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.1006888 | Distinct susceptibility of HIV vaccine vector-induced CD4 T cells to HIV infection | The concerns raised from adenovirus 5 (Ad5)-based HIV vaccine clinical trials, where excess HIV infections were observed in some vaccine recipients, have highlighted the importance of understanding host responses to vaccine vectors and the HIV susceptibility of vector-specific CD4 T cells in HIV vaccination. Our recent study reported that human Ad5-specific CD4 T cells induced by Ad5 vaccination (RV156A trial) are susceptible to HIV. Here we further investigated the HIV susceptibility of vector-specific CD4 T cells induced by ALVAC, a canarypox viral vector tested in the Thai trial RV144, as compared to Ad5 vector-specific CD4 T cells in the HVTN204 trial. We showed that while Ad5 vector-specific CD4 T cells were readily susceptible to HIV, ALVAC-specific CD4 T cells in RV144 PBMC were substantially less susceptible to both R5 and X4 HIV in vitro. The lower HIV susceptibility of ALVAC-specific CD4 T cells was associated with the reduced surface expression of HIV entry co-receptors CCR5 and CXCR4 on these cells. Phenotypic analyses identified that ALVAC-specific CD4 T cells displayed a strong Th1 phenotype, producing higher levels of IFN-γ and CCL4 (MIP-1β) but little IL-17. Of interest, ALVAC and Ad5 vectors induced distinct profiles of vector-specific CD8 vs. CD4 T-cell proliferative responses in PBMC, with ALVAC preferentially inducing CD8 T-cell proliferation, while Ad5 vector induced CD4 T-cell proliferation. Depletion of ALVAC-, but not Ad5-, induced CD8 T cells in PBMC led to a modest increase in HIV infection of vector-specific CD4 T cells, suggesting a role of ALVAC-specific CD8 T cells in protecting ALVAC-specific CD4 T cells from HIV. Taken together, our data provide strong evidence for distinct HIV susceptibility of CD4 T cells induced by different vaccine vectors and highlight the importance of better evaluating anti-vector responses in HIV vaccination.
| Development of a safe and efficacious HIV vaccine is a critical global health priority. Recombinant viral vectors are an important platform for HIV vaccine delivery. Recent clinical trials testing candidate HIV vaccines based on Ad5 vectors failed and reported excess HIV infections in some vaccine recipients, underscoring the necessity to investigate HIV susceptibility of viral vector-specific CD4 T cells in HIV vaccination. By using PBMC samples from clinical trials that examined two important HIV vaccine vectors (canarypox viral vector ALVAC and human Ad5 vector), we here report that compared to Ad5 vector, the ALVAC-specific CD4 T cells are more resistant to HIV infection, providing evidence for distinct HIV susceptibility of CD4 T-cell populations induced by different HIV vaccine vectors. Our findings present new insights into our understanding of HIV vaccine-induced immunity and help improve the design and immune assessment of viral vectors for the development of HIV vaccines.
| Over 30 years after the discovery of HIV as the causative agent of acquired immunodeficiency syndrome (AIDS), HIV/AIDS continues to be a significant challenge for global public health. More than 36 million people are currently living with HIV, with over 2 million new infections and 1 million AIDS-related deaths per year [1]. Development of a safe and effective HIV vaccine remains a high research priority. Recombinant viral vectors are an important platform for HIV vaccine development. To date, a number of HIV vaccine vectors derived from different viral families have been developed, including adenovirus [2] and poxvirus [3, 4]. Several clinical trials (Step and Phambili) testing candidate HIV vaccines based on human Ad5 vector (rAd5) have failed due to lack of efficacy and/or transiently increased HIV infections in some vaccinated individuals [5–7]. These unanticipated results from clinical trials have brought to light the importance of understanding host immune responses induced against viral vectors in HIV vaccination [8, 9].
CD4 T cells are central to host immunity by providing help signals to other components of the immune system [10]. The protective role of CD4 T cell responses has been documented for various pathogenic infections, including HIV [11–14]. However, CD4 T cells are also major target cells for HIV infection. During an antigen-specific immune response, activation and expansion of responding CD4 T cells is required [15], which is usually desired in most vaccine strategies but could become a potential problem in HIV vaccination due to the fact that HIV preferentially infects activated CD4 T cells [16–19]. Recent research from our group and others has shown that human CD4 T cells specific for different antigens differ in their susceptibility to HIV infection [20–26]. In particular, we have reported that human Ad5-specific CD4 T cells generated in response to both natural Ad5 infection and rAd5 vaccination are highly susceptible to HIV and are preferentially depleted in HIV-infected individuals [21]. Although potential mechanisms for Ad5 vector-associated excess HIV infections in the Step and Phambili studies are thought to be complex and could be affected by different factors such as the quantity, quality and in vivo localization of CD4 T cells induced during vaccination, our findings suggest that understanding the HIV susceptibility of vector-specific CD4 T-cell populations induced by different vaccine vectors may provide new insights into our understanding of host immunity in HIV vaccination.
In addition to rAd5, another important HIV vaccine vector that has been tested in late-stage clinical trials is ALVAC, a recombinant canarypox virus vector. The ALVAC prime/gp120 boost HIV vaccine regimen tested in the “Thai” RV144 trial demonstrated modest efficacy (~31%) [27]. Building upon the partial success of RV144, multiple ongoing trials further evaluating ALVAC-based HIV vaccine regimens are currently being conducted [28, 29]. In this study, we sought to understand anti-vector T cell responses with a focus on the phenotype and in vitro HIV susceptibility of vector-specific CD4 T cells induced by vaccination with ALVAC compared to Ad5. Cryopreserved peripheral blood mononuclear cells (PBMC) from RV144 vaccine recipients were analyzed in comparison with PBMC from HVTN204, a phase II trial evaluating rAd5-HIV vaccine (DNA prime/Ad5 boost) [30], using the in vitro HIV susceptibility assay reported in our previous studies [20, 21, 23]. We also measured vector-induced CD8 T-cell response in these PBMC samples. Our data show that vector-specific CD4 T cells induced by different HIV vaccine vectors manifest marked difference in their susceptibility to HIV infection; compared to Ad5-specific CD4 T cells in HVTN204 PBMC, the ALVAC-specific CD4 T cells in RV144 PBMC are substantially less susceptible to both R5 and X4 HIV infection in vitro. The differential HIV susceptibility between these two groups of vector-specific CD4 T cells is closely associated with their differences in phenotype, cytokine expression, and interestingly, the profiles of vector-specific CD8 vs. CD4 T-cell proliferative response induced by these two vectors.
To compare the HIV susceptibility of different HIV vaccine vector-induced CD4 T cells (ALVAC vs. Ad5) in human vaccine recipients, we employed the in vitro HIV infection assay reported in our previous studies [20, 21, 23] (Summarized in S1 Fig). In brief, PBMC samples of vaccine recipients in RV144 (ALVAC) and HVTN204 (Ad5 vector) were first stained with CFSE, a fluorescent dye used to track T-cell proliferation, and then stimulated with the corresponding empty vector for three days to induce the expansion of vector-reactive CD4 T cells, followed by infection with either CCR5-tropic (R5; US-1 strain) or CXCR4-tropic (X4; 92/UG/029 strain) HIV. Three days post-infection (dpi), flow cytometry was used to measure T-cell proliferation (indicated by decreased CFSE fluorescence intensity; CFSE-low) and HIV infectivity in vector-specific CD4 T cells (intracellular HIV p24+ rate in CFSE-low CD4 T cells) (S1A Fig). We have previously verified this in vitro system by demonstrating that the CFSE-low, proliferating CD4 T cells are mostly antigen specific (S1B Fig) and closely resemble their in vivo phenotypes (S1C Fig).
Based on this system, we first observed that both ALVAC and Ad5 vector induced significant levels of CD4 T-cell proliferation in PBMC of vaccine recipients (ALVAC for RV144 and Ad5 for HVTN204) (Fig 1). Regarding HIV susceptibility, we found that compared to Ad5 vector-induced CD4 T cells in HVTN204 PBMC, which were highly susceptible to R5 HIV infection (mean p24+%: 26.9%), the ALVAC-induced CD4 T cells in RV144 PBMC were markedly less susceptible to R5 HIV (mean p24+%: 1.27%) (p<0.01) (day 3 post-infection) (Fig 1A). We also monitored HIV infection in vector-induced CD4 T cells for up to 9 days post exposure and found that ALVAC-induced CD4 T cells remained resistant to HIV on day 9 post viral exposure (p24+: 0.5%), whereas Ad5 vector-specific CD4 T cells were still readily susceptible (p24+: 11.8%) (S2 Fig). Consistent with the results of R5 HIV infection, a similar lower susceptibility to X4 HIV (92/UG/029 strain) was also observed for ALVAC-induced CD4 T cells (mean p24+%: 1.82%) as compared to Ad5 vector-induced CD4 T cells (mean p24+%: 16.2%) (p<0.01) (Fig 1B). As controls, we showed that the two vectors induced very little T-cell proliferation in pre-vaccine PBMC of the same individuals (S3A Fig), suggesting that the T-cell proliferation observed in post-vaccine PBMC in our system were specific to vector with minimal non-specific proliferation. In addition, very little intracellular p24 (<0.1%) was detected in the same proliferating CD4 T cells when HIV was not added, supporting that intracellular p24 staining in our system is specific (S3B Fig). As another control, RV144 and HVTN204 PBMC were polyclonally activated by anti-CD3/CD28. We showed that anti-CD3/CD28-activated CD4 T cells in RV144 and HVTN204 PBMC were susceptible to HIV infection at comparable level (S4 Fig). Furthermore, we noted that in Ad5-stimulated PBMC, the CFSE-hi CD4 T cells appeared to be more sensitive to HIV as well compared to those in ALVAC-stimulated PBMC (Fig 1A). This might be related to the lower secretion of β-chemokines in the Ad5-stimulated PBMC culture, which will be presented later.
Transmitted founder virus (TFV) is important in HIV transmission. In addition to R5 US-1 and X4 92/UG/029 strains used, we also tested the susceptibility of vector-induced CD4 T cells to AD17 HIV molecular clone, a TFV [31, 32]. Consistently, we observed that ALVAC-induced CD4 T cells were also less susceptible to AD17 TFV infection (p24+ %: 0.6%) as compared to Ad5 vector-induced CD4 T cells (p24+ %: 3.8%) (S5 Fig), although the overall infectivity of AD17 TFV in these CD4 T cells was lower than that of the US-1 and 92/UG/029 strains (S5 Fig).
In vector HIV vaccination, insert-specific CD4 T cells are also induced in addition to vector-specific CD4 T cells. Therefore, we measured HIV susceptibility of vaccine Env-specific CD4 T cells using the same assay and found that unlike vector-specific CD4 T cells, Env-specific CD4 T cells in both RV144 and HVTN204 PBMC were readily susceptible to R5 and X4 HIV infection with no significant difference detected (S6 Fig). Taken together, these data suggest that the vector-specific CD4 T cells induced by different HIV vaccine vectors manifest marked differences in their susceptibility to both R5 and X4 HIV infection in vitro, with ALVAC-specific CD4 T cells being less susceptible than Ad5 vector-specific CD4 T cells.
We and others have shown that differential HIV susceptibility of human antigen-specific CD4 T cells can occur at both HIV entry and post-entry levels [20, 33]. An important factor that influences HIV infection of target cells at the entry level is the surface expression of the HIV co-receptors CCR5 and CXCR4. To understand potential mechanisms underlying the differential HIV susceptibility of ALVAC and Ad5 vector-specific CD4 T cells described above, we examined CCR5 and CXCR4 expression on these two groups of vector-specific CD4 T cells. We found that ALVAC-specific CD4 T cells expressed significantly lower frequencies of CCR5+ CD4 T cells (CCR5+%: 8.4 ± 1.8) than Ad5 vector-specific CD4 T cells (CCR5+%: 31.9 ± 5.1) (p<0.005) (Fig 2A). A similar difference was also observed for CXCR4 expression on ALVAC- and Ad5 vector-specific CD4 T cells (CXCR4+ % for ALVAC vs. Ad5: 8.3 ± 1.6 vs. 38.6 ± 7.4) (p<0.001) (Fig 2B). These data suggest that limited expression of CCR5 and CXCR4 represents an important mechanism for the lower susceptibility of ALVAC-specific CD4 T cells to R5 and X4 HIV, respectively, compared to Ad5 vector-specific CD4 T cells.
To better understand the relative contribution of co-receptor expression to the overall HIV susceptibility of vector-induced CD4 T cells in our system, we further analyzed HIV infection in co-receptor+ and co-receptor- (CCR5+/- and CXCR4+/-) subsets of Ad5-specific CD4 T cells as compared to that in ALVAC-specific CD4 T cells. Not surprisingly, we found that majority of HIV infection was observed in CCR5+ or CXCR4+ subsets of Ad5-specific CD4 T cells (Fig 2C). We also noted that the HIV infection rate in the CCR5- subset (p24+: 23%) or CXCR4- subset (p24+: 5.3%) of Ad5-specific CD4 T cells (Fig 2C) remained higher than the overall HIV infection rate in ALVAC-specific CD4 T cells (Fig 1). This data suggests that other factors may also contribute to the differential HIV susceptibility between Ad5- and ALVAC-specific CD4 T cells besides co-receptor expression.
At the post-entry level of viral infection, HIV infectivity is associated with innate antiviral status and the activation state of target cells. Our recent study has demonstrated that ALVAC and Ad5 vector manifest distinct innate stimulatory properties with ALVAC being able to activate strong innate responses in antigen-presenting cells (APCs) [34]. This could potentially affect the antiviral status of CD4 T cells in vector-stimulated PBMC. We therefore compared the antiviral status of vector-specific CD4 T cells in our system. CFSE-low CD4 T cells were sorted from vector-stimulated PBMC and subjected to gene-expression analysis for antiviral genes and common HIV restriction factors, including A3G, MxB, SAMHD1, Tetherin and TRIM5. We found that expression of the genes was comparable between Ad5- and ALVAC-specific CD4 T cells (Fig 3A). Consistent with this result, blockade of type-I IFN signaling in ALVAC-stimulated PBMC [34] did not significantly alter the HIV infection in ALVAC-specific CD4 T cells (Fig 3B). These data suggest that the differential HIV susceptibility of vector-specific CD4 T cells may not be related to their innate antiviral status. Next, we assessed immune activation status of vector-specific CD4 T cells by examining the expression of T-cell activation markers (CD25 and CD69). While no significant difference CD69 expression was observed between ALVAC- and Ad5-specific CD4 T cells, Ad5-specific CD4 T cells appeared to express slightly higher level of CD25 than ALVAC-specific cells (Ad5 vs. ALVAC: 81% vs. 65%) (Fig 3C). This activation status of vector-specific CD4 T cells is generally consistent with their susceptibility to HIV infection.
Human antigen-specific CD4 T cell populations manifest different phenotypes in memory differentiation, T helper (Th) lineages, and cytokine profiles which are associated with their susceptibility to HIV infection [20–23, 25, 35, 36]. We next characterized major phenotypes of ALVAC- and Ad5 vector-specific CD4 T cells. Based on expression of CCR7 and CD45RO, human CD4 T cells can be categorized into central memory (CM: CCR7+CD45RO+) and effector memory subsets (EM: CCR7-CD45RO+). By focusing on the CFSE-low CD4 T cells, we found that both ALVAC- and Ad5 vector-specific CD4 T cells predominantly manifested an EM-like phenotype 2 weeks after the final vaccination, and no significant difference in memory phenotypes was observed between ALVAC- and Ad5 vector-specific CD4 T cells (Fig 4A). Mucosal homing is another important characteristic of CD4 T cells that influences HIV pathogenesis. Mucosal compartments represent a major site for HIV infection and CD4 depletion in HIV disease [37]. Integrin α4β7 is an important mucosal homing receptor, directing migration of CD4 T cells to gut mucosa [38]. We found that compared to Ad5 vector-specific CD4 T cells, which expressed high levels of α4β7 as reported in previous studies [16, 21], ALVAC-specific CD4 T cells expressed significantly lower levels of α4β7 (Fig 4B).
Next, we examined T-helper lineage and cytokine profile of ALVAC- and Ad5 vector -specific CD4 T cells. As described, CFSE-stained PBMC from vaccine recipients were stimulated with ALVAC or Ad5 vector to induce vector-specific CD4 T cell expansion. Since cytokine expression in activated T cells is usually transient and the CFSE-low, vector-specific CD4 T cells in our system undergo days of proliferation, in order to measure cytokine production in CFSE-low CD4 T cells the culture was re-stimulated with the global PMA/ionomycin stimulus on day 6 for cytokine de novo re-synthesis in T cells as we reported previously [21, 23]. Since Th17 CD4 T-cell subset has been shown to be highly susceptible to HIV as compared to Th1 subset [21, 23], we first measured expression of IFN-γ, IL-17 and IL-2 in vector-specific CD4 T cells (Fig 4C), and found that a significantly higher fraction of ALVAC-specific CD4 T cells expressed IFN-γ than Ad5 vector-specific CD4 T cells (64.6% ± 6.98 vs. 43.0% ± 5.96; p<0.05), typical of a strong Th1-like response. In contrast, a higher fraction of Ad5-specific CD4 T cells expressed IL-2 (39.0% ± 4.68 vs. 17.6% ± 4.90; p<0.01) and IL-17 (8.71% ± 1.55 vs. 3.50% ± 0.77; p<0.01), suggesting a mixed Th1/Th17 response (Fig 4C and 4D). This result is in agreement with our previous report that examined Ad5 vector-specific CD4 T cells in PBMC from the RV156A trial [21]. Therefore, since IL-17- and IL-2-producing CD4 T cells are known to be more susceptible to HIV infection than IFN-γ-producing CD4 T cells, this differential Th1 vs. Th1/Th17 phenotype for ALVAC- and Ad5-specific CD4 T cells is consistent with their susceptibility to HIV infection.
Besides Th1 and Th17 markers, we also examined other major T-cell associated phenotypes for vector-specific CD4 T cells, including T-follicular helper (Tfh), regulatory T cells (Treg) and PD-1 (T-cell exhaustion marker). First, we observed that a significant fraction of both ALVAC- and Ad5-specific CD4 T cells expressed IL-21, a lineage-specific cytokine for Tfh cells. However, unlike IFN-γ and IL-17, no significant difference in IL-21 expression was found between ALVAC- and Ad5-specific cells (S7A Fig). Further analysis identified that HIV infection in IL-21+ Tfh-like subset (p24+: 7.6%) was not higher than IL-21- subset (p24+: 10.3%) (S7B Fig), suggesting that in our system HIV does not preferentially infect Tfh-like CD4 subset [39]. Furthermore, we measured expression of Treg markers (CD25 and FoxP3) and exhaustion marker PD-1 in vector-specific CD4 T cells and found that, similar to Tfh, no significant difference in expression of Treg markers (S7C Fig) and PD-1 (S7D Fig) was observed between ALVAC- and Ad5-specific CD4 T cells. Altogether, these data suggest that Tfh, Treg and PD-1 phenotypes may not account for the differential HIV susceptibility of ALVAC- and Ad5-specific CD4 T cells in our system.
β-chemokines (MIP-1α, MIP-1β, and RANTES) are CCR5 ligands and can block CCR5-tropic (R5) HIV infection at entry level by competitively binding to CCR5 [33, 40, 41]. Therefore, we examined MIP-1β (CCL4) expression in the CFSE-low, vector-specific CD4 T cells. Not surprisingly, we found that a significantly higher fraction of ALVAC-specific CD4 T cells expressed MIP-1β than Ad5 vector-specific CD4 T cells (57.10% ± 5.67 vs. 36.84% ± 4.16; p<0.01) (Fig 5A). To evaluate the potential impact of β-chemokine production on HIV susceptibility of vector-specific CD4 T cells in our system, in vitro HIV infection (CCR5-tropic; US-1) was conducted in the presence of neutralizing antibodies against these β-chemokines (CCL3/MIP-1α, CCL4/MIP-1β, and CCL5/RANTES). We found that blocking β-chemokines could modestly, but significantly, increase the susceptibility of ALVAC-specific CD4 T cells to R5 HIV (p <0.01) (Fig 5B), suggesting a role for β-chemokines in protecting ALVAC-specific CD4 T cells from R5 HIV. However, we also found that even in the presence of β-chemokine neutralization, ALVAC-specific CD4 T cells were still significantly less susceptible to R5 HIV than Ad5 vector-specific CD4 T cells (Fig 5B; Fig 1A), suggesting that the higher production of β-chemokines contributes only partly to the lower susceptibility of ALVAC-specific CD4 T cells to HIV as compared to Ad5 vector-specific CD4 T cells in our system.
By simultaneous analyses of both CD8 and CD4 T cells, we found that ALVAC and Ad5 vector elicited distinct profiles of vector-specific CD8 vs. CD4 T-cell proliferative response in PBMC. ALVAC stimulated robust vector-specific CD8, but relatively weak vector-specific CD4, T-cell proliferation in RV144 PBMC, whereas Ad5 vector predominantly induced vector-specific CD4, but not CD8, T-cell proliferation in HVTN204 PBMC (Fig 6A). When we analyzed the cumulative results from multiple vaccine recipients (n = 14), although no significant difference in the magnitudes of vector-specific CD4 T-cell proliferation was observed between ALVAC and Ad5 (13.43 ± 3.118 vs 19.62 ± 4.633, respectively; p = 0.2776), ALVAC induced significantly higher levels of vector-specific CD8 T-cell proliferation in RV144 PBMC (31.94 ± 5.085 vs 8.908 ± 2.172; p = 0.0004) than Ad5 vector did in HVTN204 PBMC (Fig 6B). We further analyzed the ratio of vector-induced CD8 vs. CD4 T-cell proliferation within the same individuals and compared between ALVAC and Ad5 (Fig 6C), and found that ALVAC induced a much higher ratio of CD8/CD4 T-cell proliferation than Ad5 vector did (3.137 ± 0.5696 vs 0.5615 ± 0.1364; p = 0.0003) (Fig 6C). In contrast, the vaccine insert antigen envelope (Env) induced strong CD4 and weak CD8 T-cell proliferation in RV144 PBMC, but comparable levels of CD4 and CD8 T-cell proliferation in HVTN204 PBMC (S8 Fig), consistent with the results of Env-specific CD4/CD8 T-cell response measured by ex vivo ICS in previous studies [27, 30]. Taken together, these data suggest that ALVAC induces a distinct profile of vector-specific CD8 to CD4 T-cell proliferative response from that induced by Ad5 vector in vitro.
The importance of CD8 T cells in anti-HIV immunity, including control of viral replication and limiting HIV-infected CD4 T cells, has been well established [42–44]. In our system, we have observed low levels of CD4 T-cell proliferation in RV144 PBMC after ALVAC stimulation as compared to that in HVTN204 PBMC after Ad5 vector stimulation, which could possibly reflect the inhibition of CD4 T cell proliferation by ALVAC-induced CD8 T cells. Therefore, we next explored the potential impact of vector-induced CD8 T cells on vector-specific CD4 T cell proliferation. CD8 T cells were depleted from PBMC using magnetic cell sorting (MACS) prior to CFSE staining and vector re-stimulation. Efficient depletion of CD8 T cells from PBMC was confirmed (Fig 7A). Subsequently, proliferation of CD4 and CD8 T cells in the whole or CD8-depleted PBMC was measured on day 6 by flow cytometry. We showed that depletion of CD8 T cells from ALVAC-stimulated PBMC led to a significant increase in the proliferation of ALVAC-specific CD4 T cells in RV144 PBMC (p = 0.0068), whereas no such effect was seen in CD8-depleted HVTN204 PBMC when stimulated by Ad5 vector (p = 0.1747) (Fig 7B). These results suggest that, unlike Ad5 vector-induced CD8 T cells, ALVAC-induced CD8 T cells can inhibit the expansion of autologous vector-specific CD4 T cells in PBMC.
To explore potential mechanisms by which ALVAC-stimulated CD8 T cells inhibit autologous ALVAC-specific CD4 T-cell proliferation, we conducted trans-well experiments where CD8 T cells were first depleted from PBMC and then added back to the culture in trans-well. We found that addition of CD8 T cells in trans-well could largely, though not completely, restore the inhibitory effect of CD8 T cells on ALVAC-specific CD4 T-cell proliferation (from 26.3% to 16.9%, compared to 13.3% for whole PBMC) (Fig 7C), suggesting that CD8 T cells inhibit ALVAC-specific CD4 proliferation via a cell-contact-independent mechanism. CD25+FoxP3+ regulatory CD8 T cells are an emerging CD8 subset with strong suppressive activities [45]. We measured CD25 and FoxP3 expression in vector-activated CD8 T cells on day 6 after initial vector stimulation and found that a much higher fraction of ALVAC-activated CD8 T cells were CD25+FoxP3+ (22.4%) as compared to Ad5-activated CD8 T cells (7.52%) (Fig 7D), suggesting that ALVAC-induced CD25+FoxP3+ CD8 T cells could play a role in inhibition of autologous vector-specific CD4 T-cell proliferation.
In addition to CD4 T cell inhibition, we also explored potential cytolytic effects of CD8 T cells from RV144 vaccine recipients on autologous CD4 T cells in response to ALVAC stimulation. Three conditions of RV144 PBMC were prepared as described above, including whole PBMC, CD8-depleted PBMC, and CD8 T cell addition back to trans-well culture (Fig 7E). On day 3 after ALVAC stimulation, before significant cell proliferation occurred in the culture (S9 Fig), the viability of total cells (CD3+ T cells and CD3- non-T cells) was measured by flow cytometry based on aqua blue staining (Fig 7E). We observed that compared to the whole PBMC that had only 15.1% live T cells, CD8-depleted PBMC had higher levels of live T cells (26.5%) (Fig 7E and 7F). Addition of the depleted CD8 T cells back to the trans-well culture decreased the level of live T cells (18.8%) (Fig 7E and 7F). The percent of live CD4 T cells (after subtracting CD8 T cells from the total live CD3+ T cells) in each condition was summarized and shown in Fig 7F. This data suggests that in ALVAC-stimulated PBMC, CD8 T cells can manifest a cytotoxic effect on the autologous CD4 T cells, which involves a cell-contact-independent mechanism. This cytotoxic effect of CD8 T cells may also contribute to the overall inhibition of ALVAC-specific CD4 T-cell expansion in our system.
In the context of vector HIV vaccination, it has been speculated that vector-induced CD4 T cells can be potential targets for HIV, which may affect the risk of HIV acquisition in vaccine recipients and overall outcome of vaccination [8]. Therefore, limiting the numbers and/or HIV susceptibility of vector-induced CD4 T cells in HIV vaccination is thought to be critical. We next explored the impact of vector-induced CD8 T cells on HIV susceptibility of autologous vector-specific CD4 T cells in PBMC, by using the above CD8-depletion assay. Whole or CD8-depleted PBMC were CFSE-labeled, and stimulated with vector antigen for 3 days, followed by infection with R5 or X4 HIV. Three days after infection, HIV infectivity in vector-specific CD4 T cells was measured by flow cytometry based on intracellular HIV p24 expression in CFSE-low CD4 T cells. We found that compared to whole PBMC, depletion of CD8 T cells from ALVAC-stimulated PBMC led to considerable increase in both R5 and X4 HIV infection of ALVAC-specific CD4 T cells (R5 HIV for CD8+ and CD8-: 2% vs. 6%; X4 HIV for CD8+ and CD8-: 7.4% vs. 13.8%) (Fig 8A). Analyses of PBMC from multiple subjects showed strong statistical significance between whole and CD8-depleted PBMC (p = 0.0006) (Fig 8B); in contrast, depletion of CD8 T cells in Ad5 vector-stimulated PBMC (HVTN204) had no significant impact on HIV infection rate of Ad5 vector-specific CD4 T cells (Fig 8A and 8B). Of interest, it should be noted that even in the absence of CD8 T cells (CD8 depletion), ALVAC-specific CD4 T cells were still significantly less susceptible to HIV infection than Ad5 vector-specific CD4 T cells (5.63 ± 1.84 vs 28.56 ± 5.16; p = 0.0002) (Fig 8B), suggesting that CD8 T cells contributed only partly to the low HIV susceptibility of ALVAC-specific CD4 T cells as compared to Ad5 vector-specific CD4 T cells. These data indicate that unlike Ad5 vector, ALVAC may induce vector-specific CD8 T cells that can not only inhibit the expansion of autologous vector-specific CD4 T cells, but also limit their susceptibility to HIV infection.
CD8 T cells can control viral infections through various mechanisms, including cytolytic activity and the secretion of soluble HIV-suppressive factors [46]. We next characterized potential mechanisms underlying CD8-mediated HIV inhibition in autologous ALVAC-specific CD4 T cells. First, we observed that CD8 depletion did not significantly affect the expression of CCR5 and T-cell activation markers (CD25 and CD69) on ALVAC-specific CD4 T cells (S10 Fig). We then performed a similar CD8 trans-well experiment to explore if the HIV inhibition by CD8 T cells is dependent of cell contact or soluble factors. We found that CD8 T cells could still inhibit R5 HIV infection in ALVAC-specific CD4 T cells even in the absence of direct cell contact (p24% for CD8- vs. trans-well CD8+: 11.1% vs. 5.1%) (Fig 8C), indicating that soluble HIV suppressive factors may play a role in this process. Consistent with this observation, we found that compared to Ad5 vector, ALVAC-induced CD8 T cells produced markedly higher levels of MIP-1β (MIP-1β+ % in Ad5 vs. ALVAC-induced CD8 T cells: 27.5% vs. 85.2%) (Fig 8D). Since in ALVAC-stimulated PBMC, high levels of CD8 T cells were induced (Fig 6), and β-chemokines were shown to mediate R5 HIV inhibition in ALVAC-specific CD4 T cells in our system (Fig 5), secretion of more β-chemokines might represent a mechanism for HIV inhibition in ALVAC-specific CD4 T cells by CD8 T cells. Lastly, we observed that compared to CD8-depleted PBMC, the presence of CD8 T cells in PBMC led to higher level of cell death (based on aqua blue staining) in CFSE-low, ALVAC-specific CD4 T cells (aqua blue staining in CD8+ vs. CD8-: 16.5% vs. 5.82%) (Fig 8E). This data suggests that the cytotoxic effects of CD8 T cells may also contribute to overall HIV inhibition in ALVAC-specific CD4 T cells.
Lastly, we characterized the poly-functional profile of ALVAC- and Ad5 vector-induced CD8 T cells by examining expression of antiviral and cytolytic effectors. CFSE-stained PBMC from RV144 or HVTN204 were re-stimulated with ALVAC or Ad5 vector, respectively, as described above. Six days after stimulation, cells were briefly treated with PMA and ionomycin for 6 hours to induce de novo re-synthesis of cytokines or effector molecules. Expression of IFN-γ, MIP-1β, CD107a, granzyme B (GZMB), and perforin in CFSE-low, vector-induced CD8 T cells was measured by flow cytometry. We found that compared to Ad5 vector, significantly higher percentages of ALVAC-induced CD8 T cells expressed IFN-γ (78.74 ± 12.50 vs 36.86 ± 7.57; p = 0.0210), MIP-1β (88.38 ± 4.753 vs 33.00 ± 4.51; p < 0.0001) and perforin (75.86 ± 9.139 vs 27.91 ± 8.369; p = 0.0047) (Fig 9). No significant difference in expression of GZMB (ALVAC vs. Ad5: 31.60 ± 9.720 vs 19.94 ± 5.913; p = 0.4261) and CD107a (ALVAC vs. Ad5: 15.32 ± 6.853 vs 6.893 ± 1.199; p = 0.2713) was observed between ALVAC- and Ad5-induced CD8 T cells (Fig 9). Altogether, these data suggest that ALVAC-induced CD8 T cells manifest a stronger antiviral and cytolytic phenotype than Ad5 vector-induced CD8 T cells.
In the present study, by using PBMC samples from two important HIV vaccine trials, we investigated host anti-vector T-cell responses induced by ALVAC and Ad5 vector in human vaccine recipients with a focus on the HIV susceptibility of vector-specific CD4 T cells. Our major finding is that different HIV vaccine vector-induced CD4 T cells manifest distinct susceptibility to HIV infection; while Ad5 vector-specific CD4 T cells are readily susceptible to HIV [21], ALVAC-specific CD4 T cells in RV144 PBMC are more resistant to both R5 and X4 HIV infection. Associated with this are the differences in phenotypes and cytokine profiles of these two groups of vector-specific CD4 T cells. Another major finding of our study is that in contrast to the lack of vaccine insert-specific CD8 T-cell response reported from the RV144 trial [27, 47], we demonstrate that ALVAC vector induces strong proliferative response of vector-specific CD8 T cells, which can limit the proliferation and HIV susceptibility of the autologous ALVAC-specific CD4 T cells.
The unexpected outcomes of human trials testing HIV vaccine regimens involving different viral vectors have suggested that assessment of both protective and potentially detrimental immune responses induced by vaccination is important [8, 48]. Development of a safe and efficacious HIV vaccine poses a unique challenge in that HIV infects the very CD4 T cells which are usually required to mount an effective adaptive response; this is of especial concern for viral vector vaccines because the expansion of vector-specific CD4 T cells following immunization can provide potential HIV target cells [8, 21], while presumably not contributing to anti-HIV immunity. From this point of view, it would be advantageous to employ vectors which generate fewer and/or less HIV-susceptible vector-specific CD4 T cells. Human CD4 T cells specific for different antigens or pathogens manifest differential susceptibility to HIV [20–26]. Our previous study has reported that human Ad5-specific CD4 T cells induced by natural infection or rAd5 vaccination are more susceptible to HIV infection [21]. This finding suggests that although Ad5 vectors have been commonly employed for vaccine development due to their potent immunogenicity [49], the advantages of Ad5 as a vector may be dampened by the high HIV susceptibility of CD4 T cells it induces. Our current study shows that unlike Ad5 vector, the vector-specific CD4 T cells induced by ALVAC in RV144 are markedly less susceptible to HIV infection (Fig 1). This finding is relevant to HIV vaccine development, considering that in the context of HIV vaccination, if vaccine-induced protective immunity is comparable between different vaccine regimens, the relative HIV susceptibility of vector-specific CD4 T cells may be an important factor that can affect the overall outcome of HIV vaccination. Future studies are being planned to examine the HIV susceptibility of CD4 T cells induced by other important HIV vaccine vectors, especially the adenovirus rare serotypes Ad26 and Ad35.
