of P. aeruginosa PA14 [30] and used the frequency of cross-resistance in all possible pairwise combinations of 8 of the drugs considered in this study. Briefly, the FCR counts the number of populations resistant to drug A that show collateral resistance to drug B, and vice versa, relative to the total number of populations resistant to A and B. Values close to 0 indicate reciprocal collateral sensitivity, and those close to 1 denote cross-resistance. We categorized the obtained values into 4 different groups and built a collateral sensitivity network (Fig 2): complete collateral sensitivity (FCR ≤ 0.25), partial collateral sensitivity (0.25 < FCR ≤ 0.5), partial cross-resistance (0.5 < FCR < 0.75), and complete cross-resistance (FCR ≥ 0.75).
Based on the interaction profile and the collateral sensitivity and/or resistance [30] scores, we selected a total of 38 different combinations for a series of evolution experiments (Fig 3A). For all combinations, we included 5 different proportions of the combined antibiotics, an uninhibited control, and an M9 control, resulting in 44 populations per combination, randomly distributed in a 96-well plate (2 combinations were included in a single plate), for a total of 1,672 populations. The concentration was set for each individual drug to inhibit bacterial growth by 75% (IC75). We considered 10 transfers (hereafter referred to as seasons) of 1% volume into fresh plates every 12 h (approximately 120 generations). For each season, OD600 measurements were taken every 15 min, resulting in 48 measurements per replicate and season and a total of 781,440 measurements across all replicate populations. All plates were frozen at −80 °C with 1:4 (v/v) of 86% glycerol.
To validate our OD measurements as a proxy for bacterial growth during evolution, we replicated the conditions of the first season for 4 selected combinations (only the 1:1 proportion), 6 corresponding single-drug treatments, and a no-drug control. We focused on those combinations and the corresponding monotherapies for which we also evaluated the influence of initial drug inhibitory level (Fig 7) and the evolution of resistance (S5 Fig). Each treatment was replicated 8 times. After 12 h of evolution, we performed a dilution series and standard plating techniques to count viable colony-forming units (CFUs) for all replicates and treatments. The obtained CFUs were then correlated with the endpoint OD measurements (S4 Fig). We found a significant correlation between our OD measurements and the CFU counts at the end of season 1 (Spearman rank correlation test, ρs = 0.782, P < 0.001). To further validate the OD measurements, we performed a similar correlation analysis for the same combinations and corresponding monotherapies, using evolved bacteria from the final transfer of the separate, focused evolution experiment, in which the influence of initial drug inhibitory levels was assessed. The evolved material was thawed from the frozen stock cultures, then exposed to 1 full season of experimental evolution under the exact treatment conditions already experienced by populations during the evolution experiment. Thereafter, CFUs were counted using a dilution series on Agar plates, as outlined above, and then compared to the OD measures obtained during the above repetition of a full season. As before, CFUs were significantly correlated with the corresponding OD measurements (Spearman rank correlation test, ρs = 0.339, P = 0.002).
We further validated the suitability of changes in growth characteristics as a proxy for evolutionary adaptation and therefore genetically fixed alterations by re-assessing cryo-preserved material from the last transfer of experimental evolution. This analysis was performed with material from the separate evolution experiment, which tested the influence of initial inhibitory levels, and further details are outlined below in the description of this experiment.
We first calculated the growth rate r as described above for each evolving population, treatment, and season. Subsequently, we considered the rate of adaptation for each evolving line as defined previously [16]:
Radapt=Δr2×tadapt,
such that Δr represents the change in growth rate over 10 seasons of growth, and the time of adaptation, tadapt, corresponds to the interpolated time at which a population reached half of its maximum growth rate. This measurement reflects how quickly resistance spreads in a population in a serial transfer experiment.
To determine to what extent adaptation to the drug combinations was determined by adaptation to each of the individual drugs, we measured which of the individual components in a drug pair led to lower and higher rates of adaptation. The single antibiotic in a pair that alone led to lower rates of adaptation was considered as the stronger of the components and the other as the weaker one. The adaptation rate of each combination was then standardized by the adaptation rate of either its weaker or stronger component drug. The 2 types of standardized adaptation rates were visualized in ACE networks and statistically evaluated (see below).
We used BN analysis to assess the directional relationship between 4 variables, including the inferred drug interaction type, the frequency of collateral sensitivities, the adaptation rates, and the frequency of population extinctions. The entire BN analysis was repeated with the different types of inferred adaptation rates, including those obtained for the combinations in the main experiment and then those that we standardized by either the stronger or the weaker component drug.
The BN analysis generally followed 2 steps. In the first step, the approach identifies variables that are related to each other and visualizes these as nodes in a network between variables. In this step, it further infers the direction of each relationship and represents these as arrows in the network, thereby implying a causality between the connected variables [31]. To achieve this first step, the model first infers the graphical structure of the network by analyzing the probabilistic relations between all nodes and thereafter constructs the network by setting directions for the identified connections while satisfying an acyclicity constraint [58]. We implemented BN analysis employing a constraint-based interleaved incremental association–optimized algorithm [59] to reduce the likelihood of obtaining false positives and to obtain possible probabilistic dependencies between our variables: drug interaction type (categorical: synergism, additivity, or antagonism), FCR (categorical: complete collateral sensitivity, partial collateral sensitivity, partial cross-resistance, and complete cross-resistance), proportion of extinction (numerical), and rates of adaptation (numerical). We only included combinations with complete sets of data and then followed the algorithm’s default parameters. From the obtained dependencies, we estimated the conditional probabilities associated with the linked variables over an array of different values. All tests were performed in R using the “bnlearn” package [60].
To validate the inferred dependencies from the BN analysis, we additionally performed correlation analysis combined with permutation tests, following the approach previously established for a similar analysis of ACE in E. coli [16]. For each round of permutation, we calculated correlation coefficients, ρs, between any two given variables x and y by permuting the values of x while keeping y constant, as in [16]. For each test, we considered 10,000 permutations and estimated the P value as the proportion of the obtained distribution of correlation coefficients that had an absolute value larger than the absolute value obtained for the observed ρs [16]. This approach was used to correlate the measures of collateral effects and drug interaction to proportion of extinction and, later on, to the standardized adaptation rates.
Furthermore, to account for the effect of adaptation to the single drugs (z) in the main analysis with nonstandardized adaptation rates, we performed a partial correlation analysis with z as a covariate, generally following the previously established approach [16]. For this, we first obtained the residuals from the linear regression of x on z and those of y on z, such that y corresponds to the adaptation rates of the combination. Then, to estimate the correlation coefficient between x and y, with z as a covariate, we employed the permutation test as explained above using the residuals of the corresponding regressions [16].
To evaluate the effect of the starting inhibition level of the combinations, we considered a second round of evolution experiments as described above. This time, the level of inhibition of the combination was fixed instead of that of the individual drug treatments. Briefly, concentrations of each drug were mixed 1:1 so that each would inhibit between 50% and 75% of growth. These were then diluted to obtain a range of different inhibition levels and to evaluate their effect on growth in P. aeruginosa after 12 h of incubation at 37 °C. Evolution experiments were then initiated for 4 different combinations that included 11 different treatments: a no-drug control, the individual monotherapies, and 8 different inhibition levels ranging from approximately IC50 to >IC90 of each combination. Each treatment was replicated 8 times and distributed randomly in 96-well plates.
We used the focused set-up of the above separate evolution experiment to validate the suitability of growth measurements as a proxy for evolutionary adaptation. Evolutionary adaptation assumes that changes are genetically fixed rather than due to phenotypic (i.e., physiological) responses. To assess this, we studied cryo-preserved material from the last drug-free season of the evolution experiment and regrew them under defined antibiotic conditions. Purely phenotypic adaptations to antibiotics are unlikely to have persisted for this material, which was grown under antibiotic-free conditions for 12 to 16 h (equivalent to a minimum of 6 generations) and additionally subjected to a cryo-preservation step. Therefore, any persistent changes in growth characteristics under antibiotic exposure are likely based on genetic changes and thus indicate evolutionary adaptation.
For this analysis, we considered material evolved in the presence of 2 synergistic (i.e., GEN plus CAR and STR plus PIT) or 2 antagonistic combinations (i.e, CIP plus GEN and CIP plus TOB), in all cases set to either IC50 or >IC90, and also included material from the corresponding monotherapies. A total of 4 replicate populations was studied for each of the various evolution treatments and compared to the ancestral PA14. Changes in growth characteristics were inferred from dose-response curves in a 2-fold dilution series of each of the antibiotics included in the pair. The evolved relative changes in resistance were calculated as the area under the curve (AUC) of the dose-response curve for each of the populations and then divided by that of the ancestral PA14. The results are shown in S5 Fig. They highlight a general increase in growth characteristics and thus resistance across the various treatment groups even if not significant in all cases (based on a 1-sample Wilcoxon test with μ = 1). We conclude that, overall, the observed changes in growth characteristics have a genetic basis and are not exclusively due to phenotypic responses. Therefore, we consider the recorded changes in growth characteristics to provide a meaningful proxy for evolutionary adaptation.
|
10.1371/journal.pcbi.1005719 | Infectious reactivation of cytomegalovirus explaining age- and sex-specific patterns of seroprevalence | Human cytomegalovirus (CMV) is a herpes virus with poorly understood transmission dynamics. Person-to-person transmission is thought to occur primarily through transfer of saliva or urine, but no quantitative estimates are available for the contribution of different infection routes. Using data from a large population-based serological study (n = 5,179), we provide quantitative estimates of key epidemiological parameters, including the transmissibility of primary infection, reactivation, and re-infection. Mixture models are fitted to age- and sex-specific antibody response data from the Netherlands, showing that the data can be described by a model with three distributions of antibody measurements, i.e. uninfected, infected, and infected with increased antibody concentration. Estimates of seroprevalence increase gradually with age, such that at 80 years 73% (95%CrI: 64%-78%) of females and 62% (95%CrI: 55%-68%) of males are infected, while 57% (95%CrI: 47%-67%) of females and 37% (95%CrI: 28%-46%) of males have increased antibody concentration. Merging the statistical analyses with transmission models, we find that models with infectious reactivation (i.e. reactivation that can lead to the virus being transmitted to a novel host) fit the data significantly better than models without infectious reactivation. Estimated reactivation rates increase from low values in children to 2%-4% per year in women older than 50 years. The results advance a hypothesis in which transmission from adults after infectious reactivation is a key driver of transmission. We discuss the implications for control strategies aimed at reducing CMV infection in vulnerable groups.
| Human cytomegalovirus (CMV) is a herpes virus causing lifelong infection. In high-income countries, the probability of infection increases gradually with age such that at old age up to 100% of the population is infected. CMV is thought to be transmitted mainly by transfer of saliva or urine, but little quantitative evidence is available about the transmission dynamics. We analyze serological data to estimate age- and sex-specific rates of infection, re-infection, and reactivation. The analyses show that infectious reactivation (i.e. reactivation of the virus in an infected person that is sufficient for it to be transmitted to another person) is essential to explain the data. We propose that infectious reactivation in adults is an important driver of transmission of CMV.
| Human cytomegalovirus (CMV) is a highly prevalent herpesvirus that infects between 30% and 100% of persons in populations throughout the world [1]. Usually thought to be a relatively benign persistent infection, CMV is able to cause serious disease in the immunocompromised and offspring of pregnant women with an active infection [2–5]. CMV also has been implicated in a variety of diseases in healthy persons [4, 6–8], and plays a role in aging of the immune system [9–12], perhaps thereby reducing the effectiveness of vaccination in older persons [13–15].
Although the importance of CMV to public health is acknowledged, and even though the development and registration of a vaccine has been declared a priority [16, 17], little quantitative information is available on the transmission dynamics of CMV. At present, the only population-level data derive from serological studies, aiming to uncover which part of the population is infected at what age. These studies show that i) a sizable fraction of infants is infected perinatally (before 6 months of age), ii) seroprevalence increases gradually with age and is usually higher in females than in males, and iii) the probability of seropositivity is associated with both ethnicity and socioeconomic status, with non-western ethnicity and lower socioeconomic status being associated with higher rates of seropositivity [1, 18–21].
CMV infection has a profound impact on the human immune system. Most prominently, it is able to mould the T cell immune repertoire, in particular by expansion of the CMV-specific CD8+ memory T cell pool, a phenomenon called memory inflation [12]. Similar result have been found for memory B cell immunity [22]. With regard to humoral immune responses, high levels of CMV-specific IgG antibodies are increasingly considered a biomarker for lack of control by the immune system of the host, and have been associated with high probability of reactivation ([23, 24], see [12] and references therein). In view of this, it is not surprising that evidence is accumulating of an association between high levels of CMV-specific IgG antibodies, inflammation, vascular disease, and mortality [6, 7].
Person-to-person transmission of CMV from an infected to an uninfected person can occur from a primary infected person, or from a person who is experiencing a reactivation episode or from a person who has been reinfected [4]. Here, we analyze data from a large-scale serological study to obtain quantitative estimates of the relative importance of these transmission routes [21]. We fit mixture models linked to age- and sex-specific transmission models to the data to study the ability of different hypotheses explaining the serological data. Specifically, we quantify the incidence and transmissibility of primary infection, re-infection, and reactivation. Throughout, our premise is that measurements of antibody concentrations provide information on whether or not a person has been infected, and whether or not re-infection or reactivation have occurred. Persons with low measurements are considered uninfected (susceptible), while persons with intermediate and high antibody concentrations are infected with and without subsequent re-infection or reactivation, respectively.
The analyses show that infectious reactivation in adults is necessary to explain the data, and is expected to be an important driver of transmission. The results have implications for control of CMV by vaccination, but also in the broader context of T cell immune memory inflation, vascular disease, and immunosenescence [12, 25, 26].
The study was approved by the Medical Ethics Testing Committee of the foundation of therapeutic evaluation of medicines (METC-STEG) in Almere, the Netherlands (clinical trial number: ISRCTN 20164309). All participants or their legal representatives had given written informed consent.
The analyses make use of sera from a cross-sectional population-based study carried out in the Netherlands in 2006-2007. Details have been published elsewhere [21, 27]. Briefly, 40 municipalities distributed over five geographic regions of the Netherlands were randomly selected with probabilities proportional to their population size, and an age-stratified sample was drawn from the population register. A total of 19,781 persons were invited to complete a questionnaire and donate a blood sample. Serum samples and questionnaires were obtained from 6,382 participants. To exclude the interference of maternal antibodies, we restrict analyses to sera from persons older than 6 months (6,215 samples). We further select Dutch persons and migrants of Western ethnicity to preclude confounding by ethnicity (5,179 samples) and stratify the data by sex [21], yielding 2,842 and 2,337 samples from female and male participants, respectively. The data are available at github.com/mvboven/cmv-serology.
We use the ETI-CYTOK-G PLUS (DiaSorin, Saluggia, Italy) Elisa to detect CMV-specific IgG antibodies. The assay yields continuous measurements (henceforth called ‘antibody concentration’). A small number of samples is right-censored (140 persons). We perform a Box-Cox transformation of the data (λ = 0.3), yielding a distribution of low antibody concentrations (-2.8< x ≤-0.5) that is approximately normal. According to the provider of the assay, samples with (transformed) measurement lower than -0.8 U/ml should be considered uninfected, while samples with measurement greater or equal than -0.8 U/ml should be classified as infected. Right-censoring is applied to the 140 samples above the upper limit of 3.41 U/ml. The data with model fit (see below) are shown in Fig 1.
The data are analyzed statistically using a mixture model with sex- and age-specific mixing functions. We distinguish three distributions, describing samples of low (susceptible, S), intermediate (latently infected, L), and high (latently infected with increased antibodies, B) antibody concentrations. The L and B distributions are modeled using normal distributions with means and standard deviations independent of age and sex. The S distribution is modeled by a mixture of a spike and a normal distribution (an inflated normal distribution), as there appears a spike at -2.91 U/ml in the data (263 persons). In this way, samples with concentration at the spike belong to the susceptible component with probability 1.
We model the probability of each of the three outcomes in terms of log-odds, taking the probability of being in the S component as reference. This allows us to write the log-odds of being in component L or B as linear functions of age and sex. The design matrix of the resulting multinomial logistic model consists of natural cubic splines with interior knots at 20, 40 and 60 years and boundary knots at 0 and 80 years. Hence, the mixing functions (prevalences) have flexible shape, which allows these to be optimally informed by the data. In the results, sex is put in the model as main effect, as analyses show no improvement in fit when including age by sex interaction.
We estimate parameters in a Bayesian framework using R and JAGS [28, 29]. Non-informative normal prior distributions are set on the means of the three component distributions (N ( 0 , 0 . 001 )) (mean and precision). Label switching is prevented by prior ordering of the means. The precisions of the components are given flat Gamma prior distributions (Γ(0.5, 0.005)). The spline parameters are also given non-informative normal prior distributions (N ( 0 , 0 . 001 )). We apply a QR-decomposition to the design matrix to improve mixing and run 10 MCMC chains in parallel, yielding a total of 10,000 samples. We apply an 1/10 thinning to give a well-mixed 1,000 samples from the posterior distribution.
Next to the mixture model analyses, we estimate parameters of transmission models to investigate the ability of different transmission hypotheses explaining the data. To facilitate comparison between transmission models, take the medians of the estimated mixture distributions as input. In line with the above, we focus on a sex- and age-structured model in which persons are probabilistically classified as uninfected (S), latently infected (L), and latently infected after reactivation or re-infection (B). As the infectious period is short relative to the lifespan of the host (weeks versus decades), the infectious periods are modeled implicitly using the short-disease approximation [30]. Further, we focus on the endemic equilibrium of the transmission model so that all variables are time-independent [30, 31]. Fig 2 shows a schematic of the model. For sexes i ∈ {♀, ♂}, the differential equations for the age-specific relative frequencies S(a), L(a), and B(a) (S(a) + L(a) + B(a) = 1) are given by
d S i ( a ) d a = - λ i ( a ) S i ( a ) d L i ( a ) d a = λ i ( a ) S i ( a ) - ( ρ i ( a ) + z λ i ( a ) ) L i ( a ) d B i ( a ) d a = ( ρ i ( a ) + z λ i ( a ) ) L i ( a ) , (1)
with forces of infection
λ i ( a ) = ∑ j ∈ { ♀ , ♂ } ∫ 0 M c i j ( a , a ′ ) ( β 1 λ j ( a ′ ) S j ( a ′ ) + β 2 ( ρ j ( a ′ ) + z λ j ( a ′ ) ) L j ( a ′ ) ) d a ′ . (2)
In Eqs (1) and (2), zλj(a) and ρj(a) are the age-specific re-infection and reactivation rates, z is the susceptibility to re-infection of latently infected persons relative to the susceptibility of uninfected persons (0 ≤ z ≤ 1), cij(a, a′) represents the contact rate between persons of age a′ and sex j, and those of age a and sex i [32, 33], β1 and β2 are proportionality parameters determining the transmissibility of primary infection and reactivation/re-infection, and M is the maximum age. As the data do not extend beyond 80 years we take M = 80 years. Notice that λj(a)Sj(a) and (ρj(a) + z λj(a))Lj(a) are the incidence of primary infection and the incidence of reactivation and re-infection, so that β1λj(a)Sj(a) and β2(ρj(a) + z λj(a))Lj(a) are the infectious output generated by primary infection and reactivation/re-infection, respectively [30].
As in earlier studies, contact rates are hard-wired into the model using data on social contact patterns, thereby adopting the social contact hypothesis [32–34]. Here we use the mixing matrix based on reported physical contacts [32]. The discretized contact function and demographic data are available at github.com/mvboven/cmv-serology.
Below, we consider a suite of simplifications and variations of the full model specified by Eqs (1) and (2). In the simplifications, we assume that (i) there is no re-infection (z = 0), (ii) there is no reactivation (ρi(0) = 0), or (iii) reactivation and re-infection are not infectious (β2 = 0). We also consider a variation of the model in which re-infection and reactivation do not only occur upon transition from L to B, but also in the B compartment. In these models the infectious output generated by reactivation and re-infection in Eq (2) (β2(ρj(a′) + zλj(a′))Lj(a′)) is replaced by β2(ρj(a′) + zλj(a′))(Lj(a′) + Bj(a′)).
The differential equations can be solved in terms of the forces of infection using the variation of constants method. Here we assume, based on results of the mixture model, that a non-negligible fraction of infants is infected in the first six months of life and the fraction infected is equal in female and male infants [21]. Hence, we have S♀(0) = S♂(0) = S0, L♂(0) = L♀(0) = 1 − S0, and B♀(0) = B♂(0) = 0 as initial conditions, and the solution of (1) is given by
S i ( a ) = S 0 i e − ∫ 0 a λ ( τ ) d τ L i ( a ) = ( 1 − S 0 i ) e − ∫ 0 a ρ ( τ ) + z λ ( τ ) d τ + S 0 i ∫ 0 a λ ( τ ) e − ∫ 0 τ λ ( τ ′ ) d τ ′ − ∫ τ a ρ ( τ ′ ) + z λ ( τ ′ ) d τ ′ d τ . (3)
Insertion of Eq (3) in Eq (2) yields two integral equations for the age-specific forces of infection in females and males [34–37]. These equations cannot be solved explicitly in general. It is possible, however, to solve the equations for specific functions.
Here, we assume that reactivation and contact rates are constant on certain predefined age-intervals. From Eq (2), it then follows that the force of infection is piecewise constant as well. Throughout, we consider age intervals of fixed size Δa = 5 years, so that the limits of the n = M/Δa = 16 age classes are defined by the vector a = (0, Δ a, 2Δ a, …, nΔ a). Hence, the j-th class (j = 1, …, n) contains all persons with age in the interval [a[j], a[j + 1]), where a[j] denotes the j-th element of a. Subsequently, the forces of infection λi(a) and reactivation rates ρi(a) are replaced by their counterparts λ j i and ρ j i. Similarly, Si(a), Li(a), and Bi(a) at the borders of the age-intervals are given by S j i, L j i, and B j i. Insertion in Eq (3) and integrating over the (constant) rates yields
S j i = S 0 e - Δ a ∑ k = 1 j λ k i L j i = ( 1 - S 0 ) e - Δ a ∑ k = 1 j ρ k i + z λ k i + S 0 ∑ k = 1 j λ k i e - Δ a ( ρ k i + z λ k i ) - e - Δ a λ k i ( 1 - z ) λ k i - ρ k i e - Δ a ( ∑ ℓ = 1 k - 1 λ ℓ i - ∑ ℓ = k + 1 j ρ ℓ i + z λ ℓ i ) , (4)
where i ∈ {♀, ♂} and B j i = 1 - S j i - L j i. Insertion of Eq (4) in Eq (2) and making use of the fact that the cumulative incidences of infection and reactivation/re-infection in age class j are given by ∫ a [ j ] a [ j + 1 ] λ i ( a ) S i ( a ) d a = S i ( a [ j ] ) - S i ( a [ j + 1 ] ) and Bi(a[j + 1]) − Bi(a[j]), yields 32 equations (16 per sex) for the 32 forces of infection.
As in the mixture model with spline mixing parameters, the log-likelihood of each observation is given by a mixture distribution, where the spline functions are replaced by Si(a), Li(a), and Bi(a). For instance, the likelihood contribution of a sample with antibody measurement c in a person of sex i and age a is given by
S i ( a ) f S ( c ) + L i ( a ) f L ( c ) + B i ( a ) f B ( c ) ,
where Si(a), Li(a), and Bi(a) are the age specific prevalences in sex i, and fS(c), fL(c), and fB(c) are the densities of the mixture distributions at antibody concentration c.
In both sexes, reactivation rates are modeled by piecewise constant functions with steps at 20 and 50 years, i.e. with rates that are constant on the intervals [0, 20), [20, 50), and [50, 80) years. Hence, the reactivation rates are characterized by three parameters in each sex, viz. ρ [ 0 , 20 ) i, ρ [ 20 , 50 ) i, and ρ [ 50 , 80 ) i (i ∈ {♀, ♂}).
Bayesian parameter estimates are obtained using Markov chain Monte Carlo (MCMC). Initially, results were obtained using tailored Mathematica code, using a single-component random walk metropolis algorithm while solving the consistency equations for the forces of infection using a Quasi-Newton (secant) method. As this became exceedingly slow for specific models, we recoded the models using Hamiltonian Monte Carlo with Stan (mc-stan.org). Here, the discretized equations for the forces of infection (2) are solved by specifying that the differences between the left- and right-hand sides are small, and approximately N ( 0 , 10 - 4 ) (mean and scale) distributed. Cross-checking of the two methods yielded very similar results. All programs are available at github.com/mvboven/cmv-serology.
Prior distributions of the parameters are as follows: β 1 ∼ N ( 0 . 1 , 10 ) (mean and scale), β 2 ∼ N ( 0 . 1 , 10 ), z ∼ U ( 0 , 1 ), μ ρ ∼ N ( 0 , 10 ), 1 / σ ρ ∼ N ( 0 , 10 ), and ρ x i ∼ N ( μ ρ , σ ρ ) for all i and x. Whenever applicable, distributions are truncated to be positive. With these prior parameter distributions, the joint posterior distribution is strongly dominated by the data. Ten chains of 3,000 iterations are run in parallel, of which the first 500 iterations (warmup) are discarded. We apply 1/5 thinning, yielding a total of 5,000 samples per model scenario. For all parameters, effective sample sizes usually lie between 3,000 and 4,500. Convergence of chains is assessed visually, and by assessment of the empirical variance within and between chains [38]. To prevent the occurrence of divergent transitions we set ADAPT_DELTA = 0.99. Parameter estimates and bounds of credible intervals are represented by 2.5, 50, and 97.5 percentiles of the posterior samples. Results are usually obtained in 1-3 hours on a personal computer.
Model selection is based on WAIC, a measure for predictive performance, and WBIC, a measure for identifying the most likely model generating the data [39–41]. WAIC is obtained directly from the posterior likelihood using the R-package loo (cran.r-project.org). WBIC is calculated in a separate run as the average log likelihood over the posterior samples, using a sampling ‘temperature’ determined by the number of observations [39].
Fig 1 presents the data stratified by sex and age, with fit of the statistical model. The data and model fit show peaks at low antibody measurements (-2.9 U/ml and ≈-2 U/ml), corresponding to uninfected persons (denoted by S). In both sexes, there is a third peak at higher measurements (1-3 U/ml) that shifts to higher values with increasing age. This peak is composed of persons who are infected (denoted by L) and persons who are infected with high antibody concentrations (denoted by B). Overall, the model appears to describe the data well.
This is confirmed in Fig 3, which shows the estimated components of the mixture distribution and diagnostic characteristics of the classification. The component distribution of uninfected persons hardly overlaps with the two component distributions for infected persons, while there is some overlap between the distributions of infected persons. This can be made more precise using detection theory. Specifically, in Fig 3 we graph the specificity Sp (the probability of correctly classifying a negative subject) and sensitivity Se (the probability of correctly classifying a positive subject) in a receiver operating characteristic (ROC) graph with antibody concentration specifying a cut-off for binary classification as parameter [42–44]. Subsequently, we use the maximal Youden index (i.e. max(Se + Sp − 1)) to choose an optimal cut-off, and find that classification of persons as uninfected versus infected is near perfect (Youden index: 0.97, at cut-off -0.70 U/ml), while classification of persons with high antibody concentrations is good (Youden index: 0.71, at cut-off 1.81 U/ml). These results show that the classification is supported by the data (i.e. has high probability yielding an informed decision).
We further investigate whether mixture models with fewer or more components are able to provide an even better description of the data, and found that a model with two mixture components does not perform well (ΔWAIC = 300.2 in favor of the three-component mixture distribution), while performance of models with four components depends sensitively on choice of prior distribution of the fourth distribution, and often yields broad posterior antibody distributions with small estimated prevalence that overlap with the other three component distributions. Hence, a mixture model with three components gives an optimal description of the data.
Fig 4 shows the estimated prevalences in females and males as a function of age [42–44]. The prevalence of uninfected persons decreases gradually with age, from approximately 0.80 in infants (females: 0.81, 95%CrI: 0.77-0.85; males: 0.80, 95%CrI: 0.76-0.84) to 0.27 (95%CrI: 0.22-0.34) and 0.38 (95%CrI: 0.32-0.45) at 80 years in females and males, respectively. In both females and males the latently infected prevalence remains approximately constant, ranging from 0.15 to 0.20 in females and from 0.18 to 0.28 in males. In contrast, the prevalence of persons with increased antibodies increases strongly with age, especially in females. In fact, the prevalence of persons with increased antibodies increases from 0.09 (95%CrI: 0.06-0.13) at 20 years to 0.57 (95%CrI: 0.47-0.67) at 80 years in females, and from 0.04 (95%CrI: 0.03-0.07) to 0.37 (95%CrI: 0.28-0.46) in males. Hence, in older persons the prevalence of persons with increased antibodies is 54% (or 20 per cent points) higher in females than in males.
Of particular interest is the prevalence of infection in females of childbearing age, as this group is at risk of transmission to the fetus or newborn. Using the above analyses, we find that the prevalence of infection (i.e. the combined prevalence in the L and B compartments) is 0.30 (95%CrI: 0.27-0.33) in 20-year-old females and 0.42 (95%CrI: 0.39-0.46) in 40-year-old females. If we combine these figures with the observation that approximately 20% of children are infected at six months of age, and that less than 5% of children in the Netherlands in 2007 had a mother under 20 years or over 40 years, we deduce that the probability of perinatal transmission could be between 0.20/0.42 = 0.48 and 0.20/0.30 = 0.67, with the exact figure depending on the distribution of ages at which mothers give birth. In addition, one could envisage that the highest risk of (severe) infection of the fetus or newborn is when mothers are infected or experience a reactivation episode. The estimated rates at which susceptible females of 20 and 40 years are infected are 0.0055 per year (95%CrI: 0.0036-0.0077) and 0.0092 per year (95%CrI: 0.0069-0.011) per year, respectively. The rates at which latently infected females of 20 and 40 years are re-infected or experience a reactivation episode are of similar magnitude, and are estimated at 0.0059 per year (95%CrI: 0.0038-0.0086) and 0.0093 per year (95%CrI: 0.0064-0.012), respectively. The overall rates of infection, reactivation, and re-infection in 20 and 40 year-old females are given by the sum of the above estimates, and are approximately 1% and 2% per year, respectively.
To evaluate the ability of different transmission hypotheses explaining the data, and to obtain parameter estimates that have a biological interpretation, we analyzed the data with transmission models. A comparison of model scenarios based on the information criteria WAIC and WBIC is given in Table 1. Overall, the analyses show that models with the possibility of multiple infectious reactivations perform best (Models E and F; lowest WAIC and WBIC), that models with at most one infectious reactivation perform worse (Models A and B; ΔWAIC and ΔWBIC ≈10 − 15), and that models without reactivation or with reactivation not being infectious have very low support (Models C, D, and G). These results indicate that infectious reactivation is key to adequately explain the data with transmission models. This is true in our model with contact structure based on reported physical contacts [32], and also in an alternative model formulation that assumes a uniform contact structure (ΔWAIC = 151.9 in favor of the model with reactivation over the model without reactivation and no re-infection).
Within the set of models with infectious reactivation there are only small differences between models that do and do not incorporate re-infection (Model A versus Model B, and Model E versus Model F). This indicates that while infectious reactivation is essential to adequately describe the data, the analyses are inconclusive with respect to whether or not infectious re-infection should be included.
Fig 5 and Table 2 show parameter estimates of the model with highest statistical support (as judged by WBIC). The preferred model (Model E) includes multiple reactivations and re-infections, infectious reactivation, and infectious re-infection. In this model, the estimated transmissibility of primary infection (β1) is much lower than the transmissibility of reactivation/re-infection (β2). In fact, the posterior median of β2 is more than an order of magnitude larger than the posterior median of β1. Further, the relative susceptibility to re-infection (i.e. the probability of re-infection in a contact that would lead to infection if the contacted person were uninfected) has a broad posterior distribution, and cannot be estimated with meaningful precision from the data (z ^ = 0.32 ; 95%CrI: 0.017-0.84). Similar findings are obtained in other model scenarios, in particular Models A-B and E-F (Table 1).
Estimates of the reactivation rates are quantitatively close in models with high support (Models E-F). Reactivation rates generally increase with increasing age, and are substantially higher in females than in males. In the preferred model (Model E), the estimated reactivation rate is 0.013 per year (95%CrI: 0.0042-0.021) in 0-20 year-old females, which increases to 0.021 per year (95%CrI: 0.013-0.029) in 20-50 year-old females, and then increases further to 0.028 per year (95%CrI: 0.017-0.040) in 50 + -year-old females (Table 2). The corresponding reactivation rates in males are 0.0054 per year (95%CrI: 0.0035-0.013), 0.011 per year (95%CrI: 0.0035-0.018), and 0.013 per year (95%CrI: 0.0043-0.021). These estimates are slightly higher and slightly more precise in the model without re-infection (Model F), and somewhat higher in models with a single reactivation/re-infection event (Models A-B).
In the two models with highest support (Models E-F), estimates of the force of infection increase from approximately 0.012-0.013 per year in the youngest age group to 0.014-0.017 per year in 10-15 year-old girls (Fig 6). Owing to the slightly higher contact rates in females than in men, the estimated force of infection is usually slightly higher in females than in males in the age groups 10-25 years [32]. In older age groups, estimates of the forces of infection decrease to lower values (≈0.01 per year). Noteworthy, the extreme age-specific differences in the force of infection usually observed for directly transmitted infectious diseases, with high infection rates in children and much lower rates in adults, are much less pronounced here due to infectious reactivation in older age strata combined with age-assortative mixing [32, 34, 35].
In models with re-infection, estimates of re-infection rate (zλi(a)) are considerably smaller than estimates of the reactivation rates (ρi(a)) because the estimated forces of infection (λi(a)) are usually lower than the reactivation rates, especially in females (Fig 6). Hence, re-infection contributes little to boosting of the antibody concentrations in those age groups where most of the boosting occurs (>20 years; Fig 4). In fact, in adult females it is not uncommon that the reactivation rate is more than an order of magnitude higher than the estimated re-infection rate (log10(ρ♀(a)/(zλ♀(a))) > 1).
Our study of population-wide serological data shows that IgG antibody concentrations contain a wealth of information on the transmission dynamics of CMV. Specifically, the analyses reveal that (i) the prevalence of CMV increases gradually with age such that at old age the majority of persons in the Netherlands are infected; (ii) except for the very young, the prevalence of CMV is systematically higher in females than in males. This is mainly due to a higher incidence of infection in adult women than in adult men of similar age; (iii) antibody concentrations in seropositive (i.e. infected) persons increase monotonically with age, especially in women; (iv) the above findings (i)-(iii) cannot be explained by simple transmission models in which only primary infection is infectious. This is caused by the fact that transmissibility of primary infection determines the rate at which age-specific prevalence increases; if transmissibility of primary infection would be high then a high prevalence of infection is expected in children. In other words, the fact that seroprevalence increases gradually with age puts an upper bound on the force of infection, and this in turn constrains the transmissibility of primary infection to low values.
While aforementioned findings (i)-(iii) have been noticed before in other settings ([1] and references therein, [21]), our analyses are the first to provide precise estimates using a large population sample. Moreover, the results lead us to a new transmission hypothesis in which infectious reactivation is a key driver of transmission of CMV in the population. Since several other studies have found a gradual increase in seroprevalence [1], this explanation may not be restricted to the Dutch situation, but hold in general. Underpinning this hypothesis, next to the well-known observations of shedding of CMV in breast milk and cervical material in the third trimester of pregnancy [45–47], detectable virus also has been found in healthy adults in one study [24], while in another study CMV DNA has been detected in urine of the majority of older persons [23].
The main implication is that the majority of CMV infections may not be caused by transmission among children after primary infection, even though levels of shedding can be high in infants [46, 48], but rather by older persons who go through one or more reactivation episodes. This contrasts with common childhood diseases such as measles, mumps, rubella, and pertussis. For these pathogens, infection in unvaccinated populations generally occurs at a young age, and children are the drivers of transmission. It also contrasts with other herpes viruses such as varicella zoster virus and Epstein-Bar virus for which well over 50% of the population is infected at the age of 10 years [34]. It may be comparable with other herpes viruses such as HSV1 and HSV2, which show a slowly increasing age-specific seroprevalence [49]. A corollary is that persistence of CMV in the population is not possible with transmission from primary infected persons only, and is dependent on infectious reactivation. Currently, we are focusing on making this idea more precise by calculation of the basic reproduction number, and the reproduction numbers of perinatal transmission, primary infection, and reactivation [50]. This will help put bounds on the relative contribution of each of the transmission routes.
With infectious reactivation and perinatal infection being putative drivers of transmission, it is to be expected that elimination by vaccination may prove more difficult than for directly transmitted pathogens, as it will require the pool of latently infected persons to dwindle to zero by demographic turnover. This can take up to the lifetime of one generation, and perhaps more if vaccination cannot prevent perinatal transmission to infants who are too young for vaccination. Thus, a question is whether vaccination formulations and strategies exist that minimize the probability of transmission to young infants. This is all the more of importance as a main source of morbidity is by congenital infection, and the timescale on which reductions in congenital disease are expected determines the projected health impact of vaccination [51]. In this context, next to the ability of a vaccine to prevent infection it may perhaps be equally important that a vaccine is able to reduce the probability of reactivation. Such reductions are likely mediated by T-cell responses of the host, and several (but not all) vaccines under development are expected to induce boosting of T-cell immune responses [52–54].
A number of limitations and assumptions deserve scrutiny. First, the transmission model analyses assume that the population is in endemic equilibrium. For a single cross-sectional data set such as the one considered in the present study this assumption is unavoidable if one does not want to introduce additional parameters that cannot be estimated by the data. Reassuringly, the patterns of infection present in the serological data have been found in several serological studies carried out in high-income countries over the past decades [1]. Also, no systematic patterns of increasing or decreasing seroprevalence over time have been found, and this is further reason to believe that there have not been major changes in the epidemiology of CMV over time [1]. Second, we assume that antibody measurements not only give information on CMV infection status, but also whether or not reactivation or re-infection have taken place. Unfortunately, there is no direct empirical evidence confirming or falsifying this assumption, and this is an area where in-depth comparison of the infection and immune status of persons with low and high antibody concentrations is urgently needed. Third, the analyses assume that person-to-person transmission is proportional to observed human contact patterns [32, 33]. Although this assumption is commonly made and has met with considerable success (e.g., [33, 44, 55, 56]), it is conceivable that transmission of CMV does not abide by the social contact hypothesis, and that a more complex contact structure would be able to explain the patterns of seroprevalence in a simple transmission model. To investigate the impact of the contact structure, we have analyzed transmission models with a uniform contact structure, and found that models with infectious reactivation still provide the best fit to the data (ΔWAIC > 100; Results). As a final limitation we would like to add that, in principle, it is conceivable that the data can be explained alternatively by an intricate interplay between variation in the susceptibility to infection in conjunction with age-specific variations in the strength of the antibody response. Alas, evidence for or against this hypothesis is lacking.
Our inferential analyses indicate that the transmissibility of primary infection is much lower than the transmissibility after reactivation. This seems to be at odds with the observation that prolonged and high-level virus shedding can occur in bodily fluids after primary infection in children [46, 47]. However, it could be that transitions from the infected class to the infected class with increased antibodies are in effect not the result of a single reactivation or re-infection event, but rather the result of multiple underlying reactivations or re-infections. If this were true, as seems plausible, estimates of the reactivation and re-infection rates as well as the transmissibility of reactivation and re-infection should be interpreted as compound parameters that take into account multiple reactivations and re-infections occurring over the lifetime of an infected person.
|
10.1371/journal.ppat.1003623 | The CLIP-Domain Serine Protease Homolog SPCLIP1 Regulates Complement Recruitment to Microbial Surfaces in the Malaria Mosquito Anopheles gambiae | The complement C3-like protein TEP1 of the mosquito Anopheles gambiae is required for defense against malaria parasites and bacteria. Two forms of TEP1 are present in the mosquito hemolymph, the full-length TEP1-F and the proteolytically processed TEP1cut that is part of a complex including the leucine-rich repeat proteins LRIM1 and APL1C. Here we show that the non-catalytic serine protease SPCLIP1 is a key regulator of the complement-like pathway. SPCLIP1 is required for accumulation of TEP1 on microbial surfaces, a reaction that leads to lysis of malaria parasites or triggers activation of a cascade culminating with melanization of malaria parasites and bacteria. We also demonstrate that the two forms of TEP1 have distinct roles in the complement-like pathway and provide the first evidence for a complement convertase-like cascade in insects analogous to that in vertebrates. Our findings establish that core principles of complement activation are conserved throughout the evolution of animals.
| Mosquitoes are vectors of numerous human diseases including malaria. Disease transmission requires that microbes overcome the robust mosquito immune system. In the African malaria mosquito, the TEP1 protein that is homologous to mammalian complement factor C3 is shown to play a central role in mosquito immunity to malaria parasites and bacteria. In this study, we report that another mosquito protein belonging to a class of non-catalytic enzymes that are specific to arthropods is a core component of the mosquito complement-like immune pathway. We found that this new protein, named SPCLIP1, regulates the accumulation of TEP1 on malaria parasites and bacteria, and show that this can lead to distinct defense reactions including lysis and melanization of the pathogen. This work is valuable because it reveals novel insight into the regulation of mosquito complement on microbial surfaces such as those of the malaria parasites. Unraveling the molecular mechanisms regulating these defense responses may ultimately lead to the design of novel disease blocking strategies in the vector.
| The mosquito Anopheles gambiae is the main vector of Plasmodium falciparum malaria in sub-Saharan Africa and hence directly responsible for the death of hundreds of thousands of people every year and for a devastating socioeconomic burden especially in endemic countries. Mosquitoes launch a potent immune attack leading to the killing of the majority of invading Plasmodium parasites. Multiple mechanisms are thought to participate in these anti-Plasmodium reactions, amongst them a latent pathway resembling vertebrate complement [1]. RNAi knockdown (kd) studies, based on the injection of double stranded RNA (dsRNA) into adult A. gambiae mosquitoes, have revealed important roles of components of the complement-like pathway in defense against the murine malaria parasite Plasmodium berghei [2]–[6]. There is also significant evidence for a role of this pathway in defense against the human parasite, P. falciparum, in laboratory infections of A. gambiae [3], [7]–[9].
Recent studies with natural A. gambiae populations revealed that the gene encoding the C3-like protein TEP1, a key player of the complement-like pathway, and the genomic locus encoding its interacting partner APL1C are under strong directional selection in an M form population but subject to balancing selection in another S form population [10], [11]. Despite the fact that distinct TEP1 alleles have been associated with resistance to Plasmodium parasites [2], [8], [11]–[13], the selective pressure on TEP1 is hypothesized to be driven by pathogens in larval habitats rather than those encountered by adults. This is further supported by the rather generic immune specificity of TEP1 that functions also in anti-bacterial [3], [14] and anti-fungal defense [15]. The polymorphic nature of TEP1 also suggests that the different alleles might follow different kinetics in interacting with LRIM1/APL1C as well as other TEP1 regulatory proteins, which could influence the efficiency of parasite killing or microbial clearance. Therefore, a better understanding of the mechanisms regulating complement activation and identification of the proteins involved will permit deciphering the functional relevance to Plasmodium of allelic interactions within this immune module on resistance.
The hallmark of activation of the mosquito complement-like pathway is the binding of TEP1 to microbial surfaces through a thioester bond, a reaction that is tightly linked to microbial killing [14]. TEP1 circulates in the mosquito hemolymph in two forms: the full-length form TEP1-F and the proteolytically processed form TEP1cut, corresponding to pro-C3 and the mature C3 protein after processing in the ER, respectively [14], [16]. Unlike C3, TEP1 lacks an anaphylatoxin domain and the exposed thioester bond of TEP1cut is unstable [17]. TEP1cut is stabilized in the hemolymph through interactions with a heterodimer of the leucine-rich repeat (LRR) proteins LRIM1 and APL1C, which seems to confer specificity upon TEP1 activity [5], [16]. While the structure and function of TEP1 and its C3 homolog are largely conserved from insects to mammals, LRIM1 and APL1C are thought to be specific to mosquitoes [18] raising interesting questions about the degree of structural and/or functional conservation between other modules of the complement pathway such as those that stabilize or amplify complement on microbial surfaces. The research presented here aimed to address these questions and provide novel mechanistic insights into the activation of the mosquito complement pathway during infection.
To identify novel components of the mosquito complement pathway, we searched for genes that exhibited significant co-regulation with LRIM1 in a developmental transcriptome dataset of Expressed Sequence Tags (ESTs; [19]). Pearson correlation coefficient (PCC) identified 4 EST clusters showing similarity to LRIM1 developmental expression greater than 0.95. Importantly, 3 of the 4 clusters were found to encode proteins that had been previously shown to physically interact with LRIM1, including APL1C (PCC 0.964), TEP1 (PCC = 0.978) and TEP4 (PCC = 0.965) [5], [16], [20]. The fourth EST cluster (PCC = 0.980) did not correspond to any gene model in the A. gambiae genome. It encodes a protein with CLIP and serine protease domains, previously identified as SPCLIP1 and shown to be involved in defense against P. falciparum, P. berghei, Escherichia coli and Staphylococcus aureus [3]. SPCLIP1 maps within a genomic region encompassing 12 additional genes encoding proteins with CLIP and serine protease domains (Figure S1A). All residues corresponding to the serine protease catalytic triad (Asp-His-Ser) are substituted in SPCLIP1 indicating that it is non-catalytic (Figure S1B). Phylogenetic analysis places SPCLIP1 in the highly divergent CLIPE subfamily of non-catalytic CLIP-domain serine protease homologs (SPHs; Figure S1C).
Co-regulation with LRIM1 and the previously reported knockdown phenotypes [3] were suggestive of SPLCLIP1 involvement in the A. gambiae complement-like pathway. To characterize SPCLIP1, we raised a polyclonal antibody against the entire protein and used it in western blots of adult mosquito hemolymph separated by non-reducing SDS-PAGE. The results showed that SPCLIP1 migrates at approximately 45 kDa, near its predicted 42 kDa molecular weight (Figure 1A). We examined whether the steady state levels of SPCLIP1 in the hemolymph are affected by silencing LRIM1 or TEP1. While TEP1 kd had no effect on SPCLIP1 levels, these were markedly reduced in LRIM1 kd (Figure 1B). This decrease of SPCLIP1 parallels the near complete loss of TEP1cut from the hemolymph of LRIM1 or APL1C kd mosquitoes due to its accumulation on self-tissues (Figure 1B) [5], [16]. To determine if the reduction of SPCLIP1 in LRIM1 kd is dependent on TEP1, we silenced LRIM1 and TEP1 simultaneously. Under these conditions, SPCLIP1 was restored to its baseline levels (Figure 1C). In contrast, silencing LRIM1 and SPCLIP1 together did not restore TEP1cut levels, suggesting that SPCLIP1 functions downstream of TEP1cut, and that in LRIM1 kd mosquitoes SPCLIP1 is likely to be sequestered on self-tissues together with TEP1cut.
We investigated the role of SPCLIP1 in TEP1 binding to P. berghei. It has been previously established that TEP1 binds to the surface of P. berghei ookinetes as they traverse the mosquito midgut epithelium and come into contact with the hemolymph [2]. In SPCLIP1 kd mosquitoes, TEP1 staining on the ookinete surface was inhibited (Figure 2A). This, together with the TEP1-dependent reduction of SPCLIP1 from the hemolymph following LRIM1 kd, led us to hypothesize that SPCLIP1 is recruited to the parasite surface during infection. To test this, SPCLIP1 was immunolocalized in midgut epithelium 26 h after infection. We observed robust SPCLIP1 signal on dead ookinetes, judged by the loss of their cytoplasmic GFP signal (Figure 2B). Given that TEP1 is also highly prevalent on dead ookinetes, this result indicates that SPCLIP1 and TEP1 likely co-localize to the same ookinetes; however, we could not simultaneously assay their distribution since both antibodies were raised in the same host species. No SPCLIP1 staining was observed in midgut epithelia dissected from SPCLIP1 kd mosquitoes, showing that the antibody is specific. Importantly, SPCLIP1 staining on the ookinete surface was inhibited after TEP1 kd. This suggests that the localization of TEP1 and SPCLIP1 to ookinetes is mutually dependent (Figure 2B).
TEP1 present on microbial surfaces during infection may originate either from the TEP1cut or the TEP1-F pools. To clarify this point and investigate further the functional relationship between the two forms of TEP1 and SPCLIP1, we developed an alternative infection model that allowed us to monitor temporally and quantitatively the dynamics of the examined proteins after injection of E. coli bioparticles (chemically killed bacteria) into the hemocoel. This infection model offers the advantage of tight temporal monitoring of rapid immune responses such as those of complement, which occur within minutes of microbial exposure to the hemolymph. Hemolymph was collected from groups of mosquitoes at 15, 60, 120, 240 and 360 minutes post injection with bacteria or PBS (i.e. control) and proteins were analyzed by western blot. The results showed strong reduction in SPCLIP1, the LRIM1/APL1C complex, and TEP1-F levels in mosquito hemolymph after injection of E. coli bioparticles (Figure 3A). A marked reduction of these proteins was already observed at 60 min after injection and persisted up to 240 min when LRIM1/APL1C and TEP1-F levels began to rise. The kinetics of TEP1-F reduction demonstrate that this form of TEP1 is consumed quickly in the immune response to infection, in contrast to TEP1cut, which does not seem to vary significantly during that process, at least within the examined timeframe. In addition to the well-defined TEP1-F and TEP1cut bands, we also observed a broadly stained TEP1-specific smear at 50–60 kDa exhibiting depletion kinetics following bioparticle challenge similar to that of TEP1-F (Figure 3A). These C-terminal TEP1 fragments have been previously described [14]; whether they represent functional forms of TEP1 or are products of TEP1-F turnover remains to be determined.
LRIM1/APL1C and SPCLIP1 exhibited similar depletion kinetics as TEP1-F following bioparticle injections (Figure 3A), suggesting that these proteins are either required for TEP1-F utilization or are independently consumed in the immune reactions. To address this, we monitored the effect of SPCLIP1 silencing on the infection-dependent depletion of TEP1-F. Western blot analysis of hemolymph collected from SPCLIP1 and control LacZ kd mosquitoes challenged with E. coli bioparticles demonstrated that the loss of TEP1-F is abolished in SPCLIP1 kd mosquitoes compared to controls (Figure 3B), indicating that SPCLIP1 acts upstream of TEP1-F and is indeed required for the infection-induced loss of this protein. In contrast, the depletion of LRIM1/APL1C was not restored in the hemolymph of SPCLIP1 kd mosquitoes. Together, these data suggest that activation of mosquito complement by the LRIM1/APL1C/TEP1cut complex is a separate event upstream of the SPCLIP1-dependent complement amplification process that is poised to transform initial pathogen recognition into a robust attack.
An important aspect of the complement system is its specific activation on microbial surfaces. In order to address whether the observed reduction in SPCLIP1 and TEP1-F levels in the hemolymph after injection of E. coli bioparticles is due to their sequestration on bioparticle surfaces, we designed an assay that allows quantitative assessment of E. coli-bound versus hemolymph soluble pools of these proteins. E. coli bioparticles were injected into mosquito hemocoel, and hemolymph was extracted 15 min after injection. Bioparticles were separated from the hemolymph by centrifugation, washed extensively and their surface-bound proteins eluted for western blot analysis (Figure 4A). The results showed that SPCLIP1 was present in the E. coli-bound fraction in dsLacZ control mosquitoes (Figure 4B), which explains its reduced levels in the hemolymph after bacterial challenge and is consistent with its localization to ookinetes. In TEP1 kd mosquitoes, SPCLIP1 was lost from the E. coli-bound fraction and became enriched in the soluble fraction, indicating that TEP1 is required for SPCLIP1 recruitment to bacterial surfaces.
This assay also allowed us to monitor which of the two forms of TEP1 associates with the bacterial surface. In dsLacZ treated mosquitoes, TEP1-F was not detected in the E. coli-bound fraction, despite being almost fully depleted from the soluble material, in contrast to TEP1cut, which was clearly present. These data are consistent with those reported previously using a cell culture assay and showing that bacteria only bound TEP1cut when incubated with the conditioned medium of a hemocyte-like cell line that contained both forms of TEP1 [14]. Importantly, TEP1cut signal in the bound material was dramatically reduced by SPCLIP1 kd, concomitant with the detection of TEP1-F in the soluble fraction. These data indicate that TEP1cut accumulating on the surface of E. coli is generated from TEP1-F and that its conversion requires recruitment of SPCLIP1 and a yet unidentified protease to the bacterial surface.
The functional association between SPCLIP1 and TEP1 including their cooperative recruitment to microbial surfaces suggested that these two proteins might physically interact. To examine this possibility, we performed an immunoprecipitation (IP) assay on hemolymph samples collected from mosquitoes following challenge with E. coli bioparticles using beads cross-linked to an affinity purified SPCLIP1 antibody. IP beads lacking antibody and mock bioparticle challenge (PBS injection) served as controls. The results revealed that SPCLIP1 was less abundant in the unbound fraction and significantly enriched in the bound fraction (Figure 5). In contrast, SPCLIP1 was not detected on control beads and the protein remained highly abundant in the unbound fraction. When samples were probed for TEP1, a signal for TEP1cut and a faint but clear TEP1-F signal were observed in the SPCLIP1 IP bound fraction. These bands were detectable only in samples collected from bioparticle challenged mosquitoes. These data indicate that SPCLIP1 and TEP1 can interact and that this interaction is induced by infection. These data raise the possibility that these proteins interact first in the hemolymph prior to their localization on microbial surfaces. Alternatively, membrane-bound complexes containing TEP1 and SPCLIP1 may leach off the surface during sample preparation. Whether this interaction is direct or mediated by another factor remains to be determined.
It has been previously shown that bacterial inoculation into the mosquito hemolymph leads to rapid activation cleavage of CLIPA8, a key SPH regulator of bacteria [21] fungi [15], and Plasmodium melanization [22]. We examined whether SPCLIP1 is required for CLIPA8 activation in the mosquito hemolymph following E. coli bioparticle injection. As shown in Figure 6A, silencing SPCLIP1 inhibited completely CLIPA8 cleavage, suggesting that SPCLIP1 is required for activation of the melanization cascade.
The final steps of melanization are catalyzed by phenoloxidase (PO) which is secreted as a pro-enzyme (PPO) and activated by proteolytic cleavage in response to infection. We directly examined whether SPCLIP1 is essential for PPO activation by monitoring PO activity in the mosquito hemolymph after bacterial injection. Indeed, SPCLIP1 kd resulted in a strong decrease in PO activity relative to dsLacZ-injected controls, which is comparable to that observed in CLIPA8 kd mosquitoes (Figure 6B). Similar to SPCLIP1 kd, silencing TEP1 also resulted in strong inhibition of both CLIPA8 cleavage and PPO activation (Figure S2). These data demonstrate that activation of the melanization cascade is dependent on SPCLIP1-mediated TEP1 accumulation on the bacterial surface.
We next tested the function of SPCLIP1 in P. berghei melanization using as a model CTL4 kd mosquitoes which melanize nearly all ookinetes soon after they traverse the mosquito midgut and before they develop into oocysts [4]. Indeed, silencing CTL4 alone resulted in a marked decrease of the oocyst numbers and a reciprocal increase in melanized ookinetes, but concomitant silencing of SPCLIP1 completely blocked ookinete melanization and led to an increase in oocysts comparable to that of SPCLIP1 kd alone (Figure 6C). A similar inhibition of parasite melanization has been observed after silencing TEP1 or LRIM1/APL1C [2], [4], [5]. These data reveal that, as with bacterial melanization, SPCLIP1-mediated accumulation of TEP1 on the ookinete surface is required for parasite melanization.
Here we characterize SPCLIP1, a non-catalytic CLIP-domain serine protease of the malaria vector mosquito A. gambiae, which localizes to the surface of P. berghei ookinetes and E. coli promoting rapid accumulation of the complement C3-like protein TEP1. Our results demonstrate that SPCLIP1 regulates a complement convertase-like activity henceforth referred to as TEP1 convertase. The TEP1 convertase is functionally analogous to the vertebrate C3 convertase, the formation of which is triggered by binding of antibodies or innate pattern recognition proteins on the microbial surfaces, or by spontaneous activation of C3 following hydrolysis of its thioester. The trigger for the formation of the TEP1 convertase is thought to be the binding on the microbial surface of TEP1cut which circulates in the mosquito hemolymph together with the LRIM1/APL1C complex (Figure 7). LRIM1 and APL1C possess LRR domains, a feature that is versatile in its binding properties and common in pattern recognition receptors involved in host defense in animals and plants [23]. Therefore, the LRIM1/APL1C complex may play a dual role in the mosquito complement-like pathway by stabilizing TEP1cut in the hemolymph and delivering it to the microbial surface upon infection. Given that the LRIM1 and APL1C belong to a mosquito-specific family of LRR proteins [5] whereas TEPs are widely conserved [24], different triggers of complement activity are likely to exist in other insects. A number of different putative pattern recognition receptors have been identified to play a role in TEP1-dependent defense against bacteria and malaria parasites [4], [6], [25]–[27] raising the possibility that mosquitoes may also have multiple recognitions systems that can activate the TEP1 convertase. It has been proposed that nitration of malaria parasites during their passage through the mosquito midgut epithelium is required for TEP1 binding [28]. Whether microbe nitration can trigger recognition by LRIM1/APL1C or other putative recognition receptors remains to be determined.
A study using recombinant proteins and an allele of TEP1 from mosquitoes that are refractory to Plasmodium has shown that the LRIM1/APL1C complex binds TEP1cut lacking an intact thioester, and that TEP1cut precipitates out of solution in the absence of LRIM1/APL1C [17]. A more recent study using a TEP1cut allele from susceptible mosquitoes has revealed that the LRIM1/APL1C complex can interact with TEP1cut with an active thioester [29]. These in vitro studies have led the authors to speculate that a complex between LRIM1/APL1C and TEP1cut may function in vivo as a TEP1 convertase. It remains unknown whether TEP1cut lacks an intact thioester in vivo, and whether its localization on mosquito tissues in the absence of LRIM1/APL1C is the result of protein precipitation or autoimmune attack by an active thioester motif [5], [16]. The TEP1cut dependent SPCLIP1 depletion favors the hypothesis of an autoimmune attack that is tightly regulated to prevent collateral damage to host tissues. Indeed, SPCLIP1 loss from the hemolymph following artificial induction of TEP1cut attack of self-tissues is not accompanied by TEP1-F depletion, suggesting that downstream negative regulators prevent the full formation of the TEP1 convertase and/or that additional positive factors similar to vertebrate properdin may be required to stabilize the convertase on microbial surfaces.
SPCLIP1 lacks catalytic serine protease activity and likely acts as a regulatory component of the TEP1 convertase. Hence, an unidentified protease and possibly other factors are expected to also contribute to the mature convertase, catalyzing the activation cleavage of TEP1-F. The role of non-catalytic serine proteases as cofactors for active proteases is well documented in insects with examples from Holotrichia diomphalia [30], Manduca sexta [31] and Drosophila melanogaster [32].
The SPCLIP1-dependent rapid loss of TEP1-F from the hemolymph of bioparticle injected mosquitoes and the observation that SPCLIP1 kd in naive mosquitoes does not alter TEP1-F levels, suggests that the TEP1cut cargo circulating as a complex with LRIM1/APL1C is generated through a different mechanism than that produced by the TEP1 convertase. Of note, while bioparticle injection almost depletes TEP1-F from the hemolymph, only a minor reduction in TEP1cut levels is observed most significantly at 60 min post injection. A plausible explanation for this observation is that TEP1-F is converted to TEP1cut prior to binding the bacterial surface, a fraction of which remains soluble in the hemolymph throughout the timeframe of the experiment.
Regardless of the activation mechanism, the C3 and TEP1 convertases function in very similar ways to recruit additional C3 and TEP1, respectively, from precursor pools onto the microbial surface, and to initiate diverse effector cascades. In vertebrates, accumulation of the C3 cleavage product, C3b, on microbial surfaces triggers phagocytosis as well as assembly of the membrane attack complex that causes microbial lysis. In mosquitoes, in addition to triggering phagocytosis of bacteria [14], [33] and lysis of malaria parasites [2], [3], TEP1 accumulation on microbial surfaces triggers the PO cascade leading to melanization. Therefore, the strategy of complement driving diverse effector functions is ancient and not specifically co-opted by vertebrates. It remains to be further investigated whether this system is indeed an example of convergent evolution rooted to the functional conservation of thioester-containing proteins, a hypothesis consistent with our earlier findings that this pathway appears to have evolved de novo in each mosquito species by “bricolage” assemblages of the most suitable available components [34].
This study was carried out in strict accordance with the United Kingdom Animals (Scientific Procedures) Act 1986. The protocols for maintenance of mosquitoes by blood feeding and for infection of mosquitoes with P. berghei by blood feeding on parasite-infected mice were approved and carried out under the UK Home Office License PLL70/7185 awarded in 2010. The procedures are of mild to moderate severity and the numbers of animals used are minimized by incorporation of the most economical protocols. Opportunities for reduction, refinement and replacement of animal experiments are constantly monitored and new protocols are implemented following approval by the Imperial College Ethical Review Committee.
A. gambiae G3 strain was maintained and assayed for infection with P. berghei CONGFP strain as described previously [20]. Single and double knockdown experiments and parasite counts in dissected midguts were performed as described previously [5]. Primers used for synthesis of double stranded RNA have been reported elsewhere LRIM1, TEP1, CTL4 [4], [35]; SPCLIP1 [3].
The entire SPCLIP1 open reading frame lacking the endogenous signal peptide and stop codon was cloned into the pIEx10 insect cell expression plasmid (Novagen) incorporating a C-terminal 10× HIS-tag using the primers:
For: GACGACGACAAGATGAACTTCCCCGTTGGGAAATTTC
Rev: GAGGAGAAGCCCGGTTTATCGAAGCTGATCGGATCGGG
The underlined sequences are extensions to allow ligase-independent cloning [5]. Sf9 cells adapted for growth in serum-free medium (Invitrogen) stably secreting SPCLIP1HIS were generated by selection with 1 mg/mL G418 following co-transfection using Escort IV (Sigma) of pIEx10-SPCLIP1HIS and pIE1-neo (Novagen). Clones of resistant cells were analyzed by western blot for the presence of SPCLIP1HIS in their conditioned medium and the line with the highest expression was chosen for protein production. SPCLIP1HIS was purified from 500 mL of conditioned medium using a 1 mL HisTrap column attached to an ÄKTA purifier (GE Healthcare). Bound protein was eluted in 15 mL of PBS containing 500 mM imidazole pH 8.0. Purified SPCLIP1HIS was quantified by Bradford assay and by coomassie staining of SDS-PAGE gels. The purified protein was used to generate a rabbit polyclonal antibody (Eurogentec). SPCLIP1 antibody was affinity purified from the positive immune serum by passage over an AminoLink column (Pierce) containing covalently bound SPCLIP1HIS.
A 20 mg/mL suspension of fluorescein or pHrodo labeled E. coli K-12 strain bacterial bioparticles (Invitrogen) in sterile PBS was injected into the mosquito hemocoel (∼4×105 bacteria in 69 nL). Hemolymph was collected directly into non-reducing SDS-PAGE sample buffer from groups of 30–40 mosquitoes 15, 60, 120, 240 and 360 min after the challenge and analyzed by reducing and non-reducing western as described previously [5]. Bioparticles surface extraction was performed by collecting in protein LoBind tubes (Eppendorf) hemolymph from 60 mosquitoes into 60 µL of 15 mM Tris (pH 8.0) containing 1× protease inhibitor cocktail (complete EDTA free, Roche) 15 min after bacterial injection. The soluble (unbound) fraction was collected after pelleting the bacteria by centrifugation for 4 min at 6000 g at 4°C and then supplemented with SDS-PAGE buffer. The bacterial pellet was washed with 400 µL of 15 mM Tris (pH 8.0) and the bound fraction was extracted with 25 µL SDS-PAGE sample buffer. Western blot analysis was performed using 25 µL of each sample.
Western blot analysis for TEP1, LRIM1/APL1C, SRPN3, CLIPA8 and PPO6 was performed as previously described [5], [21]. The affinity purified rabbit α-SPCLIP1 antibody was used to probe western blots at a 1∶1000 dilution of antibody in PBS containing 0.05% Tween 20 and 3% milk for 1 h at room temperature using. Co-immunoprecipitation reactions were performed using the Pierce Co-IP kit according to the manufacturer's protocol (ThermoScientific). Hemolymph was collected from 100 mosquitoes into 200 µL ice-cold PBS containing 0.05% Triton X-100, supplemented with 1× protease inhibitor cocktail 15 min after PBS or E. coli bioparticle injection (69 nL of 4 mg/mL; ∼8×104 particles). The samples were centrifuged at 4000 g for 5 min to remove mosquito and bacterial cells. 40 µL of a 1∶1 slurry of PBS and agarose beads containing crosslinked affinity purified α-SPCLIP1 antibody or control beads were added to the cleared hemolymph samples and mixed overnight at 4°C on a rotating wheel. The unbound fraction was collected and supplemented with SDS-PAGE buffer. Then the beads were washed five times with collection buffer and bound material was eluted two times with 100 µL of elution buffer (0.2% SDS and 0.1% Tween-20 in 50 mM Tris pH 8.0). The eluents were pooled and supplemented with SDS-PAGE buffer. Western blot analysis was performed by loading 40 µL of each sample. Reducing samples were made by addition of 2-mercaptoethanol to a final concentration of 2.5%.
Cleavage of CLIPA8 was assayed in samples of hemolymph analyzed under reducing conditions as described previously [21]. PPO activation was determined assaying the conversion of L-DOPA to Dopachrome in samples of mosquito hemolymph collected after bacterial challenge [36].
TEP1 and SPCLIP1 were immunolocalization to ookinetes 26 h after P. berghei infection. Mosquito midguts were prepared and analyzed as previously described [5]. The SPCLIP1 antibody was used at a 1∶250 dilution. Images were acquired on a Zeiss LSM 710 META confocal.
LRIM1, AGAP006348; APL1C, AGAP007033; TEP1, AGAP010815; TEP4, AGAP010812; CLIPA1, AGAP011791; CLIPA2, AGAP011790; CLIPA4, AGAP011780; CLIPA5, AGAP011787; CLIPA6, AGAP011789; CLIPA7, AGAP011792; CLIPA8, AGAP010731; CLIPA9, AGAP010968; CLIPA12, AGAP011781; CLIPA13, AGAP011783; CLIPA14, AGAP011788; CLIPB2, AGAP003246; CLIPB3, AGAP003249; CLIPB4, AGAP003250; CLIPB8, AGAP003057; CLIPB9, AGAP013442; CLIPB10, AGAP003058; CLIPB13, AGAP004855; CLIPB14, AGAP010833; CLIPB15, AGAP009844; CLIPC1, AGAP008835; CLIPC2, AGAP004317; CLIPC3, AGAP004318; CLIPC5, AGAP000571; CLIPC6, AGAP000315; CLIPC9, AGAP004719; CLIPC10, AGAP000572; CLIPD4, AGAP002811; CLIPD6, AGAP002813; CLIPD7, AGAP008998; CLIPD8, AGAP002784; CLIPE2, AGAP011782; CLIPE4, AGAP010530; CLIPE5, AGAP010547; CLIPE6, AGAP011785; CLIPE7, AGAP011786; PPO6, AGAP004977; CTL4, AGAP005335; SRPN3, AGAP006910.
|
10.1371/journal.pntd.0005470 | Using simulation to aid trial design: Ring-vaccination trials | The 2014–6 West African Ebola epidemic highlights the need for rigorous, rapid clinical trial methods for vaccines. A challenge for trial design is making sample size calculations based on incidence within the trial, total vaccine effect, and intracluster correlation, when these parameters are uncertain in the presence of indirect effects of vaccination.
We present a stochastic, compartmental model for a ring vaccination trial. After identification of an index case, a ring of contacts is recruited and either vaccinated immediately or after 21 days. The primary outcome of the trial is total vaccine effect, counting cases only from a pre-specified window in which the immediate arm is assumed to be fully protected and the delayed arm is not protected. Simulation results are used to calculate necessary sample size and estimated vaccine effect. Under baseline assumptions about vaccine properties, monthly incidence in unvaccinated rings and trial design, a standard sample-size calculation neglecting dynamic effects estimated that 7,100 participants would be needed to achieve 80% power to detect a difference in attack rate between arms, while incorporating dynamic considerations in the model increased the estimate to 8,900. This approach replaces assumptions about parameters at the ring level with assumptions about disease dynamics and vaccine characteristics at the individual level, so within this framework we were able to describe the sensitivity of the trial power and estimated effect to various parameters. We found that both of these quantities are sensitive to properties of the vaccine, to setting-specific parameters over which investigators have little control, and to parameters that are determined by the study design.
Incorporating simulation into the trial design process can improve robustness of sample size calculations. For this specific trial design, vaccine effectiveness depends on properties of the ring vaccination design and on the measurement window, as well as the epidemiologic setting.
| The urgency, as well as the logistical and sometimes ethical challenges of clinical trials for interventions during epidemics of emerging diseases prompts the need for novel designs and analytic strategies. The successful use of a novel cluster-randomized ring-vaccination trial to test an Ebola vaccine in Guinea raises the general question of what circumstances would favour the use of trials of similar design and how the properties of the population, the vaccine and the trial would influence the necessary sample size and the expected results. We present a generalized transmission dynamic model for a ring vaccination trial to address these questions. This work is an example of the general phenomenon that mechanistic, transmission-dynamic simulations can aid in the design and interpretation of intervention trials for infectious diseases, when the trial itself can have non-obvious effects on transmission dynamics that may not be fully captured by effect- and sample-size calculations for noncommunicable diseases.
| The West African Ebola epidemic highlighted the need to identify a range of trial designs to evaluate vaccine effects rapidly, efficiently and rigorously during emerging disease outbreaks. The ring-vaccination trial approach employed in the Ebola ça suffit trial in Guinea is one innovative approach [1], which produced valuable evidence that the vaccine could prevent Ebola infection [2]. Other approaches considered include individual randomization and a stepped-wedge design [3, 4]. In such trials it is difficult to estimate the likely effect of an infectious disease intervention because of indirect effects, and this issue is compounded by complex trial design. Sample size calculations are based on group-level quantities such as intervention effect and are therefore potentially inaccurate. By creating a transmission dynamic model for a ring vaccination trial, we show that we can make sample size calculations based on disease characteristics and individual intervention efficacy. With this framework in place we are then able to examine the estimated vaccine effect and sample size under a range of assumptions about the properties of the vaccine, the trial, and the study population.
Although the only implementation of the ring trial design has been in Guinea during the Ebola epidemic, lessons can be learned and extended to other diseases and contexts. Here, we examine the tail end of an epidemic of a disease with a latent and asymptomatic phase with effective contact tracing to illustrate a more widely-applicable set of findings. In particular, we use baseline parameters values consistent with Ebola in West Africa in 2014–6, but we vary several assumptions over broader ranges than those occurring in the Ebola ça suffit trial, with the aim of being relevant to a range of potential future situations.
The simulation is based on a stochastic, susceptible-exposed-infectious-detected-removed-vaccinated (SEIDRV) model for individual disease events, and it represents progression of the disease in a small cluster (henceforth ‘ring’) with homogeneous mixing. The ring represents both contacts and contacts of contacts so the assumption of homogeneous mixing is a simplifying assumption, which we can relax by modelling ‘contacts’ and ‘contacts of contacts’ as separate compartments with the highest transmission among the contacts. New cases arise through direct contact between an infectious individual and a susceptible individual within the ring, and through external infectious pressure, denoted by F, which is constant and fixed for all members of the ring. Members of the ring undergo surveillance by the study team, meaning that infectious individuals are detected and isolated with a daily probability pH, ending their infectious period. We assume in the baseline scenario that detection rate in the trial is equivalent to routine surveillance, reflecting the fact that the trial doesn’t interrupt or enhance disease control efforts. If infectiousness ends naturally, individuals can no longer be detected.
A ring is enrolled into the trial when a case is detected through routine surveillance. This first detected case is defined as the index case for the purposes of the trial, but may or may not be the true index case of the outbreak in the ring. Once a ring enters the trial all its members are randomly assigned to immediate vaccination (on day 1) or delayed vaccination (on day 22). In the baseline scenario we assume no ineligibility or non-consent, so that all susceptible and exposed individuals in the ring are vaccinated, and that there is no heterogeneity or administrative delay affecting the day of vaccination.
The mechanism of the vaccine in an individual is as follows: multiplicative leaky efficacy [5] increases linearly from 0 to VE (set at baseline to be 0.7) over a period of Dramp days following vaccination, after which there is no change in efficacy over the study period [6].
Statistical analysis of the trial is based on cumulative incidence in the rings by end of follow-up and a 95% confidence interval is calculated and reported [7]. The required sample size to test a vaccine effect with 80% power is based on a difference in cumulative incidence [8], using parameters output by a simulated trial with 15,000 rings. We chose this analysis method because of the existence of simple closed-form sample size and vaccine efficacy formulae. Because both arms receive the vaccine, cases that contribute towards the cumulative incidence in each arm are only counted during a window in which the immediate arm is presumed to be protected by the vaccine, and the delayed arm is not protected. The window length is set to 21 days, equal to the vaccination delay between the arms. Because the disease has an asymptomatic phase and the vaccine has a ramp-up period during which it is not fully efficacious, the window starts at 16 days, the sum of the average asymptomatic period length and Dramp, in an attempt to exclude cases in the immediate arm who were infected before they were fully protected by the vaccine. We did not explicitly implement clustering in the simulation, instead assuming that transmission dynamics in all rings are independent. However, clustering of cases within rings arises naturally due to dependent happenings. We measure this clustering using the intracluster correlation coefficient (ICC), calculated as per Shoukri et al [9], adjusting for the covariate of trial arm and accounting for variable ring size where appropriate.
In conducting the statistical analysis we assume full knowledge of the vaccine mechanism, and that cases are only included if they are detected before their infectious period ends, and their symptoms appeared during the window.
For additional details on the disease transmission model, ring initiation, and analysis of the trial see the supplementary appendix.
Table 1 shows the parameters used in the model, their meanings, values under baseline assumptions, and references or justifications.
In order to align this model with the presumed context of the Ebola ça suffit trial, we modelled an entirely susceptible study population at the end of an epidemic, so that Reff has fallen to below one due to behaviour change. To calibrate the model, we set Reff to reproduce a monthly detected attack rate of 2% when starting from one infected individual in a ring of 50 unvaccinated susceptible individuals, in the presence of case detection at a rate pBH.
Under the baseline parameter assumptions listed above, the sample size necessary in each arm to achieve 80% power to detect a difference in cumulative incidence between the two arms is 89 rings, each containing 50 individuals, making a total of 8,900 study participants. This trial would on average return a total vaccine effect estimate of 69.81%, with average 95% CI (28.5, 87.2).
Under baseline parameters in this model, the median total vaccine effect calculated from performing 100 trials with 89 rings in each arm was 70%. This value should include direct and indirect effects, so we would expect it to exceed the direct effect of 70%. However, while direct effects begin immediately, indirect effects are only important in the second generation of preventable cases onwards. There are cases in this generation that occur in the case-counting window because Reff is small and the window duration is not much longer than a typical disease generation (17 days), so the indirect effects are small.
Fig 1 shows the effect of six variables on the point estimate of vaccine effect: daily probability of detection, true individual vaccine efficacy, proportion of infections from outside the ring, baseline attack rate in the unvaccinated population, administrative delay in vaccination, and start day of case-counting window.
Firstly, if there is enhanced surveillance in both arms of the trial leading to more rapid isolation of infectious cases (pH>pBH), this will modestly reduce effectiveness estimates (Fig 1A). Secondly, as individual vaccine efficacy properties increase the estimated vaccine effect increases (Figs 1B and S1). Thirdly, the percentage of infections from within the ring shows a weak negative association with the estimate of vaccine effect (Fig 1C). While the magnitude of indirect effects is modest as discussed above, they are almost negligible when most infections are from outside the ring, because preventing infections within the ring does not confer as much protection to susceptible individuals. The increase in vaccine effect with higher attack rate seen in Fig 1D is driven by the increase in indirect vaccine effects in the immediate arm. Finally, delay between ring formation and vaccination means that by the beginning of the time window the vaccine has had less time to prevent cases in the immediate arm. Thus the reduction in incidence in the immediate arm does not reflect the true effect of the vaccine and the vaccine effect estimate is reduced (Fig 1E).
A major determinant of the effect estimate is the choice of time window in which to count cases, as seen in Fig 1F. Not surprisingly, starting the window too early reduces the estimated effects because it includes a period of time during which the vaccine cannot affect the incidence of cases becoming symptomatic–many cases becoming symptomatic on day 8, for example, will have been infected by the index case prior to isolation, or will have been infected by a contact on (say) day 3, before the vaccine had time to induce protection.
Starting the window later than the baseline of 16 days allows the trial to capture later generations in the chain of transmission, from a vaccinated person to another vaccinated person. This increases the vaccine effect estimate as it includes indirect effects. One might expect to see that starting the window too late would reduce effect estimates because it would include a period when the delayed group was also protected by the vaccine. This does not appear to be the case, at least up to a start time of 35 days (Fig 1F)–see the supplementary material for an explanation of this phenomenon.
Fig 2 shows the effect of the same six variables on the required sample size: baseline attack rate in unvaccinated population, start day of case-counting window, daily probability of detection, true individual vaccine efficacy, administrative delay in vaccination, and force of external infection.
The effect of each parameter on the sample size can be understood through its effect on one or more of the three factors that determine the power of this trial: the number of events, how they are distributed between the two arms, and the level of clustering of cases within rings. Respectively these factors are represented by the attack rate in the controls, the cumulative incidence difference between the arms, and the intracluster correlation coefficient (ICC) [8].
Variables that decrease the incidence rate in the controls and cases will decrease the power because for the same sample size the trial will observe fewer events. The baseline detected attack rate among unvaccinated individuals is a simple example of such a parameter (Fig 2A). Two other parameters act on the overall incidence in the trial. Firstly, making the start of the case-counting window later decreases incidence in both arms because with Reff<1 the incidence is on average declining, so across all rings in the trial the number of cases decreases over the follow-up period (Fig 2B). Secondly, the case detection decreases detected incidence rate at both extremes (Fig 2C). When case detection is high, transmission chains are interrupted by case isolation and the true incidence decreases. When case detection is low, many cases die or recover before they can be detected and consequently the detected incidence decreases.
Variables that make the two arms of the trial appear more different will increase the power of the trial as the ability to differentiate between them is increased, and Fig 1 identifies such variables. Vaccine characteristics, in particular vaccine efficacy (Fig 2D), are simple examples of such a parameter, since the immediate arm receives greater protection against disease compared to the delayed arm. Changes to two other parameters increase the incidence difference in this way, as explained above: reducing the delay between ring formation and vaccination (Fig 2E) and starting the case-counting window earlier (Fig 2B).
The effect of the timing of starting to count cases thus reflects two opposing forces on the sample size: it decreases sample size by increasing the incidence difference, and it increases sample size by decreasing the overall incidence. When the window is early, the former of these effects dominates as seen by the increase in sample size for early time windows in Fig 2B. When the window is late, the latter effect dominates, as seen by the increase in sample size for late time windows in the same figure.
Finally, the level of clustering within rings inflates the sample size, because more clustering means that each individual case provides less information. It is often not intuitive to predict the direction in which a parameter will cause the ICC to change, and in many cases the ICC is not sensitive to the parameter. One exception is the infection from outside the ring (Fig 2F). The most significant effect of introducing external infection and reducing within-ring transmission is to make infection probability for one individual within a ring independent from the infection prevalence within the same ring. This reduces clustering in incidence (making it more Poisson-like), thus reducing the ICC and the necessary sample size.
The width of the confidence intervals is affected in the same way by the three variables described above. In particular, low incidence in either arm, high ICC and a small incidence difference between the arms all lead to a wider confidence interval. The formula for the confidence interval is different from the formula used to make the power calculation, so the trends do not completely align because the size of the effect of each of the three factors is different for the confidence interval and the sample size.
For an investigation of the sensitivity of the total vaccine effect estimate and sample size to other parameters in the model, see the supplementary material. For an interactive tool to explore the sensitivity of the trial parameters, see https://matthitchings.shinyapps.io/ShinyApps/.
The ring-vaccination, cluster-randomized design has two key strengths that make it a good candidate when disease transmission exhibits spatiotemporal variation. Firstly, by including members of the study population who are contacts of cases, the trial preferentially selects those at higher risk of disease acquisition, leading to an increase in efficiency while preserving false-positive rate through randomization. Indeed, when a vaccine with 0% efficacy was tested in our simulations the false positive rate was maintained at 5%. Secondly, even those study subjects who are randomized to delayed vaccination are theoretically in close contact with the study team meaning that individuals from the source population who are at the highest risk are followed closely and benefit from the trial even in the absence of vaccination [12].
In addition, vaccination of clusters when they arise allows for gradual inclusion, meaning that this design is appropriate when logistical constraints make immediate vaccination of all participants impossible or inappropriate. In this respect it is similar to a stepped-wedge cluster trial, in which prespecified clusters within the study population are vaccinated in a random order. Although we have not made a direct comparison in this study, Bellan et al [13] showed that the stepped-wedge design is underpowered when the incidence is declining because it cannot prioritize the vaccine for those at highest risk. The ring vaccination design, on the other hand, is inherently risk-prioritized because all study participants should be at higher risk than the general population.
All trials should be correctly powered in order to avoid erroneous rejection of an efficacious vaccine. For a trial design with several complexities such as the one presented here, a sophisticated approach to sample size calculation is merited. A standard approach to sample size calculation for this trial would involve specifying the attack rate among the controls, the desired effect of the vaccine on the population level, and the ICC. In the context of a serious epidemic, these parameters are unlikely to be estimated with certainty; for example, the ICC requires cluster-level data to be estimated accurately. The ICC is an important parameter in designing cluster-randomized trials, yet in the absence of data it is often assumed to be 0.05. In our simulations the range of ICCs observed was 0.01–0.04, suggesting that the value of this uncertain parameter should not always be assumed to be fixed at 0.05. Therefore, the modelling approach replaces assumptions about these cluster-level quantities with assumptions about population-level parameters and disease characteristics, which are more likely to be available through analysis of data from the outbreak.
A second advantage of the modelling approach is that, based as it is on a simulating the transmission of disease within a trial, it is possible to explore the impact of parameters describing the design of the trial and the properties of the disease. The added detail gained from specifying the disease model allowed us in this study to identify some key issues with the design that are worth considering.
Firstly, as seen in Fig 2C, increasing case-finding efficiency above background rate has a negative impact on power, as fast isolation of cases in both arms leads to an overall decrease in cases observed by the trial. In future trials it is worth considering if there are alternative or composite endpoints, if the disease in question permits, that can be used to allow for efficacy estimates while maintaining close follow-up.
Secondly, a key design consideration in the delayed-arm ring-vaccination trial is when to count cases. An intuitively appealing approach is to place the window so that the immediate arm is receiving full protection and the delayed arm none. This should in theory minimize bias caused by misclassification of unvaccinated individuals as vaccinated and vice-versa. While this placement achieves nearly maximal power, it does not maximize the VE estimate. Indirect effects that are important later in time increase the VE estimate for later time windows, while at the same time declining incidence within each ring decreases power for later time windows.
Finally, the above point draws attention to the fact that caution is required when interpreting the VE estimate produced by the trial. As seen in Fig 1, many parameters that are not characteristics of the vaccine can influence the estimated effect. Whether this is due to misclassification (for example, when the time window is too early) or due to indirect effects (for example, when the attack rate is high enough to cause long transmission chains), the context of the trial should be taken into account when interpreting the VE estimate. While in the baseline scenario the trial appears to correctly estimate the individual efficacy, this is the result of misclassification and indirect effects cancelling each other out. This claim is supported by the fact that the median VE estimate falls below the individual-level vaccine efficacy when most or all infections are from outside the ring (Fig 1C) and indirect effects are negligible.