Parameters that influence HIV acquisition risk in HIV vaccination are thought to be complex, among which the level, quality (e.g. phenotypes, cytokine profile, and HIV susceptibility) and in vivo localization of induced CD4 T cells play important roles. Our data suggest that the high HIV susceptibility of Ad5 vector-specific CD4 T cells may be a contributing factor for the observed excess HIV infections in some Ad5-HIV vaccine recipients [5–7]. In addition, our ongoing studies examining in vivo localization and phenotypes of CD4 T cells following ALVAC and Ad5 immunization show that ALVAC immunization induces substantial lower levels of CCR5+CD4+ and CCR5+ α4β7+CD4+ T cells in various immune compartments, especially in the gut mucosa, of the immunized mice as compared to Ad5 immunization. Based on these findings, we propose that to better understand immune parameters associated with HIV acquisition risk in vector HIV vaccination, future studies are warranted to more thoroughly assess the frequency, quality and in vivo localization of vaccine-induced CD4 T cells in animal models and/or human trials.
HIV infection of antigen-specific CD4 T cells can be regulated at both entry and post-entry levels, and is closely associated with the phenotypic and functional characteristics of these CD4 T cells [20, 25, 33, 35]. CCR5 and CXCR4 as HIV entry co-receptors play major roles in regulating the susceptibility of target cells to HIV at entry level [50]. Our data show that ALVAC-specific CD4 T cells express markedly lower levels of CCR5 and CXCR4 than Ad5 vector-specific CD4 T cells (Fig 2), providing an explanation for the lower HIV susceptibility of ALVAC-specific CD4 T cells. We further identified that HIV infection rate in CCR5-/CXCR4- subset of Ad5-specific CD4 T cells remained higher than that in ALVAC-specific CD4 T cells (Fig 2C), suggesting that factors other than co-receptor expression are also involved in regulating the differential HIV susceptibility of vector-specific CD4 T cells in our system. Regulation of HIV co-receptor expression on target cells has been investigated previously in HIV pathogenesis [33, 51]. However, currently little is known about co-receptor regulation in HIV vaccination. Evidence from our ongoing studies suggests that innate signals derived from vector-infected APCs play a role in regulating CCR5 on CD4 T cells. Further understanding mechanisms that regulate HIV co-receptor expression on vaccine-induced cells is an interesting topic and should be pursued in future studies.
Another important factor that regulates HIV infection of CD4 T cells at entry level is β-chemokines, including CCL3 (MIP-1α), CCL4 (MIP-1β), and CCL5 (RANTES) [33, 40, 41]. Our data show that compared to Ad5 vector, ALVAC-induced T cells (CD4 and CD8) produce much higher levels of β-chemokines (MIP-1β) (Fig 5A); however, interestingly, neutralization of β-chemokines in ALVAC-stimulated PBMC only slightly increased HIV infection in ALVAC-specific CD4 T cells (Fig 5B), suggesting a modest role of β-chemokines in this process. In addition to co-receptors and β-chemokines, cytokine profiles of CD4 T cells are closely associated with HIV infection. It has been shown that IL-17-producing CD4 T cells are more susceptible to HIV than IFN-γ-producing CD4 T cells [20, 21, 23, 52, 53]. In our study, we demonstrate that while Ad5 vector-specific CD4 T cells manifest a mixed Th1/Th17 phenotype, producing high levels of IL-17 and IFN-γ [21], ALVAC-specific CD4 T cells display a polarized Th1-like phenotype, producing high level of IFN-γ but very little IL-17 (Fig 4C and 4D). This different cytokine profile of ALVAC- and Ad5-specific CD4 T cells is consistent with their susceptibility to HIV infection in our system.
CD8 T cells play important roles in anti-HIV immunity, including control of HIV replication and limiting HIV-infected cells [43, 44]. An interesting observation in the current study is that ALVAC and Ad5 vector stimulate distinct CD8 vs. CD4 T-cell proliferative responses; Ad5 vector stimulate predominantly CD4 T-cell proliferation, whereas ALVAC stimulate strong CD8 T-cell proliferation (Fig 6). This finding is somewhat unexpected since the ALVAC/gp120 vaccine regimen in the RV144 trial was reported to elicit a weak insert-specific CD8 response [27], whereas Ad5-HIV vaccines have been shown to induce a strong insert-specific CD8 response [5, 6, 30]. These findings suggest that the induction of anti-vector and anti-insert T-cell responses in vector HIV vaccination may be differentially regulated. In this study, mechanisms for differential stimulation of vector-specific CD8 vs. CD4 T-cell proliferation by ALVAC and Ad5 remain unknown. However, a prominent difference between ALVAC and Ad5 vector is related to their intracellular locations for replication. After entry into target APCs, poxvirus replicates in cytoplasm [54], whereas adenovirus replicates in nucleus [55]. This may lead to engagement of different antigen presentation pathways (e.g. MHC class I vs. II) and therefore differential induction of CD8 vs. CD4 T-cell responses to these two vectors. Nevertheless, elicitation of vector-specific CD8 vs. CD4 responses by different vaccine vectors in vivo and the immune pathways involved remain less clear and should be further investigated.
Another interesting finding of this study is that unlike Ad5 vector, ALVAC-activated CD8 T cells can inhibit the proliferation and HIV infection of autologous vector-specific CD4 T cells (Fig 7 and Fig 8). Evidence presented in our study supports that the process may involve both lytic and non-lytic effects of CD8 T cells [46]. First, our trans-well experiments showed that CD8 T cells could still inhibit ALVAC-specific CD4 T cell proliferation (Fig 7C) and HIV susceptibility (Fig 8C) even in the absence of cell contact, indicating that soluble factors play a role in mediating the inhibitory effects of CD8 T cells. Indeed, we demonstrate that compared to Ad5 vector, ALVAC-activated CD8 T cells manifest a stronger Treg potential (CD25+FoxP3+) (Fig 7D) and produce higher levels of β-chemokines (Fig 8D), which may respectively inhibit ALVAC-specific CD4 T-cell proliferation and HIV susceptibility. Other than the non-lytic mechanisms, our data suggest that the cytotoxic effects of CD8 T cells may also play a role. We found that the presence of CD8 T cells in either whole PBMC or in trans-well culture (depleted CD8 T cells were added back) caused significant cytotoxic effect on total CD4 T cells (Fig 7E and 7F) as well as on ALVAC-specific CD4 T cells (Fig 8E). In support, we further demonstrate that compared to Ad5 vector, ALVAC-activated CD8 T cells manifest a stronger cytolytic and antiviral phenotype, expressing elevated levels of perforin, IFN-γ, and MIP-1β (Fig 9). Collectively, our observation that preferential induction of strong vector-specific CD8, but not CD4, T-cell proliferation by ALVAC as compared to Ad5 vector provides some new insights into our understanding of vaccine-induced immunity in HIV vaccination.
In summary, we here present strong evidence that CD4 T cells activated via different HIV vaccine vectors manifest distinct susceptibility to HIV infection, which is closely associated with their phenotypic and functional characteristics. Our findings suggest that future efforts should focus on candidate vaccine vectors that can maximize immunogenicity while minimizing potential HIV susceptibility, for example, by inducing low levels of vector-specific CD4 T cells with high HIV resistance. Future studies will seek to extend this analysis to other important HIV vaccine vectors and to further explicate the mechanism underlying differential HIV susceptibility of vector-specific CD4 T cells. Research that aims to understand how vector-specific CD8 T cells may exert anti-HIV activity and the immune pathways by which ALVAC stimulates strong vector-specific CD8 T-cell proliferation should also be of interest.
The study involves use of PBMC samples from two HIV vaccine clinical trials: RV144 (NCT00223080) (ALVAC-HIV prime/gp120 protein boost) and HVTN204 (NCT00125970) (DNA prime/rAd5 boost). De-identified, cryopreserved PBMC collected from vaccine responders of these two trials were used. All samples were analyzed anonymously and investigators of this study have no access to any subject identification information. The study was determined as non-human subject research and approved by the University of Texas Medical Branch’s IRB. Written informed consents were obtained from study participants.
PBMC were maintained at 37°C, 5% CO2 in RPMI medium (Invitrogen) supplemented with 10% human serum, 100 U/mL penicillin G, 100 U/mL streptomycin sulfate, and 1.17mM sodium glutamine. R5 (US1) and X4 (92/UG/029) HIV-1 (original stock from NIH) was used for in vitro infection of PBMC. HIV transmitted founder virus (TFV) strains (including AD17 clone) were a kind gift from Dr. Jason Kimata of Baylor College of Medicine. Empty ALVAC vector was obtained from Sanofi, and empty rAd5 vector was obtained from the Vaccine Research Center (VRC) of NIH.
PBMC were CFSE labeled as described previously with slight modifications [20, 21, 23]. Thawed and washed PBMC at a concentration of 20 x 106 PBMC/mL were stained in 1μM CFSE for 8 minutes at 25°C. Cells were then quenched with 2 mL of warm normal human serum for 5 minutes. Empty ALVAC or rAd5 vector corresponding to the original vaccine was used to re-stimulate CFSE-labeled PBMC (MOI of 3). Unstimulated PBMC were included as a control. Three days after stimulation, cells were exposed to pre-titrated R5 HIV, X4 HIV, or TFV HIV for in vitro infection. Three days after HIV exposure, HIV infection in CD4 T cells was analyzed by flow cytometry based on intracellular HIV p24 expression. For viral kinetics experiments, HIV infection rate was measured at 3 and 9 days post infection. In some experiments, anti-MIP-1α (5μg/mL; clone 93321; R&D Systems), anti-MIP-1β (5μg/mL; clone 24006; R&D Systems), and anti-RANTES (5 μg/mL; clone 21418; R&D Systems) were added to the cultures throughout the experiments to neutralize β-chemokines. In some experiments, anti-human IFNAR antibody (Abcam, final concentration: 5 μg/ml) was added to the cultures throughout the experiments to block type-I IFN signaling.
In some experiments, CD8+ cells were depleted from PBMC using the EasySep Human CD8 Positive Selection Kit (Stem Cell Technologies, cat #17833) for comparison with whole PBMC. In the trans-well co-culture experiment, CD8 T cells were isolated from PBMC of RV144 vaccine recipients using the EasySep™ Human CD8+ T Cell Isolation Kit (StemCell Technologies) according to the manufacturer’s protocol after CFSE labeling. After CD8 T cell isolation, CD8 depleted PBMC and the corresponding whole PBMC were infected with ALVAC (MOI = 1), followed by HIV infection as describe above. In addition, isolated CD8 T cells were added back to the trans-well culture to explore mechanisms underlying CD8 T cell-mediated inhibition. Briefly, CD8-depleted PBMC were placed in the bottom chamber of the trans-well co-culture system, and the isolated autologous CD8 T cells were added back to the top chamber. The trans-well culture was also stimulated by ALVAC and infected with HIV as described above. HIV susceptibility and cellular phenotypes for different conditions (whole PBMC, CD8-depleted PBMC, CD8-depleted PBMC with added CD8 T cells in trans-well) were similarly measured by multi-color flow cytometry as described.
CFSE staining, vector stimulation and in vitro HIV infection of PBMC were conducted as described above. On day 6 after vector stimulation (3 days after HIV infection), cells were subjected to immune staining and flow cytometric analysis to examine the phenotypes and HIV susceptibility of vector-specific CD4 T cells. Cells were first stained with LIVE/DEAD fixable aqua dead cell stain (ThermoFisher Scientifc, cat #L34957) and antibodies to surface markers including CD3, CD4, CD8, CCR5, α4β7-APC (NIH AIDS Reagent Program), CCR7, PD-1, CD25 and CD45RO. Except α4β7, all surface antibodies were from BD Bioscience. Cells were then fixed, permeabilized (BD Biosciences cat #554722), and stained for HIV p24 (Beckman Coulter) for measuring HIV susceptibility of vector-specific CD4 T cells in PBMC (p24+ rate in CFSE-low CD4 T cells). In some experiments that also measured the expression of intracellular cytokines in vector-specific CD4 cells, cells were treated with phorbol 12-myristate 13-acetate (PMA) and ionomycin for 5 hours prior to staining in order to stimulate de novo cytokine production. After fixation and permeabilization, cells were also stained for intracellular cytokines IFN-γ, IL-2, IL-17, IL-21, (Biolegend), MIP-1β (BD Biosciences). In experiments that measured the antiviral and cytolytic profile of vector-specific CD8 T cells, anti-CD107a antibody (BD Biosciences) was added during cell stimulation. After fixation and permeabilization, cells were also intracellularly stained for perforin and Granzyme B (BD Bioscience). In experiments that measured regulatory T cells, cells were permeabilized using a FoxP3 Staining Buffer Set (eBioscience cat #00-5523-00) and stained for FoxP3 (Biolegend). Antibody capture compensation beads (BD Biosciences) stained with individual antibodies were prepared for compensation. Cell samples and compensation beads were acquired at LSR-II (BD). Flow cytometric data were analyzed using FlowJo Version 10 software (TreeStar).
Vaccine trial PBMC were CFSE stained and vector stimulated as described above. After 6 days of proliferation, cells were stained for CD3, CD4 and viability (Live/Dead Fixable Violet). The CFSE-low, CD3+CD4+ T cells were sorted from PBMC using FACSAria IIU (BD Biosciences). Total RNA was the sorted cells using Quick-RNA MicroPrep Kit (Zymo) according to the manufacturer’s protocol. Gene expression was quantified using iTaq Universal SYBR Green Supermix (Bio-Rad) and the CFX Connect Real-Time PCR Detection System (Bio-Rad) after reverse transcription from RNA into cDNA using iScript Reverse Transcription Supermix for RT-qPCR (Bio-Rad). Primer sequences for quantification of gene expression are shown in S1 Table. The relative quantity of gene expression was calculated using the 2−ΔΔCt method.
Statistical analysis was performed using GraphPad Prism 6 (GraphPad, Inc.) Two-tailed, unpaired Student’s T tests were performed and a p value ≤ 0.05 considered significant. Ratio-paired T tests were performed where appropriate.
|
10.1371/journal.pmed.1002835 | Biannual mass azithromycin distributions and malaria parasitemia in pre-school children in Niger: A cluster-randomized, placebo-controlled trial | Mass azithromycin distributions have been shown to reduce mortality in preschool children, although the factors mediating this mortality reduction are not clear. This study was performed to determine whether mass distribution of azithromycin, which has modest antimalarial activity, reduces the community burden of malaria.
In a cluster-randomized trial conducted from 23 November 2014 until 31 July 2017, 30 rural communities in Niger were randomized to 2 years of biannual mass distributions of either azithromycin (20 mg/kg oral suspension) or placebo to children aged 1 to 59 months. Participants, field staff, and investigators were masked to treatment allocation. The primary malaria outcome was the community prevalence of parasitemia on thick blood smear, assessed in a random sample of children from each community at study visits 12 and 24 months after randomization. Analyses were performed in an intention-to-treat fashion. At the baseline visit, a total of 1,695 children were enumerated in the 15 azithromycin communities, and 3,029 children were enumerated in the 15 placebo communities. No communities were lost to follow-up. The mean prevalence of malaria parasitemia at baseline was 8.9% (95% CI 5.1%–15.7%; 52 of 552 children across all communities) in the azithromycin-treated group and 6.7% (95% CI 4.0%–12.6%; 36 of 542 children across all communities) in the placebo-treated group. In the prespecified primary analysis, parasitemia was lower in the azithromycin-treated group at month 12 (mean prevalence 8.8%, 95% CI 5.1%–14.3%; 51 of 551 children across all communities) and month 24 (mean 3.5%, 95% CI 1.9%–5.5%; 21 of 567 children across all communities) than it was in the placebo-treated group at month 12 (mean 15.3%, 95% CI 10.8%–20.6%; 81 of 548 children across all communities) and month 24 (mean 4.8%, 95% CI 3.3%–6.4%; 28 of 592 children across all communities) (P = 0.02). Communities treated with azithromycin had approximately half the odds of parasitemia compared to those treated with placebo (odds ratio [OR] 0.54, 95% CI 0.30 to 0.97). Parasite density was lower in the azithromycin group than the placebo group at 12 and 24 months (square root–transformed outcome; density estimates were 7,540 parasites/μl lower [95% CI −350 to −12,550 parasites/μl; P = 0.02] at a mean parasite density of 17,000, as was observed in the placebo arm). No significant difference in hemoglobin was observed between the 2 treatment groups at 12 and 24 months (mean 0.34 g/dL higher in the azithromycin arm, 95% CI −0.06 to 0.75 g/dL; P = 0.10). No serious adverse events were reported in either group, and among children aged 1 to 5 months, the most commonly reported nonserious adverse events (i.e., diarrhea, vomiting, and rash) were less common in the azithromycin-treated communities. Limitations of the trial include the timing of the treatments and monitoring visits, both of which took place before the peak malaria season, as well as the uncertain generalizability to areas with different malaria transmission dynamics.
Mass azithromycin distributions were associated with a reduced prevalence of malaria parasitemia in this trial, suggesting one possible mechanism for the mortality benefit observed with this intervention.
The trial was registered on ClinicalTrials.gov (NCT02048007).
| The Macrolides Oraux pour Réduire les Décés avec un Oeil sur la Resistance (MORDOR) trial found that distributing the antibiotic azithromycin to all preschool children in communities in Niger was effective for preventing childhood mortality.
It is unclear how azithromycin might prevent mortality, though this antibiotic has moderate activity against malaria parasites.
Thirty communities from the same study area as MORDOR were randomized to receive the same interventions: either mass distribution of azithromycin or mass distribution of placebo to all children ages 1 to 59 months every 6 months over a 2-year period.
A blood smear test for malaria was performed on a random sample of 40 children per community before treatment and at 12 and 24 months after the initial distribution.
Communities randomized to the antibiotic had fewer malaria parasites at months 12 and 24 than those randomized to placebo.
Mass azithromycin distributions may reduce the burden of malaria in Niger, which may in turn contribute to the mortality benefit of this intervention.
Future studies will be important to study how frequently to give treatment, the effectiveness of azithromycin during various seasons of the year, and the generalizability of this finding to other settings in sub-Saharan Africa.
| Mass azithromycin distributions have lowered childhood mortality in Africa, although the mechanism explaining this effect is unknown. Macrolides Oraux pour Réduire les Décés avec un Oeil sur la Resistance (MORDOR) was a cluster-randomized placebo-controlled trial conducted in Malawi, Niger, and Tanzania that found a 14% reduction in childhood mortality in communities randomized to biannual mass azithromycin distributions targeted to 1- to 59-month-old children. The protective effect of azithromycin distributions was especially high in Niger, where malaria accounts for a large proportion of childhood deaths [1]. Azithromycin has activity against the malarial apicoplast and has demonstrated modest antimalarial activity in vitro and in numerous field studies [2–13]. Thus, the mortality benefit observed with mass azithromycin distributions may be due, in part, to decreased malaria.
MORDOR was designed as a large simple trial the outcome of which was childhood mortality as assessed on biannual census. Detailed health assessments were not performed in MORDOR in order to minimize co-interventions that could have biased the result of the trial. Instead, additional communities drawn from the MORDOR study area were enrolled into a parallel trial with the same interventions that also included annual monitoring visits to investigate possible mechanisms by which azithromycin impacted mortality. The present report details the results of blood smears processed for malaria in the parallel trial from Niger. Cluster randomization was employed to account for both the direct effects of the antibiotic as well as any indirect spillover effects of widespread community antibiotic use [14]. We hypothesized that mass azithromycin distributions would reduce the community prevalence of malaria parasitemia relative to placebo.
A parallel-group, cluster-randomized trial was performed in the Boboye and Loga departments of Niger from 23 November 2014 to 31 July 2017 (Fig 1). Communities were randomly selected from the same pool of communities as the main MORDOR trial in order to allow the findings of this trial to be generalizable to the parent study population (Fig 2). A set of 30 communities was randomized in a 1:1 ratio to biannual (i.e., every 6 months) mass treatment of preschool children with a single dose of either azithromycin or placebo (i.e., the same interventions offered in the main trial). Communities were followed annually with detailed morbidity assessments, including blood smears. Ethical approval was obtained from the Committee on Human Research at the University of California, San Francisco and the Institutional Review Board of the Nigerien Ministry of Health. The trial was reported according to Consolidated Standards of Reporting Trials (CONSORT) guidelines (S1 Checklist). Details of the study design were prespecified in a trial protocol (S1 Protocol).
The unit of randomization for the trial was the grappe, a government-defined health catchment area, termed “community” in this report. Communities with a population between 200 and 2,000 on the most recent government census were eligible. No mass azithromycin distributions for trachoma had been administered in the study area for the previous 5 years, and no seasonal malaria chemoprevention programs were implemented during the study period. All children aged 1 to 59 months of age who weighed 3,800 g or more were eligible for treatment. A random selection of 40 children aged 1 to 59 months per community was invited to provide a finger-stick blood specimen for thick and thin smear. Guardians of children provided oral informed consent for both treatment and monitoring.
Communities were randomly assigned in equal proportions to 1 of 6 letters by the trial biostatistician with the statistical package R (R Foundation for Statistical Computing, Vienna, Austria). The Nigerien study coordinator then enrolled communities and assigned the allocated intervention. Allocation was concealed at the cluster level by enrolling all communities before randomization and at the individual level by administering the treatment to all eligible children. Bottles of study drug were labelled by the manufacturer (Pfizer, New York, NY) with one of the 6 treatment letters, with 3 letters corresponding to azithromycin and 3 letters to placebo. Study bottles, packaging, and the appearance of the drug were identical. Participants, field personnel, laboratory staff, and all investigators except the trial biostatistician were masked to treatment allocation.
All households in each community were enumerated approximately every 6 months over a 2.5-year period using similar methods as the main trial [1]. All children aged ≤12 years in the household were documented.
A monitoring visit was performed during the first census period before distribution of the study drug (median date 26 March 2015) and then approximately 12 and 24 months later (median dates 23 June 2016 and 25 April 2017, respectively). The monitoring visits at months 12 and 24 occurred before distribution of study medication for the respective study period, as well as approximately 6 months after the treatment from the previous phase. Thus, the 12-month visit was performed approximately 6 months after the second round of mass treatment, and the 24-month visit was performed approximately 6 months after the fourth round of treatment (Fig 3). As part of the assessments, each child had a finger stick performed, with hemoglobin measurements estimated with a Hemocue Hb 201+ device (Ängelholm, Sweden) and malaria parasitemia assessed via thick and thin smear applied to a single slide. Slides were labelled with a random number sticker to mask the laboratory personnel. After the blood had dried, the portion of the slide with the thin smear was fixed with methanol; slides were subsequently stained with 3% Giemsa. Ill-appearing children also had a drop of blood tested with a rapid diagnostic test for malaria and were referred to the local government health facility if the test was positive. The thick smear was assessed for the presence and density of parasites and gametocytes by 2 independent experienced laboratory workers at the Centre de Recherche Médicale et Sanitaire (Niamey, Niger). Parasite density, measured in parasites per microliter, was estimated as the ratio of asexual parasites to white blood cells after inspecting a minimum of 200 white blood cells, multiplied by 8,000 (i.e., an arbitrary yet accepted convention for the number of white blood cells per microliter). Thin smears were assessed for malaria species. Laboratory workers were masked to treatment allocation. Discrepancies were adjudicated by additional independent laboratory workers until a majority consensus was achieved. A consensus for density measurements required a ≤1 log10-unit difference between the majority of measurements; the mean density from the consensus was taken as the study measurement.
A single, directly observed dose of study drug was administered to all eligible children approximately every 6 months. Study treatment was administered after monitoring was complete for all children, on a different day from the monitoring visit. Field staff distributed study drug in a single community at a time. The same smartphone-based mobile application used in the census was also used for treatment. All children aged 1 to 59 months entered into the most recent census were offered treatment, and children who were absent or not eligible at the time of the previous census were entered into the mobile application and offered treatment. Azithromycin was dosed at 20 mg/kg, calculated by weight for small children and approximated by height for children who could stand [15]. The same volume of an identical-appearing placebo suspension was administered in the placebo communities. Guardians were instructed to contact a village representative if their child experienced any adverse events within 7 days of receiving study drug; the representative then informed the study coordinator. A formal adverse event survey was performed for children aged 1 to 5 months and is reported separately [16].
The prespecified primary malaria outcome was the presence or absence of at least 1 parasite on thick blood smear in children 1 to 59 months of age, assessed as a community-level prevalence. Prespecified secondary outcomes included parasitemia density, gametocyte density, hemoglobin concentration at the individual level, and presence of anemia (hemoglobin < 11 g/dL) at the community level [17].
The MORDOR study sites in Malawi, Niger, and Tanzania conducted independent trials of morbidity outcomes with slightly different methods and outcomes. In consultation with the trial’s Data Safety and Monitoring Committee, we prespecified a separate analysis for each morbidity outcome at each site (S1 Protocol). Community-level prevalence outcomes from months 12 and 24 were modeled in a mixed-effects linear regression model that included a fixed term for treatment arm, baseline prevalence, and date of sample collection and a random effect for community. Individual-level outcomes from months 12 and 24 were analyzed in mixed-effects linear or logistic regression models that included fixed effects for treatment arm and date of sample collection and nested random effects for individuals within communities. Square root transformations were used for the parasitemia prevalence, anemia prevalence, and parasite density outcomes to improve the normality of residuals. Intraclass correlation coefficients (ICCs) were derived from the regression models at the level of the randomization unit. P values were determined by Monte Carlo permutation (10,000 permutations). All analyses were performed in an intention-to-treat fashion with the statistical software R version 3.4.0. The trial was overseen by a Data Safety and Monitoring Committee.
Based on previous studies in Niger, we anticipated a prevalence of malaria parasitemia in untreated communities of 25% and an ICC of 0.056. Assuming an alpha of 0.05, enrolling 40 children in each of 15 communities per arm would provide more than 80% power to detect a 13% absolute difference between the 2 treatment arms.
Baseline characteristics of the 15 azithromycin-treated communities and 15 placebo-treated communities are shown in Table 1. Communities in the placebo group were on average larger than those in the azithromycin group, with a total of 1,695 children aged 1 to 59 months enumerated in the azithromycin arm and a total of 3,029 children enumerated in the placebo arm at baseline. The age and sex distribution within communities was similar between the 2 groups. All communities received their allocated study medication, and none were lost to follow-up. Across all 4 study visits, study drug was distributed to 78.7% (95% CI 74.9%–82.6%) of eligible children in the azithromycin group and 81.7% (95% CI 78.9%–84.5%) of eligible children in the placebo group (Fig 1).
No hospitalizations or life-threatening illnesses were reported in either group over the duration of the study. Guardians of 1- to 5-month-old children were surveyed about adverse events approximately 1 month after treatment given the paucity of azithromycin safety data in this population. Detailed results have been published elsewhere [16]. In summary, the most common guardian-reported adverse events were as follows: diarrhea, which was reported in 110 out of 571 (19.3%) children in the azithromycin group and 321 out of 1,141 (28.1%) children in the placebo group (P = 0.03); vomiting, reported in 91 out of 571 (15.9%) from the azithromycin group and 240 out of 1,141 (21.0%) from the placebo group (P = 0.07); and rash, reported in 70 out of 571 (12.3%) from the azithromycin group and 155 out of 1,141 (13.6%) from the placebo group (P = 0.07).
The prevalence of parasitemia among 1- to 59-month-old children for each community over time is shown in Table 2 and Fig 4. At baseline, the mean prevalence of malaria parasitemia was 6.7% (95% CI 4.0%–12.6%) in the placebo group and 8.9% (95% CI 5.1%–15.7%) in the azithromycin group. At month 12, the mean prevalence was 15.3% (95% CI 10.8%–20.6%) and 8.8% (95% CI 5.1%–14.3%) in the placebo and azithromycin groups, respectively, and at month 24, these estimates were 4.8% (3.3%–6.4%) and 3.5% (1.9%–5.5%), respectively. Parasitemia prevalence was significantly lower in the azithromycin group at months 12 and 24 after adjusting for baseline parasitemia and the date of sample collection (P = 0.02; ICC = 0.26; square-root–transformed outcome; prespecified primary outcome). If the mean malaria prevalence was 10% in the placebo-treated communities (i.e., the post-treatment average observed in the present study), then azithromycin would be predicted to reduce the prevalence of malaria parasitemia by an absolute difference of 5.3% (95% CI −1.4 to −8.0%). The conclusions did not change in a sensitivity analysis of individual-level data (odds ratio [OR] 0.54 relative to the placebo group, 95% CI 0.30–0.97, adjusted for date of sample collection).
The average parasite density among parasitemic children is shown for each treatment arm in Table 3. Parasite density was lower in the azithromycin group in a mixed-effects linear regression of the 12- and 24-month values adjusted for the date of sample collection, with density estimates 7,540 parasites/μl lower (95% CI −350 to −12,550 parasites/μl) than the placebo arm assuming a mean parasite density in the placebo arm of 17,000 as observed in this study (ICC = 0.02; P = 0.02; square-root–transformed outcome). No significant differences in gametocyte prevalence or density were observed between the 2 groups (S1 Table, S2 Table).
Mean hemoglobin measurements are summarized for each study visit and treatment arm in Table 3. No significant difference was observed between the 2 groups after adjusting for date of sample collection, although the azithromycin group had slightly higher hemoglobin measurements (mean 0.34 g/dL higher than the placebo group, 95% CI −0.06 to 0.75 g/dL; ICC = 0.09; P = 0.10). Similarly, the prevalence of anemia was not significantly different between the 2 groups in the permutation test (P = 0.06; ICC = 0.08; square-root–transformed outcome; S3 Table), although the regression model predicted less anemia in the azithromycin-treated communities (absolute difference 6.7%, 95% CI −0.2 to −12.8%, assuming a 75% prevalence of anemia as approximately observed in the placebo-treated communities).
In this placebo-controlled study, mass azithromycin distributions resulted in a significantly lower prevalence of malaria parasitemia and lower parasite density levels when assessed at 2 annual post-treatment study visits. Hemoglobin levels were slightly greater and the prevalence of anemia slightly lower in azithromycin-treated communities, although neither of these were statistically significant. These results raise the possibility that reductions in malaria may in part explain the childhood mortality benefit of azithromycin.
The prevalence of malaria parasitemia was lower in this study than in several other population-based studies done in other parts of Niger [9, 13]. This may be explained by geographic and year-to-year variations in malaria prevalence [18–20]. The relatively low malaria prevalence could also be due to the timing of the annual monitoring visits, which were done in the spring, before the major seasonal peak in the fall. The prevalence of malaria would be expected to increase later in the summer. Indeed, the malaria estimates were higher in each group at the month 12 assessments, which on average occurred several months later than at the other 2 time points. The preponderance of malaria during the autumn months in Niger was suggested by the parent MORDOR trial since the largest number of childhood deaths occurred during this time and malaria was the most common attributed cause of death from verbal autopsies, although it is important to note that verbal autopsy has poor diagnostic accuracy for malarial deaths and can lead to overestimates in areas with endemic malaria like Niger [1, 21, 22]. The strength of the statistical association between mass azithromycin treatments and malaria could depend on the timing of treatments as well as the timing of assessments, so it is possible that the present trial did not maximize the chance of finding an antimalarial effect of azithromycin. Nonetheless, our finding of reduced parasitemia in the azithromycin group even in the presumably low-prevalence spring months is consistent with azithromycin exerting a sustained reduction in malarial burden throughout the year.
The present study is consistent with several previous studies that have assessed the antimalarial activity of azithromycin. As is the case with some other antibiotics, azithromycin acts against the parasite apicoplast and exerts slow antimalarial activity [6]. In clinical trials, azithromycin has been found to improve the treatment efficacy of both artesunate and chloroquine relative to each agent alone [4, 5, 11]. Azithromycin has also been shown to prevent malaria infections, although its efficacy appears to be inferior to that of doxycycline [2, 3]. Malaria outcomes have been measured during studies of mass azithromycin distributions for trachoma, with reductions in malaria parasitemia observed in some studies but not in others [9, 13, 23, 24]. The present study improved upon previous studies of mass azithromycin distribution in that it was placebo controlled and enforced strict masking procedures both in the field and by laboratory staff.
The public health impact of MORDOR remains to be seen. The mortality benefit of mass azithromycin distributions was especially strong for the youngest children and for the Niger site, so it may ultimately make sense to target antibiotic distributions to those most likely to benefit. Doing so would have the added benefit of reducing the total volume of antibiotics distributed, which may limit antimicrobial resistance. Studies like this one may help elucidate where to target antibiotics. Namely, our results suggest that mass azithromycin distributions might be more effective in reducing mortality in places with prevalent malaria. Subsequent studies assessing the impact of mass azithromycin on other common causes of childhood mortality will be important, as will studies in areas without prevalent malaria. For example, if the mortality benefit of mass azithromycin is mediated only through malaria reduction, then seasonal malaria chemoprevention might be a more efficient way to prevent childhood mortality. In fact, a recent household-randomized trial found that adding azithromycin to seasonal malaria chemoprevention did not provide a benefit for malaria or mortality, perhaps because any antimalarial effect of azithromycin was reduced when given along with the more effective antimalarial drugs [25]. However, that same study also found that azithromycin treatment was associated with reductions in diarrheal and respiratory infections. If other studies confirm these associations, wider implementation of mass azithromycin distributions may be warranted.
Although the most likely explanation for the observed result is a direct effect of azithromycin on those children who took the drug, it is important to note that mass antibiotics may also have indirect spillover effects [14]. By reducing the overall burden of parasitemia, mass azithromycin distributions reduced the reservoir of parasites in the community, which may have supplemented any direct effect of the antibiotic.
The present trial has several limitations. Study drug was generally administered in the dry seasons. Although mathematical models have suggested that this may in fact be the optimal time to reduce the community burden of malaria, the ideal time to administer antibiotics remains unclear [26]. Likewise, due to logistical reasons, monitoring visits were conducted in the dry season, which may not provide a complete picture of the impact of the intervention during the peak malaria transmission season. Grading of thick smears is inherently subjective, although the possibility of misclassification error was mitigated by having at least 2 independent laboratory staff assess each slide. Finally, the generalizability of the results outside Niger remains unclear, and even the generalizability within Niger is not assured given considerable variation in malaria transmission within the country [18–20]. The other 2 MORDOR sites in Malawi and Tanzania had a lower reduction of mortality in subgroup analyses, and thus the malaria results for these sites will provide important context for the interpretation of the present report and may ultimately alter our understanding of the causal relationship between mass azithromycin, malaria, and mortality.
In conclusion, a placebo-controlled, cluster-randomized trial found that biannual mass azithromycin distributions targeted to preschool children in Niger resulted in a reduction in malaria parasitemia in 1- to 59-month-old children. Study communities were selected from the same set of communities and treated with the same intervention as the parent MORDOR trial, so results of this study can be extrapolated to the larger trial. As such, this study suggests that the mortality benefits conveyed by mass azithromycin distributions may have been in part due to the antimalarial activity of azithromycin. Similar trials performed in other settings will be important to confirm the association found in this trial.
|
10.1371/journal.pcbi.1002078 | Enzyme Kinetics of the Mitochondrial Deoxyribonucleoside Salvage Pathway Are Not Sufficient to Support Rapid mtDNA Replication | Using a computational model, we simulated mitochondrial deoxynucleotide metabolism and mitochondrial DNA replication. Our results indicate that the output from the mitochondrial salvage enzymes alone is inadequate to support a mitochondrial DNA replication duration of as long as 10 hours. We find that an external source of deoxyribonucleoside diphosphates or triphosphates (dNTPs), in addition to those supplied by mitochondrial salvage, is essential for the replication of mitochondrial DNA to complete in the experimentally observed duration of approximately 1 to 2 hours. For meeting a relatively fast replication target of 2 hours, almost two-thirds of the dNTP requirements had to be externally supplied as either deoxyribonucleoside di- or triphosphates, at about equal rates for all four dNTPs. Added monophosphates did not suffice. However, for a replication target of 10 hours, mitochondrial salvage was able to provide for most, but not all, of the total substrate requirements. Still, additional dGTPs and dATPs had to be supplied. Our analysis of the enzyme kinetics also revealed that the majority of enzymes of this pathway prefer substrates that are not precursors (canonical deoxyribonucleosides and deoxyribonucleotides) for mitochondrial DNA replication, such as phosphorylated ribonucleotides, instead of the corresponding deoxyribonucleotides. The kinetic constants for reactions between mitochondrial salvage enzymes and deoxyribonucleotide substrates are physiologically unreasonable for achieving efficient catalysis with the expected in situ concentrations of deoxyribonucleotides.
| The powerhouses of human cells, mitochondria, contain DNA that is distinct from the primary genome, the DNA in the nucleus of cells. The mitochondrial genome needs to be replicated often to ensure continued generation of ATP (adenosine triphosphate) which is the energy currency of the cell. Problems with maintenance of mitochondrial DNA, arising from genetic mutations as well as from antiviral drugs, can lead to debilitating diseases that are often fatal in early life and childhood, or reduced compliance to therapy from patients suffering drug toxicity. It is therefore important to understand the processes that contribute to the upkeep of mitochondrial DNA. The activities of a set of enzymes, which together generate the chemical building blocks of mitochondrial DNA, are important in this regard. We used computational methods to analyze the properties of these enzymes. Results from our approach of treating these enzymes as a system rather than studying them one at a time suggest that in most conditions, the activities of the enzymes are not sufficient for completing replication of mitochondrial DNA in the observed duration of around 2 hours. We propose that a source of building blocks in addition to this set of enzymes appears to be essential.
| Mitochondrial DNA (mtDNA) replication [1], and the mitochondrial nucleoside salvage pathway that generates the precursor deoxyribonucleoside triphosphates (dNTPs) for mitochondrial DNA replication, have generally been believed to function independently of nuclear DNA (nDNA) replication and cytoplasmic nucleotide metabolism. However, the observation associating mutated RRM2B (a p53 inducible ribonucleotide reductase subunit) with mtDNA depletion and at least one observation of mtDNA replication restricted to S phase in DGUOK (deoxyguanosine kinase) deficient cells now make it clear that mtDNA replication and maintenance are not always completely independent of the cytoplasmic state [2], [3]. Older evidence supported the view that mitochondrial nucleotides may be isolated from the corresponding cytoplasmic pools [4], but more recent studies support a metabolic cross-talk between the mitochondria and the cytoplasm and show that nucleotide import from the cytosol very likely contributes to mitochondrial dNTP pools in both cycling and quiescent cells [5], [6]. The mechanism of this import is unknown since the discovery that the carrier SLC25A19 (solute carrier family 25, member 19) actually is a thiamine pyrophosphate transporter and not a deoxyribonucleotide transporter [7]. Similarly, the mitochondrial monophosphate kinases of dG and T deoxyribonucleotides (key elements of the purported salvage pathway) still have not been identified. In the current picture of the mitochondrial nucleoside salvage pathway, DGUOK and TK2 (thymidine kinase 2) are the nucleoside kinases; NT5M (mitochondrial 5′,3′-nucleotidase) is a nucleotidase; CMPK2 (cytidine monophosphate kinase 2), and isoforms of adenylate kinase (AK) are the monophosphate kinases; and NME4 is the major nucleoside diphosphate kinase (Figure 1A). Deoxyribonucleosides (dNs) are converted to dNTPs through three sequential enzyme-catalyzed phosphorylations. This is a complex process with some reactions occurring in parallel for the four deoxyribonucleosides, and some reactions using the same enzyme (for example, the first phosphorylation of dT and dC are both catalyzed by TK2) in addition to the presence (not shown in Figure 1A) of feedback mechanisms (for example, dTTP and dCTP inhibition on TK2 [8]).
The physical structure of the mitochondrion provides another complication that is rarely considered in this context. The mitochondrion has an intermembrane space (between the inner and outer membranes) and a matrix compartment within the inner membrane (Figure 1B). Several contact sites exist between the inner and outer membranes. In addition to the mitochondrial enzymes listed in Figure 1A, the cytoplasmic enzymes Thymidine Phosphorylase (TYMP) and RRM2B are included in the diagram since mutations in these two enzymes are known to cause phenotypes involving defects in mtDNA maintenance [2], [9]. The mtDNA are tethered to the inside of the inner membrane, within the matrix, so it would be expected that the enzymes of the salvage pathway would also be located within the matrix. In the simplest picture of the mitochondrial salvage pathway deoxyribonucleosides are transported through the inner membrane by the ENT (equilibrative nucleoside transporter) and then phosphorylated to dNTPs within the matrix. However, evidence exists to suggest that the AK2 adenylate kinase as well as NME4 nucleoside diphosphate kinase might actually be localized to the mitochondrial intermembrane space [10], [11], not in the matrix. It is possible that other isoforms of these enzymes might localize to the mitochondrial matrix [10], [11]. If not, it is hard to understand how the salvage pathway would function without an unnecessarily complicated transport of deoxyribonucleotides back and forth across the inner membrane (arrows marked with question marks in Figure 1B).
In this paper we analyze the experimentally measured enzyme kinetics of these known enzymes of this pathway. Our analysis of the mitochondrial nucleotide metabolism pathway reveals that the majority of the enzymes of this pathway are not particularly effective in the synthesis of mtDNA precursors (phosphorylated deoxyribonucleosides) either due to the affinities of the enzymes for ribonucleotides and other non-DNA precursors (dI and dUMP for example) or due to a disparity in their affinities for deoxyribonucleotides versus the expected mitochondrial concentrations of those deoxyribonucleotide substrates. Computational simulations of the function of this pathway support our analysis and indicate that a source of deoxyribonucleotides in addition to those provided by mitochondrial salvage is essential to account for the experimentally observed mtDNA replication duration of 1 to 2 hours in cycling cells.
As far as possible, we restricted our analysis to data from human enzymes. Exceptions are noted below. We assumed Michaelis-Menten kinetics for all enzymes except TK2 which has negatively cooperative kinetics with Hill coefficient less than 1 with thymidine [8].
Km values were obtained from the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] or the BRENDA database [23]. For most enzymes, we could only find a single report of kinetic parameters. For the nucleoside kinases DGUOK and TK2, we did find multiple reports of kinetic parameters. In these cases, we selected the reference providing the most comprehensive information. To compute kcat values, we first obtained reported Vmax values [8], [12], [14], [15], [16], [18], [19], [20], [21], [22] and molecular weights [13], [22], [23], [24] of the various enzymes from the literature or the BRENDA database. If the enzyme was reported to be a multimer, we added the molecular weights of the subunits to calculate the molecular weight of the holoenzyme. The quantity kcat/Km (M−1 s−1) was calculated from the reported values of Vmax/Km (with units of µmol min−1 mg−1 µM−1) using the following conversion,where Wenzyme is the enzyme molecular weight. Reported values for Km, Vmax, and the calculated kcat/Km values are provided in Table S1.
Values for the concentrations of the deoxyribonucleoside, deoxyribonucleotide, ribonucleoside, and ribonucleotide substrates were used to calculate ‘(substrate) Concentration/(substrate) Km’ ratios. These values were used for a comparison of activities of the enzyme with different substrates and are not meant to be precise. Instead, rough order-of-magnitude concentration values were used to compare values for this ratio, which often varies by several orders of magnitude within a single enzyme for different substrates. Literature reports suggested that mitochondrial dNTP pools are higher in actively cycling cells compared to quiescent cells [25], [26], [27], [28]. We assumed a 10-fold lower concentration of deoxyribonucleotides in quiescent cells, and chose 10 µM and 1 µM as reasonable representative estimates of mitochondrial dNTP concentrations in cycling and quiescent cells respectively. The basis of these estimates are the concentrations calculated from published values in HeLa cells [29] and quiescent fibroblasts [30] respectively. We used a value of 0.82 ml/g mitochondrial protein [31] to calculate concentrations from the measured pool sizes in HeLa cells, and we used the value of 92.3 µm3 for mitochondrial volume per cell [32] to obtain the concentrations from the measured pool size in quiescent fibroblasts. For simplicity, we assumed ribonucleotides and deoxyribonucleotides to be equally concentrated in the three phosphorylation states (mono, di, or tri-phosphorylated). Again for simplicity we assumed all four nucleotides (dAXP, dCXP, dGXP, dTXP where X = phosphorylation state) to have equal concentrations. Nucleoside concentrations were assumed to be equilibrated between plasma, cytoplasm, and mitochondria and set at a constant 0.5 µM using a reported value for plasma concentration [33]. Lower nucleoside concentrations have also recently been reported [34], [35]. We have kept the higher value in our analysis since this is the most conservative choice. Lower nucleoside concentration values would make the problems that we point out in this analysis even more severe. Ribonucleotide concentrations were assumed to be constant and set at 100 µM, that is, one order of magnitude higher in cycling cells and two orders of magnitude higher in quiescent cells compared to deoxyribonucleotide concentrations. This is a fairly conservative (i.e. low) choice for the ribonucleotide concentrations. For other special cases of substrates (such as dUMP, dI, or IMP) concentrations data are not readily available so we again assumed low concentration values for these substrates. The complete list of assumed concentrations is provided in Table S1.
In the case that we could not find Ki values of for enzyme inhibitors, we assumed competitive inhibition so that the Ki for the inhibitor was set to be equal to the Km for that chemical as a substrate. Inhibition kinetics data [8], [12], [16], [19], [21], [22], [36] are provided in Table S1.
Our group has previously published a computational model of mitochondrial deoxyribonucleotide metabolism [25]. Parameter values for the model were based, whenever available, on published experimental values [8], [12], [14], [16], [17], [19], [20], [21], [22], [23], [24], [25], [36], [37], [38], [39], [40], [41], [42], [43]. As part of the present work, we updated the model to reflect the findings since the original model was defined. We refer the readers to the previous publication for a complete explanation of the basic framework of the model [25]. Briefly, enzymatic reactions were modeled with Michaelis-Menten equations (except TK2, which is modeled by the Hill equation) and rates of change of metabolites were modeled using ordinary differential equations. The updates to the model include adding (e.g. CMPK2) and removing (e.g. SLC25A19 or DNC) pathway components and updated kinetics (e.g. inhibition terms and kinetic constants). The model was written in Mathematica 7. The model files are available as supporting information (Text S1). The model constants are also available as supporting information in plain text (Text S2) and PDF (Text S3).
Deoxynucleoside transport was modeled through the ENT protein as equilibrative between the cytoplasm and mitochondria. Thus, the net rate of deoxynucleoside transport was defined using the Michaelis-Menten equation as follows:where j represents the four deoxynucleoside species (dA, dC, dG, dT) and i represents inhibitors. Vmax and Km were taken to be the same for both directions of transport.
The various enzymatic reactions (i.e., phosphorylations and dephosphorylations) were modeled using the Michaelis-Menten equation. Thus, the reaction velocity waswhere S stands for substrate and [C] stand for the concentration of any competitive inhibitors. For the reaction of dT with TK2, the above equation was modified by raising the Km and [S] terms to the power 0.5 to represent the Hill coefficient.
The model of the mtDNA polymerization process was explained in the previous publication [25]. It models polymerization using fractions of the four deoxynucleotides in the mtDNA sequence, setting the prevalence of each base in the mtDNA light and heavy strands separately to match the prevalence in the rCRS reference sequence [44]. We have modeled mtDNA replication as asynchronous [45] using the locations of the origins of replication of the light strand and the heavy strand.
Differential equations for the concentrations of the various metabolites were defined by adding and subtracting the relevant reaction velocity equations. For example, for dNMPs (deoxyribonucleoside monophosphates), the following differential equation models the rate of change of a particular dNMP:where NK represents the nucleoside kinase reaction, NT represents the nucleotidase reaction, NMPK represents the forward and reverse monophosphate kinase reactions. The kinetic constants and inhibition parameter values are available in Table S1.
We used this updated model to test the hypothesis that a source of deoxyribonucleotides in addition to intra-mitochondrial salvage is essential for completing mtDNA replication in cycling cells in the experimentally observed time of 1–2 hours [45]. To be conservative, we set the ‘target’ replication time to be 2 hours (requiring an average replication rate = 33136 (nucleotides)/120 (minutes) = ∼276 nucleotides/minute). We ran simulations with a simulation time of 120 minutes (2 hours), with all dynamics including mtDNA replication starting immediately at the beginning of the simulation. We also tested a target replication time of 10 hours (requiring an average replication rate = 33136 (nucleotides)/600 (minutes) = ∼55 nucleotides/minute) – reasoning that in quiescent cells the time constraints for completing mtDNA replication may be more relaxed.
Transport of deoxynucleotides from the cytoplasm to the mitochondrial matrix was modeled in a simple manner, by setting a constant production term of either deoxynucleosides, dNMPs, dNDPs (deoxyribonucleoside diphosphates), or dNTPs. Transport was modeled as occurring at only one phosphorylation level at a time, in order to assess the effectiveness of transport at each level. The essence of our simulation experiments was to test whether mtDNA replication was completed in the target time under varying levels of added molecules, including no addition, of various (A, C, G, T) deoxynucleosides and deoxynucleotides. We note that in principle the additional source of deoxynucleotides in this model does not necessarily have to be import from the cytoplasm, but could also be from other unknown intra-mitochondrial sources. However, considering the evidence that nucleotide transport does occur between the cytoplasm and mitochondria [5], [6], we assume that the additional source we have modeled corresponds to import from the cytoplasm. We tested multiple ‘transport profiles’. A transport profile is composed of simply the rate of the transported deoxynucleosides and deoxynucleotides. For each transport profile, we ran 100 simulations each beginning with a different, randomly selected (with uniform probability) set of initial mitochondrial concentrations of each deoxynucleoside and deoxynucleotide. As an initial test of the level of exogenous precursor transport needed, we set equal rates of import for all four (A, C, G, T) nucleosides (or nucleotides) at a particular phosphorylation level and then let the rate of import vary from 0 to 1200 molecules per minute, in increments of 100. Thus, for example, for testing whether transport of deoxynucleosides alone suffices, we ran 13 sets of 100 simulations. In each of those 13 sets, deoxynucleosides alone were imported at equal rates for each of the four nucleosides, in increments of 100 starting from 0 and up to 1200. Such simulation sets of 13 different import levels were conducted similarly for each phosphorylation level of the four deoxynucleoside species.
The initial conditions of the simulations were set randomly with a uniform distribution over a set range. The allowed range (minimum and maximum) of initial deoxynucleoside concentrations was 0.05 µM to 5 µM and the range of initial deoxynucleotide concentrations was 0.1 µM to 10 µM. We set the concentrations of ribonucleosides, ribonucleotides, and non-canonical deoxynucleosides and deoxynucleotides to be proportional to the randomly selected dN and dNXP concentrations (see Table S1 for details), and held these concentrations (which only acted as inhibitors) constant throughout the time course of the simulation. The simulations were repeated 100 times with varying initial conditions.
We extended the transport analysis further by obtaining the minimum number of molecules of each transported dNTP required for mtDNA replication to be completed in 2 hours (representing cycling cells) or 10 hours (representing quiescent cells). For the simulations to determine the minimum transport profiles, we tested whether the replication rate exceeded 55 (‘quiescent cells’, fixed initial concentrations: dNs = 0.5 micromolar and dNXPs = 1 micromolar) or 276 (‘cycling cells’, fixed initial concentrations: dNs = 0.5 micromolar and dNXPs = 10 micromolar) nucleotides per minute. We started at equal import of all four dNTPs at a rate such that replication would be completed in slightly less than the target time (2 hours or 10 hours). Next, we decreased the import of one dNTP at a time to check whether the target replication rate was observed. We continued this relaxation process until we obtained the minimum transport for each individual deoxyribonucleotide species necessary to support the target replication rate.
In Michaelis-Menten kinetics, kcat and Km are the basic parameters of an enzyme-substrate reaction pair. The parameter kcat is the number of substrate molecules catalyzed per enzyme molecule per unit time and Km is the substrate concentration at which the reaction proceeds at half-maximal velocity. High kcat and low Km values imply a fast and efficient reaction, and thus, a high kcat/Km ratio indicate that this substrate is catalytically preferred by the enzyme. We searched the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] and databases [23] and gathered the available data on the reaction kinetics of enzymes of mitochondrial nucleotide salvage. Figure 2A shows a plot of kcat/Km values. Each group of bars is for one enzyme, and within each group the bars are arranged from lowest to highest so that the best substrates lie to the right on each plot. For clarity, the substrates that are DNA precursors (presumed to the ‘proper’ substrates of these enzymes) are in green, and non-DNA precursor substrates are in red. The kcat/Km values cover a very wide range and so are plotted on a logarithmic scale.
Figure 2A shows that each of these enzymes has significant reactions with non-DNA precursors. More importantly, except for TK2, none of the mitochondrial enzymes have DNA precursors as their preferred substrates, as seen from the fact that the substrates which lie to the right in each group of bars are non-DNA precursors. Prior work [46] has estimated the theoretical maximum of kcat/Km for an enzyme-substrate pair. This maximum is constrained by the diffusion limit, and was estimated to be ∼108 per M per second [46]. Compared to the diffusion limit, the kcat/Km values for reactions of the mitochondrial salvage pathway with DNA precursors are orders of magnitude lower (range = 888 to 5.63×105 per M per second). In summary, in both absolute and relative terms these enzymes of the mitochondrial salvage pathway (with the possible exception of TK2) do not appear to be optimized for discriminating mtDNA precursor substrates from chemically related non-precursor substrates.
To put the kcat/Km results in Figure 2A in perspective, Figure 2B is a plot of kcat/Km for the various substrates of the mitochondrial DNA polymerase gamma (POLG). In contrast to the enzymes of mitochondrial nucleotide metabolism, Figure 2B shows that, as expected, DNA precursors are preferably discriminated by POLG. This is true both absolutely and relatively. The kcat/Km values for the dNTP substrates approach the diffusion limit of ∼108 per M per s, and the values for dNTP substrates are many orders of magnitude larger than the kcat/Km values of the ribonucleotide substrates. GTP and UTP kinetics data are not shown because the POLG kinetics with these potential substrates have not been measured.
While the ratio kcat/Km captures the efficiency of a reaction between an enzyme and a substrate, it does not take into account the expected physiological concentration of the substrate, which may vary by several orders of magnitude between ribonucleotide and deoxyribonucleotide substrates. The ratio of ‘(substrate) Concentration/(substrate) Km’ provides information that is complementary to that revealed in the previous section by the ratio kcat/Km. When the substrate concentration is much smaller compared to the Km, the enzyme is sensitive to substrate concentration and can thus operate at a range of velocities. However, the velocities in this range would be smaller than the maximum possible velocity. Depending on the relation between maximum possible velocity and the required rate of enzymatic output, substrate concentrations smaller than Km can be a detriment. This is the case for the mitochondrial salvage enzymes because mtDNA replication has to satisfy certain time constraints. We searched the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] and databases [23] for Km values of the mitochondrial salvage enzymes for various substrates and their expected in situ concentrations. Figures 3A and 4A show a plot of Concentration/Km values for all of the enzyme-substrate pairs for which we could find data. As before, each group of bars is for one enzyme, and within each such group the bars are arranged from lowest to highest value of the ratio. Preferred substrates would be expected to have higher concentrations relative to the reaction Km and thus would fall to the right in each enzyme. Substrates that are DNA precursors are plotted in green, and non-DNA precursor substrates are in red.
Figure 3A shows Concentration/Km values at higher mitochondrial concentrations (‘cycling cells’) of the deoxyribonucleotide substrates (10 µM). The Concentration/Km values for DNA precursor substrates range from 0.001 to 0.19. Thus, none of the reactions involving DNA precursor metabolism in the mitochondria would be running at maximal reaction velocity. In fact, since all reactions involving DNA precursor substrates have Concentration/Km values less than 1, none of these reactions would be expected to be running at even half-maximal velocity. It is apparent that these enzymes of mitochondrial nucleotide metabolism have significant affinities for non-DNA precursors. In many cases, the enzymes have higher affinities for non-DNA precursors than for the DNA precursors. In the case of the nucleoside kinases TK2 and DGUOK, although they have higher affinities for DNA-precursors, there is less than 10-fold difference from their preference of non-DNA precursors. For some reactions, the expected substrate concentrations are orders of magnitude lower than the reaction Km values (range: 0.001 to 0.19). The same trends exist in the values of Concentration/Km assuming lower mitochondrial concentrations of substrates (Figure 4A). Moreover, comparing Figure 4A (low deoxyribonucleotide concentrations) to Figure 3A (high deoxyribonucleotide concentrations), the disparity between DNA precursors and other substrates is more striking with an order of magnitude decrease in the Concentration/Km ratio of the DNA precursor substrates (range: 0.0007 to 0.19). This was expected as we assumed mitochondrial ribonucleotide concentrations to be constant and independent of high or low mitochondrial deoxyribonucleotide concentrations.
To place these enzyme kinetics values in context Figures 3B and 4B show positive and negative examples justifying the principle of using the ratio Concentration/Km as a measure of substrate preference. The Concentration/Km ratio for dNTP substrates for POLG is about an order of magnitude larger than the ratios for reactions with ribonucleoside triphosphates (rNTPs) (Figure 3B). It is noteworthy that POLG is the only enzyme of the mitochondrial ‘salvage’ (DNA replication) pathway whose DNA precursor substrates have expected concentrations that are larger than the enzyme Km values. For our negative example, we considered SLC25A19 (formerly named the deoxynucleotide carrier (DNC), now identified as the thiamine pyrophosphate carrier) [7] to be a suitable choice. In contrast to POLG, the Concentration/Km ratios of DNA precursor substrates of SLC25A19 are low both in the absolute and the relative sense. Dolce et al [13] published data on the Km and Ki values of substrates (we used Ki as a proxy for Km if Km was not reported) that were tested for transport by the SLC25A19 protein, and it is seen in Figures 3B and 4B that DNA precursor substrates (green bars) are not the preferred substrates of this enzyme. Eventually, it was discovered that the function of SLC25A19 had been misinterpreted [7], [47]. When we compare the concentration/Km plots of the mitochondrial nucleotide metabolism (Figures 3A and 4A) to those of POLG and SLC25A19 (Figures 3B and 4B) we observe that the Concentration/Km values of DNA precursors with the enzymes of mitochondrial nucleotide metabolism are at the same level as the Concentration/Km values of the DNA precursors with SLC25A19, even though these DNA precursors are not the physiological substrates of SLC25A19. We note that SLC25A19 was not used a negative example in Figure 2B because the enzyme kinetics values (kcat) were not available for the relevant deoxyribonucleotide or ribonucleotide substrates.
As a side observation, we are intrigued by the fact that at lower dNTP concentrations the Concentration/Km values for rNTPs are essentially equal to those for dNTPs for polymerization by POLG (Figure 4B). This observation reveals that discrimination by POLG in this case is perhaps almost completely dependent on the corresponding reaction Vmax. As the Vmax (or kcat) of rNTPs with POLG are much lower than those for dNTPs, it is possible that in quiescent cells POLG faces more interference by ribonucleotides, thus obstructing the polymerization of deoxyribonucleotides into the DNA molecule being synthesized and at the same time promoting the incorporation of ribonucleoside triphosphates in the DNA strand. This is consistent with the reported incorporation of ribonucleotides in replicating mtDNA [48].
As an initial analysis of the function of the salvage pathway, we used the Michaelis-Menten equation to calculate reaction rates under assumed substrate concentrations. We ignored the effect of inhibitions. This implies that the reaction rates we calculated (number of substrate molecules catalyzed per enzyme molecule per minute) were the upper-bound of the rates at the estimated concentrations, because inhibitions would act to lower these rates. We call such reaction rates ‘effective velocities’. As we assumed deoxyribonucleoside concentrations to be constant at 0.5 µM and deoxyribonucleotide concentrations to be either 10 µM (high, ‘cycling cells’) or 1 µM (low, ‘quiescent cells’), we obtained two sets of effective velocities for the enzymes of mitochondrial nucleotide metabolism – one approximating the behavior in cycling cells, and one approximating the behavior in quiescent cells.
Figure 5A is a plot of the effective velocities of nucleoside kinases versus NT5M. Remember from Figure 1A that NT5M is the nucleotidase that reverses the action of the nucleoside kinases, so the amount of material fed into the salvage pathway depends in part on the balance between these two groups of enzymes. The substrate dCMP is absent for Figure 5A because no reaction was observed between NT5M and dCMP [19]. Note that the nucleoside concentrations were assumed to be constant, so the high and low concentration rates are only given for NT5M. Because of inhibitions and competing reactions, the deoxyribonucleoside output from NT5M would be much lower than represented here, but it is still instructive to compare objectively the disparity between the forward and reverse reactions at the first phosphorylation level. It is clear that the theoretical maximum velocities (at the assumed concentrations) of NT5M reverse reactions are many-fold higher than the maximum velocities from nucleoside kinases. While the situation is poor for the dG and dT substrates, it is extremely poor for the dA substrate where the reverse reaction has well over an order of magnitude advantage over the forward reaction. Furthermore, NT5M may not be the only nucleotidase in the mitochondria, thus exacerbating this issue [19].
In addition to these qualitative comparisons of substrate preferences of mitochondrial nucleotide metabolism enzymes, we analyzed the reaction kinetics further to approximately quantify the flow of substrates through this enzymatic pathway. For simplicity, we ignored the inhibition terms in the Michaelis-Menten equations (inhibitions would further reduce reaction velocities). We could then investigate the effect of kcat, Km, and substrate concentrations on the upper-bound of velocity of the reactions at assumed substrate concentrations and compare the estimated velocities to the expected requirements for completing one round of mtDNA replication in a specified amount of time. It has been reported that one round of mtDNA replication in cell culture takes ∼1–2 hours to complete [45]. To be conservative, we assumed that mtDNA replication takes 2 hours to complete. To replicate 16,568 bases pairs on two mtDNA strands in 2 hours, ∼276 nucleotides are required per minute on average (with the log scale on Figure 5B it is unnecessary to precisely divide this quantity into the specific numbers of dATP, dCTP, dGTP, and dTTP molecules needed for the human mtDNA sequence). Figure 5B shows the effective velocities of some of the enzymes of mitochondrial nucleotide metabolism (DGUOK, TK2, CMPK2 and AK2). These are all the enzymes for which we found data that would enable us to calculate effective velocities. To facilitate comparison across these enzymes, some data for DGUOK and TK2 are repeated in Figure 5B from Figure 5A. As before, nucleoside kinase velocities in Figure 5B are the same for high or low concentration conditions because nucleoside concentrations are assumed to be constant. There exists a many-fold difference in the output of the four dNMPs, with dA nucleosides being fed into the salvage pathway by DGUOK at a rate many orders of magnitude lower than that required to support mtDNA synthesis. Assuming a 2 hour replication duration and an approximately 276/4 nucleotides per minute substrate requirement, the number of molecules of the DGUOK enzyme per mitochondrion required to catalyze the requisite output of dAMP is close to 3000. The poor kinetics of DGUOK with dA is not the only problem with the dA pathway. Although there could be multiple AK isoforms in the mitochondria, some of them are reported to be lacking kinase activity and none of them appear to catalyze dAMP phosphorylation with comparable efficiency to that of AMP phosphorylation [49]. This is verified for AK2 as seen in Figures 2A, 3A, and 4A.
A calculation of dCDP production by CMPK2 at low assumed dCMP concentrations shows that more than 1000 CMPK2 enzymes per mitochondrion would be required to produce the necessary dCMP output per minute (assuming an approximate requirement of 276/4 nucleotides per minute). This result is important considering that CMPK2 expression was undetectable in many tissues [22], thus implying that CMPK2 function may not be essential for the production of mtDNA precursors as has been noted previously [22].
The data on the kinetic parameters of the human mitochondrial nucleoside diphosphate kinase (NME4 in Figure 1) is scarce (Km for dTDP of approximately 1 mM, which is 100 to 1000 times the physiological concentration of dTDP) [17], which is why it is not included in Figure 5B. An NDPK isolated from the pigeon mitochondrial matrix preferred ribonucleoside diphosphates over deoxyribonucleoside diphosphates by several fold [42]. Surprisingly, it appears that both AK2 [11] and NME4 [10] are localized in the mitochondrial intermembrane space, thus suggesting that if their reaction products participate in the mtDNA precursor synthesis, they would then have to be imported into the mitochondrial matrix. Although dAMP is not the preferred substrate for AK2 (Figures 2A, 3A, and 4A), AK2 still has a very fast reaction with dAMP (Figure 5B). Good efficiency with dAMP and the localization of AK2 in the intermembrane space instead of in the mitochondrial matrix seem to contradict each other regarding the role of AK2 in mtDNA precursor synthesis.
To test our conclusions and to build upon them, we used an updated computational model to perform simulations of deoxyribonucleotide dynamics and mtDNA replication within the mitochondrion. Our comprehensive computational model allowed us to investigate the dynamics and origins of mitochondrial dNTPs. Our modeling is based on experimentally measured kinetics and model results enable us to quantitatively track the concentrations as well as the balance of the various deoxynucleosides and deoxynucleotides over time within an individual mitochondrion. Furthermore, the mitochondrial salvage pathway is complex and a systems analysis of this pathway as a whole is an important companion to the study of the individual enzyme kinetics.
Figure 6 shows our simulation results. The X-axis represents the number of molecules of each nucleotide supplied to the mitochondrion in the form of a ‘source’ term in the differential equations in addition to the output from salvage within the mitochondrion. Each value on the X-axis is the sum of molecules supplied of all four species. For example, the X-axis value of 400 means that 100 molecules per minute of each of the four (A, C, G, T) species were supplied. The Y-axis represents the average (over 120 minutes of simulation time) mtDNA replication rate that we observed, calculated as number of nucleotides replicated divided by the time taken to replicate them. Initial values for the substrate concentration in the mitochondrion were randomly varied over a set range as described in the Methods section. Each Y value corresponds to the mean of 100 replication rates from 100 simulations with differing initial substrate conditions. The standard deviations were far smaller than the mean values (and are therefore not shown in Figure 6) indicating that the simulation was not sensitive to the initial substrate conditions. We compared the observed replication rates to those required to complete mtDNA replication in 2 hours (‘cycling cells’) or 10 hours (‘quiescent cells’). For the mtDNA length of 33,136 nucleotides (replicating both strands), these would be 33136 (nucleotides)/120 (minutes) and 33136 (nucleotides)/600 (minutes) respectively or approximately 276 nucleotides per minute and 55 nucleotides per minute respectively. Since the mean observed replication rates with no additional nucleotides supplied (0 on the X-axis) fall below the 2 hour line, it is clear that the output from mitochondrial salvage cannot account for an mtDNA replication duration of 2 hours. In fact, even when a 10 hour replication target was set, mitochondrial salvage alone is an inadequate source of dNTPs, though only a slight amount of additional substrate supplied by transport is needed in this case. Next, we note that both deoxynucleoside as well as deoxynucleoside monophosphate import are insufficient to support a 2 hour replication target. Transport of either dNDPs or dNTPs is sufficient to achieve the target replication rate. The profiles of dNDP and dNTP transport are indistinguishable from one another on Figure 6 because of the extremely fast kinetics of NME4.