The focus of this model was to explore parameters within each ring and understand how they affect the quality of data coming from the trial. As a result, we did not consider the wider context of the population disease dynamics, and in particular how and when the rings arise. For example, we calibrated Reff to a secondary attack rate in a cluster was 2%, which is not necessarily comparable to the monthly cumulative incidence in the population. If transmission takes place mainly in clusters then population cumulative incidence could be somewhat lower than cluster secondary attack rate, increasing the efficiency of a ring-vaccination trial relative to a stepped-wedge cluster trial or individual RCT. Linking this model to a model of disease within the general population would allow us to make direct comparisons to other trial designs such as the stepped-wedge cluster trial and the individually-randomized trial investigated elsewhere [13, 14], but it would require detailed information about the nature of clustering of the disease in this context, and for simplicity we focused on the within-ring dynamics only.
As with every model, there are limitations to these simulation results. The strength of the modelling approach compared with a standard approach is that it better estimates the parameters on which the sample size depends. However, some of the model parameters might still be uncertain in a situation in which such a model might be useful. For example, we may have limited information about the characteristics of a disease, in particular its latent and incubation period, and its Reff. The simulation results are dependent on these assumptions, and so they cannot be used at the very outset of epidemic, or else they risk being highly inaccurate. Even at the end of the West African Ebola epidemic, there were no more than four or five reliable estimates of the latent and infectious periods of EVD, and indeed there is perhaps evidence that our understanding of the natural history of the disease remains limited [15]. In addition, we have considered only the simplest method of analysis for the trial–a comparison of attack rates between the two arms after correction for clustering of cases within rings. More sophisticated methods, including time-to-event analyses incorporating ring-level random effects, as performed in the Ebola ça suffit trial, would have somewhat different sample size requirements. However, we believe that the trends seen here would be similar for other methods, because the VE estimates returned by various methods will be similar for a rare outcome [5]. In building the model we made some simplifying assumptions, and although we tested the robustness of the results to these assumptions (see supplementary material) it is possible that a more sophisticated model would provide more accurate results, particularly if superspreading events are not rare in this study population.
For a vaccine trial in an epidemic, when the level of indirect effects is hard to predict, power calculations can be sensitive to parameters about which very little is known. Simulations such as these can be important aids in understanding a range of values for these parameters before a trial is carried out, and thus ensuring that the trial has sufficient power to detect an efficacious vaccine. In this trial, a finding significantly different from the null likely indicates one or more types of vaccine efficacy at the individual level, but the magnitude of the effect and the power to detect the effect will vary across settings.
|
10.1371/journal.pntd.0006667 | Pre-post effects of a tetanus care protocol implementation in a sub-Saharan African intensive care unit | Tetanus is a vaccine-preventable, neglected disease that is life threatening if acquired and occurs most frequently in regions where vaccination coverage is incomplete. Challenges in vaccination coverage contribute to the occurrence of non-neonatal tetanus in sub-Saharan countries, with high case fatality rates. The current WHO recommendations for the management of tetanus include close patient monitoring, administration of immune globulin, sedation, analgesia, wound hygiene and airway support [1]. In response to these recommendations, our tertiary referral hospital in Tanzania implemented a standardized clinical protocol for care of patients with tetanus in 2006 and a subsequent modification in 2012. In this study we aimed to assess the impact of the protocol on clinical care of tetanus patients and their outcomes.
We examined provision of care and outcomes among all patients admitted with non-neonatal tetanus to the ICU at Bugando Medical Centre between 2001 and 2016 in this retrospective cohort study. We compared three groups: the pre-protocol group (2001–2005), the Early protocol group (2006–2011), and the Late protocol group (2012–2016) and determined associations with mortality by univariable logistic regression.
We observed a significant increase in provision of care as per protocol between the Early and Late groups. Patients in the Late group had a significantly higher utilization of mechanical ventilation (69.9% vs 22.0%, p< 0.0001), provision of surgical wound care (39.8% vs 20.3%, p = 0.011), and performance of tracheostomies (36.8% vs 6.7%, <0.0001) than patients in the Early group. Despite the increased provision of care, we found no significant decrease in overall mortality in the Early versus the Late groups (55.4% versus 40.3%, p = 0.069), or between the pre-protocol and post-protocol groups (60.7% versus 50.0%, p = 0.28). There was also no difference in 7-day ICU mortality (30.1% versus 27.8%, p = 0.70). Analysis of the causes of death revealed a decrease in deaths related to airway compromise (30.0% to 1.8%, p<0.001) but an increase in deaths due to presumed sepsis (15.0% to 44.6%, p = 0.018).
The overall mortality in patients suffering non-neonatal tetanus is high (>40%). Institution of a standardized tetanus management protocol, in accordance with WHO recommendations, decreased immediate mortality related to primary causes of death after tetanus. However, this was offset by an increase in death due to later ICU complications such as sepsis. Our results illustrate the complexity in achieving mortality reduction even in illnesses thought to require few critical care interventions. Improving basic ICU care and strengthening vaccination programs to prevent tetanus altogether are essential components of efforts to decrease the mortality caused by this lethal, neglected disease.
| Tetanus is a disease characterized by violent, repetitive muscular spasms that are frequently lethal. It rarely occurs in high-income countries due to a highly effective vaccine coupled with a strong immunisation system. Tanzania is a large low-income country in East Africa. Its size and rapidly growing population are straining factors for the immunisation programme. Women typically receive tetanus booster vaccinations during antenatal care, while men remain at risk for tetanus as there is no additional system in place to ensure post-infancy vaccination. In 2006, our hospital in northwest Tanzania implemented a standardized management protocol for the frequent cases of tetanus that we admit. We assessed the impact of this protocol on the care and survival of these patients, most of whom are between the ages of 20 and 50 years. Although we observed improving clinical care over a 16-year period, this increased care did not result in the reduction in mortality that had been expected. Our work reflects the complexity of this neglected disease, the challenges associated with protocolised care provision in such settings, and the importance of monitoring the effects of interventions following their implementation in low-income countries. Given the devastating personal and economic consequences of tetanus for patients and their families, we also highlight the urgent need to ensure immunisation for these vulnerable men who are the economic backbone for households and the country.
| Despite being a vaccine-preventable disease, tetanus is frequently encountered in sub-Saharan Africa [2,3]. The incidence of non-neonatal tetanus cases has fallen since the initiation of the vaccination programme but the number of cases remains high, with 4,604 non-neonatal cases reported in 2016 in the African region,[4] and likely many more that were not reported [5]. Among global tetanus deaths, 44% occur in sub-Saharan Africa and the highest proportion of these is in East Africa [6].
Inadequate vaccination is cited as the primary causative factor for tetanus despite the availability of a highly effective vaccine [7,8]. In Tanzania, national tetanus vaccine coverage is 87% in children less than 1 year, as determined by history and vaccination cards, but regional discrepancies are high [9]. In Mwanza, where our hospital is located, only 70% of infants received all basic vaccines in 2015 [10]. In addition, because the Tanzanian vaccination programme focuses on children under the age of 1 and pregnant women who attend antenatal clinic, there is currently no system in place to ensure booster vaccination for men past the infancy doses, despite recommendations by the WHO [11]. This would explain why young men are the most at risk of tetanus infection in Tanzania and many other sub-Saharan African countries [3]. A recent study from Tanzania showed that only 28% of men older than 15 years were seroprotected against tetanus [12]. This has tremendous economic and social consequences: in 2015 the WHO calculated the cost of a single dose of tetanus vaccine at $0.14 [13] whereas the cost of caring for patients with tetanus in low- and middle-income countries ranges from $78 [14] to $900 [15]. In Tanzania, 55% of the population lives in extreme poverty (less than $1.25 per day) and men are the main financial providers, so the socioeconomic impact is tremendous.
Lack of medications, inadequate implementation of proven treatment interventions, high treatment cost for patients, and long distance to specialised centres have been cited as additional key reasons for high tetanus-associated mortality in sub-Saharan Africa [3,16,17]. Studies of targeted approaches to address these barriers are lacking. Therefore, our goal was to conduct a quality-improvement project to assess whether the utilization of a standardized hospital protocol for management of tetanus was effective in reducing mortality. Our study was possible because in 2006 our Tanzanian referral hospital implemented a hospital protocol to be used for management of all patients admitted to our ICU with tetanus, with further updates in 2012. We sought to determine whether the implementation of this protocol had any impact on provision of clinical care and patient outcomes. We hypothesised that protocol-driven implementation of proven tetanus interventions would increase over the study period, and that tetanus mortality would decrease.
Ethical approval for the conduct of this study was obtained from the joint Catholic University of Health and Allied Sciences (CUHAS)/BMC Research Ethics and Review Committee (BREC/001/18/2008), the National Institute for Medical Research (NIMR/HQ/R.8c/Vol. IX/1085), and Weill Cornell Medicine (1108010827).
We conducted a retrospective cohort study of all patients who had a diagnosis of tetanus and were admitted to the Intensive Care Unit (ICU) of Bugando Medical Centre (BMC), Tanzania, from May 2001 to September 2016. BMC is a public tertiary referral hospital located in Mwanza, a north-western city on the shores of Lake Victoria. Mwanza is the second largest city in Tanzania and BMC serves the 15 million people of the Lake Zone.
The ICU admits approximately 500 patients annually from all disciplines, with specialists from internal medicine and anaesthesia providing the majority of the ICU care. The ICU has 13 beds with 7 mechanical ventilators, pressurised wall oxygen, suction, and bedside monitors. The monitors display non-invasive blood pressures, saturations and electrocardiography. There is regular availability of intravenous (IV) fluids, antibiotics, adrenaline, and dopamine. Central lines and noradrenaline are intermittently available. The nurse to patient ratio is 1:2, with most of the nurses having no formal training in intensive care medicine.
In view of high mortality rates and in an effort to improve the quality of care for patients, BMC implemented a hospital protocol to optimize care of tetanus patients in 2006, in accordance with WHO recommendations. The hospital protocol can be seen in the supporting information (Supporting information (S1); Fig 1. Bugando Medical Centre tetanus management protocol) In 2012, several additional key interventions were implemented in the ICU, including employment of a dedicated ICU physician and an agreement with the surgical department that tracheostomies would be performed as soon as possible for tetanus patients admitted to the ICU.
The stepwise protocol is organised based on clinical urgency of interventions. The initial stage focuses on airway management either in the form of oral intubation or immediate tracheostomy. There was no specific criterion or threshold for initiation of mechanical ventilation during the entire study period and this decision was made at the discretion of the on-duty physician. This is followed by emphasis on early prescription and administration of immune globulin, antibiotics, spasm control with magnesium sulphate and benzodiazepines with accompanying analgesia, fluid management to avoid acute kidney injury and deep vein thrombosis prophylaxis. Wound care is also included in the early stages with surgical consultation for wound debridement if needed. Farmers and manual workers were identified, from previous work, as the key at-risk group, and most present with identifiable wounds.
The next stage focuses on monitoring the patient’s response to interventions and conducting investigations. Instructions are provided on how to adjust prescribed medication to get the desired outcomes. The final stage focuses on the recovery phase of care with instructions of down-titrating the medications and ensuring immunisation prior to ICU discharge.
According to our hospital protocol, all patients presenting to the hospital with tetanus are admitted directly to the ICU. We identified patients with non-neonatal tetanus using either the ICU admission registry or the separate inpatient medical registry. Patients were diagnosed based on clinical findings of rigidity and/or spasms often preceded by a penetrating injury. Medical notes were sought from the medical records department. Detailed information on care was only available for patients after 2008. The main sources of information were the ICU admission notes, ward round notes and daily charts. We collected data on age, sex, place of residence, time taken to present to hospital, provision of clinical care, time to care provision, and outcomes.
We categorized patients into 3 groups based on the year of presentation. The pre-protocol group included patients admitted with tetanus prior to the protocol implementation in 2006 (2001–2006). The Early group included patients admitted between 2006 and 2011, and the Late group included patients admitted between 2012 (the year in which the hospital tetanus protocol was modified) and 2016.
The primary study outcomes were care provision for patients, 7-day mortality in the ICU, and overall mortality. The specific care interventions that we examined included administration of immune globulin (early in the emergency department or late in the ICU), surgical wound care, administration of antibiotics, and airway management in the form of mechanical ventilation and tracheostomy placement.
Statistical analysis was performed using STATA 14.0 (College Station, Texas, USA) and all data was anonymized during the analysis. Descriptive analysis of baseline variables was performed to summarize patient characteristics. Categorical variables were described using proportions and continuous variables were described using medians and interquartile ranges. We compared overall mortality between the pre- and post-protocol groups. We explored the differences between the Early and Late groups by chi-squared test for categorical variables and Wilcoxon rank-sum test for continuous variables. Overall and 7-day case-fatality rates were calculated and compared between the groups. We assessed potential factors associated with mortality by univariable and multivariable logistic regression for the Early and Late groups. All statistical tests were performed at a 5% significance level.
A total of 277 patients were admitted to the BMC ICU with tetanus between May 2001 and September 2016. No cases of neonatal tetanus were admitted; all patients were 8 years older and above. Thirty-one (11.2%) patients were admitted before January 2006 and were classified in the pre-protocol group, 133(48.0%) patients were classified in the Early group, and 113 (40.8%) were classified in the Late group. Medical files were available for 162 patients (59 in the Early group and 103 in the Late group).
When comparing the pre and post-protocol groups, both were mostly male (12/14 (85.7%) and 211/246 (85.8%) respectively, p = 0.995), and there was no significant difference in age (30.0 [21–42] years and 30.0 [19–46] years respectively, p = 0.86, Table 1). Overall case fatality was similar in both the pre- and post-protocol group (17/28 (60.7%) and 119/238 (50.0%), p = 0.28). The case-fatality rate by year of admission is presented in Fig 1.
When comparing the Early and Late groups, for whom detailed medical record data was available, we found no differences between the two groups with respect to sex, age, or distance travelled to reach BMC (Table 1). The groups had similar median time intervals from the onset of symptoms to hospital presentation, though significantly more people in the Early group presented later (3 [1–7] versus 3 [1–4] days, p = 0.042). The Early group also experienced significantly more wounds to the back than the Late group. No other significant differences in clinical presentation were observed.
Care interventions significantly differed between the two post-protocol groups (Table 2). We observed significant increases in surgical wound care, initiation of mechanical ventilation and performance of tracheostomies in the Late group compared to the Early group. Both groups had similar in-hospital rates of immune globulin administration, but 57/95 (60.0%) of patients in the Late group received early administration of immune globulin in the emergency department compared to 9/50 (18.0%) of patients in the Early group (p<0.001). We additionally observed a reduction in the time taken to initiating mechanical ventilation (2.0 [2.0–4.0] days in the Early group versus 0.0 [0.0–2.0] in the Late group, p = 0.0030) and in performing tracheotomies in those for whom mechanical ventilation was initiated (12.0 [8.5–18.5] days in the Early group versus 1.0 [0.0–8.0] in the Late group, p = 0.023). All patients in the Early and Late group received antibiotics.
Contrary to our hypothesis, we did not find a significant reduction in the overall hospital case-fatality ratios between the Early and the Late groups. In contrast, there was a trend towards an increase in mortality from 56/126(44.4%) in the Early group to 63/112 (56.3%) in the Late group (p = 0.069). Additionally, there was no change in 7-day case-fatality ratio (43/126 (34.1%) versus 35/112 (31.3%), p = 0.64). The Early group had 6/20 (30.0%) deaths attributed to loss of airway, mostly due to laryngospasm, whereas in the Late group, this accounted for only 1/56 (1.8%) death (p = 0.001). Sepsis accounted for 3/20 (15.0%) of all deaths in the Early group and 25/56 (44.6%) in the Late group (p = 0.029) (Table 3). Of note, there were no other clear changes between the Early and Late groups such as ICU staff: patient ratios, nurse training, ICU admission numbers, timeliness of antibiotic administration, or decontamination procedures.
All variables listed in Table 1 were analysed as possible predictors of overall ICU mortality in the Early and Late groups and are presented in Table 4. In the Early group, patients with longer time from symptom onset to presentation at the hospital had lower odds of death compared to patients with shorter time to presentation (OR = 0.68 [0.52–0.90] for each day delay in presenting for care, p = 0.007). Patients for whom mechanical ventilation was initiated had higher odds of death compared to patients not receiving mechanical ventilation (OR = 4.53 [1.24–16.58], p = 0.022). The increased odds of death may in part have been related to development of sepsis, though numbers were too small to draw conclusions. In the Late group, 25 out of 72 (34.7%) patients who were mechanically ventilated developed sepsis, compared with 3 out of 13 (23.1%) in the Early group (p = 0.53).
In the Late group, older patients as well as patients with longer time from presentation to initiation of mechanical ventilation had higher odds of death (OR = 1.05 [1.02–1.07] for each increasing year of age, p<0.001 and OR = 1.63 [1.06–2.52] for each day delay in receiving mechanical ventilation, p = 0.028 respectively). Multivariable analysis showed that the factors associated with mortality in the Early group were age (OR = 1.04 [95% CI 1.00–1.08], P = 0.03), mechanical ventilation (OR = 26.7 [95% CI 2.06–274.07], p = 0.006) and time from symptom onset to attending care (OR = 0.50 [95% CI 0.32–0.81), p = 0.006). In the Late group, the factors that remained significantly associated with mortality were age (OR = 1.03[95% CI 1.00–1.06], p = 0.01) and time from presentation to initiation of mechanical ventilation (OR = 1.74[95% CI 1.07–2.81], p = .024).
Our work demonstrates that the use of a standardized protocol was associated with a significant improvement in the implementation of specific interventions that have been shown to reduce mortality from tetanus. However, this led to an increase in invasive procedures, and therefore the anticipated reduction in overall mortality over a 10-year period was not seen. While the deaths due to respiratory failure and airway obstruction decreased, those due to sepsis increased. This ultimately led to longer ICU stays over the 10-year period without improving mortality. To our knowledge, this is the first study from sub-Saharan Africa to look at the impact of protocolised ICU care for tetanus patients. Our findings suggest the urgent need for additional work to optimize ICU care to help offset the mortality from secondary causes which arise due to gains from improvement in immediate survival. Furthermore, data from our study shows mortality from tetanus remains high and that efforts to improve adult immunization to prevent this neglected disease should be prioritized.
Most previous studies have highlighted poor provision of clinical care as a key contributory factor to observed high mortalities. In the Late period of our protocol implementation, patients experienced higher levels of clinical care as compared to other studies. Rates of mechanical ventilation (69.9% in our study vs 10.5% in other studies) [16,17], administration of immune globulin (93.1% vs 9–65%) [17,18], surgical wound care (41.0% vs 11.8%) [18], administration of antibiotics (100% vs 58%) [17], and tracheotomies (39.2% vs 11%) [16] were all higher in the Late period of our study compared to rates reported in other studies. A structured approach via a standardized protocol was likely a key contributory factor to the increase in the clinical care. In spite of the protocol, the mortality rates we have reported in both time periods of our study are very similar to rates reported in other studies from similar settings.
There is mixed evidence for the effectiveness of protocol standardization for medical care both globally [19,20] and also in low- and middle-income countries [21,22]. Nonetheless, protocolised care is still common, especially in environments where specialised care is not readily available and where providers’ levels of training and expertise may be variable. Despite highly significant increases in clinical care provision with earlier administration of immune globulin, an increase in surgical wound care, and an increase in mechanical ventilation and tracheostomies, our results showed no improvement in mortality. In fact, we even observed a non-significant trend towards increased mortality in the Late years of implementation of the protocol compared to the Early years (56.3% versus 44.4%, p = 0.069). The similarity in demographics between the two groups reinforces the hypothesis that the differences in mortality may be due to the protocol itself.
Implementation of the hospital protocol for tetanus management appears to have led to a shift in the causes of death among tetanus patients. In the Early years after protocol implementation, airway obstruction and respiratory failure were the most frequent causes of death, consistent with findings from other studies [3,16]. In contrast, sepsis was the most likely cause of death in the Late group, increasing from 15% at baseline to 45% in the Late group. It is likely that increased rates of mechanical ventilation and performance of tracheostomies contributed to a decline in the number of airway/respiratory deaths but that this overall increase in interventions led to an increase in the number of deaths by sepsis, for at least three reasons. First, the Late group underwent more invasive procedures including mechanical ventilation, surgical tracheostomies and surgical wound treatment, each of which increases the risk of hospital-acquired infections (HAIs). Second, our clinical experience suggests that patients in the Late group also more frequently had central venous catheters inserted, which poses an additional risk factor for HAIs. Third, patients in the Late group spent more days in the ICU, often immobilized and at increased risk for other causes of in-hospital mortality such as pulmonary embolism, urosepsis and bedsores. Of note, at present there is no specific aspect of the protocol that focuses on reducing HAIs or other critical illness-associated morbidity.
There are approximately 27 studies involving more than 25,000 patients with tetanus over a 60-year period issued from the African continent [3]. Most of the reported studies have been done at tertiary level health facilities, similar to our hospital. The median age, male predominance, median time to presentation, clinical presentation, and hospital lengths of stay are similar between our two groups and also similar to other studies of tetanus patients in Africa [3]. Most of the studies reported similar high overall hospital fatalities [3]. All of this suggests that our hospital setting is similar to many others in sub-Saharan Africa and that our finding that tetanus protocol implementation did not improve mortality is likely to be generalizable as well.
Our results are to be interpreted in light of some limitations. First, due to our inability to locate medical records for some of the tetanus patients, we were not able to document the care provided in all cases but only to document basic demographic characteristics and outcomes. Furthermore, additional data that would have been informative, such as trends in antibiotic resistance over time and the specific antibiotics administered, were not available. These limitations highlight the complexity of implementing a hospital protocol and the urgent need for additional studies in this area.
In summary, we have demonstrated a significant increase in clinical care in accordance with a standardized protocol for the treatment of tetanus patients. The protocol has not led to the anticipated reduction in patient mortality. The unchanged mortality rate, with a shift in causes of death, highlights several key points for consideration when protocols are implemented in resource-limited settings. First, implementation of protocolised care in resource-limited settings is highly complex and requires in-depth monitoring and assessment of patients, staff, and procedures. In our hospital, we are now working to implement infection control policies and determine antibiotic resistance patterns in an effort to decrease HAIs. We will continue to monitor the effect of this intervention and to consider other possible interventions to decrease the mortality of tetanus patients. In addition, management and early recognition of sepsis is extremely complex in resource-limited settings and more surveillance is needed. Finally, we strongly call for an increase in vaccination coverage for at-risk men in sub-Saharan Africa, beginning with the highest-risk groups such as farmers and motorcyclists [5,23], with the aim of eliminating this preventable, lethal disease.
|
10.1371/journal.pgen.1006103 | Cellular and Molecular Features of Developmentally Programmed Genome Rearrangement in a Vertebrate (Sea Lamprey: Petromyzon marinus) | The sea lamprey (Petromyzon marinus) represents one of the few vertebrate species known to undergo large-scale programmatic elimination of genomic DNA over the course of its normal development. Programmed genome rearrangements (PGRs) result in the reproducible loss of ~20% of the genome from somatic cell lineages during early embryogenesis. Studies of PGR hold the potential to provide novel insights related to the maintenance of genome stability during the cell cycle and coordination between mechanisms responsible for the accurate distribution of chromosomes into daughter cells, yet little is known regarding the mechanistic basis or cellular context of PGR in this or any other vertebrate lineage. Here we identify epigenetic silencing events that are associated with the programmed elimination of DNA and describe the spatiotemporal dynamics of PGR during lamprey embryogenesis. In situ analyses reveal that the earliest DNA methylation (and to some extent H3K9 trimethylation) events are limited to specific extranuclear structures (micronuclei) containing eliminated DNA. During early embryogenesis a majority of micronuclei (~60%) show strong enrichment for repressive chromatin modifications (H3K9me3 and 5meC). These analyses also led to the discovery that eliminated DNA is packaged into chromatin that does not migrate with somatically retained chromosomes during anaphase, a condition that is superficially similar to lagging chromosomes observed in some cancer subtypes. Closer examination of “lagging” chromatin revealed distributions of repetitive elements, cytoskeletal contacts and chromatin contacts that provide new insights into the cellular mechanisms underlying the programmed loss of these segments. Our analyses provide additional perspective on the cellular and molecular context of PGR, identify new structures associated with elimination of DNA and reveal that PGR is completed over the course of several successive cell divisions.
| Lampreys possess a fascinating genome biology wherein large portions of the genome, including large numbers of genes, are programmatically deleted during development. The lamprey therefore represents a uniquely informative system with respect to several broad areas of biology, including genome stability/rearrangement, epigenetic silencing, and the establishment and maintenance of pluripotency. However, little is known regarding the cellular context or mechanism of deletion, partly due to the challenges of observing rearrangements in situ. Here we present analyses and new techniques that significantly advance our understanding of the subcellular context of programmed rearrangements and interactions between programmed deletion and canonical DNA silencing mechanisms. These analyses demonstrate that DNA elimination occurs earlier in embryogenesis than was previously recognized and reveal several new cellular and molecular aspects of programmed DNA loss. Specifically we show that eliminated DNA exhibits a unique migration pattern during cell division, is packaged into discreet subcellular structures later in the cell cycle, and undergoes epigenetic silencing through DNA and histone methylation. These observations provide new insight into the mechanisms underlying programmed DNA loss and suggest a functional link between programmed DNA loss and other, more conserved gene silencing pathways.
| The sea lamprey (Petromyzon marinus) represents one of the few vertebrate species known to undergo large-scale programmatic elimination of genomic DNA over the course of its normal development [1–4]. Programmed genome rearrangements (PGRs) result in the reproducible loss of ~20% of the genome from somatic cell lineages and a reduction of chromosome number from ~198 to ~164 (2N) [4–6]. Previous studies have shown that DNA is physically eliminated during the transition between gastrula and blastula stages: between the second and third day of development [4]. Given that most aspects of lamprey’s developmental and cellular biology are conserved with other vertebrates [7–10], PGR holds the potential to provide novel insights related to maintenance of genome stability and interactions between various cellular mechanisms responsible for the proper segregation of chromosomes.
Lampreys are by no means the only organisms that undergo large-scale programmed rearrangement of their genomes. Organisms known to undergo PGR include diverse protozoan, invertebrate and vertebrate taxa, and the mechanisms underlying PGR are thought to be similarly diverse [11–19]. Studies of these independent acquisitions have revealed common themes that speak to the underlying logic of PGR and its integration with other epigenetic silencing pathways [11–14]. In many taxa PGR is known to occur early in development and results in the targeted elimination of specific genomic segments from essentially all somatic cell lineages, with targeted segments being retained exclusively by the germline. Studies in lamprey and the nematode Ascaris suum have shown that eliminated DNA encodes genes that are expressed in mature gonads and embryonic germ cells [6, 20], supporting the interpretation that PGR likely serves as an irreversible mechanism of silencing genes within somatic cell lineages.
Studies performed on diverse taxa suggest that PGR-mediated silencing may often interact cooperatively with other silencing pathways. In the ciliates both DNA methylation/hydroxymethylation and methylation of histone H3 at lysine 9 (H3K9me) are associated with programmed elimination [11, 21]. In sciarid flies embryonic elimination of the paternal X chromosome is associated with retention of H3S10 hyperphosphorylation (H3S10P) during late anaphase, which may contribute to silencing by preventing decondensation and access to H3K9 by methyltransferases [12, 13]. Similarly, in zebra finch a single germline-restricted chromosome is heavily marked by both trimethylated H3K9 (H3K9me3) and acetylated H4K16 in meiotic testes (the chromosome is eliminated at the end of male meiosis and only transmitted by oocytes, although embryonic elimination has not been directly observed) [14].
Little is known regarding the mechanistic basis or cellular context of PGR in any vertebrate lineage. Given the high fecundity of lampreys and the fact that fertilization and all stages of embryonic development occur externally, lamprey provides a powerful system for observing and manipulating cells during the process of PGR. Here we describe epigenetic correlates of PGR and the spatiotemporal dynamics of DNA elimination in lamprey. In situ analyses revealed that the earliest DNA methylation events target specific extranuclear structures (micronuclei) that contain DNA eliminated by PGR. The spatiotemporal resolution of these analyses also permitted the discovery of other reproducible subcellular features that are associated with the differential segregation of retained vs. eliminated DNA and the packaging of eliminated DNA into micronuclei. Specifically, eliminated DNA appears to be packaged into chromatin that does not migrate with somatically retained chromosomes and is superficially similar to lagging chromosomes that are observed in some cancer subtypes [22–24]. Closer examination of “lagging” chromatin reveals distributions of repetitive elements, cytoskeletal contacts and apparent chromatin contacts that provide new insights into the cellular mechanisms underlying the programmed loss of these segments.
Programmed DNA elimination is sparsely distributed across the tree of life and likely arose several times over metazoan evolution [19] yet in several species programmed elimination of DNA has been shown to act cooperatively with other, more conserved, epigenetic silencing pathways [11–14]. To investigate possible interactions between PGR and early gene silencing events, we applied indirect immunofluorescence labeling using antibodies against 5-methylcytosine (5meC), histone 3 trimethylated at lysine 9 (H3K9me3) and histone 3 trimethylated at lysine 27 (H3K27me3) to characterize the distribution of these modifications during early embryogenesis. In general these repressive modifications were essentially absent at the earliest developmental time points and increased in abundance during the first week of development. Similar patterns have been observed for several vertebrate and invertebrate species, reflecting reprogramming events that are involved in the initial establishment of pluripotency following fertilization (i.e. global demethylation) and the subsequent onset of zygotic genome activation [25, 26]. However, the subcellular localization of two modifications (5meC and H3K9me3) deviated from the typical pattern that has been described for other taxa. During the first two days of development, days post fertilization (dpf) 5meC and H3K9me3 immunofluorescence localized almost exclusively to DAPI-positive extranuclear structures (micronuclei–MNi, Fig 1A). To more thoroughly test whether micronuclei are associated with the elimination of DNA via PGR, we performed in situ hybridization with the germline-enriched repetitive element Germ1. This sequence is highly abundant within the germline and only localizes to two somatically retained chromosomes [6]. These analyses revealed that a majority of MNi, though not all, contain the Germ1 repeat, consistent with the interpretation that these micronuclei contain material destined for elimination from somatic lineages via PGR (Fig 1C and 1D, S1 Table).
Two-color immunolabeling of 5MeC and H3K9me3 revealed that these heterochromatic marks occur in largely non-overlapping sets of MNi and vary in prevalence over the first several days of embryogenesis (Fig 1A and 1B, S2 Table). At 1 dpf, 5MeC was essentially absent from both nuclei and micronuclei, whereas ~60% of micronuclei showed strong immunolabeling for H3K9me3. At 1.5 dpf the proportion of H3K9me3 positive MNi remained relatively stable, and the first 5MeC positive MNi were observed, albeit at a relatively low frequency (~7% of MNi). At 2 dpf the proportion of 5MeC positive MNi significantly increased and the proportion of H3K9me3 positive MNi significantly decreased, with localization of H3K9me3 transitioning to the primary nucleus (Fig 1A, S1D Fig, S2 Table). Similar patterns of 5MeC and H3K9me3 immunolabeling were observed at 2.5 and 3 dpf. Coordinate with changes in the distribution of epigenetic modifications, the abundance of MNi also changed dynamically over the first week of embryogenesis, rising sharply at 1.5 dpf, peaking at 2dpf and approaching zero by 7 dpf (Fig 1B and 1D).
The developmental profile and subcellular localization of H3K9me3 and 5MeC marks suggest that these epigenetic modifications may mark MNi in different phases of elimination. Micronuclei with elevated levels of H3K9me3 were predominantly located in close proximity to the primary nucleus, whereas MNi with elevated levels of 5MeC were typically located at more distal sites (Fig 1A, S1C and S1D Fig). The timing and location of MNi with chromatin repressive marks suggest that H3K9 tri-methylation marks recently formed MNi and that 5MeC may mark older MNi. It seems plausible that DNA methylation might act to ensure transcriptional silencing of material in MNi prior to its complete elimination. Notably, similar interactions between H3K9 and DNA methylation have been observed during heterochromatin formation and chromatin-remodeling in organisms that do not undergo PGR (fungi [27], plants [28], and mammals [29]).
In comparison to 5MeC, the repressive histone mark H3K9me3 showed a somewhat more complex pattern over the course of the cell cycle. This mark localizes to condensed chromosomes during metaphase and persists through telophase/cytokinesis but is essentially absent from interphase nuclei (S1A–S1C Fig, Fig 1A). The presence of H3K9me3 in newly formed micronuclei suggests that micronuclear H3K9me3 marks are remodeled more slowly than their primary nuclear counterparts following M phase. Cell cycle-dependent changes in histone H3 methylation have been reported for mammalian systems, which do not undergo PGR, and appear to be necessary for proper mitotic segregation [30–32]. Moreover, studies in both mammalian and non-mammalian systems have shown that H3K9 methylation is critical for anchoring heterochromatin to the nuclear envelope [33]. Immunolabeling of nuclear envelope markers lamin B1, and nuclear pore o-linked glycoprotein in rearranging embryos reveals that both of these proteins localize to interphase nuclei, but are absent from micronuclei (S2A and S2B Fig). In human, depletion of LMN-B1 and pore complex proteins are associated with nuclear membrane defects in the context of cancer [34]. Taken together, these studies indicate that retention of H3K9me3 in newly formed MNi might play functional roles in maintaining chromatin compaction, positioning eliminated chromatin, or recruiting other structural components of MNi.
Our in-depth analyses of MNi and their associated chromatin modifications revealed other cellular features that appear to be associated with PGR. The most striking among these were numerous anaphases with large amounts of lagging chromatin (Fig 2). Although these lagging anaphases were often visible in sections, the spindle apparatus often spanned more than 50 micrometers in rearranging embryos. As such, wax sections rarely permitted observation of entire anaphases (Fig 2F). To study detailed morphology of lagging anaphases we adapted the passive CLARITY technique (PACT) to whole lamprey embryos [35]. This approach increases the permeability of cells with minimal impact on the morphology of the embryos and effectively eliminates autofluorescence associated with yolk platelets (e.g. Figs 1 and 2). To complement this clearing method, we also optimized methods for DNA staining, fluorescence in situ hybridization, and β-tubulin immunolabeling of cleared lamprey embryos. Altogether, these analyses provide critical perspective on the developmental context of PGR and the dynamic behavior and packaging of eliminated DNA within rearranging cells.
We were able to establish a timeline for the onset and completion of PGR by examining PACT-cleared embryos across the first several days of development, leveraging natural variation in cell division rate during the first day post fertilization. Lagging anaphases were essentially absent during the first five to six cell divisions (e.g. in embryos with 30–60 cells) but abundant in embryos with more than 64 cells, suggesting PGR is initiated at approximately the onset of the seventh cell division (Fig 3). Lagging anaphases were also present at similarly high abundance at 2 dpf but dramatically decreased in frequency thereafter (S3 Table, S3A and S3B Fig). Notably, lagging chromosomes are observed earlier in development than MNi and peak in abundance at earlier developmental stages (Fig 3F; S3B and S3C Fig). We interpret the earlier appearance of lagging anaphases relative to MNi, as indicative of eliminated material initially slated for elimination during metaphase or early anaphase, and secondarily packaged into MNi.
Detailed examination of embryos at 1–3 dpf also revealed a graded series of cellular morphologies that appear to track the progression of DNA loss both within and between cell cycles. These morphological features provide additional perspective on the cellular and mechanistic details of elimination. Below we describe several salient features of eliminated chromatin, including its subcellular organization across the cell cycle and its association with cytoskeletal components.
Within a cell cycle, eliminated chromatin is first identifiable as thread-like structures that are situated between groups of poleward-oriented chromosomes immediately after the metaphase/anaphase transition. As anaphase progresses, eliminated material begins to exhibit distinguishable differences in its apparent motion relative to retained chromosomes. Lagging chromatin is typically oriented parallel to the interpolar microtubules and appeared to be tightly associated with spindle filaments (Fig 4A; S4 Fig). As cells enter telophase, retained sister chromatids begin to decondense and adopt lobate structures consistent with decondensation of somatic chromosomes and recruitment of nuclear envelope proteins. Notably, lagging chromatin does not appear to decondense at this same time and associates with tubulin prior to being packaged into compact MNi (S4 Fig and S5 Fig).
In situ hybridization with Germ1 and other repetitive sequences (Cot1 and 2 DNA) revealed that lagging chromatin was distributed symmetrically across the metaphase plane. Hybridization with Cot1 DNA revealed that the polar ends of migrating (retained) chromosomes are enriched in highly repetitive DNA (S6 Fig), consistent with the interpretation that Cot1 DNA strongly hybridizes to centromeres, as has been observed for other species [36–38]. Notably, labeled Cot1 DNA also localized to the distal ends of some lagging fragments, suggesting that these segments contain active centromeres that are capable of engaging the kinetochore microtubules and (slower) poleward motion [36, 37] (Fig 4B, S1 Movie). Moreover, poleward-oriented regions of lagging chromatin are highly enriched in H3K9me3 (S1B Fig), which is considered a hallmark of constitutive pericentromeric heterochromatin [39, 40]. We interpret the symmetry of labeling and polar orientation of centromeric regions of lagging chromosomes as indicating that a substantial fraction of eliminated material was replicated in the previous cell cycle, packaged into sister chromatids at metaphase and drawn poleward at anaphase, albeit at a slower rate than somatically retained chromosomes.
Direct confocal imaging of fluorescently stained chromosomes (Fig 4C, S2 Movie) and in situ hybridization of Cot2 DNA (Fig 4D) revealed that the equatorial ends of symmetrically stretched sister chromatids often lay in close proximity to one another throughout anaphase. These apparent contacts between sister chromatids exhibit enhanced hybridization to Cot2 DNA, suggesting the possibility that an as-yet undefined class of repetitive sequences may contribute to PGR by anchoring sister chromatids to one another during anaphase (Fig 4D). Notably, Germ1 is not present at these points of contact and is generally located in regions closer to the presumptive centromeres (Fig 4E). Taken together, these observations indicate that some of the eliminated material consists of entire chromosomes or large chromosomal segments and suggest that chromatin/chromatin (or DNA/DNA) contacts between telomeric segments of sister chromatids might contribute to the decelerated migration of these large eliminated fragments.
In addition to these large and longitudinally stretched segments, we also observed globular (presumable acentric) conglomerates of chromatin localized to the equatorial region (Fig 2C, see also Fig 5C). The presence of these conglomerates lends support to the idea that recombinational processes (intra- or inter-chromosomal) or DNA breakage contributes to PGR [4]. It seems plausible that these acentric fragments could be driven toward the equatorial region by the same polar ejection forces that normally act to orient chromosome arms during cell division [41]. The observation that eliminated material consists of both entire chromosomes and smaller chromosomal fragments mirrors observations from hagfish and parasitic nematodes, wherein both entire chromosomes and chromosomal fragments are lost from somatic lineages [2, 19, 42]. To shed further light on patterns of DNA breakage during PGR, we performed immunolabeling with an antibody to the histone variant γ-H2AX, which binds double stranded DNA breaks and recruits repair machinery [43, 44], and employed fluorescent TDT-mediated dUTP nick-end labeling (TUNEL) labeling to more generally detect DNA breaks. Although all other histone variants yielded interpretable signals, attempts to immunolabel γ-H2AX yielded no signal in embryos at 1–5 dpf. The absence of γ-H2AX immunolabeling could reflect either a paucity of double stranded breaks or failure to react with a lamprey γ-H2AX homolog. On the other hand, TUNEL labeling yielded strong and reproducible staining that was localized exclusively to MNi (S7 Fig). Given evidence that MNi represent the last visible sites of eliminated DNA, it seems plausible that TUNEL labeling reflects the degradation of germline-specific DNA within MNi. Taken together these observations indicate that DNA elimination proceeds through an ordered series of events, wherein germline-specific sequences 1) are initially slated for elimination during early anaphase (perhaps metaphase), 2) exhibit slower poleward movement in comparison to retained chromosomes and 3) condense to form MNi where they are methylated and ultimately degraded.
Thus far, analyses of PGR in lamprey have revealed that patterns of gene loss are indistinguishable among diverse somatic cell lineages, which might be interpreted as supporting a simplistic model wherein all germline-specific sequences are eliminated during a single cell cycle [4, 6]. However, in situ hybridization of intact cells with Germ1 appears to support a somewhat more complex model. As mentioned above, lamprey somatic cells possess a single pair of chromosomes that hybridize to the Germ1 probe (S6A Fig), whereas Germ1 hybridizes to several additional chromosomes in germ cells and embryonic cells that have not completed PGR [4]. As such, this marker can be used to track the progression of PGR. In early cell divisions (at 1 dpf) anaphases were observed that contained multiple Germ1 signals interspersed among retained (normally migrating) chromosomes and relatively small amounts of lagging material, consistent with partial elimination of germline-specific sequences (Fig 5A). Variation in the process of elimination is also apparent in later developmental stages, as some anaphases possess two somatic Germ1 signals and small amounts of Germ1-negative lagging chromatin (Fig 5B and 5C). These patterns suggest that cells had undergone at least one previous cycle of DNA elimination, over which they lost all germline-specific copies of Germ1, and were engaged in eliminating additional material at the time of fixation. The interpretation that PGR plays out over several cell cycles is further supported by the frequent observation of lagging chromatin and peripheral MNi within the same cell (S8 Fig). Presumably these peripheral MNi contain material that was eliminated in the previous cell cycle(s).
In this context, it is also worth noting that the earliest elimination events (1 dpf) appear to be associated with subcellular structures that are not observed at later stages. These appear as dense, presumably heterochromatic, structures located near the cleavage furrow with filamentous extensions oriented toward the enveloped nuclei (Fig 4D; S9 Fig). In general, these morphological features seem consistent with the interpretation that some (early) elimination events are characterized by persistent chromatin/chromatin or DNA/DNA contacts and that many of these same segments maintain an association with spindle microtubules. While it is possible that this variation is related to the fact that PGR is occurring in cells of vastly different sizes at 1 vs. 2 dpf (Fig 5E and 5F), it seems possible that the unique structures observed at 1 dpf might also reflect variation in the underlying mechanisms of PGR across early development.
Our analyses underscore the fact that evolution can arrive at diverse solutions to a common problem. Multicellular organisms employ a diversity of epigenetic silencing pathways, including covalent chemical modification of DNA or histones, expression of DNA binding factors (chromatin proteins and noncoding RNAs) that mediate the accessibility of DNA for transcription, and expression of short RNAs that promote degradation or prevent translation of transcripts. In general, these pathways are distributed broadly across diverse eukaryotic lineages, although individual pathways are evolutionarily labile [45–50], being retained in most lineages but absent from others. Programmed DNA elimination is sparsely distributed across the tree of life and likely arose several times over metazoan evolution [19], and in some cases PGR has been shown to act cooperatively with other silencing pathways (e.g. ciliates [11], sciarid flies [12, 13] and zebra finch [14]). It seems likely that each of these independent lineages has evolved its own approaches to achieve the reproducible elimination of DNA, prevent the loss of retained segments, and integrate these mechanisms with existing silencing pathways. As such, each of these lineages holds the potential to provide unique insights into a diversity of conserved (and derived) cellular mechanisms, including those that contribute to the proper segregation of chromosomes, epigenetic silencing, reconstitution of the nuclear envelope, and the maintenance of genome stability.
One notable feature of lamprey PGR is the variability in the content and form of eliminated chromatin across the first three days of development. Observations suggest that elimination events occurring ~1.5–2 dpf often target large regions (entire chromosomes) and appear to involve physical interactions between homologous chromosomes or sister chromatids. Earlier and later elimination events appear to target smaller fractions of the genome. The presence of variability across development raises several questions with respect to the mechanisms and outcomes of lamprey PGR. For example, does DNA loss involve a fixed number of steps/cell cycles? Do all elimination events share a common mechanism, or do new mechanisms/interactions arise later in development? Are later events uniform, or do they result in minor genetic variation across somatic cell lineages [6]? The variability observed over the time course of lamprey PGR is somewhat reminiscent of chromosome elimination in Acricotopus lucidus (Diptera, Chironomidae) [51]. In most cases, all germline-limited chromosomes are lost in a single mitosis, but rarely, one or several chromosomes escapes elimination and segregates with the somatically-retained chromosomes. Based on these observations, it has been suggested that a threshold exists wherein a certain number of hypothetical marks are necessary to drive elimination of A. lucidus chromosomes. As yet, it remains to be determined whether the observed variation apparent among lamprey elimination anaphases is programmatic, cell lineage specific, inherently noisy, or explained by threshold effects.
The analyses presented here reveal several new cellular and molecular details related to developmentally programmed genome rearrangements in lamprey, a species that undergoes PGR in the context of a developmental and cellular biology that is largely conserved with other vertebrates [7–10]. Our analyses indicate that individual segments are slated for elimination during metaphase and are ultimately packaged into compact structures (micronuclei), a subset of which are enriched for repressive chromatin marks. These studies also demonstrate that PGR is initiated at an earlier developmental stage than was previously indicated via PCR-based assays [4] and strongly indicate that PGR is a more protracted process, being completed over the course of several successive cell divisions.
Based on our these new findings, we suggest that efforts to further dissect the mechanisms underlying lamprey PGR should include studies aimed at defining 1) the sequence of, and interactions between, repetitive sequences that occur in regions of contact between some eliminated chromosomes, 2) the role of epigenetic modifications (particularly silencing) in PGR and 3) interactions between eliminated DNA and components of the spindle apparatus/cytoskeleton. In addition to providing critical insights into the cellular and mechanistic basis of PGR, such studies are expected to aid in translating this information to systems wherein large-scale rearrangements and DNA losses are less programmatic and generally deleterious.
Clearing procedure was performed according to Yang et al. [35]. Paraformaldehyde fixed embryos were incubated in hydrogel monomer solution with 5% acrylamide supplemented 0.5% VA-044 overnight. Polymerization was performed at 37°C for 2.5 hours then embryos were washed briefly with PBS, and incubated in 8%SDS, 1x PBS for 5 days at 37°C with gentle shaking. FISH and immunolabeled samples were washed in 1x PBS with 5 buffer changes over the course of a day and transferred into staining solution (1x PBS, pH = 7.5, 0.1 Triton X-100, 0.01% sodium azide).
Spreads of somatic metaphase chromosomes were generated from embryos at 11 dpf. After overnight treatment with 0.1% colchicine, embryos were ground in Dounce homogenizer, incubated with 0.075 M KCl hypotonic solution for 45 minutes at room temperature and fixed in methanol:acetic acid (3:1). Cell suspensions were placed on glass slides and air-dried. Embryos for this and other experiments were produced under the University of Kentucky IACUC protocol number 2011–0848.
Paraffin sections were prepared for immunolabeling and FISH as follows. Sections were deparaffinized in two changes of xylene, gradually rehydrated in a dilution series of ethanol (100, 80, 70% in water), rinsed in water, and placed for overnight incubation in 10 mM sodium citrate buffer (pH 6.0) at 37°C to reduce auto-fluorescence and aid in antigen retrieval. Slides were then washed in PBS before hybridization and immunolabeling.
Probes for in situ hybridization were labeled by nick-translation using direct fluorophores Cyanine 3-dUTP (Enzo Life Sciences, ENZ-42501) or Fluorescein-12-dUTP (Thermo Scientific, R0101) as described previously [52, 53]. Germ1 repeat was obtained from a previously characterized BAC-clone [4] using extraction with Qiagen Large Construct kit (Qiagen Science, 12462). Cot1 and Cot2 fractions were isolated from genomic DNA according to kinetics of reassociation [54], using S1 nuclease to digest single stranded (low copy) DNA [38, 55]. Cot DNA isolation was performed in 1.2XSSC solution as follows: 120°C heating for shearing and denaturing, reannealing at 60°C, and S1 nuclease digestion for 1 hr at 42°C [55].
Whole embryo FISH was performed using modified procedure for cryosections [56]. Briefly, embryos were incubated in 10 mM sodium citrate buffer, pH = 6.0 overnight at 37°C in a rotating incubator, washed in 1x PBS for 1 hour, and then placed in 50% formamide in 2XSSC for 2–3 hours. For hybridization, formamide/SSC solution was replaced with 30 μl hybridization mix consisting of 50% formamide, 10% dextran sulfate, 0.01% sodium azide, and 150 ng labeled DNA-probe. Embryos were pre-incubated for overnight at 37°C to permit penetration of probes, after which probe and target DNA was denatured by heating samples to 75°C for 3 minutes. Following overnight incubation at 37°C samples were washed in 50% formamide in 2XSSC and in 0.4XSSC, 0.3% IGEPAL® CA-630 (Sigma Cat. no. I8896) at 45°C for 10 min each, then in 2XSSC, 0.1% IGEPAL for 10 min at room temperature. DAPI and SYTO-24 counterstain was performed in staining solution at room temperature for, at least, 1 hour for embryos at 2–3 dpf and overnight for 1 dpf.
Fluorescence in situ hybridization of embryonic sections and mitotic spreads was carried out according to standard protocols [56, 57] with minute modifications [52]. Deparaffinized section slides were incubated in 8% sodium thiocyanate solution overnight, pretreated with 10 μg/ml RNase and 0.01% pepsin solutions, denatured in 70% formamide with subsequent dehydration in ethanol series (70, 80, 100%) and hybridized with 100–200 ng of probe overnight in humid chamber at 37°C. For chromosome spreads prehybridization treatments with sodium thiocyanate, RNase, and pepsin solutions were skipped.
Primary antibodies for immunolabeling were as follows: monoclonal anti-5-Methylcytosine (Epigentek, A-1014), polyclonal anti-Histone H3-K9 Trimethyl (Epigentek, A-4036), polyclonal anti-Histone H3-K27 Trimethyl (Epigentek, A-4039), monoclonal anti-Beta Tubulin (Abcam, ab179513), polyclonal anti-Lamin B1 (Boster, PB9611) and monoclonal anti-Nuclear Pore-O-linked Glycoprotein (Thermo, MA1-071). Primary antibodies were diluted 1:100 in 1x PBS, applied to slides, and incubated overnight at 4°C. After washing in PBS and PBST twice for 10 min each, slides were incubated with secondary antibodies to their respective host species at a 1:100 dilution using the following antibodies: Alexa Fluor 488 F(ab')2 fragment of rabbit anti-mouse (LifeTechnologies, A-21204), Alexa Fluor 594 F(ab’)2 fragment of goat anti-mouse, Alexa Fluor 488 chicken anti-rabbit (LifeTechnologies, A-21441). After washing as described in PBS and PBST solutions, slides were mounted using VectaShield-DAPI media (Vector Laboratories, H-1400). Whole embryo immunofluorescence labeling was carried out according to methods previously described for single cell phenotyping [35]. Briefly, PACT-cleared embryos were incubated with primary antibodies (1:100, in PBS containing 10% normal serum of secondary antibody host species (rabbit), 0.1% Triton X-100 and 0.01% sodium azide) for 3 days, replacing antibodies daily. Unbound antibody was removed via PBS washes, and samples were incubated with secondary antibodies (1:100) for 2–3 days then washed for 1 day in PBS prior to incubation with DAPI (50 ng/ml) and imaging media (RIMS: 88% Histodenz (Sigma, D2158) in PBS with 0.1% tween-20 and 0.01% sodium azide, pH to 7.5). All staining and mounting steps were conducted at room temperature with gentle shaking.
TUNEL reactions were performed on paraffin sections from 2 dpf embryos. Slides were deparaffinized, treated with sodium citrate solution as above, and labeling was performed using the Click-iT Plus TUNEL Assay (Life Technologies Cat. no. C10617), according to manufacturer instructions. Samples were permeabilized with proteinase K for 30 min at 37°C. A positive control was generated by treatment with a 1:50 dilution of DNAseI (ThermoFisher Scientific Cat. no. EN0525) in reaction buffer, followed by incubation at room temperature for 30 min. TUNEL assays were performed on experimental and positive control slides simultaneously, then slides were mounted with VectaShield-DAPI media (Vector Laboratories, H-1400).
After FISH and immunolabeling, slides were analyzed with an Olympus-BX53 microscope using filter sets for DAPI, TexasRed, and FITC. Images were captured using CellSence software. For thicker samples, such as sections and embryonic cells after PACT clearing, we used Extended Focal Imaging (EFI) function in order to generate a single deep-focus image. Three-dimensional images of anaphases were obtained using a scanning confocal microscope (Nikon C2) equipped with NIS-Elements AR software. Three-dimensional images were converted in two-dimensional format in NIS Element Viewer. Pseudocolor corrections were performed using Adobe Photoshop CS6. Video recordings were made in NIS Element Viewer using QuickTime media player “Screen recording” function.
The frequency of MNi in paraffin sections was assessed by counting DAPI-stained primary nuclei and small extra-nuclear DAPI-positive structures. For FISH and immunolabeling experiments, MNi were counted as signal-positive when they yielded visible DAPI emission and fluorescence in the specific wavelength corresponding to the fluorophore used for detection. Between 20 and ~200 primary nuclei were counted per slide, depending on stage of development. Fewer nuclei were counted for earlier stages due to the fact that these embryos consist of smaller numbers of larger cells. Counts of MNi, anaphases and lagging anaphases were performed after hybridization of whole embryos with fluorescently-labeled Cot2 DNA in order to improve visualization of eliminated DNA. Frequencies of MNi, anaphases and lagging anaphases were compared between adjacent time points using Pearson’s chi-square test and by calculation of Bayesian central confidence intervals [58].
|
10.1371/journal.pcbi.1006451 | Functional conservation of sequence determinants at rapidly evolving regulatory regions across mammals | Recent advances in epigenomics have made it possible to map genome-wide regulatory regions using empirical methods. Subsequent comparative epigenomic studies have revealed that regulatory regions diverge rapidly between genome of different species, and that the divergence is more pronounced in enhancers than in promoters. To understand genomic changes underlying these patterns, we investigated if we can identify specific sequence fragments that are over-enriched in regulatory regions, thus potentially contributing to regulatory functions of such regions. Here we report numerous sequence fragments that are statistically over-enriched in enhancers and promoters of different mammals (which we refer to as ‘sequence determinants’). Interestingly, the degree of statistical enrichment, which presumably is associated with the degree of regulatory impacts of the specific sequence determinant, was significantly higher for promoter sequence determinants than enhancer sequence determinants. We further used a machine learning method to construct prediction models using sequence determinants. Remarkably, prediction models constructed from one species could be used to predict regulatory regions of other species with high accuracy. This observation indicates that even though the precise locations of regulatory regions diverge rapidly during evolution, the functional potential of sequence determinants underlying regulatory sequences may be conserved between species.
| Regions of the genome that do not encode genes but affect expression of other genes, such as enhancers and promoters, are referred to as regulatory regions. Because of their regulatory functions, it was thought that enhancers and promoters should be evolutionarily conserved. Regulatory regions can be now epigenomically identified because they are marked by specific modifications of histone tails at the chromatin level. Interestingly, when we compare epigenomically identified regulatory regions from different mammals, the specific positions of regulatory regions are often divergent between species. Enhancers in particular are highly divergent between species. In this study, we show that we can find sequence fragments that are statistically enriched in enhancers and promoters of different species, and that the degree of statistical enrichment can explain different levels of evolutionary sequence conservation between enhancers and promoters. We further constructed predictive models of enhancers and promoters using the enriched sequence fragments, and show that these models can not only accurately predict enhancers and promoters of the same species, but works comparably well when applied to other species. These results indicate that even though the specific positions of regulatory regions have diverged between species, the functions of sequence fragments that comprise those regions may be conserved.
| Epigenomic modifications such as histone modifications and DNA methylation play critical roles in development, regulation, and diseases. The study of epigenetic modifications has made great strides in recent decades, and the specific combinations of different epigenome components in distinct biological conditions are rapidly being discovered [1]. In particular, epigenomic profiling is widely used to empirically identify regulatory regions including enhancers and promoters using chromatin immunoprecipitation with massively parallel DNA sequencing (ChIP-seq). For example, genomic regions enriched for histone H3 lysine 27 acetylation (H3K27ac) are considered as active enhancers [2, 3]. On the other hand, enrichment of histone H3 lysine 4 trimethylation (H3K4me3), in particular together with H3K27ac, indicates active promoters [4–6].
Beyond identifying regulatory regions, the next challenge is deciphering what factors determine and affect epigenomes. Among potential factors, the importance of cis-regulatory sequences on the epigenome is well appreciated. Several cis-regulatory sequences based predictive models have been constructed to classify regulatory regions [7–10]. For example, a recent study reported random forest classifier models from the human genome that could predict regulatory regions marked by H3K27ac and H3K4me3 modifications with relatively high accuracy [11].
Even though our understanding of the true nature of the relationship between specific histone modifications and regulatory regions is sure to undergo much more revisions, these technical advances in genome-wide epigenomic profiling brought new approaches to study evolution of regulatory regions. Instead of having to rely on experimentally characterized comparative transcription factor binding assays [12–14] and/or regions that retain sequence similarities [15–18], enhancers and promoters can be identified based on the distribution of specific epigenomic modifications such as H3K4me3 and H3K27Ac across different species [6, 19]. Interestingly, these studies show that at the genome-scale, chromosomal locations of enhancers are highly divergent between species [6, 20, 21]. Promoters are also found in divergent locations, although their positions are more constrained than enhancers, since promoters are typically adjacent to transcription units (e.g. [6]). Thus, while regulatory regions can be reliably predicted from sequences within specific genomes [7–11], the precise locations of regulatory regions, in particular of enhancers, diverge rapidly during evolution [6, 18, 20, 21].
It is not necessarily straightforward to reconcile these two aspects of regulatory regions. In the simplest scenario, functional regions such as enhancers and promoters should be evolutionarily conserved since they are subject to purifying selection. Indeed, this idea has been successfully used to identify non-coding sequences with regulatory functions [16, 17, 22, 23]. However, at the genome-scale, regulatory regions harbor little sequence similarities and their locations are highly divergent. Rapid turnover of transcription binding sites [12, 24, 25] and transcription rewiring [26–28] can explain some aspects of regulatory sequence evolution, but many questions still remain [29, 30].
Here, utilizing the wealth of comparative data on epigenomically determined enhancers and promoters, we investigated whether we could identify specific sequence fragments that constitute enhancers and promoters, and if so, whether such sequence fragments were evolutionarily conserved between species. We first performed an exhaustive search to identify sequence fragments that are statistically over-represented in experimentally identified enhancers and promoters of several mammals [6]. A unique aspect of our study is that we focused on distinguishing regulatory regions from nearby regions. Genomic sequences of mammals such as humans are highly heterogeneous in many aspects such as GC contents, transposable element contents, genic contents, and other aspects [31, 32]. By comparing regulatory regions to their nearby non-regulatory regions, we identified sequence fragments that distinguished regulatory regions from its local genomic backgrounds. Our comprehensive exhaustive search revealed numerous sequence fragments that were significantly enriched in regulatory regions compared to nearby regions. Due to the nature of the exhaustive search, some of the identified sequence fragments may be inter-related. To overcome this limitation and identify a subset of sequence fragments that are statistically independent, and to construct prediction models to test evolutionary hypotheses, we employed a machine learning method. Specifically, we used the least absolute shrinkage and selection operator (LASSO) method [33], which can effectively select one variable among the set of highly correlated variables [34]. The LASSO method is also excellent at prediction accuracy [11, 35].
From these procedures, we discovered numerous sequence fragments that are statistically enriched in experimentally verified regulatory regions (referred to as ‘sequence determinants’ henceforth). Intriguingly, sequence determinants obtained from enhancers and promoters show remarkable differences with respect to their impact on functional regions. Moreover, even though sequence determinants themselves exhibit only moderate overlaps between species, prediction models constructed using sequence determinants from different species could be inter-changed to perform as well as prediction models from the focal species. We discuss potential implications of these findings.
We used experimental annotations of liver enhancers and promoters from a previous study [6]. Following the definition in this study [6], we considered enhancers to be regions marked only with the H3K27ac mark and promoters to be regions marked with H3K4me3 (with or without H3K27ac). We selected data from seven ‘high-quality’ mammalian genomes as indicated in [6], including Home sapiens (human), Macaca mulatta (macaque), Bos taurus (cow), Sus scrofa (pig), Canis familiaris (dog), Rattus norvegicus [32], and Mus musculus (mouse). Each enhancer or promoter was designated as foreground, and a segment of the same length 100,000 base-pairs (100kb) apart from the foreground was selected as the background. We used these ‘regional’ backgrounds to control for potential chromosome effect and/or regional effects. The distance of 100kb between the foreground and background was selected since several genomic features such as linkage disequilibrium blocks and GC contents show correlations that extend to ~ 100kb [32, 36]. We obtained the genome sequences using the R Bioconductor libraries “BSgenome” [37]. Backgrounds that had greater than 50% of nucleotides missing (not sequenced) were discarded (Table 1), and put information on overlapped proportions between foreground and background in S1 Table.
Those enhancers and promoters found in orthologous locations across species were identified as conserved (Table 1). Specifically, for each human enhancer or promoter we retrieved the 17 eutherian EPO multiple alignment using Ensembl REST API [38] and determined if the region was conserved or not based on whether all other 6 species also showed the same histone mark(s) in the orthologous region. For species with different genome assemblies in the alignment, we converted the coordinates using Ensembl assembly converter [39].
We examined whether specific sequence fragments in the foreground were over-represented compared to the backgrounds by statistical testing. We used sliding windows with a specific length (from 6-mers to 15-mers), moving from the 5’ end to the 3’ end in each foreground or background (Fig 1). As the window moved by a base-pair (bp), a sequence fragment within that bin was captured and recorded. Following this sliding window analysis, counts of each sequence fragment in the foreground and background were obtained. For each sequence fragment, we constructed a 2×2 contingency table that contained counts of a sequence determinant in each of foreground and background region (Table 2), and we used the odds ratio (OR) as a measure of over-representation in foreground, compared to background. The magnitude of OR indicated how strongly over-enriched a specific element was in regulatory regions, which we also referred to as ‘effect size’ in this study.
We used the χ2 test to test the following null and alternative hypotheses:
H0:OR=1,
(1)
H1:notH0.
(2)
If the expected count of a sequence fragment in any of the cell in the 2×2 contingency table was lower than 5, we used the Fisher’s exact test instead. The resulting P-values were corrected for multiple testing using the false discovery rate (FDR) approach [40]. Following these procedures, a ‘sequence determinant’ in the statistical sense was identified as a sequence fragment whose FDR Q-value was equal to or less than 0.05 and the OR was greater than 1. In the process, we tested only sequence determinants that appeared over 100 times to avoid selecting rare sequence determinants of negligible biological relevance. For example, for 15-mers in the human enhancer data set, most sequence fragments (63 million out of 70 million) occurred only once. We repeated this procedure for each of the seven species and identified ‘species sequence determinants’.
We identified ‘common sequence determinants’ as sequence fragments that are enriched in foreground regions compared to the background regions across the seven mammalian species. For the purpose, we used the Cochran-Mantel-Haenzel (CMH) test [41] to identify enrichment of sequence determinants from multiple data sets using a conditional variable, which is a nominal covariate such as the species index [41, 42]. The CMH test is also equivalent to the score type test of logistic regression, which has advantages in the handling of sparse count data sets [42]. Consequently, we used the CMH to test the null hypothesis,
H0:OR|species=1,whereOR|speciesistheconditionalORinpresenceofthespeciesindex.
(3)
H1:notH0.
(4)
Common sequence determinants were then defined as those whose OR|species>1 for all species and FDR Q-value from CMH ≤ 0.05.
We constructed prediction models that yield predictive scores for each region. We used the least absolute shrinkage and selection operator (LASSO) method [33], which excels at prediction accuracy as well as covariate selection [11, 35]. In the LASSO model, each foreground or background region was regarded as a binary observation (foreground = 1, background = 0). The relative frequency of each sequence determinant was regarded as an explanatory variable. Because the space of all significant sequence determinants was extremely large (S2 and S3 Tables), including all determinants in the LASSO model was not computationally feasible. Instead, we selected 10,000 sequence determinants, sampled according to their distribution of GC content and fragment length, to incorporate in the LASSO models using a stratified sampling approach [43]. Specifically, we stratified the whole sequence determinants by the combination of GC content (ten uniform intervals: [0~0.1],…, (0.9~1.0]) and length (ten lengths: 6,…,15bp). Then we selected samples from each of the stratified subsets so that its number out of the 10,000 was proportional to the number of determinants in the specific subset among the total determinants. To train LASSO models and estimate coefficient of each determinant, we used the R function “glmnet” from the package “glmnet” using R 3.4.0.
To construct prediction models, we used both the 10,000 species sequence determinants and the 10,000 common sequence determinants as input variables, so that we can compare the prediction performances of species determinants and common determinants. We performed two types of predictions. First, we performed same-species prediction, which evaluates prediction AUC through a 10-fold cross-validation process [11, 35, 44, 45]. During the 10-fold cross-validation process, an optimal penalty parameter that provides the smallest test AUC is chosen. We regarded the smallest test AUC as same-species prediction AUC. For inter-species prediction, we used the optimal parameter to construct a prediction model from whole data set of a species and applied the model to the other species to calculate inter-species prediction AUCs. Workflow from the exhaustive search to LASSO is depicted in Fig 1. In most prediction results, we provided two types of AUC, the first one is receive operating characteristic AUC (ROC-AUC) for general performance of prediction and the second one is precision-recall AUC (PR-AUC) for robustness of performance regardless of the ratio between numbers of foreground and background [46].
Among several machine-learning methods, we selected LASSO because of its ability to reduce the number of input variables so that those are not redundant and are statistically meaningful. However, other machine learning methods might be useful as well. For example, when many of sequence determinants have strong relationship in terms of correlation, elastic net that can capture more input variables would be useful to improve prediction performances [47].
We examined the presence of transcription factor binding sites (TFBS) in the sequence determinants using TOMTOM [48]. This tool assesses the similarity between individual sequence input and specific TFBS databases and provides P-values and Q-values adjusted by FDR. Known TFBS compiled in the JASPAR 2014 Core vertebrate database [49], the HOCOMOCOv10_HUMAN and the HOCOMOCOv10_MOUSE [50] were used. We summarized the proportion of significant (P <0.05) TFBS hits as ‘TFBS frequency’. For example, each human sequence determinant was compared to the 641 known TFBS in the HOCOMOCOv10_HUMAN database. The number of significant comparisons out of the total 641 comparisons was referred to as ‘TFBS frequency’. Due to the probabilistic nature of TF binding and the fact that sequence determinants might encode partial or full TFBS, TFBS frequency indicates versatility of a sequence determinant that can be a motif for TFBS binding. For instance, the CAGCCC determinant from the human genome yielded 18 of 641 significant hits, thus TFBS frequency of the determinant was 2.8%. We also used–log10min(P) instead of TFBS frequency to evaluate the best match between a k-mer and the motifs in the database.
Sequence determinants from the exhaustive search as well as from the LASSO prediction models were further analyzed to explore relationships between their effect sizes and several biological factors such as GC content and TFBS binding properties. For this analysis, we used the following linear model;
log2(OR)i∼GCcontenti+TFBSfrequencyi+GCcontenti×TFBSfrequencyi+εi,
(5)
where i is the index of each sequence determinant and εi ~N(0,σ2). In this model, we log2 transformed the OR values to improve normality. We applied the model to enhancer and promoter sequence determinants from common, human, and mouse sets.
To identify sequence fragments that are significantly enriched in enhancers or promoters compared to nearby background regions (sequence determinants), we first performed an exhaustive search. Briefly, we examined sequence fragments of lengths from 6 to 15 bp, using a sliding window approach (Fig 1). We tested statistical over-representation of the specific sequence fragment in the enhancers or promoters compared to their backgrounds using a contingency table test based on their ORs. The P-values were adjusted via the false discovery procedure [40] (Materials and Methods).
Following these procedures, we identified numerous sequence determinants associated with enhancers and promoters of each species (referred to as ‘species sequence determinants’, Materials and Methods). Fig 2(A) and 2(B) show the numbers of significant sequence determinants from human enhancers and promoters based on their OR and length. The majority of sequence determinants in enhancers and promoters were found in 7–11 bps. Human enhancer determinants were slightly yet significantly longer than promoter determinants (mean lengths for human enhancers and promoters were 9.20 and 9.01, P <1×10−5 by two sample t-test). However, there was no consistent pattern across the seven mammals when comparing the length of sequence determinants in enhancers and promoters. Sequence determinants were also generally GC-rich and TFBS-rich compared to non-significant sequence fragments (see below). Remarkably, with respect to OR, sequence determinants from enhancers and promoters were highly distinct. Strongly enriched sequence determinants, such as those with OR ≥ 2.0, were 140-fold more abundant in promoters than in enhancers (Fig 2). Accordingly, the ORs of sequence determinants were significantly higher in promoter sequence determinants than in enhancer sequence determinants (P < 10−15 by Wilcoxon’s rank sum-test in all seven species, Fig 3).
We then examined sequence determinants that occurred more frequently than expected in all seven mammalian species, which we referred to as ‘common sequence determinants’ (Materials and Methods). Similar to the results from the above analysis, common sequence determinants had higher ORs in promoters than in enhancers (P < 10−15 by Wilcoxon’s rank sum-test, Fig 2(C) and 2(D), S4 Table). When we compared the entries of common sequence determinants to those of species sequence determinants, we found that 39% and 57% of all human enhancer and promoter determinants overlapped with common enhancer and promoter sequence determinants, respectively (S5 Table). Therefore, regardless of their species-wise distribution, sequence determinants that mark promoters tended to have significantly greater OR thus presumably stronger effects on regulatory potential of target regions in terms of marginal effect size, compared to those found in enhancers.
The exhaustive search allowed us to identify all sequence determinants that were marginally enriched. However, some sequence determinants might be highly correlated with each other, because they were extracted from overlapping regions (Fig 1). The LASSO approach is capable of selecting one variable among the highly correlated variable sets, in addition to selecting variables of substantial effect [33]. Therefore, we next used the LASSO approach to select essential variables among the many correlated variables, and to construct prediction models that discriminate enhancers and promoters from their corresponding background regions (Materials and Methods). The total numbers of sequence determinants from the human enhancers and promoters were 107,287 and 101,625, respectively (S2 and S3 Tables). AUCs increased as the number of input sequence determinants increased, to stabilize around 7,000 sequence determinants (S1 Fig). We thus chose 10,000 sequence determinants for each set of sequence determinants using a stratified sampling approach [43], to select a subset that is representative of the original distribution with respect to GC contents and lengths (Materials and Methods). Following these steps, prediction models were constructed for both same-species prediction and inter-species prediction.
We investigated the distribution of ORs and the lengths of selected sequence determinants from the LASSO approach (‘LASSO-selected sequence determinants’), and from same-species prediction. The same-species prediction model of human enhancers and promoters had a total of 4321 and 1343 LASSO selected sequence determinants, respectively (S6 and S7 Table). Consistent with the results from the exhaustive search, marginal ORs from the enhancer models were significantly lower than those from the promoter models in all species (Wilcoxon test, P < 10−15, S2 Fig).
We investigated the relative frequencies of individual LASSO-selected sequence determinants in foreground and background regions, shown as density plots in S3 Fig. In promoters, marginal density of the relative frequencies of LASSO-selected sequence determinants is highly distinct from that of the background, which is consistent with the high effect size of LASSO-selected promoter sequence determinants. On the other hand, marginal densities of LASSO-selected enhancer sequence determinants are similar to those in the background. This observation indicates that in addition to having weaker marginal effects than promoter sequence determinants, the frequency distribution of enhancer sequence determinants is similar between foreground and background.
Interestingly, LASSO selected sequence determinants were significantly longer for enhancers than for promoters (mean lengths of 9.22 in enhancers and 8.32 in promoters in human, P <1×10−15 by two sample t-test, S6 and S7 Table). This pattern was consistent in other species (P < 10−15 by two-sample t-test in all cases). When we applied LASSO approach to 10,000 common sequence determinants, we observed similarly significant differences of effect size and length between enhancer and promoter sequence determinants (S2 and S4 Figs).
We examined two aspects of sequence determinants to understand what features affect enhancer and promoter potentials of specific sequence fragments. Specifically, we used a linear model to analyze the effect of the frequency of G and C nucleotides (GC content) and the frequency of transcription factor binding sites (TFBS frequency). The effect sizes of sequence determinants were response variables, and GC content, TFBS frequency, and their interaction term were explanatory variables. When we analyzed the results of the LASSO-selected sequence determinants, several patterns became clear. First, this model explained a large amount of variation observed in promoter sequence determinants, but only a modest portion of those in enhancer sequence determinants (Table 3). Nevertheless, we found that main factors of GC content and TFBS frequency were positively correlated with the log2-transformed OR of sequence determinants both in enhancers and promoters (Table 3, S8 and S9 Tables). However, interaction terms between the two main factors were significantly negative only in promoters. Thus, while GC content and TFBS frequency worked additively to determine the strength of regulatory potential for enhancer sequence determinants, these two factors were antagonistic with each other in promoter sequence determinants (Fig 4(A) and 4(B)). This observation is consistent with previous studies that found a lack of transcription factor binding enrichment at GC-rich promoters compared to GC-poor promoters [51]. We also evaluated–log10min(P) instead of TFBS frequency to evaluate the best match between a k-mer and the motifs in the database, and obtained highly similar results for the same models (S10 Table).
In summary, TFBS frequency was positively correlated with effect size in both of enhancer and promoters when GC content was low. On the other hand, the estimated coefficients of GC content and TFBS frequency were higher in promoters than in enhancers, indicating that the effects of these factors were stronger in promoters compared to in enhancers. Accordingly, the R2 of the linear models were substantially higher for promoters than for enhancers (Table 3, S8 and S9 Table). Second, the relationships between GC contents and TFBS frequency were negative in both of enhancer and promoter analysis (Fig 4(C) and 4(D)). Accordingly, sequence determinants that were GC-rich tended to lack TFBS, and low GC sequence determinants tended to harbor more TFBS than high GC sequences [51]. The whole set of sequence determinants obtained from exhaustive search yielded similar results (S11 Table).
The prediction accuracy of the human promoter same-species prediction model was very high, with an AUC of 0.97 (Fig 5). Same-species prediction models from other six species exhibited similarly high AUCs (S12 and S13 Table), indicating that promoters can be accurately predicted from sequence determinants. We also evaluated prediction AUCs using 10,000 non-sequence determinants, while matching the distributions of GC content and length as those of sequence determinants. We then constructed prediction models using LASSO for enhancers and promoters in human and mouse, respectively. We iterated the process five times to measure variability of the AUCs. Results are shown in S6 Fig. The AUCs of models using non-sequence determinants were lower than AUCs with sequence determinants. For example, human and mouse enhancer prediction AUCs with non-sequence determinants showed 0.507 and 0.002, and 0.500 and 0.007 for mean and standard deviation, respectively. These results indicate that non-sequence determinants had poor prediction performances. In case of promoters, the mean and standard deviation of AUCs were 0.636 and 0.006 for human, and 0.608 and 0.004 for mouse, respectively. These values were higher than those of enhancers, likely reflecting the effect of GC contents (e.g., [52]). Nevertheless, they were substantially lower than the AUCs with sequence determinants, indicating that sequence determinants have superior prediction performances than non-sequence determinants.
Next, we tested if prediction models constructed from one species could be used in different species, to investigate if different genomes use similar sequence determinants to encode promoters. Indeed, when we calculated AUCs of inter-species prediction between seven species of promoters, the AUCs were all above 0.9, indicating high accuracy (Fig 5).
On the other hand, the LASSO prediction models of enhancers had the following differences from those of promoters. First, the enhancer models using 10,000 species determinants had 2.5- to 4.2-fold greater numbers of explanatory variables than the promoter models (S12 Table). However, their AUCs were generally lower than those of the promoter models (Fig 5). We found that same-species prediction AUCs for enhancer models were greater than 0.7, and the highest was when mouse model were used to predict mouse enhancers, 0.76 (S12 Table). Nevertheless, inter-species prediction results using enhancer models showed similar AUCs to same-species enhancer predictions (Fig 5).
We tested if the high inter-species prediction accuracies were driven by the presence of highly conserved regulatory elements across different mammalian species. The proportions of conserved enhancer regions among the seven species were much smaller than those of promoter regions, as previously described [6] (Table 1). Interestingly, we observed similar AUCs before and after removing highly conserved regulatory regions at both enhancers and promoters (S14 Table), suggesting that conserved regulatory regions were not responsible for the high predictabilities across species. We then extracted 10 subsets of 10,000 sequence determinants from human enhancer and promoter sequence determinants (all subsets were mutually exclusive with each other subset) and constructed LASSO models to apply to the same-species (human) prediction and inter-species (mouse enhancer) prediction. We found that the AUCs of these 10 subsets were highly similar (S6 Fig). Thus, even though the regulatory regions themselves were not conserved in terms of their precise location, mammalian enhancers and promoters have inter-changeability in terms of prediction between species.
We also constructed LASSO models using 10,000 common sequence determinants from all seven species. AUC values for promoter prediction were highly similar to those obtained from models using species sequence determinants (Fig 5, S13 Table), indicating that sequence fragments that were commonly enriched in all 7 species harbor sufficient signals for promoter prediction. On the other hand, enhancer prediction results using 10,000 common sequence determinants showed slight decrease of AUC compared to same-species prediction (mean AUC of prediction with species determinants: 0.723, that with common determinants: 0.679). Interestingly, mean numbers of LASSO selected common sequence determinants were significantly lower than the species ones in enhancers (1613 and 3970 for common sequence determinants and species sequence determinants, respectively; P <1×10−5 by paired t-test), while they were not significantly different in promoter models (1138 and 1342 for common sequence determinants and species sequence determinants, respectively; P = 0.2114 by paired t-test). This implies that each of the common enhancer sequence determinants may have higher predictive capabilities than species sequence determinants.
While background and foreground of enhancers exhibit similar GC distribution, foreground regions of promoters are substantially skewed towards GC-rich regions (the average difference was 10.0%, higher in promoters than in enhancers) (S7 Fig). Therefore, we investigated how GC content difference between foreground and background might affect prediction analyses. First, to measure the impact of GC content alone in prediction performances, we calculated AUCs using only GC content as a predictor (Table 4). Second, we constructed LASSO models using sequence determinants of low-GC content (GC content ≤0.5) to measure prediction performances without effects of high GC content sequence determinants. For this analysis, we randomly selected 10,000 sequence determinants with stratification of GC content and sequence length. These results were then compared to those of the original AUCs.
We found the AUCs using only GC content reflected the amount of GC content differences between foreground and background (S7 Fig). For example, average AUCs using only GC content were 0.589 and 0.837 in enhancers and promoters, respectively. However, both of those AUCs were considerably lower than the original AUCs (differences of 0.134 and 0.107 in enhancers and promoters, respectively), meaning that GC content could not explain all of the variation between foreground and background. This observation is consistent with a prior study utilizing a similar approach [52]. Moreover, models with low-GC sequence determinants had higher AUCs than those using only GC contents. In other words, models without high GC content sequence determinants outperformed the AUCs with only GC contents. Interestingly, mean AUCs with low-GC sequence determinants in enhancers were even higher than those of the original AUCs, which may imply that low-GC enhancers sequence determinants had better prediction performances than high-GC sequence determinants when they were jointly used for prediction. In conclusion, prediction performances of the sequence determinants detected by LASSO cannot be attributed to their GC contents.
Understanding specific histone modifications marking enhancers and promoters has opened the way to identify these regions using ChIP-seq, which complements and scales up traditional transcription factor binding assays [1, 6, 53]. Even though our understanding of the exact molecular nature of regulatory regions continues to improve, technical advances in epigenomic assays have opened a new opportunity to study evolution of regulatory regions using unbiased genome-wide epigenomic profiling. We were motivated by two observations: that regulatory regions identified from epigenomic assays can be predicted with high accuracy in case of same-species prediction [7–11], yet that they are highly divergent between different species [6, 18, 20, 21]. The fact that regulatory regions can be predicted with high accuracy implies that specific sequence fragments can encode regulatory function. Indeed, previous studies often referred to such fragments as cis-regulatory motifs. Since they encode function, they are likely to be subject to natural selection (largely purifying selection) and thus evolutionarily conserved. However, genome-wide studies indicate that regulatory regions, especially enhancers, are highly divergent between species. To investigate this potentially paradoxical pattern of evolution of regulatory regions, we used a powerful approach to examine every possible sequence fragments for their statistical enrichment in experimentally verified enhancers and promoters of seven mammalian species. This approach, which we named exhaustive search, revealed that numerous sequence fragments were statistically over-represented in enhancers and promoters (which we named as sequence determinants).
Sequence determinants underlying enhancers and promoters exhibited intriguing differences with respect to their degree of enrichment (effect size), GC content, and the frequencies of known TFBS. Notably, the degree of statistical enrichment was significantly higher for promoter sequence determinants compared to enhancer sequence determinants. This observation suggests that sequence determinants may have greater impacts on the regulatory potential of promoters than of enhancers. This idea is also consistent with the fact that promoters are more evolutionarily conserved than enhancers [6].
We next applied a machine-learning method, LASSO, to reduce interdependence among sequence determinants and construct prediction models based on the non-redundant sequence determinant set. Same-species prediction models generated from these sequence determinants had high AUCs for enhancers and promoters (Fig 5 and S12 and S13 Tables), affirming the predictor power of sequence determinants [11, 52]. The AUCs from these models are on par with those from previous studies that utilized different approaches (e.g., [11]). We observed that enhancer models utilized greater numbers of predictors yet exhibited lower accuracy compared to promoter models, which can be explained by promoter sequence determinants associated with significantly higher effect sizes compared to enhancer sequence determinants (Figs 2 and 3, S2 and S4 Figs). Furthermore, we applied prediction models generated from one mammal to other mammals, to directly test whether sequence determinants from one species could be used to predict regulatory regions in other species. Remarkably, even though the sequence determinants themselves had only moderate overlaps between species (S5 Table), models constructed from one species could predict promoters in other species with high accuracies (S12 and S13 Tables). As for enhancer models, AUCs from inter-species prediction models were also comparable to same-species predictions (Fig 5). In other words, the extent to which prediction models could be inter-changed between species was similar between enhancers and promoters (Fig 5).
We used a cutoff effect size for sequence determinants as 1, for the following reasons. First, many sequence determinants have extremely low p-values despite low effect sizes due to their abundance, especially those with shorter lengths. For example, 25% of human enhancer sequence determinants among those of top 10,000 lowest p-values have effect sizes smaller than 1.2. Second, when we constructed a human enhancer prediction model using randomly selected 10,000 sequence determinants with effect sizes smaller than 1.2, the resulting AUC was 0.715, which is equivalent to the original AUC. Moreover, when we applied this model to mouse, the inter-species AUC was 0.680, even higher than the original AUC (0.647). Therefore, setting an arbitrary cutoff value is likely to result in the loss of true sequence determinants that are important in terms of prediction performances.