Transport of approximately 48 molecules per minute for each of the four nucleotide species was required to complete mtDNA replication in 2 hours. The longer replication time of 10 hours required a transport of 15 dNTP molecules per minute for each of the four nucleotide species. These rates were determined from the simulation by transporting all four nucleotide species at equal rates and with fixed initial concentrations (as described in Methods).We next addressed the question of the minimum transport of each dNTP species necessary to support the target replication. As described in the Methods section, the assumption of equal transport of the four dNTP species was relaxed to find the minimal amount of transport separately for each dNTP species required to meet the mtDNA replication rate goal. To achieve a replication rate of at least 276 nucleotides per minute (‘cycling cells’), 47, 31, 48, and 48 molecules per minute of dTTP, dCTP, dATP, and dGTP were required. Thus, for this condition of relatively fast replication, transport of all four nucleotide species at similar rates is necessary. The total dNTP transport rate sums to 174 nucleotides per minute, a large fraction of the 276 dNTPs per minute consumed by the mtDNA replication. For the slower mtDNA replication with a target of 10 hours, Figure 6 shows that a relatively small amount of transport of dNTP molecules per minute suffices. To achieve the replication rate target of at least 55 nucleotides per minute (representing slow mtDNA replication in ‘quiescent cells’), individual dNTP transport rates of 0, 3, 8, and 15 molecules per minute of dTTP, dCTP, dATP, and dGTP were required. Due to the complexity of the system (a nonlinear one because of feedbacks and inhibitions), slightly different but often practically similar transport profiles were observed to result in similar replication rates. For example, for cycling cells the transport profile of 41, 34, 48, and 41 molecules per minute also achieved the replication rate target. For quiescent cells, the profiles of 2, 2, 8, and 15 and 0, 2, 8, and 15 molecules per minute (practically identical to the transport profile given above) also achieved the replication rate target.
In summary, rapid replication of mtDNA requires a substantial additional source of all four dNTPs (or dNDPs) to supplement the limited kinetics of the mitochondrial salvage pathway. Under the conditions of quiescent cells, the primary requirement is for the transport of dATP and dGTP molecules, and the vast majority of the dNTPs consumed by the mtDNA replication can be provided by the salvage pathway.
Based on this analysis of the enzyme kinetics three properties of the mitochondrial nucleoside salvage pathway are thus apparent:
From the kinetics perspective mitochondrial nucleotide metabolism as defined by this set of enzymes (Figure 1) cannot be expected to be the primary source of dNTP substrates for the rapid replication of mtDNA molecules. Since ribonucleotides exist at higher concentrations than deoxyribonucleotides, enzymes that take both ribonucleotide and deoxyribonucleotide substrates will, in situ, not favor the catalysis of deoxyribonucleotides. This is certainly true for enzymes that possess higher affinities for ribonucleotides, but also for those enzymes that have only slightly better kinetics for deoxyribonucleotides. In these cases ribonucleotide substrates will simply out-compete the deoxyribonucleotides substrates owing to the relative abundance of ribonucleotides.
One plausible interpretation of this analysis is that import of cytoplasmic deoxyribonucleotides is the primary source that supplies the direct precursors for the replication of mitochondrial genome while the mitochondrial salvage pathway acts as a back-up metabolism with a minimal role to play in cycling cells. The occurrence of deoxyribonucleotide transport between the mitochondria and cytoplasm and the substantial contribution of cytoplasm deoxynucleotides towards intra-mitochondrial dNTP pools have been demonstrated [5], [6]. Our results make it possible to comment on why this must be so, due to the kinetic properties of the enzymes of mitochondrial salvage. Our results also enable us to conclude that import of deoxyribonucleotides is in fact essential to support an mtDNA replication time of ∼2 hours. Furthermore, simulations based on these enzyme kinetics indicate that this import occurs either at the dNDP or dNTP level. In cells where cytoplasmic deoxynucleotide concentrations are low, mitochondrial salvage would assume a greater role and, in combination with some other supply such as RRM2B mediated reduction of ribonucleotides in the cytoplasm followed by deoxyribonucleotide transport into the mitochondrion, would produce the dNTPs for both the replication of mitochondrial DNA and perhaps repair of nuclear DNA. Possibly, the dNDPs produced by RRM2B activity might first undergo the terminal phosphorylation by NME4 in the intermembrane space (Figure 1B) and may then be imported into the mitochondrion matrix at the dNTP level to combine with the dNTP pool from intra-mitochondrial salvage. Indeed, this is consistent with defects in the mitochondrial salvage pathway having their most severe phenotype in post-mitotic tissues. That mitochondrial salvage has only a back-up role in supporting mtDNA replication is one explanation why DGUOK and TK2 deficiency phenotypes are tissue-restricted and not systemic.
The kinetic characteristics of the cytoplasmic counterparts of mitochondrial salvage enzymes expose informative parallels and distinctions between the cytoplasmic and mitochondrial pathways of nucleotide metabolism. The good activity of the mitochondrial enzymes (except the nucleoside kinases TK2 and DGUOK) with ribonucleotide substrates implies that these enzymes might play as important a role in ribonucleotide production to support RNA synthesis as they do in supporting DNA synthesis. Based on our analysis, we would argue that future studies of the kinetics of the mitochondrial salvage enzymes would benefit from a broader characterization of the kinetics, particularly the activity of the enzymes with ribonucleotide substrates relative to the activity with deoxyribonucleotide substrates. The majority of the cytoplasmic counterparts also show a preference for non-DNA precursors (such as dUMP) and ribonucleotide substrates [12], [43], [50]. However, the role of nucleoside salvage as a source of dNTPs for nuclear DNA replication is generally assumed to be minimal. In the S-phase of the cell cycle, ribonucleotide reductase irreversibly converts ribonucleoside diphosphates to deoxyribonucleoside diphosphates, and subsequently, deoxyribonucleotides originating from de novo sources proceed to become the predominant precursors to nDNA replication. Thus, ribonucleotide affinities of these cytoplasmic enzymes not only provide the ribonucleoside diphosphates for ribonucleotide reduction but also ensure an adequate supply of RNA substrates. The terminal kinase (NDPK) of the cytoplasmic salvage pathway accepts both ribonucleoside and deoxyribonucleoside diphosphates, and the products can then be appropriately diverted for either RNA or DNA synthesis. Salvage enzymes of thymidine metabolism fit nicely into such a model - examples being the excellent kinetics of TK1 and TK2 with dT, and those of cytoplasmic deoxythymidylate kinase (essential for both salvage and de novo pathways of dTTP synthesis) with dTMP – since thymidine is not an RNA substrate and because of the crucial allosteric control exerted by thymidine nucleotides on ribonucleotide reductase [51] as well as feedback control on mitochondrial TK2 [21]. Such similarities in the enzyme kinetics of the parallel mitochondrial and cytoplasmic metabolisms lead to the question of a ribonucleotide reductase connection to mitochondrial nucleotide metabolism. Such a connection is hinted at by the data supporting a connection between the mitochondrial dNTP pool and the ribonucleotide reductase RRM2B [2], [6]. There has been at least one report of ribonucleotide reductase activity within the mitochondrion [52], though this has never been confirmed as far as we are aware of.
Our simulations show that mitochondrial salvage is inadequate to account for the observed replication time of ∼1–2 hours in cycling cells. It is likely that the deficit is supplied by import from the cytoplasm. We propose that deoxyribonucleotide import into the mitochondria not only does occur, but is in fact essential to replicate and maintain mtDNA in cycling cells. Furthermore, in our simulations, import at the monophosphate level was not able to support mtDNA replication under the constraint of a replication duration of 2 hours or less. Our observation that either dNDP or dNTP transport are able to nearly identically support mtDNA replication is due to the extremely fast kinetics of NME4, the nucleoside diphosphate kinase. The fact that the NME4 kinetics for the conversion of dNDP to dNTP are fast lends weight to the hypothesis that transport occurs mainly at the dNDP level, and not at the dNTP level which would bypass the NME4 activity.
Our results are not necessarily in disagreement with previous reports that observed that supplementation with external dA and dG or dAMP and dGMP rescued mtDNA depletion [3], [28]. In those cases, it was undetermined whether these externally supplied substrates changed their phosphorylation level prior to or after entering the mitochondria, or even the cell. In the study conducted by Saada [28], in patient fibroblasts harboring DGUOK defects while dGTP pools were reduced compared to controls, dATP pools were only moderately affected. In this study when these patient fibroblasts were given external supplementation of both deoxyguanosine and deoxyadenosine, mitochondrial dGTP notably increased, while the increase in mitochondrial dATP was less pronounced. Our observation on the inefficient kinetics of DGUOK with dA is consistent with these findings. We are not aware of any studies on the effects of pyrimidine supplementation in TK2 deficiency.
We have chosen a somewhat arbitrary target replication time of 10 hours for the mtDNA in quiescent cells. It has been reported that even in quiescent cells (rat hepatocytes), mitochondrial DNA is subject to rapid turnover [53]. Moreover, it is plausible to suspect that long replication durations might compromise the integrity of either or both the template and the synthesis strand by increasing the probability of damage to the exposed DNA or unfaithful replication (deletions, frameshifting, etc). Therefore, it is possible that the mtDNA replication time may be practically constrained to a shorter duration than 10 hours. In that case deoxynucleotide import could be essential even in quiescent cells. There is a lack of data on mtDNA replication times in quiescent cells, a critically important gap in our knowledge since quiescent cells are the most severely affected cells in most forms of Mitochondrial DNA Depletion Syndromes (MDS).
The fact that clinical conditions arising from altered intra-mitochondrial dNTP pools mostly manifest in postmitotic tissues is consistent with our results. The possibility of there being more than one deoxynucleotide transporter, say one for purine deoxynucleotides and one for pyrimidine deoxynucleotides, might explain why mutations in TK2 and DGUOK which are both nucleoside kinases produce different phenotypes. It is plausible that there exists more than one mitochondrial deoxynucleotide transporter whose expression levels, possibly in conjunction with other factors, contribute to tissue specificity of mtDNA depletion syndromes. There have been reports [54], [55] asserting a role of PNC1 (pyrimidine nucleotide carrier encoded by solute carrier family 25, member 33 or SLC25A33) in nucleotide import into mitochondria as well as mitochondrial maintenance. PNC1 was able to transport a variety of metabolites, including purine and pyrimidine ribonucleotides and deoxyribonucleotides, with a preference for UTP. Intra-mitochondrial UTP accumulation decreased in response to siRNA-transfection against PNC1. Mitochondrial ADP, ATP, and GTP levels were not significantly altered but the effect on dNTPs was not investigated. Suppression of PNC1 was associated with reduced mtDNA while overexpressed PNC1 was associated with increased mtDNA relative to controls. Since UTP is a cofactor of the mitochondrial helicase (PEO1 or twinkle), mtDNA levels might have been altered through increased or decreased UTP [54]. It is also possible that these consequences resulted from a lack of RNA primers or lack of mtDNA precursors that might be substrates of PNC1. However, PNC1 mRNA was undetectable in skeletal muscle [55], a tissue that is a target of TK2 defects. Interestingly, ribonucleotide reductase overexpression caused mtDNA depletion in skeletal muscle of mice [56]. Also, per mg protein, PNC1 appeared to transport roughly 1.5 times more UTP compared to dTTP, the next most transported substrate. At this time, the role of PNC1 in transporting deoxyribonucleotides for mtDNA synthesis is inconclusive. Import of radioactively labeled dTMP into mitochondria has been observed [57]. However, it was also observed that a fraction of the labeled dTMP was degraded as well as phosphorylated in the growth medium, leading to the possibility that the transport of phosphorylated states other than the monophosphate may have occurred. A transport activity with preference for dCTP has also been observed [58].
It has been proposed that low basal TK2 expression in muscle renders the tissue vulnerable to TK2 defects, while overlapping substrate specificity of cytosolic dCK prevents mtDNA depletion from mutant DGUOK in tissues where dCK expression is high [59]. While mtDNA defects that have a basis in mutated salvage enzymes might conceivably be rescued by other factors such as overlapping substrate specificity of cytoplasmic enzymes, this hypothesis cannot account for phenotypes relating to POLG defects. Importantly, the fact that phenotypes from mutations in POLG are also tissue-specific and not systemic indicate that other factors, such as rates of mtDNA turnover or energetic demand of tissues might also be a factor in the basis of tissue selectivity. In a recent review, Liya Wang discussed deoxynucleoside salvage enzymes and their association with tissue specific phenotypes of mtDNA depletion [60]. It was hypothesized that since mtDNA turnover rates are different in different tissues and also because dNTP pools show organ-specific differences, it would be expected that the regulation of dNTP pools would also be different for different tissues. Because both muscle and liver have high amounts of mtDNA and also of mtDNA turnover, and since the dTTP pool is lowest in muscle and the dGTP pool is smallest in liver, it was proposed that these tissues would be especially vulnerable to mutations in TK2 and DGUOK respectively. Other contributing factors could include limiting RRM2B, thymidylate synthase, or nucleotide transporter activity. In our opinion, it is probable that there is more than one underlying principle that explains tissue specificity – vulnerability of tissues to mutations might be from a combination of various factors such as transcriptional compensation, turnover rates, energetic demand, etc and that different forms of mtDNA depletion syndromes may trace their etiology to different factors.
Based on their experiments with perfused rat heart, Morris et al concluded that in isolated perfused heart, there is no de novo synthesis of dNTPs [61], stressing the importance of TK2 in rat heart. This could indicate that our observations on the inadequacy of mitochondrial salvage enzymes may not hold across all tissue types. It is also possible that the deoxyribonucleotide pools in rat heart arose in part through salvage mediated by residual TK1 activity. In a recent report of a TK2−/− H126N knockin mouse [62], the authors observed TK1 to be the main thymidine kinase component in heart, compared to TK2 in the brain. In this mouse, phenotypic manifestation of TK2 deficiency was related to TK1 down-regulation and transcriptional compensation. Although by postnatal day 13 both brain and heart had suffered substantial mtDNA depletion, in contrast to brain, heart was spared as respiratory chain proteins were still at normal levels in this organ when assayed at postnatal day 13. This could indicate a difference in the importance of TK2 in the heart tissue of rats compared to mice. A recent report claimed that a cytosolic localization of TK2 is present in many rat tissues [63]. For the knockin mouse [62], a compensatory mechanism involving increased mtDNA transcription through suppression of MTERF3 (mitochondrial transcription termination factor 3) expression was implicated in alleviating some of the effects of mtDNA depletion. It was unclear why dTTP but not dCTP levels were affected and whether cytosolic ribonucleotide reduction had any influence in this observation. In humans, even in quiescent patient fibroblasts with only 5–40% of residual TK2 activity, mitochondrial and cytosolic dTTP pools were unaltered [30]. This finding would be consistent with the possibility that the activities of mitochondrial salvage enzymes may not be strictly necessary even for quiescent cells. Alternatively, it is also possible for there to be practical important differences between species with regard to this metabolism. In their study of the rat heart, Morris et al [61] noted that although known as a substrate of TK2, dU was not converted to dUMP possibly due to ENT1 nucleoside transporter not being localized to mitochondria in rodents, unlike humans, suggesting that dU may not be transported into the mitochondria in rodents. It has been noted that genes involved in MDS (mtDNA depletion syndromes) etiology are essential for life in mouse models [60]. However, the severe phenotype of knock-out mice is not identical to the phenotype in humans [64], although multi-organ phenotypes have come to light in humans also [65]. This divergence could perhaps be due to species differences or because of the complete absence of enzyme activity in knock-out models [64].
One limitation of modeling biochemical pathways is that kinetic parameters as reported in the literature and obtained from recombinant enzymes may not reflect the in situ reality, for instance, if the enzyme conformation is unknowingly affected in the in vitro analysis, or if the assay conditions do not represent the cellular environment adequately. Similarly, we have relied on the literature and our judgment for selecting appropriate concentrations and enzyme copies within the mitochondrion. In our analysis we have assumed a nucleoside concentration of 0.5 µM [33]. There have been reports of nucleoside concentrations of approximately 50-fold lower [34]. Such lower concentrations would have two effects on this analysis. First, the problems that we point out concerning the function of the nucleoside kinases TK2 and DGUOK would be even worse with significantly lower nucleoside concentrations. Second, there is a more subtle problem that the enzyme kinetics for TK2 were measured at much higher substrate concentrations (1 µM to more than 100 µM) [21]. If the true substrate concentrations were on the order of 10 nM, then the kinetics would have to be extrapolated to much lower concentrations, which could introduce additional uncertainty in the kinetic constants. Finally, our estimate of time taken to replicate mtDNA (2 hours) comes from a study of mouse cells [45]. It is worth mentioning that POLG kinetics suggest that polymerization itself is capable of proceeding at a rate much faster than 2 hours [14]. A more comprehensive investigation into mtDNA replication durations in a variety of human cells and particularly in the cell types affected by mtDNA depletion syndromes would thus be very beneficial.
For simplicity, we assumed that only one mtDNA molecule is replicating at any given time in a particular mitochondrion. If two or more mtDNA molecules were replicating simultaneously, then the deficit in the required dNTPs would be even larger than our analysis indicates. It should also be noted that mitochondria are very dynamic and undergo continuous fusion and fission. However, the effects of fusion and fission on the mitochondrial dNTP content would most likely average out. While fusion of two mitochondria would result in a larger dNTP pool (measured as number of molecules per organelle), fission would result in a smaller dNTP pool.
Since the known elements of the mitochondrial salvage pathway do not have sufficient enzyme kinetics to support mtDNA replication in the observed duration of ∼1–2 hours, then, an alternative source of mtDNA precursors must be essential. Despite the intensive focus of research on this pathway associated with mitochondrial depletion syndromes, it seems likely that our knowledge of mitochondrial nucleotide metabolism is still incomplete and that this pathway might need to be considerably expanded in the future to include new enzymes, mechanisms, nucleotide transporters and modes of regulation.
|
10.1371/journal.ppat.1002573 | E2F1 Mediated Apoptosis Induced by the DNA Damage Response Is Blocked by EBV Nuclear Antigen 3C in Lymphoblastoid Cells | EBV latent antigen EBNA3C is indispensible for in vitro B-cell immortalization resulting in continuously proliferating lymphoblastoid cell lines (LCLs). EBNA3C was previously shown to target pRb for ubiquitin-proteasome mediated degradation, which facilitates G1 to S transition controlled by the major transcriptional activator E2F1. E2F1 also plays a pivotal role in regulating DNA damage induced apoptosis through both p53-dependent and -independent pathways. In this study, we demonstrate that in response to DNA damage LCLs knocked down for EBNA3C undergo a drastic induction of apoptosis, as a possible consequence of both p53- and E2F1-mediated activities. Importantly, EBNA3C was previously shown to suppress p53-induced apoptosis. Now, we also show that EBNA3C efficiently blocks E2F1-mediated apoptosis, as well as its anti-proliferative effects in a p53-independent manner, in response to DNA damage. The N- and C-terminal domains of EBNA3C form a stable pRb independent complex with the N-terminal DNA-binding region of E2F1 responsible for inducing apoptosis. Mechanistically, we show that EBNA3C represses E2F1 transcriptional activity via blocking its DNA-binding activity at the responsive promoters of p73 and Apaf-1 apoptosis induced genes, and also facilitates E2F1 degradation in an ubiquitin-proteasome dependent fashion. Moreover, in response to DNA damage, E2F1 knockdown LCLs exhibited a significant reduction in apoptosis with higher cell-viability. In the presence of normal mitogenic stimuli the growth rate of LCLs knockdown for E2F1 was markedly impaired; indicating that E2F1 plays a dual role in EBV positive cells and that active engagement of the EBNA3C-E2F1 complex is crucial for inhibition of DNA damage induced E2F1-mediated apoptosis. This study offers novel insights into our current understanding of EBV biology and enhances the potential for development of effective therapies against EBV associated B-cell lymphomas.
| Aberrant cellular proliferation due to deregulation of E2F1 transcriptional activity as a result of either genetic or functional alterations of its upstream components is a hallmark of human cancer. Interestingly, E2F1 can also promote cellular apoptosis regardless of p53 status by activating a number of pro-apoptotic genes in response to DNA damage stimuli. Epstein-Barr virus (EBV) encoded essential latent antigen EBNA3C can suppress p53-mediated apoptotic activities. This study now demonstrates that EBNA3C can further impede E2F1 mediated apoptosis by inhibiting its transcriptional ability, as well as by facilitating its degradation in an ubiquitin-proteasome dependent manner. This is the first evidence, which shows through targeting EBNA3C function linked to the E2F1-mediated apoptotic pathway, an additional therapeutic platform could be implemented against EBV-associated human B-cell lymphomas.
| The role of the pRb-E2F pathway in the regulation of cell-cycle progression, particularly the G1-S transition, is well established [1]. Several lines of evidence have suggested different roles for individual members of the E2F family of proteins in regulating cell proliferation [2], [3]. There are eight different E2F genes (E2F1-8) belonging to this family in mammals and can be sub-grouped into two classes on the basis of their transcriptional activity [3], [4]. E2F1-3, referred to as the ‘activator E2Fs’, bind to pRb and their ectopic expression was shown to be sufficient for driving cells into S-phase [4]. E2F4-8 largely function as transcriptional repressors and are referred to as the ‘repressor E2Fs’ [4]. The repressor E2Fs can be further divided into two subfamilies. E2F4-5 repress gene expression in an Rb family-dependent manner, whereas E2F6-8 exert transcriptional repression through Rb-independent mechanisms [4].
Interestingly, only E2F1 was shown to play a dual role in controlling both cell growth and apoptosis [2], [5], [6]. For example, elevated expression of E2F1 promotes cell-cycle progression by driving quiescent cells into S phase [7], and in cooperation with activated ras, E2F1 can transform rat embryo fibroblast cells [8]. However, E2F1 expression can also induce apoptosis in the absence of proliferative signals [9]. A physiological role for E2F1-mediated apoptosis has been documented in several studies. E2F1−/− knockout mice develop tumors with high incident rate, signifying that E2F1 is also engaged with growth inhibitory and tumor suppressive activities [10], [11]. Moreover, over-expression of E2F1 in mouse embryonic fibroblasts results in cells entering premature S phase and significant apoptosis [6].
E2F1 mediated apoptosis is known to be associated with both p53 dependent and independent mechanisms [4]. E2F1 accelerates p53 mediated apoptotic activity either by inducing the expression of p19/p14ARF, an inhibitor of the Mdm2 ubiquitin ligase that specifically targets p53 for ubiquitin-proteasome mediated degradation or by enhancing p53 phosphorylation [2], [4]. Moreover, E2F1 can also induce apoptosis by transactivating the p53 homologue p73 and Apaf-1 (apoptosis activating factor-1) in response to DNA damage signals [2], [4], [12], [13], [14].
The signaling events that lead to E2F1 induction upon DNA damage response have also been documented [9], [15], [16], [17]. In response to DNA damage, unlike other members of the E2F family, E2F1 is uniquely induced by both ATM (ataxia telangiectasia mutated) and ATR (ATM and Rad3-related) through specific phosphorylation at serine 31 [17]. E2F1 is also shown to be phosphorylated by Chk2 [16]. In general, these phosphorylation events lead to stabilization and activation of E2F1 [16]. In addition to phosphorylation, both acetylation as well as ubiquitination have also been recognized to play an important role in activation and stabilization of E2F1 in response to DNA damage [9], [15], [18]. Thus, it appears that several DNA damage signaling pathways are actively engaged with the induction of E2F1 mediated apoptosis.
EBV is a lymphotropic γ-herpesvirus that asymptomatically persists in more than 90% of the world population [19], [20]. However, EBV intermittently causes a self-limiting disease, infectious mononucleosis in adolescents and has been shown to be associated with the development of several B-cell lymphomas and epithelial cancers primarily in immuno-compromized individuals [19], [21]. In vitro, EBV can efficiently transform quiescent B-cells into continuously proliferating lymphoblastoid cell lines (LCLs), providing a surrogate model for EBV associated B-cell tumorigenesis [19], [21], [22]. These latently infected LCLs carry the viral genome as extra-chromosomal episomes that express only a small subset of genes including six nuclear antigens (EBNA- 1, 2, 3A, 3B, 3C and LP), three membrane associated proteins (LMP- 1, 2A, and 2B) and several non-coding RNAs [19], [21]. Genetic studies using recombinant viruses from a number of different groups have established that EBNA1, EBNA2, EBNA3A, EBNA3C and LMP1 are important for EBV mediated transformation of naïve B-cells in vitro [19], [21], [23], [24], [25]. Interestingly, EBNA-LP is not absolutely required for in vitro B-cell transformation, but necessary for efficient long-term growth of transformed B-cell lines [26].
EBNA3C was initially identified as a transcriptional modulator that can efficiently regulate the transcription of both viral and cellular genes [27], [28], [29]. Coupled with RBP-Jκ, EBNA3C mediated regulation of Notch-induced transcription was shown to be one of the major signaling pathways important for LCL propagation [30], [31], [32]. In addition, EBNA3C was also shown to interact with a wide range of transcription factors and modifiers, such as c-Myc [33], SUMO1 [34], SUMO3 [34], HDAC1 [35], CtBP [36], DP103 [37], p300 [38], prothymosin-α [38], Nm23-H1 [39], p53 [40] and its regulatory proteins Mdm2 [41], ING4 [42] and ING5 [42]. Recently, we showed that EBNA3C can repress p53 dependent apoptotic activity by either blocking its transcriptional activity or recruiting Mdm2 activity for ubiquitin-proteasome mediated degradation [40], [41]. Moreover, EBNA3C attenuates the p53 function through blocking the interaction between p53 and its regulatory proteins, inhibitor of growth family proteins ING4 and ING5 [42].
In this study we address the influence of EBNA3C on suppressing E2F1 mediated apoptosis, independent of p53 in EBV transformed LCLs. We find that EBNA3C can prevent cells from entering E2F1-dependent apoptosis both at early and latent stage of infection and possibly that this effect is critically dependent on the specific interaction between the DNA binding domain of E2F1 and EBNA3C. Regulation of E2F1 mediated apoptosis correlates with EBNA3C-dependent inhibition on E2F1 transcriptional activity at apoptosis related genes, including p73 and Apaf-1. Moreover, EBNA3C specifically targets E2F1 for an ubiquitin-proteasome mediated degradation. Most importantly, we show that E2F1 plays a dual role in regulating LCLs outgrowth. In the presence of growth factors, E2F1 promotes cell proliferation while DNA damage signals can trigger E2F1 mediated apoptosis and cell death. Our data define a new interplay between EBNA3C and E2F1 mediated apoptosis that occurs independently of p53. Overall, this study supports a model where EBNA3C can antagonize the apoptotic properties of both E2F1 and p53 to maintain EBV transformed cells in a continuous state of growth stimulation.
HEK 293, HEK 293T, and both p53 and pRb null Saos-2 cells were maintained as described previously [41], [43]. Burkitt's lymphoma cell lines DG75, Ramos, BJAB, BJAB stably expressing EBNA3C clones (E3C7 and E3C10) and the in vitro EBV-transformed lymphoblastoid cell lines (LCL1 and LCL2) have been previously described [40], [41], [43].
pEGFP-EBNA3C expressing GFP-fused wild-type EBNA3C (residues 1–992) and myc-tagged EBNA3C constructs (expressing amino acids 1–992, 1–365, 366–620, 621–992, 1–159, 1–129 and 1–100) in pA3M vector have been previously mentioned [41], [43]. Other myc-tagged truncated EBNA3C constructs (encoding amino acids 1–300, 1–250, 50–300, 130–300, 160–300, 200–300, 621–950, 621–850, 621–800, 621–750 and 700–900) were generated by PCR amplification followed by directional cloning in pA3M vector at EcoRI and NotI restriction sites. pGEX-E2F1 plasmid expressing GST-fused wild-type E2F1 (encoding residues 1–437) was kindly provided by Pradip Raychaudhuri (University of Illinois, Chicago, IL, USA) and used to generate wild-type and different truncated versions of E2F1 (expressing residues 1–437, 1–400, 1–310, 1–243, 243–437 and 1–150) fused with either C-terminal flag-epitope or N-terminal GST-tag into pA3F [33], and modified pGEX-2TK vectors [43], respectively at EcoRI and NotI restriction sites. E2F1 reporter plasmids (pGL2-basic-3X-WT-E2F1-Luc and pGL2-basic-3X-Mut-E2F1-luc) containing either three wild-type E2F1 binding sites (CTGCAATTTCGCGCCAAACTT) or three mutant E2F1 binding sites (CTGCAATTGCTCGACCAACTT) fused upstream of the luciferase gene were generously provided by Stefan Gaubatz (Philipps University, Marburg, Germany) [44]. Human wild-type p73 [45] and Apaf-1 [46] promoters linked to luciferase gene were obtained as kind gifts from Mirko Marabese (Istituto di Ricerche Farmacologiche “Mario Negri”, Milan, Italy) and Kristian Helin (European Institute of Oncology, Milan, Italy), respectively. Lentiviral packaging vectors, sh-RNA expressing lentiviral vectors directed against either EBNA3C (pGIPZ-Sh-E3C.1) or control that lacks any complementary sequence in the human genome (pGIPZ-Sh-Con) were previously described [43]. Sh-RNA directed against E2F1 was cloned into pGIPZ vector (Open Biosystems, Inc. Huntsville, AL). The sense strand of E2F1 sh-RNA sequence #1 is 5′-tcgagtgctgttgacagtgagcgaGACTGTGACTTTGGGGACCTtagtgaagccacagatgtaAGGTCCCCAAAGTCACAGTCgtgcctactgcctcggaa-3′ [47]. The sense strand of another E2F1 sh-RNA sequence #2 is 5′-tcgagtgctgttgacagtgagcgaGACTGTGACTTTGGGGACCTtagtgaagccacagatgtaAGGTCCCCAAAGTCACAGTCgtgcctactgcctcggaa-3′ [47]. Upper-case letters indicate 20-nucleotide (nt) E2F1 target sequence and lowercase letters indicate hairpin and sequences necessary for the directional cloning into pGIPZ at Xho I and Mlu I restriction sites. All constructs and mutations were verified by DNA sequencing (University of Pennsylvania DNA sequencing facility).
Rabbit polyclonal antibodies reactive to E2F1 (C-20), Apaf-1 (H-324), Cyclin E (M-20) Ubiquitin (FL-76); goat polyclonal antibody against p73 (S-20); and mouse monoclonal antibodies against PARP1 (F-2) and GFP (F56-BA1) were obtained from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). Mouse monoclonal antibody against GAPDH was bought from US-Biological Corp. (Swampscott, MA). Mouse monoclonal antibodies reactive to myc epitope (9E10), flag epitope (M2), EBNA3C (A10), LMP1 (S12) and EBNA2 (PE 2) have been described previously [40], [41], [43], [48], [49]. Rabbit polyclonal antibody specific for EBNA3C was obtained from Cocalico Biologicals, Inc. (Reamstown, PA) and has been described previously [35].
Adherent cells were transiently transfected either by electroporation with a Bio-Rad Gene Pulser II electroporator as previously described [40], [41], [43], or using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) according to manufacturer's protocol. LCLs were transfected with 50 ìg of plasmids via electroporation (Bio-Rad Gene Pulser II; 230 V, 975 µF). Transfected LCLs were cultured in RPMI medium with 10% FBS for 48 h.
10 million transiently transfected cells or 50 million B-cells were harvested, washed with ice cold PBS and subsequently lysed in 0.5 ml ice cold RIPA buffer [1% Nonidet P-40 (NP-40), 10 mM Tris pH 8.0, 2 mM EDTA, 150 mM NaCl, supplemented with protease inhibitors (1 mM phenylmethylsulphonyl fluoride (PMSF), 1 µg/ml each aprotinin, pepstatin and leupeptin]. Lysates were precleared with normal control serum plus 30 µL of a 1∶1 mixture of Protein-A/G Sepharose beads (GE Healthcare Biosciences, Pittsburgh, PA) for 1 h at 4°C. Unless and otherwise stated, approximately 5% of the precleared lysate was saved for input control and the protein of interest was captured by rotating the remaining lysate with 1 µg of specific antibody overnight at 4°C. Immuno-complexes were captured with 30 µl Protein-A/G beads, pelleted and washed 5X with ice cold RIPA buffer.
Input lysates and IP complexes were boiled in laemmli buffer [50], fractionated by SDS-PAGE and transferred to a 0.45 µm nitrocellulose membrane for WB analyses. The membranes were then probed with specific antibodies followed by incubation with appropriate infrared-tagged secondary antibodies and viewed on an Odyssey imager. Image analysis and quantification measurements were performed using the Odyssey Infrared Imaging System application software (LiCor Inc., Lincoln, NE).
IF experiments were performed essentially as described previously [41], [43]. Briefly, Saos-2 cells plated on coverslips were transfected with expression vectors as indicated, using Lipofectamine 2000. After 36 h of transfection, cells were fixed by ice cold acetone: methanol mixture (1∶1) for 10 min at −20°C. LCLs were air-dried and fixed similarly. Ectopically expressed E2F1 was detected using M2-antibody, and GFP-EBNA3C was detected by GFP fluorescence. In LCLs, endogenously expressed EBNA3C and E2F1 proteins were detected using their specific antibodies. The slides were examined with a Fluoview FV300 confocal microscope (Olympus Inc., Melville, NY).