Integrating the main findings that 1) there are a large number of sequence determinants that potentially contribute to the regulatory roles of enhancers and promoters; 2) the strength of statistical enrichment of sequence determinant is greater for promoters, which are more evolutionarily conserved than enhancers; 3) prediction accuracies of models generated using sequence determinants from different species are comparable to each other, we hypothesize the following. Even though the specific motifs that encode regulatory regions are different between species [6, 18, 20, 21], the function of specific sequence determinants could be conserved between species. There may exist a large reservoir of potential sequence determinants that can contribute to regulatory regions of many species.
|
10.1371/journal.pbio.1002541 | Relative Citation Ratio (RCR): A New Metric That Uses Citation Rates to Measure Influence at the Article Level | Despite their recognized limitations, bibliometric assessments of scientific productivity have been widely adopted. We describe here an improved method to quantify the influence of a research article by making novel use of its co-citation network to field-normalize the number of citations it has received. Article citation rates are divided by an expected citation rate that is derived from performance of articles in the same field and benchmarked to a peer comparison group. The resulting Relative Citation Ratio is article level and field independent and provides an alternative to the invalid practice of using journal impact factors to identify influential papers. To illustrate one application of our method, we analyzed 88,835 articles published between 2003 and 2010 and found that the National Institutes of Health awardees who authored those papers occupy relatively stable positions of influence across all disciplines. We demonstrate that the values generated by this method strongly correlate with the opinions of subject matter experts in biomedical research and suggest that the same approach should be generally applicable to articles published in all areas of science. A beta version of iCite, our web tool for calculating Relative Citation Ratios of articles listed in PubMed, is available at https://icite.od.nih.gov.
| Academic researchers convey their discoveries to the scientific community by publishing papers in scholarly journals. In the biomedical sciences alone, this process now generates more than one million new reports each year. The sheer volume of available information, together with the increasing specialization of many scientists, has contributed to the adoption of metrics, including journal impact factor and h-index, as signifiers of a researcher’s productivity or the significance of his or her work. Scientists and administrators agree that the use of these metrics is problematic, but in spite of this strong consensus, such judgments remain common practice, suggesting the need for a valid alternative. We describe here an improved method to quantify the influence of a research article by making novel use of its co-citation network—that is, the other papers that appear alongside it in reference lists—to field-normalize the number of times it has been cited, generating a Relative Citation Ratio (RCR). Since choosing to cite is the long-standing way in which scholars acknowledge the relevance of each other’s work, RCR can provide valuable supplemental information, either to decision makers at funding agencies or to others who seek to understand the relative outcomes of different groups of research investments.
| In the current highly competitive pursuit of research positions and funding support [1], faculty hiring committees and grant review panels must make difficult predictions about the likelihood of future scientific success. Traditionally, these judgments have largely depended on recommendations by peers, informal interactions, and other subjective criteria. In recent years, decision makers have increasingly turned to numerical approaches such as counting first or corresponding author publications, using the impact factor of the journals in which those publications appear, and computing Hirsch (i.e., h-index) values [2]. The widespread adoption of these metrics and the recognition that they are inadequate [3–6] highlight the ongoing need for alternative methods that can provide effectively normalized and reliable data-driven input to administrative decision making, both as a means of sorting through large pools of qualified candidates and as a way to help combat implicit bias.
Though each of the above methods of quantitation has strengths, accompanying weaknesses limit their utility. Counting first or corresponding author publications does on some level reflect the extent of a scientist’s contribution to his or her field, but it has the unavoidable effect of privileging quantity over quality and may undervalue collaborative science [7]. Journal impact factor (JIF) was for a time seen as a valuable indicator of scientific quality because it serves as a convenient, and not wholly inaccurate, proxy for expert opinion [8]. However, its blanket use also camouflages large differences in the influence of individual papers. This is because impact factor is calculated as the average number of times articles published over a 2-y period in a given journal are cited; in reality, citations follow a log-normal rather than a Gaussian distribution [9]. Moreover, since practitioners in disparate fields have differential access to high-profile publication venues, impact factor is of limited use in multidisciplinary science-of-science analyses. Despite these serious flaws, JIF continues to have a large effect on funding and hiring decisions [4,10,11]. H-index, which attempts to assess the cumulative impact of the work done by an individual scientist, disadvantages early-career stage investigators; it also undervalues some fields of research by failing to normalize raw citation counts [6].
Many alternative methods for quantifying scientific accomplishment have been proposed, including citation normalization to journals or journal categories [12–19]; note that one of these is a previously described Relative Citation Rate [19], which should not be confused with the method we describe here. Other methods include citation percentiles [14,20], eigenvector normalization [21,22], and source normalization [13,23]; the latter includes both the mean normalized citation score (MNCS) [17] and source-normalized impact per paper metrics [15,17,22–26]. Although some of these methods have dramatically improved our theoretical understanding of citation dynamics [27–30], none have been widely adopted. To combine a further technical advance with a high likelihood of widespread adoption by varied stakeholders, including scientists, administrators, and funding agencies, a new citation metric must overcome several practical challenges. From a technical standpoint, a new metric must be article level, field-normalized in a way that is scalable from small to large portfolios without introducing significant bias at any level, and correlated with expert opinion. From an adoption standpoint, it should be freely accessible, calculated in a transparent fashion, and benchmarked to peer performance in a way that facilitates meaningful interpretation. Such an integrated benchmark, or comparison group, is not used by any currently available citation-based metric. Instead, all current measures aggregate articles from researchers across disparate geographical regions and institutional types, so that, for example, there is no easy way for primarily undergraduate institutions to directly compare the work they support against that of other teaching-focused institutions or for developing nations to compare their research output to that of other developing nations [31]. Enabling these and other apples-to-apples comparisons would greatly facilitate decision making by research administrators.
We report here the development and validation of the Relative Citation Ratio (RCR) metric, which is based on the novel idea of using each article’s co-citation network to field- and time-normalize the number of citations it has received; this topically linked cohort is used to derive an expected citation rate (ECR), which serves as the ratio’s denominator. As is true of other bibliometrics, article citation rate (ACR) is used as the numerator. Unlike other bibliometrics, though, RCR incorporates a customizable benchmarking feature that relates field- and time-normalized citations to the performance of a peer comparison group. RCR also meets or exceeds the standards set by other current metrics with respect to the ambitious ideals set out above. We use the RCR metric here to determine the extent to which National Institutes of Health (NIH) awardees maintain high or low levels of influence on their respective fields of research.
Choosing to cite is the long-standing way in which one scholar acknowledges the relevance of another’s work. However, the utility of citations as a metric for quantifying influence has been limited, primarily because it is difficult to compare the value of one citation to another; different fields have different citation behaviors and are composed of widely varying numbers of potential citers [32,33]. An effective citation-based evaluative tool must also take into account the length of time a paper has been available to potential citers, since a recently published article has had less time to accumulate citations than an older one. Finally, fair comparison is complicated by the fact that an author’s choice of which work to cite is not random; a widely known paper is more likely to be referenced than an obscure one of equal relevance. This is because the accrual of citations follows a power law or log-normal pattern, in accordance with a process called preferential attachment [27,33,34]. Functionally this means that, each time a paper is cited, it is a priori more likely to be cited again.
An accurate citation-based measure of influence must address all of these issues, but we reasoned that the key to developing such a metric would be the careful identification of a comparison group, i.e., a cluster of interrelated papers against which the citation performance of an article of interest, or reference article (RA), could be evaluated. Using a network of papers linked to that RA through citations occurred to us as a promising possibility (Fig 1). There are a priori three types of article-linked citation networks [35]. A citing network is the collection of papers citing the RA (Fig 1A, top row), a co-citation network is defined as the other papers appearing in the reference lists alongside the RA (Fig 1A, middle row), and a cited network is the collection of papers in the reference list of the RA (Fig 1A, bottom row).
All three types of networks would be expected to accurately reflect the interdisciplinary nature of modern biomedical research and the expert opinion of publishing scientists, who are themselves the best judges of what constitutes a field. By leveraging this expertise, networks define empirical field boundaries that are simultaneously more flexible and more precise than those imposed by traditional bibliometric categories such as “biology and biochemistry” or “molecular biology.” An analysis of the co-citation network of a sample RA illustrates this point. The RA in the bottom panel of Fig 1B describes the identification of new peptides structurally similar to conotoxins, a little-known family of proteins that has begun to attract attention as the result of recent work describing their potential clinical utility [36]. Although the papers in this network are all highly relevant to the study of conotoxins, they cross traditional disciplinary boundaries to include such diverse fields as evolutionary biology, structural biology, biochemistry, genetics, and pharmacology (Fig 1C).
Unlike cited networks, citing and co-citation networks can grow over time, allowing for the dynamic evaluation of an article’s influence; as illustrated by the example above, they can also indicate whether or not an article gains relevance to additional disciplines (Fig 1B and 1C). An important difference between citing and co-citation networks, however, is size. Since papers in the biomedical sciences have a median of 30 articles in their reference lists, each citation event can be expected to add multiple papers to an article’s co-citation network (Fig 1D) but only one to its citing network. The latter are therefore highly vulnerable to finite number effects; in other words, for an article of interest with few citations, small changes in the citing network would have a disproportionate effect on how that article’s field was defined. We therefore chose to pursue co-citation networks as a way to describe an individual paper’s field.
Having chosen our comparison group, we looked for a way to test how accurately co-citation networks represent an article’s field. One way to characterize groups of documents is to cluster them based on the frequency at which specific terms appear in a particular document relative to the frequency at which they appear in the entire corpus, a method known as term frequency–inverse document frequency (TF-IDF) [37]. This is not a perfect approach, as it is possible to use entirely different words to describe similar concepts, but positive matches can be taken as a strong indication of similarity. Frequency of word occurrence can be converted into vectors so that the cosine of the angle between two vectors is a measurement of how alike the two documents are. To evaluate co-citation networks relative to journal of publication, which is often used as a proxy for field, we selected all papers published from 2002 to 2011 in each of six different journals that received exactly five citations during that same time frame. We used cosine similarity analysis to compare the titles and abstracts of these 1,397 works to the titles and abstracts of each article in their co-citation network and then separately to those of each article in the journal in which they appeared; selecting publications with five citations was merely a concession to limit the heavy computational workload that this analysis entailed (249,981 pairwise comparisons within co-citation networks, and 28,516,576 pairwise comparisons with other articles from the same journal). Strikingly, this analysis showed that diagnostic words are much more likely to be shared between an article and the papers in its co-citation network than between that same article and the papers that appear alongside it in its journal of publication (Fig 2). As might be expected, the data in Fig 2 also indicate that articles published in disciplinary journals are more alike than articles published in multidisciplinary journals; the latter are shown as negative controls and highlight the difference in the degree of cosine similarity between an article and its co-citation network versus its journal of publication.
After demonstrating that co-citation networks accurately represent an article’s field, our next step was to decide how to calculate the values that numerically represent the co-citation network of each RA. The most obvious choice, averaging the citation rates of articles in the co-citation network, would also be highly vulnerable to finite number effects. We therefore chose to average the citation rates of the journals represented by the collection of articles in each co-citation network. If a journal was represented twice, its journal citation rate (JCR) was added twice when calculating the average JCR. For reasons of algorithmic parsimony we used the JCRs for the year each article in the co-citation network was published; a different choice at this step would be expected to have little if any effect, since almost all JCRs are quite stable over time (S1 Fig; S1 Table). Since a co-citation network can be reasonably thought to correspond with a RA’s area of science, the average of all JCRs in a given network can be redefined as that RA’s field citation rate (FCR).
Using this method (Fig 3; S2 Fig; Supporting Equations S1 and S2 in S1 Text), we calculated FCRs for 35,837 papers published in 2009 by NIH grant recipients, specifically those who received R01 awards, the standard mechanism used by NIH to fund investigator-initiated research. We also calculated what the FCR would be if it were instead based on citing or cited networks. It is generally accepted that, whereas practitioners in the same field exhibit at least some variation in citation behavior, much broader variation exists among authors in different fields. The more closely a method of field definition approaches maximal separation of between-field and within-field citation behaviors, the lower its expected variance in citations per year (CPY). FCRs based on co-citation networks exhibited lower variance than those based on cited or citing networks (Table 1), suggesting that co-citation networks are better at defining an article’s field than citing or cited networks. As expected, FCRs also display less variance than either ACRs (p < 10−4, F-test for unequal variance) or JIFs (p < 10−4, F-test for unequal variance, Fig 3B, Table 1).
We next asked how stable the FCRs in our dataset remain over time, particularly when the starting co-citation network is small. To answer this question, we calculated FCRs for the 262,988 papers published by R01 grantees between 2003 and 2011 and cited one or more times through the end of 2012 and then recalculated FCRs for the same articles, this time using citations accrued through the end of 2014 (Fig 3C). Comparison of the two values shows that earlier FCRs are well aligned with later ones, even when the initial co-citation network was built on a single citation (Pearson correlation coefficient r of 0.75 versus 2 y later). The FCR quickly converged within five citations, passing r = 0.9 at that point (Fig 3C). The consistency that FCR values display should not be surprising, given the manner in which additional citations rapidly grow an article’s co-citation network (Fig 1D), each node of which represents a citation rate that is itself derived from a large collection of articles (Supporting Equations S1 and S2 in S1 Text). In this way, our method of calculation provides a low-variance quantitative comparator while still allowing the articles themselves to cover a highly dynamic range of subjects.
Having established the co-citation network as a means of determining an FCR for each RA, our next step was to calculate ACR/FCR ratios. Since both ACRs and FCRs are measured in CPY, this generates a rateless, timeless metric that can be used to assess the relative influence of any two RAs. However, it does not measure these values against any broader context. For example, if two RAs have ACR/FCR ratios of 0.7 and 2.1, this represents a 3-fold difference in influence, but it is unclear which of those values would be closer to the overall mean or median for a large collection of papers. One additional step is therefore needed to adjust the raw ACR/FCR ratios so that, for any given FCR, the average RCR equals 1.0. Any selected cohort of RAs can be used as a standard for anchoring expectations, i.e., as a customized benchmark (Supporting Equations S3–S6 in S1 Text). We selected R01-funded papers as our benchmark set; for any given year, regression of the ACR and FCR values of R01-funded papers yields the equation describing, for the FCR of a given RA published in that year, the ECR (Fig 3D and S2 Table). Inserting the ACR as the numerator and FCR of that RA into the regression equation as the denominator is the final step in calculating its RCR value, which incorporates the normalization both to its field of research and to the citation performance of its peers (Fig 3E and S1 Text).
We considered two possible ways to regress ACR on FCR in the benchmarking step of the RCR calculation. The ordinary least squares (OLS) approach will benchmark articles such that the mean RCR is equal to 1.0. OLS regression is suitable for large-scale analyses such as those conducted by universities or funding agencies. However, in smaller analyses in which the distribution of data may be skewed, OLS may yield an RCR less than 1.0 for the median, i.e., typical, article under consideration. In situations such as these—for example, in the case of web tools enabling search and exploration at the article or investigator level—quantile regression is more desirable, as it yields a median RCR equal to 1.0.
For the work presented here, we chose as a benchmark the full set of 311,497 RAs published from 2002 through 2012 by NIH R01 awardees. To measure the degree of correspondence between our method and expert opinion, we compared RCRs generated by OLS benchmarking of ACR/FCR values with three independent sets of postpublication evaluations by subject matter experts (details in S1 Text). We compared RCR with expert rankings (Fig 4) for 2,193 articles published in 2009 and evaluated by Faculty of 1000 members (Fig 4A and S3 Fig), as well as rankings of 430 Howard Hughes Medical Institute- or NIH-funded articles published between 2005 and 2011 and evaluated in a study conducted by the Science and Technology Policy Institute (STPI, Fig 4B and S4 Fig), and finally, 290 articles published in 2009 by extramurally funded NIH investigators and evaluated by NIH intramural investigators in a study of our own design (Fig 4C; S5, S6 and S7 Figs). All three approaches demonstrate that RCR values are well correlated with reviewers’ judgments. We asked experts in the latter study to provide, in addition to an overall score, scores for several independent subcriteria: likely impact of the research, importance of the question being addressed, robustness of the study, appropriateness of the methods, and human health relevance. Random Forest analysis [38] indicated that their scores for likely impact were weighted most heavily in determining their overall evaluation (S6 Fig).
In addition to correlating with expert opinion, RCR is ranking invariant, which is considered to be a desirable property of bibliometric indicators [39,40]. In short, an indicator is ranking invariant when it is used to place two groups of articles in hierarchical order, and the relative positions in that order do not change when uncited articles are added to each group. The RCR metric is ranking invariant when the same number of uncited articles is added to two groups of equal size (Supporting Equations S7–S9 in S1 Text). RCR is also ranking invariant when the same proportion of uncited articles is added to two groups of unequal size (Supporting Equations S10–S11 in S1 Text). This demonstrates that the RCR method can be used effectively and safely in evaluating the relative influence of large groups of publications.
The ideal bibliometric method would provide decision makers with the means to perfectly evaluate the relative influence of even widely divergent areas of science. One way to test whether RCR represents a step towards this ambitious goal is to compare its performance in various scenarios to that of existing metrics. To begin, we asked whether RCR can fairly value a field that is disadvantaged by the use of two of the most widely recognized markers of influence: JIF and CPY. Two areas of science in which the NIH funds research are neurological function and basic cell biology. Both subjects are deserving of attention and resources; however, papers in the former field, which includes subjects such as dementia and mental health, tend to appear in lower impact factor journals and receive fewer CPY than those in the latter. In contrast, the distribution of RCR values for these two areas of study is statistically indistinguishable (Fig 4D). Although this is a single example, it does illustrate one way in which RCR provides value beyond either of these two alternative metrics.
While impact factor and CPY are two of the most commonly used evaluative measures, they are also arguably less sophisticated than other field-normalized methods advanced by bibliometricians. A recent publication has reported that RCR is better correlated with expert opinion than one of these, MNCS, but slightly less well than another, source-normalized citation score 2 (SNCS2) [41]. However, SNCS2 has an important disadvantage relative to RCR; it grows continually over time (like raw citation counts) and so is biased in favor of older papers. This disadvantage can be countered by calculating SNCS2 over a fixed time window, but doing so can obscure important dynamic aspects of citation behavior. In other words, if an article becomes more or less influential compared to its peers after that fixed window has passed, as for example can occur with “sleeping beauties” [42], this would not be apparent to users of SNCS2. These authors [41] also found that RCR is better correlated with expert opinion than citation percentiles, which is the measure bibliometricians have previously recommended as best for evaluating the impact of published work [43]. A different team of researchers has also recently reported that a simplified version of the RCR algorithm is better at identifying important papers than Google’s PageRank [44], which has previously been adapted to quantitate the influence of an article or author [45–47].
One concern that arises whenever citations are field-normalized is that papers in disciplines with intrinsically low citation rates might be inappropriately advantaged. To evaluate how RCR meets this challenge, we compared our approach to an adaptation of the MNCS method, the Thomson-Reuters (TR) ratio. Like RCR, the TR ratio uses CPY as its numerator; unlike RCR, the TR denominator is based on the average citation count of all articles published in a given year in journals that are defined as part of the same category (see S1 Text). This journal categorization approach has some problems; bibliometricians have expressed concern that it is not refined enough to be compatible with article-level metrics [48], and in the case of TR ratios, the use of proprietary journal category lists renders the calculation of the metric somewhat opaque. Still, it is a more refined measure than JIF or CPY, and like RCR, it seeks to field-normalize citations based on choice of comparison group in its denominator.
A TR ratio is available for 34,520 of the 35,837 PubMed Indexed papers published in 2009 by recipients of NIH R01 grants. For this set of articles, the TR ratio denominator ranges from 0.61 to 9.0, and a value of 2.0 or less captures 544 papers (the bottom 1.6%; Fig 4E). The average TR ratio for these papers is 1.67, and the average RCR is 0.67. Since RCR is benchmarked so that the mean paper receives a score of 1.0, it is immediately obvious that these works are having relatively little influence on their respective fields, in spite of their intrinsically low FCRs. In contrast, TR ratios are not benchmarked, so it is difficult to know whether or not it would be appropriate to flag these works as relatively low influence. Nevertheless, we can compare how each method ranks papers with these very low denominators by calculating the fraction with values above the RCR and TR ratio medians. Of the 544 low-denominator articles, 290 have a TR ratio greater than 1.07, the median value of the 34,520 papers, whereas 205 articles are above the RCR median of 0.63. This pattern holds when comparing the number of low-denominator articles that each approach places in the top 5% of the overall distribution; the journal category method identifies 17 such papers, whereas the RCR method identifies 8. Therefore, in avoiding inappropriate inflation of the assessment of articles with the lowest FCRs, RCR is at least as good as, and arguably better than, MNCS.
Importantly, RCR is an improvement over existing metrics in terms of accessibility. While citation percentiles and TR ratios are only available through an expensive institutional subscription service, RCR values for PubMed indexed articles are freely available through the web-based iCite calculator, a screenshot of which is shown in Fig 5. For each PMID entered into iCite, users can download an Excel spreadsheet showing the total number of citations and the number of CPY received by that publication; the number of expected CPY, which are derived from a benchmark group consisting of all NIH R01 grantees, and the FCR are also reported for each article. Detailed, step-by-step help files are posted on the iCite website, and the full code is available on GitHub.
One of the unique strengths of RCR is the way in which a paper’s co-citation network dynamically defines its field. Each new citation an article receives, then, can be thought of as originating either from within or from outside its existing network. As a work gains relevance to additional disciplines, it seems intuitively possible that a new out-of-network citation might lead to a disproportionate increase in FCR and thus to a drop in RCR. Such an occurrence might be thought undesirable [41]; alternatively, it might be considered an accurate reflection of the reduced relative influence the work in question has on a new and larger group of scholars, many of whom previously may not have had a reason to encounter it. Regardless, we felt it was important to determine how frequently such a hypothetical scenario [41] occurs. Among the more than 200,000 articles published between 2003 and 2011 for which we calculated FCRs, only 0.2% experienced a drop in RCR of 0.1 or more between 2012 and 2014; less than 2% experienced any sort of drop at all. This low incidence is consistent with the stability we observe in FCRs (Fig 3C) and with the theoretical properties of citation networks, which are known to be scale-free and thus resistant to perturbation [49].
While 0.2% is a very small number, we wondered whether interdisciplinary science might be overrepresented among the articles that did experience a drop in RCR. An impediment to testing this hypothesis, though, is the lack of a precisely circumscribed consensus definition of interdisciplinary research; one reason it is difficult to arrive at such a definition is that disciplines are themselves dynamic and undergo continuous evolution. For example, biochemistry may have been considered highly interdisciplinary in 1905, the year that term first appears in the PubMed indexed literature [50], but most biomedical researchers today would consider it a well-established discipline in its own right. Others might still view it as an interdisciplinary field in the strictest sense, as it occupies a space between the broader fields of biology and chemistry. To some extent, then, interdisciplinarity is in the eye of the beholder, and this presents another challenge. The question only becomes more vexed when considering more recent mergers such as computational biology, neurophysiology, or developmental genetics; are these established fields, interdisciplinary fields, or subfields? As a first approximation, then, we chose to ask whether articles produced by the NIH Interdisciplinary Research Common Fund program, which funded work that conforms to the definition of interdisciplinarity adopted by the National Academy of Sciences [51,52], were more or less likely to experience a drop in RCR than other NIH-funded articles (Fig 6). Interestingly, these interdisciplinary papers were actually 2-fold less likely to experience a drop in RCR than papers funded either by other Common Fund programs or by standard R01 support (Fig 6A). We currently lack an explanation why this might be so; given how infrequent these small drops are, we cannot yet rule out the possibility that statistical noise is responsible.
We also analyzed publications funded by NIH’s Human Microbiome Project (HMP), which was established in 2007 to generate research resources to facilitate the characterization and analysis of human microbiota in health and disease. From 2009–2011, scientists funded by the HMP published 87 articles for which citation information is available in our dataset. As a comparison group, we identified 2,267 articles on the human microbiome that were published during the same time period but were not funded by the HMP. Articles from the HMP outperformed the comparison group (Fig 6B; HMP mean RCR 2.30, median RCR 1.06; comparison RCR mean 1.23, median 0.74; p < 0.001, Mann-Whitney U test), demonstrating that sorting by funding mechanism has the potential to identify works of differential influence.
We next undertook a large case study of all 88,835 articles published by NIH investigators who maintained continuous R01 funding from fiscal year (FY) 2003 through FY2010 to ask how the RCR of publications from individual investigators changed over this 8-y interval. Each of these investigators had succeeded at least once in renewing one or more of their projects through the NIH competitive peer review process. In aggregate, the RCR values for these articles are well matched to a log-normal distribution (Fig 7); in contrast, as noted previously by others, the distribution of impact factors of the journals in which they were published is non-normal (Fig 7A and 7B) [53,54]. Sorting into quintiles based on JIF demonstrates that, though journals with the highest impact factors have the highest median RCR, influential publications can be found in virtually all journals (Fig 7C and 7D); these data call into question the assertion that journal of publication is the strongest quality signal for researchers who are young or otherwise unknown to a field [55]. Focusing on a dozen representative journals with a wide range of JIFs further substantiates the finding that influential science appears in many venues and reveals noteworthy departures from the correlation between JIF and median RCR (see S1 Text). For example, NIH-funded articles in both Organic Letters (JIF = 4.7) and the Journal of the Acoustical Society of America (JIF = 1.6) have a higher median RCR than those in Nucleic Acids Research (JIF = 7.1; Fig 7E).
As part of this case study, we also calculated the average RCR and average JIF for papers published by each of the 3,089 NIH R01 principal investigators (PIs) represented in the dataset of 88,835 articles. In aggregate, the average RCR and JIF values for NIH R01 PIs exhibited log-normal distributions (Fig 7F and 7G) with substantially different hierarchical ordering (S8 Fig). This raised a further question concerning PIs with RCR values near the mode of the log-normal distribution (dashed line in Fig 7F): as measured by the ability to publish work that influences their respective fields, to what extent does their performance fluctuate? We addressed this question by dividing the 8-y window (FY2003 through FY2010) in half. Average RCRs in the first time period (FY2003 through FY2006) were sorted into quintiles, and the percentage of PIs in the second time period (FY2007 through FY2010) that remained in the same quintile, or moved to a higher or lower quintile, was calculated (Fig 8). The position of PIs in these quintiles appeared to be relatively immobile; 53% of PIs in the top quintile remained at the top, and 53% of those in the bottom quintile remained at the bottom (Fig 8A). For each PI, we also calculated a weighted RCR (the number of articles multiplied by their average RCR); comparing on this basis yielded almost identical results (Fig 8B). It is worth noting that average FCRs for investigators were extremely stable from one 4-y period to the next (Pearson r = 0.92, Table 2), Since FCRs are the quantitative representation of co-citation networks, this further suggests that each co-citation network is successfully capturing the corresponding investigator’s field of research.
Another possible interpretation of the above data is that PI RCRs perform an unbiased random walk from their initial state with a large diffusion rate. Considered from this frame, it could be said that 47% of PIs who started in the top quintile moved out of it during the second 4-y period we analyzed. To test this hypothesis directly, we performed a mean reversion test, which determines whether or not the set of values under consideration will return to an average, or mean, value over time. If drift in PI RCR were simply a random walk, then the change in RCR should by definition be independent of starting RCR, and plotting these two values against each other should result in a straight line with a slope of zero. However, the results show that change in RCR is dependent on starting RCR value (p < 0.001, linear regression analysis, n = 3,089, Fig 8C). Furthermore, randomly shuffling PI RCRs from the second 4-y period gives a slope that is significantly different than that observed for the real data (p < 0.001, extra sum-of-squares F-test, n = 3,089, Fig 8C), ruling out the possibility that these values are randomly sampled from the same distribution in each time interval.
The relationship between scientists and JIFs has been likened to the prisoner’s dilemma from game theory: because grant reviewers use JIFs in their evaluations, investigators must continue to weigh this in their decision making or risk being outcompeted by their peers on this basis [56,57]. A groundswell of support for the San Francisco Declaration on Research Assessment (http://www.ascb.org/dora) has not yet been sufficient to break this cycle [56–61]. Continued use of the JIF as an evaluation metric will fail to credit researchers for publishing highly influential work. Articles in high-profile journals have average RCRs of approximately 3. However, high-impact-factor journals (JIF ≥ 28) only account for 11% of papers that have an RCR of 3 or above. Using impact factors to credit influential work therefore means overlooking 89% of similarly influential papers published in less prestigious venues.
Bibliometrics like JIF and h-index are attractive because citations are affirmations of the spread of knowledge amongst publishing scientists and are important indicators of the influence of a particular set of ideas. Though tracking the productivity of individual scientists with bibliometrics has been controversial, it is difficult to contradict the assertion that uncited articles (RCR = 0) have little if any influence on their respective fields or that the best-cited articles (RCR > 20) are impressively influential. We have not determined whether smaller differences, for example, those with average or slightly above-average RCRs (e.g., 1.0 versus 1.2), reliably reflect differential levels of influence. Further, citation-based metrics can never fully capture all of the relevant information about an article, such as the underlying value of a study or the importance of making progress in solving the problem being addressed. The RCR metric is also not designed to be an indicator of long-term impact, and citation metrics are not appropriate for applied research, e.g., work that is intended to target a narrow audience of nonacademic engineers or clinicians.
It is also very important to note that like all other citation-based metrics, an RCR value cannot be calculated immediately after an article is published. Instead, enough time must pass for a meaningful number of citations to accrue, and the work we describe here provides some rough guidance as to what that meaningful number might be. Specifically, we have found that 93% of co-citation network-based FCRs stabilize after a work has been cited five times (Fig 3C); also in agreement with previously published work, citation rates for approximately the same percentage of articles peak within 2 to 3 y after publication (S2 Fig). Before one or both of those benchmarks have been reached, RCR values might be viewed as provisional; even after that point, neither RCR nor any other citation-based metric should be taken as a substitute for the actual reading of a paper in determining its quality. However, as citation rates mark the breadth and speed of the diffusion of knowledge among publishing scholars, we maintain that quantitative metrics based on citations can effectively supplement subject matter expertise in the evaluation of research groups seeking to make new discoveries and widely disseminate their findings.
We believe RCR offers some significant advantages over existing citation-based metrics, both technically and in terms of usability. Technically, prior attempts to describe a normalized citation metric have resulted in imperfect systems for the comparison of diverse scholarly works (S1 Text and Fig 4E), either because they measure only the average performance of a group of papers [62] or because the article of interest is measured against a control group that includes widely varying areas of science [17,32,48]. An example of the latter is citation percentiling, which the Leiden manifesto [43] recently recommended as a best practice in bibliometrics. Theoretically, the RCR method is an improvement over the use of citation percentiling alone, since masking the skewed distribution of citations and article influence, while statistically convenient, can disadvantage portfolios of high-risk, high-reward research that would be expected to have a small proportion of highly influential articles [63]. Furthermore, we have shown that co-citation networks better define an article’s field than journal of publication (Fig 2), so RCR is a more precise measure of influence than journal-based metrics, a category that includes both citation percentiles and MNCS methods such as the TR ratio. RCR is less likely to unfairly advantage publications in fields with a low citation rate than the TR ratio (Fig 4E). Finally, by incorporating a way to benchmark to a meaningful comparison group, RCR makes it easy for users to know whether a set of articles is above or below expectations for their region or agency; the need for such a feature has already been prominently discussed [31].
In terms of usability, both RCR values and their component variables, including FCRs, CPY, and total citations, are freely available to the public through our iCite tool. As many of the source citation data are proprietary, we are prevented from identifying all of the citing papers; presently, all bibliometrics face this challenge, as limited open source citation data are available. We feel that RCR and iCite represent a large improvement in transparency relative to citation percentiles and TR ratios, which are not cost-free and are furthermore dependent on the proprietary classification of journals into one or another area of science. Our method and tool are also far more transparent than impact factor, the calculation of which has recently come under scrutiny after allegations of manipulation [58,64,65].
Any metric can be gamed, and we have thought carefully about how a single author might try to game RCR. ACRs could be inflated through a combination of self-citation and frequent publication; this strategy has its limits, though, as the top 10% of RCR values for NIH-funded publications on average receive more than 25 CPY, and it is rare for a biomedical scientist to publish more than four or five times over that period. A more promising strategy might be to strive for the lowest possible FCR. An author taking this approach would need to stack the reference section of his or her work not just with poorly cited articles, or with articles in poorly cited fields, but with articles that are co-cited with articles in poorly cited fields. Since citing behavior is also constrained by content, this might be difficult to accomplish; at the very least, it seems likely that reviewers and editors would be able to identify the resulting reference list as unusual. Of course, if enough authors start to reference works in poorly cited areas, that field’s citation rate will go up, and the RCR of the papers in it may go down; in that respect, efforts to game RCR might ultimately prove to be self-defeating.
An important point to keep in mind when interpreting RCR values, though, is that citations follow a power law or log-normal distribution [9], wherein one researcher’s choice of a particular RA is at least partly informed by the choices that other researchers have previously made. There is a certain amount of noise inherent in that selection process [27], especially in the early days of a new discovery when a field is actively working towards consensus. The results of a landmark study on the relationship between quality and success in a competitive market suggest that the ultimate winners in such contests are determined not only by the intrinsic value of the work but also by more intangible social variables [66]. Consistent with this conclusion, a different group of authors has shown that including a “reputation” variable enables an algorithm to better predict which papers in the interdisciplinary field of econophysics will be the most highly cited [67]. Although there is on average a positive relationship between quality and success [66], it is for this reason we suggest that RCR should primarily be considered as a measure of influence, rather than impact or intellectual rigor.
Within these bounds, bibliometric methods such as RCR have the potential to track patterns of scientific productivity over time, which may help answer important questions about how science progresses. In particular, co-citation networks can be used to characterize the relationship between scientific topics (including interdisciplinarity), emerging areas, and social interactions. For example, is the membership of an influential group of investigators in a given field or group of fields stable over time, or is it dynamic, and why? Our data demonstrate the existence of an established hierarchy of influence within the exclusive cohort of NIH R01 recipients who remained continuously funded over an 8-y time frame. This may mean that investigators tend to ask and answer questions of similar interest to their fields. Additionally or alternatively, stable differences in investigators’ status, such as scientific pedigree, institutional resources, and/or peer networks, may be significant drivers of persistently higher or lower RCR values. Future statistical analyses may therefore reveal parameters that contribute to scholarly influence. To the extent that scientific (im)mobility is a product of uneven opportunities afforded to investigators, there may be practical ways in which funding agencies can make policy changes that increase mobility and seed breakthroughs more widely.
There is increasing interest from the public in the outcomes of research. It is therefore becoming necessary to demonstrate outcomes at all levels of funding entities’ research portfolios, beyond the reporting of success stories that can be quickly and succinctly communicated. For this reason, quantitative metrics are likely to become more prominent in research evaluation, especially in large-scale program and policy evaluations. Questions about how to advance science most effectively within the constraints of limited funding require that we apply scientific approaches to determine how science is funded [68–71]. Since quantitative analysis will likely play an increasingly prominent role going forward, it is critical that the scientific community accept only approaches and metrics that are demonstrably valid, vetted, and transparent and insist on their use only in a broader context that includes interpretation by subject matter experts. Widespread adoption of this or any other new metric should follow, not precede, extensive testing in a wide variety of real-world circumstances. Further, the RCR method was designed to assess neither the productivity of individual researchers nor the quality of their science.
Recent work has improved our theoretical understanding of citation dynamics [27–29]. However, progress in solving scientific challenges, rather than citation counts, is the primary interest of funding agencies. The NIH particularly values work that ultimately culminates in advances to human health, a process that has historically taken decades [72]. Here too, metrics have facilitated quantitation of the diffusion of knowledge from basic research toward human health studies, by examining the type rather than the count of citing articles [73]. Insights into how to accelerate this process will probably come from quantitative analysis. To credit the impact of research that may currently be underappreciated, comprehensive evaluation of funding outputs will need to incorporate metrics that can capture many other types of outputs, outcomes, and impact, such as the value of innovation, clinical outcomes, new software, patents, and economic activity. As such, the metric described here should be viewed not as a tool to be used as a primary criterion in funding decisions but as one of several metrics that can provide assistance to decision makers at funding agencies or in other situations in which quantitation can be used judiciously to supplement, not substitute for, expert opinion.
The Thomson Reuters Web of Science citation dataset from 2002–2012 was used for citation analyses. For FCR stability analysis, this dataset was extended to include 2014 data. Because of our primary interest in biomedical research, we limited our analysis to those journals in which NIH R01-funded researchers published during this time. For assigning a JCR to a published article, we used the 2-y synchronous JCR [39,74] for its journal in the year of its publication. Publications from the final year of our dataset (2012) were not included in analyses because they did not have time to accrue enough citations from which to draw meaningful conclusions, but references from these papers to earlier ones were included in citation counts. For analysis of the stability of FCRs in Fig 3, the Web of Science dataset was extended to include the years 2002–2014.
Grant data were downloaded from the NIH RePORTER database (https://projectreporter.nih.gov/). Grant-to-publication linkages were first derived from the NIH SPIRES database, and the data were cleaned to address false positives and false negatives. Grant and publication linkages to PIs were established using Person Profile IDs from the NIH IMPAC-II database. To generate a list of continuously funded investigators, only those Person Profile IDs with active R01 support in each year of FY2003–FY2010 were included.
Co-citation networks were generated in Python (Python Software Foundation, Beaverton, Oregon). This was accomplished on a paper-by-paper basis by assembling the list of articles citing the article of interest and then assembling a list of each paper that those cited. This list of co-cited papers was deduplicated at this point. Example code for generating co-citation networks and calculating FCRs is available on GitHub (http://github.com/NIHOPA). Data that were used for analysis can be found as csv files in the same repository. Further calculations were handled in R (R Foundation for Statistical Computing, Vienna, Austria). Visualizations were generated in Prism 6 (GraphPad, La Jolla, California), SigmaPlot (Systat Software, San Jose, California), or Excel 2010 (Microsoft, Redmond, Washington). Code used to generate the database used in the iCite web application (https://icite.od.nih.gov) can be found in the GitHub repository. A preprint version of this manuscript can also be found on bioRxiv [75]. For box-and-whisker plots, boxes represent the interquartile range with a line in between at the median, and whiskers extend to the 10th and 90th percentiles.
When comparing citations rates to other metrics (e.g., postpublication review scores), citation rates were log-transformed because of their highly skewed distribution, unless these other scores were similarly skewed (i.e., Faculty of 1000 review scores). For this process, article RCRs of zero were converted to the first power of 10 lower than the lowest positive number in the dataset (generally 10−2). In the analysis of PI RCRs, no investigator had an average RCR of zero.
The commercially available text mining program IN-SPIRE (Pacific Northwest National Laboratories, Richland, Washington) [76] was used for content-based clustering of citations (Fig 1). For comparison of JIF CPY and RCR (Fig 4D), papers in the fields of cell biology and neurological function were those supported by grants assigned to the corresponding review units within the NIH Center for Scientific Review. For the data in Fig 2, articles were selected from six journals; three of these were disciplinary (Journal of Neuroscience, Blood, and Genetics), and the other three were multidisciplinary (Nature, Science, and PNAS). Articles published between 2002 and 2011 that accrued exactly 5 citations during that same time frame (n = 1,397) were selected; this subset, rather than all articles published in these six journals, was chosen for analysis in order to limit the number of pairwise text comparisons. Cosine similarity scores [37] were calculated for these 1,397 RAs against each article in their co-citation network and separately against each article appearing in the same journal. This resulted in 249,981 pairwise comparisons with articles in the co-citation networks and 28,516,576 pairwise comparisons with articles from the same journals. The journal comparisons and the co-citation comparison were both done with primary articles as well as reviews. For the comparisons, abstracts (from PubMed) were concatenated with titles to comprise the information for each document. Words were converted to lower case and stemmed. Any numbers, as well as words consisting of one or two letters, were removed from the corpus along with words appearing less than ten times. Term-document matrices were weighted for TF-IDF [77]. For one analysis, the term-document matrix was trimmed to the top 1000 TF-IDF-weighted terms, and in the other analysis, no additional term trimming was performed.
|
10.1371/journal.ppat.1006741 | The pUL37 tegument protein guides alpha-herpesvirus retrograde axonal transport to promote neuroinvasion | A hallmark property of the neurotropic alpha-herpesvirinae is the dissemination of infection to sensory and autonomic ganglia of the peripheral nervous system following an initial exposure at mucosal surfaces. The peripheral ganglia serve as the latent virus reservoir and the source of recurrent infections such as cold sores (herpes simplex virus type I) and shingles (varicella zoster virus). However, the means by which these viruses routinely invade the nervous system is not fully understood. We report that an internal virion component, the pUL37 tegument protein, has a surface region that is an essential neuroinvasion effector. Mutation of this region rendered herpes simplex virus type 1 (HSV-1) and pseudorabies virus (PRV) incapable of spreading by retrograde axonal transport to peripheral ganglia both in culture and animals. By monitoring the axonal transport of individual viral particles by time-lapse fluorescence microscopy, the mutant viruses were determined to lack the characteristic sustained intracellular capsid motion along microtubules that normally traffics capsids to the neural soma. Consistent with the axonal transport deficit, the mutant viruses did not reach sites of latency in peripheral ganglia, and were avirulent. Despite this, viral propagation in peripheral tissues and in cultured epithelial cell lines remained robust. Selective elimination of retrograde delivery to the nervous system has long been sought after as a means to develop vaccines against these ubiquitous, and sometimes devastating viruses. In support of this potential, we find that HSV-1 and PRV mutated in the effector region of pUL37 evoked effective vaccination against subsequent nervous system challenges and encephalitic disease. These findings demonstrate that retrograde axonal transport of the herpesviruses occurs by a virus-directed mechanism that operates by coordinating opposing microtubule motors to favor sustained retrograde delivery of the virus to the peripheral ganglia. The ability to selectively eliminate the retrograde axonal transport mechanism from these viruses will be useful in trans-synaptic mapping studies of the mammalian nervous system, and affords a new vaccination paradigm for human and veterinary neurotropic herpesviruses.
| Neuroinvasive members of the alpha-herpesvirinae include human (i.e. herpes simplex virus type I; HSV-1) and veterinary (i.e. pseudorabies virus; PRV) pathogens that routinely invade the peripheral nervous system of an immunocompetent host in the absence of overt tissue damage. We have identified an essential, and conserved, component of these viruses that directs incoming viral particles into the neural ganglia. Viruses carrying mutations in this protein effector region propagate normally at peripheral sites of inoculation but fail to invade the nervous system by retrograde axonal transport, cannot establish life-long latent infections, and are avirulent. These properties define a promising new class of live-attenuated vaccines that protect from subsequent nervous system invasion and encephalitic disease.
| Neuroinvasive members of the alpha-herpesvirinae include human (i.e. herpes simplex virus type I; HSV-1] and veterinary (i.e. pseudorabies virus; PRV) pathogens that establish life-long latent infections in neurons of the peripheral nervous system (PNS) [1]. Infections by these viruses can be subtle or manifest in severe disease, the latter ranging from herpes simplex encephalitis in humans to Aujeszky’s disease in swine [2,3]. Unlike other neurotropic viruses, the neuroinvasive alpha-herpesvirinae routinely access the nervous system, and do so in the absence of obvious tissue trauma. The robust nature of the invasive process is evidenced by the presence of herpesvirus DNA in the nuclei of peripheral neurons in most adults [4]. Deciphering the mechanisms underlying virus delivery from the mucosa to neuronal cell bodies of sensory and autonomic ganglia is of interest not only for its remarkable biology, but also to exploit it as a gene delivery vehicle and to produce a new generation of live-attenuated vaccines [5,6].
The alpha-herpesvirinae are unusual among neurotropic viruses in that upon entering nerve endings they do not rely on endosomes for axonal trafficking to neuronal cell bodies [7]. Instead, these viruses fuse their envelopes with a cellular membrane to deposit the 125 nm diameter icosahedral capsid into the cytosol. An internal protein matrix surrounding the capsid in the virion, referred to as the tegument, is deposited in the cytosol with the capsid [8–12]. Although the majority of the tegument disassociates from the capsid at this stage, three tegument proteins remain capsid bound: pUL36, pUL37, and pUS3 [13–17]. Deciphering the contributions of pUL36 and pUL37 during initial infection is made complicated by their subsequent roles during virus assembly and egress; eliminating either tegument protein severely compromises production of HSV-1 and PRV virions [18–28].
The pUL36 tegument protein is directly anchored to the capsid surface [29] where it promotes trafficking to the nucleus by tethering the capsid to the dynein/dynactin microtubule motor complex [30], and by docking the capsid at a nuclear pore and triggering genome release [31–39]. pUL36 is also a neuroinvasive factor, which uses ubiquitination as a molecular switch to promote virus spread into the nervous system [40–42]. Less is known about pUL37 and its contribution to neuroinvasion. Like pUL36, pUL37 promotes translocation of incoming capsids to the nucleus [43]. The HSV-1 version of the protein is a deamidase and may increase cell permissivity to infection through interactions with TNF receptor associated factor 6 (TRAF6), melanoma differentiation-associated protein 5 (MDA5), and retinoic acid-inducible gene I (RIG-I); although residues associated with deamidation and TRAF6 binding are not conserved across the neuroinvasive herpesviridae [44,45]. Only a null mutant has been extensively examined in vivo, and although the mutant was impaired for invasion of the nervous system, this defect was more generally associated with a broad inability to propagate in all tissues examined [46]. We recently determined the crystal structure of the N-terminal half of pUL37 and identified three evolutionarily conserved surface-exposed regions: R1, R2 and R3 [47]. The three regions are distal to the portion of pUL37 that is associated with deamidase activity, TRAF6 binding, and interactions with other virion structural proteins [44,45,48,49]. Although these N-terminal regions were dispensable for replication in an epithelial cell line [47], in this report we demonstrate that one of these regions, R2, is critical for invasion of the nervous system.
Mutation of R2 in representative members of the two neuroinvasive genera of the alpha-herpesvirinae, PRV (varicellovirus genus) and HSV-1 (simplexvirus genus) in both instances eliminated viral invasion of the PNS in vivo. Time-lapse imaging of cultured primary sensory neurons following infection with fluorescent reporter viruses revealed that mutations in R2 profoundly impaired retrograde axonal transport. This defect was not due to a loss of capsid intracellular transport but rather stemmed from a loss of targeted transport. Capsids of the mutant viruses engaged in aberrant bidirectional motion that did not support long-distance axonal trafficking to neural soma. In contrast, R2 was dispensable for axonal transport during the egress phase of neuronal infection, with virions delivered to distal axons at the same efficiency as wild-type infections, and indicating that axonal transport at these two stages of infection occurs by independent mechanisms. These results define a critical component of the viral apparatus that mediates the neuroinvasion of the alpha-herpesviruses and indicate that the R2 region enables sustained retrograde transport to neural soma.
The ability to selectively eliminate the neuroinvasive property of these viruses, without otherwise impairing their replication, has potential as a new class of live-attenuated vaccines. In principle, the selective attenuation afforded by the R2 mutants should prevent vaccine-associated neuroinvasive disease and life-long latent infections, which would otherwise become problematic if the individual’s immune system were to later become compromised, while simultaneously preserving robust replication and spread in peripheral tissues required to elicit a pronounced immune response. To this end, we include an initial demonstration that R2-mutant viruses efficiently protect from wild-type HSV-1 and PRV challenges and offer robust protection from ocular disease, nervous system invasion, and encephalitis.
We previously mutated three evolutionary conserved surface regions of the pUL37 tegument protein (R1, R2, R3) in PRV [47]. Each mutated region consisted of five alanine substitutions (S1 Table). Although plaques produced by the R2 mutant were smaller than wild type, all three mutants propagated to wild-type titers [47]. To extend this characterization, each mutant virus was examined in vivo following intranasal instillation into mice. Whereas wild-type PRV caused rapid lethality, the derivatives of PRV mutated for R1 and R3 were attenuated. Remarkably, the R2 mutant was avirulent despite propagating to wild-type titers in culture (Fig 1A) [47]. To determine if the stark R2 mutant phenotype resulted from misfolding of the mutant pUL37 protein, the crystal structure of the pUL37 N-terminus was solved and overlaid with the previous WT structure [47]. The five point mutations did not affect the overall fold of the N-terminal half of pUL37 as judged by the alignment of the wild-type and R2 mutant crystal structures (rmsd = 0.55 Å) (Fig 1B). These findings demonstrated that introduction of the mutations did not contort the pUL37 backbone, and encouraged us to further examine the R2 mutant in cell culture infection models to identify the specific source of its avirulence.
Because the ability of the pUL37 R2 mutant to propagate to wild-type titers was inconsistent with an assembly or egress defect, we examined whether R2 contributed to the initiation of infection. Although PRV entry into epithelial cells was not significantly delayed by R2 mutation (Fig 2A), immediate-early viral gene expression was reduced (Fig 2B). The R2 mutant delay in viral expression was equivalent to the delay imposed on wild-type infections treated with nocodazole (a microtubule depolymerization drug), yet nocodazole treatment did not further reduce gene expression from the R2 mutant (Fig 2B). These results provided an initial indication that R2 functions at a post-entry microtubule-dependent step early in infection.
Our results suggested that the PRV R2 mutant was impaired for microtubule-based transport, which is absolutely required for neuronal infection via long-distance axonal transport [50]. Therefore, studies in primary cultured neurons were next pursued. To expand the scope of these studies to additional alpha-herpesviruses, equivalent mutations were introduced into HSV-1 (Fig 3A). As was observed with PRV, the resultant HSV-1 UL37 R2 mutant propagated to wild-type titers but had a reduction in plaque diameter (Fig 3B and 3C) [47]. To determine if retrograde transport was impaired in primary sensory neurons, live-cell microscopy was used to examine capsid dynamics in axons during initial infection. Whereas wild-type HSV-1 and PRV capsids preferentially moved in the retrograde direction in axons (towards the neural soma), capsids of the R2 mutants moved in an aberrant bidirectional manner best described as “ping-ponging” (S1 Movie). Ping-pong motion was associated with near-zero net displacement of the capsids (Fig 4A–4C).
To test whether the loss of sustained retrograde transport prevented capsid trafficking to neuronal nuclei from axon terminals, dorsal root ganglia (DRG) were cultured ex vivo as intact explants and co-infected with equal amounts of wild-type and R2 mutant viruses encoding eGFP- and mCherry-capsids, respectively. Under these conditions, the sensory neurons were maintained in their surrounding tissue but had extended axons that were exposed to the inoculum, which permitted a comparative assessment of the ability of the two viruses to deliver capsids to nuclear rims following retrograde axonal transport [30]. Only capsids from the wild-type HSV-1 or PRV arrived at nuclear rims of neurons within the explants (Fig 4D). This result reinforced the inability of the R2-mutant viruses to traffic retrogradely in axons.
Although R2 mutation did not impair entry of PRV into transformed epithelial cells (Fig 2A), we could not rule out that the aberrant motion observed with R2 mutants during neuronal infection represented virions that failed to undergo fusion-based entry into axons, and instead were either residing on the axon surface or within endosomes. To specifically monitor fusion-based virus entry into axons, an enzyme-based assay was developed based on an approach that was used to monitor HIV fusion with T lymphocytes [51]. Primary sensory neurons were loaded with the β-lactamase substrate CCF2, which contains a 7-hydroxycoumarin ring linked to fluorescein via a beta-lactam ring. Excitation of the coumarin group produces Förster resonance energy transfer (FRET) to the fluorescein group and emission of green florescence at 518 nm. β-lactamase prevents FRET by cleaving the beta-lactam ring and liberating coumarin fluorescence at 447 nm [52]. To trigger CCF2 cleavage, recombinants of PRV were produced encoding β-lactamase fused to the pUL35 capsid protein (β-lac-capsid). Fusion-based entry of the recombinant viruses was monitored in CCF2-loaded primary sensory neurons. Deposition of the β-lac-capsid into the cytosol exposes the β-lactamase to the cytosolic CCF2 substrate resulting in reduced FRET. As controls, neurons were infected either with PRV that did not encode for beta-lactamase or with beta-lactamase-expressing PRV that was first treated with Accutase (Thermo Fisher) to remove the ectodomains of viral glycoproteins that are essential for entry. The two controls provided a combined baseline for the ratiometric FRET measurements over the experimental time course. The findings supported that wild-type and R2-mutant PRV induced CCF2 cleavage at an equivalent rate, and indicated that mutation of R2 did not impair the kinetics of entry into neuronal cells (Fig 5A).
Herpesvirus entry into cells is coupled with a selective disassembly of tegument proteins from the capsid. To confirm that mutation of R2 did not impair virion disassembly, a collection of dual-florescent viruses that encode mCherry-capsids and eGFP fused to protein constituents of the envelope and tegument was produced. These viruses permitted monitoring of loss of the eGFP marker as an indirect indication of fusion-based entry of individual virus particles in axons [15,53]. Similar to wild-type viruses, R2 mutant capsids moving in axons lacked the gD envelope protein and outer tegument proteins that dissociate from capsids upon entry. Furthermore, mutation of R2 did not compromise the normal retention of pUL36 and pUL37 on capsids following entry (Fig 5B; S2 Movie). Taken together, these findings demonstrated that virus entry and disassembly in neurons was unimpaired by R2 mutation, and further indicated that the capsids displaying aberrant ping-pong motion resided within the axon cytosol.
Paradoxically, R2 mutant capsids displaying ping-pong motion accumulated throughout the length of axons in the ex vivo DRG cultures. This was unexpected given that virus entry typically occurs at axon terminals, and the R2 mutant phenotype should prevent the capsids from migrating beyond the point of entry [54]. Therefore, the origin of these capsids was scrutinized using neurons cultured in microfluidic chambers, such that exposure to the virus was restricted to the distal regions of the axons near the terminals. Under these conditions no capsids of the R2 mutant were observed mid-axon, away from the site of inoculation, confirming that R2 mutant capsids were incapable of net retrograde displacement (S3 Movie). We infer that the presence of R2 capsids initially observed throughout the length of axons was due to unrestricted entry of virus particles mid-axon.
Alpha-herpesvirus rely on bi-directional microtubule-based axonal transport during initial and late infection, with preferential dynein-based retrograde transport during ingress switching to biased kinesin-based anterograde motion during egress [55]. To determine if R2 was a global regulator of microtubule transport, DRG were dissociated and cultured as isolated neurons, exposing the neuronal soma to the virus inoculum and removing the requirement for retrograde axonal transport to reach the nucleus. Under these conditions, the neurons were productively infected with the PRV R2 mutant and subsequent progeny viral particles were found to transport in the anterograde direction to distal axon terminals equivalently to the wild-type virus (Fig 6). These findings demonstrate that, as in transformed epithelial cells, R2 was not required for the delivery of capsids to nuclei over short distances and was also dispensable for long-distance anterograde transport in axons during egress.
Our results with cultured primary sensory neurons predict that the R2 region may serve as an effector of neuroinvasion in vivo. To test this hypothesis, we examined the ability of HSV-1 to infect the mouse nervous system following inoculation of peripheral tissues. For HSV-1, mice were infected at the corneal surface and transmission of infection to the trigeminal ganglia (TG) and subsequently to second order neurons in the brain stem was assessed (Fig 7A). In contrast to wild-type HSV-1, the HSV-1 R2 mutant was not detected in either site despite producing equivalent focal infections on the cornea (Fig 7B). To exclude the possibility that the fluorescently-tagged capsid interfered with HSV-1 R2 neuroinvasion, similar experiments were performed using untagged HSV-1 and viral load was examined in tear film (eye swabs), TG, and brain by previously established methods [56,57]. The near wild-type levels of R2 mutant detected in the tears, coupled with the inability to recover R2 mutant virus from the peripheral (TG) and central (brain) nervous system, added support for an essential role for R2 during neuroinvasion (Fig 7C). Alternatively, the R2 mutant virus may be prone to entering a latent state upon reaching the nervous system, and thereby becoming undetectable by the above approaches. To exclude this possibility, TGs were examined for latent virus by harvesting from mice at 14 dpi and co-culturing the tissue with Vero cells. Cells were monitored continuously for signs of cytopathic effect (CPE) for 10 days post explantation as an indication of the emergence of reactivated virus. Whereas Vero cells co-cultured with 80% of the TGs (8 of 10) from mice infected with wild-type HSV-1 became positive, no reactivation was observed from TGs (0 of 8) harvested from mice infected with the HSV-1 R2 mutant. The presence of HSV-1 DNA in the nervous system was also assessed by qPCR. Consistent with the reactivation assay, viral DNA was detected in tissues from mice infected with wild-type HSV-1, but not in tissues from mice infected with the R2 mutant (Fig 7D).
The link between R2 and neuroinvasion was examined in greater detail in an established rat encephalitic model of PRV infection [42,58]. Firstly, the PRV R2 mutant failed to spread to the TG following inoculation into the anterior chamber of the eye, which was consistent with the HSV-1 R2 mutant defect in vivo, and pinpointed R2 as a conserved effector of alpha-herpesvirus neuroinvasion. Furthermore, R2 was essential for retrograde spread of PRV to the parasympathetic ciliary ganglion (CG), sympathetic superior cervical ganglion (SCG), and oculomotor nucleus (OMN) of the midbrain, indicating a loss of neuroinvasive properties in PNS sensory and autonomic neurons, and CNS motor neurons (Fig 8). Combined, these data indicate that R2 is essential for retrograde spread from peripheral tissues to both the PNS and the CNS in vivo. To confirm that the failure of the PRV R2 mutant to spread in retrograde circuits was due to a loss of long-distance retrograde axonal transport, as opposed to a block in transmission from mucosal surfaces to innervating axon terminals, PRV was injected directly into the retinorecipient superior colliculus (SC) of the midbrain (S1 Fig). In this paradigm, the injected virus is directly exposed to axons projecting from the retina as well as local neurons. Whereas wild-type PRV had spread retrogradely to the retina by two days post-infection, the PRV R2 mutant failed to spread to the retina (S1 Fig), and remained absent from the retina at five days post-infection.
To confirm that R2 was dispensable for anterograde transport in vivo, retinal ganglion cells (RGCs) of rats were directly exposed to the PRV R2 mutant by intravitreal injection. RGCs project axons via the optic nerve to visual centers of the brain, which allows for monitoring of anterograde spread within the CNS (S1 Fig). The PRV R2 mutant spread anterogradely from RGCs to retinorecipient regions of the brain, and furthermore continued to spread via anterograde transneuronal transport to second order nuclei of the visual system (S1 Fig). We conclude that the R2 mutant remained both neurotropic and competent to spread through anterograde circuits within the CNS when the initial requirement for retrograde transport from the periphery was bypassed. Anterograde spread of the R2 mutant through the visual system was delayed but was more comprehensive than that of the wild-type virus, the latter causing a more rapid onset of death that limited its spread. The cause of the slower anterograde spread in vivo was not predicted from our studies in cultured neurons, and we suspect that the delay may have manifested from inefficient retrograde transport over short distances in dendrites or the neuronal soma following spread across sites of synaptic contact. This hypothesis is supported by our data in non-neuronal cells, which demonstrated that over short distances R2 promoted efficient gene expression in a microtubule dependent manner (Fig 2B). Taken together, the in vivo studies recapitulated the results obtained with primary cultured sensory neurons: R2 is required for retrograde axonal transport but is dispensable for anterograde axonal transport. This phenotype is noteworthy both for its novelty and its utility. For example, an alpha-herpesvirus specifically lacking the ability to transmit through retrograde circuits is a resource for mapping anterograde circuits of the mammalian nervous system.
The ability of R2 mutant viruses to replicate and spread in peripheral tissues while being unable to engage in long distance retrograde transport to reach sites of latency offers a new approach to vaccine design [5,6]. As a proof of concept, mice were exposed intranasally with a single dose of the PRV R2 mutant and challenged at varying days post vaccination with 10,000 times the 50% lethal dose of wild-type PRV (S2 Fig). Whereas PRV normally evokes a lethal encephalitis terminating at ~ 2 days, 75% of mice challenged with wild-type PRV at 14 and 28 days post vaccination showed no signs of disease and survived until the end of the experiment at 21 days post challenge (Fig 9). We believe this is the first success at protecting rodents from the unusually high virulence exhibited by PRV that normally causes rapid lethality, and this protection was noteworthy given the extreme circumstances of this high dose challenge and single dose vaccination.
To determine if the efficacy of the R2 mutant as a live-attenuated vaccine extended to HSV-1, mice were infected ocularly with a single dose of the HSV-1 R2 mutant and challenged with wild-type HSV-1. Unlike PRV, the strain of HSV-1 used for these studies (strain F) is not highly virulent in mice [59], but nevertheless invades the PNS and CNS following corneal inoculation via TG sensory neurons (Fig 7A) [60]. Mice vaccinated with the HSV-1 R2 mutant and challenged with wild-type HSV-1 strain F had no detectible virus in either the PNS or the CNS, demonstrating effective protection of the nervous system from HSV-1 invasion (Fig 10A). Furthermore, the protection offered by R2 vaccination prevented lethal disease associated with infections by the more virulent McKrae strain of HSV-1 [61] (Fig 10B). We also note that mice vaccinated with the HSV-1 R2 mutant were protected from the development of periocular disease following wild-type challenge (Fig 10C) [62]. Importantly, the R2 vaccine itself did not cause periocular disease, which is consistent with previous studies that have suggested periocular skin infection requires zosteriform spread from the cornea to periocular skin via sensory ganglia [63]. While further studies are required to assess the immune response that results from R2 vaccination, we hypothesize that the ability of R2 mutant viruses to replicate effectively in peripheral tissues produces an adaptive immune response that effectively blocks the ability of wild-type viruses from invading the nervous system. Achieving this degree of protection reinforces a prior expectation that effective vaccination against HSV would require an attenuated strain that remains fully competent for propagation while having a selective loss of neuroinvasive potential [5].
Few viruses routinely enter the mammalian nervous system, which generally requires crossing the blood-brain barrier or transmission of virions from peripheral tissues to axonal endings and subsequent long-distance retrograde transport in axons [7]. Most neuroinvasive viruses, including rabies virus and poliovirus, reach the neuronal soma by traveling retrogradely in axons within endosomes [64–66]. In contrast, the ability of alpha-herpesviruses to sustain retrograde motion in the axon cytosol suggests that these viruses engage cellular transport motors directly and selectively [30,67]. Upon fusion-based entry into an axon terminal, alpha-herpesvirus transport is bi-directional but is heavily biased for retrograde transport to promote long-distance delivery to the neuronal soma [53,54,68]. The bi-directionality likely results from the ability of capsid/tegument complexes to recruit the dynein/dynactin motor and an opposing kinesin motor [67,69–71]. Our results indicate that the conserved R2 region in the pUL37 inner tegument protein rectifies this motion to produce sustained retrograde bias. Adenoviridae, which are rarely neuroinvasive, also bind opposing microtubule motors but unlike herpesviruses do not coordinate motor activity. The stochastic motion produces zero net displacement on average but is sufficient to deliver adenovirus particles to nuclei over short distances [72–74]. The aberrant motion displayed by HSV-1 and PRV R2 mutant viruses is similar to the zero-net displacement natively exhibited by adenovirus, and showcases the added level of control that the alpha-herpesviruses impose on microtubule-based transport: a property that can now be attributed to R2 effector function.
The reliance on R2 for retrograde capsid trafficking in axons, the latter of which can be upwards of a meter in length, is underscored by estimates that in the absence of directed transport it would take a herpesvirus capsid over a century to diffuse 1 cm in the cytoplasm [75]. The requirement for R2 during neuroinvasion was consistent across several paradigms used in this study, which included two related viruses, two rodent infection models, and an avian neuronal culture model. The latter finding deserves some additional commentary due to its utility. Studies of virus delivery from axon endings to the neural soma in culture are often performed in chambers, because the chamber provides a physical barrier that allows for selective virus exposure at the distal axons (S3 Movie). In these systems, extracellular virus cannot diffuse to the neural soma; only virus capable of trafficking retrogradely in axons can reach the soma and productively infect the neuron. The application of the R2 mutant viruses described here establishes that neurons in intact DRG explants are also shielded from extracellular virus inoculum by virtue of the intact tissue architecture, and demonstrate that retrograde delivery examined in explants offers an expedient alternative to traditional chambered systems. R2 mutant viruses can also be considered as controls for traditional cell chamber models, as they can serve to help rule out unintended extracellular diffusion of inoculum due to a leaky chamber.
Because R2 was discovered in part by its conservation, and because we show here that R2 function is conserved across two neuroinvasive herpesviruses of different genera, we expect that all neuroinvasive members of the herpesvirinae are likely dependent on this activity to reach the nervous system of their respective hosts, and therefore to establish life-long latent infections [47]. Although these studies indicate that the mechanism of R2-mediated neuroinvasion is to regulate microtubule transport in axons and guide incoming capsids to neural soma, more work will be needed to elucidate the molecular mechanism of this action. We can infer that R2 is a trafficking factor that coordinates the activities of microtubule motors already bound at the capsid surface, as R2 mutants retain motion that is indicative of microtubule transport by opposing motors. This interpretation is consistent with the finding that the pUL36 tegument protein, which tethers pUL37 to the capsid surface, is the effector that recruits the dynein motor following entry into cells [30]. The requirement for capsids to coordinate opposing motors during initial infection was not necessarily expected, and indicates that either a kinesin motor is effectively performing an antiviral function, which pUL37 overcomes, or a kinesin motor contributes to the guidance of capsids to their target nuclei in the neural ganglia. Deciphering this will require the identification of the kinesin recruitment proteins on the capsid surface.
Whereas several activities were previously mapped to the C-terminal half of pUL37 [44–46,76], R2, along with R1 and R3, are located in the N-terminal half of pUL37. Mutants of all three were attenuated in vivo, despite being dispensable for propagation to wild-type titers in cultured epithelial cells (Fig 1) [47]. These findings are suggestive that the N-terminus of pUL37 may be a multifunctional effector of neuronal infection. It is noteworthy that mutation of R2 did not impair the anterograde transport of newly assembled viral particles to distal axon terminals. This result reinforces previous findings indicating that retrograde and anterograde axonal transport occurs by distinct mechanisms [53]. Potential contributions of R1 and R3 include modulating the C-terminal half of pUL37 and its interactions with cellular transport effectors including dystonin/BPAG1, and the intracellular targeting of capsids during ingress and egress phases of neuronal infection [20,77,78]. Understanding the precise contributions of R1 and R3 to infection will require follow up studies of the R1 and R3 mutants in cultured neurons and in vivo.
Beyond increasing our understanding of how alpha-herpesviruses invade the nervous system, the production of mutant viruses specifically lacking the capacity to transport retrogradely in axons has several practical applications. First, R2 mutant viruses can be used to specifically trace anterograde circuits within the mammalian nervous system. This new tool, in combination with preexisting mutant viruses that possess retrograde specificity, can now be used to precisely map multi-synaptic circuits within the CNS [79]. The low virulence of R2 mutants has the added benefit of allowing for high-order mapping of neurons. We are currently examining this potential in more detail. Second, R2 mutation may improve the safety of oncolytic herpesvirus vectors by impeding their transport away from the site of injection in brain tumors, thereby restricting lytic activity to the intended malignant tissue. Third, as demonstrated here, R2 mutants have the potential to serve as effective live-attenuated vaccines.
Recent attempts to develop a live-attenuated HSV vaccine have been based on broadly compromising viral propagation [80], or by attempting to abrogate the ability of the virus to invade the nervous system while retaining replication in non-neuronal cells. The latter approach includes mutant viruses that lack expression of specific envelope proteins that play auxiliary roles in HSV entry. Encouragingly, HSV lacking the gE envelope protein fails to spread to mouse dorsal root ganglia following inoculation of abraded skin, and protects from subsequent wild-type challenges [81]. However, gE-null HSV-1 retains baseline neuroinvasive properties and is competent to spread from epithelia to neurons in culture models, albeit at a reduced frequency [82]. In a related approach, HSV-1 encoding a truncated gK envelope protein inefficiently enters cells, including neurons, and when combined with a truncated pUL20 membrane protein renders the virus incapable of causing disease in mice [83]. The decreased capacity of the gE and gK/pUL20 mutant viruses to infect the nervous system is not fully understood and does not appear to be a conserved property of these proteins in related herpesviruses, but nevertheless show exciting potential [84,85].
Our motivation for performing an initial exploration of the pUL37 R2 mutant as a vaccine candidate was based on: (1) its inability to invade the nervous system both in animals and culture despite robust propagation in peripheral tissues and epithelial cell culture, (2) its clearly defined defect that accounts for the ablated neuroinvasive phenotype, and (3) the conservation of the phenotype across the two neuroinvasive herpesvirinae genera. The application of this approach to PRV showcases the unique potential of these recombinant viruses as vaccines. Current live-attenuated PRV vaccines for use in pigs retain sufficient virulence to be lethal in mice [86]. In contrast, the PRV R2 mutant is avirulent in mice, yet offers protection against lethal wild-type PRV challenges. Wild-type PRV is particularly virulent in mice, more so than HSV-1 or HSV-2, and provides a stringent test case of vaccine efficacy. For this reason, R2 mutation may offer a platform for the development of vaccines against many neuroinvasive herpesviruses. The inability of the R2 mutants to infect the peripheral nervous system makes them unable to establish latent infections, while replication in peripheral tissues is benefited from maintaining the integrity of genes required for normal propagation outside of the nervous system. We hypothesize that the robust propagation at the periphery allows for an adaptive immune response of sufficient magnitude to protect mice from a challenge of highly neuroinvasive wild-type HSV-1 or PRV following a single exposure to the R2 mutant strain.
The potential to prevent the establishment of life-long alpha-herpesvirus infections following vaccination may offer a way forward to developing prophylactic vaccines against several noteworthy clinical and veterinary alpha-herpesvirus pathogens. HSV-1 and PRV are representatives of the two principle branches of neuroinvasive herpesviruses, and we expect that this level of protection should extrapolate well to other alpha-herpesviruses. In particular, we anticipate that live-attenuated vaccines based on R2-deficient viruses may offer long sought-after protection against HSV-1 and HSV-2, thereby reducing the incidence of herpes simplex keratitis, herpes simplex encephalitis, and life-threatening neonatal infections [87–89].
The plasmid pJP55 encodes the PRV UL37N-R2 mutant gene in frame with a cleavable N-terminal His6-SUMO tag. This plasmid was constructed using 2 rounds of splicing by overlap extension (SOE) PCR using the plasmid pJP23 (WT PRV UL37N) as template in the first round [47], and the intermediate plasmid (pJP56) as the template in the second round. For the first round of SOE-PCR, creating the Q324A, D362A, and R365A mutations, the 5’ region was amplified using the outer forward primer 5’-GCAACCGGTGTTTGGGAAGCAGTTGCAGCAAGCGCA-3’ and reverse primer 5’-GCAACCAACTGCTGCCTGTGCTGCGGTCAGACGAGG-3’, while the 3’ region was amplified using the forward primer 5’-CCTCGTCTGACCGCAGCACAGGCAGCAGTTGGTTGC-3’ and the outer reverse primer 5’-CGCAAGCTTCTAGGCTGCGCTGGTCGGTGC-3’ (restriction sites are underlined and mutated nucleotides are in bold). Next, the PCR fragments generated above were mixed and used as template for the next step of SOE-PCR using the forward primer of the 5’ reaction and the reverse primer of the 3’ reaction. The PCR product from the last step of SOE PCR was then sub-cloned into pJP23 using the AgeI and HindIII restriction sites generating the intermediate plasmid pJP56. To introduce the H421A and H425A mutations, a second round of SOE PCR was carried out using pJP56 as template. The overlapping primers used for the first step of the second round of SOE-PCR were the reverse primer for the 5’ region 5’-CGGTGCCATACATGCACGAACAACTGCTTCAACACCTTT-3’ and the forward primer for the 3’ region 5’-AAAGGTGTTCAAGCAGTTGTTCGTGCATGTATGGCACCG-3’. The outside primers (forward for 5’ region and reverse for 3’ region) remained the same as in the first round of SOE-PCR. These same outside primers were used to complete the second step of SOE-PCR yielding the UL37N-R2 mutant gene containing all mutations. Similar to the first round of mutagenesis, the resulting PCR fragment from the second step of SOE-PCR was sub-cloned into the pJP23 plasmid using the restriction sites AgeI and HindIII and sequence verified yielding the plasmid pJP55.
UL37N-R2 protein was expressed in T7 Express E. coli (New England BioLabs) and purified according to the protocol previously used to produce WT UL37N [47]. Purified UL37N-R2 was concentrated to 3.8–4.0 mg/ml using an Ultra-15 30-kDa cutoff concentrator (Millipore) and stored at 4°C in 20 mM PIPES [pH 7.0], 50 mM NaCl, and 0.1 mM TCEP.
Crystals of UL37N-R2 were grown by vapor diffusion at room temperature in hanging drops containing 2 μl protein at ~ 4 mg/ml and 2 μl crystallization solution (30–32% PEG1000, 0.3 M Ca(CH3C00)2 and 0.1 M imidazole [pH 8.0]). Unlike WT UL37N, the UL37N-R2 mutant required a higher concentration of PEG1000 for crystallization. Large plate-like crystals appeared between 2 and 7 days and were harvested 1–4 weeks later. During harvesting, crystals were soaked in well solution containing 10% glycerol for 1–2 minutes prior to flash freezing in liquid nitrogen. Diffraction data were collected at 100 K on beamline 24-ID-C at the Advanced Photon Source at Argonne National Labs and processed in HKL2000 [90]. Whereas the WT UL37N (PDB ID 4K70) crystallized in space group P21 with 2 molecules per asymmetric unit [47], crystals of the UL37N-R2 mutant took space group P2221 with 1 molecule per asymmetric unit (S3 Table). The non-crystallographic 2-fold symmetry axis in the WT UL37N crystals became the crystallographic 2-fold symmetry axis in the UL37N-R2 crystals. Phases were obtained by molecular replacement using a single copy of WT UL37N structure (PDB ID 4K70) [47] as a search template, as implemented in Phaser [91]. The molecular replacement solution was used as a starting model for refinement in phenix.refine [92] using data to 2.48 Å resolution. Prior to refinement, 9% of the data were set aside for cross-validation, and residues Q324, D362, R365, H421, and H425 were substituted with alanines. The model was refined in phenix.refine [92] with iterative rounds of rebuilding in Coot [93]. A composite omit simulated annealing map was used to check the model. Refinement included rigid-body refinement, gradient minimization refinement of XYZ coordinates, individual atomic displacement parameter refinement, and refinement of TLS parameters, all as implemented in phenix.refine. The final Rwork is 17.81% and Rfree is 23.66%. All relevant crystallographic statistics are listed in atomic coordinates (S3 Table), and structure factors for UL37N-R2 were deposited to the RCSB Protein Data Bank under accession number 5J2Z.
All recombinant PRV (strain Becker) were derived from the pBecker3 infectious clone and are listed in S2 Table [94]. The PRV R1, R2, and R3 mutant viruses were previously described [47]. Variants of the pBecker3 infectious clones encoding a translational fusion of eGFP to envelope or tegument proteins were produced by an additional round of En Passant mutagenesis using the pEP-EGFP-in template (a kind gift from Nikolaus Osterrieder) with previously described primers [16,95]. New EP primers were designed for translational fusion of eGFP to PRV pUL37: 5’- CGTGGATCGGCTCCTGGCCGAGGTCGACGCCGTGTCCAAAGTGAGCAAGGGCGAGGAG and 5’- GGCTGAAATAACACACGCGCGTGGGGGAAAAATCTTTTTACTTGTACAGCTCGTCCATGC. For translational fusion of beta-lactamase to PRV pUL35, a new En Passant template plasmid was first constructed: pEP-Bla-in. This was accomplished by first producing a pUC-based plasmid encoding resistance to kanamycin and chloramphenicol (pGS3670). The kanamycin cassette was then removed by AvrII digestion and self-ligation, resulting in a pUC plasmid only encoding resistance to chloramphenicol (pGS3672). This latter plasmid served as the vector for cloning a partial duplication of the beta-lactamase coding sequence containing an aphAI gene (kanamycin resistance) and an I-SceI cleavage site between the duplicated sequences. This was achieved by digesting pBR322 with HindIII and PstI to remove a portion of the beta-lactamase coding sequence, and ligating in a PCR product of an oversized copy of the removed fragment. The PCR was performed with primers 5’ CC-AAAGCTTATCGATGATAAGCTGTC (reproduction of the vector HindIII site is underlined) and 5’ GCGATGCAT-AGATCTTTATCAGCAATAAACCAGCCAG (NsiI and BglII sites underlined). The NsiI site derived from the second primer was used to clone the PCR product into the vector PstI site, producing the desired duplication within the beta-lactamase coding sequence along with a BglII restriction site at its center (pGS3552). The duplicated beta-lactamase with I-SceI + kanamycin insertion was then PCR amplified in its entirety using primers 5’ GGAAGCTTGCTCACCCAGAAACGCTG (HindIII site underlined) and 5’ CGGGATCCCCAATGCTTAATCAGTGAGGC (BamHI site underlined), which was then cloned into pGS3672 cut with HindIII + BamHI, resulting in pGS3681. Finally, the aphAI gene and I-SceI cleavage site cassette from pEP-EGFP-in was subcloned into the unique BglII site by PCR amplification using primers carrying 5’ BglII sites, resulting in pEP-Bla-in. En Passant mutagenesis was performed with primers 5’- GGGCGCGCACAGACGCGCGCTCCCCGCCGAGCCATCATGGCTCACCCAGAAACGCTG and 5’- CTGCGCGGTGATCGTCCGGGGATTGTTCGGGTCGAAGGACCAATGCTTAATCAGTGAGGC off of pEP-Bla-in to insert beta-lactamase into PRV UL35.
The HSV-1 R2 mutant was derived from a variant of the HSV-1 strain F infectious clone, pGS5923 [96,97]. For the latter, codon changes were introduced through three rounds of En Passant mutagenesis. The first set of primers, 5’- CTGCGAGGCGGTCGGCCTGTCGGGGGGCGTTCTGAGCGCGACGCTGGCGGCTATCATGGGTCCGGCCAGGATGACGACGATAAGTAGGG and 5’- CCGCAGGCTCGCCAGATGTTCCGTCGGCACGGCCGGACCCATGATAGCCGCCAGCGTCGCGCTCAGAACGCCCCAACCAATTAACCAATTCTGATTAG, were used to introduce the Q511A/R515A mutations and produced pGS6151. The second set of primers 5’-GCGGCGGGCGTCCCCGCGCGGACGCCCGCCGGCCACGGACTCGGCGCAGTCCAGGCGCTCTTTGGGTGCATTAGGATGACGACGATAAGTAGGG and 5’-CAGACCAAACACGTTCGACCCCGCGAGGGCAATGCACCCAAAGAGCGCCTGGACTGCGCCGAGTCCGTGGCCCAACCAATTAACCAATTCTGATTAG, were used to introduce the E452A/Q55A mutations into pGS6151 to produce pGS6207. The third set of primers 5’- CAGCAGCGGCTGCTGGCTCTGCTGCAGCAGACGTGGACGTTGATCGCGAATACCAATTCGCCCAGGATGACGACGATAAGTAGGG and 5’- AGCGTCGATCAGGGTGTTGATCACCACGGAGGGCGAATTGGTATTCGCGATCAACGTCCACGTCAACCAATTAACCAATTCTGATTAG, were used to introduce the Q403A mutation into pGS6207 to produce the completed R2 mutant: pGS6264. A variant of pGS6264 encoding a translational fusion of the mCherry tag with the pUL25 capsid protein was produced by an additional round of En Passant mutagenesis as previously described, resulting in pGS6298 [98]. Variants of pGS4553 and pGS6298 infectious clones encoding a translational fusion of eGFP to pUL47 were produced by En Passant mutagenesis using primers 5’-GGTAGCGGACATCCGATAACCCGCGTCTATCGCCACCATGGTGAGCAAGGGCGAG and 5’-TGGATGCGCGCCTCCTGCGCCCCGCGGGTTCGCGAGCCGACTTGTACAGCTCGTC.
Vero (African green monkey kidney epithelial, ATCC), and PK15 (pig kidney epithelial, ATCC) cells were grown in DMEM (Dulbecco’s Modified Eagle Medium, Invitrogen) supplemented with 10% BGS (bovine growth supplement, HyClone), and were confirmed mycobacterium free and authenticated by the source. BGS levels were reduced to 2% during infection. All PRV strains were produced by electroporation of pBecker3 derivatives into PK15 cells as previously described [13]. The harvested virus was passaged once more on PK15 cells at low multiplicity to create a working stock. Titers of working stocks were obtained by plaque assay on PK15 cells as previously described [26]. HSV-1 strains were produced by electroporation of infectious clones into Vero cells. Using an ECM 630 electroporation system (BTX Instrument Division, Harvard Apparatus) cells were pulsed once with the following settings: 220V, 950μF, 0Ω. Serum levels were reduced to 2% BGS approximately 12 h after electroporation. Virus was harvested at a time at which 100% of the cells displayed pronounced cytopathic effect (CPE) (typically 3–5 days post electroporation).
For single-step growth curves, HSV-1 was harvested from Vero cells and supernatants at 2, 5, 8, 12, and 24 hours post infection [99]. Titers were determined by plaque assay on Vero cells overlaid with 2 ml methocel media (DMEM supplemented with 2% BGS and 10 mg/ml methyl cellulose) and allowed to expand for four days. Plaque diameters were measured using Vero cells plated in 6-well trays infected with approximately 100 plaque forming units (PFU) per well. Images of at least 30 isolated plaques from each infection were acquired with a Nikon Eclipse TE2000-U inverted microscope fitted with a 0.30 numerical aperture (NA) 10 × objective and RFP filter set. To determine the plaque diameter, the average of two orthogonal diameter measurements was calculated for each plaque using ImageJ software [100]. Plaque diameters were expressed as a percentage of the diameter of wild-type HSV-1, which was always measured in parallel. Data sets were plotted using GraphPad Prism 6 (GraphPad Software Inc.).
Dorsal root ganglia (DRG) from embryonic chicken (E8-E9) were cultured for 2 to 3 days on poly-DL-ornithine- and laminin-treated coverslips in 2 ml of F12 media (Invitrogen) containing nutrient mix: 0.08 g/ml bovine serum albumin fraction V powder (VWR), 0.4 mg/ml crystalline bovine pancreas insulin (Sigma-Aldrich), 0.4 μg/ml sodium selenite (VWR), 4 μg/ml transferrin (Intercell Technology) and 5 ng/ml nerve growth factor (NGF; Sigma-Aldrich). In a subset of experiments, DRG were cultured in polydimethylsiloxane (PMDS) chambers prepared using an epoxy mold (kindly provided by Eran Perlson, Tel Aviv University) [64]. Chambers were placed on plasma-cleaned coverslips treated overnight with poly-DL-ornithine and subsequently overnight with laminin. Laminin solution was replaced with F12 media prior to the plating of a single DRG explant in the proximal well adjacent to the chamber microgrooves (see illustration within S2 Movie). The explant was cultured in the chamber for 3 days to allow axons to grow to the distal channel. Infection was performed by replacing the F12 media in the distal wells and channel with 7.0 x 107 PFU PRV. Time-lapse imaging of mCherry and eGFP emissions was achieved by automated sequential capture using 100 ms exposures for each channel between 0.5–1.0 hpi.
PK15 cells were grown to confluence in 12-well plates. At the time of the assay, cells were placed on ice and washed once with PBS prior to infection with PFU/well of PRV diluted in serum free media. Cells remained on ice for 1 hour and were then moved to 37°C until processing at the indicated times post infection. During processing the viral inoculum was removed and replaced with either 2 ml of PBS or 2 ml of a citrate solution (40 mM citrate, 10 mM KCl, 0.135 M NaCl [pH 3]). Cells were incubated at room temperature for 1 minute at which time the PBS was removed and replaced with 2 ml methocel media. For citrate-treated wells the citrate was removed and cells washed twice with PBS prior to addition of methocel media. Cells were then incubated at 37°C for an additional 3 days and the number of plaques that formed were counted. For each time point the percent internalization was calculated using the equation: (No. of plaques on citrate-treated well/No. of plaques on PBS-treated well) × 100. Three independent experimental replicates were performed and the results were plotted using GraphPad Prism 6.