Escherichia coli BL21 competent cells were transformed with plasmids for each Glutathione S-transferase (GST) fusion protein and protein complexes containing the tagged proteins were purified essentially as previously described [40], [41], [43].
For in vitro GST-pulldown experiments, GST fusion proteins were incubated with in vitro-translated 35S-labeled protein in binding buffer (1× phosphate-buffered saline [PBS], 0.1% NP-40, 0.5 mM dithiothreitol [DTT], 10% glycerol, supplemented with protease inhibitors). In vitro translation was done with the TNT T7 Quick Coupled Transcription/Translation System (Promega Inc., Madison, WI) according to the manufacturer's instruction.
Promoter assays were performed as previously described with few modifications [40], [42]. Briefly, either 10 million Saos-2 (pRb−/−) cells were transiently transfected by electroporation with indicated plasmids. Cells were additionally transfected with pEGFP-C1 and pCMV-βgal constructs for measuring the transfection efficiency. After 36 h of transfection, cells were harvested, lysed in reporter lysis buffer (Promega Inc., Madison, WI) and the luciferase as well as β-galactosidase activities were measured using either an LMaxII384 luminometer (Molecular Devices, Sunnyvale, CA) or VERSAmax microplate reader (Molecular Devices, Sunnyvale, CA), respectively. The results are shown as representation of duplicate experiments.
Chromatin immunoprecipitation (ChIP) assay was performed as previously described [51]. Briefly, 20 million HEK293 cells were transiently transfected by electroporation with E2F1 reporter plasmid and expression vectors for flag-E2F1 and myc-EBNA3C. 36 h of post-transfection, cells were cross-linked by 1% formaldehyde, harvested, sheared DNA to an average length of 700 bp by sonication. Cross-linked DNA was immunoprecipitated by anti-flag antibody and subjected for PCR analysis using primers designed either for E2F1-responsive promoter fused with luciferase gene or control SV-40 promoter region of pGL2-basic vector. Primers for E2F1-promoter: 5′-TTGCCGATTTCGGCCTATTG-3′ and 5′-CATCCTCTAGAGGATAGAATGG-3′; primers for SV-40 promoter: 5′-CGTTGTTGTTTTGGAGCACGGA-3′ and 5′-TTGGACTTTCCGCCCTTCTTG-3′. For EBV-positive cells (LCL1 with either Sh-Con or Sh-E3C), 50 million cells were collected, immunoprecipitated with either control rabbit antibody or E2F1 specific antibody and processed as above. Eluted DNA fragments were analyzed by subsequent PCR with primers specific for the p73 and Apaf-1 human promoters. Primers used in this assay are: for p73 promoter: 5′- TGAGCCATGAAGATGTGCGAG-3′ and 5′- GCTGCTTATGGTCTGATGCTTATG-3′ [52]; for Apaf-1 promoter: 5′-GCCCCGACTTCTTCCGGCTCTTCA-3′ and 5′-GGAGCTGGCAGCTGAAAGACTC-3′ [53].
10 million Saos-2 (p53−/−) cells were transfected by electroporation with indicated expression plasmids for flag-E2F1 and myc-EBNA3C. Cells were additionally transfected with a GFP expression vector (pEGFP-C1, BD Biosciences Clontech). After 24 h of transfection the cells were serum starved (DMEM with 0.1% FBS) with 5 µM etoposide (MP Biomedicals) for 12 h, followed by selection with DMEM supplemented with 5 mg/ml G418 (Invitrogen, Inc., Carlsbad, CA) for 2 weeks. After selection, cells were fixed on the plates with 4% formaldehyde and scanned for GFP expressed colonies using Typhoon 9410 imaging system (GE Healthcare Biosciences, Pittsburgh, PA). The area of the colonies (pixels) in each dish was calculated by Image J software (Adobe, San Jose, CA). The data are shown as the average and standard deviation of two independent experiments.
Saos-2 cells were transfected and selected as above. Approximately 2 million puromycin selected LCLs, EBV positive LCLs (LCL1 and LCL2) or EBV negative Burkitt's lymphoma cell lines DG75 and Ramos were exposed to serum starvation (RPMI with 0.1% FBS) and 5 µM etoposide treatment for 12 h. For E2F1 knockdown LCLs, cells were additionally treated with an increasing concentration of etoposide (0, 5, 10 and 20 µM) for 12 h in absence of serum. Subsequently cells were collected, fixed in cold 70% ethanol for 2 h at −20°C, washed with 1× PBS and stained with PI staining buffer (10 mM Tris, pH 7.5; 0.2 mg/ml RNase A, and 50 mg/ml propidium iodide) for 2 h in the dark at room temperature. The stained cells were analyzed using FACScan (BD Biosciences, San Jose, CA) and FlowJo (Tree Star, Inc. Ashland, OR) software.
Approximately 1×105 Saos-2 cells selected for either flag-E2F1 or flag-E2F1 plus myc-tagged EBNA3C constructs (either wild-type, residues 1–992 or truncated version, residues 366–620) were plated into each well of the 6-well plates, exposed to serum starvation (DMEM with 0.1% FBS) and 5 µM etoposide treatment for 12 h, followed by culturing in regular medium for 6 days at 37°C. Viable cells from each well were counted by trypan blue exclusion method daily using a Bio-Rad TC10 Automated cell counter. In parallel assays cells were harvested, lysed in RIPA buffer and subjected for western blot analyses using indicated antibodies.
For B-cells, approximately 1×105 cells (Ramos, DG75, LCL1 and LCL2) were plated into each well of the 6-well plates and cultured at 37°C in either complete RPMI medium or RPMI supplemented with 0.1% FBS plus 5 µM etoposide. Cells were counted similarly for 6 days. Both experiments were performed in duplicate and were repeated two times.
The TUNEL assay was performed by using an In Situ Cell Death Fluorescein Detection kit (Roche, Indianapolis, IN) in accordance with the manufacturer's instructions. Saos-2 cells transfected with the indicated expression plasmids for flag-E2F1 and myc-EBNA3C were selected for 2 weeks with G418. After 12 h treatment with 0.1% FBS containing DMEM plus 5 µM etoposide for 12 h, terminal transferase reaction was performed on 1∶1 methanol∶acetone fixed cells in 6 well plates. Apoptosis was measured by counting green cells using a Fluoview FV300 confocal microscope (Olympus Inc., Melville, NY).
Lentivirus production and transduction of EBV-transformed B-cells (LCLs) were essentially carried out as previously described [43].
Generation of BAC GFP-EBV was previously described [22]. EBNA3C mutant (BAC GFP-EBVΔE3C) was generated from wild-type BAC GFP-EBV construct. For the generation of BAC GFP-EBVΔ E3C we selected the region from 91822 to 102891 bp from wild-type BAC GFP-EBV plasmid and deleted the corresponding EBNA3C region from 98370–101424 bp. Southern blot analysis and junction PCR was performed to confirm the mutant generation.
For infection, peripheral blood mononuclear cells (PBMC) from healthy donors were obtained from University of Pennsylvania Immunology Core. As previously described [22], [43], approximately 10 million PBMC were mixed with virus (either wild-type or mutant) supernatant in 1 ml of RPMI 1640 with 10% FBS for 4 hr at 37°C in 6-well plates. Cells were centrifuged for 5 min at 500 g, discarded the supernatant, pelleted cells and resuspended in 2 ml of complete RPMI 1640 medium in 6 well plates. EBV GFP expression visualized by fluorescence microscopy was used to quantify infection. The protein and mRNA level of the infected cells was detected after indicated days of post-infection.
Total RNA was isolated by using TRIzol reagent according to the instructions of the manufacturer (Invitrogen, Inc., Carlsbad, CA). cDNA was made by using a Superscript II reverse transcriptase kit (Invitrogen, Inc., Carlsbad, CA) according to the instructions of the manufacturer. The primers were for E2F1, 5′- GGCCAGGTACTGATGGTCA-3′, and 5′-GACCCTGACCTGCTGCTCT-3′, for p73 5′-CCCCATCAGGGGAGGTG-3′, and 5′-AGGGGACGCAGCGAAAC-3′, for Apaf-1 5′- CCTCTCATTTGCTGATGTCG-3′ and 5′-TCACTGCAGATTTTCACCAGA-3′, for cyclin E 5′-GTTATAAGGGAGACGGGGAG-3′ and 5′-TGCTCTGCTTCTTACCGCTC-3′, and for GAPDH 5′-TGCACCACCAACTGCTTAG-3′ and 5′-GATGCAGGGATGATGTTC-3′. Quantitative real-time PCR analysis was done using StepOnePlus Real-Time PCR System (Applied Biosystems, Foster City, CA) in triplicate as previously described [43].
100 µg of cell extracts from Saos-2 cells transfected with flag-E2F1 with or without myc-EBNA3C expression vector were incubated with 200 ng of the indicated biotinylated oligonucleotides (wild-type or mutant) immobilized with streptavidin accordingly to the manufacturer protocol, in the absence or presence of a 200 molar excess of the corresponding non biotinylated oligonucleotide. Oligonucleotide-bound E2F1 protein was washed 3X with RIPA buffer and detected by western blotting using anti-flag antibody. The band intensities were scanned using KODAK 1D Image Analysis software. The oligonucleotides [9] used in this assay are: for p73 wild-type promoter 5′-GCCGCCTTTTGGCGCGCGTCGCTCCTGCAGAG-3′; for p73 mutant promoter 5′-GCCGCCTTGTAGAGTGCGTCGCTCCTGCAGAG-3′; for Apaf-1 wild-type promoter 5′-AGTCAAATCCCGCCGGATCCACCCAGCCCGGA-3′; for Apaf-1 mutant promoter 5′-AGTCAAATTCAGTCAGATCCACCCAGCCCGGA-3′.
10 million Saos-2 (pRb−/−) cells were transiently transfected using electroporation with flag-tagged E2F1 with or without myc-tagged EBNA3C expression plasmids. Cells were additionally trasfected with GFP-expressing plasmid for measuring the trasfection efficiency. After 36 hours transfection, cells were treated with 40 µg/ml cyclohexamide (CalBiochem, Gibbstown, NJ). For LCLs, 20 million cells were treated with 100 µg/ml cyclohexamide (CalBiochem, Gibbstown, NJ) either in normal serum medium or in 0.1% FBS containing DMEM plus 5 µM etoposide. Subequently, lysates were prepared at indicated time periods and subjected to immunoblot analyses with appropriate antibodies. Band intensities were quantitated using Odyssey 3.0 software provided by Odyssey imager (LiCor Inc., Lincoln, NE).
15 million HEK 293 cells were transfected by electroporation with appropriate plasmids expressing HA-Ub, flag-E2F1 and myc-EBNA3C. Cells were incubated for 36 h and pretreated for an additional 12 h with 20 µM MG132 (Enzo Life Sciences, Inc. Farmingdale, NY) before harvesting. Flag-E2F1 was immunoprecipitated with M2 antibody and resolved by SDS-PAGE. The extent of ubiquitination of flag-tagged proteins was determined by western blot analysis using the anti HA-antibody (12CA5). For LCLs (Sh-Control and Sh-EBNA3C), cells were treated with 40 µM MG132 (Enzo Life Sciences, Inc. Farmingdale, NY) and immunoprecipitated with anti-E2F1 polyclonal antibody and subjected for western blot with indicated antibodies.
Several lines of evidence suggest that EBNA3C manipulates G1 cell-cycle restriction point through disruption of Cyclin/CDK-pRb-E2F pathway in EBV infected human cells [43], [54], [55], [56], [57]. For example, EBNA3C directly targets pRb, but not other pocket family proteins including p107 and p130, for ubiquitin-proteasome mediated degradation [54], relieving the negative regulatory pressure on E2F transcriptional factors to facilitate the G1 to S transition [43], [57]. It is therefore tempting to investigate whether or not EBNA3C has any influence in modulating functions of E2F1, the major transcriptional factor in E2F family and whose active participation in both cell-proliferation and apoptosis regulation is unquestionable.
First, we determined whether EBNA3C can form a complex with E2F1 in EBV infected human B-cells. Endogenously expressed EBNA3C was immunoprecipitated from two EBV-transformed lymphoblastoid cell lines (LCL1 and LCL2) or a Burkitt lymphoma (BL) cell line -BJAB stably expressing EBNA3C (E3C7 and E3C10) using an EBNA3C reactive rabbit polyclonal antibody, and co-immunoprecipitation (co-IP) of E2F1 was monitored by immunoblotting using an E2F1 specific antibody (Figure 1A and 1B, respectively). EBV-negative BL lines DG75 and BJAB were used as controls (Figure 1A and 1B, respectively). The results clearly demonstrated that EBNA3C formed a stable complex with E2F1 in human cells (Figure 1A and 1B). Virtually identical results were obtained when we used a different EBNA3C reactive mouse monoclonal antibody (A10) for co-immunoprecipitation experiments using these cell lines (Figure S1A and S1B). However, since all these B-cells have endogenous pRb expression, we could not rule out the possibility that pRb could serve as a bridging molecule between EBNA3C and E2F1 binding interface.
In order to validate whether this interaction between EBNA3C and E2F1 is either pRb dependent or independent, we next performed binding experiments using two different strategies. We utilized a mutant E2F1 construct (expressing residues 1–400) lacking the pRb interaction domain at the C-terminal region and a pRb-deficient cell line, Saos-2 (pRb−/−). Both HEK 293 (pRb+/+) and Saos-2 (pRb−/−) cells transiently expressing myc-EBNA3C in the presence of either empty vector, or flag-tagged wild-type E2F1 (residues 1–437) or flag-tagged E2F1 mutant (residues 1–400) constructs were harvested after 36 h of transfection and subsequently subjected for IP with anti-flag antibody (Figure 1C and 1D). HEK 293 (pRb+/+) cells stably express adenovirus E1A; however, the endogenous pRb-E2F1 complex was shown to be resistant to E1A-mediated disruption [58]. The results showed that EBNA3C was clearly co-immunoprecipitated with both wild-type and mutant E2F1 but not with the vector control in both pRb+/+ and pRb−/− cell lines (Figure 1C and 1D, respectively). Importantly, the mutant E2F1 (residues 1–400) bound to EBNA3C with stronger association compared to wild-type protein (Figure 1C and 1D, compare lanes 2 and 3). Perhaps, deletion of the pRb binding region at the C-terminal domain of E2F1 leads to a conformational change of overall E2F1 secondary and tertiary structure, which further allows greater access to EBNA3C's binding site(s). Analysis of both endogenous as well as ectopic expression data strongly demonstrated that EBNA3C forms a pRb-independent complex with E2F1.
For additional support of the binding data and to visualize the sub-cellular localization pattern of E2F1 in the presence of EBNA3C, colocalization experiments were performed with an EBV-transformed cell line, LCL2 (pRb+/+) (Figure 1E). Immunofluorescence staining using antibodies specific to E2F1 and EBNA3C proteins demonstrated that both proteins had distinctive nuclear staining with a speckled pattern (Figure 1E). The results showed that E2F1 partly colocalized with EBNA3C in human cells, as visualized by yellow fluorescence when both signals were merged (Figure 1E). The colocalization study was further extended using ectopically expressed flag-tagged E2F1 with the GFP-tagged EBNA3C in Saos-2 (pRb−/−) cells. The results showed that E2F1 noticeably colocalized with EBNA3C, as indicated by the merged yellow fluorescence signals within the nucleus (Figure 1F). Further, these colocalization data suggested that EBNA3C shares similar nuclear compartments with E2F1 in a pRb independent manner.
We wanted to determine the functional residues of EBNA3C that specifically interact with E2F1. Two successive binding experiments were performed with different truncated polypeptides of EBNA3C covering the entire length of the molecule (residues 1–992, 1–365, 366–620 and 621–992). First, HEK 293 cells were transfected with flag-tagged E2F1 (residues 1–400) in combination with either the control vector or aforementioned myc-EBNA3C expression constructs. The results showed that E2F1 (residues 1–400) was co-immunoprecipitated with both EBNA3C N-terminal (residues 1–365) and the C-terminal (residues 621–992) domains along with the full-length protein (residues 1–992) (Figure 2A). No co-immunoprecipitation was observed with either the vector control or EBNA3C middle region (residues 366–620) indicating a strong level of specificity of this experiment (Figure 2A).
Next, in order to support the abovementioned binding study, an in vitro GST pulldown experiment was conducted using similar EBNA3C fragments. Bacterially purified recombinant GST and GST-E2F1 proteins were incubated with different in vitro-translated, S35−Met radiolabeled fragments of EBNA3C (residues 1–365, 366–620 and 621–992) using T7 TNT coupled transcription-translation system [41], [59]. Interestingly, in contrast to the immunoprecipitation assay (Figure 2A) the in vitro binding assay (Figure 2B) showed a different trend. The results showed that only the N-terminal domain of EBNA3C (residues 1–365) along with the full-length molecule strongly bound with GST-E2F1 (Figure 2B). However, the remaining EBNA3C domains starting from 366 to 992, including the GST control, showed little or no binding (Figure 2B). A parallel Coomassie blue-stained gel showed the amounts of recombinant GST proteins employed in this assay (Figure 2B). The results implicated that the N-terminal residues of EBNA3C is likely to interact directly with E2F1 while the C-terminal region associates with a complex which includes E2F1 in cells (Figure 2A and 2B). However, we could not entirely rule out the possibility that the N-terminal domain of EBNA3C may also form a complex with E2F1 utilizing a protein present in the rabbit reticulocyte lysate as a potential bridging factor.
Earlier studies have shown that both the N- and C-terminal domains of EBNA3C are important to interact with a number of critical cell cycle-regulatory molecules [33], [38], [43], [56], [60], [61]. To narrow down the interacting domain within the N- as well as C-terminal regions of EBNA3C, two successive binding assays were performed using a series of truncated EBNA3C fragments. First, an in vivo immunoprecipitation experiment was set up by co-transfecting a flag-tagged E2F1 (residues 1–400) expressing construct with either empty vector or plasmid DNA expressing myc-tagged C-terminal truncated fragments of EBNA3C in HEK 293 cells. The results from immunoprecipitation using anti-flag antibody showed that the C-terminal residues starting from 621 to 700 of EBNA3C is capable of associating with E2F1 (residues 1–400). EBNA3C residues 621–992, 621–950, 621–850, 621–800 and 621–750 were co-immunoprecipitated with flag-E2F1 (residues 1–400), while no co-immunoprecipitation was observed with either the vector control or EBNA3C residues 700–900 (Figure 2C).
Further, to map within the N-terminal domain of EBNA3C (residues 1–365) an in vitro GST-pull down experiment was conducted using a series of small truncations of N-terminal EBNA3C (Figure 2D). In vitro precipitation experiments with recombinant GST-E2F1 showed strong interaction with EBNA3C residues 1–365, 1–300, 1–250, 1–159, 1–129, 50–300, 130–300 and 160–300 (Figure 2D) but not with EBNA3C residues 1–100 and 200–300 (Figure 2D). All fragments of EBNA3C failed to interact with the GST control, strongly suggesting that the observed binding was specific for E2F1 (Figure 2D). The results demonstrated that E2F1 interacts with two distinct regions of EBNA3C, one at N-terminal residues 100–200 and another at residues 621–700 (Figure 2E).
E2F1 consists of many domains, including the N-terminal Cyclin A binding domain, DNA binding domain (DBD), dimerization domain, and a C-terminal transactivation domain [2], [62]. The pRb binding region is located at the transactivation domain within residues 400–437 [2]. To identify the domains of E2F1 that are required for binding to EBNA3C, an in vivo binding assay was performed by immunoprecipitating flag-tagged E2F1 expression constructs encoding various domains of E2F1 with myc-tagged EBNA3C protein. The co-immunoprecipitation of EBNA3C was detected using anti-myc antibody. The data presented in Figure 3A showed that the E2F1 residues 1–437, 1–400, 1–310 and 1–243 strongly bound to EBNA3C, whereas the C-terminal region of E2F1 residues 243–437 did not bind to EBNA3C. In order to corroborate this binding experiment a subsequent GST-pull down experiment was carried out by incubating in vitro translated radio-labeled EBNA3C with different GST-fused E2F1 proteins, including residues 1–437, 1–310, 1–243, 1–150 and 243–437 (Figure 3B). The results indicated that the EBNA3C binding region within E2F1 lies predominantly within the N-terminal amino acids 1–243 of E2F1 containing Cyclin A and DNA binding domains (Figure 3C).
The binding studies led us to investigate whether EBNA3C was capable of modulating E2F1 mediated transcriptional activity. To specifically test the transcriptional activity of E2F1 we used two different reporter plasmids containing three copies of either wild-type (3X-WT-E2F1-luc) or mutant (3X-Mut-E2F1-luc) E2F1 responsive sequence element cloned upstream of the luciferase gene (Figure 4A). HEK 293 (pRb+/+) cells were subsequently transfected with these reporter plasmids in the presence of either vector control or flag-tagged E2F1 with or without myc-tagged EBNA3C (Figure 4B). Cells were additionally trasfected with both GFP expression and β-galactosidase reporter constructs under CMV promoter in order to check transfection efficiency. The results clearly demonstrated that ectopic expression of E2F1 leads to transcription activation from the wild-type promoter, while the mutant promoter had no response (Figure 4B). Interestingly, when co-expressed with EBNA3C, the transcriptional activity of E2F1 was significantly reduced to more than 70% (Figure 4B). EBNA3C expression alone also showed a reduction in luciferace activity perhaps preventing endogenous E2F1 activity from accessing the wild-type promoter region, whereas it showed no activity on the mutant promoter, indicating the specificity of EBNA3C mediated E2F1 transcriptional repression (Figure 4B). β-Galactosidase activity was measured to evaluate equal transfection efficiency (Figure 4B). The expression levels of transiently expressed EBNA3C and E2F1 along with GAPDH as an internal loading control and GFP expression as a transfection efficiency control were analyzed by western blots (Figure 4B, bottom panels). The results showed that co-expression of EBNA3C led to a reproducible reduction in E2F1 expression level (Figure 4B), indicating that EBNA3C mediated repression of E2F1 transcriptional activity may also mediate through targeting E2F1 degradation.
Previous studies have indicated that pRb blocks E2F1 transcriptional activity by forming a stable complex with E2F1 [63]. In order to distinguish our results from the pRb effect, we conducted a similar experiment in a pRb null background using Saos-2 cells (pRb−/−) (Figure 4C). The reduction of E2F1 transactivation activity by EBNA3C was not pRb dependent, as the normalized luciferase activity presented in Figure 4C showed that in the presence of EBNA3C there was a reduction of E2F1 transcriptional activity by greater than 50% when compared to that of E2F1 alone (Figure 4C). Interestingly, EBNA3C co-operated with pRb to inhibit E2F1 transcriptional activity (Figure 4C). Moreover, increasing amounts of EBNA3C resulted in a dose-dependent inhibition of E2F1 transactivation in Saos-2 (pRb−/−) cell line (Figure 4D). Similarly, increasing amount of EBNA3C expression caused a gradual decrease in E2F1 expression levels without affecting GFP expression levels (Figure 4D, bottom panels). This further indicates that EBNA3C may regulate E2F1 stability besides affecting its transactivition process.
To further define the domain(s) of EBNA3C important for this activity and also to determine if the binding domain(s) of EBNA3C is essential for inhibition of E2F1-mediated transactivation, the reporter assays were extended using different truncated domains of EBNA3C (residues 1–365, 366–620 and 621–992). All truncated EBNA3C mutants were able to localize in nucleus since the wild-type EBNA3C contains three functional nuclear localization signals (NLS) located at residues 72–80, 412–418 and 939–945, respectively (Figure 2E) [64]. From the truncated EBNA3C mutants that were tested, only the N-terminal binding region of EBNA3C (residues 1–365) showed an almost similar ability to repress E2F1-dependent transcriptional activity as of wild-type EBNA3C (Figure 4E). The C-terminal binding domain (residues 621–992) showed an approximately 50% activity when compared to either the full-length (residues 1–992) or the N-terminal binding domain (residues 1–365) (Figure 4E). However, the non-binding middle region (residues 366–620) of EBNA3C had no effect on E2F1 transcriptional activity (Figure 4E). Results from corresponding western blots indicated that co-expression of both full-length (residues 1–992) and the C-terminal binding domain (residues 621–992) led to a reduction in E2F1 expression levels but not in the presence of either N-terminal binding domain (residues 1–365) or non-binding middle region (residues 366–620) of EBNA3C (Figure 4E, bottom panels). Similarly, ectopic expression of EBNA3C proteins showed no effect on GFP expression, demonstrating that EBNA3C may regulate E2F1 transcriptional activity by multiple mechanisms.
Overall, the data suggest that EBNA3C represses the E2F1 transactivation activity by forming a complex with E2F1 at its N-terminal DNA binding domain, perhaps by interfering with its ability to access the target promoters. To test this hypothesis we performed a ChIP assay where we used a similar reporter plasmid containing 3X E2F1 responsive element transfected with vectors expressing flag-E2F1 with or without myc-EBNA3C in HEK 293 cells (Figure 4F). The ethidium bromide stained agarose gel of end products as well as the Ct values from real time PCR results showed that flag-E2F1 strongly bound to the E2F1 responsive sites, but this interaction was drastically impaired in the presence of EBNA3C expression (Figure 4F, top). The specificity of this experiment was confirmed by amplifying a similar size PCR product from the SV40 promoter region of the reporter plasmid, which showed no binding with E2F1 (Figure 4F, bottom). These findings demonstrate that EBNA3C efficiently blocks the recruitment of E2F1 to its responsive sites by inhibiting its DNA binding activity.
It has been well established that in response to DNA damage E2F1 induces apoptosis through both p53-dependent and independent mechanisms [4]. The p53-dependent pathway is mediated through the activation of p19ARF expression which eventually blocks Mdm2 activity [4]. On the other hand, the p53-independent pathway is mediated through the activation of pro-apoptotic genes including p73 and Apaf-1 [4], [9]. To specifically determine the significance of EBNA3C in terms of its direct regulation of E2F1 function, we wanted to verify whether or not EBNA3C could affect E2F1 mediated apoptosis. To investigate the regulation of E2F1 mediated apoptotic activity independent of p53, we used a p53-deficient cell line (Saos-2), since it has been shown earlier that the over-expression of E2F1 can lead to apoptosis in Saos-2 cells in response to DNA damage [65].
First, to determine whether co-expression of EBNA3C can control the growth suppressive effects of E2F1 in response to DNA damage, we performed colony formation assays in Saos-2 cells (Figure 5A). Saos-2 cells were transfected with the expression plasmids for vector control, myc-tagged EBNA3C alone and flag-tagged E2F1 with or without myc-tagged EBNA3C (Figure 5A). Cells were additionally transfected with a GFP expression vector. After 24 h of transfection, cells were exposed to serum starvation plus etoposide (5 µM) treatment for 12 h, subsequently selected with G418 cultured in regular medium, and the number of colonies per plate was screened 14 days later. Figure 5A shows representative plates and average colony counts (bar diagram) from three independent experiments. The results showed that when Saos-2 cells were transfected with E2F1 alone; an approximately 2-fold reduction in efficiency of colony formation was observed compared to cells transfected with the empty vector (Figure 5A). However, cells co-transfected with E2F1 plus EBNA3C showed an approximately 7–8 fold increase in efficiency of colony formation compared to E2F1 alone (Figure 5A), indicating that EBNA3C expression neutralizes the growth inhibitory effect of E2F1 in response to initial DNA damage signals. Interestingly, EBNA3C alone exhibited a drastic effect in colony-formation efficiency compared to cells either expressing empty vector or E2F1 (Figure 5A).
In order to corroborate the previous experiment, we next performed a cell proliferation assay, where Saos-2 cells were transfected with plasmids expressing either flag-tagged E2F1 alone or in the presence of either myc-EBNA3C (wild-type, residues 1–992) or the non-binding EBNA3C domain (residues 366–620). After selection of the transfected cells with G418 similarly as stated above for 2 weeks, the proliferation rate of the selected cells was measured by an automated cell counter for 6 days (Figure 5B). Dead cells (approximately 5%) were excluded using Trypan Blue staining. The results showed that the cell-proliferation rate of cells stably expressing E2F1 plus EBNA3C was approximately 4-fold higher compared to either E2F1 alone or when co-expressed with EBNA3C residues 366–620 (Figure 5B). The results indicated that the interaction between EBNA3C and E2F1 is necessary to block E2F1-mediated anti-proliferative effects in response to DNA damage.
Can EBNA3C expression affect the ability of E2F1 to induce apoptosis? To answer this question, the levels of apoptotic cells in the stably transfected Saos-2 cells (as mentioned above) were examined in response to DNA damage signals by two successive methods, flow cytometric analysis and TUNEL assay (Figure 5C and 5D, respectively). Saos-2 cells stably expressing E2F1 resulted in induction of a significant level of apoptosis compared to the basal level of apoptosis in cells stably expressing either empty vector or EBNA3C alone (Figure 5C–D). However, co-expression of EBNA3C resulted in inhibition of E2F1 mediated apoptosis by approximately 50% (Figure 5C). This effect was more dramatic in TUNEL assays, with a decrease in approximately 75% (Figure 5D). However, in the presence of the non-binding region of EBNA3C (residues 366–620), there was no sign of reduction in E2F1-mediated apoptosis (Figure 5C). Instead there was a slight increase in the level of apoptosis (approximately 10%) (Figure 5C). Interestingly, the basal level of apoptosis of cells expressing EBNA3C residues 366–620 was relatively higher (approximately 5%) compared to both cells stably expressing vector control and wild-type EBNA3C (Figure 5C). Nevertheless, these data clearly indicated that EBNA3C can provide cells with a significant level of protection from E2F1 mediated apoptosis. Altogether our results suggest that EBNA3C plays a critical role in regulating the apoptotic and anti-proliferative functions of E2F1 independent of p53 in response to DNA damage.
It has been shown earlier that EBNA3C blocks p53 dependent apoptosis [40], [42]. In addition, the abovementioned data clearly revealed that EBNA3C also negatively regulates E2F1 mediated apoptosis in a p53 null cell background. To determine the apoptotic cells in response to DNA damage signals, cells were cultured in medium with reduced serum (0.1% FBS) conditions and treated with 5 µM etoposide for 12 h prior to analyze by flow cytometry for sub G1 content (Figure 6A and 6D). Analysis of both serum starved and etoposide treated EBV negative Burkitt's lymphoma cells Ramos and DG75 showed an increased level of apoptotic cells compared to two different clones of LCLs (LCL1 and LCL2), which is approximately 3–4 fold higher (Figure 6A). To further test whether or not EBNA3C regulates the endogenous E2F1 activity in EBV transformed cells, LCLs knockdown for EBNA3C (Sh-E3C) were generated using lentiviruses that express short hairpin RNA against EBNA3C gene. LCLs with sh-control (Sh-Con) represent a non-complementary RNA element to the human genome sequence. As expected, reduction of EBNA3C expression in LCLs resulted in a significant increase in apoptosis (∼3-fold) compared to LCLs with sh-control in response to serum starvation and etoposide treatment (Figure 6D). In agreement to the flow cytometry results, western blot data also showed an elevated level of PARP cleavage in both EBV negative cell lines compared to EBV transformed LCLs (Figure 6B). As expected, EBNA3C knockdown LCLs revealed more PARP cleavage compared to LCLs with sh-control (Figure 6E). To further validate these observations, cell-death assays were conducted using these cell lines in the absence of growth stimuli for a period of 6 days (Figure 6C and 6F). The results showed that DNA damage caused by etoposide treatment and serum starvation resulted in a drastic increase in cell death (approximately a 4-fold difference) in EBNA3C knockdown cells (Sh-E3C) compared to wild-type (LCL1 and LCL2), as well as control LCLs (Sh-Con) (Figure 6C and 6F). These results further extends previously published data [66], [67], [68], which indicated that EBNA3C is absolutely necessary to block DNA damage response in LCLs.
The abovementioned apoptotic phenomena due to downregulation of EBNA3C expression in LCLs could be attributed as a cumulative effect of both E2F1 and p53 mediated apoptosis. The inhibitory effect of EBNA3C on E2F1 mediated transcriptional activity led us to further investigate the basal expression levels of E2F1 both in primary as well as in latent infection model systems. For primary infection, approximately 10 million peripheral blood mononuclear cells (PBMC) from healthy donors were infected with either wild-type (WT) or EBNA3C knockout (ΔE3C) BAC-GFP EBV as previously described [22], [43]. In order to check EBV infection, GFP fluorescence was assessed using fluorescence microscopy (data not shown). Infected PBMCs with the wild-type virus were initially assessed for mRNA expression levels of both EBNA3C and E2F1 at different times of post-infection (0, 2, 4, 7 and 15 days) (Figure 7A). Real time PCR analysis demonstrated that EBNA3C activation typically occurred at 2 days post-infection and its expression was maintained at a constant level throughout the experiment, which was up to 15 days post-infection (Figure 1A). However, the E2F1 transcript levels was seen particularly robust at 2 days post-infection and gradually declined to a lower expression level similar to uninfected PBMC (Figure 1A). These data strongly corroborated the previous findings that the EBV-induced DNA damage response caused by an early period of hyperproliferation [67] is also linked to the cellular E2F1 expression level, which is further attenuated during LCL outgrowth. In order to determine a definitive role for EBNA3C in attenuating this E2F1 mediated DNA damage response, we generated EBNA3C knockout BAC-GFP virus and infected PBMCs for 2 days to analyze E2F1 transcript levels at hyperproliferative state (Figure 7B). Interestingly, EBNA3C knockout virus infected cells displayed a drastic increase (∼6–7 fold) in E2F1 activation compared to wild-type infection (Figure 7B). A similar trend, however to a lesser extent (∼2–3 fold) was observed 15 days post-infection (Figure S1C). These data clearly supported a concept that EBNA3C expression regulates genotoxic stress at the early stages of infection.
To further determine whether these changes also correlated in latent infection, we assessed both transcript and protein levels of E2F1 and its related apoptotic markers in EBNA3C knockdown LCLs. The results showed that knockdown of EBNA3C in LCLs resulted in an increased level of E2F1 expression both at the protein as well as transcript level, compared to control cells (Figure 7C and 7E, respectively), indicating that EBNA3C expression blocks E2F1 transcriptional activity in EBV transformed cells. Elevated expression of E2F1 also caused an enhanced level of E2F1 targeted gene expression, including p73 and Apaf-1 at the protein and mRNA levels (Figure 7C and 7E, respectively). Similarly, we also observed increased expression of Cyclin E, another bona-fide target of E2F1 during cell-cycle regulation, both at protein as well as transcript level (data not shown). This indicates the specificity of EBNA3C effect on E2F1 transcriptional activity, which is not only confined to E2F1 mediated apoptotic activities but also extended to E2F1 mediated cell-proliferation, possibly maintaining a feedback regulation of uncontrolled cell growth.
In order to corroborate this finding, the LCLs with EBNA3C knockdown cells were transiently transfected with plasmids expressing either vector control or myc-tagged EBNA3C. After 48 h of transfection, cells were harvested and subjected to western blot analysis (Figure 7B). The results showed that rescue of EBNA3C expression in EBNA3C knockdown LCLs had similar results to the wild-type (compare Figure 7C and 7D) as expression levels of E2F1, p73 and Apaf-1 were substantially reduced compared to cells with vector control (Figure 7D). Overall, these results clearly provide a possible explanation for the elevated level of apoptosis in EBNA3C knockdown LCLs. Since downregulation of EBNA3C in LCLs could affect the expression of other critical EBV latent proteins [69], we investigated the effect of EBNA3C inactivation on the expression of EBNA2 and LMP1 proteins using specific monoclonal antibodies. The results showed that the expression levels of both EBNA2 and LMP1 in EBNA3C knockdown LCLs were not affected when compared to control cells (Figure 7F).
To more rigorously evaluate the EBNA3C effect on E2F1 recruitment to its targeted promoters, we performed a ChIP assay on both endogenous p73 and Apaf-1 promoter using E2F1 antibody in these cell lines (both Sh-C and Sh-E3C). Indeed, the results showed that in response to DNA damage EBNA3C knockdown resulted in increased recruitment of E2F1 (2–3 fold) on these promoters compared to LCLs with Sh-control (Figure 7G). We next determined if the repressive effects of EBNA3C on the transcription level of both p73 and Apaf-1 seen in LCLs would also be observed using an exogenous system (Figure 7H–I). The results showed that in Saos-2 (pRb−/−) cells, co-transfection of both wild-type p73 and Apaf-1 promoters with an E2F1 expression vector resulted in activation of transcription, which was repressed by approximately 50% when co-expressed with EBNA3C (Figure 7H and 7I, respectively). Interestingly, EBNA3C expression alone also caused approximately 50% reduction of basal promoter activity (Figure 7H–I), perhaps by inhibiting the endogenous E2F1 activity in Saos-2 cells. It also suggested that EBNA3C may act more broadly to repress these promoter activities by blocking the recruitment of other transcription factors onto these promoters, as for example p53 recruitment on Apaf-1 promoter [70].
To address this phenomenon more directly and to nullify the effect from other transcriptional factors, we employed a direct oligo-pulldown assay where cell extracts from Saos-2 cells transfected with expression vectors for flag-tagged E2F1 with or without myc-tagged EBNA3C were incubated with biotinylated oligonucleotides containing only E2F1 responsive elements specific to either p73 or Apaf-1 promoters as described schematically in Figure 7J. Oligonucleotide-bound E2F1 protein was detected by immunoblotting using anti-flag antibody (Figure 7K). In agreement with the ChIP and reporter assays, these results also showed that the E2F1 DNA-binding activity to both oligonucleotides was hindered in the presence of EBNA3C (Figure 7K, compare the top and middle panels), providing a possible explanation for EBNA3C regulation of E2F1-mediated apoptosis. The specificity of this experiment was verified by using either mutant oligonucleotides or performing a competitive binding assay with 200 molar excess of the corresponding non-biotinylated oligonucleotide (Figure 7K). Taken altogether, these results showed that EBNA3C can block E2F1 mediated apoptosis by downregulating both p73 and Apaf-1 expression through inhibiting DNA-binding ability in EBV transformed cells.
To further assess E2F1 mediated apoptotic activities in EBV transformed cells, we generated LCLs stably knockdown for E2F1 (Sh-E2F1) using similar lentivirus technique as mentioned before. In order to minimize the off-target effects we chose two different Sh-RNA sequences which are previously reported [47]. First, LCLs were transiently transfected with these Sh-RNA containing plasmids and validated by western blot 48 h post-transfection (Figure S1D). The results showed that both these Sh-RNAs efficiently silenced E2F1 expression in transfected LCLs (Figure S1D). Subsequently, corresponding lentiviruses were made from Sh-E2F1 #1 expressing vector and stably trasfected LCLs were generated. As expected, western blot analysis of these cells showed that reduction of E2F1 level also resulted in a significant decrease in expression levels of E2F1 regulated apoptotic markers including both p73 and Apaf-1 as well as cell-cycle regulatory protein Cyclin E (Figure 8A). This indicates that downregulation of E2F1 may have an effect on both cell-proliferation and apoptosis in EBV transformed cells.
To further explore E2F1 function in regulating LCLs growth, proliferation assays were conducted both in the absence and presence of DNA damage response (Figure 8B and 8C, respectively). Interestingly, the results showed that upon E2F1 knockdown, LCLs response to varying stimuli was distinctly different (Figure 8B–C). As expected from the western blot results (Figure 8A), the growth rate of LCLs knockdown for E2F1 (Sh-E2F1 #1) showed a significant reduction (approximately 2-fold) compared to control cells (Sh-Con) in the presence of mitogenic stimuli (normal cultured medium with growth factors) (Figure 8B). Importantly, the cell cultures used in these assays had greater than 98% viability, as determined by trypan blue exclusion method (data not shown). However, in the absence of growth stimuli, etoposide (5 µM) treatment induced marked cell death in control cells (Sh-Con) than that of E2F1 knockdown LCLs (Sh-E2F1 #1) over a period of 6 days, probably by inducing a greater level of apoptosis (Figure 8C). To further support this notion, we performed an apoptosis assay to quantitatively determine the apoptotic response in these cell lines with an increasing concentration of etoposide (Figure 8D–F). The representative histogram shows the analysis of multiple experiments which clearly demonstrated that LCLs knockdown for E2F1 (Sh-E2F1 #1) were less responsive (approximately 2-fold) to apoptotic stimuli compared to control cells (Sh-Con) as seen with an increasing dose of etoposide treatment (Figure 8D–E). Western blot of PARP cleavage in these cells additionally confirmed the flow-cytometric analysis, which showed more cleavage in control cells (Sh-Con) compared to E2F1 knockdown LCLs (Sh-E2F1 #1) in a dose dependent manner (Figure 8F). These observations clearly suggest that E2F1 plays a dual role in EBV positive cells and the active engagement of EBNA3C and E2F1 is necessary to block E2F1 induced apoptosis in response to DNA damage signals in LCLs.
E2F1 expression is strictly cell-cycle dependent and its protein level is unstable due to active degradation through the ubiquitin–proteasome pathway [71]. So far, our results showed that EBNA3C expression led to a decrease in steady state level of E2F1 expression. One mechanism, which we clearly demonstrated was that EBNA3C can efficiently block E2F1 mediated transcription. Since, EBNA3C was earlier shown to play a critical role in modulating the ubiquitin-proteasome machinery to regulate many important cell-cycle components [41], [43], [54], [60], we wanted to further determine whether or not EBNA3C can also regulate E2F1 degradation. To examine our hypothesis, transiently transfected HEK 293 cells with flag-tagged E2F1 with or without EBNA3C were treated with the proteasome inhibitor, MG132 (Figure 9A). Cells were additionally transfected with GFP expression plasmid to check the transfection efficiency. The results showed that co-expression of EBNA3C led to a considerable decrease in E2F1 ectopic expression levels (∼1.5 fold), whereas no change was observed in GFP expression levels (Figure 9A). However, after treatment with MG132 for a period of 12 h, the loss of E2F1 expression level was rescued compared to mock treatment. This strongly indicates that the decreased level of E2F1 observed in the presence of EBNA3C was a result of destabilization of E2F1 through the ubiquitin-proteasome degradation pathway.
To directly assess EBNA3C mediated destabilization of E2F1, HEK 293 cells were transfected with flag-E2F1, myc-EBNA3C and GFP expression vectors. At 36 h post-transfection, cells were treated with protein synthesis inhibitor cycloheximide (CHX), and samples were collected at different intervals - 0, 2, 4, and 6 hours. Western blots probed with flag antibody showed that the stability of the E2F1 protein was significantly reduced by EBNA3C co-expression, whereas GFP expression was unaltered (Figure 9B). The decreased stability of E2F1 in the presence of EBNA3C, prompted us to investigate whether EBNA3C facilitates poly-ubiquitination of E2F1 and thus enhances its degradation. To explore this possibility, a ubiquitination experiment was set up, where HEK 293 cells were transiently co-transfected with expression constructs for HA-ubiquitin, flag-E2F1 and myc-EBNA3C and the ubiquitination of E2F1 was measured by immunoprecipitation followed by Western blotting with anti-HA antibody (Figure 9C). The results clearly demonstrated a significant elevation in E2F1 poly-ubiquitination level in the presence of both EBNA3C and ubiquitin (Figure 9C). Overall, the results of these experiments suggest that EBNA3C can destabilize E2F1 by regulating its targeted degradation likely through recruitment of the ubiquitin-proteasome degradation system.
To more rigorously assess this in an endogenous background, we analyzed E2F1 stability as well as its ubiquitination levels using EBNA3C knockdown LCLs (Figure 9D and 9E). The results showed that in wild-type LCLs (Sh-Con), E2F1 was degraded to near completion by 4 h, whereas in EBNA3C knockdown LCLs, E2F1 stability was significantly extended after addition of CHX (Figure 9D). E2F1 half-life was determined to be ∼2 h in EBNA3C expressing wild-type LCLs; however, it was noticeably extended to more than 6 h when EBNA3C expression was compromised (Figure 9C). The results also indicated that in the presence of DNA damage signals (reduced serum and etoposide treatment) caused a significant increase in E2F1 expression as well as its stability in both control as well as EBNA3C knockdown LCLs (Figure 9D). Interestingly, EBNA3C stability also seem to be affected in response to DNA damage signals compared to mitogenic stimuli, where no sign of EBNA3C degradation was observed (Figure 9D, compare panels 1 and 3). The consequences of the transient decrease in EBNA3C level in response to DNA damage was thus manifested in an increase of E2F1 stability, which may explain the E2F1 dependent apoptotic phenomenon in EBV transformed LCLs. As expected, western blot analysis with anti-ubiquitin antibody of immunoprecipitated E2F1 revealed a significant decrease in ladder of higher molecular weight E2F1 species in EBNA3C knockdown LCLs compared to control cells (Figure 9E). This ladder is even more evident with MG132 treatment (data not shown).
The dynamic changes in E2F1 expression level in the presence of EBNA3C and MG132 treatment support our reporter assays above indicating that ubiquitin-proteasome dependent degradation is also associated with EBNA3C mediated E2F1 transcriptional suppression that is attenuated in response to DNA damage signals. To determine whether these changes correlated with E2F1 transcriptional activity, we performed similar promoter assays as described before using wild-type (3X-WT-E2F1-luc) E2F1 responsive reporter construct in the absence and presence of MG132 (Figure 9F). The results showed that addition of MG132 caused an inhibition of EBNA3C mediated blocking of E2F1 transcriptional activity (Figure 9F), which was also evident from E2F1 western blots (Figure 9F, bottom panels). However, MG132 addition could not completely reverse EBNA3C mediated E2F1 transcriptional inhibition, suggesting that even if the total E2F1 protein was enhanced via deregulation of the ubiquitin-proteasome machinery, free E2F1 species was still scarce due to the active EBNA3C-E2F1 complex. We next assessed whether MG132-mediated activation of E2F1 transcriptional activity affects apoptotic regulation in LCLs (Figure 9G–H). The presence of MG132 led to an increase in E2F1 total protein levels in both control as well as EBNA3C knockdown LCLs causing a considerable elevation in apoptosis as evident from PRAP cleavage (Figure 9G). However, the extent of apoptosis was much lower in control cells compared to EBNA3C knockdown LCLs (Figure 9G). Importantly, MG132 treatment did not show any significant effect in LCLs silenced for E2F1 (Figure 9H), suggesting that MG132 specifically acts to alleviate E2F1 protein stability through blocking of the ubiquitin-proteasome degradation pathway rather than via a different mechanisms. Overall the data suggest that besides blocking E2F1 transcriptional activity, EBNA3C actively participates to regulate E2F1 degradation in a ubiquitin-proteasome dependent manner (Figure 9I).
Cancer development critically depends on the subtle balance between cell proliferation and apoptosis mediated cell death. p16INK4a-Cyclin D/CDK-Rb-E2F cascade is thought to be a major determinant in regulating cell fate. Deregulation of E2F family member activities occurs due to the functional deviation of the upstream molecules in this pathway, which includes inactivation of Rb pocket proteins (pRb, p107, p130), p16INK4a tumor suppressive functions, genetic manipulation of cyclin D (D1, D2 and D3) oncogenes and its kinase partners CDK4/6, which confers a growth advantage and thus has become a hallmark of human cancer [72], [73]. In addition to regulation of cell proliferation, compelling evidence now indicates that E2F1 can also induce apoptosis under various cellular events regardless of p53 status [9], [47], [53]. Given the frequent inactivation of the tumor suppressor proteins pRb and p53 in human cancers [74], [75], E2F1 mediated apoptosis may provide an additional tumor surveillance mechanism. The E2F1 mediated apoptosis pathway is therefore emerging as a promising therapeutic target in controlling cancer development [2], [76].
Previous reports have suggested that the EBV essential latent antigen EBNA3C critically manipulates upstream components of E2F1 in this pathway. For example, EBNA3C mediated repression of p16INK4a expression was shown to be essential for LCLs growth [23], [77]. Recently, we have shown that EBNA3C facilitates S phase entry through stabilizing and enhancing Cyclin D1/CDK6 activity [43]. Moreover, EBNA3C recruits SCFSkp2 E3 ligase activity for ubiquitin-mediated degradation of pRb [54]. EBNA3C was also shown to interact with pRb in the presence of proteasome inhibitor [54]. It is therefore compelling to investigate whether EBNA3C can further regulate the function of E2F1; the downstream effector of this pathway in order to control proliferation of EBV associated cancer cells. We initiated our study with binding experiments and we conclusively show that EBNA3C and E2F1 can form a pRb independent complex. Using a series of truncated EBNA3C and E2F1 proteins, we show that the N-terminal DNA binding domain of E2F1 (residues 1–243) is sufficient to interact with two distinct sites of EBNA3C, one lies at N-terminal residues 100–200 and another at C-terminal region comprising residues 621–700. Earlier studies have shown that this N-terminal region of E2F1 is responsible for apoptotic induction but also contains a Cyclin A interaction motif [65], [78]. Interestingly, the N-terminal domain of EBNA3C binds to E2F1 directly, while the C-terminal domain associates in a complex with E2F1 in cells. This N-terminal binding region of EBNA3C was shown to be particularly important as it binds to many critical cell-cycle regulators, including Cyclin A, Cyclin D1 and pRb and SCFSkp2 [43], [54], [56], [60]. Furthermore, genetic study using recombinant EBV expressing conditionally active EBNA3C demonstrated the importance of this particular N-terminal domain of EBNA3C as upon deletion of this N-terminal region there was a significant reduction in LCLs growth [31], whereas the C-terminal domain (residues 621–700) was dispensable [31].
E2F1 is an essential transcriptional activator of many cellular genes required for the G1 to S phase transition [3], [7], [79]. The active participation of EBNA3C in controlling G1-S phase [43], [57], combined with our binding data, prompted us to investigate the feasibility of EBNA3C mediated regulation of E2F1 transcriptional activity. A recent study using a genetically engineered EBV Bacmid has also shown that EBNA3C strongly attenuates DNA damage response induced during EBV-mediated B-cell transformation [67]. In agreement with this data our results show that EBNA3C knockout virus is incapable of suppressing E2F1 mediated DNA damage response during the early stages of infection. Our results also show that EBNA3C represses E2F1 mediated transcriptional activity by blocking the E2F1-DNA binding ability in latent infection using EBNA3C knockdown LCLs as confirmed by endogenous ChIP experiments. Interestingly, in support of our finding a recent publication by White et al. also showed in a microarray analysis that E2F1 transcript is specifically elevated by EBNA3C knockout virus infection compared to wild-type EBV [80]. However, in this paper this observation is entirely ignored and unaccredited. In certain human tumors genetic amplification and over-expression of E2F3 has been documented, but there were no clear indication of an oncogenic role for the other ‘activators’ of the E2F family members (E2F1 and E2F2) [81]. In addition to its well-established function in controlling cell proliferation, E2F1 is also capable of DNA damage-induced apoptosis by targeting several related genes including p73, Apaf-1, and caspases [9], [12], [13], [14]. The data presented here allow us to propose a model in which association of EBNA3C with E2F1 inhibits its DNA-binding ability as well as transcriptional activity that eventually blocks E2F1 mediated apoptosis in response to DNA damage by downregulating the target genes p73 and Apaf-1.
A number of DNA damage signaling events are clearly involved in the induction of E2F1 and its stabilization. However, the mechanism by which these modifications can lead to E2F1 stabilization remains unclear. E2F1 protein is known to be regulated through an ubiquitin-proteasome pathway in a cell-cycle dependent manner [82], [83], [84], which relies upon its dissociation from pRb and its binding to specific E3-ubiquitin ligases. One of the E3-ubiquitin ligases involved in E2F1 ubiquitination and degradation is SCFSkp2 [85], [86]. As previously described EBNA3C was also shown to interact and recruit this E3 ligase activity for pRb degradation [54]. Thus, one can expect that EBNA3C may also be involved in regulating E2F1 protein stability through modulation of its ubiquitination status. Indeed, our results showed that EBNA3C facilitates E2F1 degradation in an ubiquitin-proteasome dependent manner. However, we are not certain whether EBNA3C recruits solely SCFSkp2 activity for E2F1 degradation, as there are a number of molecules which actively targets E2F1 for degradation [87]. Further a comprehensive study is required to evaluate the E2F1 degradation pathway in an EBV background. Our results support and extend our model in which two distinct events, the control of DNA binding and protein stability contribute to the downregulation of the transcriptional activation function of E2F1 in EBV transformed LCLs.
We could not rule out the contribution of other ‘unknown’ events that may control EBNA3C mediated inhibition of E2F1 induced apoptosis. For example, identification of ATM-Chk2 signaling pathway as a mediator that specifically stabilizes E2F1 through phosphorylation in response to DNA damage provides us a conceptual framework to understand the critical interplay between cell proliferation and apoptosis regulated by E2F1 [16], [17]. Specifically, Chk1 and Chk2 were shown to promote E2F1 stabilization and activity after genotoxic stress and thereby contribute to E2F1-induced upregulation of p73 and consequently apoptosis [52]. In addition, 14-3-3τ, a phosphoserine-binding protein, stabilizes E2F1 via inhibition of ubiquitination [15]. Interestingly, we have previously shown that EBNA3C targets Chk2 to bypass G2/M transition under genotoxic stress [68]. A recent study also showed that EBNA3C attenuated ATM-Chk2 DNA damage responsive signaling pathway to establish B-cell immortalization [67]. It would therefore be important to understand the precise molecular regulation of E2F1 induced apoptosis during initial as well as persistent EBV infection in primary B-lymphocytes. This is currently under investigation in our lab.
Beside protein phosphorylation, acetylation is also known to be a conserved mechanism modulating the activity of several pro-apoptotic proteins in response to DNA damage, so as to selectively induce apoptosis [9], [88]. It has been shown earlier that E2F1 post-translational modification that occurs after DNA damage is important in directing E2F1 on the promoter of the proapoptotic gene p73 [9], [88]. It is as yet unknown from our study whether EBNA3C can also affect E2F1 acetylation specifically in the presence of DNA damage signals to regulate apoptosis. Additional studies are required to fully elucidate the combinatorial effects of these different mechanisms and the intricate network by which EBNA3C affects both the levels and activity of E2F1 to regulate apoptosis.
It is well established that the ability of E2F1 to drive apoptosis is distinct from its ability to drive cell division [79]. Our data show a fascinating observation that E2F1 plays a dual role in EBV positive LCLs. As expected, LCLs knockdown for E2F1 exhibited a reduced growth rate, whereas, in response to DNA damage E2F1 knockdown LCLs were more resistant to apoptosis, indicating that E2F1 acts as both an oncogene and a tumor suppressor in response to different stimuli. It is possible that different E2F1 target genes were selectively modulated by the EBNA3C-E2F1 complex, during the cell-cycle and in response to DNA damage. However, this issue could not be directly addressed until the E2F1 target genes essential for both routes are clearly identified. The factors that determine the decisions of inducing either cell division or cell death need to be further investigated in EBV positive cells. Overall, our findings suggest that, in addition to its seemingly contradictory roles as oncogene and tumor suppressor in tumorigenesis, E2F1 actively promotes DNA-damage induced apoptosis in LCLs and thus it could serve as an important determinant for chemosensitivity in EBV associated human cancer therapy, irrespective of p53 status.
|
10.1371/journal.pntd.0000724 | Isolation and Characterization of New Leptospira Genotypes from Patients in Mayotte (Indian Ocean) | Leptospirosis has been implicated as a severe and fatal form of disease in Mayotte, a French-administrated territory located in the Comoros archipelago (southwestern Indian Ocean). To date, Leptospira isolates have never been isolated in this endemic region.
Leptospires were isolated from blood samples from 22 patients with febrile illness during a 17-month period after a PCR-based screening test was positive. Strains were typed using hyper-immune antisera raised against the major Leptospira serogroups: 20 of 22 clinical isolates were assigned to serogroup Mini; the other two strains belonged to serogroups Grippotyphosa and Pyrogenes, respectively. These isolates were further characterized using partial sequencing of 16S rRNA and ligB gene, Multi Locus VNTR Analysis (MLVA), and pulsed field gel electrophoresis (PFGE). Of the 22 isolates, 14 were L. borgpetersenii strains, 7 L. kirschneri strains, and 1, belonging to serogoup Pyrogenes, was L. interrogans. Results of the genotyping methods were consistent. MLVA defined five genotypes, whereas PFGE allowed the recognition of additional subgroups within the genotypes. PFGE fingerprint patterns of clinical strains did not match any of the patterns in the reference strains belonging to the same serogroup, suggesting that the strains were novel serovars.
Preliminary PCR screening of blood specimen allowed a high isolation frequency of leptospires among patients with febrile illness. Typing of leptospiral isolates showed that causative agents of leptospirosis in Mayotte have unique molecular features.
| Leptospirosis has been recognized as an increasing public health problem affecting poor people from developing countries and tropical regions. However, the epidemiology of leptospirosis remains poorly understood in remote parts of the world. In this study of patients from the island of Mayotte, we isolated 22 strains from the blood of patients during the acute phase of illness. The pathogenic Leptospira strains were characterized by serology and various molecular typing methods. Based on serological data, serogroup Mini appears to be the dominant cause of leptospirosis in Mayotte. Further molecular characterization of these isolates allowed the identification of 10 pathogenic Leptospira genotypes that could correspond to previously unknown serovars. Further progress in our understanding of the epidemiology of Leptospira circulating genotypes in highly endemic regions should contribute to the development of novel strategies for the diagnosis and prevention of this neglected emerging disease.
| Leptospirosis, a zoonotic disease with a worldwide distribution, is an important emerging infectious disease [1]. Rodents are a main reservoir of the pathogenic agents of this disease, spirochetes of the genus Leptospira, excreting the bacteria in their urine. Humans are usually infected through contaminated water. This increasingly common disease affects impoverished populations from developing countries and tropical regions [2]. Leptospirosis is an endemic disease in rural regions of developing countries because of the exposure to a large number of animal reservoirs [3], [4]. Furthermore leptospirosis is an emerging health problem in urban slums where inadequate sanitation has produced the conditions for rat-borne transmission of the disease [5], [6], [7]. Outbreaks of leptospirosis are associated with heavy seasonal rainfall [5], [8], [9], [10] and extreme climatic events, such as hurricanes [6], [11] and El Niño [12], [13]. Over the last decade, outbreaks of leptospirosis were also associated to adventure tourism [14], [15]. More than 500,000 cases of severe leptospirosis are currently reported each year, with case fatality rates exceeding 10% [16]. However, its prevalence is still underestimated due to low awareness among the medical community and an absence of specific symptoms and readily available tests. In addition, some patients may also experience transient or mild manifestations [2].
Mayotte is a French-administrated territory located northwest of Madagascar in the Comoros archipelago, southwestern Indian Ocean (Figure 1A). The climate of this volcanic island is generally tropical and mild. The estimated population in 2007 was 186,387. The population is primarily of African origin. Unemployment is high and 39% of the population consists of immigrants from the neighboring Comoro Islands. Many of the inhabitants rear dairy herds and/or cultivate rice, cassava roots (manioc), maize, bananas or pineapples. Most households lack proper sanitation.
Leptospirosis is endemic in Mayotte [17]. The annual incidence of leptospirosis was previously estimated to be 3.8 patients per 100,000 individuals between 1984 and 1989 [17]. In 2008 and 2009, the annual incidence rate was estimated to be 45 patients per 100,000 individuals.
Leptospires are usually classified into species and serovars or serogroups. There are over 200 recognized pathogenic serovars; these serovars are currently clustered into 24 serogroups [1], [18], [19]. There have been no extensive studies on leptospiral serovars obtained from humans in Mayotte. Hence, we investigated the serological and genetic characteristics of leptospiral isolates isolated from leptospirosis patients in 2007–2008. Twenty-two plasma samples from heparinized blood specimens of individuals with leptospirosis-like illness testing positive for Leptospira spp. by PCR, were cultured and characterized by serology, sequencing of 16S RNA and ligB, and pulsed field gel electrophoresis (PFGE). In this study, our approach allowed a high rate of isolation of Leptospira from patients. We also report the existence of ten potentially new pathogenic Leptospira genotypes, which cause acute leptospirosis in Mayotte.
Blood samples (heparinized blood for culture and EDTA plasma for DNA extraction) were obtained from patients during the acute phase of illness (fever of 38°C or higher for no more than 7 days, accompanied by headache and/or myalgia) after oral assent after reading a script, which was approved by the Ethical Committee of the Centre Hospitalier de Mayotte, that informed of the possible use of blood samples for scientific purpose. Informed consent was recorded in writing in the patient's file as required by the Ethical Committee.
Ten drops (250 µl) and 20–40 drops (500–1000 µl) of plasma from heparinized blood were transferred into two tubes containing 9 ml of EMJH liquid medium [20], [21]. Cultures were incubated at 30°C and examined weekly, for 3 months, by dark field microscopy. In case of contamination, cultures were filtered through 0.22 µm pores to remove contaminants. Reference strains were obtained from the collection maintained by the National Reference Laboratory for Leptospira, which is also a WHO Collaborating Center, at the Institut Pasteur (Paris, France).
Serological characterization of clinical isolates was performed at the National Reference Center for Leptospira. A Microscopic agglutination test (MAT) was performed to determine the serogroup of Leptospira isolates using rabbit antisera against reference serovars representing a standard battery of 24 serogroups (Text S1). High rates of agglutination of the serum with one particular antigen are used to identify the presumptive serogroup of the infecting bacterium [22].
To determine if clinical isolates would induce an infection in laboratory animals, a group of four 28-day-old gerbils (Charles River Laboratories, http://www.criver.com) were inoculated intraperitoneally with 101, 102, 103, 104, and 106 leptospires from strain 2007/01203. A group of control was also inoculated with EMJH medium. Animals were monitored daily for clinical signs of leptospirosis (i.e., prostration, jaundice, etc) and survival for up to 21 days post infection. Protocols for animal experiments were prepared according to the guidelines of the Animal Care and Use Committees of the Institut Pasteur.
Histopathologic analysis was perfomed after necropsy of infected animals which received 104 leptospires at the day of death (6 or 7 days post -inoculation) and non infected animals. Liver, kidneys, and lungs were removed and fixed in 4% buffered formaldehyde for standard microscopic analysis; serial sections were stained with hematoxylin and eosin (HE) and Warthin–Starry silver impregnation as previously described [23]. The pathologist viewed the histopathological preparations without knowing the infection status of the animals.
Genomic DNA was extracted from 400 µl of EDTA plasma using a MagNaPure Compact instrument (Roche Molecular Diagnostics), and yielded 50 µl of eluate. Leptospires in plasma were detected by quantitative real-time PCR (qPCR) using the Light cycler LC480 system (Roche) or the Cobas TacMan 48 system (Roche) as previously described [24]. A standard curve with DNA extracted from 10-fold dilutions of known numbers of leptospires was used for quantification. Samples with a threshold cycle (Ct) value >45 were considered negative. However, patient 13 with a Ct of 44.7 (Table 1) close to the cutoff was also included in this study.
Genomic DNA was extracted from EMJH cultures using the Cell DNA Purification kit (Maxwell, Promega, Madison, WI). DNA was amplified using Taq polymerase (Amersham) under standard conditions. The amplified products were analyzed by 1% agarose gel electrophoresis. The 16S rRNA gene was amplified with the primers LA (5′-GGCGGCGCGTCTTAAACATG-3′) and LB (5′-TTCCCCCCATTGAGCAAGATT-3′) [25]. Partial ligB sequences were amplified with the primers PSBF 5′-ACWRVHVHRGYWDCCTGGTCYTCTTC-3′) and PSBR (5′-TARRHDGCYBTAATATYCGRWYYTCCTAA-3′) [26]. Sequencing was performed at the Genotyping of Pathogens and Public Health Platform (Institut Pasteur, Paris, France). The assembled sequence was then aligned against other 16S rRNA sequences available in GenBank using BLAST (http://www.ncbi.nlm.nih.gov/BLAST). Genotyping was also performed by multiple-locus variable-number tandem repeat analysis (MLVA) using the loci VNTR4, VNTR7, VNTR10, Lfb4 and Lfb5 as described by Salaun et al. [27].
For pulsed-field gel electrophoresis (PFGE), cells were embedded in agarose plugs as previously described [28]. DNA plugs were restriction digested with NotI. PFGE was performed in a contour-clamped homogeneous electric field DRIII apparatus (Bio-Rad Laboratories, Richmond, CA). Programs with a ramping from 1 to 70 s for 36 h at 150 V and from 10 to 100 s for 40 h at 150 V were used to resolve of restriction patterns.
We used blood sample cultures to investigate patients with febrile illness in Mayotte. We assayed 388 human plasma samples from blood collected in heparin by quantitative real-time PCR between April 2007 and September 2008 (not including serial samples from a same patient). The specific amplification of pathogenic Leptospira spp. was detected in 53 (13,7%) samples (Figure 2). Blood samples from 29 patients diagnosed with leptospirosis (as determined by PCR) were examined by culture. Samples from 26 patients (89.7%) were positive by culture, but leptospires in 4 cultures were lysed on arrival at the National Reference Center of Leptospira (Institut Pasteur). Twenty-two clinical isolates were therefore included in this study. The median time to culture positivity for the primary culture was 19.3 days (7 to 77 days).
Ct values of samples that were positive by culture were between 25.2 and 44.7 (median 34.9) (Table 1). Past experience has shown that prior antimicrobial drug treatment reduces the chances of detecting leptospires in the blood by real-time PCR and culture (data not shown).
We collected demographic and epidemiologic data (age, sex, gender, occupation or hobbies, exposure to water and/or animals) of the twenty-two leptospirosis patients. Many patients were male (77.2%). The median age was 34 (13–78) years old. Patients originated from various locations, illustrating the fact that this disease is distributed across the whole country. Most cases (16/22) occurred during the hot and rainy season from December to April (temperate dry season from May to October). Among the leptospirosis cases with information on occupation (6/22), all had occupations associated with contact with surface water or animals (Table 1).
The clinical manifestations of leptospirosis range from a mild febrile illness to a severe and potentially fatal illness, characterized by jaundice, renal failure, thrombocytopenia, and hemorrhage (Weil's disease). Two patients did not survive. Clinical data generally included fever, myalgia, headache, and elevated bilirubin, creatinine, and transaminase. Patients did not exhibit severe forms of pulmonary leptospirosis.
Serological identification of the clinical isolates was performed using the microscopic agglutination test (MAT) method with reference sera that were representative of the major Leptospira serogroups (Text S1). Serogrouping of the 22 isolates revealed agglutination titers for antisera raised against serogroups Mini, Hebdomadis, Pyrogenes, and Grippotyphosa, and negative reactions with antisera raised against the remaining serogroups. The vast majority of isolates (20/22) exhibited high agglutination reactions (400–12,800) with the antiserum raised against serogroup Mini, of which 6/20 also showed detectable agglutination reactions (200–3,200) with the Hebdomadis antiserum. Strains 2008/01774 and 2007/01872 showed a high level of reaction with rabbit antiserum raised against serovars Grippotyphosa and Pyrogenes, respectively (Table 1).