To measure penetration kinetics into primary sensory neurons, an adaptation of a beta-lactamase assay was developed. Chick dorsal root ganglia explants were cultured for 2–3 days as described above. Media was removed and replaced with CCF2/AM Live-Blazer dye solution (Invitrogen): 0.6 μl CCF2/AM, 5.4 μl Solution B, 79 μl Solution C, 15 μl 0.1 M Probenecid, and 500 μl F12 media supplemented with nutrient mix (see above) and NGF. Explants were loaded with the CCF2/AM dye solution for 45 minutes at 37°C then washed three times with F12 media. The CCF2-loaded explants were infected with 8.0 x 107 pfu/ml of the indicated recombinant PRV strains. Axons were imaged from 10–75 minutes post infection using an inverted wide-field Nikon Eclipse TE2000-U microscope fitted with a 60x/1.4 NA objective, a Photometrics CoolSnap HQ2 camera, and a Beta-lactamase Ratiometric filter set (Chroma). Sequential images were captured using DIC, followed by excitation with a HQ405/20x filter and capture with HQ460/40m and HQ530/30m emission filters. 100 ms exposures were used for all images. To avoid unintended bias, CCF2 cleavage was quantitated by drawing a region of interest (ROI) along an axon on the DIC image, which was then transferred to the 460 nm and 530 nm images. The average fluorescence intensity of the 460 nm ROI was divided by the corresponding average fluorescence intensity of the 530 nm ROI. Ratiometric values from 20–30 ROIs were recorded for each time point. This procedure was repeated for three independent experiments. The MetaMorph software package (Molecular Devices) was used for image analysis.
PK15 cells were grown to confluence in 6-well plates and infected with PRV at a MOI of 1. Total RNA was isolated at 4 hpi with TRIzol (Invitrogen) according to the manufacturer’s instructions. Contaminating DNA was removed with RQ1 RNase-free DNase (Promega). RNA concentration was determined by absorbance at 260 nm with a NanoDrop 8000 (Thermo scientific). Gene-specific complementary DNA (cDNA) was synthesized using the SuperScript III First-Strand Synthesis System (Invitrogen). SYBR Green-based quantitative real-time PCR was performed on cDNA in a LightCycler 480 II (Roche). All reactions were carried out in 10 μl volumes: 2 μl cDNA, 5 μl LightCycler 480 SYBR Green I Master (Roche), 0.5 μl forward primer and 0.5 μl reverse primer (0.5 μM each), and 2 μl water. The running conditions and primer sequences were based on a previous study [101]. A control sample lacking the reverse transcriptase enzyme was included in parallel. Specificity of primer binding was confirmed by melting point analysis. The crossing points (CP) for each transcript was determined by “Fit Point Method” using LightCycler 480 II software 1.5.0 SP3 (Roche). IE180 mRNA levels were normalized to S28 rRNA, and the fold change of between R2 and WT PRV infections was averaged across three independent experiments. A subset of cells was treated with 9 μM nocodazole for 1 hr prior to infection and maintained in nocodazole throughout the 4 hr infection.
Virus transport dynamics were monitored in primary DRG explants. Explants were infected in 2 ml of media with 3.5 x 107 PFU/ml of PRV (WT and R2 mutant) or 1.3 x 107 PFU/ml of HSV-1 (WT and R2 mutant) from 3–4 hpi. Time-lapse images were captured using an inverted wide-field Nikon Eclipse TE2000-U microscope fitted with a 60x/1.4 NA objective and a CascadeII:512 electron-multiplying charge-coupled device (EM-CCD; Photometrics). The microscope was housed in a 37° environmental box (In Vivo Scientific). Moving particles were detected by time-lapse fluorescence microscopy in the red-fluorescence channel at 10 frames per second (continuous 100 ms exposures) for 100 or 500 frames. Particle trajectories were traced in the 100 frame time-lapse image stacks using a multi-line tool with a width of 20 pixels and average background subtraction, and a kymograph was produced using the MetaMorph software package (Molecular Devices). The multi-line tool was again employed to trace kymograph paths, and the fraction of time that a particle was stopped, moving anterograde, or moving retrograde was calculated for each particle. Graphs were created in GraphPad Prism 6.
Virus composition in axons of DRG explants was monitored using dual-fluorescent viruses encoding pUL25/mCherry and either eGFP-pUL47, pUL49-eGFP, gD-eGFP, pUL36-eGFP, or pUL37-eGFP (S2 Table). Time-lapse imaging of mCherry and eGFP emissions was achieved by automated sequential capture with 100 ms exposures for each channel at 3–4 hpi. Red-fluorescent capsids that moved > 2.5 μm were scored for the presence of a coincident eGFP signal. Extracellular viral particle composition was analyzed by collecting supernatants from infected PK15 (PRV) or Vero (HSV-1) cells as previously described [95]. Briefly, cells in 10 cm dishes were infected at a MOI of 5 PFU/cell and incubated for 18 hrs in F12 media lacking phenol red (Gibco) and supplemented with 2% (vol/vol) BGS (HyClone). Supernatants were harvested and cleared of cell debris with a 10 min 3,000 × g centrifugation. Next, 8 ml of cleared supernatant was transferred to a SW41 centrifuge tube and underlaid with a 1 ml cushion of 10% (wt/vol) Nycodenz (Accurate Chemical) in PBS. Samples were centrifuged at 38,500 × g for 60 min. Media and the Nycodenz cushion were gently aspirated from the virus pellet, which was resuspended in 0.1 ml PBS. Prior to imaging, the resuspended virus particles were diluted 1:50 in PBS and 0.07 ml was transferred into a microscope slide chamber consisting of a plasma-cleaned No. 1.5, 22 × 22-mm coverslip. Images were captured using 1.7 s exposures with a CoolSnap HQ2 camera (Photometrics).
Imaging of capsid delivery to nuclear rims of DRG sensory neurons was conducted following co-infections with pUL25/mCherry and pUL25/eGFP encoding viruses. DRG explants were infected in 2 ml of media with 3.5 × 107 PFU/ml of each virus (PRV co-infections) or at 1.0 × 107 PFU/ml of each virus (HSV-1 co-infections). Infected DRG were imaged between 3–4 hpi using a Nikon Ti inverted microscope fitted with a 100 × 1.45 NA objective (Nikon Instruments) and coupled to a CSU-W1 confocal head (Yokogawa Electric Corporation) and a CascadeII:1024 EM-CCD (Photometrics). Illumination was provided by a Sapphire 561 laser (Coherent) and custom laser launch (Solamere Technology Group, Inc.). Nuclei were identified using differential interference contrast imaging.
Anterograde axonal transport following virus replication in DRG sensory neurons was monitored as previously described [55]. Briefly, triturated neurons were seeded on poly-DL-ornithine treated coverslips at low density and cultured for 2 days prior to infection in 2 ml of media with 2.5 × 106 PFU/ml of either wild-type or the R2-mutant PRV. Actively transporting PRV particles were imaged between 10–13 hpi at 10 frames/s (100 ms exposures) for 100 frames, and transport dynamics were measured by kymograph analysis as described above. Particle accumulation at axon terminals was imaged using 1.7 sec exposures on an inverted wide-field Nikon Eclipse TE2000-E fluorescence microscope fitted with a 60 × 1.4 NA objective and a Photometrics CoolSnap HQ2 camera. Particle counts at axon terminals were determined manually across three independent experiments with at least 10 terminals examined per sample.
All procedures conformed to NIH guidelines for work with laboratory animals and were approved by the Institutional Animal Care and Use Committee of the University of Nebraska, Lincoln (Protocol: 1086) and Northwestern University (Protocol: IS00003334). Fertilized chicken eggs were obtained from Sunnyside, Inc. and tissue were harvested between embryonic day 8 and 10.
Intranasal inoculations were performed as previously described [98]. Briefly, male CD-1 mice (6 wk old; Charles River) were maintained for at least 2 weeks under a 12:12 hr light/dark cycle, with 2–3 mice per cage and food and water freely available. Intranasal application of PRV was administered to animals anesthetized by 2.5% isoflurane inhalation. Viral stocks were stored frozen at −80°C and were used immediately after being thawed. Animals received 5 μl of PRV (1.0–6.4 x 108 PFU/ml) in either one or both nostrils. For vaccination studies, animals received 5 μl of the PRV R2 mutant (6.4 x 108 PFU/ml) in each nostril and a second intranasal instillation of 5 μl of WT virus (8.8 x 109 PFU/ml) at 3 to 28 days post vaccination. Behavior was continuously video monitored with images captured every 10 min. Survival times post-inoculation were rounded to the nearest hour. Surviving animals were euthanized at the end of the experiment.
Intraocular injections were performed as previously described [41]. Briefly, male Long Evans rats (8 wk old; Charles River) were maintained two per cage under a 12:12 hr light:dark cycle for at least 2 weeks with food and water freely available. Under isoflurane inhalation anesthesia (2.5%), animals received either an anterior chamber injection of 2 μl of PRV (0.2–4.0 x 109 PFU/ml) or an intravitreal chamber injection of 10 μl of PRV. Injections were performed over a 2 min interval using a Hamilton syringe fitted with a 30 g needle. A fresh stock of virus was thawed for each experiment. Animals were maintained in a biosafety level 2 facility for up to 144 hrs post-inoculation.
Intracranial injections were performed as previously described [40]. Briefly, under isoflurane inhalation anesthesia (2.5%), male Long Evans rats (6 wk old, Charles River) were positioned in a Kopf stereotaxic apparatus (David Kopf Instruments). A craniotomy was performed, and a micropipette attached to a Nanoject II nanoinjector (Drummond Scientific Co.) was lowered to the appropriate position and approximately 0.7 μl of PRV was injected. The micropipette was kept in place for 2 min, elevated 100 μm and another 0.7 μl injected, and kept in place an additional min before being withdrawn. The craniotomy was subsequently packed with gel foam, the incision was sutured, and the animals were monitored in a recovery area until fully mobile. Animals were maintained in a biosafety level 2 facility for 48–102 h post-injection and were prepared for histological analysis. Briefly, under deep anesthesia, animals were intracardially perfused with 0.9% saline followed by 4.0% paraformaldehyde in 0.1 M phosphate buffer (pH 7.3). Retinas were removed and imaged as flat mounts whereas brains were post-fixed overnight in the same fixative containing 20% sucrose and sectioned at 40 μm with a CM1850 cryostat (Leica). Sections were mounted on slides, coverslipped with Vectashield (Vector Laboratories) and imaged with a DM5500B fluorescence microscope (Leica) equipped with a C11440 Orca-flash 4.0 digital camera (Hamamatsu).
Imaging of fluorescently-tagged HSV-1 infected tissues was performed following corneal inoculation of DBA2 mice (6 wk old; Charles River). The animals were anesthetized (2.5% isoflurane inhalation), the corneas gently abraded with a 25-gauge needle, and 5 μl of HSV-1 (6–10 x 107 PFU/ml) was administered to the corneal surface. Two to six days later animals were anesthetized and prepared for histological analyses as described above. Corneas were prepared as flat mounts and imaged. Brains and trigeminal ganglia were post-fixed, sectioned at 40 μm and imaged as described above. In separate experiments, HSV-1 infections of BALB/c mice (9 wk old; Jackson Lab) were carried out in animals anesthetized with an intraperitoneal injection of a ketamine (86.98 mg/kg) and xylazine (13.04 mg/kg) mixture. Each cornea was lightly abraded 10 times in a crosshatched pattern with a 25-gauge needle, and 7 μl of HSV-1 (9 x 107 PFU/ml) was administered to the cornea surface. Prior to infection, the virus stock was sonicated and centrifuged for 2 min at 300 x g to remove cell debris. Eye swabs were collected by lightly anesthetizing the mice with isoflurane, gently proptosing each eye, and wiping a DMEM-moistened sterile cotton swab three times around the eye in a circular motion and twice across the center of the cornea in an “X” pattern. The swabs were placed in 1 ml of DMEM and stored at −80°C. Before titering, the swabs were thawed and vigorously vortexed for 30 seconds. At the indicated day post infection each trigeminal ganglia and the brain were removed and individually homogenized in 1 ml DMEM, sonicated, and stored at −80°C. Titers of recovered HSV-1 from eye swabs and tissues were determined on Vero cells as described above. For reactivation studies whole trigeminal ganglia were harvested and bisected prior to plating on a monolayer of Vero cells. Cells were monitored for CPE for 7 days. If no CPE was observed, cells and tissue were collected, homogenized, sonicated, and 100 μl of the homogenate was plated on a monolayer of Vero cells. Cells were monitored for 10 days for signs of CPE. Tissue was scored positive if CPE was detected from the intact or homogenized tissues.
For HSV-1 vaccination studies, BALB/c mice were inoculated in the left eye with either 7 μl of the HSV-1 R2 mutant stock (9 x 107 PFU/ml), or 7 μl of conditioned media following corneal scarification. Conditioned media was harvested from Vero cells grown in DMEM supplemented with 10% BGS: cells and media were collected together and frozen, and prior to use were thawed, sonicated, and centrifuged to remove cell debris to mimic the handling of the HSV-1 stocks. At 14 days post vaccination mice were challenged with 7 μl of wild-type HSV-1 strain F (1.0 x 108 PFU/ml) or 5 μl of HSV-1 McKrae (6.0 x 108 PFU/ml) by inoculation of the right eye following corneal scarification.
DNA was isolated from homogenized brain and trigeminal ganglia using the DNeasy Blood and Tissue Kit (Qiagen). For brain samples the DNA was used at a final concentration of 100 ng/μl. For TG samples the final DNA concentration varied between 10–20 ng/μl. Each sample was run in triplicate using a 10 μl reaction volume consisting of: 5 μl of CyberGreen Mastermix (Roche), 0.5 μl forward primer (30 μM), 0.5 μl reverse primer (30 μM), 1.5 μl water, and 2.5 μl of DNA. Run settings were 95°C for 10 min, 50 cycles of 95°C for 15 sec, and 60°C for 30 sec. The forward and reverse primer sequences were previously published [102]: HSV-1 UL35Fwd: GTCTTGGCCACCAATAACTC; HSV-1 UL35 Rev: GGGTAAACGTGTTGTTTGCG; mGAPDHFwd: GATGGGTGTGAACCACGAG, and mGAPDHRev: GTGATGGCATGGACTGTGG.
The statistical tests for all data are justified and the data meets the assumptions of the test. There is an estimate of variation within each group of data, however, the variance is similar between the groups that are being statistically compared. Significance between specific data sets is described in the respective figure legends.
|
10.1371/journal.pgen.1000982 | Arabidopsis thaliana Chromosome 4 Replicates in Two Phases That Correlate with Chromatin State | DNA replication programs have been studied extensively in yeast and animal systems, where they have been shown to correlate with gene expression and certain epigenetic modifications. Despite the conservation of core DNA replication proteins, little is known about replication programs in plants. We used flow cytometry and tiling microarrays to profile DNA replication of Arabidopsis thaliana chromosome 4 (chr4) during early, mid, and late S phase. Replication profiles for early and mid S phase were similar and encompassed the majority of the euchromatin. Late S phase exhibited a distinctly different profile that includes the remaining euchromatin and essentially all of the heterochromatin. Termination zones were consistent between experiments, allowing us to define 163 putative replicons on chr4 that clustered into larger domains of predominately early or late replication. Early-replicating sequences, especially the initiation zones of early replicons, displayed a pattern of epigenetic modifications specifying an open chromatin conformation. Late replicons, and the termination zones of early replicons, showed an opposite pattern. Histone H3 acetylated on lysine 56 (H3K56ac) was enriched in early replicons, as well as the initiation zones of both early and late replicons. H3K56ac was also associated with expressed genes, but this effect was local whereas replication time correlated with H3K56ac over broad regions. The similarity of the replication profiles for early and mid S phase cells indicates that replication origin activation in euchromatin is stochastic. Replicon organization in Arabidopsis is strongly influenced by epigenetic modifications to histones and DNA. The domain organization of Arabidopsis is more similar to that in Drosophila than that in mammals, which may reflect genome size and complexity. The distinct patterns of association of H3K56ac with gene expression and early replication provide evidence that H3K56ac may be associated with initiation zones and replication origins.
| During growth and development, all plants and animals must replicate their DNA. This process is regulated to ensure that all sequences are completely and accurately replicated and is limited to S phase of the cell cycle. In the cell, DNA is packaged with histone proteins into chromatin, and both DNA and histones are subject to epigenetic modifications that affect chromatin state. Euchromatin and heterochromatin are chromatin states marked by epigenetic modifications specifying open and closed conformations, respectively. Using the model plant Arabidopsis thaliana, we show that the time at which a DNA sequence replicates is influenced by the epigenetic modifications to the surrounding chromatin. DNA replication occurs in two phases, with euchromatin replicating in early and mid S phase and heterochromatin replicating late. DNA replication time has been linked to gene expression in other organisms, and this is also true in Arabidopsis because more genes are active in euchromatin when compared to heterochromatin. The earliest replicating DNA sequences are associated with acetylation of histone H3 on lysine 56 (H3K56ac). H3K56ac is also abundant in active genes, but the patterns of association of H3K56ac with gene expression and DNA replication are distinct, suggesting that H3K56ac is independently linked to both processes.
| DNA replication is a fundamental process required for the growth and development of all eukaryotes. This process is regulated both spatially and temporally so that all DNA sequences are replicated exactly once during S phase, insuring that each daughter cell receives a complete copy of the genome. DNA replication initiates from discrete locations on chromosomes known as replication origins (origins) where proteins required for DNA synthesis are recruited by the origin recognition complex (ORC). Once initiated, DNA replication proceeds by elongation to regions where opposing replication forks converge (termination zones). This organization of DNA sequences into regions of initiation, elongation and termination define a replicon – a segment of DNA replicated as a unit by replication forks originating from a single origin [1]–[5]. The time of replication for any particular DNA sequence within a replicon is determined by three factors: its proximity to an origin, the efficiency of initiation at that origin, and the rate of DNA elongation in that region.
The pattern of DNA replication has been determined for multiple eukaryotic genomes ranging from the compact genome of budding yeast to the moderately sized genome of Drosophila melanogaster and the large human and mouse genomes [6]–[14]. In budding yeast, DNA sequences acting as origins have a conserved consensus motif, and origin activation appears to follow a strict temporal program [6]. However, recent single molecule studies of DNA replication in yeast [15], [16] suggest that the temporal program likely represents the average replication program for a population of cells, with considerable variation in the order of origin activation in individual cells [17]–[20]. In higher eukaryotes, no consensus sequence for origin DNA has been identified, and some known origins are organized as broad initiation zones containing multiple potential origins [2]–[4]. It is unclear whether origin activation follows a temporal sequence in higher eukaryotes, but origin activation in Drosophila is most prevalent in early and late S phase, suggesting some degree of temporal regulation [14]. In mammals, clusters of replicons frequently display coordinate origin activation and are organized into larger replication domains [1], [5], [21]. The organization of replication domains appears to be cell type specific, as differentiation of embryonic stem cell lines to neural precursor cells resulted in the widespread reorganization of replication domains [13]. Differences in replication patterns between cell types have been linked to changes in gene expression and epigenetic modifications [13], [14].
The relationship between gene expression and replication time has been examined in yeast, Drosophila, mouse and human cells. In budding yeast, there is little correlation between replication time and gene expression [6]. In higher eukaryotes with more complex genomes, there is a positive correlation between early replication and gene expression, and this correlation is strongest when integrated over large chromosomal domains [7], [8], [10]–[14], [22], [23]. The fact that an open chromatin conformation is necessary but not sufficient for both DNA replication and gene expression may underlie the correlation between these processes [2]–[4], [11], [21], [24].
In general, euchromatin replicates early in S phase and heterochromatin replicates late, although specific types of heterochromatin replicate in early S phase in yeast [6], [8], [21], [25], [26]. Chromatin is subject to a plethora of epigenetic modifications including histone methylation, histone acetylation and DNA cytosine methylation (5mC). The combinatorial effect of these modifications, as well as the association of other chromatin-binding proteins, determines whether DNA adopts a heterochromatic or euchromatic conformation [27]–[29]. Epigenetic modifications associated with heterochromatin and characteristic of silenced genes and transposable elements include tri- and dimethylation of histone H3 lysine 9 (H3K9me), hypoacetylation of histones, and abundant 5mC [27]–[34]. There are conflicting reports for the correlation between heterochromatic marks and late replication, which is surprising given the tight relationship between late replication and heterochromatin [12], [13], [35]. Modifications associated with euchromatin and active or potentially active genes include tri-, di- and monomethylation of histone H3 lysine 4 (H3K4me), hyperacetylation of histones, and 5mC localized to gene coding sequences [27]–[30], [32]–[34], [36]–[38]. Several replication timing studies showed a positive correlation of early replication with H3K4me [12], [13], [35], which may be indirect because H3K4me is associated almost exclusively with genes and gene-rich regions which tend to replicate early [7], [8], [11].
Several lines of evidence suggest that the link between histone acetylation and replication time is more direct. Hyperacetylaton of histone H3 on lysines 9 and 14 (H3K9/14ac) associates with origins in human cells [39]. Hyperacetylation of histone H3 lysine 56 (H3K56ac) associates with early firing origins in budding yeast [40]. Hyperacetylation of histone H4 lysine 16 (H4K16ac) associates with early replicating regions in Drosophila cells [14]. In addition, late-firing origins in budding yeast are regulated by a histone deacetylase complex [41]. These and other experiments suggest that histone acetylation may be the best epigenetic determinant of replication time [24], [42]–[44].
Very little is known about the regulation of DNA replication in plants [45]. The core proteins required for DNA replication are conserved between yeast, plants and animals [46], [47]. The replication machinery of plants is more similar to animals than yeast, but many of the genes encoding these proteins have multiple homologs in Arabidopsis thaliana suggesting that functional diversification has occurred [47]. DNA fiber autoradiography studies revealed that Arabidopsis possesses two families of replicons, one initiating replication early and the other later in S phase [48]. These likely correspond to euchromatic and heterochromatic replicons because, like most eukaryotes, plants replicate heterochromatin later than euchromatin [25].
In contrast, knowledge of epigenetic modifications in Arabidopsis has kept pace with other systems, and with few exceptions, these modifications are functionally conserved between plants and animals [28], [29], [49]. The relationship between epigenetic modifications and DNA replication in plants is virtually unexplored. However, DNA replication is required to maintain the repressed state of a negative regulator of flowering in Arabidopsis [50], suggesting that the interplay of these processes is crucial for plant growth and development. Similar to the replication machinery, the genes encoding DNA and histone modifying enzymes often have multiple homologs in plants [51]–[53].
Arabidopsis with its small, well-characterized genome is an excellent model system for examining the global relationship between DNA replication and chromatin state in higher eukaryotes. The genome of Arabidopsis is gene-dense in comparison to mammalian genomes, with roughly the same number of genes encoded by a genome one-twentieth the size [54]–[56]. The genome size of Drosophila is similar but encodes half the number of genes [57]. This characteristic of Arabidopsis may provide insight into the influence of gene density on DNA replication. In addition, analysis of Arabidopsis DNA replication has the potential to uncover features that are unique to plants. The diversification of genes encoding replication-associated proteins and chromatin modifiers suggests that plants may have developed unique mechanisms to regulate DNA replication and to establish and maintain chromatin states. These mechanisms may be related to developmental pathways that are common in plants but rare in other systems. For example, endoreduplication plays a prominent role in plant development and totipotency of plant cells is not limited to germline or embryonic cells. We used a combination of fluoresence-activated cell sorting (FACS) and genomic tiling arrays to profile DNA replication of Arabidopsis chr4 in early, mid and late S phase cells. We investigated the relationship between DNA replication, gene expression and chromatin state in analyses of our data and the extensive genomic data available for Arabidopsis chr4.
We used an established Arabidopsis Col-0 suspension cell line for the analysis of replication time and optimized the culture conditions to provide ample nuclei from replicating cells for fractionation by FACS (Figure S1 and Text S1). This cell line was also used in recent studies that examined the effects of cell culture on specific epigenetic modifications [34]. We first characterized the relationship between DNA content and replication in this cell line by monitoring the incorporation of the nucleotide analog bromodeoxyuridine (BrdU). An asynchronous population of cells was labeled with BrdU for 1 hour and fixed. Nuclei were isolated, stained with propidium iodide, labeled with a fluorescent anti-BrdU antibody, and analyzed by FACS for DNA content and BrdU incorporation. Nuclei in S phase that incorporated BrdU appeared as a distinct “arc” above the population of cells in G1 and G2/M (Figure 1A). Surprisingly, almost 30% of the S phase nuclei fractionated above the G1 peak, and we designated this population early S phase (Figure 1A and Table S1). Similarly, we designated the 50% of the S phase nuclei that fractionated above the G2/M peak as late S phase. The remaining 20% of S phase nuclei between the G1 and G2/M peaks were designated mid S phase. We estimated the DNA content of the early, mid and late S phase populations at 1.16, 1.49 and 1.95C, respectively (Figure 1B). This distribution of S phase nuclei and DNA content indicated that to get a complete picture of DNA replication during S phase we needed to analyze DNA replication in nuclei that co-sorted with G1 (early S phase) and G2/M (late S phase) peaks.
We profiled DNA replication independently in early, mid and late S phase. We could not sort the early, mid and late S phase nuclei based on BrdU content because visualization of the BrdU degrades DNA. Instead, nuclei were sorted based on DNA content, and BrdU-labeled DNA was separated by immunoprecipitation (Figure 1C). Nuclei in the early S/G1, mid S and late S/G2/M sorts contained different fractions of nuclei in S phase, with the early S/G1, mid S and late S/G2/M sorts containing 4.2, 42.3 and 18.3% BrdU-positive nuclei, respectively (Figure 1A and Table S1). Because of these differences, it was necessary to account for cross contamination associated with sorting (Figure S2A), especially contamination of mid S phase nuclei into the early S/G1 sort (Table S2 and Figure S2B). When corrected for the percentage of nuclei in S phase, we determined that the early, mid and late S phase purity was 69, 94 and 85% respectively (Table S2). In the worst case, 28% of S phase nuclei in the early S/G1 sort were actually in mid S phase (Table S2 and Figure S2B). However, this contaminating population had a DNA content from the lower tail of the mid S phase distribution (Figure S2B).
BrdU-labeled DNA from early, mid or late S phase nuclei was hybridized separately to a tiling microarray that covers 99% of the sequenced regions of chr4 of Arabidopsis thaliana with 22,761 PCR-generated probes averaging 1 kb in length [33]. This array was used previously to profile specific epigenetic modifications in this cell line [34]. Microarray results were confirmed by qPCR analysis of 14 selected regions (Tables S3 and S4, Figure S3 and Text S1).
Figure 2 shows a schematic representation of chr4 including plots for gene and transposable element (TE) coverage (Figure 2A) and GC content (Figure 2B). Chr4 is unusual in that it has three regions of constitutive heterochromatin – the nucleolar organizing region (NOR) at the end of the short arm (not shown), a 700 kb heterochromatic knob centered at 2 Mb, and 2.5 Mb of pericentromeric heterochromatin centered at 4 Mb (Figure 2C) [58], [59]. These heterochromatic regions were used as boundaries to subdivide chr4 into six regions for subsequent analyses – the distal short arm, the heterochromatic knob, the proximal short arm, the pericentromere, the proximal long arm and the distal long arm (Figure 2C). The boundaries of most of these regions are evident from gene and transposable element (TE) coverage and to some extent from the GC content profile (Figure 2A and 2B). The boundary between the proximal and distal long arms is less evident and was chosen based on the replication time results presented below.
The replication profiles were generated from the microarray data by applying a loess algorithm in a 150-kb window to smooth the probe-level data (Figure 2D). The early and late profiles display remarkable complementarity (R = −0.83), i.e. regions of chr4 enriched for BrdU in early S phase cells are depleted in late S phase cells. Early replication is most prevalent in the distal long arm, a euchromatic region rich in genes with few TEs. Late replication predominates in the heterochromatic knob and pericentromere of chr4, but regions of late replication are also dispersed in other parts of chr4, most notably the proximal long and short arms.
The replication profiles for early and mid S phase cells are surprisingly similar (R = 0.87) (Figure 2D). The most evident difference is a broadening and merging of early replicating regions in the mid S phase profile. The DNA replicating in mid S phase represents nearly the same population of sequences as that replicating in early S phase even though FACS analysis demonstrates that the early and mid S phase nuclei have notably different DNA content (Figure 1, Figure S2 and Tables S1 and S2). Like early S phase, the mid S phase profile is distinct from the late profile (R = −0.85). The similarity of the early and mid S phase profiles is not consistent with a fixed order of origin activation and, instead, suggests that origin activation in early and mid S phase is stochastic. Together, the early, mid and late S phase profiles suggest that DNA replication in Arabidopsis cells is biphasic, a result consistent with a previous report that Arabidopsis DNA replication takes place in two distinct stages [48].
To facilitate further analyses, we performed a two-step segmentation of the early, mid and late S phase profiles to assign a replication time for each microarray probe. Figure 3 illustrates this process for two chr4 regions representative of early and late replicating regions. In the first step, we identified contiguous segments of probes showing coordinate replication times (log2 ratio >0) within each smoothed profile, thereby defining segments of early, mid or late replicating DNA (Figure 3C and 3D). In the next step, we reconciled the replication times between experiments by determining the regions of overlap between the early, mid and late segments (Figure 3D). This analysis identified segments of DNA replicating only in early S phase (E), in both early and mid S phase (EM), only in mid S phase (M), in both mid and late S phase (ML), only in late S phase (L), in early and late S phase (EL), throughout S phase (EML), and segments of indeterminate replication time (I) that did not show enrichment in any experiment (Table S5).
The majority of chr4 replicates as either EM (37%) or L (44%) when segment length is taken into account (Figure 4A). Only 4% of chr4 replicates exclusively in mid S phase (M), while 6% replicates as ML and 6% replicates as EML. The positions of these segment types with respect to EM and L segments suggest that many of the M, ML and EML segments are regions of DNA elongation between EM segments or transition zones between early to late replication (Figure 3). In regions of predominately late replication, M, ML and EML segments are often located between larger flanking L segments (Figure 3), suggesting that they contain the DNA replication origins for the flanking regions. The EL segments comprise only 2% of chr4 and are enriched for repetitive sequences (Table S6). Thus, at least some EL segments are likely to be artifacts created by cross hybridization on the microarray. I segments, which comprise 2% of chr4, also have an elevated repeat content (Table S6). Another possible explanation for EL and I segments is that replication time in these regions is driven by allele-specific gene expression and/or epigenetic modifications (see below) [12], [60].
We also determined the replication times of the six chr4 regions defined in Figure 2. The heterochromatic knob and pericentromere replicate almost exclusively as L segments while the gene-rich distal long arm replicates predominately as EM segments (Figure 4A). The replication time of the distal short and proximal short regions is more complex, perhaps influenced by the flanking heterochromatic regions (Figure 2). The proximal long arm displays a surprising amount of late replication despite the fact that this region is not constitutive heterochromatin [59], although it does have lower gene and higher TE content than the distal long region (Figure 2).
Within a given replicon, the DNA closest to the origin will replicate earliest while the DNA located at termination zones, regions where opposing replication forks converge, will replicate latest. Replication time profiles have been used to identify both initiation and termination zones [6], [7], [14]. Initiation zones manifested as local maxima in the early and mid S phase profiles and as local minima in the late S phase profile (Figure 3C). Conversely, termination zones manifested as local minima in the early and mid S phase profiles and as local maxima in the late S phase profile (Figure 3C). We identified initiation and termination zones by computationally determining probes occurring at local maxima and minima in the loess smoothed profiles. We did not treat individual probes as initiation or termination zones and, instead, defined zones as 10 kb segments centered at the identified probes. Any zones that overlapped were then merged into a single zone. Replication time for each zone was determined from constituent probes.
The number of initiation and termination zones was consistent between experiments (Table S7 and Figure 3C). However, their positions were more consistent between the early and mid S phase profiles versus comparisons with the late S phase profiles, e.g. 80% of the initiation zones identified in the early S phase profiles are within 20 kb of an initiation zone identified in the mid S phase profiles while this figure dropped to 65% when comparing the early and late S phase initiation zones (Table S8). This difference is not unexpected given that initiation zones are more likely to replicate in early or mid S phase while termination zones are more likely to replicate in late S phase.
We then examined the frequency of initiation and termination zones as a function of replication time. In Drosophila, initiation sites are more abundant in late replicating DNA than in early replicating DNA with very little initiation occurring in mid S phase [14]. In Arabidopsis, we found that the distribution of replication times for the initiation zones reflected the distribution of replication times for chr4 with initiation zones prominent only in EM (37%) and L (42%) segments (compare Figure 4A and 4B). Thus, unlike Drosophila, initiation sites are no more abundant in late than in early replicating DNA. However, Arabidopsis appears similar to Drosophila [14] in that the majority of termination zones are located in L (61%) rather than EM (19%) segments (Figure 4B). These results indicate that DNA replication in late S phase includes elongation from origins that have fired earlier in S phase as well as initiation and elongation from late firing origins.
Higher eukaryotes do not possess replicons in the strictest sense of the term, but rather the concept of a “relaxed replicon” likely applies [2]–[4]. In this model, replication origins are not rigidly defined, and replicon boundaries can vary from cell to cell. We defined the boundaries of these relaxed replicons (hereafter referred to as replicons) using a subset of the termination zones. Where possible, we used termination zones that were identified in early, mid and late S phase cells. Where termination zones differed between experiments, we preferentially used termination zones enriched in late S phase cells or local minima from early or mid S phase cells for EM-replicating segments (Figure 3C and Table S7 and Materials and Methods). In this way, we identified 164 termination zones that defined 163 putative replicons across chr4 with a median length of 107 kb. This replicon size is consistent with previous measurements of single replicons in Arabidopsis [45], [48], although we cannot exclude the possibility that at least some of these replicons are clusters of smaller replicons. The majority (154) have at least one putative initiation zone (Figure 3C and Table S9). This strategy worked well for the euchromatic regions of chr4, particularly the distal long and distal short arms, where the predicted termination zones were consistent between early, mid and late S phase cells (Figure 3C). There was less agreement between profiles for the other chr4 regions, and replicon boundaries in the late-replicating regions are defined primarily from the late S phase profiles (Figure 3C and Table S7).
The assignment of a specific replication time to individual replicons is complex because a replicon can be comprised of DNA segments with replication times that cover the entirety of S phase. To simplify the analysis, we classified replicons based on the replication time of the probes comprising the greatest proportion of a replicon, e.g. a replicon comprised of 45% EM probes, 40% L probes and 15% M probes would be classified as EM. Figure 4C (top panel) shows a schematic representation of chr4 replicons with the replication times for the constituent probes. The complexity of replication time within replicons likely reflects several factors including time and efficiency of origin firing, the number of origins within initiation zones, and the rate of elongation by DNA polymerase in specific contexts [2]–[4], [24].
In Drosophila, the interval between termination zones varies between early S phase and late S phase, with increased initiation in late S phase resulting in more closely spaced termination zones [14]. The size of the Arabidopsis replicons does not vary significantly between EM and L replicons (Figure 4D). While M, ML and EML replicons are smaller than either EM or L replicons, the difference in size is not statistically significant (Figure 4D). The similar size of EM and L replicons follows from the previous observation that initiation zones are no more abundant in late replicating regions than in earlier replicating regions (Figure 4B).
In mouse cells, replicons are organized into replication domains consisting of large clusters of replicons with similar replication times [13], [22]. In Drosophila cells, clustering is less evident with replication profiles showing distinct peaks of early replication [14]. Arabidopsis appears more similar to Drosophila in this regard, but the 163 chr4 replicons could be organized into 41 replication domains based on their replication time (Figure 4C, middle panel, and Table S10). There are a few large replication domains, including a 4.5 Mb L domain (coordinates 2.6–7.1 Mb) that encompasses the entire pericentromere and portions of the proximal short and long arms, and a 2.3 Mb EM domain (coordinates 16.2–18.5 Mb) in the distal long arm (Figure 4C, middle panel). However, the mean length of chr4 replication domains is 450 kb which is considerably smaller than the 1 Mb reported for mouse cells [13]. This difference in replicon organization may be related to genome size. The genome sizes of Arabidopsis and Drosophila are similar at 115 and 122 Mb, respectively [54], [57], while the mouse genome is estimated at 2500 Mb [56].
Replication time has been correlated with both genetic and epigenetic features in other model systems [7], [8], [11]–[14]. The replication profiles (Figure 2) show that on the scale of the entire chromosome, EM replication is associated with euchromatic regions while L replication is associated with heterochromatic regions. To examine the relationships between replication time and both genetic and epigenetic features in more detail, we generated a database for computational analysis that incorporates our replication time data, the Arabidopsis TAIR 8 genome annotation [61], and epigenetic information for the Arabidopsis cell line [34]. We performed our analyses both on the level of individual probes and within the context of replicons.
To compare the genetic and epigenetic features of probes with different replication times, we partitioned the data into six smaller data sets based on the chr4 regions (Figure 2). This approach was necessary because heterochromatin replicates almost exclusively late, so any analysis that does not account for this fact merely compares heterochromatin to euchromatin. We then used a series of one-sample statistical tests to query whether probes with specific replication times were enriched or depleted for a specified genetic or epigenetic feature relative to the mean for that feature within a given region. This analysis is equivalent to comparing replication segments, but has the advantage of controlling for segment length by using probe numbers. Results for the proximal and distal portions of long arm are presented in Table 1. (The complete analysis is in Table S11.)
In animal systems, early replication positively correlates with gene and GC content when integrated over large domains [7], [12], [13], [62]. We found that the GC content of EM probes is depleted relative to the distal long arm, whereas the L probes are GC-enriched. EML probes have a GC content similar to EM probes, but M and ML probes are also GC-enriched. These results are likely linked to the gene coverage of these probes, with EM and EML probes showing depleted gene coverage and M, ML and L probes showing enriched gene coverage. The sequence content of the proximal long arm is different from the distal long arm, showing both a lower GC and gene content. However, the EM probes still show a lower GC content relative to the entire region. This depletion of gene and GC content in early-replicating regions contrasts with mammalian systems, and may reflect differences in genome structure.
In both animals and plants, H3K4me is almost exclusively genic and correlates with gene expression [27], [29], [30], [32], [34], [36], [37], with H3K4me3 having the strongest positive effect on gene expression in Arabidopsis [37]. H3K4me3 has been linked to early replication in mouse cells [13], and all forms of H3K4me correlate with early replication in human cells [12], [35]. We found that H3K4me1/2 is depleted in EM probes and enriched in ML and L probes in the distal long arm, consistent with the gene coverage. Despite its lower gene coverage, the proximal long arm has an abundance of H3K4me1/2 similar to that of the distal long arm, due in part to the gain of H3K4me1/2 by certain classes of TEs [34]. While we detected a depletion of H3K4me1/2 in EM probes, we did not detect a significant enrichment of H3K4me1/2 in L probes relative the proximal long arm as a whole.
DNA cytosine methylation (5mC) is found in the coding region of genes in the euchromatic regions of Arabidopsis, often in conjunction with H3K4me1 [33], [34], [37], [38], [63]. Like H3K4me1/2, 5mC is depleted in EM probes and enriched in ML and L probes in the distal long arm. The distribution of 5mC differs between the proximal long arm and the distal long arm. While 88% of 5mC is genic in the distal long arm, the percentage drops to 60% in the proximal long arm, and much of the 5mC in this region is associated with TEs and other repetitive sequences located in heterochromatin [33], [34]. We found a depletion of 5mC in EM, M and ML probes and an enrichment in L probes, which likely reflects the heterochromatic character of L probes in the proximal long arm.
To confirm this hypothesis, we examined the distribution of histone H3K9me2, which is associated with heterochromatin in Arabidopsis [31], [34], [64]. While H3K9me2 is not an abundant feature in the distal long arm, it is depleted in EM and M probes and enriched in L probes in this region, suggesting that some L probes are located in cryptic or facultative heterochromatin [28], [31], [65]. H3K9me2 is much more abundant in the proximal long arm, principally due to the elevated TE and repeat content of this region [33], [34], [58]. Again, H3K9me2 is depleted in EM, M and ML probes and enriched in L probes. The abundance of H3K9me2, 5mC and late replication in the proximal long arm suggests that much of this region should be considered cryptic or facultative heterochromatin.
Finally, we examined the correlation between H3K56ac and replication time. H3K56ac is associated with multiple biological processes that require an open chromatin conformation, including DNA replication, repair and transcription [40], [66]–[71]. H3K56ac is enriched in gene promoter regions in Arabidopsis suggesting a role in transcription [34]. In both the distal and proximal long arms, we detected enrichment of H3K56ac in EM probes and depletion in L probes. H3K56ac is also enriched in EL probes in both the proximal and distal long arms and in EML probes in the proximal long arm. The enrichment of H3K56ac in regions depleted for genes and the epigenetic marks associated with genes raises the possibility that some of the H3K56ac detected in our cells may be related to DNA replication rather than gene transcription.
To explore the relationship between genetic and epigenetic features and replication time in more detail, we performed further analyses in the context of the replicons identified above, again restricting our analysis to the long arm of chr4. We compared the overall content of genetic features and epigenetic modifications between EM and L replicons. We found that gene coverage/content, GC content and H3K56ac are higher in EM than in L replicons, whereas L replicons are enriched for TEs, H3K9me2 and DNA 5mC (Table 2). H3K4me1/2 is similar in EM and L replicons (Table 2). While these results are more consistent with animal systems, the results for gene coverage, GC content and H3K4me1/2 seem to conflict with the probe-level analysis presented above. However, two factors must be considered. First, 62 of the 66 EM replicons are located in the distal long arm while 31 of the 42 L replicons are located in the proximal long arm, and the distal long arm has a higher gene content and GC content than the proximal long arm (Table 1). Second, many of the EM and L replicons are comprised of DNA segments that replicate in various parts of S phase, e.g. the termination zones of EM replicons often replicate in late S phase (Figure 3 and Figure 4C). Thus, integration of genetic and epigenetic features over large regions such as replicons may obscure finer relationships.
To further resolve these relationships, we devised an analysis that examined the distribution of features within an “average” replicon. A similar strategy was used to examine the distribution of epigenetic modifications across genes [33], [34]. Each putative replicon in the proximal and distal long arms was divided into 10 intervals, each comprising 10% of its length. Unlike genes that have a definite polarity, most replicons are products of bidirectional fork progression and can be treated as symmetrical [1]–[5]. Hence, we combined our 10 intervals into 5 bins with the two innermost intervals near initiation zones comprising bin 1 and the two outermost intervals near termination zones comprising bin 5. We determined the occurrence of gene-rich, AT-rich, H3K4me1/2, H3K9me2, H3K56ac and 5mC probes within each bin across EM and L replicons separately (Figure 5).
We detected spatial correlations for both genetic and epigenetic features in EM replicons (Figure 5). Both AT-rich (top 25%) and H3K56ac probes are more abundant near initiation zones and depleted near termination zones (Figure 5). In contrast, the distribution of gene-rich (top 25%), H3K4me1/2 and 5mC probes show opposite trends (Figure 5). H3K9me2 is sparse in EM replicons, and there is no spatial correlation (Figure 5). These results suggest that DNA replication initiates in AT-rich intergenic regions with an open chromatin conformation and proceeds by elongation into gene-rich regions where the epigenetic features associated with the gene regulation specify a more complex chromatin structure. Most of the spatial correlations do not apply to L replicons, although there is a clear enrichment of H3K56ac near initiation zones (Figure 5). This analysis reconciles the probe-level (Table 1) and replicon analyses (Table 2), demonstrating that genetic and epigenetic features have both short and long range influences on replication time.
To determine if the increased H3K56ac near initiation zones is linked with gene expression, we looked more closely at the relationship between replication time, gene expression and epigenetic modifications. Previous analysis of these cells showed that H3K56ac is enriched at the 5′ end and promoters of genes, while H3K4me1/2 and 5mC are enriched in the body of genes [34]. To discern broad patterns of epigenetic modification and gene expression, we generated heat maps of the epigenetic data using a loess algorithm as we did for replication time. We determined gene expression in our cells using existing microarray data [34] and used two metrics to measure gene expression. The presence/absence of a transcript was determined using the Affymetrix Micro Array Suite 5.0 algorithm (MAS5) [72]. If the transcript was present, we considered the gene to be active. Gene expression levels were estimated using the gcRMA algorithm [73]. For the heat maps, we mapped the gcRMA expression values to the microarray probes prior to applying the loess algorithm. Representative late and early replicating regions of chr4 are shown in Figure 6. Elevated levels of H3K56ac are frequently associated with regions near replicon initiation zones whereas elevated levels of H3K4me1/2, H3K9me2 and 5mC are often near termination zones. Gene expression showed less clear-cut results sometimes colocalizing with H3K4me1/2 near termination zones and sometimes with H3K56ac near initiation zones.
We then examined the effect of epigenetic modifications on gene expression and replication time at the level of genes. The 2844 chr4 genes with available expression and epigenetic data were classified into 16 groups based on the pattern of all possible combinations of the four epigenetic modifications examined in our cells [34]. Replication time for each gene was derived from the overlapping probes. Using MAS5 presence/absence calls, we estimated that 61% of chr4 genes are active in our cells. Using this as a baseline, we ranked the 16 epigenetic patterns by increasing gene activity, with genes displaying pattern 1 having the highest probability of expression and genes with pattern 16 having the lowest (Table 3). Genes with pattern 1, which constitute the largest group, are positive for H3K4me1/2, H3K56ac and 5mC (Table 3). The presence of H3K4me1/2 and 5mC on expressed genes is consistent with previous studies showing that these marks can potentiate gene expression in Arabidopsis [37], [38], [63]. Strikingly, H3K56ac is the only epigenetic modification found in all patterns that show increased gene activity (Table 3). A positive correlation between gene expression and H3K56ac has been shown in other organisms [66], [70], [71], [74], and we show that this correlation exists in Arabidopsis. For the remaining patterns, H3K9me2 showed a clear association with reduced gene activity while genes lacking detectable H3K4me1/2, H3K9me2, H3K56ac or 5mC also showed low activity (Table 3).
Studies in other model organisms have shown a positive correlation between gene transcription and early replication [7]–[14]. When examined independent of epigenetic modifications, genes are significantly more likely to be expressed if they replicate EM rather than L (Table 4). Of chr4 genes, only genes with patterns 3 and 4 are more likely to replicate EM. Interestingly, genes with patterns 3 and 4 are distinguished from genes with patterns 1 and 2 by the lack of 5mC (Table 3). Despite their high frequency and levels of expression, genes with pattern 1 showed a slight tendency to replicate L and genes with pattern 2 showed no clear bias for either EM or L replication. Genes with patterns 7, 14 and 15 are more likely to replicate L than EM, and each of these patterns is characterized by the presence of H3K9me2 and 5mC (Table 3). In summary, the increased expression of EM-replicating genes is associated with enrichment of this population for genes displaying H3K56ac but lacking 5mC as well as with depletion of genes bearing the repressive combination of H3K9me2 and 5mC.
Allele-specific differences in replication time have been observed in animals [12], [60]. This can occur when one allele of a gene bears activating epigenetic modifications while the other allele bears repressive modifications, and could give rise to EL, EML or I replication time. Genes with patterns 6 through 9, 11 and 14 bear such modifications, and we did observe a slight enrichment of pattern 9 for EL genes and pattern 7 for EML genes (Table 3). However, the majority of the EL, EML and I segments cannot be explained by allele-specific replication timing. In many cases, genes that replicate EL, EML or I have only activating or repressive marks (Table 3). As stated above, many of these segments are associated with TEs and other repetitive elements.
The heat maps suggested that much of the H3K56ac on chr4 is associated with early replication and not gene expression (Figure 6). To examine this more closely, we determined whether the H3K56ac near the initiation zones of replicons in the long arm of chr4 was due to genes with epigenetic patterns 1 through 4 or reflected H3K56ac in intergenic sequences as well. Genes with pattern 3, positive only for H3K56ac, show a slight enrichment near initiation zones of EM replicons (Figure S4). An analysis of intergenic regions of chr4 revealed that the two most abundant epigenetic patterns are 3 (H3K56ac only) and 13 (no detected modifications) (Table S12). In the long arm of chr4, intergenic regions with pattern 3 are enriched near initiation zones and depleted near termination zones, but intergenic regions with pattern 13 are uniformly distributed across replicons (Figure S5). To determine if this enrichment for intergenic H3K56ac near initiation zones is associated with the promoters of expressed genes, we analyzed the distribution of expressed genes (regardless of epigenetic modifications) across replicons. This analysis showed that expressed genes are uniformly distributed (Figure S4), allowing us to conclude that much of the intergenic H3K56ac is associated with early replication and not gene expression.
DNA replication has been profiled in Drosophila, mouse and human genomes [7]–[14], [22], [23], [60], [75]. Arabidopsis and Drosophila have a similar genome size (∼120 Mb each) and gene density (250 and 111 genes/Mb respectively) [54], [57], so it is not surprising that their replication profiles are similar. In contrast, the human and mouse genomes are substantially larger (3300 and 2500 Mb respectively) and have a much lower gene density (10 genes/Mb each) [55], [56]. Mammalian genomes are also characterized by large regions of uniform GC and gene content known as isochores [62], [76], [77]. In both human and mouse cells, replication time has been shown to correlate with isochore structure, and high GC, gene-rich isochores tend to replicate early in S phase [13], [62]. In contrast, it is not clear that a functionally equivalent isochore structure exists in Arabidopsis or Drosophila [77], [78]. Such differences in genome structure may explain why gene content and expression and the associated epigenetic modifications have a more subtle influence on replication time in Arabidopsis than in mammals. For example, in human cells, distance to the closest expressed gene is strongly correlated with replication time [23]. However, these distances are on the order of megabases. Such a correlation is meaningless in Arabidopsis where the median intergenic distance is less than one kilobase [54], [61]. Accordingly, we tailored our analysis to suit this compact genome, revealing many similarites and a few differences in the DNA replication programs of these model systems.
A common approach to determine DNA replication timing utilizes the direct hybridization of BrdU labeled early and late S phase DNA to genomic tiling arrays to construct a replication profile that indicates the enrichment of a given sequence in early relative to late S phase [7]–[9], [13], [14]. In this approach, DNA replication in mid S phase is inferred rather than directly evaluated. We measured Arabidopsis DNA replication in early, mid, and late S phase cells in separate microarray experiments producing three independent replication profiles. This strategy revealed that the replication profiles for early and mid S phase cells are very similar to each other and clearly distinct from the late S phase profile.
The majority of euchromatin in chr4 replicates in early and mid S phase, and the bulk of the heterochromatin replicates in late S phase (Figure 2). Temporal separation of DNA replication for euchromatin and heterochromatin was first observed at least five decades ago in both plants and animals [25] and is consistent with recent findings in Drosophila, mouse and human cells [21]. Fiber autoradiography experiments in Arabidopsis identified two temporal classes of replicons but did not distinguish euchromatin from heterochromatin [48].
Surprisingly, there is little difference between the early and mid S phase replication profiles (Figure 2), even though FACS profiles for the early and mid S phase cells are distinct (Figure 1 and Figure S2). When interpreting these results, it is important to remember that while FACS takes a DNA content measurement for each cell, the replication profiles are derived from a population of cells. If DNA replication followed a totally random program, a population of early S phase cells could produce a replication profile that encompasses the entire genome. At the other extreme is a strict temporal program in which the order of origin activation is highly consistent between cells in a population. With our experimental design, such a program would produce early S phase profiles showing replication of approximately 20% of the genome, while mid S phase profiles would be distinct from the early S phase profiles because they would encompass an additional 30% of the genome.
Our results are an intermediate case between these two extremes and are best explained by a biphasic model of replication for Arabidopsis. In this model, the bulk of euchromatin replicates in early to mid S phase and the heterochromatin replicates late. Origin utilization is largely the same in early and mid S phase, suggesting that the temporal order of origin activation in the first half of S phase is stochastic. While we did not attempt to identify origins per se, we did identify initiation zones, and we detected few, if any, initiation zones specific to mid S phase cells (Figure 3C and Figure 4B). The segmentation analysis showed some merging of early S phase segments to form larger mid S phase segments, but this effect most likely reflects elongation of replicons rather than activation of additional origins (Figure 3). The relative enrichment for initiation zones is similar in early and mid S phase cells suggesting that there is no quantitative difference in origin activation (Figure 3C). In contrast, there are many putative initiation zones specific for late S phase (Figure 3C and Figure 4B).
The idea that DNA replication follows a strict temporal program derives largely from seminal work in budding yeast [6], [79]. Budding yeast is characterized by sequence specific origins in a compact genome and, as such, might not be a good model for eukaryotes with much larger genomes and no clear origin sequence specificity [1]–[4]. Single molecule studies showed that even in budding and fission yeast, origin activation is stochastic and varies from cell to cell in a population [15], [16]. Whole genome studies in Drosophila and mouse cells are also consistent with a biphasic model of DNA replication. In Drosophila, initiation zones are most abundant in early and late S phase [14], while mouse replicons and replication domains tend to segment as either early or late [13]. Increasingly, origin activation is being interpreted as a largely stochastic process at the level of individual cells, with temporal profiles corresponding to the most probable sequence of origin activation for a population of cells [17], [18], [20].
The replication time of any given DNA segment is related to three factors – distance from the closest origin, activation time of that origin, and rate of DNA elongation upstream of the segment. Chromatin conformation can influence the latter two factors, and chromatin remodeling factors have been shown to be critical for DNA replication [80]–[83]. Our analyses of replication time with respect to both genetic and epigenetic features revealed correlations that may reflect the effect of chromatin conformation on origin specification, origin activity and the rate of DNA elongation.
The heterochromatic knob and pericentromeric heterochromatin are entirely late replicating (Figure 4A). Both of these regions are depleted in genes, rich in TEs, and display abundant H3K9me2 and 5mC (Figure 2 and Table S11). This constitutive heterochromatin exists in a compact conformation throughout most of the cell cycle [59]. This conformation likely restricts both origin activation and DNA elongation [2]–[4]. In both budding and fission yeast, pericentromeric heterochromatin replicates in early S phase [6], [26], but pericentromeric DNA replicates in late S phase in animal cells [9], [84], [85]. In both cases, replication of heterochromatin is dependent on chromatin remodeling complexes [80], [82], [83], and it will be interesting to identify the complexes utilized by plants.
We focused our analyses on the long arm of chr4 because it represents a large contiguous, genomic segment generally regarded as euchromatic [33], [34], [58], [59]. However, we were surprised by the predominance of late replication in the proximal portion of the long arm (Figure 2 and Figure 4A). Probe and replicon level analyses revealed that relative to the distal long arm, the proximal long arm has considerable heterochromatic character, including decreased gene coverage/content, increased TE coverage/content, and elevated levels of both H3K9me2 and DNA 5mC (Figure 2, Table 1, and Table 2). Much of the proximal long arm likely adopts a chromatin state known as cryptic or facultative heterochromatin [28], [65]. Such regions share some of the biochemical features of constitutive heterochromatin, including hypoacetylation, H3K9me2 and DNA 5mC, but do not adopt the long range, highly condensed structure of constitutive heterochromatin. In mouse cells, replication domains that switch replication time upon differentiation are believed to be facultative heterochromatin [13], [21].
Despite the overall differences in replication time for the proximal long and distal long arm regions, we detected several correlations between replication time and genetic and epigenetic features that were similar in both regions. For example, EM-replicating probes show increased AT content, decreased gene coverage and decreased DNA 5mC (Table 1). Further, the histone modifications, H3K4me1/2 and H3K9me2, are decreased while H3K56ac is increased. The pattern is opposite for L-replicating probes. These observations suggest that DNA replication initiates in AT-rich intergenic regions with an open chromatin conformation and proceeds into regions where the epigenetic modifications associated with gene expression specify a more complex chromatin conformation. The distribution of genetic and epigenetic features within replicons further supports this hypothesis (Figure 5).
The EM replicons display trends that are consistent with a replicon model that has been termed the “relaxed replicon” model [2]–[4]. This model incorporates several mechanisms to explain ORC binding and replicon structure in higher eukaryotes. Mechanisms consistent with our work include a higher affinity of ORC for open chromatin and AT rich sequences [86], [87], transcriptional interference preventing ORC binding [88], and inhibition of ORC binding by DNA methylation [89]. The structure of EM replicons may be driven by the probability of both ORC binding and origin activation. Regions proximal to initiation zones have a higher AT content and elevated H3K56ac and may have a higher probability of binding ORC to form an origin (Figure 5). The lower gene density, lower H3K4me1/2 and reduced 5mC in these regions would also favor origin formation. Termination zones show opposite trends for these characteristics, consistent with a lower probability of binding ORC. In addition, elevated levels of H3K4me1/2 and 5mC may impede the progress of replication forks in these regions. Chromatin modified by DNA 5mC adopts a more compact conformation and impedes the progress of RNA polymerase [63], [90].
The trends for EM replicons are readily apparent when the epigenetic modifications are integrated over large regions (Figure 6). Most of the trends do not hold for L replicons, which in comparison to EM replicons, have greatly elevated and evenly distributed levels of H3K9me2 and 5mC indicative of a heterochromatic state (Table 2, Figure 5, and Figure 6). Replication may be delayed in these regions because it requires the activity of chromatin remodeling complexes, as discussed above for the heterochromatic knob and pericentromere. Additionally, L replicons may have a lower density of potential origins.
Gene expression shows a positive correlation with early replication in all higher eukaryotes examined to date [7]–[9], [11]–[14], [22], [23]. This correlation is strongest when integrated over large regions because there are many exceptions at the level of individual genes. We identified a similar correlation in Arabidopsis, with genes in EM replicating regions more likely to be expressed than genes in L replicating regions (Table 4). However, the relationship of specific epigenetic modifications to gene expression and replication time is complex (Table 3). From the standpoint of replication time, two effects are prominent. H3K56ac with a lack of H3K9me2 is favorable for both gene expression and early replication, whereas H3K9me2 with a lack of H3K56ac correlates with lower expression and late replication. Genes associated with both H3K9me2 and H3K56ac also tend toward low expression and late replication, but the effect is less clear-cut than H3K9me2 alone. Genes with both marks are similar to the “pan S” or “biphasic” genes in human cells which bear both active and repressive chromatin marks due to interallelic variation [12], [60]. We also observed an increase in EML replication for these genes in Arabidopsis (Table 3). Unlike the epigenetic modifications discussed above, integration of gene expression over large regions did not reveal a correlation between gene expression and replicon structure (Figure 6). This lack of correlation probably reflects the fact that the expression of an individual gene is more strongly modulated by epigenetic modifications specific to that gene rather than by the global characteristics of large regions containing many genes.
H3K56ac is thought to occur on all newly synthesized H3 histones and be required for nucleosome assembly [66], [69], [91]. H3K56ac is associated with regions of nucleosome exchange such as active promoters [70], [71], sites of DNA repair [67], [69], and nascent chromatin [40], [68]. In budding yeast, H3K56ac is most abundant during S phase and localizes to early origins in a cell cycle dependent manner [40], [68], [69]. Intriguingly, H3K56ac correlated with EM replication and was enriched at the center of Arabidopsis replicons (Figure 5 and Figure 6). Interpretation of this data must be tempered by the fact that the epigenetic profiling was performed on an unsorted population of cells so both replication dependent and independent H3K56ac is represented. Although there was a positive correlation between H3K56ac and gene expression (Table 3), integration of H3K56ac over large regions, including intergenic regions, showed a clear association with replication time and not with gene expression (Figure 6). H4K16ac correlates with early replication in Drosophila [14], while H3K56ac associates with early origins in budding yeast [40]. We have provided the first evidence, to our knowledge, linking H3K56ac to replication time in a higher eukaryote. Unlike H4K16, H3K56 is located in the core of the histone and is inaccessible to acetylation in the context of a fully assembled nucleosome [66], [92]. Therefore, H3K56ac might be associated with nascent DNA behind active replication forks rather than the disassembly of chromatin ahead of replication forks [68]. Nevertheless, H3K56ac may prove to be a valuable epigenetic mark for identifying replication origins.
We have presented a high-resolution analysis of the replication program for a plant chromosome. Arabidopsis DNA replication is biphasic, with euchromatin replicating in the first half of S phase and heterochromatin replicating in the last half. This pattern is similar to other eukaryotes [9], [84], [85], although exceptions do occur in yeast [6], [26]. Within each phase, origin activation appears to be largely stochastic because we could discern few differences between replication profiles for early and mid S phase cells. This result provides additional support for the emerging model of stochastic origin activation rather than strict temporal regulation [15]–[18], [20]. The replication profiles allowed us to construct a replicon map for chr4 and to correlate replication time with gene expression and specific epigenetic modifications. We showed that initiation zones are enriched for epigenetic features associated with open chromatin, providing support for the “relaxed replicon” model, which proposes that origin specification and activity are strongly influenced by both sequence content and chromatin conformation in higher eukaryotes [1]–[4]. Finally, we showed that early replicating regions and initiation zones are enriched for H3K56ac. This histone modification continues to be an area of intense research because of its role in DNA replication, DNA repair and gene expression. We provide evidence that H3K56ac has both replication independent and dependent roles in plants by showing that genes bearing H3K56ac have a higher probability of expression, whereas large regions with elevated H3K56ac levels are associated with early replication. Replication time and H3K56ac data in conjunction with other experiments may help us identify replication origins in plants. This study linking DNA replication and replicon structure to chromatin conformation provides a foundation for future studies that will investigate the impact of these processes on plant growth and development.
The Arabidopsis cell line (Col-0, ecotype Columbia) was maintained in Gamborg's B5 basal medium with minor salt (Sigma G5893) supplemented with 1.1 mg/L 2,4-dichlorophenoxyacetic acid, 3 mM MES and 3% sucrose. The cells were grown on a rotary shaker at 160 rpm under constant light at 23°C and subcultured every 7 days with a 1∶10 (inoculum∶fresh medium) dilution [34].
BrdU labeling was maximized using a ‘7-d split culture' by mixing 25 mL of fresh medium and 25 mL of the Arabidopsis culture at 7 days post subculture. The 7-d split culture was grown for 16 h and then labeled for 1 h with 100 µM BrdU (Sigma B9285). Labeled cells were fixed in 1% paraformaldehyde for 15 min, washed in 1× phosphate buffered saline (PBS) three times, and snap frozen in liquid nitrogen. Time course experiments showed that BrdU incorporation is highest between 12 and 16 h post-labeling (Figure S1C). Cells from six 7-d split cultures were combined for each biological replicate.
The frozen cell pellet for each biological replicate was ground at 4°C in 150 mL lysis buffer (15 mM Tris-HCl pH 7.5, 2 mM EDTA, 80 mM KCl, 20 mM NaCl, 15 mM β-mercaptoethanol, and 0.1% Triton X-100) using a commercial blender. The ground cell suspension was incubated at 4°C for 5 min and filtered through a 3-tiered nylon mesh (100, 50, and 30 µm). The filtrate was centrifuged at 200 ×g for 5 min at 4°C, and the nuclei were resuspended in 8 mL of lysis buffer containing 2 µg/mL DAPI and 50 µg/mL RNase A. The isolated nuclei were filtered through a 20-µm nylon filter before flow cytometric analysis and sorting.
Nuclei were sorted and recovered using an InFlux cell sorter (BD Biosciences) equipped with a 355-nm UV laser and a 488-nm sapphire laser. STE buffer (10 mM Tris-HCl pH 7.5, 1 mM EDTA, and 100 mM NaCl) was used as a sheath fluid, and nuclei were sorted into a 50-mL tube containing 5 mL STE buffer.
An analytical FACS profile for BrdU incorporation and DNA content was generated as described [93] with some modifications. BrdU-labeled cells were fixed in 70% ethanol on ice for 1 h and frozen in liquid nitrogen. Nuclei were isolated, denatured in 2N HCl and 0.5% Triton X-100 at room temperature for 30 min, neutralized by adding 0.1 M Na2B4O7 (pH 8.5), and washed twice with PBS-TBR (1x PBS, 1% BSA, 0.5% Tween-20 and 50 µg/mL RNase A). The nuclei were resuspended in PBS-TBR containing a 1∶50 dilution of anti-BrdU Alexa Fluor 488 conjugate (Invitrogen) by gentle agitation overnight at 4°C in the dark. The nuclei were washed once with PBS-TBR, incubated in PBS-TBR containing 10 µg/mL propidium iodide for at least 30 min, filtered through a 20-µm nylon filter, and analyzed by FACS. FlowJo (Version 8.8.6) software was used for the data analysis.
To reverse the crosslinks, the sorted nuclei were treated with 50 mM EDTA, 1% sodium lauroyl sarcosine and 200 µg/mL proteinase K for 1 h at 42°C and then overnight at 65°C in the dark. The mixture was supplemented with 4 mg/mL phenylmethanesulphonylfluoride and incubated for 40 min at room temperature prior to extraction of genomic DNA using phenol/chloroform/IAA in a phase lock gel (Sigma). The upper aqueous phase was mixed with 150 µg/mL GlycoBlue (Ambion) and precipitated with 0.3 M sodium acetate and 2 volumes of cold ethanol. The DNA was centrifuged and the pellet was washed with 70% ethanol once, dried for 5 min using a SpeedVac in the dark, and resuspended in sterile water.
BrdU-labeled DNA was immunoprecitated as described [94] with some minor modifications. Genomic DNA extracted from the sorted nuclei was sonicated in 450 µL of ChIP dilution buffer (0.1% BSA, 1.2 mM EDTA, 16.7 mM Tris-HCl pH 8, and 167 mM NaCl) to a shear-size of 500 to 1000 bp, followed by addition of Triton X-100 (1.1%). The sheared DNA was denatured at 95°C for 5 min and immediately cooled on ice for at least 5 min. One mL of cold ChIP dilution buffer containing 1.1% Triton X-100 was added and the sheared DNA was incubated with 0.5 µL anti-BrdU antibody (Invitrogen) for 3 h at 4°C. DNA containing BrdU was immunoprecipitated by adding 100 µL of 50% protein G-sepharose beads (Sigma) and incubating overnight in the dark at 4°C with gentle agitation. The beads were washed as previously described by Gendrel, et al. (2005). BrdU-labeled DNA was eluted from the beads with 0.2 M glycine (pH 2.5) and neutralized by adding 10% (v/v) of 1 M Tris-HCl (pH 8). Eluted DNA was treated with proteinase K for 1 h at 45°C, extracted with phenol/chloroform/IAA, and precipitated with sodium acetate and ethanol. Precipitated DNA was resuspended in RT-PCR grade water (Ambion) and used as template for random amplification and real-time quantitative PCR.
BrdU immunoprecipitated DNA (target DNA) and input DNA (reference DNA) samples were amplified as described [95], purified and concentrated to 200–250 ng/µL using a QIAquick PCR Purification Kit (QIAGEN). Each amplified DNA sample (1.5 mg) was labeled with either Cy3 or Cy5 fluorescent dye and purified using a BioPrime Array CGH Genomic Labeling System (Invitrogen). The Cy dye-labeled target and reference samples were co-hybridized on a custom-printed tiling array [33] with a dye-swap experimental design. Each experiment comprised six microarrays representing the three biological replicates and the corresponding dye swaps. Microarray hybridization and washing were previously described [34] but modified to include DyeSaver2 coating reagent (Genisphere) to minimize oxidation of Cy5. Hybridized microarrays were scanned using a PerkinElmer ScanArray Express scanner and quantified using GenePix Pro software (version 6.01).
Calculation of microarray probe enrichment ratios, loess and quantile normalizations were done in the R statistical computing environment with the limma package using default settings [96]–[98]. Probe ratios were loess-smoothed in a 150-kb window for replication profiles and identification of initiation and termination zones. Segments of contiguous replication time were defined as regions where smoothed probe ratios were greater than zero for a minimum of 10-kb. This filter minimized excessive replication time changes in regions with low probe enrichment ratios. Merging of the segmentations for early, mid and late S phase cells was done by determining the regions of overlap. The 10-kb length minimum was not used at this step. Initiation and termination zones were identified as the inflection points of the loess-smoothed profiles as described in the results. Zones were then defined as the 10-kb regions centered at the inflection point. Overlapping zones were merged into a single zone. Replication boundaries were chosen from the three sets of termination zones based on the following order of precedence: 1) termination zones present in early, mid and late S phase cells, 2) termination zones enriched in late S phase and 3) termination zones that manifest as local minima but enriched in early and/or mid S phase.
All data manipulation and statistical analysis was performed with R and Bioconductor [96], [99]. A database incorporating probe ratios for replication time, histone modifications, DNA 5mC and the TAIR8 Arabidopsis genome annotation [61] was constructed to facilitate analysis. Gene and TE coverage values for probes and larger regions are the percentage of bases in that region that overlap with any gene or TE respectively. Overlapping genes or TEs were treated as one so that coverage values do not exceed 100%. Statistical comparisons of GC content and gene or TE coverage were performed by one-sample t-tests. AT-rich and gene-rich probes were defined as the top quartile of all probes on the array. AT-rich, gene-rich and probes positive for histone modifications or DNA methylation data were treated as binomial data, and a one-sample binomial test was used for analyses. Gene expression values were determined using the affy package in R [73]. MAS5 presence or absence calls and gcRMA expression values were calculated using default settings. The pattern of epigenetic modifications for chr4 genes was determined from the modifications of the overlapping probes again treating the modifications as binomial data. Heat maps for epigenetic modifications were generated by smoothing probe ratios in a 150-kb window as for replication profiles and ranking the data by deciles for the whole of chr4. Heat maps for gene expression were generated similarly but gcRMA expression values were used rather than probe ratios. R scripts for all analyses and figures are available upon request.
|
10.1371/journal.pntd.0006947 | Geostatistical mapping of the seasonal spread of under-reported dengue cases in Bangladesh | Geographical mapping of dengue in resource-limited settings is crucial for targeting control interventions but is challenging due to the problem of zero-inflation because many cases are not reported. We developed a negative binomial generalised linear mixed effect model accounting for zero-inflation, spatial, and temporal random effects to investigate the spatial variation in monthly dengue cases in Bangladesh. The model was fitted to the district-level (64 districts) monthly reported dengue cases aggregated over the period 2000 to 2009 and Bayesian inference was performed using the integrated nested Laplace approximation. We found that mean monthly temperature and its interaction with mean monthly diurnal temperature range, lagged by two months were significantly associated with dengue incidence. Mean monthly rainfall at two months lag was positively associated with dengue incidence. Densely populated districts and districts bordering India or Myanmar had higher incidence than others. The model estimated that 92% of the annual dengue cases occurred between August and September. Cases were identified across the country with 94% in the capital Dhaka (located almost in the middle of the country). Less than half of the affected districts reported cases as observed from the surveillance data. The proportion reported varied by month with a higher proportion reported in high-incidence districts, but dropped towards the end of high transmission season.
| A better understanding of spatial and temporal variation in dengue risk is invaluable since it guides intervention strategies and facilitates effective health resource allocation. Transmission of dengue depends on the distribution and abundance of the mosquito vectors which are sensitive to climatic and environmental factors including temperature, rainfall, and population density. By modelling dengue-climate relationships, the burden of dengue can be estimated in locations where data on transmission are not available permitting identification of high-risk areas. This reveals the extent of under-reporting in dengue surveillance data in resource-limited countries which is essential to estimating the changing burden of dengue.
| Dengue is a neglected tropical disease caused by the dengue virus (DENV) and is transmitted by female Aedes mosquitoes, predominantly Aedes aegypti and Aedes albopictus. The severe forms of the disease are potentially fatal. The World Health Organization (WHO) estimates that about 52% of the people at risk of dengue worldwide live in 10 countries of the WHO South-East Asia Region [1]. Bangladesh, located in South Asia and surrounded by India and Myanmar (Fig 1) where dengue is endemic, experienced its first epidemic of dengue in 2000 [2]. Since then cases have been reported every year, most commonly among adults and older children living in metropolitan cities [3]. Of the 64 Bangladeshi districts (Bangladesh’s second largest administrative unit), 29 reported dengue cases between 2000 and 2009, with Dhaka consistently reporting the largest number of cases [4]. Heterogeneity in the distribution of hospitals across the country and differentials in treatment-seeking behaviour based on location are likely to cause under-reporting [3]. However, despite no effective control program being introduced and no changes in surveillance until 2010 when serological confirmation was mandated for case reporting, nationally reported cases have declined since 2002 [3]. This might be partially attributable to increased prevalence of immunity and reduction in mosquito breeding sites resulting from public awareness [3]. However, there is considerable under-reporting inherent in the passive hospital-based surveillance system [3]. A study investigating the global distribution of dengue burden estimated an average of 4,097,833 symptomatic infections (95% Bayesian credible interval: 2,952,879–5,608,456) occurred in Bangladesh in 2010 [5] but only 409 were reported to authorities [4].
Climatic factors (mainly temperature and rainfall) influence the survival and development rate of vector and virus. The aquatic larval and pupal stages of the Aedes mosquitoes require fresh water. Outdoor artificial containers filled with rain water serve as breeding sites [6]. Average temperature, as well as diurnal temperature range (DTR, the difference between daily maximum and minimum temperature), influence mosquito development, the mosquito biting rate, the extrinsic incubation period, and vector-virus-host interaction [7–10].
Climatic similarity and the movement of viraemic individuals in geographically neighbouring areas introduce spatial correlation in dengue incidence [11,12] and might lead to spurious model-based incidence estimates if ignored. Bayesian geostatistical modelling approaches are powerful for disease mapping, explicitly accounting for spatial correlation in disease data while incorporating uncertainty in data and model parameters [13,14]. Within a Bayesian paradigm, inference about model parameters is based on the posterior distribution derived from the combination of data and pre-existing knowledge of parameter values, and therefore does not require a large sample size assumption as does the frequentist approach. Therefore, more robust estimates can be obtained when the disease is rare [15] or data on case numbers are limited.
Regional variation in dengue must be studied to allocate resources proportionate to burden but this has not been done in Bangladesh. The only spatial mapping study of dengue in Bangladesh identified Dhaka as the most likely cluster for dengue transmission during 2000–2009 with a small number of secondary clusters in the southern part of the country in 2000 [16]. However, under-reporting was not accounted for, nor were potential causes of geographical variation in dengue transmission (such as climate and socio-demographic factors) considered. Our aim was to produce maps of the monthly spatial variation in dengue incidence at the district-level in Bangladesh in relation to climatic and demographic factors with adjustment for under-reporting. The resulting maps can facilitate the efficient distribution of vector control interventions to areas of highest need at the appropriate time. The method is useful for identifying locations where DENV transmission occurs but incidence data are lacking.
The study was approved by The Australian National University Human Research Ethics Committee. National dengue surveillance data were anonymised.
Bangladesh has a hot, humid, tropical climate with monsoons occurring during June to September. Monsoon rainfall, about four-fifths of the mean annual rainfall, ranges from 1,527mm in the west to 4,197mm in the east, and the mean monthly monsoon temperature of 29°C averaged across the country is generally suitable for dengue transmission.
High population density (964 people per square kilometre in 2011) and unplanned urbanisation leading to over-crowding in divisional cities with inadequate water supply, and inefficient drainage and waste disposal increase risk for dengue since the vector mosquitoes breed in water storage containers in and around houses [6,17,18].
Dengue notification data, consisting of suspected, probable, and confirmed cases reported to the Directorate General of Health Services between January 2000 and December 2009 were analysed. Under-ascertainment of symptomatic dengue cases is highly likely since the passive surveillance system only reports cases admitted to hospital [3]. Daily rainfall (mm), and minimum and maximum temperatures (°C) measured from 35 stations were sourced from the Bangladesh Meteorological Department with around 2, 3, and 3% missing data, respectively. Diurnal temperature range (DTR) was calculated from the daily maximum and minimum temperatures. Weather values for days with missing data were filled by averaging data from adjacent days. Monthly averages were then calculated from daily records. Missing values for months were supplemented with the average of non-missing values from three neighbouring stations. Bayesian kriging [19] was used to interpolate weather values in districts without a station. Mean monthly rainfall, mean monthly temperature, mean monthly DTR, and monthly total dengue cases over the 10 years were then aggregated by month and used in the model development discussed below.
The population density (people/km2) for each district was estimated by dividing the district population by the district area (km2). The monthly population of each district were calculated by linear interpolation between population estimates from the 1991, 2001, and 2011 census data of the Bangladesh Bureau of Statistics [20].
More than half of Bangladeshi districts did not report any dengue cases during the study period 2000 to 2009. It was unclear whether these missing values corresponded to true zeros or a lack of reporting, so we used a zero-inflated model which allows for two different interpretations for the occurrence of zero cases; either no case occurred, or cases occurred but were not reported.
To model the spatio-temporal pattern of dengue, district-wise dengue notification data aggregated by month over the period 2000 to 2009 were modelled via a negative binomial generalised linear mixed effect model with a logarithmic link function and the population of the districts as an offset. Let yit be the number of reported dengue cases for the ith (i = 1, 2,…, 64) district in the tth (t = 1, 2,…, 12) month. Preliminary non-spatial analysis indicated that the following factors should be included in the analysis: mean monthly temperature, mean monthly DTR, interaction between mean monthly temperature and DTR, and mean monthly rainfall at lag one and two months, and population density [21]. The variable “Border” indicating whether a district bordered India or Myanmar was included in the model. This variable was used as a proxy for movement across borders with neighbouring dengue endemic countries. An indicator for outbreak months with case numbers exceeding the 10-year mean plus two standard deviations was added. Population age structure which is approximately the same across the districts was not considered [22]. The generalised linear mixed effect model is specified by:
log(μit)=log(Populationit)+α+β1Ti,t−1+β2Ti,t−2+β3DTRi,t−1+β4DTRi,t−2+β5(Ti,t−1*DTRi,t−1)+β6(Ti,t−2*DTRi,t−2)+β7Ri,t−1+β8Ri,t−2+β9*Borderi+β10*Outbreakit+β11*Popdenit+∅i+AR(1)t;i=1,2,…,64andt=1,2,…,12
where μit denotes the mean dengue counts for the ith district in the tth month. T, DTR, (T * DTR), and R, are the district wise mean monthly values of temperature, diurnal temperature range, interaction between temperature and diurnal temperature range, and rainfall. AR(1)t is a first order autoregressive month effect that captures the correlation in dengue cases between consecutive months. ∅i represents the spatially structured random effects at district-level that take into account the spatial dependency in dengue cases by assuming a Matérn covariance using the stochastic partial differential equations (SPDE) approach [23]. All continuous covariates were standardised (zero mean and variance one) to ensure that the influence of each covariate parameter was comparable. The model fitting was done using the R-package Integrated Nested Laplace Approximation (INLA) [24].
Following a Bayesian model specification, prior distributions were assigned to model parameters. Independent diffuse Gaussian priors (with mean 0, precision 1 × 10−3) were chosen for the intercept (α) and regression coefficients (β) to allow the data to predominate in calculating the posterior distributions. The precision parameter for the random effect was assigned a logGamma (1, 0.00005) prior.
The deviance information criteria (DIC) was used to check the goodness of fit of three zero-inflated negative binomial models (ZINB) developed sequentially as a temporal model, a spatio-temporal random effects model, and a spatio-temporal model without any fixed effect covariates. A low DIC value is indicative of the best trade-off between model fit and complexity of the model.
Table 1 displays parameter estimates from the three ZINB models. The spatio-temporal full model fitted the data best as evidenced by the lowest DIC (1079.44 vs 1099.54 and 1132.55). Of all the climatic variables, mean monthly temperature, its interaction with DTR, and mean monthly rainfall at lag two months showed significant associations with dengue incidence. Average temperature had a positive relationship with dengue (β2 = 3.81; 95% highest posterior density [HPD] credible interval [CrI]: 1.63, 6.04), while it’s negative interaction (β6 = -1.73; 95% HPD CrI: -3.13, -0.36) with DTR indicates that increasing temperature and decreasing DTR is associated with increased dengue incidence. Mean monthly rainfall at two months lag showed a positive relationship (β8 = 1.22; 95% HPD CrI: 0.19, 2.26) with dengue. Population density was significantly (β11 = 0.69; 95% HPD CrI: 0.38, 1.03) associated with dengue. Districts not adjoining the borders with India or Myanmar were found to have significantly lower incidence compared with adjacent districts. The posterior density plots of fixed effect covariates are shown in Fig 2.
A zero-inflation parameter with mean 0.08 (95% HPD CrI: 0.06, 0.10) confirms significant zero-inflation in reported dengue data. The Moran’s index (0.12; p-value<0.0001) provided significant evidence against the null hypothesis of zero spatial autocorrelation in dengue cases in Bangladesh. Significance of the spatial effects were evident from the estimates of tau and kappa.
Fig 3 shows the spatial variation in monthly aggregated dengue cases reported over 2000–2009, compared with the corresponding fitted values obtained from the spatio-temporal full model. The model clearly captures the spatial variation between months with an increasing trend in case numbers from June with the highest in August. Southern Bangladesh observes higher transmission than the northern part of the country. The model identified several districts (mostly in the northern part of the country) with modelled transmission but without any reported cases.
Dengue transmission fluctuated over the months with 92% of estimated annual cases occurring in August and September. During this period, almost all the districts across the country were estimated to have cases, although 94% of national total were in Dhaka. National surveillance data reported cases from less than half of the affected districts estimated by the model (Fig 4).
Fig 5 shows the spatial variation in the percentage of estimated cases reported by month (July-September). In general, districts with higher case numbers had a higher percentage of estimated cases reported compared to the districts with fewer case numbers during the high transmission months of August and September.
We present the first model-based estimates of the monthly geographical distribution of dengue cases in Bangladesh. In national surveillance data more than half the districts reported zero cases but we believe disease occurred but was not reported. To address this zero-inflation, a ZINB model was developed. The spatio-temporal full model identified that increasing mean monthly temperature and decreasing DTR at lag two months were significantly associated with high incidence of dengue. The inverse relationship between temperature and DTR in our study corroborates previous observations [10,25]. Our analysis also suggested a positive association between dengue and mean monthly rainfall at two months lag. Several studies reported that increasing mean monthly temperature and rainfall significantly increase dengue risk at different time lags [26–28], reflecting the delays in the impact of weather on mosquito populations and subsequent changes in transmission patterns. Population density was found to be significantly associated with dengue risk. High transmission of dengue in densely populated areas has been reported in the literature [29,30]. Incidence was higher in districts bordering India or Myanmar, perhaps due to cross-border migration from endemic countries to Bangladesh [31].
The model identified highest transmission during August and September (92% of estimated annual cases). Transmission is spatially heterogeneous, with a higher number of cases estimated in the south compared to the northern districts, and the highest in Dhaka. An earlier study investigated space-time clusters of dengue transmission in Bangladesh from 2000 to 2009 and reported Dhaka as the most likely cluster throughout the study period, with a small number of secondary clusters in the south of the country in 2000 [16]. June-November was reported as the high transmission season when all the clusters were identified [16].
Our model also identified districts with transmission that went unreported; during the high transmission month of August, cases were reported from less than 50% of the affected districts. We estimated that the reporting percentage (the percentage of all cases reported to the surveillance system) also varies by month. During the high transmission months, reporting is generally higher in districts with high compared to those with low transmission. However, towards the end of the high transmission season, proportion reported declined in majority of districts presumably due to the lower case numbers.
A study measuring global burden of dengue estimated 4,097,833 symptomatic infections (95% Bayesian credible interval: 2,952,879–5,608,456) in Bangladesh in 2010 [5] which differs considerably from our annual estimate presumably due to the differences in data and modelling strategy. A two-stage analytical approach was taken by Bhatt et al [5] where a boosted regression tree approach was adopted to estimate the relationship between the probability of occurrence of a dengue infection and the environmental conditions sampled at each study site. Covariates included vegetation index, indicators of urbanisation and relative poverty, and an urban accessibility metric in addition to temperature and precipitation [5]. Annual estimates of infections were obtained from a hierarchical Bayesian model estimating the relationship between longitudinal incidence data from 54 cohort studies and the previously generated probability of occurrence of dengue infection [5]. The estimated apparent dengue infections referred to any infection encompassing any disruption to the daily routine of the infected individual [5], whereas our estimates are calculated based on hospital admitted patients and therefore not representative of all symptomatic infections. Sparse data points and lack of cohort studies across a range of transmission intensities in the Indian subcontinent could be responsible for the large uncertainties associated with the Bhatt et al’s estimates in this region [5].