Leptospira isolates were characterized by partial 16S rRNA and ligB sequencing, the amplification of VNTR loci, and PFGE separation of NotI-digested genomic DNA (Table 1). The amplification and sequencing of the 16S rRNA gene (rrs) showed that clinical isolates belong to the pathogenic species L. borgpetersenii (14/22), L. kirschneri (7/22), and L. interrogans (1/22). Sequence analysis of ligB, which is a pathogen-specific gene [26], allowed differentiation below the species level. Thus, L. borgpetersenii isolates are further divided in two different clusters (A and B). The dendrogram showing all ligB genotypes is shown in Figure 3.
MLVA (Multi Locus VNTR Analysis) has been mainly developed for the pathogens L. interrogans and L. kirschneri [27]. The locus VNTR 7 is not found in L. borgpetersenii; thus, two other loci (Lb4 and Lb5) are usually amplified for typing L. borgpetersenii isolates [27]. The five patterns defined by MLVA are in agreement with the four clusters determined by ligB sequencing (Table 1). None of these genotypes has been previously identified in reference strains when compared to members of serogroups Mini, Hebdomadis, Pyrogenes, and Grippotyphosa [27].
PFGE is the long-standing gold standard method for genotyping Leptospira strains [29], [30], [31]. Tenover et al. [32] proposed PFGE criteria for interpreting the relatedness of epidemiologically related bacterial isolates. In addition, several studies have shown that PFGE profiles for Leptospira strains of a serovar belonging to the same species were indistinguishable or closely related [29], [30], [31]. In this study, isolates were considered different if more than three band differences were observed. Eleven of 21 isolates (52%) generated a common pattern (pattern I) with PFGE separation of NotI-digested genomic DNA. Thus strains isolated from different patients over a one-year period (from April 2007 to June 2008) gave indistinguishable PFGE patterns. The other 10 isolates were grouped into at least six other clusters. Strains 2008/0695, 2008/01926, 2008/03703, the “Grippotyphosa” (strain 2008/01774) and “Pyrogenes” (strain 2007/01872) isolates all displayed a unique PFGE pattern (Figure 4).
MLVA and PFGE patterns were compared against fingerprints from members of serogroups Mini, Hebdomadis, Pyrogenes, and Grippotyphosa, as well as against three african strains belonging to serogroups Grippotyphosa, Sejroe, and Hebdomadis [33], [34] (Text S1).
The PFGE and MLVA patterns obtained for Leptospira clinical isolates that reacted with rabbit antisera raised against serovar Mini were quite different from the patterns obtained for reference strains of the serogroups Hebdomadis (serovars Jules, Nona, Kabura, Kambale, Kremastos, Worsfoldi, and Hebdomadis) and Mini (serovars Mini, Beye, Georgia, Swajizak, Ruparupae, and Tbaquite) (Figure 4). An interesting observation is the similarity of the PFGE reference strain restriction patterns for serovars Jules and Nona (serogroup Hebdomadis), which are L. borgpetersenii isolates of African origin (Text S1) (Figure 4). These two serovars were also indistinguishable by MLVA (data not shown). Similarly, serovars Mini and Swajizak (L. borgpetersenii strains belonging to serogroup Mini) appeared to be closely related based on their PFGE restriction patterns (Figure 4). The “Grippotyphosa” strain displayed a genotype that was different to that displayed by others members of the serogroup Grippotyphosa (serovars Grippotyphosa, Canalzanae, Dadas, Valbuzzi, Vanderhoedeni, Ratnapura, Muelleri, and Liangguang) (Figures 4 & 5). Similarly the “Pyrogenes” strain was different from the MLVA (data not shown) and PFGE (Figure 4) patterns of members that belong to the L. interrogans serogroup Pyrogenes (serovars Pyrogenes, Abramis, Camlo, Guaratuba, Manilae, Robinsoni, Biggis, and Zanoni).
A representative strain of the major genotype (2007/01203) was tested for virulence in the gerbil infection model of acute leptospirosis [35], [36]. Inoculation with as low as 10 bacteria of strain 2007/01203 induced death in 100% of the animals. Infected gerbils died within 5 to 10 days after the infection (Figure 6). Nasal bleeding was frequently observed in infected animals just before they died. Although gross hemorrhages or jaundice were not detected during necropsy, microscopic foci of hemorrhages were detected in the liver, kidneys, and lungs (Figure 7). Silver impregnation demonstrated a large numbers of spirochetes in the livers, kidneys, and lungs of infected gerbils (Figure 7).
The characterization of Leptospira isolates is essential if we are to understand the epidemiological properties of the disease. Leptospira serovars usually demonstrate specific host preferences. For example, rats serve as reservoirs for the serogroup Icterohaemorragiae, whereas house mice are the reservoir for the serogroup Ballum [1]. Local Leptospira isolates may also serve as antigens for the serodiagnosis of leptospirosis. However the isolation of Leptospira from blood specimens is usually rare because of the low sensitivity of the technique, the need for specific medium (EMJH) and a prolonged period of incubation (>1 month) [37]. Infection causes a leptospiremia within the first week of illness until the host mounts an effective acquired immune response, leaving a narrow window in which bacteria can be detected in blood. By screening blood from patients with febrile illness by real-time PCR, we were able to obtain a high rate of isolation of Leptospira. The real-time PCR technique has been shown to provide good sensitivity and a linear relationship between the bacteria copy number and cycle threshold (Ct) values. In this study, blood samples from the two deceased patients exhibited the lowest Ct values (corresponding to 1×105 and 7.7×105 leptospires per ml of blood, respectively) (Table 1). It was previously shown that a density of 104 leptospires per ml of blood is a critical threshold for the vital prognosis of patients [38], [39].
In this study, we isolated 22 strains from the blood of human leptospirosis cases in Mayotte. The isolates were pathogenic to gerbils and were identified by serology and molecular typing. Based on serological data, serogroup Mini appears to be the dominant cause of leptospirosis in Mayotte. However, the first infections due to serogroup Mini were only reported in 2007, as culture isolation techniques were not attempted for logistic reasons in Mayotte before 2007. Prior to 2007, the standard MAT was performed and it did not include the serogroup Mini in its panel of antigens. Over the last ten years (1998–2008), both serological characterization of isolates (since 2007) and serological detection of antibodies in patient sera have shown that the most prevalent Leptospira serogroups in Mayotte have been Sejroe (21%), Grippotyphosa (14%), and Pyrogenes (10%). Serogroup Icterohaemorrhagiae accounted for only 5% of cases (12 of 230 patient sera) (data from the National Reference Center of Leptospira, France). Three isolates were identified as members of the serogroup Mini in 2007 and 16 isolates were identified as members of the serogroup Mini in 2008. In 2009, 84 cases have been diagnosed by PCR, from which 41 positive cultures were identified at the serological level. The most prevalent Leptospira serogroups were Mini (28/41), Pyrogenes (2/41), Pomona (8/41), and Grippotyphosa (1/41).
The serovar Mini was originally isolated from a patient in Italy in 1940. Other serovars were subsequently isolated from humans and animals (opposum, bandicoot, raccon) and classified into the serogroup Mini (Text S1) as a function of their antigenic determinants [18]. Historically, members of serogroup Mini belonged to the larger serogroup Hebdomadis [18]. This serogroup was then divided into three autonomous serogroups: the serogroups Hebdomadis, Mini, and Sejroe. Members of these serogroups may therefore exhibit serological affinities [18] as observed between serogroups Mini and Hebdomadis for some of our clinical isolates (Table 1).
Four molecular typing methods have been used to further characterize the 22 Leptospira clinical isolates. The four molecular typing methods revealed independent polymorphisms (Table 1). 16S ribosomal RNA gene sequence analysis can distinguish leptospiral species, but it is unable to distinguish between serovars. ligB sequencing and MLVA identified four and five genotypes, respectively, whereas PFGE allowed the recognition of subgroups within these genotypes. Thus, PFGE was the best method for discriminating between strains among the tested typing methods; moreover, it showed a strong correlation with the other techniques. However, the rapidity of PCR makes MLVA and ligB sequencing appropriate for the preliminary genotyping of clinical isolates [26], [27].
The pathogens L. borgpetersenii and L. kirschneri were the dominant species among clinical isolates in Mayotte (located between Tanzania and Madagascar). These data are in agreement with previous studies that show that Leptospira isolates from Africa belong to either L. borgpetersenii or L. kirschneri species [33], [34], [40], [41]. By contrast, in other areas, L. interrogans is considered to be the major species responsible for human infections. Interestingly, the serovars belonging to serogroup Mini were found in clinical isolates belonging to both L. borgpetersenii and L. kirschneri. The “serovar” is identified based on structural heterogeneity of the O-antigen, which is the carbohydrate component of the lipopolysaccharide (LPS). Horizontal gene transfer of the LPS locus between different Leptospira species may be responsible for the presence of serovars with serologically related LPS in various species. Thus, the LPS locus of the antigen-related serovars Hardjobovis and Hardjoprajitno belonging to species L. borgpetersenii and L. interrogans, respectively, are highly similar [42].
Genomic macrorestriction with NotI followed by PFGE is considered to be a powerful typing method for classifying Leptospira strains at the serovar level [29], [30], [31], [43]. For example, classification of the Dadas I strain as a new serovar of serogroup Grippotyphosa was strongly supported by its unique pulsed-field gel electrophoresis pattern [43]. The clinical isolates from Mayotte had PFGE patterns that were different to those for the reference strains belonging to serogroups Mini, Hebdomadis, Pyrogenes, and Grippotyphosa (Text S1). These data were confirmed by MLVA. Non tested reference serovars (not present in our collection of strains) include L. weilii serovar Hekou and L. interrogans serovar Perameles [44] from serogroup Mini, L. noguchi serovar Huanaco from serogroup Grippotyphosa, and L. santarosai serovars Borincana, Goiano, Maru, and Sanmartini and the L. alexanderi serovar Manzhuang from the serogroup Hebdomadis [18]. However none of these serovars belong to the identified Leptospira species (i.e. L. borgpetersenii and L. kirschneri) and may therefore be phylogenetically distant from our clinical isolates.
Our findings suggest that clinical isolates belonging to PFGE patterns I to X may represent ten new serovars. These isolates are highly virulent as observed in experimental animals (LD50<10 leptospires for strain 2007/01203) and are associated with severe clinical forms that could lead to death (2 of the 22 patients). These results highlight the potential existence of several undiscovered Leptospira serovars in the Indian Ocean. Further characterization of these isolates should include the use of the cross agglutination absorption tests (CAAT) to determine whether these isolates correspond to previously unknown serovars [45].
A previous survey in Madagascar found no antibodies to leptospirosis or the presence of pathogenic leptospires in possible animal reservoirs or in humans in contact with these animals [46]. Interestingly, patient 20 (a 52 year-old male) reported swimming in a river in Madagascar 10 days prior the onset of symptoms (and no other risk activities during this period). Since the incubation period of the disease is usually 5–14 days, the relationship between symptoms and the water exposure suggests that this traveler acquired the disease in Madagascar, a country with no prior reports of leptospirosis. The strain (2008/03703) isolated from this patient has been grouped with the L. kirschneri serogroup Mini isolates from Mayotte. However, it does exhibit a unique genotype by PFGE (Table 1).
Since the population in Mayotte is primarily of African origin, the epidemiology of leptospirosis in Mayotte may be the result of the introduction of pathogenic strains coming from the neighboring African countries as in the case of the chikungunya [47] and Rift Valley fever [48] outbreaks in Mayotte. Apart from the serogroups Hebdomadis and Grippotyphosa already reported in Africa, there are no references, to our knowledge, to strains belonging to the serogroup Mini in the African continent.
The high incidence of leptospirosis in Mayotte may be explained by the risk of exposure to infected animals. Contact with ruminants (sheep, cattle, or goats) is frequent. Dogs and rats (Rattus rattus; Norway rats or Rattus norvegicus have never been described in Mayotte) are usually encountered in the household area. Mayotte has also an abundance of endemic fauna such as the maki, a type of lemur, and the roussette (flying fox), a large bat. In 1991, cattle, herd, dogs and the common tenrec (Tenrec ecaudatus), a small spiny insectivorous mammal that resembles a hedgehog also found in Madagascar and other islands of the Indian Ocean, were serologically surveyed (88 serum samples) for leptospirosis using MAT [49]. In total, between 50 and 87% of the samples for each group of animals were seropositive for at least one Leptospira serogroup. The most prevalent Leptospira serogroups were Grippotyphosa, Sejroe, Icterohaemorrhagiae, Pyrogenes, and Canicola. Although MAT testing was unable to detect antibodies to Leptospira in the rat sera, a more recent study did detect leptospira (5 positive samples out of 27) in rat kidneys by PCR (unpublished data). These findings demonstrate that these animals are constantly exposed to Leptospira in their environment. The high prevalence of leptospiral infection in animals represents a potential threat to human health.
In summary, we identified 10 potentially new pathogenic Leptospira genotypes. Further ecological and surveillance studies are needed in Mayotte to identify the reservoir host(s) involved in transmission and to determine the public health impact and distribution of pathogenic leptospires in the region.
|
10.1371/journal.ppat.1004220 | Defining Immune Engagement Thresholds for In Vivo Control of Virus-Driven Lymphoproliferation | Persistent infections are subject to constant surveillance by CD8+ cytotoxic T cells (CTL). Their control should therefore depend on MHC class I-restricted epitope presentation. Many epitopes are described for γ-herpesviruses and form a basis for prospective immunotherapies and vaccines. However the quantitative requirements of in vivo immune control for epitope presentation and recognition remain poorly defined. We used Murid Herpesvirus-4 (MuHV-4) to determine for a latently expressed viral epitope how MHC class-I binding and CTL functional avidity impact on host colonization. Tracking MuHV-4 recombinants that differed only in epitope presentation, we found little latitude for sub-optimal MHC class I binding before immune control failed. By contrast, control remained effective across a wide range of T cell functional avidities. Thus, we could define critical engagement thresholds for the in vivo immune control of virus-driven B cell proliferation.
| Chronic viral infections cause huge morbidity and mortality worldwide. γ-herpesviruses provide an example relevant to all human demographics, causing, inter alia, Hodgkin's disease, Burkitt's lymphoma, Kaposi's Sarcoma, and nasopharyngeal carcinoma. The proliferation of latently infected B cells and their control by CD8+ T cells are central to pathogenesis. Although many viral T cell targets have been identified in vitro, the functional impact of their engagement in vivo remains ill-defined. With the well-established Murid Herpesvirus-4 infection model, we used a range of recombinant viruses to define functional thresholds for the engagement of a latently expressed viral epitope. These data advance significantly our understanding of how the immune system must function to control γ-herpesvirus infection, with implications for vaccination and anti-cancer immunotherapy.
| The gamma-herpesviruses (γHVs) infect >90% of humans and cause diseases including nasopharyngeal carcinoma, African Burkitt's lymphoma and Kaposi's Sarcoma. Their colonization of circulating memory B cells is crucial to persistence and hence to disease ontogeny. Viral latency gene expression in B cells provides an immune target [1] that has been exploited to prevent lymphoproliferative disease in acutely immunodeficient patients by T cell transfer [2]. However, extending this approach to established cancers and developing related vaccines have proved difficult. A significant problem is that the narrow species tropisms of human γHVs severely restrict in vivo analysis, and hence an understanding of how empirical therapies such as adoptive T cell transfer work.
Immune recognition can be assayed in vitro; but while Epstein-Barr virus (EBV) latency gene products drive autonomous B cell proliferation in vitro, most in vivo infected cells are resting memory B cells that have passed though lymphoid germinal centers (GCs) [3]. This makes difficult in vitro analysis of in vivo immune control. One way to make progress is to study related viruses that are experimentally more accessible. Probably the best characterized is Murid Herpesvirus-4 (MuHV-4, archetypal strain MHV-68) [4]–[6]. MuHV-4 is more closely related to the Kaposi's Sarcoma-associated Herpesvirus (KSHV) than to EBV [7]. However it shares many features of host colonization with EBV, for example it exploits lymphoid GCs to establish persistence in circulating memory B cells [8]–[10]. Therefore it can be used to reveal fundamental mechanisms of γHV/host interaction.
MuHV-4 studies have shown that γHV-driven lymphoproliferation occurs in complex lesions incorporating T cell evasion and infected cells with distinct patterns of viral gene expression [10]. In addition to cis-acting T cell evasion during episome maintenance [11], [12], EBV inhibits the transporter associated with antigen processing (TAP) via BNLF2a [13]–[15] and MHC class I export to the cell surface via BILF1 [16], [17]; KSHV degrades MHC class I and other immune receptors via K3 and K5 [18]; and MuHV-4 degrades MHC class I and TAP via MK3 [19]–[21]. Disrupting MK3 impairs virus-driven lymphoproliferation [22].
The γHVs also evade immune recognition during latency by expressing few CTL targets. However a gene that modulates signaling through the B cell receptor - M2 in MuHV-4 [23]–[26], LMP-2A in EBV [27] and K1 in KSHV [28] - is expressed more widely than growth program genes [3], and shows protein sequence diversity [29]–[33] consistent with immune selection. More directly, the presence of an H2Kd binding epitope in M2 [34], [35] significantly reduces long-term MuHV-4 latent loads in BALB/c mice [29]. Therefore despite viral evasion, CTL help to regulate long-term infection [36], [37], and CTL recognition of M2/K1/LMP-2A, which in EBV may extend also to EBNA3A/B/C [38], [39], provides a potential point of attack. LMP-2A is also a candidate vaccine target for nasopharyngeal carcinoma [40]. Thus, how M2/K1/LMP-2A recognition works in vivo is important to understand.
CTL effector capacity broadly correlates with functional avidity, as determined by the capacity of T cell receptor (TcR) engagement to trigger CTL proliferation, cytokine production and target cell lysis at limiting antigen dose [41]. Therefore with limited γHV protein expression during latency, peptide affinity for MHC class I and TcR functional avidity are likely to be crucial for immune control. The diversity of LMP-2A, K1 and M2 prompted us to analyze in vivo the consequences of varying MHC class I binding and TcR functional avidity for a single epitope derived from M2. These parameters affected dramatically the control of virus-driven lymphoproliferation, even in the context of immune evasion. The capacity of MuHV-4 to correlate biochemical interactions with in vivo immune function allowed us to establish quantitative guidelines for infection control.
To understand the CTL recognition requirements for γHV infection control, we expressed from MuHV-4 a well-characterized, H2Kb-restricted epitope comprising amino acid residues 257–264 of ovalbumin (OVA), or APL derivatives (Figure 1A). OVA binds to H2Kb with high affinity (KD = 4.1 nM) [42]. We compared OVA and APL binding by H2Kb stabilization on TAP-deficient RMA/S cells (Figure 1B) [43]. The OVA concentration giving 50% maximal stabilization (EC50) was 40 nM, in close agreement with published data [44]. APLs Q4, V4, G4 and R4 were similar to OVA (EC50 within 2-fold), consistent with residue 4 being solvent-exposed in the H2Kb-peptide complex [45]. E1 required 6-fold more peptide for equivalent H2Kb stabilization, consistent with this residue being only partly exposed; A8, which has a mutated anchor residue, required 10-fold more peptide again; and the control peptide A5A8, with 2 mutated anchor residues, gave no significant stabilization. The H2Kb/OVA/β2M complex has an estimated half-life of 8 h [44]. Its stability is determined primarily by the peptide off-rate, so the E1 complex is likely to have a half-life of approximately 1.3 h.
We assessed the functional avidity of the H2Kb-OVA-specific TcR of OT-I [46] for each APL by ex vivo stimulation of CD8+ T cells from OT-I mice with graded peptide doses (Figure 1C). There was a clear hierarchy in dose-response, with OVA>Q4 (14-fold)>V4 (a further 279-fold)>G4 (53-fold further still), consistent with published data [47]. The R4 antagonist peptide [48], [49] gave no stimulation. As predicted E1 and A8, which have lower MHC class I binding, generated the lowest dose-responses.
We next introduced each epitope at the MuHV-4 M2 C-terminus to ensure expression in latency without compromising M2 function [29]. CTL recognition of an endogenous M2 epitope reduces long-term MuHV-4 latent loads in H2d mice [29]. The lack of an endogenous H2b-restricted M2 epitope therefore allowed us to introduce new targets in a context where this is known to be important. Each recombinant virus was also made with a yellow fluorescent protein (YFP) reporter construct [50] to aid infection tracking (Figure S1). Correct epitope insertion and assembly of the surrounding genome were demonstrated by PCR of plaque-purified viral DNA (Figure 1D). Each recombinant virus showed equivalent in vitro growth (Figure 1E), equivalent lytic replication in the lungs of intranasally (i.n.) infected C57BL/6 mice (Figure 1F) - with peak titers at 4–7 days post-inoculation and clearance by day 11 - and normal latency establishment in H2d BALB/c mice - with equivalent splenic infectious center assay titers 14 days after i.n. inoculation (Figure 1G). Therefore none showed a replication defect independent of H2b-restricted latent epitope expression.
We then tested latency establishment in H2b mice. Infectious center assays (Figure 2A) showed attenuation of any virus with an H2Kb binding epitope attached to M2 (vOVA, vQ4, vV4, vG4, vR4): splenic infection was established at day 11, but then cleared rather than amplified by days 14–21. In contrast, the virus expressing a poorly binding epitope (vA8) was indistinguishable from the epitope-negative wild-type (vWT). Interestingly vE1, which expresses an epitope with 6-fold lower EC50 for H2Kb stabilization (Figure 1B), showed an intermediate phenotype with normal titers at day 11 followed by a gradual reduction.
Not every latently infected cell necessarily reactivates its virus ex vivo. We therefore used PCR of viral DNA at limiting dilution (Figure 2B; Table 1) as a second measure of infected cell frequency. We looked at the peak of latent infection (14 days post-inoculation) and at the steady state (50 days). These results supported the infectious centre assays: vOVA, vQ4, vV4, vG4 and vR4 were all markedly attenuated (>100-fold reduction); vA8 was equivalent to vWT; and vE1 showed an intermediate phenotype, with strongly decreased acute titers but long-term titers close to vA8 and vWT. MuHV-4-specific CTL responses peak at 14–21 days post-infection [51]. Thus a weakly binding latent epitope (E1) allowed some control when CTL responses were at their peak, but not in the long-term when CTL responses decrease in size.
MuHV-4 colonizes multiple cell types in acutely infected spleens. Many are B cells, which change in phenotype as they pass through germinal centers; others are myeloid cells. The main proliferating population is GC B cells, and these also connect most directly to the long-term latency reservoir of resting memory B cells [9], [10]. Therefore to understand better the relationship between acute and long-term viral loads, we measured viral genome prevalence in flow cytometrically sorted GC B cells (Figure 2C; Table 2). They showed marked reductions for vOVA, vQ4, vV4, vG4 and vR4, equivalent frequencies for vA8 and vWT, and intermediate frequencies for vE1. These data were further supported by in situ hybridization for latently expressed viral tRNA/miRNA homologs [29] (Figure 2D), which showed abundant GC infection by vWT and vA8, severely impaired infection by vOVA, vQ4, vV4, vG4 and vR4, and intermediate infection by vE1. Therefore susceptibility to CTL attack during acute lymphoproliferation varied with cell type, and the relative sparing of vE1+ GC B cells appeared to allow high long-term viral loads.
We measured epitope-specific CTL responses with H2Kb-peptide tetramers (Figure 2E) and by staining for intracellular IFN-γ after ex vivo stimulation (Figure 2F). Responses to vA8 were uniformly low despite high viral loads, presumably because this epitope was not produced in sufficient amounts to compensate for its poor H2Kb binding. Responses to vOVA, vQ4, vV4, vG4 and vR4 were detectable, although small compared to those reported for lytic antigens [51]. Surprisingly, the largest CTL response was elicited by the intermediate phenotype virus, vE1. This could not be explained by lytic infection, since this was high in lungs for all viruses (Figure 1F).
We confirmed the functionality of vE1-specific CTL by in vivo killing of CFSE-labelled, peptide-exposed targets (Figure 2G,H): vE1-induced CTL showed target cell elimination comparable to vOVA, whereas mice infected with vWT or vA8 showed none. Therefore the relatively weak H2Kb binding of E1 was sufficient to stimulate large, functional CTL responses, but not for those CTL to curtail efficiently virus-driven lymphoproliferation. This result suggested that at least for vE1, most CTL stimulation comes from a population distinct from that engaged in lymphoproliferation.
The capacity of C57BL/6 mice to control MuHV-4-driven lymphoproliferation through the recognition of latently expressed OVA, Q4, V4, G4 or R4 indicated that the key requirement in a polyclonal TcR setting is the availability of an epitope capable of strong MHC class I binding: T cells from the naive repertoire could recognize either OVA or an APL. However responses to EBV can involve oligoclonal or even monoclonal CTL expansions [52]–[54]. Therefore to understand better the quantitative requirements of TcR functional avidity for in vivo γHV control, we focussed on the well-characterized OT-I TcR (Figure 3).
We first infected OT-I mice with MuHV-4 expressing OVA or APLs with comparable H2Kb binding (Q4, V4, G4, R4), and measured host colonization by infectious center assay of spleens 9 and 11 days later (Figure 3A). vE1 and vA8 were not utilized since they bind MHC class I less efficiently precluding analysis of T cell functional avidity because target concentrations are different. There was a clear correlation between CTL functional avidity (Figure 1C) and in vivo virus control. The antagonist epitope (R4) allowed no control - titers were equivalent to those of the epitope-negative vWT; the others showed a hierarchy of control (OVA>Q4>V4>G4) that matched exactly their hierarchy of functional avidity (and not their minor differences in H2Kb binding). Low titers of pre-formed infectious virus were found in some mice, but generally in proportion to their latent titers, consistent with reactivation of a fixed fraction of the latent viral load; we saw no evidence that M2-associated epitope presentation created a significant new lytic CTL target.
To confirm that the immune control was by CTL, we treated mice with a depleting, CD8-specific mAb from the time of infection (Figure 3B–D). Each virus then reached equivalent titers to the wild-type. While the depletions were highly effective (Figure 3C), they had little effect on the day 11 spleen titers of vWT (Figure 3D). This result was consistent with previous publications [36], [55] and with the lack of known H2b-restricted MuHV-4 latency epitopes. Thus, introducing latent epitope recognition caused new, CD8-dependent virus attenuation in proportion to the functional avidity of that epitope for the dominant TcR.
OT-I mice provided a useful starting point for in vivo analysis of single TcR function. However their limited CD4+ T cell repertoire impairs GC formation and so the ability of MuHV-4 to drive B cell proliferation. Hence, to define the impact of TcR functional avidity in an environment more conducive to lymphoproliferation, we adoptively transferred lymphocytes from Rag-1−/−OT-I mice and purified CD4+ T cells from C57BL/6 mice into TcRα−/− recipients (Figure 4A). Thus the reconstituted mice had polyclonal CD4+ T cells and a TcRαβ+CD8+ T cell compartment of modest size that was restricted to OT-I cells. (Most CD8+ T cells of TcRα−/− mice are TcRγδ+TcRαβ−.) Infecting these with vWT led to a robust proliferation of infected GC B cells (Figure S2 and S3). Infecting them with vOVA elicited a strong OT-I response (Figure 4B) and suppression of splenic colonization (Figure 4C); by contrast vR4, which expressed an antagonist epitope, elicited no OT-I response and reached high titers (Figure 4C). Therefore these mice provided a new and informative window onto how TcR engagement by a latency epitope affects virus-driven lymphoproliferation.
We then infected reconstituted mice with MuHV-4 expressing OVA or APLs (Figure 5). At day 16 post-infection OT-I T cell expansion was greatest for vOVA, reduced for vQ4, reduced further for vV4, and close to background for vG4 and vR4 (Figure 5A). Thus it correlated well with the epitope functional avidity measured in Figure 1C (OVA>Q4>V4>G4>R4). Specifically, the 14-fold avidity reduction of Q4 only modestly reduced CTL cell expansion, and the 4000-fold reduction of V4 caused further reduction but still did not ablate it entirely. The CTL response declined to background only when the avidity was reduced 200,000-fold (G4). Therefore the immune response showed a surprisingly large tolerance for sub-optimal TcR engagement.
Similar results were obtained for OT-I T cell activation (loss of CD62L, Figure 5B). We analyzed CTL function further by intracellular staining for IFN-γ (Figure 5C) and Granzyme B (Figure 5D) after ex vivo stimulation with the corresponding peptide epitope. The responses to vG4 and vR4 were hard to assess due to low CTL numbers; but those to vQ4 and vV4 showed comparable functionality to vOVA. (Note that the peptide concentration used was only just sufficient for maximal stimulation by V4 in Figure 1C) Therefore there was no sign of vQ4 and vV4 eliciting CTL responses that were functionally impaired (or functionally enhanced); they simply elicited responses that were smaller.
Virus titers (Figure 5E) were reduced markedly by OVA expression, only marginally less by Q4, and not significantly by G4 or R4. V4 expression gave an intermediate phenotype, with titers significantly below those of the vWT control and significantly above those of vOVA. The frequencies of viral DNA+ cells in spleens (Figure 5F and Table S1) showed a similar hierarchy (vWT = vG4 = vR4>vV4>vQ4>vOVA). The viral DNA+ frequencies of flow cytometrically sorted GC B cells (Figure 5G and Table 3) showed less discrimination. Nonetheless the trends were similar, and these results were further corroborated by analysis of YFP expression in GC B cells (Figure S4). Therefore high functional avidity (vOVA) gave marked CTL expansion and low virus titers; a 14-fold avidity reduction (vQ4) have remarkably similar results; a 200,000-fold avidity reduction abolished virus control (vG4); and a 4000-fold reduction gave an intermediate phenotype (vV4). OT-I TcR engagement by M2-derived OVA was therefore considerably above the threshold required for in vivo viral control, and low functional avidity compromised viral control via reduced CTL expansion, rather than by differentially affecting CTL effector function.
Gamma-herpesvirus epitope recognition by CTL has been studied extensively [1], [54], but ours is the first quantitative assessment of how epitope/MHC class I/TcR complex formation affects host colonization. Where no latency epitope expression existed, introducing one led to a profound, CTL-dependent suppression of virus-driven lymphoproliferation. This was consistent with the impact of endogenous epitope presentation in H2d mice [29]. The latter affected only long-term viral loads; OVA expression in H2b mice also conferred susceptibility to CTL during acute lymphoproliferation, when trans-acting immune evasion operates [1]. This greater effect of epitope presentation possibly reflected differences in host susceptibility to immune evasion: the MuHV-4 K3 degrades H2Kb relatively poorly [19] and degrades TAP better in H2d than H2b cells [20].
The precise cellular targets for CD8+ T cell recognition of M2-linked epitopes remain unknown. One possibility is proliferating germinal centre B cells, as B cells are a major site of M2 expression [10], [34]. Infected B cells could also be recognized before the onset of proliferation; and as myeloid cells transfer infection to B cells [56], CD8+ T cells could also suppress lymphoproliferation indirectly, by targeting infected myeloid cells [1].
A key point for physiologically relevant epitope presentation is that it conforms to normal latent gene expression. Exogenous promoters such as HCMV IE1 show activity independent of endogenous viral gene expression [57] and this can lead to attenuation [58]. Previous analysis of endogenous M2 epitope [29] established its importance for determining the different long-term latent loads of H2d and H2b mice. Here, to identify presentation thresholds, we made use of the well-characterized SIINFEKL epitope, attaching it to a neutral region of M2 (its C-terminus). This allowed the generation of a very well-defined model epitope with the kinetics and copy number of a known endogenous epitope. Epitope presentation varies with MHC class I genotype. C57BL/6 mice have only 2 MHC class I molecules and appear not to recognize an endogenous M2 epitope. In this context, M2-SIINFEKL illustrated the impact of strong epitope presentation, and wild-type M2 (or M2-vA8) that of poor epitope presentation. The SIINFEKL variants covered the range between, and so allowed us to identify functional recognition thresholds.
Small differences (<1.6-fold) in H2Kb epitope binding had no obvious impact on in vivo CTL efficacy, but a 60-fold reduction abolished protection and a 6-fold reduction showed a partial phenotype. Thus, M2-linked epitope presentation left little room for sub-optimal MHC class I binding. By contrast when H2Kb binding was maintained, reducing TcR functional avidity 14-fold had little effect, reducing it 200,000-fold abolished control, and reducing it 4,000-fold gave an intermediate phenotype. Therefore this aspect of recognition was more flexible even for monoclonal, Rag-1−/− CTL, and a polyclonal population could attack any epitope so long as its MHC class I binding was strong.
In complex viral infections, larger CTL responses are not necessarily more effective responses. These parameters can correlate: MuHV-4 lacking its K3 evasion gene elicits more CTL and achieves lower titers [22]; and our reconstituted mice showed a correlation between more CTL and less virus. But as with latent epitope presentation downstream of ORF73 [11], OVA-specific CTL responses that completely suppressed lymphoproliferation were small compared to lytic epitope responses [51]; and mice infected with vE1 made large epitope-specific responses yet showed poor virus control. We hypothesize that CTL can be stimulated by the key, self-renewing population of infected B cells, when infection is suppressed, but also by infected cells less important to host colonization, when large responses may achieve little. Crucially, viral evasion may make the self-renewing population harder to target. Thus, vE1 showed a strong acute reduction in total viral DNA+ cell frequencies, but relative sparing of GC B cells and consequently high long-term virus loads. A position 1 mutation also impairs the control by Rag-1−/−OT-I mice of MuHV-4 expressing OVA from an HCMV IE1 promoter [59]. However such mice lack B cells or CD4+ T cells, and without CD4+ T cells MuHV-4 causes a lethal, chronic lytic infection even with a strong, polyclonal CTL response [60], [61]. Our reconstituted mice maintained both virus-driven lymphoproliferation and infection control without outgrowth of CTL escape mutants. Thus we could relate directly quantitative changes in epitope recognition to the control of lymphoproliferation.