The strength of our model is the ability to generate estimates of dengue in areas with suspected under-detection. Relatively low case numbers indicate the potential of active transmission of the disease in a district which is crucial in the absence of information regarding the mosquito vector population. This is also indicative of increased transmission risk across the neighbouring districts resulting from inter-district movements of viraemic individuals. The assumption that transmission is governed by climatic suitability is reasonable, especially in the absence of effective public health strategies including mosquito control program. Our model was incapable of capturing the inter-annual variation in dengue and climate variables due to the aggregation of monthly counts over the study period which was required to model the monthly spread of dengue across the country.
In conclusion, our study provides model-based estimates of spatial variation in monthly dengue cases across Bangladesh without compromising data for model fit. We believe that our findings provide a valuable assessment of the national dengue situation accounting for under-reporting. This study will contribute important information for prioritising and targeting dengue control and elimination interventions across Bangladesh.
|
10.1371/journal.pcbi.1003447 | A Division in PIN-Mediated Auxin Patterning during Organ Initiation in Grasses | The hormone auxin plays a crucial role in plant morphogenesis. In the shoot apical meristem, the PIN-FORMED1 (PIN1) efflux carrier concentrates auxin into local maxima in the epidermis, which position incipient leaf or floral primordia. From these maxima, PIN1 transports auxin into internal tissues along emergent paths that pattern leaf and stem vasculature. In Arabidopsis thaliana, these functions are attributed to a single PIN1 protein. Using phylogenetic and gene synteny analysis we identified an angiosperm PIN clade sister to PIN1, here termed Sister-of-PIN1 (SoPIN1), which is present in all sampled angiosperms except for Brassicaceae, including Arabidopsis. Additionally, we identified a conserved duplication of PIN1 in the grasses: PIN1a and PIN1b. In Brachypodium distachyon, SoPIN1 is highly expressed in the epidermis and is consistently polarized toward regions of high expression of the DR5 auxin-signaling reporter, which suggests that SoPIN1 functions in the localization of new primordia. In contrast, PIN1a and PIN1b are highly expressed in internal tissues, suggesting a role in vascular patterning. PIN1b is expressed in broad regions spanning the space between new primordia and previously formed vasculature, suggesting a role in connecting new organs to auxin sinks in the older tissues. Within these regions, PIN1a forms narrow canals that likely pattern future veins. Using a computer model, we reproduced the observed spatio-temporal expression and localization patterns of these proteins by assuming that SoPIN1 is polarized up the auxin gradient, and PIN1a and PIN1b are polarized to different degrees with the auxin flux. Our results suggest that examination and modeling of PIN dynamics in plants outside of Brassicaceae will offer insights into auxin-driven patterning obscured by the loss of the SoPIN1 clade in Brassicaceae.
| Computational models and functional studies using the plant Arabidopsis thaliana have led to competing models for how the PIN-FORMED1 (PIN1) auxin transporter polarizes in the cell to create both the maxima required for organ initiation and the narrow streams required for vein patterning. Here we identify a previously uncharacterized PIN protein most closely related to PIN1 that is present in all flowering plants but lost in the Brassicaceae, including Arabidopsis. We localized this protein, here termed Sister-of-PIN1 (SoPIN1), along with duplicate members of PIN1 (PIN1a and PIN1b), in two grass species. Our localization data provide striking evidence for a spatial and temporal split between SoPIN1 and the two PIN1s during organ initiation in grasses. Based on our localization results we created a computational model showing that the observed patterns of expression and polarization of the grass PINs can emerge assuming SoPIN1 polarizes up the gradient of auxin concentration while the PIN1 members polarize with the auxin flux. This model reveals a minimal framework of necessary functions involved in auxin-transport-mediated patterning in the shoot and demonstrates that work outside of Arabidopsis is essential to understanding how auxin-transport mediates patterning in most flowering plants.
| Active transport of the plant hormone auxin provides key positional and environmental cues during plant development [1], [2]. Of the numerous auxin transport proteins [3], the membrane-localized PIN-FORMED (PIN) proteins appear to define the direction and rate of auxin movement in many contexts [4], [5]. The angiosperm PIN family can be divided into “short” and “long” classes based on the length of the hydrophilic region [6], [7]. Short PIN proteins are likely involved in auxin homoeostasis within the cell [8]. Long PINs show a characteristic polar localization in the cell plasma membrane that provides directionality to auxin transport [9]–[13]. The hydrophilic loop domains of long PIN proteins contain phosphorylation sites that control PIN cellular localization [14]–[16]. Thus it is likely that variation in function between PIN family members is at least in part due to differing protein domains within this region.
Localization and genetic studies have identified the long PIN group member PIN1 as the major auxin transporter involved in leaf initiation, leaf margin definition, and vascular patterning in shoots [9], [12], [13], [17], [18]. The convergence point hypothesis posits that the creation of auxin maxima by convergent localization of PIN1 defines the locations of initiating leaves, serrations, lobes and vasculature [12], [13], [17], [19], [20]. Most models of how convergent localization of PIN1 facilitates formation of auxin maxima propose positive feedback regulation where PIN1 is allocated to the cell membrane adjacent to the neighboring cell with the highest auxin concentration, thus moving auxin against the concentration gradient [17], [21]–[24]. Such up-the-gradient models are able to accurately recapitulate the initial phase of organ initiation, the formation of PIN1 convergence points and auxin maxima in the correct phyllotactic patterns. These models can also generate files of cells with aligned PIN polarities, similar to those observed during vascular development [24], but do not reproduce localization data showing PIN1 oriented away from auxin maxima, measured using the DR5::GFP reporter, during patterning of leaf veins [12], [13], [17], [19], [20].
Complementary models based on the canalization hypothesis [25], [26] propose an alternate positive feedback where auxin transport is facilitated in the direction of highest auxin flux [27]–[31]. Simulations of this with-the-flux type of polarization can accurately recapitulate formation of canalized traces and are useful in explaining how PIN1 mediates vein patterning [29]. While with-the-flux polarization models can create convergent PIN localization when PIN is assumed to polarize weakly with-the-flux in the epidermis and strongly with-the-flux in subepidermal layers [30], these models predict dynamics that do not match experimental observations. Specifically, they do not predict the observed transient localization of PIN towards the convergence point in internal layers [32]. In addition, they display a transient dip in auxin concentration at the convergence point, which is not observed experimentally [22], [32]. A model that dynamically combines up-the-gradient and with-the-flux modes according to auxin concentration is able to recapitulate the observed DR5 dynamics as well as PIN1 polarization during convergence point formation and vein canalization [32]. However, this dual polarization model requires a hypothetical signal from pre-existing veins in order for new canalization events to consistently connect to the existing vasculature, a pattern that is highly regular in vascular development [32]. Reliably connecting auxin sources and sinks is a noted problem in models of vein formation [33].
Here we describe the phylogenetic analysis of angiosperm long PIN coding sequences. We provide evidence that Arabidopsis and other members of the Brassicaceae have lost a clade of long PIN genes that is conserved in all other angiosperms sampled, a clade we designate Sister-of-PIN1 (SoPIN1). We then localize SoPIN1 along with the PIN1 clade members PIN1a and PIN1b in Brachypodium distachyon (Brachypodium) and maize. These two clades exhibit dramatically different expression and polarization patterns, suggesting a role for SoPIN1 in maximum formation, and for PIN1a and PIN1b in vein patterning. Our computational model shows that these patterns can emerge assuming a combined action of SoPIN1 polarizing up the gradient of auxin concentration and PIN1 members polarizing with the auxin flux. The model also shows how newly formed auxin transport axes may reliably connect to older organs without a hypothetical signal from pre-existing veins.
Our phylogenetic analysis defines four major long PIN clades within the sampled angiosperms (Figure 1A,B). Because of the large body of previous work on Arabidopsis thaliana (Arabidopsis) PINs, we named three clades, PIN1, PIN2, and PIN3/4/7, based on the previously characterized Arabidopsis proteins that nested within these clades. All sampled angiosperms have at least one member in each of these three canonical long PIN clades. However, we also found strong support for a fourth clade placed sister to PIN1, here designated “Sister-of-PIN1” (SoPIN1), which contains sequences from all sampled angiosperms except species within the Brassicaceae, including Brassica rapa, Arabidopsis lyrata, and Arabidopsis thaliana (Figure 1). In previous smaller phylogenetic analyses SoPIN1 proteins were placed in the same clade as PIN1 members [32], [34], [35]. In support of our phylogeny that suggests SoPIN1 is a unique clade, we identified several conserved regions within the variable cytosolic loop of both PIN1 and SoPIN1 proteins that are unique to each clade (Figure S2).
These results suggest that SoPIN1 was lost in the lineage leading to the Brassicaceae sometime after diverging from papaya. In support of this loss, we identified syntenic chromosomal regions across a subset of angiosperms and found that SoPIN1 was absent in the syntenic chromosomes of all sequenced Brassicaceae species despite overall conservation of gene order with other angiosperms that still have SoPIN1 (Figure S3). Thus, Arabidopsis and other Brassicaceae members have lost SoPIN1, one of the four canonical long PIN clades conserved in all other sampled angiosperms.
Within the grasses our phylogenies support a lineage-specific duplication in the PIN1 clade, termed PIN1a and PIN1b based on previous maize annotations (Figure 1A) [36]–[38]. Overall, both PIN1a and PIN1b protein sequences resemble other eudicot PIN1 proteins, but in some regions of the variable cytosolic loop PIN1a and PIN1b have grass-specific sequences (Figure S2). While several species-specific duplication events have occurred in rice, Setaria, and maize, both Brachypodium and Sorghum grasses contain single members within the PIN1a, PIN1b and SoPIN1 clades. The relationship of Brachypodium SoPIN1, PIN1a and PIN1b to Arabidopsis PIN1 is summarized in Figure 1C.
To explore the significance of the loss of SoPIN1 in the Brassicaceae and the duplication of PIN1 in the grasses, we examined expression and localization of PIN1a, PIN1b and SoPIN1 during Brachypodium spikelet development. Each Brachypodium spikelet meristem is indeterminate, and initiates two sterile bracts followed by 7 to 14 floral meristems in an alternate distichous phyllotaxy before terminating (Figure 2A,B) [39]. The first product of each floral meristem is the lemma, a leaf-like organ that surrounds the remaining floral organs (Figure 2B).
We examined PIN expression and localization during lemma initiation in the first few florets. This stage had several advantages for live imaging: the spikelet meristem is relatively exposed early during spikelet development, and the indeterminate nature of the Brachypodium spikelet meristem allows visualization of a developmental series of one leaf initiation event (lemma) and one axillary branch initiation event (floral meristem) at each node in a distichous phyllotaxy (Figure 2A). To visualize each PIN, we created stable transgenic plants containing full-length Citrine (a variant of Yellow Fluorescent Protein, YFP) fluorescent protein fusion constructs under their native promoters.
SoPIN1, PIN1a, and PIN1b have partially overlapping but unique expression domains in the spikelet meristem (Figure 2C–F and Video S1 in supplementary material). SoPIN1 expression is highest in the epidermal cell layer (Figure 2C), and substantial internal expression is restricted to the sites of initiating organs (Figure 2D). In contrast, PIN1a and PIN1b are primarily expressed in the internal cell layers along the presumed paths of incipient lemma veins (Figure 2E,F). PIN1b is also expressed in the center of both spikelet and floral meristems at this stage (Figure 2F). In the epidermis, PIN1a and PIN1b are only expressed in a few cells at the distal tips of both mid and lateral vein traces (Figure 2E,F and Video S1).
To visualize the entire potential path of auxin transport in the Brachypodium spikelet, we examined DR5 expression during lemma initiation. Although indirect, DR5 is a standard reporter used to estimate relative auxin concentrations during development [13], [22], [37], [40], [41]. In the Brachypodium spikelet DR5 expression is highest in the epidermis of both spikelet and floral meristems, at the tip of each lemma primordium, along the path of each incipient vein, and in a broad column down the center of the spikelet (Figure 2G). Only combined SoPIN1, PIN1a, and PIN1b expression matches the entire DR5 expression pattern (Figure 2C–H, Video S1). These data suggest that all three PINs act in concert to create the auxin transport path in the Brachypodium spikelet meristem, but each PIN may have a unique role.
The tunica-corpus theory of meristem organization divides the meristem into the tunica, where cell divisions are primarily anticlinal, or perpendicular to the meristem surface, and the corpus, which undergoes cell divisions in several planes. We examined the cell division planes in the Brachypodium spikelet meristem and found that the outer two cell layers in the meristem apex, L1 and L2, are dominated by anticlinal divisions (Figure S4). This suggests that, similar to wheat and Arabidopsis [42], [43], Brachypodium has a two-layered tunica. Cell divisions in the tunica layers that are parallel to the meristem surface (periclinal) mark the beginning of leaf morphogenesis, and allowed us to easily identify incipient lemma primordia [44] (Figure S4, arrows).
Because organ initiation in Brachypodium is distichous and the spikelet meristem indeterminate, we were able to define two stages of SoPIN1 expression prior to visible lemma morphogenesis, numbered I1 and I2 in order of their appearance (Figure 2A box, 3A). At I2, SoPIN1 expression is highest in the epidermal cell layer and polarity begins to converge, with shootward polarity in abaxial cells and rootward polarity in adaxial cells (Figure 3A, red arrows). DR5 expression at this stage is highest in the apical epidermis and is limited in the internal cell layers (Figure 3B, S5). At the I1 stage, SoPIN1 expression increases in both the epidermal and internal cell layers (Figure 3G, S6). Cellular localization of SoPIN1 at I1 shows strong convergent polarization in the epidermis as well as along the presumptive midvein axis, and is coincident with an increase in DR5 expression in the epidermis as well as internally (Figure 3G, S6). In later stage I1 primordia, SoPIN1 convergence surrounds the periclinal cell divisions in both tunica layers which are a hallmark of leaf initiation in grasses [44] (Figure 3A,H, S7). SoPIN1 is polarized toward the new cell plate in the daughter cells resulting from these periclinal divisions.
After morphogenesis begins, primordia are designated P1, P2, P3, etc., from youngest to oldest (Figure 2A). In P1 primordia, SoPIN1 convergence narrows around a DR5 maximum present in a few cells at the midvein tip (Figure 3B, S5). As the lemma expands, SoPIN1 expression briefly persists at low levels along the axis of the midvein oriented both toward the DR5 maximum at the tip and parallel to the midvein axis (Figure 3M, S8). At the same time, SoPIN1 expression increases at two secondary convergence points equidistant around the circumference of the meristem (Figure 2C circles), marking the initiation of two symmetrical secondary lemma veins (Video S1). At later stages SoPIN1 expression decreases at each midvein convergence point, and by P3 expression along the midvein is almost gone (Figure 2D, Video S1).
PIN1b expression is absent from I2 but was observed at several stages during the formation of the I1 midvein. In a small proportion of spikelets, PIN1b expression was observed to varying degrees in the center of the spikelet meristem apical dome connecting to the P1 midvein trace. This expression appears initially as a small ill-defined spike protruding from the top of the P1 midvein into the apical dome (Figure 3J,L, double arrow). In what is likely a later stage, PIN1b is expressed well inside the meristem apex, consistently limited to the corpus cell layers (Figure 3D, double arrow). PIN1b polarity at these early stages is often unclear, and expression is relatively low. In a larger proportion of the meristems examined, presumably at an even later developmental stage after convergence point formation in I1 by SoPIN1, PIN1b expression narrows and is loosely polarized along the presumptive path of the I1 midvein, connecting the I1 epidermis to the midvein trace of P1 (Figure 3E, double arrow). In summary, we infer a likely progression of PIN1b during the I1 stage, first extending from the P1 midvein (Figure 3J), into the apical dome (Figure 3D), then connecting to the I1 epidermis after maxima formation (Figure 3E).
PIN1a expression is completely absent at the I2 stage. PIN1a expression begins in only a few cells of I1, usually highest in the L2 layer (Figure 3C, S9). In later stages, expression is present in both the epidermis and internally (Figure 3I). PIN1a polarity in I1 is consistently oriented either away from the epidermis (Figure 3I) or rootward (Figure 3C). The area of highest PIN1a expression in I1 correlates with the periclinal divisions that mark the beginning of lemma morphogenesis (Figure 3I).
By P1, PIN1a and PIN1b expression increases, overlapping in a distinct midvein trace. As the P1 primordium extends, the polarity of PIN1a and PIN1b becomes more ordered, oriented along the midvein trace axis into the center of the spikelet and then rootward (Figure 3C,E, S9). While PIN1b expression remains relatively broad and extends completely across the spikelet, connecting to the midvein trace of P2 (Figure 2F), expression of PIN1a is narrow and terminates in the center of the spikelet (Figure 3C,N, S10). PIN1b expression is highest in P1 and P2 and decreases by P3, but is continuously connected between all primordia (Figure 2F). In contrast, PIN1a remains strongly expressed in a narrow path along each presumptive lemma midvein. In each successive older primordium expression extends further rootward into the center of the spikelet toward the midvein of the next older primordium (Figure 2E). We observed PIN1a connecting to older primordia only in much later stages, where the tissue thickness made clear imaging difficult (not shown).
In general, DR5 expression follows the combined expression pattern of PIN1a and PIN1b. Significant expression along the developing midvein occurs in later-stage I1 primordia, and is maintained in each successive older vein trace (Figure 2G). DR5 expression in older primordia is highest in the epidermal maximum at the midvein tip and in the central column where PIN1b is expressed (Figure 3L). Remarkably, in the P1 trace both PIN1a (Figure 3C,F large arrow, Video S1) and PIN1b (Figure 3J–L, large arrow) usually span a region of lowered DR5 expression between the central column of high DR5 and the maximum in the epidermis.
To support our reporter analysis in Brachypodium, we immuno-localized SoPIN1 in maize spikelet meristems (Figure S11A) and found a similar expression pattern where convergent localization of SoPIN1 marked the sites of incipient primordia (Compare Figure S11A to Figure 3A). We also observed a localization pattern similar to the combined expression pattern of PIN1a and PIN1b using antibodies that likely detect both PIN1 proteins in maize (Compare Figure S11B to Figure 3E), suggesting that the split between SoPIN1 and PIN1 is conserved between these two grass species.
We observed that the spatio-temporal patterns of SoPIN1, PIN1a, and PIN1b expression and polarization, as well as the pattern of auxin concentration as reported by DR5, are consistent with aspects of previous models of PIN1 polarization in response to auxin. In general, the progressive convergence of SoPIN1 is coincident with increasing DR5 expression, and thus the formation of presumed auxin maxima in the tunica layers. The resulting SoPIN1 convergence points around DR5 maxima mark the sites of initiating organs and precede PIN1a and PIN1b expression. The observed polarization of SoPIN1 is consistent with the up-the-gradient model of PIN1 polarization leading to the formation of convergence points in the tunica [21], [22]. In contrast, both PIN1a and PIN1b are expressed mainly in the corpus. As development proceeds, they become gradually polarized away from convergence points, and along presumptive vein traces. The expression and polarization of PIN1a is consistent with the with-the-flux model of vein canalization [25]–[28]. The relatively broad expression characteristic of PIN1b was considered from a theoretical perspective by Feugier et al. [31] and Stoma et al. [30], who observed that a weak polarizing response to auxin flux can generate broad regions of PIN1 polarization towards the sink. We postulate that PIN1b in Brachypodium behaves in a similar fashion. To verify whether this conceptual model can plausibly capture the experimentally observed spatio-temporal pattern of expression and polarities of the three PINs in Brachypodium during the initiation of convergence points and vascular strands, we constructed a computational model, described below.
A longitudinal section of a Brachypodium apex was modeled as a regular 2D array of hexagonal cells (Figure 4A). Associated with each cell is the concentration of auxin and the distributions of SoPIN1, PIN1a and PIN1b proteins (Figure 4B). These distributions are represented by storing concentrations of the three PINs separately for each segment of the cell membrane (colored edge of each hexagonal cell). PIN production is assumed to be auxin dependent, permitting expression levels to vary from cell to cell. See the Computational Model Description S1 (CMD S1) Eqs. 7, 8 and 10 in the supplementary materials. Each cell stores a concentration of unallocated PINs (colored circles in Figure 4B), which are moved to the cell membrane by exocytosis (green arrows) and removed from the membrane by endocytosis (red arrows). For PIN1a, exocytosis is increased by total auxin flux through the membrane, and endocytosis is increased by influx through the membrane (CMD S1 Eq. 5). For PIN1b, exocytosis is increased by auxin outflux through the membrane, and endocytosis is increased by influx (CMD S1 Eq. 4). For SoPIN1, allocation to the membrane is increased by high auxin concentration in the neighboring cell (CMD S1 Eq. 9). Cellular auxin concentration (CMD S1 Eq. 1) changes based on biosynthesis, turnover, and the flux due to auxin transport between neighboring cells (CMD S1 Eqs. 2–3). Additional mathematical details of PIN production and allocation can be found in Section 3 of the Computational Model Description S1.
To capture the tissue specific differences in the Brachypodium spikelet meristem observed in our analysis of cell division patterns during lemma initiation (Figure S4), we divided the cellular array into the tunica layers, L1 and L2, and the sub-epidermal corpus (Figure 4A, CMD S1 Section 4.1). Following previous work [12], [32], we assumed that up-the-gradient patterning is particularly strong in the tunica layers. This was implemented by increasing auxin-dependent production of SoPIN1 in the L1 and decreasing auxin-dependent production of both PIN1a and PIN1b in the L1 and L2 layers, consistent with the observed expression patterns (Figures 2 and 3). In addition, we assumed that auxin biosynthesis is increased (two fold) in the L1 [22], [32], which is consistent with observed DR5 expression in the spikelet meristem, and that communication between the L1 and inner tissues is reduced (See CMD S1 Section 4.1) [32], [45], [46].
The model is limited to a segment of meristem immediately below the central zone. Growth is introduced by adding rows of cells to the top end of this segment at regular time intervals. The impact of tissues outside the scope of the model is approximated by boundary conditions (CMD S1 Section 4.2). The effect of older primordia at the base is simulated by withdrawing auxin from the lowest bottom left cell and four bottom right cells in the L1 (Figure 4A). This asymmetry reflects the alternating distichous phyllotaxy of Brachypodium. A single sink cell, placed in the center of the bottom row, provides a target for auxin flow along the axis of the meristem (Figure 4A).
Primordia and the associated vascular strands are produced periodically in the spikelet meristem. Figure 5 shows the simulated dynamics of auxin and PIN distribution during a single period (plastochron). Column 1 shows the state of the simulation at the beginning of this period. A single maximum of auxin concentration, resulting from the convergent polarization of SoPIN1, is present in the L1 layer on the left side of the tissue section shown (Figure 5, Panel 1A). SoPIN1 expression is evident in the L1 and L2 layers, where the production of SoPIN1 is greater than in the sub-epidermal layers (Figure 3A). The convergence point is connected to the sink at the base of the tissue with a single strand of high PIN1b (Panel 1B) and PIN1a (Panel 1C) expression. The high auxin flux in this strand makes it act as an auxin sink for cells closer to the apex. Consequently, PIN1b proteins and auxin transport are polarized towards the strand (Panel 1B, blue arrows). This polarization is the strongest near the vein and gradually decreases closer to the apex. Except for near the convergence point, PIN1b and PIN1a are predominantly expressed in sub-epidermal layers, where their production was observed to be higher than in L1 and L2 (Figure 3C,D).
As the simulation progresses (Figure 5, Column 2), additional rows of cells are added to the top of the tissue (not shown) and increase auxin supply. Transient variation in the auxin concentration triggers an emergent reinforcement of concentration differences in the L1/L2 layers by SoPIN1, leading to the formation of a second SoPIN1 convergence point and auxin maximum opposite to and above the first one (Panel 2A). The auxin that forms the maximum at the new convergence point is supplied by neighboring cells, in which auxin concentration thus decreases, and by increased local auxin production in the new convergence point (see Section 4.3 in the Computational Model Description S1 in the supplementary materials). Auxin from the new maximum enters the sub-epidermal layers and is transported towards the vascular strand below by PIN1b. This transport is coupled with an increase in PIN1b expression in L1/L2 layers, a substantial increase in PIN1b polarization near the convergence point (Panel 2B, region outlined in white) and a broad strengthening of polarization in the region between the new convergence point and the previous vascular strand (Panel 2B). This pattern of PIN1b expression and localization is consistent with the observed progression during the formation of the I1 midvein in the Brachypodium spikelet (Figure 3J,D,E)
The next phase of the simulation (Column 3) is marked by the onset of PIN1a expression at the convergence point, and its gradual extension towards the vascular strand below (Panel 3C). This extension is guided by the broad expression of PIN1b (Panel 3B). This PIN1a expression generates a high-flux strand canalizing auxin transport, and refines the broad field of PIN1b expression into a narrow strand coinciding with that of PIN1a (Panel 3B). Leaving the incipient vein tip, auxin is carried by PIN1b towards the previous vein, forming an approximately triangular region of increased auxin concentration and PIN1b polarization (Panel 3B, region outlined in white). This region is continually refined, both guiding the progressive extension of the incipient PIN1a vein towards the previous strand and being guided by it. Concurrently, PIN1b proteins in the nearby cells become polarized towards this vein (Panel 3B). Combined, these dynamics are consistent with PIN1a and PIN1b expression and localization observed in the formation of the Brachypodium P1 primordia (Figure 3C,E).
The transport of auxin by SoPIN1 towards the convergence point in the L1 and L2 layers, combined with efficient transport by PIN1a and PIN1b along the emerging vein, creates a gap in auxin concentration near the convergence point (arrow in Panel 3C). This gap is consistent with experimental data (large arrow in Figure 3F,K).
As the simulation proceeds, the strand of PIN1a expression extends until it joins the previously patterned vein (Column 4). At this point, PIN1b expression and polarization largely coincide with that of PIN1a (compare Panel 4B and Panel 4C). The last two cells of the strand (Panel 4C, white arrow) exhibit a transient increase in auxin concentration as they progress from a low-flux to high-flux state (compare Panel 4C to Panel 5C). The focused flux of auxin canalized by PIN1a completes the patterning of the incipient vein (Column 5). The increased auxin flux generated by strong PIN1a expression causes PIN1b in nearby cells to polarize towards the newly formed strand, thus indicating its transformation into an auxin sink (Panel 5B).
With the addition of a new primordium and the corresponding vascular strand, the pattern of SoPIN1, PIN1a and PIN1b localization and expression, as well as auxin concentration, in the cells near the apex mirror those that existed prior to initiation of the new primordium (compare Panel 1D and Panel 5D). A new convergence point and midvein trace have been added to the stem. Iteration of this process leads to the formation of convergence points and vascular strands arranged in an alternate branching pattern (Figure 6 and Videos S2 and S3), as observed experimentally (Figure 2).
To examine the dependence of simulation results on the choice of cellular template we constructed a second model operating on a 2D array of square cells (Figure S12 and Video S4). With small changes to model parameters (see Tables S2, S3 in the supplementary materials), we were able to obtain qualitatively similar results to those described above for a 2D array of hexagonal cells. This indicates that the proposed model is not dependent on the choice of a hexagonal cellular template, and can generate equivalent patterns when other cellular topologies are employed.
The patterning mechanisms resulting in leaf initiation and vein patterning integrate three distinct processes: determination of organ placement and vein origins, guidance of newly forming veins towards appropriate end points, and the refinement of emerging veins into narrow canals patterning the procambial tissue. While in previous work these processes have been attributed to a single PIN protein in the shoot, PIN1, our results suggest that outside of the Brassicaceae they are associated with different PIN proteins.
In the grass species Brachypodium and maize the division of these processes between different PIN proteins appears to be particularly crisp. The observed expression and convergent localization of SoPIN1 in the tunica suggests a role for SoPIN1 in the formation of auxin maxima. The expression and localization of PIN1b indicates a broad domain of auxin transport preceding vein formation that connects to previous organs (Figure 3D). Following the formation of a convergence point by SoPIN1, PIN1a appears to refine the broad PIN1b-promoted auxin transport into a narrow auxin stream, leading to the formation of a procambial strand. Based on these expression and localization patterns we propose that each PIN has a distinct role in pattern formation in the shoot: SoPIN1 forms convergence points which determine the sites of organ initiation and position of veins, PIN1b directs the developing veins to target locations, and PIN1a transforms broad regions of polar auxin flux into narrow canals (Figure 7).
We used a computer model to show that the experimentally observed spatio-temporal pattern of expression and polarization of SoPIN1, PIN1a and PIN1b can result from distinct polarization regimes of the three proteins. Specifically, we assumed that SoPIN1 is polarized up the gradient of auxin concentration, PIN1b polarization is a relatively weak linear function of the flux, and PIN1a polarization is a stronger, non-linear (power) function of the flux. The concurrent operation of up-the-gradient and with-the-flux polarization modes is consistent with the previously proposed dual polarization model of primordia initiation and vein formation [32]. Our results provide another example where these two modes suffice to explain the observed localizations of PIN and concentrations of auxin reported by DR5. At the same time, the nature of molecular mechanisms polarizing PIN remains an open problem, and we cannot preclude the possibility that we observe two facets of a single lower-level molecular mechanism.
Several arguments support a functional division between the SoPIN1 and PIN1 clades. First, these two clades are conserved across most angiosperms, which suggests that both are functionally important. Second, there are significant differences in protein sequence between SoPIN1 and PIN1 proteins, which suggests that they may be functionally distinct. Most of these sequence differences are in the intervening hydrophilic loop domain (Figure S2), a region that contains phosphorylation sites involved in PIN1 localization in Arabidopsis [14]–[16]. Third, our evidence suggests that SoPIN1 and PIN1a can have opposing polarities within a single cell, consistent with the presence of different mechanisms controlling their cellular localization. The periclinal cell divisions in the I1 primordia mark a unique place and developmental time point. Our comparison of PIN1a and SoPIN1 polarization at the periclinal cell division in the L2 layer of I1 primordia shows that SoPIN1 is polarized toward the new periclinal cell plate in the L2 division (in Figure 3H), whereas PIN1a is polarized away from the new cell plate, towards the center of the inflorescence meristem (Figure 3I). This opposing polarization pattern persists along the axis of the midvein of P1 lemmas (Compare Figure 3M to 3N). While further work is needed to clarify the functional separation between the PIN1 and SoPIN1 clades, our results support a model in which epidermal convergence point formation and internal vein patterning involve different molecular mechanisms.
Our model for the overlapping roles of PIN1a and PIN1b during vein patterning further suggests that the guiding of new auxin traces to existing sinks (sink finding) can be mechanistically distinguished from patterning the final vascular trace (canalization). In our model PIN1b provides a reliable guiding mechanism allowing new veins to consistently connect with older traces and form the regular pattern of connections observed in Brachypodium. In the context of the river network metaphor proposed for the canalization hypothesis [25], [26], the broad polarization of PIN1b is analogous to a broad slope that orients the overall direction of water flow towards the river mouth. In this setting, the initial flow of water is thus guided by the direction of the slope, which subsequently orients the overall direction of canals that emerge via erosion. Likewise, as PIN1b polarizes towards auxin sinks, the “slope” it provides directs emerging strands towards these sinks. The necessity of supplementing the canalization hypothesis with a guiding mechanism that directs veins towards their target locations (sinks) was observed by Bayer et al. [32]. However, the postulated distinction between PIN1a and PIN1b provides a mechanism for guiding developing veins towards their target locations that is different from the hypothetical guidance by a diffusing substance postulated by Bayer at al.
The question arises to what degree the postulated roles of SoPIN1 in auxin maximum formation, PIN1b in sink-finding, and PIN1a in canalization (Figure 7) can be generalized to diverse angiosperms. In Arabidopsis thaliana and Cardamine hirsuta, a single PIN1 appears to both form convergence points and effect canalization [12], [47]. According to the dual polarization model, this behavior results from PIN1 combining up-the-gradient and with-the-flux polarization modes in context-dependent proportions [32] (alternative explanations have also been proposed [24], [30], as reviewed in [48]). Such integration of polarization modes appears to be limited to Brassicaceae, since all angiosperm species sampled outside of the Brassicaceae have distinct SoPIN1 and PIN1 proteins (Figure 1).
It should be noted that the dual polarization model was largely based on tomato, which has both PIN1 and SoPIN1 members, rather than Arabidopsis. In the experiments of Bayer et al. [32], the localization of the presumed tomato AtPIN1 ortholog was inferred using the AtPIN1::AtPIN1:GFP reporter. Our phylogenetic analysis suggests that the presumed tomato PIN1 ortholog in Bayer et al. is in fact a member of the SoPIN1 clade (Accession HQ127075 in Figure 1A). Remarkably, the expression of the AtPIN1::GFP reporter in tomato appears almost identical to immuno-localization using peptide antibodies targeting this member of the SoPIN1 clade [32]. Both the tomato SoPIN1 clade member and the AtPIN1::GFP reporter are expressed in convergence points as well as along the emerging veins of tomato leaf primordia. This indicates that the apparently sharp distinction between the convergence point formation by SoPIN1 and vein patterning by PIN1 in grasses may not be as precise in other species with both clades. These discrepancies could suggest that all PINs combine with-the-flux and up-the-gradient modes in some proportions, or possibly, that a so far unidentified mechanism controls PIN polarity in response to auxin.
Finally, while PIN1b in Brachypodium provides a striking demonstration of sink-finding behavior, the PIN1a/b duplication was only identified in grass species. Thus the proposed division of roles between PIN1b involved in finding vein targets and PIN1a refining veins does not hold for tomato. A possible solution to the problem of finding the vein target may be a context-dependent transition between weaker (linear) and stronger (non-linear) flux-driven polarization of the so far uncharacterized tomato PIN1 member (HQ127074 in Figure 1A), thus combining the functions of PIN1a and PIN1b. Alternatively, PIN1 in tomato may function similar to PIN1b in Brachypodium, while SoPIN1 in tomato would combine functions of Brachypodium SoPIN1 and PIN1a. One can envision different partitioning of functions between SoPIN1, PIN1 and their variants, with unique solutions in diverse species. Such cross-species comparisons highlight the risks of using well established model species as representatives for large groups. It is likely that considerable variation exists in the systems of auxin self-organization, and further comparative work is essential to outline general mechanisms. The question of whether different solutions drive morphological diversity remains to be tested.
While molecular mechanisms that polarize PIN remain an area of intensive research [49]–[52], our results point to the minimal set of functions that are needed to pattern organ initiation and vasculature in the shoot: convergence point formation, sink finding, and canalization. Our computational model of the behavior of SoPIN1, PIN1a and PIN1b shows that splitting up-the-gradient and with-the-flux modes between separate proteins can provide a robust patterning mechanism consistent with the available localization data. Further examination of the split between SoPIN1 and PIN1 is essential to the understanding of PIN function and the patterning of organs in flowering plants outside the Brassicaceae.
Coding sequences from Phytozome (http://www.phytozome.org/), NCBI (http://www.ncbi.nlm.nih.gov/), and CoGe (http://genomevolution.org/CoGe/) were analyzed with Geneious (http://www.geneious.com/). We found that the third codon position was GC base rich in monocots (60% monocots, 45% eudicots) and biased phylogenetic analyses, thus after MUSCLE alignment, this position was removed. Analysis of GC normalized sequences (33% monocots, 30.2% eudicots) was performed with MrBayes 2.0.3 (http://mrbayes.sourceforge.net/) on GreenButton (geneious.greenbutton.net) using the Jukes-Cantor model of nucleotide evolution (selected using the AIC: MEGA 5.0, http://www.megasoftware.net/). Four chains, sampled every 200 generations, were run until convergence (1,013,000 generations, standard deviation of split frequencies below 0.01). After examination of the likelihood scores, 25% of trees were discarded as burnin. P. patens - Pp1s10_17V6.1 was used as the outgroup. The final tree is available for download on Treebase: http://purl.org/phylo/treebase/phylows/study/TB2:S15020. BLAST searches using both DNA and protein sequences of SoPIN1 members identified only PIN1 clade members in the Brassicaceae. To identify the SoPIN1 syntenic regions across the angiosperms the sequence for a gene neighboring SoPIN1 in Papaya was used in CoGe.
Internal fluorescent protein fusions were generated in a similar position to [37], [40]. Brachypodium PIN promoters and the 5′ part of each coding region were cloned into pDONR-P4-P1R. The YFP variant Citrine was cloned into pDONR221 with 5× Ala linkers. The 3′ part of each coding region and downstream sequences were cloned into pDONR-P2R-P3. Genomic regions 3045, 5164, and 3147 nt upstream of the ATG, and 1652, 1512, 1403 nt downstream of the stop codon were cloned for PIN1a, PIN1b, and SoPIN1 respectively (see supplementary Table S1 for primers). Fragments were recombined using Multisite Gateway (Invitrogen, Grand Island, NY) into pH7m34GW (http://gateway.psb.ugent.be/search). The DR5 synthetic auxin signaling promoter driving expression of an endoplasmic reticulum-localized monomeric RFP (DR5 in text) was described previously [37]. Brachypodium transformation was performed as described [53]. At least three events were characterized for each PIN-Citrine reporter. Two DR5-RFP events were recovered and showed identical expression, similar to maize [37].
Images were captured on a Leica TCS-SP5 laser-scanning confocal equipped with a water-dipping 20× objective (NA 0.7) (http://www.leica-microsystems.com/) and processed with ImageJ (http://rsbweb.nih.gov/ij/). Citrine was excited at 514 nm and mRFP at 561 nm. The pinhole was set to one Airy unit. For each z-stack, transmitted light was detected and flattened using an extended-depth-of-field plugin (http://bigwww.epfl.ch/demo/edf/). Fluorescent channels were processed with a median filter to reduce noise and were recombined with the processed transmitted light image either as single z-planes or maximum projections. In most images maximum projections were limited to internal sections in order to reveal sub-epidermal localization and midvein development. Brightness and contrast were adjusted after fluorescence channels were pseudo-colored with look-up tables. PIN cellular polarity was observed through multiple confocal sections, and similar to previous work was defined by a characteristic arching shape [12], [13], [32], [40]. Multiple samples were analyzed at each developmental stage, and ratios printed below each figure label reflect the number of times phenomena discussed in the text were observed out of the total images captured. FM4-64 staining was performed as described in [54].
Purified protein of residues 188–407 of maize SoPIN1 and 188–414 of maize PIN1c were created as described in [55] and injected into Guinea pigs (Cocalico, Reamstown, PA). Western blots with primary serum showed preference of anti-SoPIN1 for SoPIN1 recombinant protein and anti-PIN1c for PIN1a/b/c recombinant proteins (not shown). Serum was used directly for immuno-localization. Dilutions: 1∶200 anti-SoPIN1, 1∶200 anti-PIN1c. Tissue was fixed in FAA and imbedded in Steedman's wax (Electron Microscopy Sciences, http://www.emsdiasum.com/microscopy/default.aspx). 9 µm sections were mounted, dewaxed in ethanol, dried, rehydrated into PBS, blocked with 5% Donkey serum in PBS, then probed. Secondary antibody dilution: 1∶200 anti-Guinea Pig cy3 (Jackson ImmunoResearch, http://www.jacksonimmuno.com/). Washes were performed with 1% Fish Gelatin in PBS (Sigma, http://www.sigmaaldrich.com/).
The computational model was implemented in C++ using the VVE system (an extension of the Vertex-Vertex (VV) system [56] used in [22], [32]), which provides a data-structure and libraries for representing cellular tissues. Details of the model are presented in the Computational Model Description S1 in the supplementary materials.
|
10.1371/journal.pcbi.1002548 | Potassium Starvation in Yeast: Mechanisms of Homeostasis Revealed by Mathematical Modeling | The intrinsic ability of cells to adapt to a wide range of environmental conditions is a fundamental process required for survival. Potassium is the most abundant cation in living cells and is required for essential cellular processes, including the regulation of cell volume, pH and protein synthesis. Yeast cells can grow from low micromolar to molar potassium concentrations and utilize sophisticated control mechanisms to keep the internal potassium concentration in a viable range. We developed a mathematical model for Saccharomyces cerevisiae to explore the complex interplay between biophysical forces and molecular regulation facilitating potassium homeostasis. By using a novel inference method (“the reverse tracking algorithm”) we predicted and then verified experimentally that the main regulators under conditions of potassium starvation are proton fluxes responding to changes of potassium concentrations. In contrast to the prevailing view, we show that regulation of the main potassium transport systems (Trk1,2 and Nha1) in the plasma membrane is not sufficient to achieve homeostasis.
| Without potassium, all living cells will die; it has to be present in sufficient amounts for the proper function of most cell types. Disturbances in potassium levels in animal cells result in potentially fatal conditions and it is also an essential nutrient for plants and fungi. Cells have developed effective mechanisms for surviving under adverse environmental conditions of low external potassium. The question is how. Using the eukaryotic model organism, baker's yeast (Saccharomyces cerevisiae), we modeled how potassium homeostasis takes place. This is because, through mathematical modeling and experimentation, we found that the electro-chemical forces regulating potassium concentrations are coupled to proton fluxes, which respond to external conditions in order to maintain a viable potassium level within the cells. Our results challenge the current understanding of potassium homeostasis in baker's yeast, and could potentially be extended to other microorganisms, including non-conventional yeasts such as the pathogenic Candida albicans, and plant cells. In the future, the fundamental bases for this descriptive and predictive model might contribute to the development of new treatments for fungal infections, or developments in crop sciences.
| Potassium is an essential cation required for many cellular processes including the regulation of cell volume, intracellular pH, protein synthesis, activation of enzymes, and maintenance of the plasma membrane potential [1]–[4]. In their natural environment, most cell types have to accumulate intracellular potassium against a strong concentration gradient. Animal cells utilize the energy stored in ATP to directly pump potassium ions into the cell via the / ATPase. This enzyme is absent in most fungi and plants [2], which have developed alternative mechanisms to control the intracellular potassium concentration. Saccharomyces cerevisiae (S.c.) cells can grow in media with a potassium concentration ranging from to . Despite extensive knowledge about the identity and function of most potassium transporters in this organism [3], a systems level understanding of the interplay and regulation of the various transport pathways is still lacking.
In S.c., uptake of potassium across the plasma membrane is driven by the membrane potential, which itself is generated by proton pumping via the -ATPase, Pma1 [5], [6]. The high affinity and high velocity transporter, Trk1, is the major uptake system for potassium. The expression levels of the other Trk protein, Trk2, are low, compared to Trk1, and therefore considered of minor importance [7], [8]. A low affinity uptake observed by electrophysiological techniques in trk1,2 double mutants has been attributed to the putative calcium blocked channel Nsc1, though the gene responsible for this transport activity has not been found yet [9], [10]. Efflux of potassium is strongly pH-dependent and coupled to sodium toxicity. The antiporter Nha1 extrudes or ions in exchange for protons under acidic environmental conditions and contributes to the continuous cyclic flux of potassium ions across the plasma membrane and to pH regulation [11], [12]. It is only at higher external pH that potassium or sodium is actively extruded by the Ena1 ATPase [13]–[15]. Another potassium efflux system is the voltage gated channel, Tok1. Electrophysiological studies revealed that Tok1 opens at positive membrane potentials, which do not occur under normal physiological conditions [16]. Potassium is also stored in intracellular compartments, in particular in the vacuole. The effect of intracellular transport is, however, not sufficiently characterized yet [3], [17].
Besides protons, a number of other ions are associated with the transport of potassium. The anion bicarbonate was shown to be important for potassium accumulation [18]. Decarboxylation reactions produce carbon dioxide, which is quickly converted to carbonic acid (), by carbonic anhydrase. Carbonic acid can either diffuse freely across the cell membrane or dissociate into bicarbonate (), and protons. While protons can be extruded via Pma1, the permeability of bicarbonate is very low compared to that of carbonic acid. The resulting accumulation of bicarbonate provides the link to potassium homeostasis; the negative charges carried by bicarbonate can be balanced by potassium cations. In principle, other weak acids could contribute in a similar way to potassium accumulation, but our results below and previous investigations suggest that the bicarbonate reaction plays an important role [18]. Potassium transport is also related to ammonium toxicity [19]. Under low external potassium conditions, ammonium leaks into the cells, presumably via potassium transporters. Toxic concentrations of ammonium are counteracted by increased production and excretion of amino acids [19].
The maintenance of a minimal potassium concentration requires the orchestration of the different transport systems under the constraints of various thermodynamic forces. In this article, we use a mathematical model in conjunction with a novel inference algorithm (the reverse tracking algorithm) and model-driven experimentation to identify the key transport mechanisms that must be regulated under the conditions of potassium shortage. We show that the activation of the proton pump, Pma1, and the activation of the bicarbonate reaction sequence are the regulators of potassium homeostasis. We also show that potassium homeostasis is an example of non-perfect adaptation: The intracellular potassium concentration depends on the external potassium concentration and is only regulated to keep minimal levels of potassium required for survival. This is different from other homeostatic systems such as osmoregulation [20], where certain stationary systems characteristics perfectly adapt, irrespective of the external conditions.
To study the response of S.c. cells to an abrupt decrease of external potassium, we performed potassium starvation experiments using and free media. Cells grown in non-limiting potassium ( KCl) were washed with -free YNB medium (YNB without amino acids and ammonium sulphate, Formedium UK, CYN7505 plus 2% glucose, traces of KCl left: 15 , hereafter referred to as Translucent -free medium [21]) and resuspended in the same medium [12]. The time course for changes in intracellular potassium concentrations for the wild type strain exhibits two different phases (Figure 1A). In the first hour of starvation there is a large net efflux of potassium indicated by the rapid decrease in the intracellular concentration. Loss of potassium slows down in the second phase and the internal concentration slowly approaches a new stationary state (Table 1). Although the cells cannot perfectly adapt to the large concentration gradients they are able to keep a certain amount of potassium required for survival (approx. ). Interestingly, the second phase of potassium loss is slower for the trk1,2 double mutant than for the wild type (wt). This is surprising, because it is believed [3] that increased uptake of potassium via Trk1 even at very low external potassium concentrations is a major mechanism of potassium homeostasis. Thus, one would have expected the concentration of internal potassium in the trk1,2 mutant to be lower than in the wild type. The time course for the nha1 mutant is not significantly different from the trk1,2 mutant (see also Figure S7 in Text S1).
Multiple signaling pathways modulate the activity of the various transport systems involved in potassium homeostasis [2]–[6], [14], [15], [22]–[24]. However, it is not entirely clear which of these signals are essential to achieve homeostasis and how they are acting under the constraints set by the thermodynamics of ion transport. To study these constraints, we developed a minimalistic mathematical model which incorporates the essential parts known to be important for potassium homeostasis. The model describes the dynamic coupling between the intracellular potassium concentration , internal pH (), carbon dioxide concentration , membrane voltage , and cell volume . A complete description of the equations and parameter values is given in the Materials and Methods section and derivations can be found in the Text S1. Here, only the basic model structure is given:(1)(2)(3)(4)(5)Equation (1) links the temporal change of the intracellular potassium concentration to the various potassium transport fluxes (Figure 1B). The model comprises the Trk1,2 system (abstracted as a single system, ), the Nha1 antiporter (), and the Tok1 channel (). To mimic the joint contribution of other, mainly non-specific transport pathways for potassium (e.g. Nsc1) we added a potassium leak current to the model. The Ena1 ATPase is neglected because it is known to be inactive at the relatively low external pH used in the experiments [15].
The dynamics of carbon dioxide (Equation (2)) is coupled to the transport fluxes of bicarbonate and carbonic acid . These transport rates are given in the Materials and Methods section (Equations (18–19)) and a detailed derivation of the bicarbonate model [25] is given in the Text S1. Carbon dioxide is produced in various metabolic processes such as the TCA cycle or pyruvate decarboxylation. It is impossible to model all these processes explicitly, but we incorporate them in the effective metabolic carbon dioxide production flux . This flux is an input to the model and was initially assumed to be constant.
The change in pH (Equation (3)) per change in proton concentration is described by the buffering capacity . In principle, is a function of the internal pH, but due to the combined action of various buffering species [26] it can be approximated by a constant for a wide range of intracellular pH values. In addition to the proton fluxes via the -ATPase Pma1 () and the Nha1 antiporter () there are many other proton transport pathways in yeast. The corresponding net flux is subsumed in the proton leak current . The effective proton flux originating from the bicarbonate reaction sequence is given by the term , where is the pH-dependent fraction of dissociated carbon dioxide.
The membrane potential (Equation (4)) is modeled as a charge balance equation (, specific membrane capacitance; , Faraday constant; , surface area of the cell) [27]. We explicitly modeled the charges carried by potassium, total protons () and bicarbonate. The remaining net charges contributing to the membrane potential are subsumed in , which is determined by the initial conditions of the dynamic variables in the model.
The cell volume (Equation (5)) depends on the balance between internal osmotic pressure , external osmotic pressure and turgor pressure [28]. Ion transport processes change the intra- and extracellular solute concentrations and thus have an osmotic effect (Equations (24–26)) in Materials and Methods). The resistance against volume changes is given by the hydraulic permeability parameter [29].
The concentration and voltage dependent kinetics of all transport systems were described by simple thermodynamic consistent relationships. The driving force for the transport fluxes of ions across the plasma membrane can be written as the difference of the membrane potential and the equilibrium potential . The equilibrium potential depends on the concentrations and stoichiometry of the ions transported, see Equations (12–14) in the Materials and Methods section. For the potassium fluxes in Equation (1) and the proton leak in Equation (3) we assumed linear relations (Ohm's law) of the form between the driving force and the transport flux , or the corresponding electrical current , respectively. For the leak currents and we initially assumed constant conductivity parameters (Equations (9) and (11) in Materials and Methods). The conductivity of the transport proteins Trk1,2, Nha1 and Tok1 was modeled as a function of the membrane voltage, see Equations (6–8) in Materials and Methods.
This minimalistic model captures the essential biophysical and thermodynamic constraints under which control of potassium homeostasis operates. Despite the simplicity of the model, the experimental data was not sufficient to uniquely identify all the parameters. We decided to use this flexibility to explore the parameter space for regions that are consistent with the data and performed extensive parameter scans and sensitivity analysis simulations. However, we were unable to identify a single parameter combination which reproduced the experimental time courses for the wild type strain observed in Figure 1A. In the model, all potassium inside the cell was rapidly and completely lost upon starvation (Figure S4A in Text S1). Based on our model simulations, this believed to be caused by a strong efflux via the Nha1 antiporter driven by the large concentration gradient across the plasma membrane. This model behavior is robust against various model variations, including the incorporation of an intracellular potassium storage mechanism that mimics the contribution of intracellular compartments to potassium retention. Thus, we conclude that further dynamic mechanisms counteracting the strong potassium gradient are essential for homeostasis. Importantly, the model described so far incorporates only the biophysics of transport but does not account for gene regulatory, signal transduction or metabolic events affecting the transporter activity.
The fact that the minimal model is not able to reproduce the experimental time courses for potassium starvation means that there are some unmodeled dynamics that are not captured by the model. Under the working hypothesis that the model covers the major biophysical effects of potassium transport we assumed that there are additional regulatory responses to a shortage of potassium. Available knowledge [2], [3] and data is currently not sufficient to develop exhaustive models for the metabolic, signal transduction and gene regulatory responses to potassium starvation. It is not even clear which of the transporters or other components are activated or deactivated for the maintenance of homeostasis. In engineering terms [30], neither the regulators nor the signals triggering their action are sufficiently characterized.
To overcome this limitation, we combined our minimal biophysical model with an inference algorithm for unmodeled dynamics. We assumed that the unknown regulatory events modulate the activity of the transport systems or other components in the model. Mathematically this means that a constant parameter in the model might in fact not be constant, but a function of time. For example, the maximum conductivity (see Equation 6) of the Trk1,2 transport system could be influenced by signal transduction events [3], [31] in response to low potassium. Any attempt to explicitly model this regulation by additional equations is hindered by insufficient knowledge of the structure and dynamics of the regulatory networks involved. However, one might recoin the question and ask: “Is there a function such that the given experimental time course of intracellular potassium and the time course predicted by the model are in sufficient agreement?”. If such a function would exist we would regard the modulation of the Trk1,2 transporter as one potential regulatory mechanism and Trk1,2 as a potential regulator of potassium homeostasis. However, there might be another parameter (e.g. ) associated with a transporter or another component in the model for which a time course exists such that experimental data can be reproduced. Our strategy was now to test different parameters and corresponding processes for being potential regulators, see Figure 2A. We define a transporter or any other component in the model to be a potential regulator if a tracking control signal exists which changes the activity of the component in such a way that the experimental time course and stationary data can be reproduced. We refer to this inference approach as the reverse tracking algorithm, a more detailed mathematical explanation is given in Materials and Methods.
We used the reverse tracking algorithm to test the transporters Trk1,2, Ena1, Nha1, Tok1 and Pma1 and the activity of the bicarbonate reaction for being potential regulators of potassium homeostasis and then compared the predicted tracking control signals to experimental observations. There is no tracking control signal for the major uptake system Trk1,2; see Figure S1 in the Text S1. This is in contrast to the prevailing view that increased uptake of potassium via Trk1 is essential for potassium homeostasis under starvation conditions. The loss of potassium after starvation is slower in trk1,2 double mutants (see Figure 1A) than in wild type cells. It was experimentally observed [12] that these double mutants have a more negative membrane potential than wild type cells under starvation conditions and also when external potassium is plentiful. This stronger membrane potential (see also Figure S3 in the Text S1) counteracts the outwardly directed potassium concentration gradient and thus explains the higher potassium levels after starvation. Taken together, these results show that the uptake of potassium via Trk1,2 is not the primary mechanism to prevent excessive loss of potassium under starvation conditions.
Although we found a tracking control signal for the Nha1 antiporter, we excluded it from our list of potential regulators based on two observations. First, as indicated in Figure 1A, the time course of potassium loss in nha1 mutants is slower than in the wild type and similar to the trk1,2 mutants. Secondly, it was demonstrated that the influence of Nha1 on the internal potassium concentrations decreases with time [11]. This is in contradiction to our predicted tracking signal (Figure S2 in Text S1), which is nonmonotonic in time.
Similarly, the unspecific transport pathways (leak currents) were excluded, because it is not plausible that unspecific transporters are regulated for the specific purpose of potassium homeostasis. This is based on the well founded assumption that all potassium specific transporters are active under our experimental conditions are known [3] and included in the model. The non-specific cation uptake system NSC1 can be excluded, because our medium contains enough calcium to render NSC1 inactive [32]. The proton flux includes many co-transport mechanisms with nutrients and other molecules. It is thus unlikely, that one of these transport mechanisms is specifically regulated in response to potassium starvation.
The remaining parts in our model are the Pma1 -ATPase and the bicarbonate reaction sequence. For both of them, the reverse tracking approach predicts a rapid burst of activity in response to the rapid removal of external potassium (Figure 2B and 2C). Activation of proton pumping by Pma1 (Figure 2B) hyperpolarizes the plasma membrane, which counteracts the large concentration gradient of potassium and thus limits potassium efflux. An increased reaction flux (see Figure 2C) through the bicarbonate system has a similar effect: The negative charges carried by bicarbonate increase the magnitude of the membrane potential and thereby compensate the potassium gradient.
To test the prediction that Pma1 is activated after potassium starvation, we measured Pma1 activity from crude membrane preparations [33] using an in vitro method that has been extensively established as a faithful measure of in vivo Pma1 function [6], [33], [34]. Indeed, the activity measurements confirm the prediction of the reverse tracking algorithm that Pma1 activity increases rapidly (timescale of 10 minutes) and slowly declines during the first hours of potassium starvation (Figure 2D). Control experiments revealed that Pma1 protein levels do not change under these conditions. Moreover, we also observe, as predicted by the model, that the Pma1 activity is higher in the trk1,2 mutant than in the wild type strain throughout the time course of potassium starvation (Figures 2B and D). To further substantiate that the activation of Pma1 is essential for the response to low potassium, we measured growth for Pma1 mutants pma1–204 and pma1–205 [35] with decreased expression and ATPase activity (33 and 50% of wild type). Figure 3 shows that the ratio of the growth rates at 1 mM and 50 mM external potassium is much lower for the mutant strains than that of the wild type. These results are in line with the recent finding that the brp1 mutant, which is a PMA1 promotor deletion, that leads to decreased Pma1 protein levels, presents markedly decreased growth in low potassium medium and defective rubidium uptake [36].
The second prediction from the reverse tracking approach was an increased reaction flux for the bicarbonate system (Figure 2C). This prediction is supported experimentally by an increased mRNA expression of the NCE103 gene coding for carbonic anhydrase, the enzyme catalyzing the bicarbonate reaction (Figure 2E). This result was part of a genome-wide transcriptomic analysis, using DNA microarrays, of the response to potassium starvation (0–120 min) to be published elsewhere (Barreto et al., submitted). It was shown earlier that protein and mRNA levels of carbonic anhydrase are highly correlated [37]. A qRT-PCR measurement confirmed the increase in NCE103 expression in wild type cells shifted to free medium. After 60 minutes of potassium starvation, the NCE103 mRNA levels increase more than four-fold ( independent experiments). These results show that activation of both Pma1 and the bicarbonate reaction sequence are essential for the control of internal potassium concentrations. In non-starved cells the expression of NCE103 is higher for the trk1,2 double mutant than for the wild type (single dot in Figure 2E). A confirmatory semi-quantitative RT-PCR measurement using the same RNA sample as in the microarray experiment and one RNA sample from independent cultures yielded a mean expression ratio of ( data points) for trk1,2 relative to the wild type. These results suggest a simple explanation for the reported hyperpolarization of the trk1,2 double mutant [12]: A high activity of the bicarbonate reaction sequence means that many protons and many bicarbonate ions are produced. Together with a more active proton pump (Figures 2B and D), this results in a more negative membrane potential that counteracts the outwardly directed potassium gradient. The consequence is a higher intracellular concentration of potassium (Figure 1A) in trk1,2 double mutants than wild type cells.
Homeostatic control of a cellular function in response to a changing environment is often mediated by a negative feedback loop. A change in the input signal (e.g. the external potassium concentration) is counteracted by this feedback loop in order to keep an essential cellular quantity (e.g. the intracellular potassium concentration) in a range sufficient for the cell's function. One particular type of feedback is integral control, where the control signal is the time integral of the difference between the reference and the actual quantity [30]. Integral control was observed for a number of cellular processes including bacterial chemotaxis [38], [39] and osmoregulation [20]. A characteristic property of integral control is perfect adaptation, where the steady state input is independent of the steady state output. For potassium this would mean, that the same intracellular potassium concentration (output) is approached irrespective of the extracellular potassium concentration (input).
The activation of proton transport by Pma1 and the activation of the bicarbonate system counteracting low external potassium indicate the existence of a negative feedback loop. To further investigate this feedback, we have modified the potassium starvation experiment. As before, cells were grown at external potassium, but now resuspended in media with different external potassium concentrations. The potassium efflux and the stationary internal concentrations are different for the different external concentrations, which is also reflected by the model (Figure 4A). To test whether these stationary intracellular concentrations are characteristic for the external potassium, we grew cells overnight in media with different external potassium concentrations (Figure 4B). When external potassium is plentiful (), the internal concentration attains an upper limit of approx. . For low external potassium (), the internal concentration is proportional to the external and agrees with the stationary states of Figure 4A. These experiments show, that perfect adaptation by integral control is not a characteristic of potassium homeostasis for low external potassium. The molecular function and characteristics of this feedback have to be further explored.
In summary, we found that direct regulation of potassium transport proteins is not sufficient for the maintenance of viable potassium levels inside the cell. Although the presence of Trk1,2 influences the dynamics of potassium loss under conditions of low potassium, the regulation of their activity is not the main regulatory process. Cells lacking these proteins have higher intracellular potassium concentrations and the loss of potassium after a rapid shift to low external potassium is slower than in wild type cells. The adaptation to low potassium requires a rapid modulation of proton fluxes as a rescue operation via the increased production of bicarbonate and the activation of the -ATPase Pma1 (Figure 5). The observation that the internal steady state potassium concentration is determined by the external concentration indicates, that potassium homeostasis is an example of non-perfect adaptation, excluding the existence of integral control. The detailed sensing and signaling mechanisms remain to be elucidated and currently we cannot distinguish whether changes in internal or external potassium are sensed directly or indirectly, e.g., as changes of the membrane potential.
Although we cannot completely rule out the possibility that other transport systems not considered in the model contribute to homeostasis, we have reason to believe that our model covers the dominant effects required for the maintenance of viable potassium levels under starvation conditions. All experiments were performed in the presence of calcium, which renders the activity of the calcium blocked non-selective cation pathway Nsc1 unlikely. In addition, non-specific transport of potassium is covered in the model by the leak current. The information about potassium storage in intracellular compartments in the literature is limited. To test the influence of intracellular potassium fluxes originating from an intracellular storage mechanism, we added a hypothetical compartment which can release potassium in response to starvation. This modification did not change the qualitative behavior of the model and was not sufficient to explain the slow efflux of potassium and the maintenance of sufficient intracellular potassium after starvation. Thus, we excluded this modification from the model.
Many cation transporters are evolutionarily conserved in other yeast species and even in higher plants [1]–[3]. However, the current knowledge for these organisms is not as detailed. Considering the importance of ion homeostasis for some pathogenic yeasts [40] and for the growth and development of plants, the question of whether the regulation of proton fluxes plays a similar dominant role as in S.c. is an interesting starting point for future research.
The development of dynamic mathematical models requires a compilation of all parts and processes which could potentially be important for a cellular mechanism under consideration. Other processes believed to be negligible are often lumped together in the parameter values of the model. The decision of which processes to incorporate or to neglect is often hampered by insufficient biological knowledge. Incorporating too many details is impractical and leads to overly complex models with many parameters and little predictive power. On the other extreme are simplistic models which potentially neglect important processes and cannot reproduce the experimental data. We believe that our strategy to start with such a minimal model and to infer unmodeled dynamics with a reverse tracking approach might be of broader interest in systems biology. The reverse tracking algorithm provides (i) candidate points of applications for regulatory signals not explicitly captured by the model and (ii) an estimate of the corresponding time dependent regulatory signal. We emphasize that these potential regulatory signals have to be checked for biological plausibility and have to be validated by experiments. It can be applied when the core model for the process of interest is “underfitted”, i.e. when it can not sufficiently reproduce the experimental data because other regulatory process influence the parameters in the model. Its main advantage is that it can be applied even when an explicit modeling of the processes generating these regulatory inputs is beyond reach. On the other hand, the algorithm can be used as a tool for prioritizing experiments. In combination with experiments, it also may also help to indicate which model extensions are most promising.
The basic structure of the mathematical model is given by Equations (1–5) in the Results section. Here we report the details of the kinetic relationships. Parameter values, initial conditions and derivations are provided in the Text S1. In the following , and denote the Faraday constant, the gas constant and the temperature.
Equations (1–5) have the form of a differential algebraic control system(28)with . Here, denotes the dynamical variables () and Equation (4) for the membrane potential corresponds to the algebraic equation . The scalar input function is given by the external potassium concentration . The solution of this system for given values of the parameters and a given input function is denoted by . Assume now, that we can observe of the components of experimentally. We collect the experimentally observable components in . This can be written as with a matrix with binary elements . For an experimentally observable variable the -th column of has a single entry and . A zero column with indicates that can not experimentally be observed and is thus excluded from .
Assume further, that we have experimental data for certain time points in response to the known input function . Most parameter estimation techniques aim to minimize the squared errorover the parameter vector in order to bring the model prediction for a given input close to the experimental data . However, it might be the case that the minimum error is still too large so that the model cannot be regarded as a reasonable description of the data. This could mean that a dynamical process not explicitly accounted for renders at least one component of the parameter vector to be a time dependent function instead of being constant. The reverse tracking algorithm aims (i) to identify, which of the components of are potentially time dependent and (ii) to predict the time course which minimizes the error. Although the unmodeled dynamical process might effect more than one component, we consider for simplicity each component separately and solve the problem(29)for each component separately. Here, denotes the parameter vector with the l-th component excluded. We then regard as a potential regulatory input, if the problem (29) has a solution with a minimum error smaller than a predefined threshold : . There might be more than one potential regulatory input and the decision of which of these are real can only be made from biological considerations or from additional validation experiments. For example, it might be that has a huge magnitude or takes unrealistic values which could be used to exclude from the list of potential regulatory inputs.
Mathematically, problem (29) is an optimal tracking problem, which often can be solved by a feedback control law [46]. This means that the function is updated according to the local error at time . For a scalar we found the integral controller [30](30)to be a good solution. During a reverse tracking run, this equation is numerically integrated in parallel with the dynamic equations (28). Here, is a least squares spline fit to the experimental data points . Details and suitable parameter values for are provided in the Text S1.
Details about the wildtype strain BY4741, the related trk1,2 mutant and the Translucent free medium can be found in [12], [21].
Cells were grown in Translucent -free medium supplemented with the indicated amount of KCl to an OD600 of 0.4–0.6. Intracellular potassium concentrations were measured by atomic emission spectrometry after extracting the cells with acid as previously described [12]. The time course of internal potassium was obtained by growing the cells in KCl, then cells were washed with Translucent -free medium (traces of KCl left: ) and resuspended to the same free medium or containing the indicated KCl concentrations. Apart from the washing procedure the medium contains 2% glucose.
Data for NCE103 expression changes upon potassium starvation was obtained in the context of a genome-wide transcriptomic analysis by DNA microarray (Barreto et al., Manuscript submitted). Microarray data has been deposited at NCBI's Gene Expression Omnibus [47] and are accessible through GEO Series accession numbers GSE24711 (trk1 trk2 data) and GSE24712 (time-course data). Briefly, wild-type strain BY4741 cells were grown in Translucent medium supplemented with 50 mM KCl to OD 0.8. Cells were centrifuged and resuspended either in fresh Translucent medium with 50 mM KCl or without potassium. Samples (20 ml) were taken at 10, 20, 40, 60 and 120 min by rapid filtration from 4 biological replicates. Total RNA was extracted by using the Ribo PureTM Yeast kit (Ambion) following the manufacturers instructions. cDNA was prepared and indirectly labeled with Cy3 and Cy5. Images with a resolution of 10 were analyzed with the GenePix Pro 6.0 software (Molecular Devices).
Microarray data was confirmed by qRT-PCR using independent RNA samples. To this end, 60 ng of RNA were amplified using oligonucleotides RT_ NCE103_5 (TCATTACCTGTCGCACTG) and RT_ NCE103_3 (CACAAAAGTTACCCCAAAA) and the QuantiTect SYBR Green PCR Kit (Quiagen).
Cell cultures were grown at in Translucent YNB medium containing KCl to OD660 0.6, then washed with Translucent - free medium and resuspended in the same medium without KCl. At the indicated times, cell samples were pelleted by centrifugation, resuspended in of fresh media (with KCl for t = 0 and without KCl for the remaining samples), incubated for 5 minutes and frozen in liquid nitrogen. For the crude membrane purification, of 3× extraction buffer (0.3 M Tris-HCl pH 8.0, 180 mM KCl, 30 mM EDTA, 6 mM DTT and Protease Inhibitor Cocktail (Roche)) was added to the thawed samples and cells were broken by vortexing in the presence of an equal volume of glass beads. of GTED20 buffer (20% glycerol, 10 mM Tris-HCl pH 7.6, 1 mM EDTA and 1 mM DTT) were added to the crude extract, which was then centrifuged 5 minutes at 2000 rpm. The supernatant was transferred to a new tube and centrifuged 20 minutes at 13000 rpm. The insoluble fraction was resuspended and homogenized in of GTED20. The total amount of protein present was estimated using the Bradford assay (BioRad). The amount of Pma1 present in this protein fraction was estimated by comparing the amount of Pma1 to a protein standard curve separated in SDS-PAGE gels stained with Coomassie Blue. In a microtiter plate, of total protein (which corresponds to of Pma1) were assayed in the presence and absence of a Pma1-specific inhibitor, dietilstylbestrol (DES, final concentration 0.2 mM). The reaction was started by adding of the reaction buffer (50 mM MES-Tris pH 5.7, 5 mM MgSO4, 50 mM , 5 mM Na Azide, 0.3 mM Molybdate, 2 mM ATP) and the plate was incubated for 20 minutes at . The reaction was stopped by adding of detection solution (2% sulphuric acid, 0.5% ammonium molybdate, 0.5% SDS, 0.1% ascorbic acid) and the color was allowed to develop for 5 minutes before reading the absorbance in microplate reader (BioRad) at 750 nm. Residual activity values in the presence of DES were subtracted from the absolute activity values to obtain the Pma1 activity measurements. The results represent the average of at least 4 measurements at each time point and essentially identical results were observed in two separate experiments. Measurements of Pma1 activity are expressed in mmol/min/g Pma1. Error bars represent the standard deviation.
|
10.1371/journal.ppat.0030053 | Molecular Basis for a Lack of Correlation between Viral Fitness and Cell Killing Capacity | The relationship between parasite fitness and virulence has been the object of experimental and theoretical studies often with conflicting conclusions. Here, we provide direct experimental evidence that viral fitness and virulence, both measured in the same biological environment provided by host cells in culture, can be two unrelated traits. A biological clone of foot-and-mouth disease virus acquired high fitness and virulence (cell killing capacity) upon large population passages in cell culture. However, subsequent plaque-to-plaque transfers resulted in profound fitness loss, but only a minimal decrease of virulence. While fitness-decreasing mutations have been mapped throughout the genome, virulence determinants—studied here with mutant and chimeric viruses—were multigenic, but concentrated on some genomic regions. Therefore, we propose a model in which viral virulence is more robust to mutation than viral fitness. As a consequence, depending on the passage regime, viral fitness and virulence can follow different evolutionary trajectories. This lack of correlation is relevant to current models of attenuation and virulence in that virus de-adaptation need not entail a decrease of virulence.
| Virulence expresses the harm that parasites inflict upon their hosts. Many studies have addressed the basis of virulence and its effect on host and parasite survival. It has generally been accepted that one of the components of parasite virulence is fitness, or the capacity of the parasite to multiply in its host. Some models have equated virulence with fitness. In the present study, we use foot-and-mouth disease virus (FMDV) to document that virulence and fitness—measured in the same biological environment provided by cells in culture—can be unrelated traits. This has been achieved by multiplying the virus in a manner that mutations accumulated in its genome. Mutations decreased fitness dramatically, but not virulence. Chimeric and mutant viruses were constructed to show that virulence is influenced by only some of the FMDV genes, while fitness is influenced by the entire genome. For this reason, virulence is more robust (“resistant”) than fitness to the effects of deleterious mutations. The fact that virulence can be unrelated to fitness has implications for the design of anti-viral vaccines because it suggests that it may be possible to design high fitness, low virulence strains to stimulate the host immune response. Furthermore, in modelling studies it cannot be assumed that virulence is equal to fitness.
| The relationship between fitness and virulence is an unsettled question, and sometimes fitness is considered a component of the virulence phenotype of parasites. RNA viruses are ideal systems to address this important question because of their high mutability and fecundity, which result in a potential for rapid evolution, and also because of the availability of quantitative fitness and virulence assays.
RNA viruses replicate as complex and dynamic mutant spectra, termed viral quasispecies. Key to quasispecies dynamics are mutation rates in the range of 10−3 to 10−5 substitutions per nucleotide copied, and competition among continuously arising variant genomes [1–4], which prompt rapid movements in sequence space, with corresponding changes of position in the fitness landscape [5]. Indeed, large population passages of RNA viruses in cell culture permit competitive optimization of mutant distributions that generally result in fitness gain [6,7], while repeated bottleneck events (experimentally realized as plaque-to-plaque transfers) lead to random accumulation of deleterious mutations (operation of Muller's ratchet [8]) and result in average fitness decreases [9–13]. Fitness recovery of low fitness foot-and-mouth disease virus (FMDV) clones occurs mainly with introduction of mutations along the genome, with very few true reversions.
An understanding of the consequences of fitness variation for viral virulence is a key question for viral pathogenesis and evolution. Here, we approach this issue with FMDV, an important viral pathogen in veterinary medicine [14], and one that fully participates of quasispecies dynamics. Our laboratory has characterized multiple FMDV variants that derive from one original biological clone, and that occupy widely different fitness levels when replicating in a defined environment in cell culture. We define fitness as the replication capacity of a mutant FMDV, relative to a reference FMDV, in direct growth-competition upon coinfection of baby hamster kidney 21 (BHK-21) cells [15–17]. Fitness of FMDV in BHK-21 cells is a multigenic trait [7].
In the present study, we define virulence of FMDV as the capacity of the virus to kill BHK-21 cells under a standard set of cell culture conditions [18]. Thus, the FMDV–BHK-21 system offered a means to investigate in a direct and comparative fashion the relationship between fitness and virulence of a virus, measured in the same biological environment provided by BHK-21 cells. We describe the behavior of an FMDV clone (
), which has a history of repeated serial plaque-to-plaque transfers in BHK-21 cells [11], that attained a very low fitness value relative to its parental reference virus (C-S8c1), and yet, its virulence for BHK-21 cells was significantly higher than that of C-S8c1. A comparative study of the capacity to kill BHK-21 cells of chimeric FMDVs constructed with cDNA copies of the two parental FMDVs indicates that the enhanced virulence for BHK-21 cells of the low fitness clone is a polygenic trait, with participation of the regions encoding capsid proteins and non-structural proteins 2A, 2B, and 2C as virulence determinants. Three specific amino acid replacements in 2C have been identified as redundant virulence determinants of FMDV for BHK-21 cells. Thus, while large population passages of the virus resulted in a gain of both fitness and virulence, subsequent bottleneck passages resulted in a decrease of fitness but not of virulence.
The results suggest that fitness is very vulnerable to mutation in any genomic region. In contrast, because of the involvement of several (but not all) viral genes in virulence, and the redundant effect of three 2C substitutions, virulence is a more robust phenotypic trait than fitness, and less vulnerable to accumulation of mutations. Therefore, we provide direct evidence that viral fitness and capacity to kill cells can (in some cases) be unrelated traits. Furthermore, the relationship between fitness and virulence, of being either linked or unrelated traits, depends on the evolutionary history of the virus. This observation has implications for viral pathogenesis, and sheds light on models of virulence proposed on the basis of theoretical and experimental studies with cellular organisms.
Several biological clones and populations were obtained by passaging FMDV biological clone C-S8c1 [19–22] in BHK-21 cells, either as large population passages or plaque-to-plaque transfers (Figure 1). The biological clones and populations differed up to 236-fold in relative fitness (Table 1). The fitness differences found are expected from previous results on fitness gain upon large population passages of RNA viruses [6,7] and fitness decrease upon plaque-to-plaque (bottleneck) transfers [9–13]. The initial experiment was aimed at testing whether
, because of its low fitness (0.11 times that of its parental C-S8c1 [12,23,24] [Table 1]), had an advantage in establishing a persistent non-cytopathic infection in BHK-21 cells as compared with its parental clone, C-S8c1 (Figure 1). A persistent FMDV infection is established by growing the cells that survive a standard cytolytic infection with FMDV [25]. Confluent monolayers of BHK-21 cells were infected either with C-S8c1 or with
at a multiplicity of infection (MOI) of 0.02–0.1 plaque-forming units (PFU)/cell (2 × 106 cells infected with 4 × 104 −2 × 105 PFU). Unexpectedly, at 24 h postinfection, the cells infected with
showed extensive cytopathology, and at 48 h postinfection, no surviving cells were observed. The frequency of surviving cells in parallel infections with C-S8c1 was 5 × 10−3–9 × 10−3, which is consistent with previous determinations [25]. No persistently infected BHK-21 cell cultures could be established with
, despite several attempts. Thus, C-S8c1, which displays a 9-fold higher relative fitness than
in BHK-21 cells, showed a capacity to kill BHK-21 cells that was at least 103-fold lower than the killing capacity of
in the infectivity assay intended to establish a persistent FMDV infection.
The capacity of
to kill BHK-21 cells despite its low fitness in BHK-21 cells led us to quantitatively examine the relationship between fitness of FMDV and its capacity to kill BHK-21 cells. To this aim, FMDV clones or populations were compared in a cell killing assay, consisting in determining the time required to kill 104 BHK-21 cells as a function of the PFU added (described in Materials and Methods). The results (Figure 2A) indicate that over the time range of 12 h to 48 h postinfection, the number of PFUs needed to kill 104 BHK-21 cells varied logarithmically as a function of time. Similar quantifications of relative virulence were obtained by measuring the PFU needed to kill 104 cells in 24 h, and then by extrapolating the PFU values to 0 h postinfection (Tables 1 and S1). Virulence of
was 29 to 35 times higher than virulence of C-S8c1, despite the latter displaying a 9-fold higher fitness (Tables 1 and S1). The high virulence of
was not due to the plaque-to-plaque transfers, since a high virulence was also quantitated for its parental clone,
, and for population C-S8p113 (Figure 2B; Tables 1, 2, and S1).
deviated from a line that correlated relative fitness of FMDV and the logarithm of cell killing capacity, as reflected in the decrease of the regression coefficient (R2) (inset in Figure 2A). Probably, this deviation is due to the fact that
lost fitness due to plaque-to-plaque transfers, and the other viruses were not subjected to plaque-to-plaque transfers. On the other hand, virulence determinants were acquired during the large population passages done between C-S8c1 and C-S8c1p113. The 29- to 35-fold higher virulence of
with respect to C-S8c1 (Tables 1 and S1), despite its low fitness, indicates that viral fitness and virulence can be two unrelated traits.
The comparison of the consensus nucleotide sequence of the
genome with that of C-S8c1 revealed a total of 47 mutations (Table S2), leading to 21 amino acid replacements affecting structural and non-structural proteins (Figure 3). To identify the genomic regions associated with the increased virulence of
with respect to C-S8c1, we measured the BHK-21 cell killing capacity of nine chimeric viruses rescued from constructs obtained by introducing fragments of cDNA of the
genome into plasmid pMT28, which encodes infectious C-S8c1 RNA [21] (Figure 4). The results (Figure 5; Tables 2 and S1) show that several genomic regions contribute to the virulence of
for BHK-21 cells, and that the major contributors map within genomic positions 2046 to 3760 (residues encoding part of VP2, VP3, and part of VP1, Figure 5A) and 3760 to 5839 (residues encoding 2A, 2B, 2C, and 3A, Figure 5B). The results exclude the internal ribosome entry site and the 3C- and 3D-coding regions as significant virulence determinants of
for BHK-21 cells (virulence of the relevant chimeric viruses ≤ 2.5, relative to C-S8c1; Tables 2 and S1). Infectious progeny production by each chimeric virus was intermediate between the production of the parental viruses pMT28 and
, with no significant differences that could be correlated with virulence (Table 2).
Amino acid substitutions in human rhinovirus protein 2C promoted cytopathology for mouse L cells [26]. Remarkably,
shares with other FMDV clones and populations, notably, MARLS and C-S8p260p3d (the two viruses showing the highest virulence for BHK-21 cells; Figure 2A; Tables 1 and S1), three amino acid substitutions in 2C: S80N, T256A, and Q263H. In addition, MARLS and CS8p260p3d include replacement M283V in 2C, relative to C-S8c1 [27,28]. To test whether any (or a combination) of the three shared amino acid substitutions in 2C contributed to the increased virulence of FMDV, each of the mutations was introduced individually into plasmid pMT28 by site-directed mutagenesis, as described in Materials and Methods. Transcripts of the three mutants, termed pMT28 (SN), pMT28 (TA), and pMT28 (QH) (Figure 4), were used to transfect BHK-21 cells, and the viruses obtained were tested with the BHK-21 cell killing assay. Viruses having any of the substitutions in 2C have a virulence intermediate between that of C-S8c1 and
(Figure 6A). To test whether the combination of the three substitutions in 2C could produce an additional increase of virulence, the three mutations were introduced in pMT28 to rescue the triple mutant pMT28 (SN, TA, QH) (Figure 4). The results (Figure 6B) show that the virulence of the triple 2C mutant is similar to the virulence of the individual 2C mutants. The 2C mutations did not significantly affect the infectious progeny production (Table 2). A testable prediction of this result is that the introduction of the wild-type 2C-3A-coding region in the genetic background of
should produce a virus with lower virulence than
. Indeed, the results with such a chimeric virus (Figure 5D) indicate that the presence of the 2C- and 3A-coding region as the only genetic region of the pMT28 in the genetic background of
resulted in an FMDV with a 2.4- to 4.8-fold lower virulence than
. We conclude that mutations in 2C contribute to virulence of FMDV for BHK-21 cells.
Thus, a virus that evolves towards low fitness levels due to the operation of Muller's ratchet may nevertheless maintain its capacity to kill the same cells in which it displays low fitness. In FMDV, the enhanced capacity to kill BHK-21 cells was multigenic, including participation of non-structural protein 2C with three amino acid substitutions acting in a redundant fashion. In conclusion, the results provide a molecular interpretation of why fitness and virulence of an animal virus can follow disparate evolutionary trajectories, culminating in two unrelated traits.
The capacity of a virus to kill cells is probably influenced by several steps in the virus life cycle, including receptor affinity (which may trigger signalling pathways and alter cell functions) and intracellular viral replication that may lead to metabolic alterations such as transcriptional or translational shut-off [29]. A parallel increase of virulence and fitness as a virus improves its adaptation to a host cell type is expected, since a key parameter that should contribute to the fitness level of a cytophatic virus is the accumulation of infectious particles and its release from cells, which are events often associated with cell killing. This expectation was fulfilled in our experiments (increase of both fitness and cell killing following large population passages of C-S8c1 [Figure 2]), and also in other virus–host systems. In a comparison of two genetically divergent isolates of the whispovirus white spot syndrome virus, whose virulence for the shrimp host Penaeus monodon was measured by in vivo cumulative mortality rates, virulence correlated with competitive fitness in vivo [30]. The onset of type 1 diabetes by coxsackievirus B strains was linked to the viral replication rate and to the infectious dose [31]. In engineered alphavirus replicons, a direct correlation between the level of RNA replication and cytopathogenicity was observed [32]. At an epidemiological level, a greater replicative fitness of historical versus current human immunodeficiency virus type 1 (HIV-1) isolates was taken as evidence of HIV-1 attenuation over time, assuming a direct connection between fitness and virulence [33]. In vivo, viral fitness may vary among specific organs, and virulence may be affected only when fitness for some specific target tissues is affected [34].
Replicative fitness is, however, but one of several factors which influence the progression of a viral infection in vivo. In a comparative analysis, R5-tropic and X4-tropic clones of HIV-1 showed similar replication capacity in mitogen-activated T cells. However, X4 clones were transferred more efficiently than R5 clones from dendritic cells to CD4(+) T cells, a fact that can contribute to the competitive advantage of X4 viruses in AIDS patients [35]. Simian immunodeficiency virus SIVmac239 infects both the sooty mangabey and the rhesus macaque, reaching high viral loads in both hosts, yet it is only virulent for the rhesus macaque [36]. Deviations of a positive correlation between viral fitness and virulence were observed also in the plant viruses cucumber mosaic virus [37] and barley stripe mosaic virus [38]. A study of the effect of lysis timing on bacteriophage fitness revealed that a strain with an intermediate lysis time had the highest fitness [39]. The time of transmission may also affect virulence. Nuclear polyhedrosis virus transmitted early to its host, the moth Lymantria dispar, was more virulent than virus transmitted late, although the latter was more productive because the virus could use more host tissue for replication [40]. In a study of the susceptibility of North American and non–North American breeds of Lymantria to several isolates of the fungus Entomaphaga maimaiga, mortality was scored in all cases. However, virulence of the fungus, quantitated by the time of death of Lymantria, was, in some cases, inversely proportional to fitness, quantitated by fungal reproduction in the moth [41]. In all these cases, the molecular basis of the lack of positive correlation between fitness and virulence is not understood.
The results with FMDV clones H5 have documented that both fitness-enhancing and virulence-enhancing mutations can be incorporated in the viral genome in such a fashion that subsequent fitness-decreasing mutations associated with bottleneck (plaque-to-plaque) transfers produce only minimal effects on virulence (Figure 2). The dissection of accompanying molecular events, achieved through quantification of virulence of recombinant and mutant genomes (Tables 2 and S1), provides an interpretation of the lack of positive correlation between virulence and fitness. Multiple fitness-decreasing mutations occur in the course of plaque-to-plaque transfers, distributed throughout the FMDV genome [11]. In contrast, determinants of virulence for BHK-21 cells are multigenic, but concentrated mainly in some FMDV genomic regions. Similar multigenic but discrete virulence determinants have been described also in other virus–host systems [42,43]. To decrease virulence, mutations occurring randomly in the course of plaque-to-plaque transfers should affect specific genomic sites, and this will occur with a lower probability than fitness-decreasing mutations, which can hit any of the multifunctional picornaviral proteins and regulatory regions [11]. This model is reinforced by the observation that three amino acid substitutions in 2C (S80N, T256A, and Q263H) had a similar effect in enhancing FMDV virulence, and the three mutations in the same genome had an effect comparable to each mutation individually (Figure 6; Tables 2 and S1). It is not clear what the basis of the contribution of 2C to virulence for BHK-21 could be. 2C is involved in RNA synthesis and contains a nucleotide-binding domain, although none of the substitutions found in
and
lie within such a domain. An unlikely triple reversion would be required to eliminate the virulence-enhancing effect of the three mutations in 2C. We propose that a higher robustness of the FMDV genome with regard to virulence for BHK-21 cells, rather than to replicative fitness in the same cells, underlies the different trajectories followed by fitness and virulence upon subjecting the virus to repeated bottleneck transfers. Obviously, we cannot exclude that parameters of the virus life cycle, other than fitness as measured in our experiments, could correlate with virulence for BHK-21.
The comparative analysis of FMDV clones and populations shows that shifts in virulence can occur even through the evolution of a single viral clone (C-S8c1), with its restricted genetic diversity prompted by different replication regimes in the same host cells, which also have a clonal origin (see Materials and Methods). We conjecture that the demonstration that fitness and virulence can follow different evolutionary courses has been possible thanks to the consequences of the extreme passage regimes to which the viral populations were subjected: competitive evolution of an ample mutant spectra during repeated large population passages, and accumulation of deleterious (with regard to fitness, but not with regard to virulence) mutations upon plaque-to-plaque transfers (predominance of genetic drift and operation of Muller's ratchet) [12,15].
It must be emphasized that fitness and virulence are relative values that pertain to a defined physical and biological environment. Virulence determinants of FMDV, identified here for BHK-21 cells, need not apply to virulence for the natural animal hosts of FMDV [44]. However, the observation of a lack of correlation between fitness and virulence in a FMDV clone is relevant to current models of attenuation and virulence, since it shows that more virulent forms of a virus need not have a reproductive advantage, and that viral virulence is not necessarily a byproduct of viral fitness. Even if virulence is regarded as an unavoidable consequence of parasite adaptation [45], virus de-adaptation (fitness loss) need not entail a decrease of virulence.
Most current definitions of virulence include both the ability of the pathogen to multiply and to cause harm to its host; some authors, however, assume a direct relationship between fitness and capacity to produce disease [46–48]. In relating the results with FMDV to general models of virulence in host–parasite systems, it must be considered that in the FMDV system, evolution of the host BHK-21 cells could not influence FMDV evolution, because clonal cell populations with a controlled passage history were supplied in constant numbers at each infection event (see Materials and Methods). Therefore, changes in host density, or mobility, as well as pathogen survival in the external environment, all of which are relevant parameters in virulence models [48,49], cannot play a role in our system. A consistent finding in serial passage experiments is that virulence of a parasite increases with passage number in a new host [50]. The results with FMDV infecting BHK-21 cells cytolytically imply that the increase of virulence can be conditioned to the history of passage regimes undergone by a virus.
The invariance of BHK-21 cells in the course of serial cytolytic passages of FMDV is in contrast with the parallel system consisting of BHK-21 cells persistently infected with FMDV C-S8c1 [25], in which the cells are passaged and coevolve with the resident virus [51]. Host–virus coevolution has generally favored a decrease of viral virulence in the field, a classical example being myxoma virus and myxomatosis in rabbits [52].
Our comparison of FMDV clones did not provide evidence of clones with high fitness and low virulence, which, with regard to natural hosts, is an aim of biomedicine to obtain vaccine strains. Yet, the existence of specific mutations that differentially affect fitness and virulence opens the way to engineer candidate vaccine strains unable to kill the host, while maintaining replicative competence. Virulence is, however, a feature of the host–parasite relationship [46], and the mutations needed to impair virulence are expected to be host-dependent [53,54].
The BHK-21 cells used in the present study were cloned by end-point dilution, followed by preparation of a cell stock from a single cell; they were passaged a maximum of 30 times before being used for FMDV infection [25,51]. Procedures for cell growth, infection of BHK-21 cell monolayers with FMDV in liquid medium, and plaque assays in semi-solid agar medium were carried out as previously described [11,19,25,27]. Mock-infected cells were handled in parallel in all infectivity and plaque assays to monitor absence of viral contamination. The FMDVs used in the present study (Figure 1) are (i) the reference clone C-S8c1, which has been assigned a relative fitness of 1.0 [11]. (ii) MARLS, a monoclonal antibody escape mutant isolated from population C-S8c1p213 [55]; MARLS has a fitness of 25 relative to C-S8c1 [24]. (iii) C-S8p260p3d, a standard FMDV virus rescued by low MOI passage of C-S8p260. The latter is a virus that evolved by passage of C-S8c1 at a high MOI, which resulted in dominance of two defective FMDV genomes (both including internal deletions) that were infectious by complementation, in the absence of standard virus [22,24,28]; C-S8p260p3d has a relative fitness of 20 [24]. (iv) REDpt60, obtained after 60 successive plaque-to-plaque transfers of RED (a monoclonal antibody escape mutant isolated from population C-S8c1p100) [20]; REDpt60 has a fitness of 1.9 relative to C-S8c1. (v) C-S8c1p113, a viral population obtained after 113 serial cytolytic passages of C-S8c1 at a high MOI in BHK-21 cells (2 × 106 BHK-21 cells infected with the virus contained in 200 μl of the supernatant from the previous infection). (vi) Clone
, a biological clone isolated from population C-S8c1p113 [11]; its relative fitness is 26 (unpublished data). (vii) Clone
, obtained after 95 successive plaque-to-plaque transfers of
[11]; its relative fitness is 0.11 [23].
The capacity of FMDV to kill BHK-21 cells was measured as previously described [18,22]. The assay consists in determining the minimum number of PFU required to kill 104 BHK-21 cells after variable times of infection. The assay was performed in M96 multiwell plates with monolayers of 104 BHK-21 cells per well infected with serial dilutions of virus. At different times postinfection, cells were fixed with 2% formaldehyde and stained with 2% crystal violet in 2% formaldehyde. Results are expressed as the logarithm of the number of PFUs needed for complete cell killing (as judged by cell staining with crystal violet, with series of control wells with known numbers of cells) as a function of time postinfection [18,22].
The relative fitness of FMDV
was determined by growth competition in BHK-21 cells as previously described [7,10,11,24,56]. FMDV
was mixed with appropriate proportions of
, which was used as reference virus. This virus has a fitness 8.5-fold higher than that of the reference clone of C-S8c1 in BHK-21 cells [7,10,11,24,56]. Four serial infections were carried out at MOI of 0.1 PFU/cell. The proportion of the two competing genomes at different passages was determined by real-time reverse transcription (RT)–PCR, employing primers 5531wtnew and
, which are able to discriminate FMDV
RNA from
RNA. The nucleotide sequences of the primers will be provided upon request. The fitness vector obtained for
corresponded to the equation y = 0,0206e1,1074x; R2 = 0.9507. The antilogarithm (base e) of the vector slope is the fitness of the assayed virus relative to the reference virus [56].
Viral RNA was extracted by treatment with Trizol as previously described [57]. Reverse transcription of FMDV RNA was carried out with avian myeloblastosis virus reverse transcriptase (Promega, http://www.promega.com) or Transcriptor reverse transcriptase (Roche, http://www.roche.com), and PCR amplification was performed by using either Ampli-Taq polymerase (PerkinElmer, http://las.perkinelmer.com) or an Expand High Fidelity polymerase system (Roche), as instructed by the manufacturers. The FMDV genome-specific oligonucleotide primers used have been previously described [22,58]. In all RT-PCR amplifications, negative amplification controls, including all reaction components except template RNA, were run in parallel to monitor absence of contamination.
Chimeric viruses containing selected regions of
in the genetic background of C-S8c1 (Figure 4) were obtained by replacing the corresponding DNA fragment of pMT28 by a cDNA copy of
RNA, using specific restriction sites. To obtain pMT28/
(436-2046), a chimera that included nucleotides 436 to 2046 of
(the residue numbering of the FMDV genome is as in [11]),
RNA was amplified by RT-PCR using primers NR2 and JH2, and the cDNA was digested with Hpa I (position 436) and Xba I (2046), and ligated into pMT28 DNA digested with the same enzymes. To obtain pMT28/
(2046–3760),
RNA was amplified by RT-PCR using primers 2R1New and pU, and then the cDNA was digested with Xba I (2046) and Avr II (3760). To obtain pMT28/
(3760–5839),
RNA was amplified by RT-PCR using primers 3R2New and 3CD1, and then the cDNA was digested with Avr II (3760) and Rsr II (5839). To obtain pMT28/
(5839–7427),
RNA was amplified by RT-PCR using primers 5531 wt new and C-Not-Pol, and then the cDNA was digested with Rsr II (5839) and Bam HI (7427). To obtain pMT28/
(436-3760),
RNA was amplified by RT-PCR using primers NR2 and JH2, and then the cDNA was digested with Hpa I (position 436) and Xba I (2046), and ligated into pMT28/
(2046–3760) DNA digested with the same enzymes. To obtain pMT28/
(3760–7427),
RNA was amplified by RT-PCR using primers 3R2New and 3CD1, and then the cDNA was digested with Avr II (3760) and Rsr II (5839), and ligated into pMT28/
(5839–7427) DNA digested with the same enzymes. To obtain pMT28/
(2046–7427),
RNA was amplified by RT-PCR using primers 2R1New and pU, and then the cDNA was digested with Xba I (2046) and Avr II (3760), and ligated into pMT28/
(3760–7427) DNA digested with the same enzymes. To obtain pMT28/
(436-7427),
RNA was amplified by RT-PCR using primers NR2 and JH2; the cDNA was digested with Hpa I (position 436) and Xba I (2046), and ligated into pMT28/
(2046–7427) DNA digested with the same enzymes. To obtain
/2C-3A(pMT28), pMT28 was digested with Bgl II (4201) and Rsr II (5839), and a DNA fragment including wild-type 2C-3A-coding region was inserted into pMT28/
(436-7427) DNA digested with the same enzymes. DNA ligation, transformation of Escherichia coli DH5α, isolation of DNA from bacterial colonies, and characterization of DNA by restriction enzyme digestion were performed by standard procedures [59]. The primers used for molecular cloning and site-directed mutagenesis are described in Table S3.
To obtain FMDV C-S8c1 containing the mutations found in gene 2C of
, plasmid pMT28 was subjected to site-directed mutagenesis using an oligonucleotide including the required nucleotide replacement, and 3R2New or 3CD1 as external oligonucleotide primer (Table S3; Figure 4). A DNA fragment termed A was obtained by subjecting plasmid pMT28 to site-directed mutagenesis using primers (reverse) mutSNu, mutTAu, and mutQHu (to introduce mutations S80N, T256A, and Q263H, respectively) and an external oligonucleotide primer (3R2New, forward). A DNA fragment termed B was obtained amplifying pMT28 with primers (forward) mutSNd, mutTAd, and mutQHd (to introduce mutations S80N, T256A, and Q263H, respectively) and an external oligonucleotide primer (3CD1, reverse). DNA fragments A and B, including the desired mutations, were recombined by shuffling PCR using equimolar amounts of DNA fragments and two external primers (3R2New and 3CD1). The DNA with the desired mutation(s) in the 2C gene was digested with Avr II (genomic position 3760) and Rsr II (position 5839), and cloned into pMT28 to generate pMT28 (SN), pMT28 (TA), and pMT28 (QH). To obtain pMT28 (SN, TA, QH), plasmid pMT28 (SN) was subjected to site-directed mutagenesis to introduce mutation T256A in a similar way as described above, and then, plasmid pMT28 (SN, TA) was subjected to site-directed mutagenesis to introduce mutation Q263H. All chimeric viruses and mutants were analyzed by nucleotide sequencing using Big Dye Terminator Cycle Sequencing kit (Abi Prism; PerkinElmer) and sequencer ABI373 as previously described [58]. Sequences were analyzed using DNASTAR 4.0 (http://www.dnastar.com), GeneDoc, and GCC (University of Wisconsin). Each sequence was determined at least twice, with products obtained using different oligonucleotide primers.
DNA from pMT28 or its recombinant and mutant derivatives was linearized with Nde I and transcribed with SP6 RNA polymerase as previously described [22,27]. Transcript RNA integrity and concentration were estimated by agarose gel electrophoresis, in parallel runs with known amounts of standard C-S8c1 RNA. BHK-21 cell monolayers (70% confluent, about 1 × 106 cells) were transfected with RNA transcripts (1 μg RNA) using lipofectin as previously described [59]. Virus was collected from the culture supernatant at 72 h post-transfection. The virus obtained by transfection was passaged twice before using it in biological studies. RNA was extracted and sequenced to ascertain that the virus maintained the genomic structure and mutations of the initial transcript.
Consensus genomic nucleotide sequences of FMDV clones were obtained by RT-PCR amplification of virion RNA using specific primers [7,22,28].
The GenBank accession numbers for the C-S8c1,
,
, CS8p260p3d, and MARLS genomic sequences are AJ133357, AM409190, AM409325, DQ409185, and AF274010, respectively. Nucleotide and amino acid sequences for picornaviruses can be found at http://www.iah.bbsrc.ac.uk/virus/picornaviridae/SequenceDatabase/3Ddatabase/3D.HTM.
|
10.1371/journal.pgen.1006617 | The Zic family homologue Odd-paired regulates Alk expression in Drosophila | The Anaplastic Lymphoma Kinase (Alk) receptor tyrosine kinase (RTK) plays a critical role in the specification of founder cells (FCs) in the Drosophila visceral mesoderm (VM) during embryogenesis. Reporter gene and CRISPR/Cas9 deletion analysis reveals enhancer regions in and upstream of the Alk locus that influence tissue-specific expression in the amnioserosa (AS), the VM and the epidermis. By performing high throughput yeast one-hybrid screens (Y1H) with a library of Drosophila transcription factors (TFs) we identify Odd-paired (Opa), the Drosophila homologue of the vertebrate Zic family of TFs, as a novel regulator of embryonic Alk expression. Further characterization identifies evolutionarily conserved Opa-binding cis-regulatory motifs in one of the Alk associated enhancer elements. Employing Alk reporter lines as well as CRISPR/Cas9-mediated removal of regulatory elements in the Alk locus, we show modulation of Alk expression by Opa in the embryonic AS, epidermis and VM. In addition, we identify enhancer elements that integrate input from additional TFs, such as Binou (Bin) and Bagpipe (Bap), to regulate VM expression of Alk in a combinatorial manner. Taken together, our data show that the Opa zinc finger TF is a novel regulator of embryonic Alk expression.
| The Alk receptor tyrosine kinase is employed repeatedly during Drosophila development to drive signaling events in a variety of tissues. The spatial and temporal expression pattern of the Alk gene is tightly regulated. Identifying factors that influence the expression of Alk is important to better understand how Alk signaling is controlled. In this paper we characterize cis-regulatory sequences in the Alk locus and the transcription factors that bind them to govern Alk expression in the Drosophila embryo. Using a robotic protein-DNA interaction assay, we identified the Zic family transcription factor Odd-paired as a factor that binds to regulatory elements in the Alk locus. Binding of Odd-paired to Alk cis-regulatory elements varies spatially, revealing a requirement for additional transcription factors such as the NK3 and FoxF orthologues Bagpipe and Biniou in a subset of Alk-expressing tissues. Our findings provide new insight into the dynamics underlying temporal and spatial regulation of the Alk receptor during embryogenesis.
| During embryogenesis, the Anaplastic Lymphoma Kinase (Alk) receptor tyrosine kinase (RTK) is dynamically expressed predominantly in the primordia of the visceral mesoderm (VM), the developing CNS, the amnioserosa (AS) and in a restricted manner in the epidermis [1]. Alk plays a critical role during VM development, where it is activated in response to the secreted ligand Jelly Belly (Jeb) driving the Ras/MAPK/ERK pathway [2–5]. This leads to expression of founder cell (FC) specific transcription factors (TFs) such as Hand [6], optomotor-blind related-1 (org-1) [4] and factors important in the muscle cell fusion process like dumbfounded/kin of irre (duf/kirre) [3–5]. Jeb/Alk signaling also leads to downregulation of fusion competent myoblast (FCM)-specific factors such as sticks and stones (sns) [7] and Verprolin 1 (vrp1) [8–10]. In addition, Alk signaling in the VM modulates the subcellular localization of the Gli-family TF Lame duck (Lmd), resulting in Lmd translocation from the nucleus to the cytoplasm [11]. Thus, signaling regulated by Jeb/Alk is critical for embryonic FC-specification and the subsequent fusion with FCMs to form a functional larval midgut muscle [2–5].
While we and others have previously identified and characterized several important components and targets of the Alk RTK signaling pathway, little is currently understood about the molecular mechanisms regulating the spatial and temporal expression of the Alk receptor itself. Development of the early VM requires the activity of the NK4/msh-2-like homeobox TF Tinman (Tin) for dorsal mesoderm differentiation, as well as the NK3 and FoxF orthologues Bagpipe (Bap) and Biniou (Bin) [12–15]. Interestingly, the expression patterns of bap and bin in the VM primordia are similar to that of Alk [15]. In addition, ChIP-on-chip studies have shown the region upstream of Alk gene to be occupied by several mesodermally expressed TFs, such as Bin, Bap, Twist (Twi), Tin and Myocyte enhancer factor 2 (Mef2) at different time points during embryogenesis [16, 17]. While binding of these factors has been documented, their importance in the regulation of Alk transcription in the VM has only been initially characterized in case of Tin [16, 17].
Here we address regulation of Alk expression during embryogenesis. We have employed a combination of in vitro and in vivo approaches to identify and characterize Alk-specific enhancer elements, including high throughput yeast one-hybrid screening (Y1H) with a library of Drosophila TFs [18]. This Y1H screen identified the zinc finger TF Odd-paired (Opa) as binding to an evolutionary conserved cis-regulatory module (CRM) within one of the Alk-associated enhancer regions. In agreement with these findings, opa mutants displayed a complete loss of Alk expression in the epidermis and reduced levels of Alk in the VM. Furthermore, CRISPR/Cas9-mediated deletion of the Opa binding site containing region in the Alk locus resulted in a reduction of VM Alk protein together with loss of Alk expression in both the AS and embryonic epidermis, indicating that Opa plays an important role in tissue-specific Alk expression during embryogenesis. We have also identified additional enhancer regions regulated by the Bin and Bap TFs, likely together with additional TFs, that work with the Opa binding CRM to regulate Alk expression in the VM in a combinatorial manner.
To study Alk expression during embryogenesis, we employed transgenic GAL4-lines containing overlapping DNA sequences corresponding to Alk 5-prime upstream regions (Fig 1A, S1 Fig), aiming to identify regulatory elements with activity in the visceral mesoderm (VM). AlkEI6.5-GAL4 was previously described [1] as driving expression in the trunk VM with stronger expression in founder cells (FCs) (Fig 1B, stage 11, arrowhead). We also noted that the AlkEI6.5-GAL4 driver was expressed in the amnioserosa (AS), in keeping with earlier observations that Alk mRNA is expressed in the dorsal-most region of the embryo corresponding to the presumptive AS at the early gastrulation stage (S2A and S2B Fig) [1]. We next analyzed AlkE4-GAL4, which contains 2.4 kb of the AlkEI6.5-GAL4 region and an additional 1.6 kb upstream (4.0 kb in total). This GAL4-driver promotes expression in a similar pattern to AlkEI6.5-GAL4, suggesting this DNA region also contains regulatory elements involved in Alk transcriptional regulation (Fig 1C). In addition, AlkE2.7-GAL4, covering a shorter sequence within AlkEI6.5 and AlkE4, displays activity in the entire trunk VM, being considerably stronger in FCs (Fig 1D, arrowhead).
To ensure the specificity of our transgenic lines for the Alk locus flanking genes we performed in situ hybridization on both neighboring genes namely CG5065 (upstream) and gprs (downstream) (Fig 1A, S3 Fig). Neither CG5065 nor gprs is expressed in a pattern similar to that of Alk in the VM, suggesting that any VM expressing region identified flanking the Alk locus may be involved in the regulation of Alk transcription.
The elevated level of expression of AlkEI6.5-GAL4 and AlkE2.7-GAL4 in FCs compared with other cells of the developing VM suggests Alk may respond to its own signaling. Since signaling in the FCs is driven by activation of Alk by its ligand Jelly Belly (Jeb), we examined expression of AlkE2.7-GAL4 in either the absence of Alk activity (Alk1/Alk10), or upon activation of Alk by overexpression of Jeb in the VM. AlkE2.7-GAL4 expression in the FCs was reduced in Alk1/Alk10 mutants (Fig 1F; arrowhead). In contrast, overexpression of Jeb resulted in robust expression of AlkE2.7-GAL4 in all cells of the VM (Fig 1G; arrowhead). These results suggest that Alk expression in the VM is positively regulated by Alk signaling, representing a positive feedback loop. Thus, we have identified CRMs in the 5’ region of the Alk locus that promote Alk expression in the presumptive amnioserosa and developing VM. Additionally, our preliminary GAL4 analysis suggests the presence of inhibitory modules within this region that likely contribute to the overall regulation of Alk expression.
ChIP experiments performed by the Furlong laboratory have identified a 547 bp CRM (MesoCRM-880) overlapping the AlkE2.7 fragment that binds Bin, Bap, Mef, Tin and Twi TFs [16] (shown schematically in Fig 1A, S1 Fig). Later analysis by the Frasch group identified a 1,984 bp region (AlkE301) in a genome wide Tin ChIP analysis that drives expression in the VM [17] (shown schematically in Fig 1A, S1 Fig). Together with our GAL4 analyses these results suggest that the Alk-RB promoter may be important for the VM expression of Alk. To functionally address the role of Alk-RB we generated deletion mutants targeting the Alk-RB isoform with CRISPR/Cas9 [19–21], employing two independent single guide RNA (sgRNA) combinations. This resulted in genomic deletions of 1053 bp (represented by AlkΔRB_1.22.2) or 1325 bp (represented by AlkΔRB_15.16.2) in the region of the Alk-RB 5’UTR (Figs 1A and 2A; S1 Fig; S1 Table). Both homozygous mutants were embryonic lethal. We further examined the visceral morphology of homozygous AlkΔRB_1.22.2 mutant embryos and control siblings using Fasciclin III (FasIII) as marker for differentiated VM. In control embryos FasIII was expressed in the visceral musculature surrounding the entire midgut, which at later stages of embryogenesis is subdivided into four chambers (Fig 2B; arrowhead). In AlkΔRB_1.22.2 embryos, FasIII-positive midgut muscles were absent while FasIII-expression could still be detected in the embryonic foregut and hindgut respectively (Fig 2C), resembling the Alk mutant phenotype [2]. In agreement with their mutant phenotype, AlkΔRB_1.22.2 mutants lacked detectable Alk mRNA and protein in the VM (Fig 2K, S4 Fig) compared to wild-type animals (Fig 2E and 2H, S4 Fig), while Alk expression levels in the CNS were similar to those observed in control embryos (Fig 2I and 2L; S4 Fig). Alk expression was also lost in the AS and epidermis of AlkΔRB_1.22.2 mutants (Fig 2J and 2K, asterisks). Therefore, expression from the Alk-RB promotor drives Alk expression in the embryonic VM, AS and epidermis and is critical for proper formation of the midgut musculature.
A 3.6 kb genomic region that covered the putative VM and epidermal Alk enhancer regions identified in our initial experiments (Fig 1A) was subjected in parallel to high throughput yeast one-hybrid (Y1H) and more detailed reporter gene analyses. Six fragments (denoted AlkEB6 –AlkEB11; S1A Fig; S2 Table) of approximately 700 bp in length, including a ~100 bp overlap between neighboring fragments, were analyzed.
Embryonic lacZ reporter activity was observed with only two of the DNA fragments studied, namely AlkEB8 and AlkEB9 (Fig 3A–3E). AlkEB8 displayed weaker activity in the VM than that observed with AlkEB9 (Fig 3B, 3B”, 3D and 3D”, arrowheads; quantified in S5 Fig). In addition to VM expression, AlkEB9 was also expressed in the AS and epidermis where it overlapped with Alk protein (Fig 3E, asterisks; S2D Fig). No expression in the AS and epidermis was observed in AlkEB8 (Fig 3C, asterisks; S2C Fig). To further confirm that AlkEB9 contains important enhancer elements for Alk, we performed rescue experiments using AlkE9-GAL4 (Fig 3F–3H). Ectopic over-expression of Alk (AlkEB9-GAL4>UAS-Alk) in an Alk1/Alk10 mutant background resulted in a rescue of the embryonic gut phenotype (Fig 3H). Therefore, the AlkEB9 genomic region contains sufficient regulatory information to allow rescue of the embryonic Alk VM expression.
High throughput Y1H was carried out on the same six fragments employing a library of Drosophila TFs fused to the yeast GAL4 activation domain [18] (Fig 1A, S1 Fig). Based on our reporter gene analysis we focused on the AlkEB9 DNA bait Y1H data set aiming to functionally characterize novel transcriptional regulators of Alk. A set of TFs was identified to bind to the AlkEB9 DNA bait by Y1H screening (Fig 4A). Among these, Odd-paired (Opa) (Fig 4B), Pointed (Pnt), Side and CG14655 bound to the AlkEB9 DNA bait and promoted growth in selective media in all biological replicates performed. We further investigated a role for TFs binding AlkEB9 in Alk transcriptional regulation in vivo, employing paired (prd)-GAL4, which drives expression in alternating parasegments and offers internal control of Alk expression levels in the epidermis. In this assay both Opa and Pnt were identified as potential regulators of Alk, with Opa inducing and Pnt repressing Alk expression (S6 Fig). Of the TFs tested in this study, Opa was the only one that resulted in an increase in Alk protein. We also overexpressed opa with the engrailed (en)-GAL4 driver which resulted in an increase in AlkEB9-lacZ reporter activity as well as Alk protein levels in the epidermis (Fig 4C–4D’), indicating that Opa is sufficient to promote Alk expression. Therefore we focused on a more detailed investigation of the role of Opa Alk transcriptional regulation.
Employing the JASPAR online prediction tool [22], we were able to identify a potential Opa binding site (BS) in the AlkEB9 sequence, JASPAR_OpaBS (GACCTCCGGCTG) (Fig 5A and 5B). In addition, we identified another Opa BS similar to the Opa consensus motif previously reported by [23] and therefore referred to as SELEX_OpaBS (GCGGGGATG) (Fig 5A and 5B). Employing the phastCons database, which identifies evolutionarily conserved elements in a multiple alignment, to analyze this sequence, we found that both binding sites are conserved among Drosophila species (Fig 5B; conservation score in green; Opa BS highlighted in yellow) [24, 25]. We next assessed the ability of Opa to specifically bind these predicted sites by electrophoresis mobility shift assay (EMSA). EMSA was performed on the SELEX_OpaBS and JASPAR_OpaBS sequences, incubating probes with cell lysates from Opa-expressing HEK293 cells in the presence of poly(dI-dC) to prevent non-specific binding. Addition of Opa lysate to the binding reaction resulted in a shift of both SELEX_OpaBS and JASPAR_OpaBS probes and was reversed by addition of 100 fold molar excess of non-labelled probe (Fig 5C and 5D). In contrast, addition of cold probes that were mutated within the SELEX and JASPAR binding sites, based on published data [23], was unable to compete the shift generated upon addition of Opa to the labelled wild-type probe. Furthermore, labelled mutated SELEX_OpaBS and JASPAR_OpaBS probes did not exhibit a mobility shift upon incubation with Opa (Fig 5C and 5D). The above observations led us to characterize the interactions of the Opa with the Alk locus by chromatin immunoprecipitation (ChIP). Consistent with Y1H and EMSA analyses, Opa association is detected with a region upstream of the Alk promoter that spans both the SELEX_OpaBS and JASPAR_OpaBS sequences in chromatin from wild-type embryos (Fig 5E).
To address the importance of the JASPAR_ and SELEX_OpaBS for in vivo Alk transcription we first attempted to identify a minimal region within the AlkEB9 region that could drive VM expression. This analysis led to the identification of a 154 bp fragment including both SELEX and JASPAR Opa binding sites (AlkEB9_OpaBS; schematically shown in S1 Fig) that drives strong VM and epidermal expression, similar to that observed with the 700 bp AlkEB9 fragment (Fig 6A–6F’). Quantification revealed that VM expression from AlkEB9_OpaBS was weaker than that of the 700 bp AlkEB9-lacZ reporter (Fig 6B’ and 6E’; S7 Fig), while expression in the epidermis appeared similar in both the 154 bp and 700 bp fragments (Fig 6C’ and 6F’; S7 Fig). In order to examine the role of the predicted Opa binding sites, we introduced the same mutations as in our earlier EMSA analysis within the 154 bp AlkEB9_OpaBS minimal region to create AlkEB9_OpaKO-lacZ. Mutation of these binding sites led to a loss of lacZ expression in both the VM and epidermis (Fig 6G–6I’, quantified in S7 Fig), implying that the predicted Opa binding sites in AlkEB9 indeed contribute to expression from this element. While mutation of Opa binding sites led to a significant reduction of reporter gene expression in the VM (Fig 6H’; quantified in S7 Fig) this was not complete, in contrast to a complete loss of detectable lacZ activity in the epidermis (Fig 6I’; quantified in S7 Fig). Taken together, these data show that the AlkEB9 genomic region contains sequence-specific binding sites for Opa that regulate expression from Alk enhancer elements.
To further dissect the potential role of Opa as a regulator of Alk expression, we examined opa expression during embryogenesis [26]. opa mRNA can be detected at stage 5 in the ectoderm and mesoderm progenitors spanning the presumptive segmented region of the embryo. At stage 9 opa expression decreases slightly and appears in the neuroectoderm persisting until late embryo stages. In the VM, opa mRNA is observed in a dynamic pattern, where it is expressed in a clustered fashion in PS 3–5 and PS 9–12 (S8 Fig).
We next examined the reporter expression of AlkEB9-lacZ in opa loss-of-function mutants (opa1/opa8). While AlkEB9-lacZ is activated in the entire VM and epidermis in wild-type embryos (Fig 7A and 7A’), opa1/opa8 mutant embryos display only weak reporter activity during embryogenesis (Fig 7B and 7B’; quantified in Fig 7C). The severe developmental defects observed in opa1/opa8 mutants make analysis difficult, however we noted lower levels of Alk protein in the VM and a complete loss of detectable Alk in the epidermis of opa mutant animals, in agreement with the loss of AlkEB9-lacZ activity (Fig 7B). These observations were supported by analysis of RNAi-induced Opa knockdown in the developing mesoderm employing 2xPE-GAL4 (Fig 7D–7F). We observed that embryos expressing opa RNAi (2xPE-GAL4>UAS-opaRNAi) displayed a reduction of AlkEB9-lacZ in the VM at later stages when compared with controls (Fig 7E’, quantified in Fig 7F).
Since Opa has been reported to be required for proper midgut formation, with opa mutants exhibiting an interrupted VM that fails to form midgut constrictions during embryogenesis [26], we also examined Alk signaling in the VM of opa mutants. opa1/opa8 mutants, examined with the FC-marker Org-1, exhibited Org-1 positive VM FCs, however, the level of Org-1 protein observed was less than in control embryos (S8B and S8C Fig). Since reductions in both Alk and Org-1 protein were seen in opa1/opa8 mutants, we asked whether Opa overexpression was sufficient to drive Alk signaling. As expected, bap3-GAL4 driven expression of Jeb in the VM resulted in an increased expression of the HandC-GFP FC marker reflecting activation of Alk signaling (S8D–S8D” Fig). In contrast, bap3-GAL4 driven expression of Opa did not increase HandC-GFP levels (S8E–S8E” Fig). Thus, while Alk signaling may be reduced in opa mutants, Opa is not sufficient to influence FC specification driven by Alk signaling in the embryonic VM.
As a complement to our analysis of opa mutants, we employed the Opa4opt-lacZ transgene as readout for Opa activity, focusing on the epidermis. Opa4opt-lacZ contains four tandem copies of the SELEX determined Opa-BS [23]. In parallel we analyzed the opa3D246 lacZ enhancer trap which reflects opa expression [26]. We observed expression of both opa3D246 and Opa4opt-lacZ in the embryonic epidermis, coinciding with Alk protein (Fig 7G and 7H), suggesting that Opa is both expressed and active in these cells. Furthermore, a mutant Opa4opt-lacZ transgene, called Opa4opt-KO-lacZ, in which the Opa binding sites are mutated, no longer displayed expression overlapping with Alk in the embryonic epidermis (Fig 7I).
Taken together, this data supports an important role for Opa in driving embryonic Alk transcription, particularly in the epidermis, through the AlkEB9 regulatory region. However, in agreement with our earlier analyses, Alk expression in the VM does not depend only on Opa activity, since Alk protein is still observed in the VM of opa1/opa8 loss of function animals (Fig 7B).
Given the presence of Opa binding sites proximal to the Alk-RB isoform promoter, together with the loss of reporter gene activity after deletion of these sites, we next addressed their in vivo relevance for Alk transcriptional regulation. CRISPR/Cas9 genome editing was again employed to delete the identified Opa binding sites (Opa-BS) in the AlkEB9 enhancer region of the Alk locus (Fig 8A; S1 Fig). This resulted in isolation of two viable AlkΔOpaBS mutants: AlkΔOpaBS_10.28.3 and AlkΔOpaBS_10.36.1 (Fig 8A and 8F–8I; S1 Fig). Loss of 151 bp containing the Opa binding sites in AlkΔOpaBS_10.28.3 mutants led to a complete loss of detectable Alk protein in the amnioserosa and epidermis (Fig 8H; S1 Table), indicating this region is essential for Alk expression in these tissues. We also observed reduced Alk protein levels in the VM when compared to control embryos at the same stage (Fig 8H, compare with Fig 8B; quantified in Fig 8N). In close proximity to the Opa binding sites we also observed a cluster of highly scoring JASPAR-predicted binding sites for mesodermal TFs (Bap, Sna and Tin) in the AlkEB9 genomic region, here designated as meso-BS (Fig 8A; S1 Fig). Deletion of this meso-BS region alone, in AlkΔmesoBS embryos, does not appear to affect either Alk protein levels or the formation of a fully developed gut (S1 and S9 Figs; S1 Table). Interestingly, AlkΔOpaBS_10.36.1 removes 178 bp including both the Opa- and the meso-BS sites allowing us to functionally address the contribution of the meso-BS region relative to the Opa binding sites. Deletion of both the meso-BS and the Opa-BS regions (AlkΔOpaBS_10.36.1) results in viable animals, albeit with reduced Alk protein levels when compared to those in control embryos (Fig 8F, S1 Table). Reduction of Alk protein levels in the VM was noticeably stronger in AlkΔOpaBS_10.36.1 when compared with AlkΔOpaBS_10.28.3 mutants (Fig 8F and 8H; quantified in Fig 8N). However, the reduced Alk protein levels observed in AlkΔOpaBS_10.28.3 and AlkΔOpaBS_10.36.1 were still sufficient to drive Jeb/Alk signaling in the VM as measured with HandC-GFP reporter expression (Fig 8G and 8I insets), and form a functional gut as visualized by FasIII staining (Fig 8G and 8I).
Since we detected VM expression activity in the overlapping Alk proximal AlkEB8-lacZ reporter (Fig 3B), we explored the contribution of the corresponding region in the Alk locus to regulation of Alk VM expression. To do this we employed CRISPR/Cas9 genomic editing to remove 808 bp covering part of AlkEB9 (312 bp) and the majority of AlkEB8 (647 bp) (represented by AlkΔEB8) (Fig 8A; S1 Fig, S1 Table). These mutants were homozygous viable, with a wild-type VM morphology (Fig 8K; S1 Fig). Investigation of Alk protein levels in AlkΔEB8 mutants revealed a decrease, but not complete loss, of Alk in the VM (Fig 8J; quantified in Fig 8N), suggesting that CRM(s) within the AlkEB8 region are not essential but contribute to VM expression of Alk. Expression of Alk in the epidermis was not affected, in agreement with a sole epidermal CRM including the Opa binding sites within the AlkEB9 region. To further exclude the possibility that an essential CRM might be located in the overlap between AlkEB8 and AlkEB9, we generated a series of overlapping reporter constructs in this area S1 Fig. We did not observe any VM expression activity in this reporter series (S10 Fig), suggesting that two CRMs, one in the region of AlkEB8 and one in AlkEB9 function together drive VM expression of Alk.
To test the contribution of additional CRMs to VM expression of Alk we extended the AlkΔEB8 deletion to include the Opa binding sites within AlkEB9. This deletion was denoted AlkΔOpaBS+EB8 (Fig 8A, 8L and 8M; S1 Fig, S1 Table). AlkΔOpaBS+EB8 mutants failed to express Alk protein in the VM, epidermis or AS (Fig 8L), and were homozygous lethal due to lack of FC specification (Fig 8M; inset), supporting our hypothesis of several independent CRMs within this area that are critical for Alk expression in the VM.
Taken together, our analysis identifies a CRM proximal to the Alk-RB isoform promotor that contains Opa binding sites as critical for Alk expression in the embryonic AS and epidermis. This region also contributes to Alk expression in the VM. Further deletion analysis reveals additional CRM(s) located within the AlkEB8 fragment that contribute to regulation of Alk VM expression.
Previous studies identified CRMs binding Bin, Bap, Twi, Tin and Mef2 in the Alk locus [16, 17]. In particular the ChIP and reporter gene analyses performed by Jin et al. (2013) suggested Tin binding to be important if not essential for Alk expression. We studied expression of Alk protein and the AlkEB9-lacZ reporter in tin346/ED6058 mutant embryos (S11 Fig). Both Alk protein and reporter gene expression, could be observed in the dorsal epidermis and amnioserosa at stage 10/11 and in the epidermis at stage 14 (S11 Fig), indicating that regulation by Tin is not critical for Alk expression outside the VM. In contrast, Alk and AlkEB9-lacZ reporter gene expression as observed in the VM of control embryos (S11 Fig) was not observed. However, this analysis was inconclusive since it is difficult to address if, and to which extent, VM formation proceeds in tin mutant embryos. Bin and Bap TFs are known to have a critical function during Drosophila VM development [12, 15]. In our initial experiments we were unable to see any effect on Alk expression on ectopic expression of either Bin or Bap alone in the epidermis employing en-GAL4 as driver (S12 Fig), however this may reflect a lack of tissue competence in our experimental approach. Therefore, we analyzed both Alk protein and AlkEB9-lacZ expression in bin and bap mutants focusing on VM expression. Although VM development does not proceed normally in either bin or bap mutants, we could observe Alk protein and AlkEB9-lacZ expression in the VM in both cases (S13B–S13C’ Fig). We next investigated AlkEB8-lacZ expression, which was reduced in both bin mutants and bap mutants (S13E–S13F’ Fig). On closer inspection of the AlkEB8 region we identified four putative Bap binding sites, which we deleted to create AlkEB8ΔBapBS-lacZ. AlkEB8ΔBapBS-lacZ failed to exhibit reporter expression suggesting that Bap may be involved in Alk expression in the VM through binding sites within the AlkEB8 region (Fig 9A and 9B). Based on these findings, we analyzed Alk protein levels in AlkΔOpaBS_10.28.3;bin1/BSC374 and AlkΔOpaBS_10.28.3;bap208/ED6058 double mutant backgrounds to test whether Alk expression was affected in a combinatorial manner. We observed a strong reduction of Alk expression in the VM of both AlkΔOpaBS_10.28.3; bin1/BSC374 mutants (Fig 9C and 9D; quantified in Fig 9G) and AlkΔOpaBS_10.28.3;bap208/ED6058 mutants (Fig 9E and 9F; quantified in Fig 9H), with loss of Bap appearing to have a stronger impact. These results suggest that additional factors, including Bin and Bap, contribute to regulate Alk expression in the VM through the AlkEB8 region of the Alk locus (Fig 9I).
In this study we report the identification of Alk cis-regulatory elements and TF binding sites that control the expression of Alk during embryogenesis. We have been able to identify regions that regulate transcription of Alk in the AS, the VM and the epidermis. We further identify the Opa TF as well as Bin and Bap as regulators of Alk transcription in these tissues during embryogenesis. Taken together our results shed light on the regulatory mechanisms controlling Alk transcription and identify important cis-regulatory sequences required for regulation of Alk gene expression.
The importance of Jeb/Alk signaling in vivo in the embryonic VM for FC specification is well established [2–5]. From this earlier work we know that activated Alk in the VM triggers not only transcriptional activation but also post-translational modifications that promote the specification of the FC fate [3–6, 11]. In contrast, very little is known about factors that mediate Alk transcriptional regulation. In this study we aimed to identify CRMs and TFs important for Alk transcription. The Alk-RA and Alk-RB transcripts encode the same protein, but differ in their 5’ non-coding regions which employ alternative promoters [1]. This potentially allows differential expression of the Alk-RA and Alk-RB mRNA isoforms both temporally and spatially. Such regulation has been described previously for genes such as the Drosophila DOA kinase [27] and the BBG PDZ-protein [28], among others. Embryos in which the promoter of the Alk-RB isoform has been disrupted fail to express detectable Alk protein in the VM, AS and epidermis, and exhibit an Alk loss of function phenotype, revealing that this promoter is critical for Alk expression in these embryonic tissues. However, expression of Alk in the embryonic CNS is not compromised by the removal of the Alk-RB promoter and upstream sequences, suggesting that CNS expression of Alk is independent of the VM, AS and epidermal enhancers identified here. Taken together, our results indicate a critical requirement for Alk-RB expression to ensure sufficient Alk protein levels in the VM for signaling and founder cell specification, as well as for Alk expression in the AS and epidermis where the function of Alk is currently uncharacterized.
Previous reports have studied sequences within the Alk locus either by reporter activity assays [1, 17] or ChIP-on-chip analyses [16, 17]. Our analysis of reporter activity has identified regions upstream of Alk that are active in the AS, VM and epidermis. These coincide temporally with Alk protein expression, allowing us to define Alk VM, AS and epidermal enhancers located proximal to the Alk-RB promoter. High-throughput Y1H screens performed in this study identified a number of TFs that potentially bind to and regulate these regions of the Alk locus. In addition, a genome-wide ChIP-on-chip screen for mesodermal TFs occupancy identified a CRM upstream of the Alk locus that is active during mesoderm development [16]. This CRM maps to 2R:16,639,969..16,640,341 (relative to Dmel_Release_6 sequence assembly) and was described to be bound by mesodermal TFs including Bin, Bap, Mef2 and Tin and Twi [12, 15, 16, 29, 30]. However, none of these factors were found in our Y1H analysis. This may reflect additional requirements for binding of some TFs, which would preclude their identification by Y1H, such as heterodimerization with co-factors or post-translational modifications. Interestingly, homozygous mutants for bin and bap still express Alk protein in the VM, suggesting that while they may be involved in the modulation of Alk expression, additional factors are also important in the regulation of Alk expression in the VM.
One such factor could be the NK-4/msh-2 TF Tinman (Tin) which has been previously reported to bind CRMs at the Alk locus [16, 17]. Indeed, expression of Alk in the VM is affected in tin mutant embryos ([17], this study) however it is not clear if this occurs due to direct regulation of Alk expression by Tin or a general lack of induction of the VM lineage. Moreover, our analysis of AlkΔOpaBS;bap double mutants uncovers a severe decrease in Alk protein in the VM suggesting only a minor direct contribution of Tin. Interestingly, opa has been reported to be directly regulated by Tin during heart development [17, 31] and Tin is critical for the expression of two key VM TFs bin and bap [4, 15]. Therefore it is likely that the importance of Tin for Alk expression relies on its activating potential for these Alk-regulating TFs. Interestingly, loss of tin does not affect Alk expression in the epidermis.
Reporter gene expression analyses suggest the VM Alk enhancer is located upstream of the Alk-RB isoform, in agreement with previously reported AlkE301-lacZ reporter spanning 1,984 bp (Fig 1A; S1A Fig) [17] and the 547 bp MesoCRM-880 [16] that both cover the AlkEB9 region. Our data suggests that Alk-RB expression can be activated through an upstream enhancer that is bound by Opa located within AlkEB9. We were also able to identify additional nearby enhancer elements in AlkEB8 that integrate information from factors such as Bin and Bap that are critical to ensure precise and robust VM expression of Alk. Taken together with the earlier ChIP analyses from the Furlong and Frasch groups, our data suggest that Opa, along with mesodermal TFs such as Bin, Bap, Mef2 and Tin and Twi function in a combinatorial manner to drive robust expression of Alk in the VM (Fig 9I).
Our efforts to identify novel TFs involved in Alk transcriptional control by in vitro Y1H assay resulted in a cluster of TFs potentially binding the AlkEB9 sequence. Of those TF hits for which UAS-transgenes were available to test, only Opa was observed to induce cell autonomous expression of Alk when ectopically expressed. opa is a pair-rule gene [32] that encodes a zinc finger protein important during embryonic segmentation and midgut formation [26, 33, 34], as well as adult head morphogenesis by direct regulation of decapentaplegic (dpp) transcription [23, 35]. opa transcript is expressed in a spatially and temporally dynamic pattern, starting from stage 5 in a broad expression domain and from stage 11 onwards in two discrete domains in the VM corresponding to the first and third midgut constrictions [26, 33].
While Opa plays a role in the differentiating midgut musculature, with opa mutants exhibiting an interrupted VM unable to form midgut constrictions during embryogenesis [26], its role during segment formation presents a challenge when attempting to decipher the contribution of this TF more precisely. One component of this may be the regulation of Alk by Opa shown here. While we observed that opa mutants display lower levels of Alk protein in the VM, Jeb/Alk signaling is not abrogated, suggesting that while reduced, Alk protein levels are not reduced to levels under the threshold critical to drive Alk signaling. The lack of a critical role for Opa in the VM expression of Alk may reflect the importance of Alk signaling in this tissue for survival of the fly, where a more complex network of TFs may be employed to ensure rigorous Alk expression.
Additional VM enhancer elements 5’ of AlkEB9 in the Alk locus are regulated in part by Bin and Bap, two TFs that are critical for VM development. Thus multiple partially redundant enhancer regions are employed to safeguard VM expression of Alk, a phenomenon that has been observed in numerous genes expressed in the Drosophila embryonic muscle [36]. Moreover, while we have tested the role of Opa and the Opa binding sites in the AlkEB9 region of the Alk locus in this work, we have done so under standard laboratory conditions, and as a result have not tested whether either Opa itself, AlkEB9 or AlkEB8 VM enhancers may play an increasingly critical role in Alk expression in more demanding environmental conditions, as it has been described for some Drosophila loci [37]. Although Alk is expressed in bin and bap mutants, our experiments combining deletion of the Opa binding region in Alk in a bin or bap mutant background suggest a combinatorial role for Bin, Bap and Opa driving VM expression of Alk [12–15]. Opa, Bin and Bap potentially act in combination with other TFs to control Alk transcription in the VM, as has been described for sloppy paired-1 (slp1) activation in the somatic blastoderm in response to Opa and Runt [38]. In addition to direct regulation of Alk expression, Opa may also impact Alk expression via indirect mechanisms during embryogenesis.
Further complexity arises when the regulation of opa itself in the VM is considered. It is known that Dpp signaling restricts the VM spatial expression pattern of opa to PS6-8, with dpp mutants showing continuous opa expression throughout the VM [26]. Opa is also known to regulate dpp expression during adult head development [23]. In addition, opa is broadly expressed in the mesoderm at stage 6 potentially driving Dpp signaling. The Dpp mesodermal response consists of up-regulation of tin and bap, important regulatory genes in the dorsal mesoderm that essentially contribute to the specification of the VM [12, 39]. Similarly, Alk activity, the FoxF forkhead domain TF Bin and the Tbx1 Org-1, are also critical factors for expression of dpp in the VM and subsequent activation of Mad signaling in the midgut endoderm [40, 41]. Moreover, loss of org-1, whose expression is maintained by Alk signaling in the VM, results in decreased opa VM expression [4, 41], revealing a complex interplay of regulation where both Alk and Opa control each other’s expression in a spatially and temporally regulated manner.
Surprisingly, in addition to a non-essential role for Opa in the regulation of Alk transcription in the VM, in this work we have been able to identify a critical role for Opa in Alk expression in the AS and epidermis. Here, in contrast to the VM, Opa appears to be required and sufficient to drive Alk expression, although the functional significance of Alk in these tissues remains uncharacterized. Expression of the AlkEB9-lacZ reporter and derivatives in which the Opa binding sites have been mutated indicate that Opa has an important function in Alk transcription through the predicted Opa BS. This is supported by the absence of detectable Alk protein in the AS and epidermis of AlkΔOpaBS mutants, where the Opa binding sites within the AlkEB9 enhancer have been deleted. Given that AlkΔOpaBS mutants are viable, it may be that Alk signaling is employed in a small population of non-essential cells that remain to be identified. Further work will be required to characterize the role of Alk in this context.
We have focused here on the regulation of Alk expression during embryonic development, however, Alk is also observed in larval and adult stages. Although Alk signaling does not seem to be critical for viability post-embryogenesis, a number of important roles in the nervous system have been described [42–47]. While we have not investigated the role of Opa, Bin or Bap in Alk expression at these other stages, nor in the CNS in this study, this would certainly be of interest to address in future experiments.
Standard Drosophila husbandry procedures were employed. Drosophila strains and crosses were maintained on a potato-meal based diet. Crosses were performed at controlled 60% humidity and 25°C conditions. Fly lines used in this study are: UAS-Alk [1], UAS-GFP (Bloomington 4775), UAS-bap.ORF.3xHA (FlyORF #F000006), UAS-bin.ORF.3xHA (FlyORF #F000281), UAS-jeb [6], UAS-lacZ (Bloomington 1776), UAS-opa [35], UAS-opaRNAi (VDRC KK108975), UAS-pnt.P1 (Bloomington 869), UAS-side (Bloomington 9679), Alk1 [2], Alk10 [2], bap208 [12], Df(3R)ED6058 (Bloomington 24140), bin1 (Bloomington 1438), Df(3L)BSC374 (Bloomington 24398), opa1 (Bloomington 3312 and 3222), opa8 (Bloomington 5335), tin346 [12], AlkEI6.5-GAL4 [1], bap3-GAL4 [15], en2.4-GAL4 (Bloomington 30564), prd-GAL4 (Bloomington 1947), twi.2xPE-GAL4 (Bloomington 2517), HandC-GFP [48], opa3D246 [26], Opa4opt-lacZ and Opa4opt-KO-lacZ [23]. Alk alleles generated in this study are summarized in S1 Table.
Transgenic flies generated in this study: AlkE4-GAL4, AlkE2.7-GAL4, AlkEB9-GAL4, eve.p:empty-lacZ, AlkEB6-lacZ, AlkEB7-lacZ, AlkEB8-lacZ, AlkEB9-lacZ, AlkEB10-lacZ, AlkEB11-lacZ, AlkEB9_OpaBS-lacZ, AlkEB9_OpaKO-lacZ, AlkEB8ΔBapBS-lacZ, AlkEB8∩EB9-lacZ, AlkEB8∩EB9+50flank-lacZ and AlkEB8∩EB9+100flank-lacZ. Molecular details of the regions covered by these fragments are described in S2 Table. Genomic coordinates refer to the Dmel_Release_6 sequence assembly [49].
Embryos were stained as described [1]. Primary antibodies used were: guinea pig anti-Alk (1:1000 [3]), rabbit anti-β-galactosidase (1:150; Cappel 0855976), chicken anti-β-galactosidase (1:200; Abcam ab9361), mouse anti-Fasciclin III (1:50; DSHB 7G10), rabbit anti-GFP (1:500; Abcam ab290), chicken anti-GFP (1:300; Abcam ab13970), mouse 16B12 anti-HA.11 (1:500; Covance #MMS-101P), rabbit anti-Org-1 (1:1000, this work), sheep anti-digoxygenin-AP fab fragment 1:4000 (Roche). Alexa Fluor®-conjugated secondary antibodies were from Jackson Immuno Research. Embryos were dehydrated in an ascending ethanol series before clearing and mounting in methylsalicylate.
Images were acquired with a Zeiss LSM800 confocal microscope or Axiocam 503 camera, processed and analyzed employing Zeiss ZEN2 (Blue Edition) imaging software. For analysis of protein levels, the laser, pinhole and PMT settings were adjusted on control siblings subsequently employed for imaging of mutant embryos.
Fluorescence intensity measurements were quantified using Zeiss ZEN2 (Blue Edition). In brief: mean fluorescence values were acquired from regions of interest (ROI), corresponding to the VM or epidermis (Alk staining) selected in confocal sections of stage 11 embryos. This mean fluorescent intensity was corrected using a background ROI chosen from a non-stained area. Measurements were taken from 10 embryos per sample analyzed. For statistical analysis we performed a one-way ANOVA using GraphPad Prism 6 software, where n.s. stands for non-significant, ***p≤0.001 and ****p≤0.0001. All plots are visualized as mean ±S.D.
Recombinant N-terminal Org-1 protein was produced from pET30a—Org-1-N as generated by [50] was purified by His affinity chromatography and injected into rabbits for antibody generation (Genscript).
For in situ hybridization, fragments of Alk, gprs, CG5065 and opa were amplified from genomic DNA with the primer combinations shown in S3 Table. PCR products were cloned into the dual promoter PCRII TOPO vector (Invitrogen) and used as template to generate DIG-labeled in situ probes with SP6/T7 polymerases (Roche). In situ hybridization of antisense probes to embryos was carried out as previously described [51]. Samples were mounted in in situ mounting media (Electron Microscopy Sciences).
pMW2-vectors containing the different Alk putative CRMs were generated by regular cloning techniques (primer combinations shown in S4 Table) and integrated into the yeast genome as described [18]. Each DNA bait yeast strain was then transformed with a library of 647 Drosophila TFs fused to GAL4. Interaction was assessed by growing transformant yeast strains on selective plates followed by data analysis as previously described [18]. Briefly, selective growth of diploid yeast colonies was analyzed by the Matlab-based image-analysis program TIDY which quantifies bright spots, representing yeast colonies to the dark background. For every biological replicate in the screen, each bait-TF interaction was analyzed in four technical replicates resulting in quandrants of yeast colonies as shown for AlkEB9 DNA bait in the results.
For generation of lacZ reporter flies, DNA sequences of for AlkEB6 to AlkEB11 were PCR amplified (S4 Table) and cloned into the eve.p-lacZ.attB vector [52]. In addition, the AlkEB9 DNA bait was cloned into pPT-GAL vector (1225, DGRC) to generate the AlkEB9-GAL4 construct. DNA sequences for AlkEB8ΔBapBS-lacZ, AlkEB9_OpaBS-lacZ and AlkEB9_OpaKO-lacZ were assembled by Genscript and cloned into eve.p-lacZ.attB vector for further PhiC31 directed genome integration. For generation of AlkE4-GAL4 and AlkE2.7-GAL4 constructs, DNA genomic regions covering 2R:16,638,503..16,642,495 and 2R:16,638,510..16,640,834, respectively, were cloned into pCaSpeR-DEST6 (1032, DGRC) by the Gateway system (primer combinations in S4 Table). Constructs were sequenced (GATC Biotech) and injected into w1118 flies, except for attB constructs which were injected into Bloomington 24482 and 24485, for PhiC31 directed integration at 51C and 68E respectively (BestGene Inc.).
Deletions within the Drosophila Alk enhancer region were generated with CRISPR/Cas9 [53]. The sgRNA targeting sequences used (listed in S1 Table) were cloned into pBFv-U6.2 expression vector (Genome Engineering Production Group at Harvard Medical School). Constructs expressing sgRNA were injected into vasa (vas)-Cas9 (Bloomington 51323) embryos by BestGene Inc. Screening of deletion events was performed by PCR and further sequencing (GATC Biotech). For additional complementation tests we employed balanced Alk10 or Df(2R)Exel7144 flies.
DNA coding sequence of opa was synthesized (Genscript) in frame with carboxy-terminal OLLAS and 6xHis tags and cloned into the pcDNA3.1(+) mammalian expression vector. Binding of Opa to the AlkEB9 was analyzed by a DNA binding assay on dsDNA oligonucleotides with cell lysates from HEK-293F cells expressing Opa-OLLAS. Binding reactions were performed as described in [54] containing 10 mM Tris-HCl (pH 8.0), 25 mM KCl and 1 mM DTT, 1 μg poly-dIdC (Sigma-Aldrich), 2.5% glycerol, 0.05% Triton X-100, 0.2 mM MgCl2 and the indicated 3’-end biotin labelled probe. After 20 min incubation at room temperature, reactions were separated on a 6% native TBE-PAGE in 0.5x TBE buffer at 100V. DNA was transferred to nylon+ membranes (Amersham), UV cross-linked to the membrane and detected by Chemoluminiscence Nucleic Acid Detection Module (Pierce) according to manufacturer’s indications. Competition assay was performed by addition of 100 fold molar excess of unlabeled competitor DNA to the reaction mix. Wild-type probes used for band shift experiments were Opa_SELEX and Opa_JASPAR. Mutated version were made according to for Opa_SELEX mutant [23], and in a similar manner for Opa_JASPAR mutant. All four EMSA probe sequences are shown in S3 Table.
Chromatin was prepared from approximately 100 mg of pooled collections of fixed 3–4 hour embryos. The embryos were homogenized for 1 min in 10 mM EDTA and 50 mM Tris (pH 8.1). After addition of SDS to a final concentration of 1% and incubation on ice for 10 min, glass beads (150–200 μm) were added and the homogenates were sonicated to give sheared chromatin preparations with an average DNA size of 300–400 bp. Chromatin immuno-precipitation was performed largely as described previously [55] using an affinity-purified anti-Opa antibody raised against a truncated recombinant protein spanning from amino acids 125–507 of Opa, a region containing the DNA-binding zinc-fingers at a concentration of 0.5 μg/ml with 100 μg of chromatin in 1 ml of 0.01% SDS, 1% TritonX-100, 1 mM EDTA, 20 mM Tris, pH 8, 150 mM NaCl and 1x Protease Inhibitor Cocktail (Roche). After overnight incubation of the chromatin and antibody at 4°C, the mixture was incubated with Protein-A Agarose (Millipore) for 2 hours at room temperature, followed by low-salt, high salt and LiCl washes as used in the Chromatin Immunoprecipitation Assay Kit (Upstate Biotechnology). After heat reversal of protein-DNA crosslinks, protein digestion, phenol chloroform extraction and purification of the nucleic acids by ethanol precipitation the amount of recovered DNA was quantified using qPCR and a standard curve generated for each primer pair with a sample of nucleic acid purified from the input chromatin. The control primer pair produces a 115 bp amplicon located 12.4 kb upstream of odorant receptor 42b, a region devoid of modEncode hallmarks of cis-regulatory DNA sequences. The DESE-Opa primer pair produces a 209 bp amplicon from a central region of the slp1 DESE enhancer that requires Opa for expression [56]. The Alk primer pair produces a 140 bp amplicon that extends from 21 bp downstream of the SELEX_OpaBS to 53 bp upstream of the JASPAR_OpaBS. The ChIP values that are reported are percent precipitation relative to input DNA with error bars representing the mean ± S.D. from three technical replicates of the qPCR. The sequences of the primers are summarized in S3 Table and are as follows: Or42b forward: 5’ TCAAGCCGAACCCTCTAAAAT 3’, Or42b reverse: 5’ AACGCCAACAAACAGAAAATG 3’, DESE-Opa forward: 5’ TGCCGTTCGAGTCCTTTATT 3’, DESE-Opa reverse: 5’ CGGAGATCGGAAGGTTAGTG 3”, Alk-OpaBS forward: 5’ TTGTGCGTTTCACCAATCG 3’, Alk-OpaBS reverse: 5’ CGGACTAGCCACATCGAAC 3’.
|
10.1371/journal.pgen.1004618 | HIPPO Pathway Members Restrict SOX2 to the Inner Cell Mass Where It Promotes ICM Fates in the Mouse Blastocyst | Pluripotent epiblast (EPI) cells, present in the inner cell mass (ICM) of the mouse blastocyst, are progenitors of both embryonic stem (ES) cells and the fetus. Discovering how pluripotency genes regulate cell fate decisions in the blastocyst provides a valuable way to understand how pluripotency is normally established. EPI cells are specified by two consecutive cell fate decisions. The first decision segregates ICM from trophectoderm (TE), an extraembryonic cell type. The second decision subdivides ICM into EPI and primitive endoderm (PE), another extraembryonic cell type. Here, we investigate the roles and regulation of the pluripotency gene Sox2 during blastocyst formation. First, we investigate the regulation of Sox2 patterning and show that SOX2 is restricted to ICM progenitors prior to blastocyst formation by members of the HIPPO pathway, independent of CDX2, the TE transcription factor that restricts Oct4 and Nanog to the ICM. Second, we investigate the requirement for Sox2 in cell fate specification during blastocyst formation. We show that neither maternal (M) nor zygotic (Z) Sox2 is required for blastocyst formation, nor for initial expression of the pluripotency genes Oct4 or Nanog in the ICM. Rather, Z Sox2 initially promotes development of the primitive endoderm (PE) non cell-autonomously via FGF4, and then later maintains expression of pluripotency genes in the ICM. The significance of these observations is that 1) ICM and TE genes are spatially patterned in parallel prior to blastocyst formation and 2) both the roles and regulation of Sox2 in the blastocyst are unique compared to other pluripotency factors such as Oct4 or Nanog.
| Pluripotent stem cells can give rise to any cell type in the body, making them an attractive tool for regenerative medicine. Pluripotent stem cells can be derived from the mammalian embryo at the blastocyst stage or they can be created from mature adult cells by reprogramming. During reprogramming, SOX2 helps establish pluripotency, but it is not clear how SOX2 establishes pluripotency in the blastocyst. We evaluated where SOX2 is present, how SOX2 is regulated, and where SOX2 is active during blastocyst formation. Our data show that the roles and the regulation of SOX2 are unique compared to other pluripotency/reprogramming factors, such as OCT4 and NANOG. SOX2 marks pluripotent cells earlier than do other factors, but does not regulate pluripotency until several days later. Rather, the earlier role of SOX2 is to help establish the yolk sac lineage, which is essential for gestation.
| To create and use pluripotent stem cells, it is essential to understand the origins of pluripotency during normal development. During mouse blastocyst formation, pluripotent epiblast (EPI) cells are established by two cell fate decisions that segregate pluripotent progenitors from extraembryonic tissues [1], [2]. During the first cell fate decision, trophectoderm (TE) is segregated from inner cell mass (ICM) prior to blastocyst formation. During the second cell fate decision, the ICM is subdivided into EPI and primitive endoderm (PE) lineages after blastocyst formation. Recent studies have examined the roles and regulation of pluripotency genes, such as Oct4, Nanog, and Sox2, during establishment of EPI cells in the blastocyst [3]–[12], but aspects of the roles and regulation of Sox2 in the blastocyst are unresolved. For example, several studies reported that Sox2 is restricted to the ICM by the blastocyst stage [3], [13]–[15], but the molecular mechanisms regulating Sox2 expression in the blastocyst are unknown.
In addition to the unresolved mechanism by which Sox2 expression is patterned, the functional roles of Sox2 in the blastocyst are not yet clear. ES cells cannot be derived from embryos lacking zygotic (Z) Sox2 [5], indicating that Sox2 is essential for pluripotency. In ES cells, Sox2 is required for the expression of pluripotency genes, such as Oct4 and Nanog, and for the repression of TE genes [16]–[20]. Therefore, Sox2 might be required for initial expression of pluripotency genes and repression of TE genes in the ICM. However, the expression of pluripotency and TE genes in Sox2 Z null blastocysts has not yet been examined at the level of individual cells. Moreover, maternal (M) Sox2 is also thought to participate in blastocyst formation, which could partially compensate for loss of Z Sox2. RNAi knockdown of M and Z Sox2 in the zygote was reported to disrupt blastocyst formation [6]. However, RNAi knockdown embryos do not always phenocopy MZ null embryos [3], [21]. Because understanding the regulation and roles of SOX2 in the blastocyst is key to understanding the molecular regulation of preimplantation development and the establishment of pluripotency, we examined both the mechanisms that pattern SOX2, as well as the functional requirements for MZ Sox2 during development.
Sox2 mRNA is enriched in ICM progenitors starting at the 16-cell stage [14], but the SOX2 protein expression pattern at this stage is unclear, as is the mechanism by which Sox2 is restricted to ICM progenitors. Using immunofluorescence and confocal microscopy, we observed that SOX2 is restricted to nuclei of ICM progenitors at the 16-cell stage and later (Fig. 1A; see Table S1 for wild-type embryo staging scheme). In morulae, a weaker signal was detected in the cytoplasm of outside cells, but this was also detected in embryos lacking MZ Sox2 (Fig. S1A), indicating that the cytoplasmic stain is non-specific. In the early blastocyst (E3.25–E3.5), SOX2 was detected in most ICM cells (Fig. 1A and Fig. S1B), and SOX2 did not colocalize with CDX2 in outside cells (n = 13 embryos; Fig. S1C). By contrast, NANOG and OCT4 are still detected in the TE at this stage (Fig. 1A and 2C) [22], [23]. Therefore SOX2 is a unique, early marker of ICM fate.
Next we examined the mechanism by which SOX2 expression is restricted to ICM. The TE-expressed transcription factor CDX2 restricts the expression of Oct4 and Nanog to the ICM by repressing Oct4 and Nanog expression in the TE after blastocyst formation [22]. We therefore asked whether CDX2 also restricts SOX2 to the ICM. Surprisingly, SOX2 remained restricted to the ICM in Cdx2 null embryos at early and late blastocyst stages (Fig. 1B), indicating that SOX2 expression is restricted to ICM progenitors through a Cdx2-independent mechanism. We therefore investigated whether the pathway that restricts CDX2 to the TE also restricts SOX2 to the ICM in parallel. We previously helped show that TEAD4 partners with YAP and WWTR1 to promote expression of CDX2 and GATA3 in TE cells [7], [24]–[26]. YAP and WWTR1 are localized to nuclei only in TE cells, where LATS kinase activity is lower [26]. We hypothesized that if TEAD4 regulates Sox2 in parallel to Cdx2, then we would detect ectopic SOX2 in the TE cells of Tead4 null embryos. To test this hypothesis, we examined SOX2 expression in Tead4 null embryos. Tead4 is essential for blastocyst formation, but not for polarization of TE cells [24], [25], enabling us to distinguish inside (apolar) and outside (polarized) cells in Tead4 null embryos. We observed that in Tead4 null embryos, ectopic SOX2 was detected in about half of outside cells, in contrast with control littermates, where SOX2 was not detected in outside cells (Fig. 1C, E). Thus, Tead4 is required to restrict SOX2 to ICM cells. This result is significant because although other pluripotency factors, such as OCT4 and NANOG are detected in outside cells of Tead4 null embryos when they die [24], [25], these factors are also detected in the outside cells of wild type embryos at this stage [22]. SOX2 is therefore the first pluripotency factor known to be restricted to ICM progenitors by TEAD4 and not by CDX2.
Finally, we misexpressed Lats2 mRNA in outside cells (Fig. 1D), which is sufficient to shift YAP and WWTR1 localization from nucleus to cytoplasm and downregulate CDX2 in TE cells [26]. As a control, we overexpressed β-Globin in a second group of embryos. As expected, overexpression of Lats2 disrupted nuclear YAP localization and led to ectopic SOX2 in outside cells, while ectopic of β-Globin did not (Fig. 1E, F). Therefore, LATS2 and TEAD4 regulate patterning of SOX2 and CDX2 in parallel, leading to the establishment of their complementary expression patterns in the blastocyst.
Our finding that SOX2 protein is restricted to ICM progenitors conflicted with prior reports suggesting that M SOX2 protein is present in TE cells and required for TE cell development [5], [6]. However, the requirement for MZ SOX2 in the TE has not been functionally evaluated using null alleles. We examined embryos lacking MZ Sox2 using a conditional Sox2 allele [27] and Zp3Cre, which is expressed in the female germ line [28]. We first confirmed that female germ line expression of Cre indeed deleted M Sox2 by quantitative RT-PCR (qPCR) analysis of oocytes from Zp3Cre; Sox2fl/fl or del females (Fig. 2A). Next, we mated these females to wild-type males to determine whether M Sox2 is required for development of their progeny. The number of offspring per litter did not differ significantly between Sox2 germ line-deleted females and control females (Fig. 2B), indicating that M Sox2 is not required for development, consistent with a recent report [29]. While variable levels of Sox2 mRNA have been detected in 1–2 cell embryos [14], we were unable to detect Sox2 mRNA or protein in wild type embryos at 1–2 cell stages, or in Sox2 Z null morulae (Fig. S2A), indicating that M SOX2 is neither present nor functional in the blastocyst.
We next examined whether loss of both M and Z Sox2 disrupts blastocyst formation by breeding Sox2 germ line-deleted females to males carrying a Sox2 null allele. Expression of TE (CDX2 and EOMES) and ICM (OCT4) markers was normal in the absence of MZ Sox2 (Fig. 2C), indicating that neither M nor Z Sox2 is required for TE specification or blastocyst formation, in contrast with the reported RNAi phenotype [6]. Moreover, quantification of the numbers of ICM and total cells in the blastocyst showed that there was no significant reduction in the average numbers of cells contributing to either lineage in the absence of MZ Sox2, compared to other genotypes (Fig. 2D). In ES cells, deletion of Sox2 leads to downregulation of Oct4 and upregulation of TE genes, including Eomes [16]. However, we did not detect ectopic expression of TE genes in ICM cells of Sox2 MZ null blastocysts at early (Fig. 2C) or late (Fig. S2B) time points. We therefore conclude that neither M nor Z Sox2 are required for segregation of ICM and TE lineages.
Our results indicated that Sox2 is not required for the first lineage decision, so we next asked if Sox2 is involved in the second lineage decision during mouse development: the subdivision of the ICM into EPI and PE cell fates. In the blastocyst at E3.75, EPI and PE cells are distributed in a salt-and-pepper fashion within the ICM. EPI cells express higher levels of NANOG, while PE cells express higher levels of SOX17, GATA6, PDGFRA, and GATA4 [4], [30]–[32]. In contrast, OCT4 is detected in both EPI and PE cells at this stage [4], [14], [23], [33]. It is not currently known whether SOX2 is restricted to EPI cells like NANOG, or if SOX2 is expressed throughout the ICM like OCT4. We therefore first examined the SOX2 pattern within the ICM in the E3.75 blastocyst.
Our analysis of early and late blastocysts indicated that the SOX2 expression pattern in the ICM resembles that of NANOG and not OCT4. At E3.75, SOX2 was detected in cells expressing NANOG and not in cells expressing SOX17 (Fig. 3A, B), although downregulation of NANOG in PE cells slightly preceded the downregulation of SOX2 in the PE cells (Fig. 3A, D). By E4.25, SOX2 was detected in EPI and not PE cells (Fig. 3C). These observations suggested that SOX2 and NANOG are restricted to EPI cells by a similar mechanism. NANOG has been shown to be restricted to EPI cells by FGF4/MEK [9], [34]–[37], so we evaluated whether FGF/MEK signaling is necessary and/or sufficient to repress SOX2 in the ICM. We cultured wild-type embryos in FGF4 and HEPARIN (FGF4/HEP) from E2.75 to E4.5, which leads to repression of NANOG, and upregulation of SOX17 in the ICM [3], [34]. Control embryos were cultured alongside in the absence of FGF4/HEP. We observed that FGF4/HEP was sufficient to repress SOX2 and upregulate SOX17 in wild-type embryos (Fig. 3E). Next, we cultured wild-type embryos in inhibitors of FGFR/MEK from E2.75 to E4.0, which leads to ectopic expression of NANOG and repression of SOX17 in the ICM of wild-type embryos [34], [36]. Treatment with FGFR/MEK inhibitors led to ectopic expression of SOX2 and repression of SOX17 throughout the ICM (Fig. 3F). We conclude that SOX2 expression is restricted to EPI cells through an FGFR/MEK-dependent mechanism.
Our results showed that, like NANOG, SOX2 expression is restricted to EPI cells at E3.75 and later. We next asked whether Sox2 is required for the expression of NANOG or for the segregation of EPI and PE cell types at this stage. Prior analysis of Sox2 Z null embryos showed that formation of the ICM is independent of Z Sox2 [5], but the requirement for M Sox2 was not evaluated. To determine whether MZ Sox2 is required for NANOG expression, we evaluated whether the ICM contained normal numbers of NANOG-expressing cells in the absence of MZ Sox2. NANOG was detected at normal levels in the absence of MZ Sox2 at E3.75 (Fig. 4A), and in a normal number of ICM cells in Sox2 M, Z, and MZ null blastocysts at E3.75 (Fig. 4B). These observations indicate that Sox2 is dispensable for regulating the initial expression of NANOG.
By contrast, the number of cells in which we detected the PE marker SOX17 was greatly reduced in Sox2 null blastocysts relative to control embryos at E3.75. In these embryos, the number of unlabeled cells, in which neither NANOG nor SOX17 was detected, was greatly increased in the absence of Sox2 (Fig. 4A, B). Notably, this phenotype was equivalent between Sox2 Z and MZ null embryos (Fig. 4B), consistent with the conclusion that there is no role for M Sox2. We also examined other PE markers and found that the proportion of ICM cells expressing a higher level of GATA6 was significantly reduced in the absence of Sox2 (Fig. 4C, D), and both PDGFRA and GATA4 were detected at much lower levels in the absence of Sox2 (Fig. 4E, F). These results indicate that Sox2 promotes PE gene expression at E3.5–E3.75.
To discover the mechanism by which Sox2 promotes PE development, we next examined the role of Sox2 in regulating Fgf4 expression, since Fgf4 is necessary and sufficient to induce PE gene expression in the ICM [34], [35], [37]. SOX2, together with OCT4, promotes expression of Fgf4 in pluripotent stem cell lines [38], [39], indicating that Sox2 may also promote expression of Fgf4 in the ICM. Consistent with this hypothesis, Fgf4 mRNA was reduced in Sox2 null blastocysts to about 30% of the wild-type level (Fig. 5A). Additionally, we found no requirement for M Sox2 in promoting expression of Fgf4 (Fig. S3A). Thus Z Sox2 is required for maximal expression of Fgf4 in the blastocyst, leading us to ask whether the observed defects in PE gene expression in Sox2 Z null blastocysts are due to reduced expression of Fgf4.
If the disruption in PE gene expression in Sox2 null embryos were due to reduced Fgf4 expression, then exogenous FGF4/HEP should restore PE gene expression. To test this prediction, we cultured Sox2 null embryos in FGF4/HEP from the 8-cell stage (E2.75) to E3.75 (Fig. 5B). As a positive control, we cultured non-mutant embryos in FGF4/HEP, and as a negative control we cultured Sox2 null embryos in the absence of exogenous FGF4/HEP. We first confirmed that embryos of all genotypes and treatment groups were equivalent to the E3.75 developmental stage in terms of cell number (Fig. 5B). Next, we evaluated expression of SOX17 and NANOG in each group. As predicted, FGF4/HEP treatment led to a significant increase in the proportion of SOX17-positive ICM cells in Sox2 null and non-mutant embryos, relative to untreated Sox2 null and non-mutant embryos (Fig. 5C, D). We also cultured Sox2 null and non-mutant embryos in FGF4/HEP for an extended period, after which 100% of ICM cells became SOX17-positive/NANOG-negative, irrespective of genotype (Fig. S3B, C), confirming that Sox2 null embryos respond to exogenous FGF4/HEP like non-mutant embryos. We conclude that Sox2 is not required for ICM cells to receive or respond to FGF4 signaling, but is required for maximal expression of Fgf4.
Our observations that Sox2 promotes PE gene expression via FGF4 predicts that Sox2 promotes PE gene expression non cell-autonomously. We tested this hypothesis by examining expression of PE genes in chimeric embryos containing a Sox2 null PE and wild-type EPI. To generate these chimeras, we aggregated wild-type, YFP-expressing ES cells [40] with precompacted 8-cell Sox2 null or non-mutant host embryos, and then cultured these chimeras to E3.75 (Fig. 5E). Performing the aggregation at this stage allows the ES cells to completely colonize the EPI compartment, such that only PE and TE cells are host-derived [3], [41]. We observed that in Sox2 null host embryos, expression of SOX17 was rescued by wild type ES cells (Fig. 5F, G). These results indicate that Sox2 in EPI cells acts non cell-autonomously to promote expression of SOX17 in PE cells by E3.75.
We next examined whether Sox2 is required to maintain the expression of PE genes after E3.75. Surprisingly, we detected SOX17, GATA6, PDGFRA, and GATA4 in Sox2 null embryos at E4.25 (Fig. 6A). By examining the time course of SOX17 expression in Sox2 null embryos, we determined that SOX17 was detected in a progressively larger proportion of ICM cells in Sox2 null embryos starting from E3.25 until E4.25, when the proportion of SOX17-expressing cells was equivalent to wild type (Fig. 6B). Similarly, the proportion of cells expressing a high level of GATA6 was also normal in Sox2 null embryos at E4.25 (Fig. S4A). These results show that Sox2 is required for the initial, but not the later, expression of SOX17, GATA6, PDGFRA, and GATA4.
We hypothesized that PE gene expression is eventually induced in the cells that were originally unlabeled in Sox2 null embryos at E3.75 (Fig. 4B). Alternatively, rare, correctly specified PE cells may have proliferated to replace the unlabeled cells in Sox2 null embryos. This latter hypothesis predicts that unlabeled cells would undergo apoptosis in Sox2 null embryos to maintain ICM cell number from E3.75 to E4.25 (Fig. S4B). However, we did not observe a difference in the number of apoptotic cells in Sox2 null embryos during this window (Fig. 6C), suggesting that PE gene expression is eventually induced in cells that were originally unlabeled in Sox2 null embryos at E3.75. We hypothesized that the delayed expression of PE genes in Sox2 null embryos is due to the lower level of Fgf4 (Fig. 5A). Consistent with this hypothesis, the expression of SOX17 in Sox2 null embryos at E4.25 was indeed dependent on FGFR/MEK signaling (Fig. 6D), arguing that low levels of FGF4 can eventually induce expression of PE genes in Sox2 null embryos. To determine whether delayed PE gene expression also delayed PE maturation, we examined expression of SOX7, which is expressed only in mature PE cells [42]. At E4.25, SOX7 was detected in Sox2 null embryos (Fig. 6E), suggesting that PE cells had matured in an age-appropriate manner, in spite of the reduced Fgf4. We conclude that Sox2 is not required for maintaining PE gene expression in the blastocyst, consistent with the observation that PE-derived cells are detected in Sox2 null embryos postimplantation [5].
Curiously, we noted that in spite of the normal expression of PE genes in the absence of Sox2, PE cells were often mislocalized in Sox2 null embryos at E4.25. Rather than forming a single, contiguous hypoblast layer between EPI and blastocoel, PE cells were often observed between EPI and polar TE in Sox2 null embryos (Fig. 6A, E). These observations raised the possibility that Sox2 promotes expression of genes thought to regulate PE cohesion and sorting, such as LAMININ and DAB2 [43]–[45], which are normally detectable in the PE at E4.25 [31], [42], [46]. Consistent with this hypothesis, expression of both LAMA1 and DAB2 was reduced in PE cells at E4.25 in Sox2 null blastocysts (Fig. 6F). Notably, expression of DAB2 and PE sorting were rescued by wild type ES cells in Sox2 null E4.25 blastocysts (Fig. S4C), and expression of DAB2 was eventually restored in implantation-delayed blastocysts (Fig. S4D), consistent with SOX2 acting non cell-autonomously to promote the initiation, but not the maintenance, of PE gene expression.
Finally, we examined the status of the EPI in Sox2 null embryos around the time of implantation, because EPI cells are not detected in Sox2 null embryos postimplantation [5]. In Sox2 null embryos at E4.25, expression of the EPI marker PECAM1 [14], [47] was undetectable (4/5 Sox2 null embryos) or reduced (1/5 Sox2 null embryos), expression of OCT4 was undetectable (1/5 Sox2 null embryos) or reduced (3/5 Sox2 null embryos), and expression of NANOG was also undetectable (1/1 Sox2 null embryos) (Fig. 6G). These observations indicate that although Sox2 is dispensable for the initiation of EPI gene expression, Sox2 is required to maintain EPI gene expression. To evaluate the role of Sox2 in the EPI at later developmental stages, we prolonged preimplantation by inducing diapause. By two to four days of delayed implantation, the number of presumptive EPI cells was reduced in Sox2 null embryos relative to wild type, while the number of PE cells was largely maintained until the latest time point (Fig. S4D). We conclude that Sox2 is required for maintaining EPI cell fate at E4.25 and thereafter.
Here we have examined the roles and regulation of SOX2 in the preimplantation embryo, with the goal of deepening our understanding of the origins of pluripotency during development. We showed that SOX2 is a unique, early marker of ICM progenitors, consistent with the reported early enrichment of Sox2 mRNA in ICM progenitors [14]. However, the significance of this early expression is not obvious, since the cell-autonomous role for Sox2 in regulating cell fate does not become apparent until late blastocyst stage. It is possible that Sox2 is initially genetically redundant with other pluripotency factors, such as Oct4 or Nanog. Although phenotypes resulting from disruption of multiple pluripotency genes have not yet been reported in mice, there is evidence of genetic redundancy among zebrafish Oct/Sox/Nanog orthologues [48]. Genetic redundancy between these factors is consistent with our observations that Fgf4 expression is reduced, but not eliminated, in the absence of either Sox2 or Oct4 [3]. Thus, in the embryo, OCT4 and SOX2 may promote expression of Fgf4, and possibly other targets, synergistically, as has been demonstrated in pluripotent stem cell lines [38], [39].
Our evidence suggests that SOX2 and CDX2 are patterned by HIPPO pathway components in parallel (Fig. 7A), but it is not yet clear whether SOX2 and CDX2 are regulated by HIPPO pathway components in the same way. In the TE, expression of CDX2 is activated by a YAP/WWTR1/TEAD complex [24]–[26]. Here we showed that TEAD4 represses expression of SOX2, but we do not yet know if TEAD4 regulates expression of Sox2 directly or indirectly. In ES cells, the YAP/TEAD complex been shown to bind upstream of Sox2 to promote its expression [49], arguing that TEAD4 could, in principle, work together with a transcriptional repressor to repress expression of Sox2 in TE cells. It will be exciting to address this hypothesis in future studies in addition to examining whether position, polarity, and/or contact regulate Sox2 expression, as has been shown for Cdx2 [26], [50]–[54]. Interestingly, the HIPPO pathway can be activated in a position-independent manner in blastomeres [53], [55], raising the possibility that multiple upstream inputs could regulate expression of genes such as Sox2 and Cdx2 in the embryo. Finally, our observations are also consistent with LATS regulating the activity of an as-yet unidentified transcription factor that promotes expression of SOX2 in ICM progenitors. This hypothesis is also supported by qPCR evidence that Lats1/2 maintains expression of Sox2 in the blastocyst [56]. Identification of LATS targets in the preimplantation embryo will therefore provide exciting new inroads into understanding the origins of pluripotency.
Our study provides insight into the regulation of extraembryonic cell types during preimplantation development. In terms of the TE lineage, we showed that SOX2 is not detected in TE cells during preimplantation, nor is it required for their specification. These observations suggest that expression of Sox2 is activated de novo in the trophoblast lineage postimplantation, where it promotes trophoblast development [5], and raise the possibility that HIPPO pathway components participate in regulation of Sox2 expression postimplantation as well. Investigation of the mechanisms by which Sox2 expression becomes activated in the extraembryonic ectoderm is an exciting opportunity to learn about the origins of trophoblast stem (TS) cells, which are derived from extraembryonic ectoderm [57], and dependent on Sox2 and Tead4 [15], [26].
While SOX2 does not activate TE gene expression, SOX2 also does not appear to repress TE gene expression in the ICM. This is in contrast to OCT4, which is known to repress expression of TE genes in the ICM and in ES cell lines [3], [7], [10], [58], [59]. Moreover, we have shown that SOX2 is neither expressed nor functional in PE cells at the time when OCT4 represses TE gene expression in PE cells [3]. These observations support the idea that SOX2 and OCT4 have important non-overlapping roles in the embryo and stem cell lines [15], [60], in addition to their widely appreciated overlapping functions. Thus in the ICM, OCT4 may act alone or with partners other than SOX2 to repress transcription of TE genes in EPI cells.
Our analysis led us to explore the genetic regulation of PE specification as well. We have shown that in the ICM, SOX2 becomes expressed in a salt-and-pepper fashion, similar to NANOG. We have also shown that the salt-and-pepper distribution of SOX2 in the ICM depends on FGFR/MEK signaling, but it is not yet clear whether FGFR/MEK signaling regulates SOX2 expression directly, or whether FGFR/MEK signaling maintains cell fate, which in turn regulates SOX2 expression. Alternatively, NANOG or GATA6 could help regulate SOX2 expression in the ICM. While the expression pattern of SOX2 in Nanog null embryos has not yet been reported, in Gata6 null embryos, SOX2 is expressed in all ICM cells [61], suggesting that GATA6 helps mediate FGFR/MEK signaling to repress SOX2 in the E3.75 ICM (Fig. 7B). Further studies of SOX2 in Gata6 and Nanog null embryos, with and without manipulations to the FGFR/MEK signaling pathway will help clarify the direct and indirect mechanisms regulating Sox2 expression in the ICM.
We also showed that SOX2 helps to maintain the appropriate level of FGF4 that is essential for timely creation of the hypoblast layer. Curiously, Sox2 null embryos do not completely phenocopy Fgf4 null embryos, since NANOG was not upregulated in Sox2 null embryos as it is in Fgf4 null embryos [35]. We hypothesize that the intermediate level of Fgf4 in Sox2 null embryos are sufficient to repress NANOG, as we have shown for Oct4 null embryos [3]. In addition, we did not observe reduced total cell number in Sox2 null embryos as has been observed in Fgf4 null embryos [35], arguing that a moderate level of FGF4 can maintain cell proliferation during preimplantation. Finally, the moderate level of FGF4 produced by Sox2 null embryos is eventually able to restore expression of PE genes (Fig 7C), which does not occur in embryos completely lacking Fgf4 [35], [37], or downstream signaling components [4]. Interestingly, the timing of PE gene expression has also been shown to be sensitive to dose of Gata6 [61], suggesting Fgf4 may be regulated by GATA6 as well. By contrast, PE gene expression is not eventually restored in Oct4 null or in Nanog null embryos [3], [9]. In light of evidence that Oct4 is not required for later expression of PE genes [59], our observations predict that Fgf4 levels, or levels of an as-yet unidentified, later-acting signal essential for maintaining PE gene expression, may be more rapidly and/or dramatically lost in Oct4 and possibly Nanog null embryos, than in Sox2 null embryos.
Our observation that SOX2 is one of the earliest known unique markers of ICM progenitors is supported by the observation that SOX2 is also one of the first pluripotency genes to localize to ICM progenitors in the morula in other mammals [62]. While Sox2 is not initially required for expression of pluripotency genes in the mouse, Sox2 eventually does promote expression of pluripotency genes in the ICM, as in ES cells. Thus, in the ICM, the role of Sox2 appears to be to maintain, and not to initiate, pluripotency (Fig. 7C). This idea is consistent with observations that pluripotency genes are initially normal in Oct4 and Nanog null embryos [3], [12], [33]. The idea of a later role for Sox2 in maintaining expression of pluripotency genes in the ICM is also consistent with evidence that both the genetic regulation and the transcriptional profile of ES cells are more similar to late than to early EPI cells [7], [63]. Recent studies have shown that HIPPO and IL6/JAK/LIF/STAT3 pathways also maintain expression of pluripotency genes in the ICM around implantation stage [50]–[52], [64]. Discovering the mechanisms of crosstalk between pluripotency pathway members at the implantation stage and shortly thereafter will therefore provide exciting new insight into the origins of pluripotent stem cells.
All animal research was conducted in accordance with the guidelines of the University of California Santa Cruz Institutional Animal Care and Use Committee or by RIKEN CDB and Kumamoto University. The following alleles or transgenes were used in this study: Sox2tm1.1Lan [27], Tg(Zp3-cre)93Knw [28], Tead4tm1Hssk [24], and Cdx2tm1.1Aral [21]. Mice carrying the Sox2 null allele (Sox2del+) were generated by crossing mice carrying Sox2tm1.1Lan with 129-Alpltm1(cre)Nagy [65].
As described previously [3], [21], mice were maintained on a 12-hour light/dark cycle. Embryos were collected from timed natural matings by flushing dissected oviducts or uteri with M2 medium. Cultured embryos were cultured in KSOM (Millipore) alone, or KSOM with a final concentration of 1 µg/ml recombinant human FGF4 (R&D Systems) and 1 U/mL Heparin (Sigma), or 100 nM PD173074 and 500 nM PD0325901 (Stemgent) at 37°C and 6% CO2. Microinjection of mRNA was performed as described [26], [50]. To delay implantation, diapause was induced as previously described [32], [66], [67].
Embryos were fixed, stained, imaged, and recovered for genotyping as previously described [68]. Primary and secondary antibody sources were described previously [3], [50], and also included rabbit anti-EOMES (Abcam), rabbit anti-DAB2 (Santa Cruz Biotech), rabbit anti-LAMA1 (Sigma), and rat anti-PECAM1 (BD Biosciences).
Chimeras were performed using Sox2 MZ null embryos as previously described [3]. Chimeras were subsequently genotyped by PCR using primers that distinguished wild-type, floxed, and deleted Sox2 alleles [27].
RNA isolation and single blastocyst qPCR was performed as previously described [21]. qPCR primers included Sox2 (GCGGAGTGGAAACTTTTGTCC and CGGGAAGCGTGTACTTATCCTT), Fgf4 (AGCAGGGGCAAGCTCTTC and GGGTACGCGTAGGATTCG), Oct4 (AGCTGCTGAAGCAGAAGAGG and AGATGGTGGTCTGGCTGAAC), and Actb (CTGAACCCTAAGGCCAACC and CCAGAGGCATACAGGGACAG).
|
10.1371/journal.pgen.1000129 | A Precisely Regulated Gene Expression Cassette Potently Modulates Metastasis and Survival in Multiple Solid Cancers | Successful tumor development and progression involves the complex interplay of both pro- and anti-oncogenic signaling pathways. Genetic components balancing these opposing activities are likely to require tight regulation, because even subtle alterations in their expression may disrupt this balance with major consequences for various cancer-associated phenotypes. Here, we describe a cassette of cancer-specific genes exhibiting precise transcriptional control in solid tumors. Mining a database of tumor gene expression profiles from six different tissues, we identified 48 genes exhibiting highly restricted levels of gene expression variation in tumors (n = 270) compared to nonmalignant tissues (n = 71). Comprising genes linked to multiple cancer-related pathways, the restricted expression of this “Poised Gene Cassette” (PGC) was robustly validated across 11 independent cohorts of ∼1,300 samples from multiple cancer types. In three separate experimental models, subtle alterations in PGC expression were consistently associated with significant differences in metastatic and invasive potential. We functionally confirmed this association in siRNA knockdown experiments of five PGC genes (p53CSV, MAP3K11, MTCH2, CPSF6, and SKIP), which either directly enhanced the invasive capacities or inhibited the proliferation of AGS cancer cells. In primary tumors, similar subtle alterations in PGC expression were also repeatedly associated with clinical outcome in multiple cohorts. Taken collectively, these findings support the existence of a common set of precisely controlled genes in solid tumors. Since inducing small activity changes in these genes may prove sufficient to potently influence various tumor phenotypes such as metastasis, targeting such precisely regulated genes may represent a promising avenue for novel anti-cancer therapies.
| Successful carcinogenesis involves the integration of both pro- and anti-oncogenic pathways. We postulated that genes critical for balancing these opposing pathways are likely to be precisely controlled in tumors, since even subtle alterations in their activity might cause substantial alterations in tumor growth and survival. Using a novel genomic approach, we identified a 48-gene “Poised Gene Cassette” (PGC) showing tight regulation specifically in human cancers but not in corresponding nonmalignant tissues. We show, using a wide variety of in vitro and in vivo approaches, that small alterations in PGC expression are consistently associated with significant differences in experimental metastasis and patient survival, and we demonstrate a direct functional role for five PGC genes (p53CSV, MAP3K11, MTCH2, CPSF6 and SKIP) in cancer invasion. Our findings support the existence of a novel class of ultrasensitive genes that may regulate various cancer-associated phenotypes such as metastasis. Such precisely controlled genes could represent appealing drug targets, since even partial alterations in their activity should prove sufficient to induce potent effects on tumors. Besides cancer, our analytical approach is quite generalizable and likely to be applicable to other disease conditions.
| The accurate processing and integration of multiple external signals is a common feature of biological networks in normal health and complex disease. As illustrated by the examples of oxygen handling [1], energy control [2], and ion homeostasis [3], such accuracy frequently involves the precise coordination of multiple cellular pathways, and mechanisms for regulating and balancing opposing activities. In cancer networks, many similar requirements for pathway balance are likewise found as successful tumorigenesis requires the robust integration of both pro- and anti-oncogenic pathways controlling cellular proliferation, apoptosis, motility, adhesion and senescence [4],[5]. The importance of balancing opposing activities in cancer is illustrated by genes such as HEF1 (NEDD9), a metastasis-related gene [6] and HMMR, a gene involved in centrosome formation (Pujana et al. 2007). Either repression or overexpression of HEF1 can cause mitotic defects [7],[8], indicating that its activity in tumors requires tight regulation. Similarly, subtle alterations of HMMR expression in normal mammary tissues may promote breast tumorigenesis, underscoring the need to keep the HMMR gene tightly regulated [9]. Such findings support the notion that balancing the activity of positive and negative effectors is likely to be a central requirement of many cancers.
At the systems-level, pathway balance is often facilitated through the use of network structures [10] conveying robustness to random fluctuations and errors [11]–[14]. However, the pivotal balancing role played by certain genetic components may at least partially explain why some networks also exhibit ultrasensitivity – a phenomena where small changes in activity at specific components can suffice to elicit qualitative changes in output [12],[13]. Ultrasensitivity may contribute to a network's ability to rapidly respond to changing environmental and genetic conditions [15],[16]. Intriguingly, there is emerging evidence that certain cancers can also display ultrasensitivity. Some remarkable examples include the dramatic responses of chronic lymphocytic leukemia cells to colchicines, occurring at concentrations 10,000-fold lower than that required for similar effects on normal lymphocytes [17],[18], and the striking clinical responses of certain solid tumors to targeted pathway inhibitors [19]. From a therapeutic perspective, such ultrasensitive components could prove particularly appealing as drug targets, as even small alterations might prove sufficient to induce potent effects on tumor phenotypes such as tissue invasion and metastasis. However, our current understanding of the role that ultrasensitivity plays in cancer is still far from complete. Identifying additional genetic components regulating pathway balance in tumors might thus improve our ability to target critical control nodes in cancer networks.
As a general strategy to identify ultrasensitive components in tumors, we hypothesized that a) such components should be precisely regulated and thus exhibit restricted levels of expression variation in cancers; and b) subtle alterations in the expression levels of these components should induce or be associated with significant phenotypic changes. We then applied these criteria to determine if such precisely-regulated genes might be inferred from databases of tumor gene expression profiles. While several groups have compared the expression profiles of multiple tumor and non-malignant tissues [20],[21], to our knowledge, no study to date has systematically attempted to investigate the issue of precise gene regulation in tumors. Employing a genome-wide computational strategy, we identified and robustly validated a novel “Poised Gene Cassette” (PGC) of genes undergoing precise regulation in a microarray database of human tumors from diverse tissue types. Furthermore, subtle alterations in PGC expression were associated with significant and measurable alterations in important tumor phenotypes such as experimental metastasis and patient survival. Our results thus suggest the existence of a generalized homeostatic mechanism in solid tumors for maintaining precise levels of PGC transcription, which may be important for various cancer-associated phenotypes, such as tissue invasion and metastasis. Importantly, the approach described in this study is quite generalizable and can be applied to other diseases.
We hypothesized that genes precisely regulated in cancer should exhibit a highly restricted level of gene expression variation across a large database of individual tumor gene expression profiles. To investigate this, we generated gene expression profiles for 270 primary tumors from six tissue types (breast, colon, liver, lung, oesophageal and thyroid) using Affymetrix U133A Genechips. For every gene, we computed gene expression coefficient of variances (CV), where genes with small CVs are considered more tightly regulated than genes with large CVs. We focused on the top 15% most tightly-regulated genes in tumors, corresponding to an empirical CV cut-off of 0.28. To identify genes whose tight regulation was tumor-specific, we used a second database of 71 adjacent matched non-malignant tissues (“control” tissues) to eliminate from this 15% genes that were also tightly regulated in non-malignant samples (CV>0.3). The use of an absolute CV threshold is permissible, as the global distribution of expression CVs between tumors and controls were highly similar (mean CVs were 0.46 and 0.45 for tumors and controls) (Figure 1A). Using this criterion, we identified a “Poised Gene Cassette” (PGC) of 48 genes exhibiting highly restricted levels of expression variation in tumors (Figure 1B). The F-test, a statistical method for comparing the variation of different data sets, confirmed that each of the 48 PGC genes was indeed associated with significantly decreased expression variation in tumors relative to controls (one tailed F-test, p = 0.0001 to 4×10−14). We also varied the CV threshold between 0.26–0.3 (+/−7%) and repeated the analysis. Similar results were obtained (Table S1), indicating that the identification of PGC is not dependent on a particular CV threshold.
We investigated whether the reduced expression variation of the PGC might be due to technical features of the Affymetrix platform or the composition of the initial training set. We considered the possibility that the reduced variance of the PGC might be due to an overabundance of ‘poor quality’ probes, which might be expected to cross hybridize with multiple genes and hence generate higher background signals [22]. However, an examination of a vendor provided list of questionable probes (i.e., with ‘_s_at’ and ‘_x_at’ suffixes), confirmed that the PGC was not significantly enriched in poor quality probes (p = 0.4). In addition, a comparison of the PGC genes against an in-house curated list of unreliable array probes based on sequence redundancy and repeat mapping [23] confirmed that unreliable probes were not overrepresented in the set of PGC genes (p = 0.8).
To investigate the influence of normalization protocol on PGC discovery, we re-processed the training set using a different normalization method (RMA, [24],[25]). In the RMA-normalized data, we found that 90% of the original PGC genes still exhibited decreased expression variation in tumors relative to controls (i.e., CV(control)>CV(tumors)) (Figure 2A). Thus, the tumor-specific restricted expression variation of the PGC does not appear to be dependent upon a specific normalization technique.
The reduced variation of the PGC is also not due to an overrepresentation of either high-expressing or low-expressing genes. As shown in Figure 2B, the PGC genes were equally distributed across a wide range of expression levels and not confined to either low or highly expressing genes in tumors or control tissues. Thus, the reduced expression variation of the PGC in cancers is unlikely to be due to the PGC genes simply being either highly expressed, rendering the PGC distinct from some studies suggesting an inverse correlation between expression variation and absolute expression levels [26]. Similarly, the PGC is also not biased in lowly expressed genes, consistent with our original selection criteria requiring these genes to be reliably detected in the majority of samples (see Methods). It is also important to note that the PGC genes do not exhibit significant differences in their absolute mean expression levels between cancers and normal tissues (Figures 1B and 2B), but instead only differ in their levels of expression variation between cancers and normal tissues. This observation, as well as others, also provides an argument that the PGC genes are unlikely to represent tissue-specific expression (see Discussion).
The discovery of the PGC is also not influenced by the overrepresentation of breast tumors in our initial training set (breast tumors comprised 68% of the training set). Specifically, we removed all the breast tissues and repeated the PGC analysis. Even without inclusion of breast tissues, 83% (40/48) of the PGC genes still exhibited reduced variation in tumors compared to controls. Of 47 genes exhibiting tumor-specific tight regulation in the breast-excluded data (CV<0.28), 24 genes were part of the original PGC, an overlap far beyond random chance (50%, p = 1.3E-11, hypergeometric test). Taken collectively, these results suggest that the identification of the PGC, and its restricted expression variation in cancers, is unlikely to be due to a technical artifact or the inclusion of a specific cancer type.
To confirm that the restricted expression variation of the PGC was specifically associated with malignancy, we determined the frequency at which a member gene of the PGC could be re-identified in a series of class-permutation tests. When the class labels of the samples (i.e., tumor or control) were shuffled to generate a series of 1000 permuted sets, almost all the PGC genes (46/48, 96%) could only be re-identified in less than 5% of the class-permuted signatures, consistent with the decreased expression variation of the PGC being tightly associated with tumor samples.
We then evaluated the robustness of the PGC by repeated random sampling (RSS), a stringent cross-validation strategy [27]. The original training set was randomly divided 1000 times into two parts, generating a large series of distinct training/test set combinations. For each of the 1000 derived RSS training sets, we identified new PGC signatures (rPGC) and compared them to the original PGC gene set. Following the guidelines of Michels et al [27], 20 genes were repeatedly selected in more than half of the 1,000 new rPGC signatures. Of these 20 genes, 19 (95%) are members of the original 48-gene PGC (Figure 2C) – the observation that only one gene not part of the original PGC signature was repeatedly selected in the RSS assay indicates that a substantial proportion of the PGC signature (40%) is robust to training set selection. To evaluate the transportability of the PGC signatures, we then applied each of the 1000 rPGC signatures to their cognate test sets. In anticipation that most independent test sets are likely to contain either tumor or control samples but not both, we considered the tumors and controls separately from one another in this analysis. In each test set, we checked if the population of tightly regulated genes, defined using the original CVT threshold (0.28), contained a significant enrichment of rPGC genes (see Methods). The rPGC signatures were significantly enriched in the population of tightly regulated genes in 80% of the tumor test sets (PGC→T, Figure 2D), and importantly were NOT significantly enriched in 100% of the control test sets (PGC→N, Figure 2D), indicating that the PGC is robust in recapitulating its precise regulation in multiple tumor data sets, but not data sets of non-malignant samples. Together, these results confirm the specificity of the PGC for tumors.
We then asked if the precision of PGC regulation in cancer could be observed in independent data sets of diverse tumors. We collected nine independent cancer cohorts, comprising in total 1105 cancer samples from >7 primary tissue types [28]–[32], including I) four tissues not represented in the original training data (gliomas, gastric, NPC, and ovarian), II) one data set (Yu_Gastric&NPC) representing a mix of two different tissues, and III) a collection of cancer cell lines (NCI60) from nine different tissues. A summary of these nine data sets can be found in Table S2 and the corresponding references. Using a similar strategy to the RSS test sets, a significant fraction of the PGC genes were tightly regulated in all nine primary tumor data sets (p-value range: 0–0.002) (Table 1A), confirming the existence of the PGC in a wide variety of solid tumors. In total, 19 out of 48 PGC genes repeatedly exhibited reduced expression variation in more than half of the 9 cancer test sets (Table S3). We also performed the reciprocal experiment and evaluated the regulation of the PGC in a series of independent non-malignant samples. Although such datasets are rarer in their availability and typically smaller than cancer datasets, we collected two distinct cohorts comprising 115 normal tissues from various organs [33],[34]. Notably, these non-malignant samples were obtained from healthy donors, and are thus free of malignancy and representative of true normal samples. In stark contrast to the cancer data sets, the PGC genes exhibited either no or only a marginal degree of tight regulation in the normal data sets (p = 0.07 and 0.01; Table 1B). Thus, these results indicate that the precise regulation of the PGC genes is largely restricted to cancer tissues, suggesting that diverse tumor types may harbor a general requirement for tightly regulating PGC expression.
A pathway analysis revealed multiple highly significant interactions between the PGC genes and prevalent tumorigenic pathways. The top-scoring molecular network for the PGC comprised 11 PGC focus genes interacting either directly or indirectly with the well-known cancer-related transcription factors Myc and TP53 (p = 10−19, see Methods) (Figure S1), and the most significantly enriched cellular functions in the PGC were cancer (p<0.0045), tumor morphology (p<0.0045) and cell cycle control (p<0.0045). The PGC was also significantly enriched in components related to integrin signaling (p = 2.33E−04; Figure S1), a complex signaling pathway implicated in both positive and negative regulation of tumor cell growth and cancer metastasis. Besides integrin signaling, other individual PGC genes, such as RPS2 and RPL7A, have also been previously implicated in the control of cellular transformation, tumor growth, aggressiveness, and metastasis [35],[36]; while the PGC gene MUS81 has recently been reported to interact with p53 to maintain genome stability [37]. Thus, an array of biological and functional evidences suggest that the PGC genes are likely to be involved in the activity of multiple cancer-related pathways, and not ubiquitous ‘housekeeping’ cellular functions. The full list of PGC genes is provided in Table S3.
The tightness of PGC regulation in tumors might be explained if small alterations in the expression levels of these components are sufficient to cause significant phenotypic changes in tumors. We employed three experimental assays to address this possibility. First, we analyzed a set of colon cancer cell lines derived from either primary tumors or distant metastases from the same patient (SW480 and SW620), which have been shown to exhibit several phenotypic differences including metastatic potential [38],[39]. Using Gene Set Enrichment Analysis (GSEA, [40]), we found that PGC expression was subtly yet significantly decreased in highly metastatic SW620 cells compared to poorly metastatic SW480 cells (p<0.001, Table S4). Second, we then analyzed patterns of PGC expression in a cohort of 30 breast cancer cell lines, where the invasive capacity of each line had been previously measured by matrigel assays [32]. The PGC genes exhibited minimal expression variation across the lines when assessed using a standard range of expression variation, consistent with their being tightly regulated in cancers (Figure 3A, left heat-map). However, when the scale of variation was amplified, we identified by hierarchical clustering two groups of cell lines showing either subtly higher or lower levels of PGC expression (Figure 3A, right heat-map). Importantly, we again found that the majority of cell lines with high to moderate invasive abilities exhibited subtle yet significant decreased expression of the PGC genes compared to poorly invasive lines (p = 0.04, chi-square test, sample groups defined on the basis of the top-level branch point). To validate the robustness of this clustering by an alternative method, we then also performed independent k-means clustering (k = 2). Using k-means, 7 out of 8 highly invasive cell lines were clustered into one group together with 4 marginally or non-invasive cell lines (p = 0.01, chi-square test for high vs. marginal/non-invasive), consistent with the groupings observed by hierarchical clustering. Third, we conducted in vivo experiments using a murine xenograft model of metastasis, where poorly metastatic HCT116 colon cancer cells were injected into the spleens of nude mice, and metastatic liver tumor nodules were harvested 6 to 8 weeks later. The liver nodules were expanded in culture and re-passaged in mice to generate a panel of lines (M1, M2, and M3) with increasing levels of metastatic capacity (Figure 3B). Examining the gene expression profiles of these lines, we found that highly metastatic cells once again exhibited subtly decreased PGC expression compared to poorly metastatic HCT116 cells (p = 0.03, Figure 3B and Table S4). These results, based on three different experimental models of metastasis, collectively suggest that small alterations in PGC expression in tumors may be associated with potent differences in tumor physiology, specifically metastatic and invasive capacity.
To directly demonstrate the functional role of PGC genes in cellular invasion, we performed siRNA experiments where five PGC genes (p53CSV, MAP3K11, MTCH2, CPSF6, SKIP) were silenced in poorly-metastatic AGS gastric cancer cells. While p53CSV is a gene required for p53-mediated cell survival [41], its role in cancer is otherwise poorly understood. Furthermore, associations between MAP3K11, MTCH2 and CPSF6 to cancer have also not been previously reported. The siRNA treatments reduced the expression levels of these five PGC genes from 45%–80%, as assessed by quantitative real-time PCR (Figure 4A), and reductions in p53CSV, MAP3K11, MTCH2 and CPSF6 resulted in a significant enhancement of in vitro invasive activity as measured in a matrigel assay (p<0.01, one-tailed t-test, Figures 4B and 4C). Furthermore, SKIP siRNA treatment resulted in a significant inhibition of cellular proliferation in AGS cells (p<0.01, Figure 4D). It is worth noting that for at least two genes (p53CSV and CPSF6), a partial reduction of gene expression of 45–60% was able to trigger a significant change in invasive phenotype. To further demonstrate the generality of this phenomenon, we then knocked down p53CSV in another poorly-metastatic colon cancer cell line, HCT116 which we previously utilized in the xenograft assay. Again, the partial silencing of p53CSV expression significantly increased the invasion activity of HCT116 cells (Figure S3). These results suggest that the PGC genes may play roles in regulating cancer invasion and metastasis.
To extend the potential role of precise PGC regulation to the clinical context, we asked if similar small changes in PGC expression might be associated with significant differences in patient survival and clinical outcome. We employed hierarchical clustering to group the tumors in each of the six data sets with survival data available by their overall level of PGC expression. A representative example is shown in Figure 5A. Once again, the PGC genes exhibited minimal expression variation across the tumors when assessed on a standard scale of expression variation, consistent with their being tightly regulated in tumors (Figure 5A, left heat-map). However, when the variation scale was amplified, we identified two groups of tumors showing either subtly higher or lower levels of PGC expression (Figure 5A, right heat-map). Remarkably, a Kaplan-Meier survival analysis revealed that in all six data sets, patients with tumors expressing PGC levels below the population average experienced significantly worse survival outcomes compared to patients with high-PGC expressing tumors (Figure 5B; all cases p<0.05 except in ovarian cancer set where p = 0.057, see Figure S4 for clustering groupings). We only observed comparable survival stratifications across the six data sets in 46 out of 10,000 randomly selected 48-member gene sets, arguing that the prognostic ability of the PGC is statistically unique. In a multivariate analysis, PGC expression behaved as an independent prognostic factor compared to other clinical variables in the breast and colon cancer cohorts, and was associated with tumor stage in ovarian, lung and glioma cancer patients (Table S5).
Importantly, the PGC exhibits very little overlap with other expression signatures reported to predict clinical behavior in multiple tumor types. A comparison of the PGC against a 128-gene metastasis signature (MS) [42], a 70-gene chromosomal instability signature (CIN70) [43], a cell cycle module [44], a wound response healing signature [45],[46], and multiple cell proliferation-related signatures (57–59) including a 874-gene cell cycle gene signature (CPS) [47], revealed that there was no direct overlap in gene content between the PGC and these other “multi-tumor” gene signatures, except for a one-gene overlap with the CIN70, and a four-gene overlap with the CPS, which was not statistically significant. This finding suggests that the specific gene content of the PGC is distinct from other previously described signatures. To ask if the PGC might target the same “poor prognosis” tumors as other published signatures capable of predicting clinical outcome in multiple tumor types, we then investigated the ability of the MS, CIN70, and CPS to stratify patient survival in the six data sets - none of these signatures exhibited comparable prognostic significance to the PGC across the six patient cohorts (data not shown). These observations suggest that the PGC is likely to target different molecular features and types of tumors than the aforementioned signatures.
Previous gene expression studies comparing tumors and non-malignant tissues have typically employed microarray analysis algorithms such as t-tests with false positive correction or SAM [20]. Genes detected by such techniques typically require both differing mean expression levels and equivalent levels of variation between two cellular states (Figure S5). However, the PGC might not be detected by such conventional techniques, as PGC genes might not exhibit distinct mean expression levels between the two groups and only be associated with differing degrees of expression variation between tumors and controls (Figure S5). Indeed, performing SAM and t-tests on the training set only identified 27% of the original PGC, after multiple hypothesis correction, and the absolute mean expression levels of many PGC genes between tumors or non-malignant tissues were highly similar (Figure S5). To ask if the unequal distributions in expression variation might underlie the failure of the PGC genes to be identified by conventional techniques, we also analyzed the original training data set using Welch's test, an adaptation of Student's t-test intended for use with two groups having unequal variance. Again, 75% of the PGC genes failed to be detected as significant using Welch's test (data not shown). These findings suggest that conventional algorithms would likely have failed to detect the PGC, thereby providing a partial explanation as to why the PGC might have been missed in previous studies.
In this study, we identified a novel cassette of genes exhibiting tumor-specific precise regulation in multiple cancer tissues. Our ability to discern the PGC was facilitated by the use of an analysis method focused on expression variance rather than expression levels. The reduced variance of the PGC in tumors is unlikely to be a technical artifact of the Affymetrix platform, as it was not related to probe selection, data normalization, absolute high or low expression levels in either tumors or non-malignant tissues, or sample set. Using both rigorous cross-validation (RSS) and multiple independent validations, we found the PGC to be robust to alterations in training set composition and repeatedly observed in diverse malignant tumor types, including several tissue types not present in the original training data. Importantly, the PGC failed to demonstrate tight regulation in several non-malignant tissue data sets, arguing that its control is cancer-specific. Interestingly, even though it was not a specific requirement in our initial analysis, the majority of PGC genes exhibited similar mean expression levels in both tumors and non-malignant tissues. This absence of a distinct difference in mean expression values resulted in the failure of standard microarray analysis methods (e.g., t-test) to detect the majority of PGC genes when applied to the same training data set. Furthermore, a standard practice in microarray data processing is to filter out genes exhibiting low variation prior to clustering or statistical analysis - such filtering would inevitably lead to a bias towards differentially expressed genes and prevent the discovery of the PGC.
One potential concern might be that the PGC genes simply reflect the activity of tissue-specific gene expression. However, five findings argue against this possibility. First, while dedifferentiated cancer cells frequently exhibit a loss of tissue-specific gene expression (Rhodes et al. 2004); such a loss would typically result in tissue-specific genes being down-regulated in their absolute expression levels compared to normal tissues. In contrast, the PGC genes do not exhibit significant differences in their absolute expression levels between cancers and normal tissues (Figure 2B). Second, the reduced variation of the PGC genes was consistently observed in multiple independent sets from diverse tissues (e.g. gliomas, lung, breast), including a data set (NCC) that combined tissues from two different sources (gastric and NPC tumors). Third, the PGC genes also showed reduced expression variation in the NCI60 test set - a mixture of cancer cell lines from 9 different tissue types. Fourth, the PGC genes consistently exhibited reduced expression variation in the repeated random sampling (RSS) cross-validation assay, where we tested 1000 distinct training set and independent test sets composed of mixed tissue types (Figure 2D). Fifth, even within each of the six tissue types in the training set (liver, colon, esophagus, thyroid, lung, and breast), the majority of the PGC genes (70%) are not differentially expressed within tumors and normals (p>0.01, t-test) (YK, data not shown). Taken collectively, it is unlikely that the consistency of the PGC would have been observed if its reduced expression variation was solely due to tissue-specific expression, supporting the notion that the PGC genes are likely to be distinct from the conventional differentially expressed gene signatures described in most microarray studies.
One possible explanation for why certain genes may require precise control is if they regulate or are involved in balancing disparate downstream pathways possessing mutually opposing activities. In cancers, the successful establishment of a malignant tumor involves multiple pro- and anti-oncogenic forces involved in cell proliferation, apoptosis, cell death, senescence, cell adhesion, and motility, all of which require delicate balance by different genetic components. For example, while loss of Ras signaling is lethal, aberrant signaling through this pathway is important for cancer development but can also drive cells into either senescence or cell proliferation, depending on cellular context [48],[49]. Another good example is the anti-apoptotic gene Akt/PKB (protein kinase B), which when constitutively activated reduced metastases in mice and inhibited the invasion of breast cancer cells [50],[51], indicating its involvement in multiple cancer pathways. Reassuringly, similar examples of balanced coordinator genes are also seen in the cohort of PGC genes. The PGC gene FUS1 (also known as FUS) has been reported as a tumor suppressor gene in lung and breast cancer [52] and a pro-oncogene in leukemia [53]. Oxidative stress, which may play an important role in cancer progression and the regulation of cancer metastasis [54], is dependent upon the critical balance between intracellular hydrogen peroxide H2O2 and superoxide O2−. Two PGC genes - p53CSV and KIAA0247 have been reported to be induced in response to oxidative stress [55], and may influence this balance and the response of tumor cells to apoptotic stimuli [56]. It is also worth noting that the PGC was significantly overrepresented in components of the integrin signaling pathway – a highly complex process involving multiple related family members with roles in many cellular functions, including ERK/MAPK and JNK/SAPK regulated gene expression, cell motility, cytoskeletal interactions, and PI3K and Wnt pathway signaling [57]. In metastasis, integrins are crucial for cell invasion and migration, not only for physically tethering cells to the matrix, but also for sending and receiving molecular signals regulating these processes [57]. Moreover, while some groups have proposed that increased integrin expression could promote malignant behavior by enhancing tissue stiffness [58], other groups have suggested that loss of integrins may promote tumor invasion and metastasis [59]. The complexity of integrin family members and their pathway components also provides a plausible explanation for why even subtle alterations in PGC expression are associated with distinct and measurable changes in metastatic behaviour in both experimental models of metastasis and clinical outcome.
What might be the mechanistic basis of precise PGC regulation? At a general level, many precisely-regulated genes are likely to possess complex regulatory systems for tightly controlling expression levels, to rapidly sense and adapt to dynamic perturbations in both the internal and external environment [60]. Such mechanisms could involve the use of both positive and negative feedback loops, analogous to the circuitry utilized by the LacI/O bacterial system to ensure precise expression [61], but in cancers could also involve eukaryote-specific mechanisms like epigenetic modifications (DNA methylation or chromatin modifications), microRNA regulation, or transcription factor binding. Interestingly, in a preliminary analysis, we attempted to extend our observations from the pathway analysis showing an association of several PGC genes with both Myc and TP53. Specifically, we investigated whole-genome transcription factor binding data for Myc and TP53 [62], and found that the PGC genes were weakly but significantly associated with Myc binding sites under Myc-overexpressed (tumorigenic) conditions (p = 0.04) but not under physiological conditions (p = 0.3) (Table S6). These preliminary results raise the possibility that transcription factor binding, specifically Myc binding, may constitute one possible mechanism for PGC regulation in cancer cells. However, deciphering the mechanism of PGC regulation will undoubtedly require further research.
Cancers have been proposed to possess robustness mechanisms for protection against various therapeutic perturbations and naturally occurring microenviromental (e.g., hypoxia) and immune responses. However, many complex systems have evolved to exhibit a ‘robust yet fragile’ structure [63],[64], and it has been proposed that studying mechanisms of cancer-specific robustness and accompanying fragilities might prove useful for the development of novel targeted therapies [65]–[67]. The PGC gene cassette reported here may indicate such fragilities in the network of tumor cells, as subtle alterations on these components significantly affected the cellular behavior of cancer cells. Beyond cancer, this approach is conceptually applicable and easily transportable to other disease conditions where gene expression data is available. It will be interesting to explore if the approach will also prove informative in identifying genes and pathways with important roles in other human pathophysiologies.
The training data set contained 270 primary human tumors (Lung = 18, Thyroid = 35, Liver = 9, Esophagus = 16, Colon = 9, Breast = 183) and 71 adjacent non-malignant tissues (Lung = 12, Thyroid = 16, Liver = 8, Esophagus = 13, Colon = 9, Breast = 13) obtained from the Tissue Repository of the National Cancer Centre of Singapore (NCCS). The phrase ‘non-malignant’ instead of ‘normal’ was used to describe the control tissues in the training set, as they were also obtained from cancer patients. Institutional approvals were obtained from the NCCS Tissue Repository and Ethics Committees. Descriptions of sample collection protocols, archiving, and histological assessments are presented in the Text S1. RNA was extracted from the tissues using Trizol reagent (Invitrogen, Carlsbad, CA) and processed for microarray hybridizations on Affymetrix U133A Genechips according to the manufacturer's instructions (Affymetrix Inc., Santa Clara, CA). The expression data has been deposited into the Gene Expression Omnibus (GEO) database (GSE5364).
Raw Genechip scans were processed using either the MAS5 algorithm (Affymetrix) normalized by median-centering (GeneData, Basel, Switzerland), or by robust multiple chip analysis (RMA) [24],[25] (see Results). To identify reliably measured genes, we discarded probes with <80% present values (P-call <80%) across the training set samples. For genes with multiple probes, we selected the best-match probes (to targets) represented by a “_at” extension. For genes with multiple “_at” extension probes, the probe with the highest P-call rate (i.e., the highest valid value proportion) was used. The final pre-processed training set comprises 5729 unique genes, each represented by a single probe.
Gene expression CVs (standard deviation divided by the mean expression level) were used to compute the variability of expression for each gene. Based on the global distribution of CVs in the training set, we selected an empirical threshold of CVT = 0.28 below which a gene was considered to be tightly regulated (see Results). Prior to comparing gene CVs between populations, we also confirmed that the global CV distributions for different sample cohorts (i.e., tumor or non-malignant) were similar.
To estimate the probability that the PGC signatures might be generated by chance, we randomly shuffled the class labels (i.e., tumor or non-malignant) of the training set to generate multiple class-permuted sample sets and determined the frequency a particular PGC gene could be re-identified in situations where the sample labels were shuffled. Repeated Random Sampling (RRS), a rigorous cross-validation strategy described in [27], was also used to determine the influence of particular training set compositions on selecting specific signature genes. Detailed descriptions of the class permutation and RSS tests are provided in the Text S1.
The hypergeometric distribution was used to test if the PGC genes were significantly over represented in the population of tightly controlled genes in each test set. First, we identified genes exhibiting tightly controlled expression in the test set, using the CVT threshold cut-off (CV(Test)<CVT). Second, we determined the overlap between the PGC gene signatures and the population of tightly regulated genes in the test set, and the hypergeometric distribution test was used to calculate the significance of the overlap. Significance was defined as p<0.01.
We used Ingenuity Pathway Analysis (IPA, Ingenuity Systems) to identify molecular networks, cellular functions, and signaling pathways associated with the PGC. The various networks were displayed as nodes (genes) and edges (biological relationships between genes). We also used IPA to identify cellular functions and signaling pathways that were significantly enriched in the PGC. The significance of a pathway association is reflected by a Fisher's exact test p-value, indicating the likelihood that the pathway would have been identified by random chance.
AGS gastric cancer cells and HCT116 colon cancer cells were cultured according to American Type Culture Collection (ATCC) recommendations. Cells were transfected with either siRNA pools of representative PGC genes p53CSV, MAP3K11, MTCH2, CPSF6 and SKIP (Dharmacon, Lafayette, CO) or non-targeting siRNA controls at 100 nM concentration, using oligofectamine reagent (Invitrogen) at 0 and 24 hr time points, in 6 well culture plates. Gene silencing was verified by Real time PCR. Invasion assays were performed using Biocoat matrigel invasion chambers (BD Biosciences, Bedford, MA) as recommended by the manufacturer. 48 hrs after siRNA transfection, equal numbers of target gene siRNA transfected cells and non-targeting siRNA transfected cells were placed in the invasion chambers, and after 24 hrs cells that successfully invaded through the matrigel invasion chambers were scored. Each experiment was repeated thrice and the percentage of invasion was calculated by comparing against the non-targeting siRNA transfected cells. To assay cell proliferation, AGS cells transfected with the PGC genes and non-targeting control siRNA in 6 well culture plates were harvested at 96 hrs after siRNA transfection and counted. Experiments were performed thrice.
Total RNA was reverse transcribed using Taqman Reverse Transcription Reagent kit (Applied Biosystems, Foster City, CA) and quantitative PCR was performed using the following Taqman probes: p53CSV (Hs00429934_g1); MAP3K11 (Hs00176759_m1); MTCH2 (Hs00819318_g1); CPSF6 (Hs00199668_m1); SKIP (Hs00273351_m1), on a 7900HT Fast Real time system (Applied Biosystems, Foster City, CA). Taqman GAPDH probes (glyceraldehyde phosphate dehydrogenase) (Hs99999905_m1) were used as internal controls. All samples were run in triplicates.
(A) Colorectal cancer model : this comprises two colon cancer cell lines derived from either primary or distant metastases from the same patient (SW480 and SW620). SW480 and SW620 cells exhibit several phenotypic differences including metastatic potential [38],[39]. Gene Set Enrichment Analysis (GSEA) was performed as described in [40]. (B) Breast cancer panel: this comprises a panel of 51 breast cancer cell lines for which gene expression data is available [32], and where the relative invasive capability of 30 lines has been measured using matrigel assays [32]. (C) Murine assay: this comprises an in vivo passage model where poorly metastatic HCT116 colon cancer cells were injected into mouse spleens, and subsequent hepatic metastases were harvested to generate increasingly metastatic cellular variants. Details of this model are provided in the Text S1. The animal work performed was approved by the National University of Singapore Institutional Animal Care and Use Committee (NUS IACUC). Cells obtained from the hepatic metastatic nodules after the first passage were named M1, and the selection procedure was repeated twice to obtain the M2 and M3 cell lines. Three independent replicates were profiled for each cell line.
Hierarchical clustering (average linkage metric with Pearson correlation) was used to cluster tumors into different groups on the basis of their PGC expression levels. Kaplan-Meier analysis (SPPC, Chicago) was used for survival comparisons between the tumor groups. P-values were calculated using the Log-rank test. |
10.1371/journal.pbio.3000133 | Lens differentiation is controlled by the balance between PDGF and FGF signaling | How multiple receptor tyrosine kinases coordinate cell fate determination is yet to be elucidated. We show here that the receptor for platelet-derived growth factor (PDGF) signaling recruits the p85 subunit of Phosphoinositide 3-kinase (PI3K) to regulate mammalian lens development. Activation of PI3K signaling not only prevents B-cell lymphoma 2 (BCL2)-Associated X (Bax)- and BCL2 Antagonist/Killer (Bak)-mediated apoptosis but also promotes Notch signaling to prevent premature cell differentiation. Reducing PI3K activity destabilizes the Notch intracellular domain, while the constitutive activation of Notch reverses the PI3K deficiency phenotype. In contrast, fibroblast growth factor receptors (FGFRs) recruit Fibroblast Growth Factor Receptor Substrate 2 (Frs2) and Rous sarcoma oncogene (Src) Homology Phosphatase 2 (Shp2) to activate Mitogen-Activated Protein Kinase (MAPK) signaling, which induces the Notch ligand Jagged 1 (Jag1) and promotes cell differentiation. Inactivation of Shp2 restored the proper timing of differentiation in the p85 mutant lens, demonstrating the antagonistic interaction between FGF-induced MAPK and PDGF-induced PI3K signaling. By selective activation of PI3K and MAPK, PDGF and FGF cooperate with and oppose each other to balance progenitor cell maintenance and differentiation.
| A central aim in understanding cell signaling is to decode the cellular logic that underlies the functional specificity of growth factors. Although these factors are known to activate a common set of intracellular pathways, they nevertheless play specific roles in development and physiology. Using lens development in mice as a model, we show that fibroblast growth factor (FGF) and platelet-derived growth factor (PDGF) antagonize each other through their intrinsic biases toward distinct downstream targets. While FGF primarily induces the Ras–Mitogen-Activated Protein Kinase (MAPK) axis to promote lens cell differentiation, PDGF preferentially stimulates Phosphoinositide 3-kinase (PI3K) to enhance Notch signaling, which is necessary for maintaining the lens progenitor cell pool. By revealing the intricate interactions between PDGF, FGF, and Notch, we present a paradigm for how signaling crosstalk enables balanced growth and differentiation in multicellular organisms.
| Receptor Tyrosine Kinases (RTKs) are a large family of membrane proteins that can activate a common set of downstream pathways, but they are also known to elicit distinct biological responses. This raises the question of how the signaling specificities of these receptors are generated. The vertebrate lens is a unique model to study the functional mechanism of RTKs. It is composed of an epithelial monolayer overlying a lens-fiber–cell core that is devoid of the complications encountered with vasculature invasion, neural innervation, and immune infiltration [1, 2]. During embryonic development, lens progenitor cells within the epithelium proliferate and migrate toward the equator of the lens until they reach the transitional zone, where they exit the cell cycle and begin to differentiate into lens fiber cells (Fig 1A). Previous studies have identified several RTKs in the lens. Among them, fibroblast growth factor receptors (FGFRs) are expressed weakly in the lens epithelium but strongly in the elongating secondary fiber cells present in the equator region [3]. Indeed, in lens explant cultures, FGFs have been shown to promote either epithelial cell proliferation or fiber-cell differentiation in a dose-dependent manner [4]. This is supported by in vivo evidence that transgenic expressions of FGFs cause premature differentiation of lens epithelial cells into fiber cells, while deletion of FGFRs or their coreceptor heparan sulfates abrogate lens fiber differentiation [5–8].
Another RTK known as platelet-derived growth factor receptor α (PDGFRα) is restricted to the lens epithelium (Fig 1A). Its ligands PDGFA and PDGFB are present in the ciliary margin zone, which abuts the transitional zone of the lens epithelium [9, 10]. PDGF in explant cultures was reported to promote the proliferation of lens epithelial cells and potentiates FGF-induced differentiation [10–12]. In support of this, transgenic overexpression of PDGFA not only increases lens size but also induces a slight elongation of lens epithelium cells, which is characteristic of lens-fiber–cell differentiation [9]. Moreover, an early examination of the Patch mutant mouse, which harbors a large genomic deletion encompassing the Pdgfrα locus, reported lens fiber defects [13]. However, this claim was later contradicted by the targeted deletion of the Pdgfrα gene, which did not appear to affect lens development. Nonetheless, it should be noted that a detailed analysis of the Pdgfrα null lens was never described [14, 15]. Therefore, the role of PDGF signaling in lens development remains a topic requiring further exploration.
In this study, we investigated PDGF signaling in lens development, aiming to understand its crosstalk with the closely related FGF-signaling pathway. We showed that PDGF signaling primarily activates the Phosphoinositide 3-kinase (PI3K)–Protein kinase B (AKT) pathway in lens development and the direct binding of PI3K to PDGFRα is required for preventing the depletion of lens epithelial cells. In contrast, FGF and Mitogen-Activated Protein Kinase (MAPK) signaling display the opposite effect in promoting lens-fiber–cell differentiation. Attenuation of MAPK signaling restored the proper balance of lens progenitor and differentiated cells in PI3K mutants, demonstrating their antagonistic roles in lens development. These two intracellular pathways converge upon Notch signaling because MAPK is required for the expression of Jagged 1 (Jag1) and PI3K is necessary for the stability of the Notch intracellular domain (NICD). Taken together, PDGF–PI3K signaling counterbalances FGF–MAPK to maintain the progenitor pool of cells within the lens epithelium.
At mouse embryonic day 14.5 (E14.5), Pdgfrα is exclusively expressed in the lens epithelium, as indicated by RNA in situ hybridization and a green fluorescent protein (GFP) reporter from the Pdgfrα locus (Fig 1B, arrowheads). We ablated Pdgfrα using the Cre deleter Le-Cre, which is active in the lens precursor cells as early as E9.5 [16]. As expected, this led to the complete loss of PDGFRα immunostaining in the Le-Cre;Pdgfrαf/f lens (Fig 1C, circled in dotted lines). At E12.5, the primary fiber cells in the Le-Cre;Pdgfrαf/f mutant failed to reach the anterior rim of the lens as those in the control samples (Le-Cre or Le-Cre; Pdgfrα f/+) did (Fig 1C, arrowheads). At E14.5 and E16.5, lens epithelial cells initiated differentiation into the secondary fiber cells at the transitional zone, which was located at the equator region of the control lens. In Pdgfrα mutants, the transitional zone had shifted anteriorly, resulting in a significant shortening of the lens epithelium (Fig 1C, arrows). Furthermore, there was an abnormal increase in cell death, as shown by TUNEL staining in the lens epithelium and reduction in lens size, while the expression of α-, β-, and γ-crystallins in lens fiber cells was preserved (Fig 1D–1F, arrowheads). These results showed that PDGF signaling was required to maintain the balance between lens epithelial and fiber-cell compartments.
Similar to other RTK family members, PDGF and FGF signaling can induce a common set of downstream effectors, notably Ras–MAPK and PI3K–AKT [17, 18]. To determine which intracellular pathway mediates the effect of PDGF signaling in the lens, we generated immortalized lens epithelial cells from neonatal pups [19], which expressed a mixture of progenitor markers Paired Box 6 (Pax6) and α-crystallin and more differentiated markers β- and γ-crystallin (S1 Fig). As shown in Fig 2A, FGF2 at 50 ng/ml induced a strong elevation in Extracellular Signal-Regulated Kinase (ERK) phosphorylation that lasted for 30 minutes, but the increase in phospho-AKT (pAKT) was much weaker and transient. Conversely, the same concentration of PDGFA produced a much higher phosphorylation rate of AKT than that of ERK. From 1 ng/ml to 50 ng/ml, PDGFA consistently stimulated higher pAKT levels than FGF2 at 5 minutes, whereas FGF2 generated stronger phospho-ERK (pERK) response than PDGFA (Fig 2B). These results were consistent with previous observations that FGF preferentially activated Ras–MAPK signaling, while PDGF was more biased toward PI3K–AKT signaling [20–24]. Indeed, pAKT staining was significantly reduced in the Le-Cre;Pdgfrαf/f mutant lens, with pERK being maintained at normal levels (Fig 2C–2E). To confirm that AKT phosphorylation in the lens depended on PDGFRα-stimulated PI3K signaling, we took advantage of the PdgfrαΔPI3K knock-in mutant that lacks the docking site for PI3K [20]. In Le-Cre; Pdgfrαf/ΔPI3K lenses, loss of the PDGFRα–PI3K interaction failed to disrupt pERK staining in the transitional zone, with the levels of pAKT being comparably reduced as in the Le-Cre;Pdgfrαf/f lens (Fig 2B–2E). Moreover, both mutants displayed the anterior encroachment of p57 staining, which marked the differentiating lens cells that had just exited the cell cycle (Fig 2B, arrows). There were fewer numbers of proliferative Ki67-expressing cells (Fig 2F), indicating a depletion of the lens progenitor cell pool. This was further demonstrated by the staining of epithelial cell markers Forkhead Box E3 (Foxe3) and E-cadherin and fiber-cell markers MAF BZIP Transcription Factor (C-maf) and Jag1, which together displayed a significant reduction in the length of the anterior lens epithelium in relation to the posterior lens circumference (Fig 2G). These observations were consistent with the notion that PDGFR–PI3K signaling maintains the progenitor population of the lens epithelium.
PI 3-kinase is a heterodimer composed of two subunits: a regulatory one known as p85 and a catalytic one known as p110. When PI3K binds to the specific phosphotyrosine residues of RTKs, p85 brings p110 to the plasma membrane to catalyze the conversion of Phosphatidylinositol 4,5-bisphosphate (PI-4,5-P) to Phosphatidylinositol (3,4,5)-trisphosphate (PI-3,4,5-P) (Fig 3A). We thus sought out to abolish PI3K signaling in the lens by crossing Le-Cre with the floxed allele of Pik3r1 encoding p85α/p55α/p50α and a knockout (KO) allele of Pik3r2 encoding p85β [25, 26]. Western blot analysis demonstrated that Le-Cre;p85α f/f;p85β KO/KO (p85 CKO) mutant lenses lost both p85 expression and AKT phosphorylation, but the pERK level was unchanged, consistent with the role of PI3K in AKT activation (Fig 3B). In line with a previous report, immunostaining showed that p85 was predominantly expressed in the transitional zone of the control lens, overlapping with pAKT staining (Fig 3C, arrows) [27]. In p85 CKO mutants, loss of p85 resulted in a smaller lens than the controls as early as E12.5 (Fig 3C). Histology analysis showed that the transitional zone in the p85 CKO lens also moved anteriorly (Fig 3C, arrowheads, quantified in Fig 4P). Altogether, the p85 CKO closely resembled the Pdgfrα mutant in their lens phenotypes, further demonstrating that PI3K is the main effector of PDGF signaling in the lens.
Knowing that phosphatase and tensin homolog (Pten) reverses the lipid phosphorylation reaction catalyzed by PI3K, we reasoned that the removal of Pten may ameliorate the p85 CKO mutant lens phenotype. Indeed, Le-Cre;p85α f/f;p85β KO/KO;Pten f/f (p85;Pten CKO) mutants displayed hyperphosphorylation of AKT in the entire lens, whereas pERK staining was only modestly expanded (Fig 3D–3F, arrows). Marked by the boundary between Ki67 and p57 staining, the transitional zone moved anteriorly in p85 CKO mutants, but it reverted back to the lens equator in p85;Pten CKO mutants (Fig 3D, arrows). Similar to Pdgfrα mutants, the p85 CKO lens exhibited aberrant cell death in the anterior epithelium, as indicated by TUNEL staining (Fig 3D, arrowheads), and reduction in cell proliferation, as shown by Ki67 expression (Fig 3G). It was previously reported that deletion of Pten alone in the lens led to slight increase in cell proliferation and apoptosis at E12.5 [28]. The loss of Pten in p85;Pten CKO mutants restored cell proliferation, but it did not prevent cell apoptosis defect. This result indicated that Pten deletion rescued the p85 deficiency in the epithelial and fiber-cell compartmentation without enhancing cell survival.
The depletion of the lens progenitor pool in Pdgfrα and p85 mutants could be due to the increase in cell death present in the lens epithelium. However, this model was challenged by the selective rescue of the lens epithelial population but not cell survival in p85;Pten CKO mutants. To address the question of whether cell death was responsible for depleting the lens progenitors, we sought to disable the intrinsic apoptotic pathway in p85 mutants. The proapoptotic B-cell lymphoma 2 (BCL2) proteins BCL2-Associated X (Bax) and BCL2 Antagonist/Killer (Bak) are required for permeabilization of the mitochondrial outer membrane and release of cytochrome c into the cytoplasm, which triggers the programmed cell death pathway [29, 30]. Previous studies have shown that knocking out Bax and Bak prevents most forms of apoptosis in vivo [31, 32]. After crossing p85 CKO mutants with Baxflox/flox;BakKO/KO mice, we confirmed by TUNEL staining that genetic removal of Bax and Bak indeed prevented cell apoptosis in Le-Cre;p85α f/f;p85β KO/KO;Baxflox/flox;BakKO/KO (p85;Bax/Bak CKO) mutant lenses (Fig 4A–4D, arrowheads), but the cell proliferation defect was not rescued, as shown by the Bromodeoxyuridine (BrdU) incorporation index (Fig 4E–4H). Moreover, Ki67 and p57 staining showed that the transitional zone was still positioned anteriorly in the p85;Bax/Bak CKO mutant lens (Fig 4I–4K, arrows). Further, the number of lens epithelial cells marked by Foxe3 staining remained significantly reduced compared to the control lens (Fig 4L–4O, arrows). As a result, the ratios of the anterior epithelium versus the posterior lens circumference were indistinguishable between p85 CKO and p85;Bax/Bak CKO mutant lenses (Fig 4P). Taken together, these results demonstrated that the cell death caused by PI3K inactivation was not a key factor in the depletion of lens progenitors.
The loss of lens epithelial progenitors observed in Pdgfra and p85 mutants was highly reminiscent of the defects observed in the Notch-signaling mutant lens [33–37]. During lens development, the Notch ligand Jag1 is expressed by the nascent secondary fiber cells at the equatorial region of the lens. It induces Notch signaling to activate Hes Family basic helix–loop–helix (BHLH) Transcription Factor 1 (Hes1) in the transitional zone, promoting the proliferation of the lens epithelial cells and preventing them from premature differentiation. Genetic deletion of Jag1 resulted in a striking loss of the anterior epithelial cells [34]. Although p85 CKO mutants displayed a similar depletion of lens epithelial cells, Jag1 mRNA and protein expression were preserved (Fig 5A, arrowheads). Instead, there was a striking loss of Hes1 expression in the transitional zone (Fig 5A, arrow), suggesting that PI3K may be required for Notch-signaling recipient cells in the lens epithelium.
To investigate the molecular mechanism of PI3K in regulating Notch signaling, we cultured the immortalized lens epithelial cells on plates coated with Jag1 and found that these cells displayed elevated levels of Hes1 expression as measured by quantitative Reverse transcription-polymerase chain reaction (RT-PCR) (Fig 5B). Importantly, the Jag1-induced Hes1 expression was repressed by both the Notch inhibitor N-[N-(3,5-Difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester (DAPT) and the PI3K inhibitor LY294002, which is consistent with the in vivo observation that inactivation of PI3K disrupted Jag1-induced Notch signaling in the lens. We next sought to bypass Jag1 in vitro by using EDTA to chelate the extracellular calcium, which stimulates shedding of the Notch extracellular domain and release of the NICD to activate Hes1 transcription [38, 39]. In the lens cell cultures, the EDTA-induced Hes1 expression was also suppressed by DAPT and LY294002 (Fig 5C). Taken together, these results suggest that PI3K may regulate the Notch signaling pathway downstream of the Jag1 ligand activation of Notch.
We next examined whether PI3K regulated the proteolytic cleavage of the Notch protein. Time course analysis by western blot indicated that the Notch1 intracellular domain (N1ICD) was generated as early as 10 minutes after EDTA treatment and was blocked by DAPT (Fig 5D). In contrast, although increasing the concentration of LY294002 progressively quenched AKT phosphorylation at 10 minutes, it did not reduce the level of N1ICD until 1 hour after EDTA treatment. Nevertheless, 50 μM LY294002 still disrupted the induction of Hes1 by EDTA. These results suggested that PI3K did not influence the initial cleavage of the Notch1 receptor but rather affected the stability of N1ICD. To further corroborate this finding, we overexpressed a Human influenza hemagglutinin (HA)-tagged rat N1ICD in lens cells. After blocking protein synthesis using cycloheximide (Chx), LY294002 treatment lead to a sharp decline of N1ICD over the course of 6 hours, while the AKT level was maintained (Fig 5E). This effect was confirmed using another potent PI3K inhibitor, PX-866 (S2 Fig) [40]. Glycogen Synthase Kinase 3β (GSK3β), a protein previously reported to affect Notch signaling, can be phosphorylated by AKT to inhibit its activity [41, 42]. We found that LY294002 reduced phosphorylation of both AKT and GSK3β in the cultured lens cells. Although the addition of a potent and specific GSK3β inhibitor (GSK3βi) CHIR99021 had little effect on AKT and GSK3β phosphorylation, it partially reversed N1ICD instability caused by the LY294002 treatment (Fig 5G). These results support that GSK3β is a downstream mediator of PI3K in regulating Notch signaling.
If the role of PI3K in maintaining the lens progenitor cell pool is to promote the stability of N1ICD, we reasoned that overexpression of N1ICD may compensate for the loss of PI3K in lens development. To test this hypothesis, we crossed p85 CKO mutants with mouse carrying a Rosa26 knock-in allele (RosaN1-IC) that expressed N1ICD after the removal of the loxP-STOP-loxP cassette [43]. It was previously shown that overexpression of N1ICD in the lens resulted in hyperproliferation of the anterior lens epithelium at the expense of fiber-cell differentiation [36]. Likewise, Le-Cre;p85α f/f;p85β KO/KO;RosaN1-IC (p85 CKO;N1ICD) embryos displayed a significant expansion of the Ki67-positive lens epithelium compared to that of the p85 CKO mutants (Fig 6A). Marked by the boundary between Foxe3 and C-maf staining, the transitional zone was also reverted back to the equatorial region (Fig 6A and 6B). As a result, there was a significant increase in lens size in p85 CKO;N1ICD mutants (Fig 6C). This genetic evidence is in line with the model that PI3K stabilizes N1ICD to control lens differentiation.
The above results showed that the PDGF–PI3K signaling enhanced the Notch pathway to maintain the lens progenitor pool, which is opposite to the known function of FGF–ERK signaling in stimulating lens cell differentiation [5, 44]. This prompted us to examine the potential interactions between these two critical signaling pathways. We have previously shown that FGF receptors engage the docking protein Fibroblast Growth Receptor Substrate 2 (Frs2) and the tyrosine phosphatase Rous sarcoma oncogene (Src) Homology Phosphatase 2 (Shp2) with the purpose of inducing ERK signaling during lens development [45]. In support of this model, the genetic ablation of Frs2 and Shp2 abolished ERK phosphorylation in the Le-Cre;Frs2 f/f;Shp2 f/f (Frs2;Shp2 CKO) lens at E12.5 (Fig 7A). In contrast to the control lens that expressed PDGFRα in the anterior and Jag1 in the posterior, Frs2;Shp2 CKO mutants contained a hollow lens with only PDGFRα expression (Fig 7A, arrow and arrowheads). In a complementary approach, we also examined a transgenic mouse line (Fgf3OVE391) that overexpressed Fgf3 under control of an αA-crystallin promoter [46, 47]. As expected, pERK staining expanded from the transitional zone to the entire Fgf3OVE391 lens, coinciding with ectopic induction of Jag1 and loss of PDGFRα expression in the anterior lens epithelium (Fig 7B, arrow and arrowheads). These results demonstrated that FGF signaling regulates the Notch pathway by inducing the expression of its ligand as well as inhibiting PDGF signaling by suppressing its receptor.
We next investigated the signaling crosstalk at the level of MAPK and PI3K. To avoid the early lens defect in Frs2;Shp2 CKO mutants, we ablated only Shp2 in the lens, which we previously showed to attenuate MAPK signaling, reduce cell proliferation, and increase apoptosis [48]. Interestingly, we observed a modest but consistent reduction in Jag1 expression in the Le-Cre;Shp2 f/f (Shp2 CKO) mutant, indicating a dosage-dependent regulation of Jag1 by MAPK signaling (Fig 7C and 7D). Combined deletion of both Shp2 and p85 resulted in a smaller lens size than either single mutant, consistent with the notion that both MAPK and PI3K are required for cell survival and proliferation in the lens. Notably, the Shp2 CKO mutant displayed a posterior shift of the transitional zone, as opposed to the anterior shift observed in the p85 CKO lens. The transitional zone in the Le-Cre; p85α f/f;p85β KO/KO;Shp2 f/f (p85;Shp2 CKO) lens, however, was restored to the equatorial region (Fig 7C and 7E). Taken together, these results revealed that MAPK and PI3K play the dual role of cooperating in promoting lens cell survival and proliferation while also antagonizing each other in the regulation of cell differentiation.
PDGF and FGF receptors are both RTKs that share a similar network of signaling pathways. Previous studies using cell-culture models have identified the potential downstream effectors of PDGF signaling as PI3K [20], Src [49], Phospholipase C γ (PLCγ) [50], and MAPK cascades [51], many of which are also activated by FGF signaling [52]. In this study, we showed that under physiological conditions, FGFRs and PDGFRs within the lens were biased toward different downstream effectors: FGFRs mainly activated Ras–MAPK, while PDGFRs mainly activated PI3K–AKT. This arrangement sets up both positive and negative interactions between these two RTK signaling pathways (Fig 7F). On the cooperative side, the FGF–Ras–MAPK pathway induces the Notch ligand Jag1, while the PDGF–PI3K–AKT one stabilizes the Notch effector NICD. This leads to the activation of Notch signaling with the purpose of dampening lens cell differentiation. On the antagonistic side, MAPK restricts PI3K activity by inhibiting PDGFRα expression while also directly promoting lens cell differentiation. By biasing progenitor cells toward distinct cell fates, the FGF–MAPK and PDGF–PI3K pathways enable the lens to achieve the optimum balance of expansion versus differentiation during development.
Both PDGF and PI3K signaling have been previously investigated for their roles in lens development. Genetic perturbation of PI3K signaling by deleting the catalytic subunit p110α or the negative regulator Pten have revealed its role in lens cell growth and survival [28, 53]. In vitro studies, however, have shown that the PDGF–PI3K signaling pathway may also promote lens-fiber–cell differentiation [12, 21, 27, 54, 55]. It was therefore a surprise to find that the complete ablation of PDGFRα and PI3K signaling in vivo resulted in premature differentiation and depletion of the lens progenitor cells. This discrepancy is likely due to the different experimental systems used. The previous in vitro studies were 2D models that utilized lens cell lines and explant cultures that expose all the cells to a uniform environment. In contrast, our in vivo study showed that the lens epithelial cells experienced not only the prodifferentiation FGF signal from the vitreous humor but also the antidifferentiation Notch signal from the fiber cells. In this 3D environment, PDGF–PI3K signaling was restricted to the transitional zone of the lens because of the convergent expressions of the ligand (PDGFA), receptor (PDGFRα), and regulator (p85). This confined the thrust of PDGF–PI3K signaling to the augmentation of Notch signaling in maintaining the lens progenitor cell population. This highlights the importance of the precise environmental context with regards to understanding the different functions of cell signaling.
Notch signaling is an evolutionarily conserved pathway controlling the proliferation and differentiation of progenitor cells [56, 57]. It is the underlying mechanism for lateral inhibition, which enables equipotent cells to take on divergent cell fates. In this model, the newly differentiated cells express high levels of the ligand, Jagged or Delta, that activates the Notch receptor in the neighboring progenitor cells. This leads to proteolytic cleavages of Notch adjacent to the transmembrane domain by Tumor Necrosis Factor, Alpha, Converting Enzyme (TACE) and at its transmembrane domain by γ-secretase. The resulting cytoplasmic peptide of Notch (NICD) translocates to the nucleus, where it binds the transcriptional factors Recombination Signal Binding Protein For Immunoglobulin Kappa J Region (Rbpj) and recruits Mastermind to activate downstream genes. These target genes, including the family of BHLH transcription factors, Hes and Hes Related Family BHLH Transcription Factor With YRPW Motif (Hey), act in progenitor cells to prevent differentiation and maintain the cells’ proliferative abilities [56, 58]. PI3K has been previously implicated in Notch signaling. In mouse embryonic fibroblast (MEF) and a variety of human cancer cell lines, PI3K has been shown to induce Jag1 expression via the Mammalian Target Of Rapamycin Kinase (mTOR) pathway to positively regulate Notch signaling [59]. In the Drosophila external sensory organ and Human embryonic kidney 293 (HEK293) cells, however, the mTOR regulator tuberous sclerosis complex (TSC) was reported to activate Notch in a target of rapamycin complex 1 (TORC1)-independent fashion [60]. Another potential mediator of PI3K regulation of Notch is GSK3β, which can be inactivated by AKT phosphorylation. GSK3β has been shown to bind and phosphorylate NICD in vitro [42, 61]. Although GSK3β-null MEF cells appeared to exhibit a reduced level of Notch signaling, it was later reported that the inhibition of GSK3β by PI3K–AKT promoted Notch signaling in a variety of different cell types that included Chinese hamster ovary (CHO) cells, T cells, and hippocampal neurons [41, 61]. Consistent with this, the genetic ablation of GSK3α and GSK3β in the developing cortex resulted in an elevated level of NICD and Hes1 [62]. In our study, we did not observe any decrease in Jag1 expression in the p85 CKO mutant lens. Instead, the inhibition of PI3K accelerated the degradation of NICD in lens cells and could be partially rescued through the blocking of GSK3β activity. These results suggested that the stabilization of NICD through the inhibition of GSK3β activity is an important mechanism for PI3K to potentiate Notch signaling in lens development.
A major goal in studying multicellular organisms is to understand the cellular algorithm by which overlapping signals exert distinct biological outcomes. Using lens development as a model, we have revealed in this study that FGF and PDGF receptors both cooperate and antagonize each other in a tightly regulated fashion. What accounts for the functional differences between these two similar RTKs? First, despite their propensities to induce the same set of downstream signaling cascades, RTKs differ in their mode of engagement with these pathways. Whereas PDGFR directly binds the p85 subunit of PI3K to activate PI3K–AKT signaling, the main signaling output of FGFR is channeled through the adaptor protein Frs2, which recruits Shp2, Growth Factor Receptor Bound Protein 2 (Grb2), and v-crk avian sarcoma virus CT10 oncogene homolog (Crk) proteins to activate the Ras–MAPK signaling cascade [47, 63]. Although Grb2 has been reported to interact with GRB2-Associated Binding Protein 1 (Gab1) to activate PI3K, we previously showed that the genetic ablation of Gab1 and its homologue Gab2 failed to disrupt FGF signaling in both MEF cells and the developing lens [45, 64]. Thus, the intrinsic disparities in the signaling networks bias PDGFR toward PI3K and FGFR toward MAPK, which has been documented in a variety of cell types and demonstrated in genetic studies [20, 22–24]. The differential responses to FGF and PDGF were also evident in early studies using lens epithelial explants that found that compared to FGF, PDGFA led to more rapid increase in AKT phosphorylation but less sustained ERK phosphorylation [12, 65]. Such signaling biases certainly do not equate to ERK activation exclusively by FGF or PI3K activation solely by PDGF, but they help to explain the distinct physiological functions of these two growth factors. Second, under physiological conditions, the ligands and receptors of RTK signaling differ in their affinities and stoichiometry, which can lead to quantitative differences in the duration and amplitude of intracellular pathways as well as qualitative differences in the specific substrates being recruited [28, 66]. Notably, such signaling specificity can also be overridden in nonphysiological conditions, which may explain why the transgenic overexpression of Pdgfa could promote lens cell differentiation and the constitutive activation of PI3K after Pten deletion could ameliorate some of the Fgfr2 lens KO phenotype characteristics [9, 28]. The third reason why PDGF and FGF signaling leads to different cell fate decisions in lens development has to do with their involvement in Notch signaling. We have shown that while FGF–MAPK signaling induces Notch ligand expression, the PDGF–PI3K pathway stabilizes the Notch effector NICD. Because of the lateral inhibition effect of Notch signaling, cell differentiation is permitted in the Notch-ligand–expressing cells but restrained in the Notch effector cells. This naturally segregates the biological functions of FGF–MAPK and PDGF–PI3K signaling into pro- and antidifferentiation outcomes in lens development. Thus, the cooperative and inhibitory interactions among FGF, PDGF, and Notch pathways can be considered as “and” and “or” gates of signaling circuitry, which are used to generate intricate programs that achieve the exquisite balance of growth and differentiation in a complex multicellular organism.
The animal experiments were approved by Columbia University Institutional Animal Care and Use Committee (protocol number: AAAR0429). The mice were killed in humane fashion via CO2 asphyxiation and cervical dislocation according to the American Veterinary Medical Association (AVMA) Guidelines.
Mice carrying Frs2αflox and Shp2flox alleles were bred and genotyped as described [67, 68]. We obtained Fgf3OVE391 from Dr. Michael Robinson (Miami University, Oxford, OH, USA), Le-Cre from Drs. Ruth Ashery-Padan (Tel Aviv University, Tel Aviv, Israel) and Richard Lang (Children's Hospital Research Foundation, Cincinnati, OH, USA), p85αflox and p85βKO from Dr. Lewis Cantley (Weill Cornell Medicine, New York, NY, USA), and PdgfrαΔPI3K from Dr. Philipo Soriano (Mount Sinai School of Medicine, New York, NY, USA) [16, 20, 25, 26, 46, 69]. Pdgfrαflox (stock no.: 006492), PdgfrαGFP (stock no.: 007669), and Ptenflox mice (stock no.: 006440) and Rosa-N1-ICDflox/+ (stock no.: 008159) and Baxflox/flox;BakKO/KO (stock no.: 006329) mice were obtained from Jackson Laboratory (Bar Harbor, ME, USA) [31, 32, 43, 70–72]. In all conditional KO experiments, mice were maintained on a mixed genetic background, and Le-Cre–only or Le-Cre and heterozygous flox mice were used as controls. Mouse maintenance and experimentation were performed according to protocols approved by Columbia University Institutional Animal Care and Use Committee.
Mouse embryos were fixed with 4% paraformaldehyde (PFA) in PBS overnight and paraffin- or cryo-embedded. The paraffin sections (10 μm) were rehydrated and stained with hematoxylin–eosin (HE) for histological analysis or TUNEL and BrdU staining as previously described [73, 74]. RNA in situ hybridization and immunostaining were performed on at least three cryosections for each embryo (10μm) [75]. The in situ probes used were Jag1 (from Doris Wu, National Institute on Deafness and Other Communication Disorders) and Pdgfrα (Marc Mercola, Harvard Medical School). The following primary antibodies were used: anti-C-maf (sc-7866), ant-Foxe3 (sc-377465), and anti-Jag1 (sc-6011) (all from Santa Cruz Biotechnology, Dallas, TX, USA); anti-p85 (#4257), pAKT (D9E), and anti-pERK1/2 (#4370) (all from Cell Signaling Technology, Danvers, MA, USA); anti-P57 (ab75974, from Abcam, Cambridge, UK); anti-E-cadherin (U3254, from MilliporeSigma, St. Louis, MO, USA); and anti-Ki67 (#550609) and anti-Pdgfrα (#558774) (both from BD Pharmingen, BD Biosciences, San Jose, CA, USA). The antibody against Hes1 was a gift from Dr. Ryoichiro Kageyama (Kyoto University, Kyoto, Japan). Antibodies against α-, β-, and γ-crystallins were kindly provided by Dr. Sam Zigler (National Eye Institute). For pERK and pAKT staining, the signal was amplified using a Tyramide Signal Amplification kit (TSA Plus System, PerkinElmer Life Sciences, Waltham, MA, USA). To analyze the fluorescent intensity, the average pixel intensity in the lens transitional zone was obtained by measuring the grayscale value using ImageJ (NIH), and each data point was calculated by taking the average from three separate images of the same lens. The lens anterior and posterior arcs and areas were measured using ImageJ and analyzed by one-way ANOVA analysis. If not otherwise stated, at least three embryos were analyzed for each genotype, and they presented a consistent phenotype.
Ten newly born mouse pups were dissected in cold PBS, and the lenses were rolled over on an autoclaved filter paper before being transferred into 0.5 ml of cold PBS. These lenses were treated with 0.125% trypsin in EMEM (Eagle’s MEM), minced using 22-gauge needles, and incubated for 5 minutes at 37°C. The EMEM plus 4 mM L-Glutamine with 10% fetal bovine serum (FBS) was added to stop trypsin activity. After centrifuge, the lens cells and tissue pieces were plated in a 35-mm petri dish. Because of lack of proliferative ability, lens fiber cells eventually died out after 4 days of culture. After 3 days of exposure to SV40 T-antigen retroviral particles collected from a Ψ2 cell supernatant, immortalized lens cells were maintained in DMEM with 10% FBS [19]. The lens cells were starved prior to stimulation by FGF2 and PDGFA.
Cells in a 24-well plate were lysed with 80 μl/well RIPA buffer (50 mM Tris-HCl [pH 8.0], 150 mM NaCl, 1% NP40, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with 1× Halt protease inhibitor cocktail (ThermoFisher, Waltham, MA, USA). Cell lysates were denatured by boiling with Laemmli SDS sample buffer for 5 minutes before protein separation on 8%–10% SDS polyacrylamide gels. After protein transfer, the Immobilon-FL PVDF membrane (Millipore) was blocked with Odyssey TBS blocking buffer (LI-COR Biosciences, Lincoln, NE, USA) at room temperature for 1 hour. Primary antibodies were diluted in the same blocking buffer with 0.1% Tween-20 and detected by corresponding secondary antibodies conjugated with IRDye 800CW or 680RD (LI-COR Biosciences). Proteins were visualized by an infrared-based Odyssey SA scanner (LI-COR Biosciences). The signal intensity was quantified using the Odyssey software. The antibodies used for western blot were mouse anti-pERK1/2 (sc-7383), anti-HA (sc-805), and anti-Hes1 (sc-25392) (all from Santa Cruz Biotechnology) and mouse anti-AKT (#4060), anti-ERK1/2 (#4695) anti-Notch1-ICD (#4147), rabbit anti-pAKT (#4060), and anti-phospho-GSK3β (pGSK3β) (#9336) (all from Cell Signaling Technology).
For the Notch activation assay, 3.5 × 103 immortalized lens cells were plated in each well of a 24-well tissue culture plate two days before the experiment began. Cells were first pretreated with the PI3K inhibitor LY294002 (10 μM or 50 μM) or the γ-secretase inhibitor DAPT (20 ng/μl) for 1 hour before treatment with 1 μM EDTA in Hanks’ balanced salt solution for 15 minutes to activate Notch. Cells were then incubated in DMEM with and without LY294002 or DAPT for another 10 minutes or 1 hour before protein or RNA extraction and analysis by western blots or qPCR.
To measure N1ICD degradation, 3.5 × 103 immortalized lens cells were transfected with 0.5 μg plasmid pCCL-Notch1IC (a gift from Dr. Jan Kitajewski, University of Illinois Cancer Center, IL) using Lipofectamine 3000 (Invitrogen, Carlsbad, CA, USA) following the manufacturer’s protocol. This plasmid contains a cDNA encoding N1ICD (rat: NM_001105721.1) with a C-terminal HA tag. 24 hours after transfection, cells were treated with 50 μM LY294002, 1 μM PX-866, or 5 μM CHIR99021 in the presence of protein synthesis inhibitor Chx (300 ng/μl). Cell lysates were collected before or after 1.5 hours, 3 hours, 4.5 hours, or 6 hours of treatment. The levels of N1ICD were quantified by western blot analysis probed with an anti-HA-tag antibody.
Cells in a 24-well plate were lysed with 500 μl/well TRIzol Reagent (ThermoFisher), and total RNA was extracted according to the manufacturer’s protocol. cDNAs were synthesized using the High-Capacity cDNA Reverse Transcription Kit (ThermoFisher). Triplicate PCR reactions were prepared with SYBR Green PCR master mix (Applied Biosystems, Foster City, CA, USA) and carried out in a StepOnePlus real-time PCR system (Applied Biosystems). The primers used for Hes1 are 5′-TCAACACGACACCGGACAAAC and 5′-ATGCCGGGAGCTATCTTTCTT. 18S rRNA (5′-GTAACCCGTTGAACCCCATT, 5′-CCATCCAATCGGTAGTAGCG) was amplified as the internal control. The CT value of each reaction was used to calculate the relative concentration of the target RNAs.
|
10.1371/journal.pcbi.1007043 | Hierarchical Bayesian inference for concurrent model fitting and comparison for group studies | Computational modeling plays an important role in modern neuroscience research. Much previous research has relied on statistical methods, separately, to address two problems that are actually interdependent. First, given a particular computational model, Bayesian hierarchical techniques have been used to estimate individual variation in parameters over a population of subjects, leveraging their population-level distributions. Second, candidate models are themselves compared, and individual variation in the expressed model estimated, according to the fits of the models to each subject. The interdependence between these two problems arises because the relevant population for estimating parameters of a model depends on which other subjects express the model. Here, we propose a hierarchical Bayesian inference (HBI) framework for concurrent model comparison, parameter estimation and inference at the population level, combining previous approaches. We show that this framework has important advantages for both parameter estimation and model comparison theoretically and experimentally. The parameters estimated by the HBI show smaller errors compared to other methods. Model comparison by HBI is robust against outliers and is not biased towards overly simplistic models. Furthermore, the fully Bayesian approach of our theory enables researchers to make inference on group-level parameters by performing HBI t-test.
| Computational modeling of brain and behavior plays an important role in modern neuroscience research. By deconstructing mechanisms of behavior and quantifying parameters of interest, computational modeling helps researchers to study brain-behavior mechanisms. In neuroscience studies, a dataset includes a number of samples, and often the question of interest is to characterize parameters of interest in a population: Do patients with attention-deficit hyperactive disorders exhibit lower learning rate than the general population? Do cognitive enhancers, such as Ritalin, enhance parameters influencing decision making? The success of these efforts heavily depends on statistical methods making inference about validity and robustness of estimated parameters, as well as generalizability of computational models. In this work, we present a novel method, hierarchical Bayesian inference, for concurrent model comparison, parameter estimation and inference at the population level. We show, both theoretically and experimentally, that our approach has important advantages over previous methods. The proposed method has implications for computational modeling research in group studies across many areas of psychology, neuroscience, and psychiatry.
| Across different areas of neuroscience, researchers increasingly employ computational models for experimental data analysis. For example, decision neuroscientists use reinforcement learning (RL) and economic models of choice to analyze behavioral and brain imaging data in reward learning and decision-making tasks [1, 2]. The field of computational psychiatry uses these models to characterize patients and people at the risk of brain disorders [3–6]. Neuroimaging studies use models of neural interaction, such as dynamic causal modeling [7, 8], as well as abstract models to analyze brain signals [2, 9]. The success of these efforts heavily depends on statistical methods making inference about validity and robustness of estimated parameters across individuals, as well as making inference on validity and generalizability of computational models. A key theoretical and practical issue has been capturing individual variation both in a model’s parameters and additionally in which of several candidate models a subject expresses, which may also vary from subject to subject.
Computational models usually rely on free parameters, such as learning rate in RL models, which often capture quantities of scientific interest but typically vary across individuals and must be estimated from data. A dataset includes a number of subjects, and often the question of interest is to characterize parameters in a population: Is choice consistency altered in patients with attention-deficit hyperactive disorders? Do cognitive enhancers, such as Ritalin, enhance the learning rate at the population level? These questions are most naturally framed in terms of hierarchical models, which characterize both the population distributions over a model’s parameters and also each individual subject’s parameters given the population distribution. Since these two levels are mutually interrelated, they are often estimated simultaneously, using methods like expectation maximization or sampling (MCMC). For example, the hierarchical parameter estimation (HPE) procedure [10, 11] regularizes individual estimates according to group statistics, producing better individual estimates and permitting reliable group-level tests. Because subjects typically share underlying structure, hierarchical Bayesian approaches can leverage this structure to yield better individual estimates and to provide better predictions for unseen data, compared to approaches that fit each subject separately [12].
A second, and seemingly logically prior, question is which of several candidate models provides the best explanation for the data. This is important both for providing the setting within which to do parameter estimation, and also for investigating questions of scientific interest. Are rodents’ reaction times best explained by independent or competing accumulators? Do compulsive gamblers rely more on model-free RL compared to controls? Importantly, in principle (and apparently in practice) the model expressed might also vary from subject to subject; thus modern model comparison techniques rely on estimating which of several models obtains for each subject [13]. Estimating such variation is important since the prior assumption that the same model obtains across all individuals (treating model identity as a fixed effect) is a very strong (and in most cases potentially unwarranted) assumption, which makes model comparison very sensitive to outliers [13]. To estimate this variation, in turn, depends on the likelihood of each subject’s data given each model (and, thus, on each subject’s parameters for each model).
Intuitively, evaluating whether a model is a good model for a subject’s data precedes estimation of its specific parameter values; and indeed, previous research has used separate tools to solve these two problems. But statistically, the two questions are actually interconnected, because individual parameters and hence individual fit depend on which subjects belong to the population that expresses the model. Here, we address this challenge from a fully Bayesian viewpoint. This work addresses issues of statistical inference over both parameters and models, which have remained elusive with the previous hierarchical methods.
Notably, although it is accepted (for the reasons discussed above) that the best-fitting model may vary from subject to subject, hierarchical parameter estimation (conducted separately) has typically assumed that the given model is expressed over all subjects, i.e. that it is a fixed effect (and if multiple models are compared, these are each fit to the entire population). This assumption biases parameter estimation, at both individual and group levels, because it entails that the estimated parameters for each individual subject equally affect group-level estimates, even though some members of the population may be better understood as expressing altogether different models. This same bias, in turn, affects the estimation of which subjects are best fit by each model.
In this work, we introduce a hierarchical and Bayesian inference method, which solves these problems by addressing both model fitting and model comparison within the same framework using variational techniques. Furthermore, our fully Bayesian approach enables us to assess uncertainty and provide a rigorous statistical test, HBI t-test, for making inference about parameters of a model at the population level, an issue that has not been addressed in some previous hierarchical models. This paper is structured as follows. First, we highlight the main theoretical advances in our approach. A full formal treatment is given in Materials and methods and S1 Appendix. We then apply the proposed method to synthetic choice datasets as well as empirical datasets to demonstrate its advantages over previous methods.
Consider a typical computational modeling study in which data of a group of subjects have been measured and a set of candidate models are considered as possible underlying computational mechanisms generating those data. Such studies have generally two main goals: 1) to compare model evidence across competing models; 2) to estimate free parameters of models for each individual and their group-level distributions. All this is typically characterized in terms of inference in a hierarchically structured model of the data, which captures how each subject’s observations depend on their parameters and the individual parameters on their group distribution.
The HPE procedure [10, 11] employs a hierarchical approach to define the priors based on statistics of the group. This method typically assumes that for a particular model k, all individual parameters are normally distributed,
p ( h k n ) = N ( h k n | μ k , V k ) ,
where hkn is a vector of the free parameters of the kth model for subject n, μk and Vk are the mean and variance parameters, respectively, indicating the prior distribution over hkn.
It is important to distinguish the statistical model itself from the algorithms or approximations used to estimate it. HPE uses the expectation-maximization algorithm [14], a well-known iterative procedure, for obtaining estimating group parameters μk and Vk and individual parameters hkn. Every iteration of this algorithm alternates two steps: 1) an expectation step in which the individual parameters are estimated in light of the group-level distribution; and 2) a maximization step in which the group parameters, μk and Vk, are updated given the current estimates of the individual parameters. Importantly, reflecting the assumption that all subjects express model k, this update weights the individual subjects’ estimates equally; for instance, the update for μk is given by the average of subject level mean estimates (denoted θkn) across all subjects:
μ k = 1 N ∑ n θ k n ,
where N is the number of subjects.
Although HPE characterizes variation across subjects in the model parameters hkn (that is, it treats those parameters as random effects), a critical assumption of the procedure is that the parameters for model k are estimated assuming that the same model is responsible for generating data in all subjects. That is, the model identity is taken as a fixed effect, in contrast to the random effects approach that assumes different models might be responsible for generating data in different subjects. The fixed effects assumption has two important implications: 1) for parameter estimation, group parameters, the group mean μk and variance Vk, are influenced equally by all subjects, even those who would be better fit by some other candidate model j ≠ k; 2) for model comparison, the straightforward procedure (e.g. iBIC from [10, 11]) is to compare models according to the sum of individual model evidences over all subjects, i.e. again treating the model identity as a fixed effect. Note that while it is possible to submit individual model evidence values (per subject and model) derived from HPE to a separate model comparison procedure that treats model identity as a random effect (such as random effects model selection [13]), these will be biased both from having been fit under the fixed effects assumption and also due to the optimization of the free group-level parameters. For this reason, HPE has typically been accompanied by fixed-effects model comparison [10, 11, 15], whereas attempts to study subject-subject variation in model identity [13] have typically been conducted using a different, non-hierarchical parameter estimation procedure. Altogether, violations of the fixed effects assumption can adversely influence both parameter estimation and model comparison.
Here, we extend HPE’s generative model with another level of the hierarchy, specifying for each subject which model generated their data. This is governed by a subject-specific multinomial random variable, itself drawn from a distribution controlling the proportion of each model in the population. This, in effect, merges the Bayesian model selection model from Stephan et al. [13] with HPE. To accomplish inference in this model, we then lay out a procedure for joint inference over model identities and parameters, including quantifying the probability that each model is responsible for generating data for each subject. To achieve this goal, we adopt a fully Bayesian framework in which the group parameters for each model, μk and Vk, are also random variables. This also gives us a straightforward way to quantify the level of certainty in group-level estimations. We use mean-field variational Bayes [16, 17], an extension of expectation-maximization [18], which is able to deal with multiple latent variables in a probabilistic model. Since HBI is a mean-field variational framework, the resulting algorithm (see Materials and methods) is an iterative algorithm. On every iteration, the HBI performs 4 steps: calculates the summary statistics, updates its estimates of the posterior over group parameters, updates its estimate of the posterior over each individual parameter and finally updates its estimates of responsibility of each model in generating each individual data. The algorithm and other important mathematical issues are given in Materials and methods. Here, we highlight three main results. The mathematical proofs are given in S1 Appendix.
As noted above, the HBI method estimates the probability of each subject’s dataset being generated by each model, or the responsibility of model k for generating data for subject n, rkn, which is expressed as (expected) probability. Larger values of rkn (i.e. close to 1) indicate that model k is likely to be the true underlying model of the nth subject. In contrast, smaller values of rkn (close to 0) indicate that model k is unlikely to be the underlying model for the nth subject. Based on the responsibilities, it is then possible to estimate the number of subjects explained by each model, N¯ k:
N ¯ k = ∑ n = 1 N r k n .
Thus N¯ k is always less than the number of subjects and indexes the predominance of model k in the population. Furthermore, the fraction N¯ k / N is called model frequency, which always lies between 0 and 1 and is a useful and intuitive metric for model comparison.
In practice, in many situations, researchers are interested in selecting a single best model (rather than relative comparisons among several) even in the face of variation in model identity across subjects. One way to accomplish this goal is to compute the exceedance probability of each candidate model, a metric commonly used for model selection [13]. Exceedance probability is the probability that model k is more commonly expressed than any other model in the model space. Furthermore, the random effects approach enables us to quantify how likely the observed differences in model evidence is simply due to chance [19]. In this case, model selection is not statistically supported, as there is no meaningful difference between models. A metric called protected exceedance probability [19], which typically is more conservative than the exceedance probability, takes into account this possibility (see Materials and methods). Altogether, the random effects approach results in a more robust model comparison and model selection, one less driven by outliers than fixed-effects methods. Note that previous attempts to do model selection at group level using exceedance probability assumed no hierarchy for parameter estimation, thus did not deal with the issue that parameter estimation was not properly conditionalized by group distributions based on model identity.
We noted above that an issue with the HPE is that the influence of subjects on the group parameters is equal, due to the assumption that the model is a fixed effect. However, by virtue of its random effects structure, the comparable parameter in our approach, the mean of posterior distribution over μk, denoted by ak, shows an important property: Algorithmically, a subject’s effect on this parameter depends on the degree to which the model is estimated to be the underlying model for that subject. Specifically, this parameter, ak, is updated at each iteration as:
a k = 1 1 + N ¯ k ( a 0 + ∑ n r k n θ k n ) ,
where θkn is the mean of the individual posterior and a0 is the prior mean over μk. The important point in this equation is that ak is a weighted average of individual parameters, in which the weights are the corresponding responsibilities, rkn. This is not specific to the group mean, but it is rather a general feature of our approach: contribution of model k to group parameters is weighted according to the responsibility of model k in generating data in the nth subject, rkn.
As mentioned above, another issue that has been incompletely treated in HPE is related to inference on parameters of a fitted model at the population level. Statistically, one needs the uncertainty of the estimated group mean, μk, to be able to make inference on the corresponding parameter at the group level. Since parameters fitted by the HPE are not independent but instead regularized according to the variance given by data, one cannot employ regular statistical tests, such as t-test, to test whether a specific model parameter is “significantly” different from zero. Using those tests on such parameters is biased in favor of generating a significant p-value (more false positives). The HBI framework solves this problem by quantifying the uncertainty of the posterior over the group parameter, resulting in a statistical test similar to the t-test, which we call it HBI t-test. Specifically, it is possible to show that the posterior over the ith group parameter in model k, μki, takes the form of standard Student’s t-distribution centered at the corresponding group mean, aki, with n k = 1 + N¯ k as degrees of freedom. The resulting t-value takes an intuitive form:
t = μ k i − a k i s k i / n k ,
where ski is the empirical deviance statistics for the ith parameter of model k. Therefore, s k i / n k plays the role of standard error, which we call it hierarchical error. Note that the degrees of freedom of the test depend on the number of subjects (i.e. evidence) in favor of model k given by N¯ k, not the total number of subjects. Other group statistics, aki and ski, are also weighted according to the responsibilities of model k in generating data of each subject (as formally obtained in Materials and methods). Using this marginal distribution for population-level group parameters, the HBI t-test enables researchers to determine whether a parameter is significantly different from an arbitrary value, notably 0. For example, the parameter is significantly different from 0 at P < 0.05 if 0 does not fall within the 95% credible interval.
In this section, we apply the proposed HBI method to synthetic datasets and compare its performance with that of HPE, as well as with a non-hierarchical inference (NHI) method estimating parameters for each subject independently according to some fixed, a priori Gaussian priors [20–23]. Importantly, these methods differ in their statistical assumptions about the generative process of data. The NHI assumes no hierarchy in parameter estimation. We then used the individual-level evidence approximated by the NHI (S1 Text) to subsequently perform random effects model comparison using the procedure introduced by Stephan et al. [13, 19]. This means that whereas the NHI procedure assumes no hierarchy across parameters, it does (via the Stephan procedure [13]) allow for a hierarchical structure over model identity. In contrast, the HPE procedure, as introduced by Huys et al. [10, 11], assumes a hierarchy over parameters, but no hierarchy over model identity: we accordingly, use it with a fixed-effects model comparison procedure. The HBI assumes that both parameters and model identities are generated hierarchically in turn. Note that related approximations, as similar as possible, have been used for making inference in these methods, which allows for a fair comparison (S1 Text) since our main points concern the statistical structure of the methods, not the estimation techniques. In particular, HPE builds upon NHI’s Bayesian inference of per-subject parameters to condition these on additional group level parameters, by using expectation-maximization [14]; and HBI extends that algorithm to condition these on an additional level of model identity variables, by using variational Bayes [16, 17]. We also use the same (Laplace) approximation to marginalize the subject-level variables in all three methods. The HBI algorithm has been given in Materials and methods and details of implementing the NHI and HPE have been given in S1 Text. The details of simulation analyses and parameters used in simulations have also been given in S1 Text.
The HBI is general and could be applied to any type of data, such as choice data, reaction times, physiological signals and neural data. Since we are primarily interested in models of choice data, we focus on decision-making experiments.
We then applied the HBI method to an empirical choice dataset from 31 subjects performing the two-step Markov decision task. The data used for this analysis have been reported elsewhere [28]. In Fig 13, we have plotted protected exceedance probabilities of each model, model frequencies and estimated group means and corresponding hierarchical errors. According to this analysis, the hybrid model is the most likely model across the group.
We also performed further analysis testing whether individual differences found by the HBI generalize to individual differences in conceptually related, yet independent, data. We reasoned that subjects showing a hybrid strategy might be slower in their choice, as the hybrid model requires combining of model-based and model-free values (which in some trials might be in conflict). Therefore, we looked at the median of response time across all first-level choices for each subject and tested whether there is a difference in response times between those subjects who (according to the separate analysis of choices) employed a hybrid strategy vs. those who employed a model-based strategy as estimated by the HBI. The subgroup attributed to the hybrid model by the HBI showed slower response time compared to those subjects attributed to the model-based account (P = 0.03, Wilcoxon test). These results suggest that HBI reveals meaningful individual differences generalizing to unseen data.
We applied the HBI to another choice dataset of Parkinson’s disease (PD) patients (N = 31), who performed a probabilistic reward and punishment learning task with binary choices (160 trials), which has been used previously for studying maladaptive learning in PD patients. All patients tested on medication. The dataset used here has been reported elsewhere [15].
Previous studies proposed that positive and negative prediction errors might be communicated through different dopaminergic receptors or striatal pathways [25, 26, 29], and thus the PD patients might have different learning rate parameters for learning from positive and negative prediction errors [29]. Therefore, we considered a model space including the RL model, the dual- RL model and a simple strategy that selects actions based on the most recent outcome. In both RL models, we also included a perseveration parameter, which models the tendency to repeat or avoid the same choice regardless of the value [15, 30]. This analysis showed that the dual-α RL model was more likely across the group. Protected exceedance probabilities, model frequencies and estimated group means and corresponding hierarchical errors are plotted in Fig 14A. We then considered data from matched control participants (N = 20), who performed the same task. The analysis with the HBI showed that the RL model is more likely for the control group (Fig 14B), suggesting that PD (or dopaminergic medication in PD) increases the discrepancy between the learning rates for positive and negative prediction errors. We finally performed a permutation test to formally test the significance of this difference (1000 permutations). For each permutation, all participants were randomly divided into control and PD groups with the same size as the real control and PD groups. The HBI was then used to fit the same model space to each random group. The relative model frequency statistics (RL vs. dual-α RL) was quantified for each permutation. This permutation test confirmed that the dual-α RL was significantly more likely than the RL model in PD patients compared with controls (P <0.001).
In this work, we have introduced a novel method, a hierarchical and Bayesian inference framework, for parameter estimation and model comparison. The HBI framework is hierarchical in the sense that parameters at the individual level are regularized by statistics across all individuals in the group. The HBI framework is Bayesian in the sense that all uncertainties at both individual and group levels are represented by probability distributions. The HBI framework has major theoretical advantages over current state-of-the-art methods, mainly because it combines, in a single hierarchical structure, two sorts of inference (about model identity and model parameters), which are interdependent but have previously been treated separately. Our simulation results demonstrated these advantages experimentally.
In this work, we took an empirical Bayes approach [31, 32], in which priors are constructed based on data. In other words, parameters at the individual level are regularized by statistics across all individuals in the group. Furthermore, we took a so-called random effects approach to model identity [13], which indicates that different models might underlie data in different subjects. This is in contrast to previous hierarchical methods for model fitting, which assume the same model underlie data in all subjects (fixed effects assumption [10, 11]). The random effects approach to hierarchical inference has important consequences for both parameter estimation and model comparison. Moreover, we took a fully Bayesian approach by quantifying uncertainty at the group level, which enabled us to develop statistical tests about group parameters and to quantify corresponding statistical errors.
Empirical Bayes methods play an increasing role in modern statistics. These methods essentially take a hierarchical approach, by assuming that individual data are generated based on the probabilistic properties of the population. This hierarchical approach has important consequences. The most important consequence is that they provide a promising solution to the classical problem of priors in Bayesian statistics by providing informative, yet objective, priors at the individual level. Furthermore, by partly sharing parameters across subjects, they reduce overfitting relative to non-hierarchical models, which in turn allows them to confidently fit more complex models with a smaller penalty for overfitting. This is because non-hierarchical methods assume that the extra parameters of a complex model are independent. For example, consider a model space in which the more complex model has one extra free parameter and there are 40 subjects in the dataset. Fitting the dataset with the complex model using non-hierarchical methods introduces 40 additional independent free parameters, driving the danger of overfitting, and accordingly an excessive penalty to account for this possibility in assessing the evidence for the model. The hierarchical approach, however, assumes that the individual parameters are dependent, as they are all generated according to the same distribution, sharing a single mean parameter and smaller deviations from it. Modeling this hierarchical dependency enables those methods to avoid penalizing complex models as excessively. Our simulation results demonstrate this point experimentally (Fig 3D). While the non-hierarchical method failed to select the correct model with one additional parameter, evidently because the overfitting penalty was too extreme, the HBI was successful in selecting the correct model (Fig 3D).
The HBI method introduced in this paper is built based on the random effects view that different models might underlie data in different subjects. Taking this view enabled us to address problems caused by taking the model identity as a fixed effect in some hierarchical parameter estimation procedures. For parameter estimation, the fixed effects assumption biases the group parameters because it assumes that all subjects contribute equally to the group parameters. The proposed HBI framework solves this problem by weighting contribution of each subject to group statistics by the degree to which that model is likely to be the true underlying model for that subject (Figs 1 and 3). For model comparison, the fixed effects assumption leads to oversensitivity to outliers as the evidence across the group is driven by the sum of individual evidence. Our simulation results (Fig 2) showed that only a few outliers lead to incorrect model selection inference made by the fixed effects assumption. The proposed HBI method solves this problem by normalizing individual evidence across all candidate models. Specifically, the HBI framework quantifies the responsibility of each model k in generating each subject data, a metric lying between 0 and 1. For every subject, the responsibility sums up to 1 across all candidate models as it partitions probability space among those models (see [13, 19] for a similar non-hierarchical approach). It is then easy to compare models by enumerating responsibilities across the group in favor of each model or by estimating the most likely model.
Another major contribution of this paper is to provide a statistical test, HBI t-test, to the inference problem at the group level using hierarchically fitted parameters. For models fitted by a non-hierarchical method, such as maximum likelihood or Laplace approximation, it is statistically valid to use classical statistical tests on fitted parameters to make inference at the group level. However, for datasets fitted by a hierarchical method in which the individual fits are regularized according to statistics of the group data, conventional statistical tests are not valid, because the parameter estimates are non-independent from subject to subject. Our fully Bayesian approach enabled us to address this issue. Our method provides an intuitive solution to this problem in the form of a t-statistic, in which all the group statistics are computed according to the estimated responsibilities of the corresponding model in generating each individual data. Thus, the HBI quantifies the uncertainty of the group parameters and thereby the corresponding hierarchical errors. Our analysis showed that the HBI performed better than both the NHI and HPE in detecting true effects and also that it was well calibrated, displaying the appropriate number of false positives when effects were absent. Therefore, the HBI framework enables researchers to make statistical claims about parameters at the group level.
It is important, however, to note that the foundation of the HBI t-test is completely different from the classical t-test, as it is a Bayesian (in contrast to frequentist) test using posterior probabilities. In particular, this test is based on the posterior distribution of the statistics of interest (i.e. group mean) marginalized over all other parameters (e.g. group variance), which is given by a Student’s t-distribution (Eq 24). Statistically, the precise claim of the HBI t-test is that whether a specific point is outside of a credible interval, which is the interval that the group parameter value falls with a particular subjective probability. For example, if the HBI t-test indicates that a parameter is significantly different from 0 at P <0.05, it means that 0 does not fall within the 95% credible interval. One important difference between Bayesian credible intervals and classical (frequentist) confidence intervals (used in classical Student’s t-test) is that Bayesian credible intervals depend on priors. However, since we used minimally-informative priors (statistically proper priors with very little effects on posteriors, see Materials and methods), the HBI t-test almost entirely depends on data. In fact, that is the reason that under the null, the HBI t-test generates p-values uniformly as shown by simulations (Fig 11). Notably, the same Student distribution can also be used to accept the null hypothesis for example using a “region of practical equivalence” procedure described by Kruschke [33]. It is also possible to employ the more common way and make inference in favor of the null hypothesis using model selection. In this case, one needs to perform a model selection between a model in which the group-level mean of the parameter of interest is fixed at the null value (null model) and compare that with a full HBI with no restriction (alternative model) using Bayes factor (i.e. difference log model evidence).
In addition to model comparison, the HBI framework can also be used for model selection in situations where the goal is to select one of the models as the best model across the group. Exceedance probability is a metric proposed [13] to perform model selection using a random effects approach. An important revision of this metric called protected exceedance probability [19] also takes into account the null possibility that none of the models in model space is supported sufficiently by data, i.e. the differences in model evidence are due to chance. As the HBI framework treats model identity as a random effect, it is possible to compute exceedance and protected exceedance probabilities (Eqs 26–28). Note that if this procedure indicates that models’ ability to explain data are not different (i.e. their difference is likely to be due to chance), one cannot rely on estimated parameters, as those are also dependent on estimated model frequencies. In this situation, we recommend to obtain parameters by fitting models separately to data using the HBI, which makes sense as there is no evidence that models are differently expressed across subjects. In our analysis with simulated and empirical data, however, we never encountered this situation as probability of the null (P0 in Eq 28) was always very small.
In this study, we compared the performance of the HBI with two alternative methods with different statistical assumptions about the generative process of data. The NHI assumes a hierarchy in model identity for generating individual data. The HPE assumes that parameters are generated in a hierarchical fashion, but assumes no hierarchy regarding model identities. The HBI assumes that both model identity and parameters are generated hierarchically. Importantly, the inference procedure for all these methods is very similar, which allows a fair comparison of them largely based on their statistical assumptions. In particular, the three methods all employ Laplace approximation for making a quadratic approximation of individual-level posteriors. Furthermore, the HBI is based on variational Bayes, which is an extension to the case of multiple latent variables of the expectation-maximization procedure used previously for implementing the HPE [10, 11] (see also [34] for a variational implementation), which itself extends the one-level Bayesian inference of NHI. There are other ways to make an inference, for example using Markov chain Monte Carlo methods. Future studies should investigate the pros and cons of those methods, compared with the variational Bayes used here, for making inference in HBI.
There are increasing efforts to exploit advances in computational modeling for understanding mental disorders [3–6]. Recent works, however, have started to tackle challenges related to quantifying uncertainty in diagnosis and also in the evaluation of treatment effects. For example, hierarchical unsupervised generative modeling, have used Monte-Carlo and variational methods to identify a cluster of subjects showing similar patterns of neural connectivity [35, 36]. HBI also offers a promising solution by quantifying uncertainty in model attribution to individuals. Our simulation analyses showed that the HBI performs better than other alternatives in model attribution. This can help us to move towards better diagnosis and precise evaluation of different treatments [37].
In summary, the HBI framework proposed in this work rests on a hierarchical view of both hypothesis testing (i.e. model comparison) and parameter estimation for multi-subject studies and thus provides a generic framework for statistical inference. Moreover, the HBI framework runs fully automatically and it does not rely on hand tuning of parameters. Therefore, we expect this method to be useful for a wide range of studies testing different hypotheses in a multi-subject setting. This includes not only computational models of learning and decision making but also any statistical models of brain or behavior.
Here, we give a formal treatment of the HBI framework in seven sections, in which we 1) define the probabilistic model underlying HBI; 2) lay out the basis of our variational approach for making inference (the full proof is given in S1 Appendix); 3) present the HBI algorithm; 4) derive the HBI t-test; 5) show how HBI can be used for making inference about a new subject; 6) define important practical points, in particular prior parameters, initialization and convergence criteria; 7) give a formal definition of the exceedance and protected exceedance probability. The HBI and its manual are freely available online as part of computational and behavioral modeling (cbm) toolbox: https://payampiray.github.io/cbm.html.
We begin by describing the probabilistic model of the HBI. Consider an observed dataset X = {x1, …, xN} where xn is the dataset (e.g. choices) of nth subject and N indicates the number of subjects and a model-space including K candidate models, M1…MK. Moreover, suppose that the prior probability of each model in the population is given by m = {m1, …, mK}. For each dataset, xn, we assume that there is a latent variable zn comprising a 1-of-K binary random vector, in which zkn is one if xn generated is by the kth model. Thus, the probability of the latent variable across all subjects, Z = {z1, …, zN}, is assumed to have a multinomial distribution,
p ( Z | m ) = ∏ n ∏ k m k z k n . (1)
Each model Mk in the model-space is supposed to compute the probability of a given dataset (e.g. a set of choices) given a set of parameters, hkn. For example, the reinforcement learning model computes the probability of choices using two parameters: a learning rate and a decision noise parameter. The number of models and their structures depend on specific scientific questions. Here, we take a general approach by making no specific assumption about the number of models, K. Thus, the kth model in the model-space, Mk, computes the probability of dataset xn given the parameter vector hkn, which is denoted by p(xn|hkn, Mk). Note that the number of parameters in model k, denoted by Dk, might be different across models. Since data for each subject is generated by one of the models, which is denoted in the binary vector zn, the probability of the observed dataset given the model-space is
p ( X | H , Z ) = ∏ k ∏ n p ( x n | h k n , M k ) z k n , (2)
where H denotes all the parameters across all participants and models. The parameters of kth model are assumed to have a multivariate normal distribution with mean μk and precision matrix Tk,
p ( H | Z , μ , T ) = ∏ k ∏ n N ( h k n | μ k , T k − 1 ) z k n , (3)
where Tk is a diagonal matrix with positive elements.
We also introduce a distribution over model frequencies, m. We use the Dirichlet distribution, which forms the conjugate prior for the multinomial distribution, as the prior:
p ( m ) = Dir ( m | α 0 ) = C ( α 0 ) ∏ k = 1 K m k α 0 − 1 , (4)
where C(α0) is the normalizing constant for the Dirichlet distribution.
We also take group parameters μ and T as random variables, which allows us to evaluate their posterior distribution given data. We introduce conjugate priors for these variables, a Gaussian-Gamma prior in which the distribution over μk depends on Tk:
p ( μ | T ) = ∏ k = 1 K N ( μ k | a 0 , ( b T k ) − 1 )
p ( T )= ∏ k = 1 K ∏ i = 1 D k G ( τ k i | v , s ) ,
where G ( . ) denotes Gamma distribution. Here, τki is the ith diagonal element of Tk. Assuming that τk is a vector containing τki, by defining Tk = diag(τk), in which diag(.) is an operator outputting a diagonal matrix with elements given by τk, we can write these two equations in a compact form:
p ( μ , τ ) = ∏ k = 1 K N ( μ k | a 0 , diag ( b τ k ) − 1 ) G ( τ k | v , s ) , (5)
where we have defined:
G ( τ k | v , s ) = ∏ i = 1 D k G ( τ k i | v , s ) ,
in which v is a scalar and s is a vector with Dk elements all equal to s. The full probabilistic model is given by,
p ( X , H , Z , μ , τ , m ) = p ( X | H , Z ) p ( H | Z , μ , τ ) p ( Z | m ) p ( μ | τ ) p ( τ ) p ( m ) . (6)
The task of Bayesian inference is to compute the posterior probabilities of latent variables given data, p(H, Z, μ, τ, m|X). Since the inference is intractable for the probabilistic model outlined in the previous section, we employ variational inference to compute approximate posteriors. We take a so-called mean-field approach [16, 17] by assuming that the posterior is partially factorized as follows:
q ( H , Z , μ , τ , m ) = q ( H , Z ) q ( μ , τ , m ) . (7)
Note that we force no factorization in the posterior between latent variables, Z and H. Using a quadratic approximation of the conditional posterior, q(H|Z), we prove in S1 Appendix that these posteriors are given by,
q ( H , Z ) = ∏ k ∏ n r k n z k n N ( h k n | θ k n , A k n − 1 ) z k n (8)
q ( μ , τ , m ) = Dir ( m | α ) ∏ k q ( μ k , τ k ) (9)
q ( μ k , τ k ) = N ( μ k | a k , diag ( β k τ k ) − 1 ) G ( τ k | ν k , σ k ) , (10)
where 0 ≤ rkn ≤ 1 is the responsibility of model k for nth subject, θkn and Akn are the subject-level mean and precision, νk and βk are scalars and σk is a vector with the same size as τk. In the next section, we provide the HBI algorithm, which iteratively updates the parameters of these distributions, rkn, θkn, Akn, α, ak, νk, βk, and σk.
After initializing the individual parameter estimates, θkn and Akn and responsibilities rkn for all subjects and models, as well as setting prior parameters a0, b, s, v and α0 (which will be defined later), the HBI algorithm performs these steps:
Calculate the summary statistics:
N ¯ k = ∑ n r k n (11)
θ ¯ k = 1 N¯ k ∑ n r k n θ k n (12)
V ¯ k = 1 N¯ k ∑ n r k n ( θ k n θ k n ⊤ − θ ¯ k θ ¯ k ⊤ + A k n − 1 ) . (13)
Update parameters of q(μ, τ, m) for all models:
a k = 1 N ¯ k + b ( N¯ k θ ¯ k + b a 0 ) (14)
β k = b + N¯ k (15)
σ k = s + 1 2 diag ( N ¯ k V ¯ k + b N ¯ k b + N ¯ k ( θ ¯ k − a 0 ) ( θ ¯ k − a 0 ) ⊤ ) (16)
ν k = v + 1 2 N¯ k (17)
α k = α 0 + N¯ k . (18)
Update the individual posterior parameters θkn, Akn and fkn, by obtaining a quadratic approximation of the function, ℓkn(h), with respect to h:
ℓ k n ( h ) = p ( x n | h , M k ) N ( h | E [ μ k ] , E [ T k ] − 1 ) , (19)
where E [ μ k ] = a k and E [ T k ] − 1 = 1 ν k diag ( σ k ). This approximation can be written as
ℓ k n ( h ) ≃ f k n exp ( − 1 2 ( h k n − θ k n ) ⊤ A k n ( h k n − θ k n ) ) . (20)
Note that any quadratic approximation can be used here. For example, using a Laplace quadratic approximation (which is a very common approximation for analyzing behavioral and neural data [20, 22, 23]), θkn, Akn and fkn are given by the mode, Hessian of log ℓkn and the maximum value of ℓkn, respectively:
θ k n = arg max hlogℓ k n ( h )
A k n = − ∇ ∇ logℓ k n ( h ) | θ k n
f k n = ℓ k n ( θ k n ) .
Update responsibilities,
r k n = ρ k n ∑ j = 1 K ρ j n , (21)
where
logρ k n = logf k n + 1 2 D klog2 π − 1 2 log| A k n | + λ k + E [ logm k ] (22)
λ k = D k 2 ( ψ ( ν k ) − logν k − 1 β k ) (23)
E [ logm k ] = ψ ( α k ) − ψ ( ∑ k = 1 K α k ) ,
in which ψ(.) is the digamma function.
Terminate if stopping criteria are met, otherwise go to 1.
An important goal of computational modeling studies is to compute the distribution of parameters given data across the whole population. From a Bayesian viewpoint, this is given by the marginal posterior over the mean of group parameters, μk, which reads
p ( μ k | X ) ≃ ∫ q ( μ k , τ k ) d τ k = ∫ N ( μ k | a k , ( β k τ k ) − 1 ) G ( τ k | ν k , σ k ) d τ k = S t ( μ k | a k , η k , n k ) ,
where n k = 2 ν k = 2 v + N¯ k is the number of degrees of freedom of the Student distribution and η k = ν k β k σ k − 1 is the inverse-scale parameter. Therefore, the random variable t = η k 1 2 ( μ k − a k ) takes a form of standard Student distribution with nk degrees of freedom. By defining s k i 2 = 2 β k σ k i, in which s k i 2 corresponds to empirical variance (c.f. Eq (16)), we can write this result in an intuitive form,
p ( μ k i | X ) = S t ( μ k i − a k i s k i / n k | n k ) . (24)
Noting the similarity between s k i / n k and the standard error of the mean, we called s k i / n k the hierarchical error. Note that if we assume v = 0.5 (which is reasonable as explained later), we obtain n k = 1 + N¯ k.
In many situations, researchers are interested to fit a new dataset to a particular model and find corresponding parameters. In Bayesian statistics, this is called the predictive distribution and it is given by marginalizing over group parameters. Suppose that x* and h k * denote the new dataset and its corresponding parameters for model k. The marginal distribution p ( x * , h k * | z k * = 1 , X ) is the predictive distribution given the observed dataset X assuming that the new data is generated by the kth model. This distribution is given by:
p ( x * , h k * | z k * = 1 , X ) = ∫ p ( x * | h k * , M k ) p ( h k * | μ k , τ k , z k * = 1 ) p ( μ k , τ k | X ) d μ k d τ k = p ( x * | h k * , M k ) S t ( h k * | a k , ( 1 + β k ) − 1 η k , n k ) ,
where ηk and nk have been defined in the previous section. This distribution can also be written in terms of standard Student distribution with nk degrees of freedom. Furthermore, if we assume that b = 2v, which is a reasonable assumption (see the next section), this distribution is given by
p ( x * , h k * | z k * = 1 , X ) = p ( x * | h k * , M k ) S t ( diag ( s k ) − 1 ( h k * − a k ) | n k ) ,
where sk is a vector of corresponding empirical deviance parameters, defined in the previous section. Using this joint distribution, one can use sampling methods to obtain the posterior over parameters, p ( h k * | z k n = 1 , X , x * ), or to obtain the maximum-a-posteriori parameters, θ k *, given by
θ k * = arg max hlogp ( x * | h , M k ) S t ( diag ( s k ) − 1 ( h − a k ) | n k ) . (25)
Note that for many degrees of freedom due to large values of N¯ k, the Student distribution tends to a Gaussian with mean ak and deviance matrix diag(sk). However, small values of N¯ k lead to a small number of degrees of freedom and heavier tailed distributions than Gaussians, which are more robust against outliers.
As the mean-field variational inference is an iterative framework, it also depends on the initialization of the parameters. In this section, we provide priors that do not bias the final solution and also provide some intuitive criteria for the initialization.
We initialize the parameters θkn and Akn by fitting all models separately to all participants (with some initial Gaussian prior), i.e., assuming as if zkn = 1. These values are then used to calculate summary statistics according to Eqs (11)–(13).
Furthermore, we need to define prior parameters. The free parameter α0 indicates prior frequency of each model. We take uninformative priors on frequency of models, which is given by α0 = 1 for all models. The prior mean, a0k, is assumed to be zero. Given Eq (15), we see that b can be interpreted as the effective number of prior samples associated with models. Also, given Eq (17), v could be interpreted as the half of the effective number of prior samples associated with models. Assuming that the priors account for one sample, which is a common assumption in Bayesian statistics, we take b = 1 and v = 1 2. Finally, since s has always an additive effect on σk according to Eq (16), we assume a small positive value for s, allowing that σk to be driven dominantly by data. In all our analyses, we assumed s = 0.01. It is also important to note that by choosing a small value for s, we ensure that if a model loses entirely (takes no responsibility), its corresponding parameters at the individual level converge to the prior mean, a0k, with a very small variance.
Finally, the HBI algorithm presented above requires stopping criteria. In our analyses, we terminated the algorithm if the change in normalized value of parameters between two consecutive iterations, j − 1 and j, defined as
d ^ = 1 K ∑ k 1 D k ∑ i ( θ ^ k i j − θ ^ k i j − 1 ) 2 ,
was smaller than 0.01. Here, θ ^ k i j is defined according to summary statistics of parameters on the jth iteration:
θ ^ k i j = θ ¯ k i / V ¯ k i 1 2 ,
where θ ¯ k i and V ¯ k i are the ith element of θ ¯ k and V ¯ k defined in (12 and 13), respectively. In our analyses, we also set 50 as the maximum number of iterations, although almost always the algorithm stopped before hitting this number.
Using the posterior over m, one can also derive the so-called exceedance probability and protected exceedance probability, as defined in previous works [13, 19]. We reproduce the equations here for completeness.
The exceedance probability of kth model, ϕk, is defined as the probability that model Mk is more likely than any other model in the model-space and it is given by
ϕ k = Prob ( m k > m j | α ) , ∀ j ≠ k . (26)
Computing protected exceedance probabilities, as defined in [19], also requires to run the HBI under the (prior) null hypothesis, H0, that there is no difference between models (i.e. α0 → ∞). The alternative hypothesis, H1, is the original case, in which α0 = 1. If we define L and L0 as the log-likelihood (actually the variational lower bound as its approximation) of all data given the model-space under H1 and H0, respectively, then the protected exceedance probability of kth model, ϕ ˜ k, is defined as:
ϕ ˜ k = ϕ k ( 1 − P 0 ) + 1 K P 0 , (27)
where
P 0 = 1 1 + exp ( L − L 0 ) . (28)
Note that if P0 is close to 1, then model frequencies should be ignored, as the difference between models in the model space is due to chance. Furthermore, if data does not support any model, i.e. P0 is close to 1, then parameters should be estimated by fitting each model separately using the HBI.
|
10.1371/journal.pntd.0001887 | Serum Proteome and Cytokine Analysis in a Longitudinal Cohort of Adults with Primary Dengue Infection Reveals Predictive Markers of DHF | Infections caused by dengue virus are a major cause of morbidity and mortality in tropical and subtropical regions of the world. Factors that control transition from mild forms of disease such as dengue fever (DF) to more life-threatening forms such as dengue hemorrhagic fever (DHF) are poorly understood. Consequently, there are no reliable methods currently available for early triage of DHF patients resulting in significant over-hospitalization.
We have systematically examined the proteome, cytokines and inflammatory markers in sera from 62 adult dengue patients (44 DF; 18 DHF) with primary DENV infection, at three different times of infection representing the early febrile, defervescence and convalescent stages. Using fluorescent bioplex assays, we measured 27 cytokines in these serum samples. Additionally, we used multiple mass spectrometry methods for iTRAQ-based comparative analysis of serum proteome as well as measurements of protein adducts- 3-nitrotyrosine and 3-chlorotyrosine as surrogate measures of free radical activity. Using multiple methods such as OPLS, MRMR and MSVM-RFE for multivariate feature selection and classification, we report molecular markers that allow prediction of primary DHF with sensitivity and specificity of >80%.
This report constitutes a comprehensive analysis of molecular signatures of dengue disease progression and will help unravel mechanisms of dengue disease progression. Our analysis resulted in the identification of markers that may be useful for early prediction of DHF during the febrile phase. The combination of highly sensitive analytical methods and novel statistical approaches described here forms a robust platform for biomarker discovery.
| While the majority of patients who exhibit febrile dengue infection recover within a week, a small proportion of the patients progress to develop severe symptoms that can be life-threatening if not managed in a hospital setting. Because there is no method to accurately identify this subgroup of patients, many dengue patients are hospitalized unnecessarily, which causes significant burden to the healthcare system. In our study, we have systematically measured a large number of molecules including cytokines and serum proteins in blood samples from a dengue patient cohort using highly sensitive mass spectrometry-based methods. We have further developed novel statistical methods that allow us to identify small panels of measureable blood markers, which can distinguish dengue patients that develop milder, self-limiting form of the disease from those that progress to develop severe symptoms. Because these markers can be applied within 48–72 hours of onset of febrile symptoms, we expect them to be useful for early classification of severe dengue disease.
| Infection with dengue virus (DENV) causes a spectrum of clinical manifestations ranging from mild dengue fever (DF) to the potentially lethal dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [1]. In humans, the major cellular targets of dengue appear to be dendritic cells of the skin, macrophages and monocytes [2]. Dengue is endemic to the tropical and sub-tropical regions of the world, which are home to over half the population of the world as well as being popular tourist destinations. It has also emerged in new areas such as south Florida and Mediterranean France. With a significant proportion of the world population at risk of infection annually, coupled with the absence of a licensed vaccine, dengue is emerging as a global health concern.
The majority of dengue patients recover uneventfully after 5–7 days of acute illness. In a small proportion of patients, however, the initial febrile period is followed by a rapid onset of vascular leakage, thrombocytopenia and hemorrhage indicating DHF. The continual loss of intravascular volume from plasma leakage can very rapidly lead to hypotension and cardiovascular collapse which, if not carefully managed, can result in death. In the absence of an effective antiviral drug, the management of dengue patients is primarily supportive. Early recognition of patients with plasma leakage is thus critical for the initiation of appropriate fluid management to prevent onset of hypovolemic shock. However, because these symptoms become evident only in the critical phase of infection, it is currently not possible to distinguish DF and DHF accurately during the early stages of illness, when the disease is less well differentiated [3].
The mechanisms that trigger transition from mild DF to more life threatening DHF are poorly understood, hampering early classification of dengue patients who will progress to DHF. This not only delays treatment but frequently results in the over-hospitalization of patients contributing significantly to the financial burden imposed by dengue [4], [5]. The availability of reliable markers that predict DHF during the early stages of infection could be useful in triaging patients for management.
In the last decade, numerous efforts have been made to identify serum markers that may predict severe dengue disease, with an emphasis on cytokines [6]. A number of studies propose that innate immune cytokines (e.g. IFN-α, IL-8) are elevated during the early febrile phase while adaptive immune cytokines (e.g. TNF-α, IL-10, IFN-γ) appear to increase during the defervescence phase of dengue infection [6]. Several factors have traditionally limited the usefulness of these studies in biomarker development. Firstly, the highly variable nature of patient cohorts (e.g. pediatric versus adults; ethnicity) used makes it difficult to compare the results of these various studies. Secondly, most studies have examined ‘case versus control’ type of sample population instead of longitudinal studies to distinguish ‘predictors’ from ‘indicators’. Finally, a lack of follow-up in larger population base to test the prognostic potential of proposed markers limits their clinical application.
The early dengue infection and outcome (EDEN) study in Singapore prospectively recruits and follows-up adult dengue patients in Singapore through early febrile, defervescence as well as convalescence stages [7] of the disease. This makes this longitudinal study highly suited for the identification of prognostic markers of severe dengue disease. In this study, we report a systematic characterization of serum cytokines, proteome, and markers of macrophage and neutrophil activity in a subset of adult dengue patients with primary dengue infections obtained from the EDEN cohort. In addition to identifying molecular signatures of disease progression, we describe a comprehensive multivariate statistical analysis to identify serum markers for early prediction of DHF.
The EDEN study is a multi-center longitudinal study of adult febrile infections that was carried out at a number of clinics island-wide in Singapore. Enrollment of eligible individuals was based on written informed consents and the protocols were approved by the National Healthcare Group (DSRB B/05/013). The study protocols have been described earlier [7]. In brief, adult patients (>21 years) presenting with acute onset fever (≥38.0°C for less than 72 hours) without rhinitis or other clinical alternatives were included in the study. Initial dengue diagnosis and viremia levels were determined by real time RT-PCR using a previously described method [8]. This was followed by serology and subsequent serotyping by virus isolation and immunofluorescence using serotype specific monoclonal antibodies (ATCC: HB46-49). Venous blood samples were also collected at fever day 4 to 7 (visit-2) and weeks 3 to 4 (visit-3), aliquoted and frozen at −80°C. ‘Fever day’ here refers to number of days post onset of fever. Classification of DF or DHF was made based on the guidelines provided by the WHO [9]. In brief, acute febrile patients positive for dengue with one or more of the following: headache, retro orbital pain, myalgia, rash, leucopenia, hemorrhage were classified as DF while patients with fever lasting 2–7 days combined with bleeding, thrombocytopenia (<100,000/mm3) as well as evidence of plasma leakage shown by a 20% or greater rise in hematocrit relative to the blood sample obtained at convalescence or pleural effusion on chest X-ray were classified as DHF. Of the 133 dengue patients that were finally enrolled in this study (September 2005–October 2006), 62 patients (44 DF, 18 DHF) tested negative for dengue IgG antibodies in the acute sera, using a commercial ELISA kit (PanBio, Brisbane, Australia). These patients were deemed to have primary DENV infection, all of which were included in this study.
A detailed hematological and virological analysis was performed and a subset of 15 clinical indicators was selected for our statistical analysis. These included white blood cell count (WBC), red blood cell count (RBC), blood hemoglobin (HGB), hematocrit (HCT), macrophage cell volume (MCV), MCH, MCHC, platelet count (PLT), lymphocyte percentage (LYMPH%), lymphocyte count (LYMPH), mixed cell distribution (MXD), neutrophil percentage (NEUT%), neutrophil count (NEUT), red blood cell distribution width-coefficient of variation (RDW-CV), and viral titers. Additionally, we used plasma samples from 50 asymptomatic healthy army recruits collected during their annual physical examination in Singapore as controls in our analyses. A comparison of cytokines in dengue patient sera with healthy plasma is shown in supplementary data (Figure S1 and Table S1). This study was approved by the National University of Singapore Institutional Review Board and samples were collected with individual informed written consents (see checklist S1).
Cytokine measurements were performed with 12.5 µl sera in duplicates using the Bioplex 27-plex human cytokine kit from BioRad as per manufacturer's instructions. The standard curves were optimized automatically by the software (Bioplex manager) and verified manually. The Bioplex manager software was used to calculate cytokine concentrations and only measurements that showed a coefficient of variability (CV) of <10% were included for further analysis. Levels of interferon-induced cytokine IP-10 in 30% of the dengue patient samples during visit-1 were above upper limit of detection. We repeated the analysis after diluting the serum 100 fold for this subset of samples. Six of the visit-1 samples in DF group and 3 in DHF group still had very high levels of IP-10 and for the statistical analyses, we included these as missing values since levels of other cytokines for these samples were within detectable range. Measurement of 9 acute phase proteins was performed using the Bioplex Pro Acute phase multiplex kit (BioRad laboratories) as per manufacturer's instructions. Two different dilutions of sera were used-1∶1000 for ferritin (FT), serum amyloid A2 (SAA), procalcitonin (PCT), tissue plasminogen activator (tPA), fibrinogen (FB) and 1∶100,000 for alpha-2-macroglobulin (A2M), haptoglobin (HPT), C-reactive protein (CRP) and serum amyloid P (SAP).
Nitrotyrosine (NT) and chlorotyrosine (CT) in human serum were measured by a liquid chromatography-triplequadrupole MS method. Briefly, 2 mg of serum protein was spiked with 4 pmol internal standards (IS) L-3-chloro-[13C9, 15N]-tyrosine and L-3-nitro-[13C9, 15N]-tyrosine, and digested in the sodium acetate solution 0.1 M (pH 7.4) with 0.4 mg pronase E (freshly treated by the size-exclusive micro bio-spin column). The mixture was incubated at 50°C for overnight (16 hrs.) and filtered by Vivospin500 3KMW centrifuge filter at 15,000 rpm to remove undigested protein. The amino acids were further purified by Agilent 1200 series HPLC system (Waldbronn, Germany) on an Xbridge TM Phenyl column (3.5 µm, 4.6×50 mm, Waters, Milford, MA). The fractions containing nitrotyrosine and chlorotyrosine, together with internal standards, were collected and dried by SpeedVac for subsequent LC/MS/MS analysis. Subsequent mass spectrometry analysis of target compounds involved separation on an Xbridge TM Phenyl column (3.5 µm, 1.0×100 mm, Waters, Milford, MA) online injection into an Agilent 6460 triple quadrupole mass spectrometer. Two microliters of each sample was injected and eluted by isocratic 25% methanol (0.1% formic acid) for 13 min at 15 µL/min. CT along with IS were analyzed by regular multiple reaction monitoring (MRM) as follows: 216/170 (CT) and 226/179 (CT, IS). NT along with IS were measured by modified MS3 based in-source fragmentation as follows: 181/117 (NT) and 190/125 (NT-IS) by elevating the potential to 135 V at the ion source. The limits of quantitation achieved were 8.1 and 7.3 nM for CT and NT, respectively.
We selected 62 adult dengue patients from the EDEN cohort, of which 44 were diagnosed as DF, and 18 as DHF. The patients selected in DF and DHF groups had similar age and ethnic distribution (Table 1). Serotyping analysis indicated that DENV serotypes-1 and -3 were the most common, followed by serotype-2, while no serotype-4 was present. Average duration between fever onset and first sample collection was <48 hours, and the average duration between samples were <80 hours (visit 1 & 2) and <21days (visit 2 & 3) respectively (Table 1).
We examined the key clinical indicators commonly used for the diagnosis of DHF. Blood platelet count dropped significantly from febrile phase to defervescence in both DF and DHF patient groups with DHF patients exhibiting significantly (p<0.05) lower platelet levels during defervescence (visit-2) than DF patients (Figure 1A). DHF patient groups also exhibited significantly (p<0.05) lower WBC and lymphocyte counts especially during defervescence (Figure 1B &C). Viral titer measured at visit-1 was higher in DHF patients (Figure 1D) consistent with several previous studies that have reported higher plasma viral loads in DHF patients [15], [16]. Our study population thus recapitulated most of the hallmark clinical features of dengue progression in DF and DHF during the early febrile, defervescence and convalescence stages of infection.
We measured the levels of 27 serum cytokines in our dengue patient cohort, using a multiplex assay. A majority of cytokines was maximally elevated in dengue patients during the early febrile phase (visit-1) of infection (Table S1, Figure S1). These included IL-1ra, IL-4, IL-7, IL-8, IL-10, IL-12, Eotaxin, G-CSF, IFN-γ, IP-10, MCP-1, MIP-1b and VEGF. Cytokines IL-1b, IL-5, IL-6, IL-9, IL-12, IL-17 and FGF-basic, remained elevated during defervescence (visit-2) and convalescence (visit-3) stages. When compared with plasma samples from an independent cohort of healthy individuals, cytokines IP-10, VEGF and PDGF-BB were found to be elevated >20 times over controls during the febrile phase of infection, while IL-4, IL-9, IL-10 and IL-1ra were elevated by 10–20 times over controls (Table S2). Cytokines IL-6, IL-7, IL-8, Eotaxin, G-CSF, IL-17 and MIP-1b were elevated 4–8 fold (Table S2). These values may however be an overestimation of the actual changes since an independent cohort may not be an ideal control for the study population.
To identify temporal patterns in cytokine flux in patient sera, we performed K-means clustering to group cytokines in DF patients exhibiting similar patterns across the three stages of disease as detailed in the methods section. The cytokine IP-10 was the sole member of cluster-1 (Figure 1E) with very high levels during the febrile phase followed by a rapid decline to near control levels at convalescence. A majority of cytokines fell into a second cluster (Figure 1E and F, Cluster-2) that exhibited a peak at the febrile phase but declined modestly, with levels remaining significantly higher than controls even at the late convalescent stage (visit-3). A third cluster of 7 cytokines (Figure 1E and F, Cluster-3) increased at febrile phase and decreased during defervescence but increased again to peak levels during late stages (visit-3). While overall clustering profile of cytokines was similar between DF and DHF, cytokines IL-1b, IL-4, IL-6, IL-8, IFN-γ, IL-17, G-CSF, VEGF, IP-10, and PDGF-BB (marked by asterisk in Figure 1F) either clustered differently or showed different slopes (not shown) between the disease stages in DF and DHF groups suggesting that there may be changes in the temporal profile of these cytokines.
Overall, our results indicated that cytokines and chemokines associated with innate immune activity (e.g. IFN-γ, IP-10), Th2 cell response (IL-4, IL-10, and IL-13), inflammation (IL-1b, IL-6, and IL-8), chemotaxis of macrophages and neutrophils (Eotaxin, MIP-1b) are all maximally elevated in dengue patients during the early febrile phase. Cytokines IL-12, growth factors FGF and PDGF increased even at convalescence. TNF-α remained below detection levels in our analysis likely because production is transient and missed in our timeline of sample collection. Similarly, levels of IL-2, IL-15, GM-CSF and MIP-1a were below the detection limit in >85% of the samples and were excluded from further analysis.
Differences in temporal profile of a subset of cytokines between DF and DHF patients, identified in the clustering analysis outlined above, prompted us to examine these cytokines more closely across different time points of infection in DF and DHF groups. We observed that DHF patients had lower levels of IFN-γ during febrile phase, a time of peak interferon activity (Figure 2A–B). Although levels of IP-10 (an interferon-induced cytokine) were also lower in the DHF group, this was statistically significant (p<0.05) only at defervescence (Figure 2B). Low levels of IFN-γ as well as IP-10 during the febrile phase point to an attenuated interferon response in DHF patients, which may be associated with diminished viral clearance. There was a marginal but significant correlation between viral titers and IFN-γ levels during the early febrile stage (visit-1, (r = 0.370; p<0.05). The correlation was especially strong between IFN-γ at visit-1 and IP-10 at visit-2 in DHF patients (r = 0.66; p<0.05).
We observed decreased levels of Th2 cytokine IL-4, in DHF patients during the febrile stage, (Figure 2C) compared with DF. Unlike DF patients, IL-1b levels in DHF patients were indistinguishable from healthy controls until the convalescence stage, indicating a depressed IL-1b response (Figure 2D). Levels of IL-17 as well as Granulocyte-Colony Stimulating Factor (G-CSF) were lower in DHF patients especially during the febrile stage (Figure 2E and F). The serum profiles for platelet-derived growth factor (PDGF-BB) as well as vascular endothelial growth factor (VEGF) were similar to G-CSF and markedly lower during the febrile phase in DHF patients compared to DF patients (Figure 2G, H). A similar comparison of other cytokines IL-6 and IL-8 that were found altered in the clustering analysis (Figure 1E &F) indicated that the differences between DF and DHF groups were not statistically significant (data not shown). The number of patients in this study was too low to allow stratification by days from fever onset (Figure S2).
Quantitative proteomics by isobaric tagging of peptides allows multiplexing of biological samples thereby reducing variability while increasing accuracy of protein quantitation [17]. We adopted an iTRAQ-based approach to quantify the serum proteome of pooled dengue patient sera during the different stages of the disease. Overall, we identified 90 proteins with high confidence, and determined their fold-change over control samples, in both DF and DHF patient groups (Table 2). Of a total of 35 proteins that showed a >1.5 fold enrichment or depletion, 25 proteins were unique to DHF patient group while 6 proteins - serum amyloid A2, leucine-rich-alpha-2 glycoprotein, hemoglobin alpha, actin, haptoglobin and alpha-1-antitrypsin, changed in both DF and DHF samples (Table 2). The acute phase reactants were the most abundant class, followed by serpin class of protease inhibitors and complement pathway proteins (Figure 3A). A majority of these proteins were maximally elevated during the febrile phase although some remained high or increased further during defervescence (Table 2). Five proteins were depleted from sera during the febrile and defervescence stage but returned to near control levels during the convalescent stage (Table 2). Overall, the proteomic analysis indicated that the most readily observable predominant serum protein response in dengue infections was the acute phase response.
A major caveat of the sample pooling approach described above is the averaging effect which may result in a gross underestimation of fold changes despite the high accuracy and sensitivity of the proteomic quantification. As an alternative, we used a commercially available multiplex fluorescent-bead based ELISA assay, which simultaneously measures levels of 9 well-known acute phase proteins including two serum proteins (serum amyloid A2 (SAA) and haptoglobin (HPT)) that were identified in our proteomics analysis (Table 2). Using this method, we analyzed individual serum samples from 10 DHF, 24 DF patients and 10 healthy asymptomatic controls. SAA and HPT were elevated in dengue patients during the early febrile (visit-1) and defervescence (visit-2) stages (Figure 3B–C). Other acute phase proteins that were elevated in dengue patients included C-reactive protein (CRP), alpha-2 macroglobulin (A2M) and ferritin (FT) (Fig. 3D–F), while serum amyloid P (SAP), pro-calcitonin (PCT), tissue plasminogen activator (t-PA) and fibrinogen (FB) remained unchanged (not shown). With the exception of SAA, which was higher in DHF patients during the febrile phase, the differences in levels of other proteins between DF and DHF patient groups were not statistically significant.
We used a previously established mass spectrometry based method [18] to measure levels of total serum 3-nitro-tyrosine (NT) and 3-chloro-tyrosine (CT), in 44 DF patients and 10 DHF patients at three different stages of the disease. Compared to healthy individuals where CT and NT levels in sera are below detection, there was a significant elevation of both CT and NT in dengue patient sera (Figure 4). Levels of CT were elevated in all dengue patients during the febrile phase compared to controls, and continued to increase during defervescence and remained high at convalescence (Figure 4A). This suggests that neutrophil activity remains high even after viral clearance. Interestingly, DHF patients displayed higher levels of CT compared to DF patients during the early febrile phase and although higher levels were also seen during defervescence and convalescence, the differences at the latter stages were not statistically significant (at p<0.05). NT peaked during the early febrile phase of the infection but declined to near basal levels during the convalescence stage (Figure 4B). We did not observe statistically significant (p<0.05) differences in NT levels between DF and DHF groups in our experiment.
We adopted a multiple-feature selection strategy to identify subsets of features from among the 47 blood parameters described above that may have predictive value in the identification of DHF during the early febrile phase. By analyzing the various feature classes (i.e. cytokines, serum proteins, protein adducts, and clinical features) measured at the early febrile phase (visit-1), both independently, as well as together we evaluated the relative predictive power of these various molecules. First, we analyzed 23 cytokines and identified a subset of 7 cytokines which displayed sensitivities and specificities >75% (Table 3). A receiver operator characteristics (ROC) curve analysis indicated that this subset performed well with area under curve (AUC) of 0.87±0.05 (Table 3, Figure 5). We next combined 15 laboratory clinical features (listed in the methods section) along with the cytokines and reanalyzed the data. This resulted in a new subset (Table 3) and achieved sensitivities and specificities >80% with an AUC of 0.92±0.03 (Table 3, Figure 5). While cytokines IFN-ϒ, IL-1b, IL-8 and IL-17 were common with subset A, combining them with lymphocyte, platelet counts and viral titers improved the predictive performance of subset-B compared to subset-A (Table 3). The addition of two more features- CT and NT- to the dataset resulted in a new subset that retained cytokines- IFN-ϒ and IL-1b, IL-8, and blood lymphocyte count, but also had additional set of cytokines (Table 3) along with CT. However, overall predictive performance of subset group-C was poorer likely due to reduction of population size (n = 54 as compared with n = 62).
Finally, we expanded the dataset to include all measured features (i.e. 23 cytokines, 5 serum proteins, 2 protein adducts and 15 clinical features). The number of patients in this analysis was much lower (n = 34) than the previous analysis (n = 62 and n = 54) due to further exclusion of samples where the data was incomplete due to missing values. The subset from this analysis included a variety of features including serum proteins (SAA and HPT), cytokines (IFN-ϒ, IL-17) and protein adducts (CT) that achieved a sensitivity and specificity of >75% and AUC of 0.90±0.06 (Table 3).
We have performed a comprehensive molecular analysis of serum molecules in a cohort of adults with primary dengue infections with the objective of identifying predictive markers of DHF. Traditionally, biomarkers studies have relied mostly on case versus control studies (reviewed in [6]) with one sample per patient, collected in a 1–10 day period. Some of these studies have reconstructed temporal profiles via data grouping based on fever day [19], [20], [21]. However, variability in sample size within groups (e.g. fever day) and lack of patient follow-up often result in poor statistical performance and inadequate modeling of individual immune responses. Prospective follow-up of patients across disease stages, although most desirable for biomarker development, are scarce. A good example is a 1997 pediatric dengue study in Thailand, where a positive dengue diagnosis was followed by daily blood sampling till one day post defervescence [22], [23]. The EDEN study combines the convenience of asynchronous patient recruitment during the early febrile phase, with patient follow-up, and is designed to specifically model adult dengue infections [7].
A detailed cytokine analysis indicated that DHF patients are characterized by an attenuated serum cytokine response especially during the early febrile phase. In DHF, low levels of IFN-γ during febrile phase correlated with reduced levels of IP-10, indicating that an inability to mount a timely anti-viral response may result in high viremia. In cell culture models, pretreatment with interferons inhibits dengue viral replication [24] although treatment after infection has no effect due possibly to active inhibition of IFN-signaling pathways by dengue viral protein NS4B [25]. Whether higher viral titers reported in DHF patients is a consequence or cause of an impaired interferon response remains to be confirmed. In recent human challenge studies, development of infection correlated with extremely low or undetectable IFN-γ production by PBMCs suggesting a role for sustained IFN-γ production in protection [26]. An attenuated innate response may in turn affect the kinetics of adaptive immune and pro-inflammatory responses, as suggested by the lower levels of Th2 cytokines IL-4 and IL-13, growth factors G-CSF, VEGF and PDGF, observed during the febrile stages in DHF patients in this study.
In contrast our findings, a number of previous studies have reported elevated levels of IFN-γ [19], IL-8 [27], IL-6, TNF-a [28], MIP-1b [19], IL-10 [29], and free VEGF [23], in DHF patients. However these studies differ significantly from the present study in the types of clinical cohorts evaluated. For example, comparable longitudinal cohort studies reporting higher levels of IL-10 and IL-6 have focused exclusively on pediatric cases [29], [30]. Importantly, primary infections made up less than ten percent of the cohorts in previous studies reflecting the higher incidence of DHF in secondary infections. Hence, different cytokine profiles observed in the present and previous studies are likely related to differences in immune responses to primary and secondary infections. The few studies that have included a subset of primary infections have reported conflicting results, with some reporting higher levels of cytokines in secondary compared to primary infections [31], [32], while others reporting no differences [33]. It is noteworthy that no DHF patients were included in these studies, and therefore it is not possible to compare cytokine profiles specific to DHF. We hypothesize that timely interferon-regulated antiviral responses are critical determinants of outcome in primary infections, whereas inflammatory mediators and regulators of antibody-dependent enhancement, including IL-6, IL-8, and IL-10 may dominate in secondary infections. Ethnic background of patients can also affect the type of cytokine responses to dengue infections [34], and may contribute to cytokine profiles described here.
In an attempt to identify serum protein markers of DHF, several groups have reported proteomic analysis of dengue patient sera [35], [36], [37]; using a variety of methods and clinical cohorts. We used a highly sensitive isobaric-tag method of quantitation that allowed us the compare the proteomic changes across different stages of infection. We focused on the most prominent functional protein group identified (i.e. acute phase reactants), and observed elevated levels of CRP, SAA, HPT, A2M and FT in individual patient samples. Maximum elevation of CRP and SAA in the early febrile phase was consistent with elevated production of IL-6, a hepatic inflammatory cytokine. Interestingly, acute phase reactants PCT, FB and SAP were not altered, and this may be related to liver dysfunction observed in dengue patients [38]. With the exception of SAA, there were no significant differences between DF and DHF groups suggesting that acute phase response is not a dominant mechanism of pathology in primary infections. Understanding the specific functional role of other proteins shortlisted in our proteomics study will require detailed validation.
Nitric oxide (NO) production by phagocytes is an important inflammatory response to pathogens and although increased levels of both inducible NO synthase (iNOS) and NO levels have been reported in dengue patients [28], [39], their role in dengue viral clearance is unknown. Protein adducts CT and NT formed from NO-mediated reactions are sensitive surrogate measures of neutrophil and macrophage activities during inflammation [18], [40]. We observed elevated levels of CT in dengue patient sera compared to healthy controls, which continued to rise from early febrile to defervescence stage indicating robust and sustained neutrophil activity. Interaction of activated neutrophils with the endothelium has been known to modulate vascular permeability [41], [42]. Whether elevated levels of CT can serve as an early indicator of plasma leakage remains to be tested. In contrast to CT, the transient nature of NT accumulation suggests that macrophage activity is limited to the acute phase of the infection, possibly linked to viral titers.
The comprehensive database of 47 blood parameters from dengue patients described in this study provides a unique opportunity to statistically query this dataset to identify -1) most significant molecules and 2) their relative importance in distinguishing DHF from DF during the early febrile stage. In the final analysis, a subset of 9 features was identified that included 5 cytokines, chlorotyrosine, blood lymphocyte count, and two serum proteins. Overall, cytokines involved in attenuated antiviral response; up regulation of acute phase proteins, and elevated neutrophil activity; together appear to be early signatures of DHF resulting from primary infections. The precise role of other cytokines IL-17, FGF-basic, and RANTES that were included in the predictive subset, in DHF pathogenesis is currently unclear and does not rule out the involvement of other cytokines in regulation of immune mechanisms in DHF patients.
Previously, a variety of statistical methods including classification and regression tree (CART) analyses [43], [44], as well as decision tree algorithms [45] have been used to identify clinical markers that achieved high sensitivity but poor specificity in classification of DHF. These clinical parameters, however, require daily monitoring. Identifying and measuring the molecules that are directly involved in pathogenesis could improve our predictive capabilities. Recently, Brasier et al used a logistic regression approach to report a 3 component biomarker panel consisting of platelet count, lymphocyte count and IL-10 that, classified DF from DHF patients with an accuracy of >85% during the first week following onset of fever [21]. In a second study Brasier et al used a multivariate adaptive regression splines (MARS) method to evaluate cytokines and plasma proteome from a cohort of secondary dengue infections and reported a panel consisting of IL-10 and seven serum proteins that achieved 100% sensitivity and specificity in prediction of DHF in the first week of fever onset [37]. However, these two studies applied a broad window of measurement, which may not capture the dynamic processes of DHF pathogenesis. It also raised the possibility that biomarkers of DHF in secondary infections may be qualitatively and quantitatively different from primary infections. Determining which of these biomarkers reflect differences in primary versus secondary infections and which inform on DHF development, whether in primary or secondary infections, will be critical for the development of robust biomarkers to stratify dengue patients for medical care.
In conclusion, this study describes a comprehensive and systematic molecular analysis of serum samples from a cohort of patients with primary dengue infection. The analytical approach and statistical workflow we have outlined forms a robust platform for both future discovery and validation of biomarkers for prediction of severe dengue disease.
|
10.1371/journal.pcbi.1005628 | High-confidence assessment of functional impact of human mitochondrial non-synonymous genome variations by APOGEE | 24,189 are all the possible non-synonymous amino acid changes potentially affecting the human mitochondrial DNA. Only a tiny subset was functionally evaluated with certainty so far, while the pathogenicity of the vast majority was only assessed in-silico by software predictors. Since these tools proved to be rather incongruent, we have designed and implemented APOGEE, a machine-learning algorithm that outperforms all existing prediction methods in estimating the harmfulness of mitochondrial non-synonymous genome variations. We provide a detailed description of the underlying algorithm, of the selected and manually curated training and test sets of variants, as well as of its classification ability.
| The mitochondrion is an organelle floating in the cytoplasm of almost all eukaryotic cells. Its primary function is to generate energy. It contains an independent DNA (mtDNA), which is inherited maternally in many organisms. This DNA is highly susceptible to mutations since it does not possess the robust DNA repair mechanisms proper of the nuclear DNA. Mutations in the mtDNA were associated to several inherited and acquired mitochondrial diseases, including Alzheimer and Parkinson diseases, and cancer. The assessment of the mutation-disease causal link is an onerous task. It requires important laboratory skills/equipment and, often, an animal facility, which are not always available to any laboratory altogether. More and more often, one falls back on software solutions that rely on structural and functional characteristics of proteins to predict the putative harmfulness of a mutation. Many have been implemented and tested on the nuclear proteins, but only a few were finely tuned to the “neglected genome”. Our work not only presents APOGEE, a machine-learning-based predictor that outperforms all existing predictors in reliability and sensitivity, but it makes freely available the APOGEE’s predictions for all the mitochondrial missense mutations in MitImpact.
| Assessing the pathogenicity of genome mutations is a notoriously onerous task both in-vitro and in-vivo, and occasionally even unviable because of the paucity of funds or of proper analytical facilities. This is particularly true when dealing with the mitochondrial DNA, which is less studied, although significantly smaller, than the nuclear counterpart [1]. This task was massively faced from a computational point of view though, and a growing number of algorithms and software packages, which elaborate sequence, structural and functional data to yield plausible evaluations of the harmfulness of variant amino acids in the form of pathogenicity scores and categorical, often dichotomous, variables, were implemented and released over time.
Their assessments of pathogenicity are actual predictions, whose global congruency was deeply investigated by a few comparative studies [2–7]. Generally, only 60–70% agreement resulted when considering all the possible human non-synonymous variants. No single predictor emerged, neither in terms of classification accuracy, nor of specificity and sensitivity [6]. Similar results were achieved when considering only a subset of 173 validated disease-causing mitochondrial mutations taken from MITOMAP [8]: 64% were deemed possibly or probably damaging by PolyPhen-2, 62% as being high or medium impact variants by MutationAssessor and 61% as being deleterious by PROVEAN. The worst performance was achieved by SIFT, with only 16% of true positives, and by FatHmm that correctly classified only one variant on 173. Even with this subset of validated mutations or with those falling in ultraconserved genomic loci, predictions were broadly incongruent. Reasons for that were ascribed to the intrinsic differences between computational/statistical methods and reference databases, or between training datasets and alignment algorithms [2–7]. These facts drove the development of the so called aggregators or meta-predictors, namely those software packages that yield an evaluation of pathogenicity based on the outcomes of other reference predictors, as well as of databases of nuclear and mitochondrial precomputed predictions [9–12]. Even these were contrasting [2].
Due to their high incongruence and since almost all existing predictors were tailored to the nuclear genome, which is an important contributing factor to their modest classification performance and incongruence, we designed APOGEE. It grounds on three milestones: it feeds third-party predictors with features that are strictly related to the 13 mitochondrial proteins, like multi-alignments and amino acids conservation estimates; its reasoning strategy was tuned on finely curated, non-overlapping, training sets of variations; its predicting model was based on decision tree learning in order to provide investigable rules of pathogenicity.
The classification engine of APOGEE was built on 100 sets of variants drawn randomly and with replacement from a training set of 864 known mitochondrial variants. This strategy left out-of-bag as many test sets of variants on which we calculated an array of performance metrics. These were additionally calculated for all aggregated individual predictors and were reported in Table 1. It is important to notice that our training set overlaps those used by most of the aggregated software predictors, which are available from http://structure.bmc.lu.se/VariBench/GrimmDatasets.php, of only 102 on 864 variants.
Performance of the considered predictors were generally low, with elevate misclassification rates (MCRs) and low Matthew’s Correlation Coefficient (MCCs) for all the investigated methods, but EFIN that achieved good specificity, accuracy and relatively low MCR. FatHmm_W outperformed FatHmm, both in terms of sensitivity and precision. CADD and FatHmm MCRs were also sensibly low. Among the meta-predictors, CAROL and COVEC WMV, which assembled only two (SIFT and PolyPhen2) and three (SIFT, PolyPhen2 and MutationAssessor) primary scores, respectively, showed decent performances. Pairwise comparisons of predictions revealed good agreement between all individual tools, but PANTHER that was mostly discordant (S1 Text). On the contrary, the outcomes of the meta-predictors were generally poorly congruent. In particular, Condel was mostly in disagreement with all the others (S1 Text).
APOGEE outperformed all by achieving the best sensitivity, accuracy, precision, FDR, MCC and MCR rates and the second highest specificity value (after FatHmm) (Table 1 and Fig 1).
The risk of overfitting was checked against two additional test sets, not overlapping with the training set (S2 Text). One was made of 153 variants appearing in the latest releases of dbSNP and MITOMAP, at the time of this writing. The classification rates of APOGEE resulted at least as high as those reported in Table 1 (cf. Table 2). The latter independent test set was made of 48 unbiased variants, on which APOGEE obtained the performance records reported in Table 3, which are in line with those previously shown.
Software predictors of the harmfulness of genomic variations were shown to be incongruent [2]. The major cause of incongruence was ascribed to two types of circularity issues affecting both training and test datasets used by data mining-based predictors [13]. Type 1 refers to the accidental, even if frequent, partial overlap between the training and test datasets. Type 2 consists in deeming all variants of some genes as pathogenic or neutral, for the mere fact of falling within a functionally critical gene. The consequence of that was a strong bias towards pathogenic or neutral predictions for them, and thus an increasingly low prediction sensitivity. Unfortunately, being the mitochondrial genome very small and each gene relatively little affected by mutations, both type 1 and 2 problems are unavoidable, even if reducible. Type 2 circularity problem has limited impact on the mtDNA, since for all 13 protein-coding genes, both true neutral and true harmful missense substitutions are reported, with a proportion of deleterious variants ranging from 15% to 35%. Considering the low number of genes and the disproportion between harmful and neutral mutations, the type 2 problem has a globally reduced effect on the predictions, and will tend to disappear with new findings. In principle, type I problem might be significantly cut down by finely curating the training sets.
The strategy implemented in APOGEE, which made it the best performer, consisted in adopting a transparent machine-learning algorithm that yielded a number of decision rules taken on larger and finely curated training sets. The LMT classifier was not claimed here to perform better than any other machine learning algorithms by far, but to perfectly fit the need for a dichotomous classifier that delivers a probability for a variant to be pathogenic, together with the rule according to which the decision is taken. The extra and decisive step consisted in tackling the longstanding problem of artifacts and misclassified variants of training sets by (i) discarding variants if originated from alignment errors; (ii) flipping the outcomes of the predictions (pathogenic and neutral) when new phenotypes or clinical evidences become available; (iii) removing false variants corresponding to alleles excluded from multi-allelic sites after periodic dbSNP update. Some variants of public datasets were indeed poorly annotated or simply artifacts. We bumped against a number of these along the previous three versions of the training sets of APOGEE. Several variants from a preceding version of the training set were updated in the subsequent, either because new experimental evidences reverted their estimated pathological effects, or because they were finally associated with any disease or in case of multiallelic sites. In particular, a few multiallelic sites were reassessed since only one allele was actually validated, with the others being deemed artifacts. Other variants were completely removed since they were found not to map to any assembled mitochondrial sequence present in dbSNP. 432 core variants were shared among all three datasets, 230 of which were considered functionally neutral and 202 pathogenic. 215 in 230 were observed to be actually neutral in all three training sets. 181 in 202 were unanimously considered deleterious.
This preprocessing step contributed to obtain a finely curated and larger training set. Most classifiers were indeed trained on a handful of known variations, as for example, MToolBox DS, which was built on the 53 damaging missense variants available from the Humsavar dataset (Table 3 in [14]). On the contrary, APOGEE was trained on a total of 223 deleterious variants.
The proportion of neutral and pathogenic amino acid changing variants that occur in a gene sequence mainly depends on the mutational pressure, genetic drift and both purifying and adaptive selection. These evolutionary mechanisms are generally considered gene-specific, thus making difficult the identification of potential deleterious mutations without any knowledge of the gene-specific level of tolerance to mutations. Therefore, taking into account the evolutionary measure of a gene or of a gene family, meant as the ratio between the non-synonymous and synonymous substitution rates, as calculated in a set of aligned orthologous sequences [15], could dramatically increase the sensitivity of the predictors. A beneficial effect might also be conferred by the RVIS index [16], which determines which nuclear genes are more intolerant to missense mutations. These two indices might provide useful insights in the understanding of the different evolutionary dynamics of genes and will be integrated in future releases of APOGEE.
The functional role of each mutant residue is greatly influenced by the co-inherited missense variants within the very same protein or within structurally/functionally associated proteins. This “coevolutionary issue” has been poorly investigated so far, although it is well known that human pathogenic mutations can also be present within other species, without no deleterious effects, because they are probably compensated by co-inherited intra- or inter- gene mutations. Currently, a novel computational strategy has been developed in order to identify human pathogenic mutations that are compensated in extra-specific genomes (Compensated Pathogenic Deviations), i.e., their damaging effects are counterbalanced by other fixed mutations that are absent in humans [17]. A relevant proportion (3–10%) of human damaging missense mutations has been identified in mammal and vertebrate protein alignments, indicating that compensatory mechanisms exist (sequences are assumed to derive from healthy animal organisms) at different evolutionary ages [17].
The identification of coevolving residue pairs is impeded, at any rate, by the paucity of appropriate experimental data. Knowledge of the ternary and quaternary structures of mitochondrial and nuclear OXPHOS proteins could contribute to resolve the inconsistencies among computational pathogenicity predictions and diseases association [18]. This aspect will also be taken into consideration in the next releases of APOGEE.
Variants with known functional effects on mitochondrial proteins were harvested from MITOMAP [8] (accessed July 2015) and dbSNP 144 [19]. In total, we collected 864 non-synonymous mutations, 228 of which were tagged as “confirmed” or “reported” in MITOMAP. 223 were already linked to known mitochondrial genetic disorders (i.e., Leber optic neuropathy (LHON), mitochondrial encephalomyopathy, lactic acidosis, stroke-like episodes, maternally inherited deafness or aminoglycoside-induced deafness), or complex diseases such as Alzheimer or cancer. 30 on 228 were confirmed to be pathogenic amino acid changing variants and most of them resulted to cause LHON. On the other hand, 5 out of 228 (8741:T>G, 8795:A>G, 9055:G>A, 8414:C>T, 3745:G>A) were reported as non-pathogenic, thus exhibiting a likely protective or compensatory effect on the carrier subjects. The remaining 699 variants were retrieved from Ncbi dbSNP 144 through the Ncbi Variation Reporter tool (http://www.ncbi.nlm.nih.gov/variation/tools/reporter). Variants with no reported pathological consequences in dbSNP and no overlap with MITOMAP were considered harmless. In detail, 63 of the 699 dbSNP variants were classified as pathogenic, being these present in MITOMAP. The remaining variants were considered neutral (cf. Table 4). Hence, the entire variant set consisted of 223 pathogenic (MITOMAP), 5 non-pathogenic (MITOMAP) and 636 (non-overlapping dbSNP) neutral variations (cf. S1 Table).
Our APOGEE classifier was trained on these datasets, as specified below, and tested also on two non-overlapping datasets. In particular, we have put together a set of 153 new functional variants, which came from the latest releases of MITOMAP (accessed in January 2017) and dbSNP (ver. 147), and additional 48 variants obtained from VariBench [13] (web-site: http://structure.bmc.lu.se/VariBench/GrimmDatasets.php). We made sure that these variants were not included in our original training sets.
Assessments of pathogenicity were computed by and collected from a number of predictors (Table 5), provided that these could process batch queries and accepted mitochondrial protein-coding gene symbols in input [20]. We used EFIN with standard parameters, after training it on SwissProt (SP) [21] and HumDiv (HD) [22] datasets. Similarly, we queried CADD 1.3 [23] and obtained two scores, the original and the phred-scaled scores. Being numeric, we dichotomized the phred scores and classified the variants that exceeded the threshold of 12 as harmful, as suggested by the authors. Variants were submitted to CADD in VCF-like data format. We further retrieved predictions from CRAVAT [24], both for mendelian (VEST) [25] and cancer (CHASM) [26] diseases. Input variants were specified as Ensembl Transcript IDs and amino acid substitutions, using the one-letter encoding. It responded to our query with pairs of p-values and FDRs, one for each input variant. If a prediction was significant, i.e., p-value <0.05 and FDR <0.2, we labeled the corresponding variant as pathogenic (in case of VEST) or driver (in case of CHASM). We applied the weighted version of the FatHmm prediction algorithm [27] to a list of Uniprot accession numbers and amino acid substitutions and obtained functional scores and categorical predictions for them. Likewise, we queried the Meta-SNP server [28], but submitting the fasta sequences of the OXPHOS proteins and the corresponding lists of amino acid mutations. It returned categorical predictions and scores for PhD-SNP, SIFT, SNAP and PANTHER [29–32].
MitImpact accounted also for the MtoolBox Disease Scores. We set the pathogenicity threshold to 0.4311, as described in [14] (details in S1 File), and considered harmful all variants exceeding it. We additionally included the COVEC 0.4 scores [33]. We run the COVEC Weighted Majority Rule algorithm and obtained a numerical score for each variant, based on a consensus of the predictions of SIFT, PolyPhen2 and MutationAssessor. A pathological status was associated to a variant if its COVEC score was positive. Similarly, we computed the Transformed Functional Impact for Cancer (TransFIC) score [34], by providing TransFIC with the SIFT, PolyPhen2 [22] and MutationAssessor [35] scores, which were already stored in the former release of MitImpact [36]. TransFIC normalized these scores on a baseline tolerance of genes, which corresponded to the level of tolerance of germline variants occurring in genes with dissimilar functions. Functional similarity was assessed on the Gene Ontology Biological Process annotation (gosbp). The tool yielded a tripartite categorical classification for each variant given in input, along with the transformed scores.
MitImpact took into consideration also cancer-related information, taken from COSMIC 68 [37]. COSMIC IDs and information on the tumor type, number of examined tumor samples and mutation frequency for the matching variants were included. Moreover, the conservation indices PhyloP100V and PhastCons100V [38] were calculated for all the mitochondrial genomic positions that cause missense substitutions by using the UCSC Gene Tables gateway. We additionally included information on protein coevolution through the MISTIC [39] webserver, a tool that predicts coevolving sites within mitochondrial protein sequence alignments. We retrieved protein alignments from the Ncbi Organelle Genome resource, restricting the study to Mammals (about 670 species-specific sequences for each gene) and using the human protein sequences as reference. We then computed the matrix of Mutual Information (MI) scores (MI Z-scores), which contains the scores of all the possible amino acid pairs, and then selected only the pairs with scores > 6.5, since these are suggested by the authors to be coevolving pairs of amino acids. Then, we calculated the frequency of the coevolving amino acids and the mean MI Z-score for each amino acid site.
These scores were computed for all 24,189 non-synonymous amino acid changes potentially affecting the human mitochondrial DNA and made freely available, as a flat-file with variants as rows and scores as columns, from MitImpact. Variants were grouped in training and test sets, as for the previous section, and used to build and verify the APOGEE classifier.
The predictions of the abovementioned tools were used to feed APOGEE (pAthogenicity Prediction thrOugh loGistic modEl trEe). Its operating logic bases on the classification model of the Logistic Model Tree (LMT). The choice of yet another meta-predictor was driven by our intent to offer a transparent classifier, finely tuned on mitochondrial variants and that gives reproducible and easy-to-understand results. LMT combines the logistic regression models with tree induction resulting in a single tree. It uniquely provides the user with decision rules that allow, easily, classifying unknown variants as neutral or harmful. Moreover, LMT has the advantage of providing explicit class probability estimates and, thus, of helping the user to intuitively grasp the actual uncertainty behind any evaluation of pathogenicity.
It builds a standard decision tree structure with logistic regression functions at the leaves. Each leaf may not contain the same function, since variables are independently selected to maximize the discrimination between neutral and pathogenic mutations. The tree-induction procedure proceeds in a top-down fashion. It recursively splits the instances (variations) space and stops when the inferred subdivisions are reasonably “pure”, in the sense that they contain observations with mostly identical class labels (pathogenic or neutral). In a standard decision tree framework, a region is labeled with the majority class of the observations in that region.
Formally, we inferred an unknown function f, which can map the predictor variables Xs to the class label Y:
Y=f(X1,…,Xp),
where Xs were the pathogenicity scores, while the response variable Y was the target class. The function f(⋅) was directly inferred from real data, which consisted of a set of n variations, carrying their pathogenicity scores p along with their classes (or labels) y of belonging.
By denoting the n × p data matrix (without labels y) with bold X and the p-dimensional vector of annotation scores for a single mutation with x, we modeled the posterior class probabilities P(Y|X) using a sigmoid function. For a two-class classification problem, for which we specify the labels of Y as y = ±1 (with 1 for neutral mutations and -1 for pathological mutations):
P(Y=y|x,w)=11+e−ywTx
or, equivalently:
P(Y=1|x,w)=ewTx1+ewTxorP(Y=−1|x,w)=11+ewTx.
Here, w is the unknown vector of p weights associated with each predictor. In order to compute the model for each class, we estimated these weights. This was achieved through the minimization of the following logistic cost function:
w^=argminw∑i=1nlog(1+e−yiwTxi).
Once the weights were computed, the final regression model for each class was determined through a LogitBoost algorithm, which selected the final predictors (xi) to be included in the model. Therefore, we obtained that:
P(Y=yj|x,w^)=eFj(x)1+∑k=1JeFk(x),∑k=1JFk(x)=0,J=2
where Fk(x) was the estimated logistic function of the kth class. The class labels of the mutations were assigned by the following formula:
y*=argmaxyP(Y=y|x,w^).
As mentioned earlier, the tree structure gives a disjoint subdivision of the whole instance space S, spanned by all pathogenicity scores (or predictors) that are present in the data, into regions St. Every region was represented by a leaf in the tree:
S=t∈TSt,St∩St′=∅fort≠t′
A logistic regression function ft was associated to each leaf t ∈ T, which included a subset Vt ⊆ V of all pathogenicity scores present in the data and that modeled the class membership probabilities as P(Y = y|x,w). The weight estimates were zero when the predictor did not contribute to the model. Generalizing the whole LMT model:
f(x)=∑t∈Tft(x)∙I(x∈St)
where I(x ∈ St) is a variable indicator that equals 1 if the observation x belongs to the region St or zero, otherwise.
Under or over-estimation of the prediction capability of APOGEE would be possible if considering only one run of the algorithm, in a similar setting with unbalanced class sizes (i.e. 223 pathogenic vs 641 benign mutations). This dimensional bias was tackled by the implementation of a bootstrap strategy that, by definition, is based on randomly drawing a sample with replacement from the observed sample of size n = 223 for pathogenic variants and n = 641 for tolerated variants. The random sampling was repeated 100 times, resulting in 100-bootstrap samples. For any given draw, approximately one-third of observations were not selected and served as test set (out-of-bag (OOB) test set). Subsequently, the LMT was applied to each of the 100-bootstrap samples and a prediction error assessed using the corresponding 100 test sets, namely those observations not included in the training set due to sampling with replacement. This measure of prediction error is referred to as leave-one-out bootstrap estimate. [40]. Thus, the fact of sampling the 70% of all pathogenic variants and the same number of the neutral variants implied that the expected frequencies of inclusion of both types of variants were 50% and 22%, respectively. In brief, for 100 iterations, we run this algorithm:
Each iteration gave an estimate of the pathogenicity of the variants in the OOB set. A variant was deemed harmful if the mean of the probabilities of being harmful, calculated for all iterations in which it was included in the OOB, resulted > 0.5. Compared to an individual run, bootstrap replaces the classification rules of an LMT model with the probability of being harmful. The classifier was implemented in R, by using the R package Rweka [41] [42].
APOGEE is freely available in MitImpact [36] at http://mitimpact.css-mendel.it/.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.