An important task with EBV is to predict in vivo CTL efficacy. Extrapolating from CTL numbers and in vitro assays alone is clearly problematic. For example, large responses to lytic epitopes in infectious mononucleosis [54] could be interpreted as important, or simply as poor latency epitope recognition when better recognition might preclude large lytic responses and avoid symptoms. The precise relatedness of EBV memory B cell colonization via GCs to MuHV-4 memory B cell colonization via GCs is unknown. But all γHVs have evolved to colonize lymphocytes with maximal efficiency, within limits set ultimately by the immune system, so similar quantitative thresholds would not be surprising. Our data therefore have important general implications for γHV-specific CTL function, and for predicting in vivo CTL efficacy from biochemical measures.
The study accorded with the Portuguese official Veterinary Directorate (Portaria 1005/92), European Guideline 86/609/EEC, and Federation of European Laboratory Animal Science Associations guidelines on laboratory animal welfare. It was approved by the Portuguese official veterinary department for welfare licensing (protocol AEC_2010_017_PS_Rdt_General) and by the IMM Animal Ethics Committee.
CD45.1 C57BL/6, OT-I, Rag-1−/− and TcRα−/− mice were obtained from Jackson Laboratories. CD45.1 Rag-1−/− OT-I mice were obtained by breeding OT-I onto a CD45.1 Rag-1−/− background. C57BL/6 and BALB/c mice were purchased from Charles River Laboratories. All mice were housed under specific pathogen-free conditions at the Instituto de Medicina Molecular and used when 6–12 weeks old. For adoptive transfers to TcRα−/− mice, CD4+ T cells were purified by negative selection from pooled lymph nodes of naïve C57BL/6 mice using the CD4+ T cell isolation kit (Miltenyi Biotech). OT-I T cells were obtained from pooled lymph nodes of naïve CD45.1 Rag-1−/− OT-I mice. 2×106 CD4+ T cells and 106 CD45.1 Rag-1−/− OT-I T cells were adoptively transferred to TcRα−/− recipients via tail vein injection one day prior to infection.
MuHV-4 recombinants were generated from BAC-cloned viral genomes [29]. OVA and APL epitopes were introduced by PCR at the M2 C-terminus. Briefly, the M2 downstream region (genomic co-ordinates 3846-4029) containing a HindIII restriction site followed by the epitope coding region and a stop codon were PCR amplified (Table S2) to attach each epitope to the M2 C-terminus. The PCR products were inserted downstream of a HinDIII/XhoI MuHV-4 genomic fragment (nt 4029–5362) in pSP72 (Promega), using a genomic BglII site (nt 3846) and the engineered HinDIII (nt 4029) restriction site. The constructs were then subcloned into a HinDIII-E MuHV-4 genomic fragment in the pST76K-SR shuttle plasmid, using genomic BlnI (nt 3908) and XhoI (nt 5362) restriction sites. All PCR-derived regions were sequenced to confirm the integrity of the introduced epitopes and the M2 flanking region. Each recombinant HinDIII-E shuttle plasmid was transformed into E.coli carrying the wild type MuHV-4 BAC (pHA3) or a YFP+ BAC [50] obtained from Dr Samuel Speck (Emory Vaccine Center, Atlanta). Following multi-step selection, recombinant BAC clones were identified by restriction digestion with HinDIII. The integrity of each BAC was confirmed by digestion with BamHI and EcoRI. All viruses were reconstituted by transfecting BAC DNA into BHK-21 cells using FuGENE 6 or X-tremeGENE HP (Roche Applied Science). The loxP-flanked BAC cassette was then removed by viral passage through NIH-3T3-CRE cells and limiting dilution cloning. The integrity of each reconstituted virus was checked by PCR of viral DNA across the HinDIII-E region and DNA sequencing across M2.
Murine RMA/S cells were cultured in RPMI 1640 with 10% fetal calf serum, 2 mM glutamine and 100 U/ml penicillin and 100 µg/ml streptomycin. NIH-3T3 (ATCC)-CRE cells [22] were grown in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal calf serum, 2 mM glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Baby hamster kidney fibroblast cells (BHK-21, ATCC) were cultured in Glasgow's modified Eagle's medium (GMEM) supplemented as above plus 10% tryptose phosphate broth. To prepare viral stocks, low multiplicity infections (0.001 PFU per cell) of NIH-3T3-CRE or BHK-21 cells were harvested after 4 days and titrated by plaque assay [29].
H2Kb stabilization was determined with TAP-deficient RMA/S cells. These were incubated overnight at 26°C to promote the export of empty H2Kb complexes, then loaded with graded concentrations of OVA or APL peptides (Thermo Scientific) for 2 h at 26°C and subsequently transferred to 37°C for 2 h to destabilize empty MHC molecules [43]. The cells were then washed twice, stained with anti-H2Kb (AF6-88.5.5.3, eBioscience), and analysed on a LSR Fortessa (BD Biosciences). Mean fluorescence intensities were determined with FlowJo (Tree Star). To measure the ex vivo stimulation of naïve OT-I T cells by OVA and APLs, CD8+ T cells from the spleens of naïve OT-I mice were purified by negative selection (CD8+ T cell isolation kit, Miltenyi Biotech); for equivalent peptide/MHC class I numbers, irradiated (7500 rads) RMA/S cells were loaded with different peptides at 26°C, then incubated at 37°C; and 5×104 OT-I T cells were cultured with 2.5×104 RMA/S cells for 72 h at 37°C. IFNγ levels in culture supernatants were measured by ELISA (DuoSet ELISA development kit, R&D Systems). The data were fitted to sigmoidal dose-response curves and EC50 values calculated using GraphPad Prism.
Groups of 6- to 8-week old BALB/c and C57BL/6 mice were inoculated i.n. with 104 PFU of MuHV-4. 8- to 12-week old OT-I and TCRα−/− mice were inoculated i.n. with 103 PFU of MuHV-4. All virus inoculations were in 20 µl of PBS under isofluorane anaesthesia. At different days post-infection lungs or spleens were removed and processed for subsequent analysis. Titres of infectious virus were determined by plaque assay of freeze-thawed lung or spleen homogenates using BHK-21 cells. Latent virus loads were quantified by explant co-culture of splenocytes with BHK-21 cells. Plates were incubated for 4 (plaque assay) or 5 (explant co-culture assay) days, then fixed with 4% formaldehyde and stained with 0.1% toluidine blue. Viral plaques were counted with a plate microscope. The frequency of MuHV-4 genome-positive cells was determined by limiting dilution combined with real time PCR [10]. Splenocytes were pooled from 4–5 mice. GC B cells (CD19+CD95hiGL7hi) were purified from pools of 4 or 5 spleens using a BD FACSAria Flow Cytometer (BD Biosciences). Cells were serially two-fold diluted and eight replicates of each dilution were analysed by real time PCR (Rotor Gene 6000, Corbett Life Science). The primer/probe sets were specific for the MuHV-4 M9 gene (5′ primer: GCCACGGTGGCCCTCTA; 3′ primer: CAGGCCTCCCTCCCTTTG; probe: 6-FAM-CTTCTGTTGATCTTCC-MGB). Samples were subjected to a melting step of 95°C for 10 min followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. Real-time PCR data was analysed on the Rotor Gene 6000 software. The purity of sorted populations was always >96%. In situ hybridization with a digoxigenin-labelled riboprobe encompassing MuHV-4 vtRNAs 1–4 and microRNAs 1–6 was performed on formalin-fixed, paraffin-embedded spleen sections [29], using probes generated by T7 transcription of pEH1.4.
Splenocytes from naïve CD45.1 C57BL/6 mice were used as targets and controls. Targets were pulsed with 1 µM OVA, E1 or A8 peptides for 1 h at 37°C, then labeled with 1 µM carboxyfluorescein succinimidyl ester (CFSE) (Molecular Probes). Controls were left unpulsed and labeled with 0.1 µM CFSE. Cells were washed three times then injected intravenously as a 50∶50 mix of CFSEhi and CFSElo cells (4×106) into mice infected with vWT, vOVA, vE1 or vA8. The same mixes were injected intravenously into vWT infected C57BL/6 controls to ensure equal transfer. On the next day splenocytes were harvested and the proportion of CFSEhi and CFSElo cells among CD45.1 splenocytes was analysed by FACS. Target cell killing was calculated as (% CFSElo/% CFSEhi), with % = 100−(ratio in vWT infected/ratio in vOVA, vE1 or vA8 infected)×100.
MuHV-4 infected OT-I mice were depleted of CD8+ T cells by 5 intraperitoneal injections of 200 µg monoclonal antibody YTS 169.4. Splenocytes from control or depleted mice were stained with anti-CD8α (53-6.7) (BD Pharmingen) and analysed on a LSR Fortessa (BD Biosciences).
Splenocytes (2×106) from infected mice were stimulated for 5 h at 37°C with 10 µg/ml peptide (OVA, APLs or VSV NP52-59) in RPMI 1640/10% fetal calf serum/2 mM glutamine/100 U/ml penicillin/100 µg/ml streptomycin/50 µM 2-mercaptoethanol/10 U/ml recombinant murine IL-2 (PeproTech)/10 µg/ml Brefeldin A. Cells were then washed, blocked with anti-CD16/32 (2.4G2) (BD Pharmingen), surface stained with anti-CD8α ± anti-CD45.1 (for OT-I T cells), fixed and permeabilized with Foxp3 staining buffer (eBioscience) and stained with anti-IFNγ (XMG1.2) (BD Pharmingen), anti-Granzyme B (NGZB) or anti-IgG2ak Isotype control (eBioscience). Samples were analysed on a LSR Fortessa (BD Biosciences).
Splenocytes were treated with red blood cell lysis buffer, blocked with anti-CD16/32 (2.4G2, BD Pharmingen, 10 min), and stained at 4°C in PBS/2% FCS 30 minutes: anti-CD95 (Jo2), anti-CD19 (1D3), anti-CD8α (53-6.7), anti-IFNγ (XMG1.2) (BD Pharmingen); anti-CD45.1 (A20), anti-CD45.2 (104), anti-CD44 (IM7), anti-CD62L (MEL-14) (Biolegend); anti-GL7 (GL7), anti-H2Kb (AF6-88.5.5.3), anti-TCRβ (H57-597), anti-GranzymeB (NGZB), anti-IgG2ak Iso control (eBR2a) (eBioscience). For biotinylated antibodies, an additional 20 minutes incubation with streptavidin was performed. MuHV-4 infected cells were identified by YFP expression. H2Kb tetramers conjugated to PE were a kind gift from Dr Hidde L. Ploegh (Whitehead Institute for Biomedical Research, Massachusetts Institute of Technology, Cambridge). Conditional ligand was exchanged for SIINFEKL (OVA), SIIQFEKL (Q4), SIIVFEKL (V4), SIIGFEKL (G4), SIIRFEKL (R4), EIINFEKL (E1) or RGYVYQGL (VSV NP52-59) peptides (Thermo Scientific). Streptavidin-APC or -PerCP (BD Pharmingen) was used to reveal biotinylated antibodies. Samples were acquired on a LSR Fortessa using DIVA (BD Biosciences) and analysed with FlowJo (Tree Star, Inc.).
Data comparisons between groups were performed by an unpaired two-tailed t-test or ordinary one-way ANOVA as appropriate. Mean +/− SEM and statistics were calculated with GraphPad Prism Software. For limiting dilution analysis 95% confidence intervals were determined as described [10].
Primers used for attaching each epitope to MuHV-4 M2 C-terminus are detailed in supplemental Table S2.
|
10.1371/journal.pcbi.1001030 | Spatial Pattern Switching Enables Cyclic Evolution in Spatial Epidemics | Infectious diseases often spread as spatial epidemic outbreak waves. A number of model studies have shown that such spatial pattern formation can have important consequences for the evolution of pathogens. Here, we show that such spatial patterns can cause cyclic evolutionary dynamics in selection for the length of the infectious period. The necessary reversal in the direction of selection is enabled by a qualitative change in the spatial pattern from epidemic waves to irregular local outbreaks. The spatial patterns are an emergent property of the epidemic system, and they are robust against changes in specific model assumptions. Our results indicate that emergent spatial patterns can act as a rich source for complexity in pathogen evolution.
| Parasites are commonly believed to evolve to make as many infections as possible. In large scale simulations of disease spread, however, natural selection can instead act to maximize outbreak frequency. Here, pathogens that cause short infections can be rewarded for their prudence by a rapid subsequent outbreak. Very surprisingly, in a simple spatial model for disease spread so-called evolutionary cycling can be observed, that is, the infection period indefinitely keeps evolving up and down. The reversal in the direction of selection is triggered by a switch in the spatial patterns from epidemic waves to irregular local outbreaks. This finding offers an alternative mechanism to explain why infectious diseases show so much variation in the lengths of their infectious periods.
| Recent studies show that in spatial models, evolutionary dynamics of infectious diseases change with respect to predictions from non-spatial model, which implicitly assume complete mixing of individuals [1]–[5]. Spatial epidemic waves have been observed for several diseases [6]–[9]. Starting with Rand et al. [10], a number of authors have analyzed a spatial pathogen-host model in which a host population with local reproduction is infected by a lethal pathogen that is transmitted through direct local contact [4], [11]–[13]. In this class of models, local extinction of host populations caused by the pathogen is balanced by re-colonization of empty space by reproduction of uninfected hosts. Evolution towards increased transmissibility is limited, because pathogens that are too infectious exhaust their local host population before enough new hosts are born for the pathogen to persist.
Spatial pattern formation can also act to limit the duration of the infectious period. We recently examined a spatial epidemic model in which infection of hosts leads to waning immunity instead of host death [14], and we found that, after the system self-organizes into epidemic waves, natural selection is directed toward increasing outbreak frequency. Outbreak frequency is optimal for infections of relatively short duration so that pathogens evolve towards short lasting infections. The mechanistic explanation for competition for frequency between waves is that, when two infection waves collide, this typically is followed by local extinction of the pathogen. Subsequently, the pathogen with the higher outbreak frequency will be the first to reinvade the susceptible host population. Frequency selection also occurs in spatial competition in chemical reactions [15], autocatalytic hypercycles [16] and parasitoid-host systems [1], [17].
Our previous analysis [14] adopted a standard evolutionary approach in that ecological dynamics were operating on a faster timescale than mutation and evolution. However, for many pathogens, in particular viruses and bacteria, this assumption need not be true, as even within a single host the pathogen can accumulate many mutants as a by-product of the large mutation rate, large copy numbers, and short generation time [18], [19]. In this paper we investigate the effects of increasing mutation rate in our spatial disease model (see Model section for a description of the model). It will turn out that co-occurrence of pathogens with a substantial difference in infection period can trigger a novel type of spatial patterns, which can temporarily reverse the selection pressure in the system. As a result, the model exhibits so-called evolutionary cycling in the length of the infection period.
In our spatial epidemic model, hosts are situated in a regular square contact network. Hosts can be in three states: susceptible (S), infected (I), or resistant to infection (R). There are three possible state transitions, namely infection, acquiring resistance, and loss of resistance. Infection is a local contact process, where each infected host can infect its eight direct neighbors, and this occurs at rate β. Acquiring resistance occurs after a fixed time τi since infection (i.e. τi can be interpreted as the duration of the infectious period). Loss of resistance occurs after a fixed duration τr. Note that only infection is a spatial process. Because the total population size of hosts is constant, infection is both density and frequency dependent. All time units are relative to the resistant period (which is set at τr = 1). In this paper, we investigate evolution of the length of the infectious period. Upon each new infection, the infectious period can, with equal probability, increase or decrease with a fixed mutation size step ±Δτi with mutation probability μ. As a consequence, strains with different infectious period will co-occur and compete for susceptible hosts. We assume full cross-resistance, and no co-infection or super-infection between strains, so that after infection by a particular strain a host cannot be infected by any other strain.
In our earlier paper [14], we used a probabilistic cellular automaton model with a small fixed time step. We tested decreasing the time step until the behavior approached a continuous time process. However, such a method requires quite small time steps, which implies a large waste of computational power, as most time is used for updating hosts that do nothing within the time step. Instead, here we adopt a much faster continuous time method by using an event based updating procedure. For each infected individual, we schedule a next infection event, using an exponential distribution with rate parameter β. A scheduled infection event will only cause a new infection if, at the time of the event, the individual is still infectious, and if a randomly chosen neighbor individual is susceptible to infection. Acquiring resistance and loss of resistance of individuals are also scheduled as events. All events are efficiently managed in a binary heap data structure [20]. The next scheduled event is always located at the top of the binary tree, and updating the tree is an efficient process (scaling with the logarithm of the number of events). The method is vey flexible, as e.g. other probability distributions for the infectious period, such as lognormal or gamma distributions, can be implemented without loss of computational speed. Furthermore, using larger neighborhoods for the infection process does not substantially increase the computation time, as the neighborhood is only evaluated at the time of a potential infection. A non-spatial analogue of the model can be investigated by implementing infection at a global scale, that is, an infected individual can infect any other individual in the system with equal probability, thus effectively removing the spatial component from the model. A deterministic continuous approximation of the mean field dynamics [21] would consist of a set of three (SIR type) delay differential equations. Typical single run simulations of the spatial model, as reported in this paper, take around one hour to run on a Pentium computer. For large simulations several nodes of the Dutch national computer cluster LISA were used overnight. A C-code for the updating procedure can be obtained from the corresponding author upon request.
In Fig. 1A through 1C (for movies see supplemental Videos S1, S2, S3), we investigate pair-wise competition between three pathogen genotypes that differ in infection period, with infection periods of τI = 0.4, τI = 0.6 and τI = 0.8, respectively. We observe that these three pathogen types have a cyclical dominance structure, where each infection period can outcompete exactly one other infection period, resulting in a so-called “rock-scissor-paper” dynamics. In each pair-wise competition two distinct genotypes are initialized each on one side of a 201×100 field. Infection and resistance is randomly initialized in large blocks, in order to speed up the pattern formation. The middle vertical row is kept susceptible for the first 10 time units, so that the patterns can fully develop before the competition is started. Fig. 1A, τI = 0.4 against τI = 0.6, and Fig. 1B, τI = 0.6 against τI = 0.8, both show that the genotype with the shorter infection period wins the competition. This outcome is surprising, as a shorter infection period is a disadvantage at an individual level, as it generates fewer secondary infections. This selection for shorter infection period was also reported in our previous paper [14], and it can be explained by selection for higher outbreak frequency. At the spatial location where the two genotypes co-occur, the disease locally goes extinct after each wave of infection. The genotype that subsequently can cause the next disease outbreak faster will reinvade the area more quickly and it will gradually increase its domain of dominance. This selection for shorter infection period depends on the spatial pattern formation, and particularly on the epidemic waves that cause local extinction of the disease. In the non-spatial analogue of our model a shorter infection period is always a competitive disadvantage, as it locally generates less secondary infections per infected individual (i.e. it has a reduced reproduction number R0; see ref. [22]). The spatial patterns cancel this selection at the local individual level, because after each epidemic wave both disease genotypes locally will go extinct.
However, in Fig. 1C, for a two-fold difference in infection period of τI = 0.4 against τI = 0.8, the genotype with the longer infection period wins. This reversal of selection is accompanied by a local change in the spatial pattern; that is, at the interface between the two genotypes the dynamics now switches to a fine-scaled irregular outbreak pattern instead of regular epidemic waves. Here, the epidemic waves from the fast genotype run into remaining resistance fragments of the slow genotype. As a consequence, the waves break up and become irregular, and, most importantly, the disease no longer locally goes extinct (Fig. 1D). In the absence of local extinction, selection will favor the disease genotype with the longer infection period, because it locally generates more secondary infections (i.e. it has a larger R0, see ref. [20]).
Summing up, we have two opposite directions of selection in our system, depending on the spatial pattern at the competition interface. When the difference between infection periods is relatively small, the spatial pattern consists of epidemic waves, and both genotypes will go extinct in between successive epidemic waves. In this spatial regime, selection is for increased outbreak frequency, because the higher frequency genotype will be able to reinvade the area more quickly. In contrast, when the difference between infection periods is large, the spatial pattern at the competition interface will be irregular (i.e. turbulent), and local extinction of both genotypes between successive epidemic waves does no longer occur. In this spatial regime, selection is for increased infection period, because the genotype with longer infection period will be able to locally cause more secondary infections. In a way, the situation is reminiscent of a life history trade-off, or in particular “r” versus “K” selection theory [23], [24]. Here, the epidemic waves create an unstable “r” strategy environment, selecting for fast regeneration and growth, whereas the local irregular outbreak pattern creates a (relatively) stable “K” strategy environment, selecting for increased local competition strength. In the next paragraph we will quantify the exact boundaries for these two modes of selection, depending on the infection periods of both competitors.
In Fig. 2A, a pair-wise competition plot is shown for infection periods up to τI = 1.0. For small differences in infection period, there is selection for increased outbreak frequency, leading to an evolutionary stable attractor (ESS) around τi = 0.2 (i.e. in the ESS, R0 = τI β = 6.4). However, if the difference in infection period is large, the fine-scaled local irregular outbreak pattern develops, and selection favors the genotype with the longer infection period. As a consequence, for a small mutation rate of μ = 0.01, an evolving population will converge to the ESS, because at any time the difference between competing genotypes will be small (Fig. 2B, dotted line). However, for a larger mutation rate of μ = 0.05, the evolutionary dynamics are qualitatively different, and they converge to large amplitude cycling of the infection period (Fig. 2B, solid line). In the spatial pattern of the cyclic evolutionary attractor (Fig. 3A) it appears that distinct areas differ substantially in infection period. In Fig. 3B, after 20 time units, the areas of irregular outbreaks have invaded the nearby epidemic wave areas, whereas the other areas have decreased in infection period (see supplemental Video S4). In Fig. 3C the local direction of selection for this 20 time unit period is plotted, showing a strong dependence of the direction of selection on the local spatial pattern. In areas of irregular outbreaks the infection period increases whereas in areas of epidemic waves it decreases over time.
When the mutation rate is slowly increased, the evolutionary dynamics show a sudden shift from the ESS to the cyclic evolutionary attractor (Fig. 4A). For increasing mutation rate, the distribution of mutants around the ESS gradually widens (due to quasispecies dynamics, see ref. [19]), until the first local irregular outbreaks can develop when neighboring genotypes differ enough in infection period. Hereafter, the dynamics quickly shifts to the cyclic attractor. If we now gradually reduce the mutation rate (Fig. 4B), it turns out that the cyclic evolutionary attractor can be maintained for quite small mutation rates, before the system falls back to the ESS. There is thus a considerable region, for 0.027≤μ≤0.036, where the system displays bi-stability between the ESS and the cyclic evolutionary attractor. This bi-stability can be explained by the fact that the cyclic evolutionary attractor reinforces itself, as it is associated with large differences in infection period across the field which induces the irregular outbreak pattern. It should be noted that, due to the stochastic nature of the model, the hysteresis is only a transient phenomenon; that is, allowing long enough simulation time, within the bi-stable region the dynamics will spend time in both alternative attractors. Increasing the system size will, within the bi-stable region, promote the cyclic evolutionary attractor, as the first irregular outbreaks can develop anywhere in the spatial domain, and increasing the size of the domain will thus increase the chance of the irregular outbreak pattern to develop. Once the irregular outbreak pattern has developed somewhere in the field, it will expand to the rest of the system. Also, on a small spatial domain, the cyclic evolutionary attractor is more easily lost, because persistence of the attractor depends on different spatial regions being in different phases of the attractor. On a small field global synchronization occurs more easily, and this will result in the system falling back to the ESS (i.e. τi = 0.2) attractor.
In the previous section, the first local irregular outbreak pattern is generated by increasing the mutation rate. One could argue that the mutation rate that is necessary to generate this pattern is quite large, and hence the evolutionary cycling might be considered unlikely to play a role in real epidemics. However, it turns out that the irregular outbreak pattern can also originate from various other sources. For instance, cyclic evolutionary dynamics can be observed for a small mutation rate of μ = 0.001 (or even smaller) if only a small region of the field has some variance in the length of the resistant period. In Fig. 5 such a small inhomogeneous area is introduced in the middle of the field. Within this inhomogeneous region, the disease will develop a small scale spatial pattern in which there is no local extinction. As a consequence, within this region, the disease will evolve to maximum length of the infection period (Fig. 5, dotted line). Subsequently, the long infection periods in middle of the field will seed the irregular outbreak pattern and the resulting evolutionary cyclic dynamics in the rest of the field (Fig. 5, solid line). The mutation rate in this case will set the timescale of the evolutionary cycle, but even at very small mutation rates the evolutionary cycling dynamics persist. The evolutionary cycling here is much more regular than in Fig. 2B, because now the maximal infection period remains continuously present in the system, whereas before it occasionally was lost and had to reemerge from de novo mutation.
The cyclic evolutionary attractor can also be induced for small mutation rates if some local movement of individuals is included in the model, or if occasional large effect mutations are considered (results not shown). Furthermore, the occurrence of the cyclic evolutionary attractor does not depend on the specific assumptions of the model. We imposed a maximum to the length of the infection period, but such an upper limit could also be obtained implicitly by e.g. implementing a trade-off between the length of the infection period and the infectivity of a genotype. We extensively tested robustness of our results against various model assumptions, for instance by changing the contact network topology (using 4 or 6 infection neighbors), changing the boundary conditions (reflecting boundaries and empty boundary conditions), and testing various statistical distributions for the duration of the infectious period and the resistant period (e.g. log-normal and gamma distributions). The key property that is necessary for the selection for shorter infection periods to occur is recurrent local extinction of the disease. This coincides with a local dynamics that is unstable (i.e. the mean field dynamics converges to large amplitude oscillations or even extinction), inducing the spatio-temporal pattern of epidemic waves. If instead, the local (and mean field) dynamics of the disease converges to a stable endemic equilibrium, selection will favor genotypes with maximal infection period.
In the spatial epidemic model we present here, the local oscillating dynamics give rise to epidemic waves and recurrent local extinction of the disease. These spatial patterns enable selection for a reduced infection period, as this will increase the outbreak frequency. However, the system can switch to a local fine-scaled irregular outbreak pattern, which emerges at the interface between pathogens with large enough difference in infection period. Here, local extinction of the disease no longer occurs, and selection is in the direction of increased infection period. These opposite directions of selection can lead to large amplitude cyclic evolutionary dynamics.
Cyclic evolution was reported before for non-spatial systems [25]–[27]. It can result from co-evolution between species or traits. In contrast, in the model we present here, we observe cyclic dynamics in a single evolving trait, namely the length of the infection period. In our system, a change in the spatial pattern acts as a switch for the direction of selection. Interestingly, both alternative spatial patterns cause selection in a direction that promotes the switch to the other pattern, resulting in a continual cycling between the two patterns and directions of selection. Alternative spatial patterns can also act to induce bi-stability, in the case that each of the two patterns causes a direction of selection that enforces the current pattern [4], [17]. Note that the observed cyclic evolutionary dynamics cannot be explained with currently popular spatial analytic tools, such as spatial adaptive dynamics [28], spatial moment approximation techniques [29] or spatial inclusive fitness measurements [30]. The fine-scaled irregular outbreak pattern is only a transient pattern that occurs at the interface between genotypes, and the local selection always favors the disease genotype with longer infection period. However, as we demonstrated, this local selection can be overruled by local extinction and recolonization, which can cause selection to act in an opposite direction. In this paper, we use numerical computational methods, such as constructing the pair-wise competition plot of Fig. 2A. We feel that the current emphasis on spatial approximation techniques acts to underestimate the potential (non-linear) effects of spatial patterns on disease dynamics. In particular, spatial patterns can undergo sudden changes, such as the transition from epidemic waves to local irregular outbreaks that is observed in this paper, and such transitions are often hard (or even impossible) to predict with the current approximation techniques. Maybe these analytic tools can be improved to include, or at least predict, such transitions; for instance by developing ‘early warning signals’ for these shifts in spatial pattern [31], but this is not easy. In the meantime, we want to advocate the complementary use of numerical computational methods that incorporate the full non-linearity of the system, as such methods can increase both our qualitative and quantitative understanding of the impact of spatial pattern formation on disease dynamics.
Spatial epidemic waves have been reported for many infectious diseases [6]–[9], and recurrent local extinction is occurring for many epidemic diseases [32]. We have shown that these spatio-temporal patterns can have profound effects on selection for disease properties. Most notably, epidemic waves and recurrent local extinction can induce selection for a short infection period, which is a property of many diseases, and which is otherwise hard to explain without invoking strong trade-off assumptions [33]–[35]. We want to emphasize that the spatial patterns can induce selection for properties that emerge at a scale beyond that of the individual or the direct neighborhood of individuals [36]. Selection for increased outbreak frequency in epidemic waves is an intriguing example where a property that appears at a scale of outbreak centers overrides the local individual selection for maximizing secondary infections.
|
10.1371/journal.pgen.1004988 | Prodomain Removal Enables Neto to Stabilize Glutamate Receptors at the Drosophila Neuromuscular Junction | Stabilization of neurotransmitter receptors at postsynaptic specializations is a key step in the assembly of functional synapses. Drosophila Neto (Neuropillin and Tolloid-like protein) is an essential auxiliary subunit of ionotropic glutamate receptor (iGluR) complexes required for the iGluRs clustering at the neuromuscular junction (NMJ). Here we show that optimal levels of Neto are crucial for stabilization of iGluRs at synaptic sites and proper NMJ development. Genetic manipulations of Neto levels shifted iGluRs distribution to extrajunctional locations. Perturbations in Neto levels also produced small NMJs with reduced synaptic transmission, but only Neto-depleted NMJs showed diminished postsynaptic components. Drosophila Neto contains an inhibitory prodomain that is processed by Furin1-mediated limited proteolysis. neto null mutants rescued with a Neto variant that cannot be processed have severely impaired NMJs and reduced iGluRs synaptic clusters. Unprocessed Neto retains the ability to engage iGluRs in vivo and to form complexes with normal synaptic transmission. However, Neto prodomain must be removed to enable iGluRs synaptic stabilization and proper postsynaptic differentiation.
| Synapse development is initiated by genetic programs, but is coordinated by neuronal activity, by communication between the pre- and postsynaptic compartments, and by cellular signals that integrate the status of the whole organisms and its developmental progression. The molecular mechanisms underlining these processes are poorly understood. In particular, how neurotransmitter receptors are recruited and stabilized at central synapses remain the subject of intense research. The Drosophila NMJ is a glutamatergic synapse similar in composition and physiology with mammalian central excitatory synapses. Like mammals, Drosophila utilizes auxiliary subunit(s) to modulate the formation and function of glutamatergic synapses. We have previously reported that Neto is an auxiliary protein essential for functional glutamate receptors and for organization of postsynaptic specializations. Here we report that synapse assembly and NMJ development are exquisitely sensitive to postsynaptic Neto levels. Furthermore, we show that Neto activity is controlled by Furin-type proteases, which regulate the processing and maturation of many developmentally important proteins, from growth factors and neuropeptides to extracellular matrix components. Such concerted control may serve to coordinate synapse assembly with synapse growth and developmental progression.
| Synapse development is a highly orchestrated process that enables proper establishment of neural circuits and development of the nervous system. Crucial to synapse assembly is the recruitment and stabilization of neurotransmitter receptor complexes at synaptic sites [1]. Receptor complexes can be inserted directly into synaptic membranes via vesicular trafficking from ER-Golgi network, or they can move into the synaptic regions by lateral diffusion from extrasynaptic pools (reviewed in [2,3]). Clustering of neurotransmitter receptors at new synapses induces expression of synaptic components and assembly of postsynaptic structures, such as postsynaptic densities (PSDs), which in turn help maintain the local density of receptors [4]. Neural activity and trans-synaptic communication between pre- and postsynaptic specializations together with intracellular signals within the synaptic partners themselves ensure the maturation, refinement and plasticity of the synaptic connections and synapse growth [5–9]. The molecular mechanisms that coordinate the recruitment and stabilization of receptors at synaptic sites and assembly of synaptic structures with synaptic growth remain unclear.
The Drosophila NMJ provides an ideal genetic system to examine the mechanisms that couple synapse assembly with synapse growth and development. The fly NMJ is a glutamatergic synapse similar in composition and physiology to vertebrate AMPA/kainate central synapses [10,11]. The fly NMJ iGluRs are tetrameric complexes composed of three essential subunits, GluRIIC, GluRIID and GluRIIE, absolutely required for assembling functional channels [12–14]. The fourth subunit can be either GluRIIA (type-A channels) or GluRIIB (type-B) [15–17]. GluRIIA and GluRIIB compete for the essential subunits, which are limiting for the formation of functional receptors. Before a muscle is innervated, low levels of iGluRs are present diffusely in the muscle membrane. Innervation triggers the clustering of iGluRs at synaptic locations and postsynaptic differentiation [18–20]. Type-A channels are the first to arrive at nascent synapses, while type-B, which desensitize ten times faster than type-A, mark more mature synapses [12,20,21]. The fly NMJ iGluRs, but not other PSD components, show very little turnover suggesting that the iGluR complexes are stably incorporated at synaptic sites [22].
At the Drosophila NMJ, clustering of iGluRs and formation of postsynaptic specializations requires an additional essential protein, Neto [23]. Neto belongs to a family of highly conserved transmembrane proteins sharing an ancestral role in the formation and modulation of glutamatergic synapses [24–26]. Vertebrate Neto proteins (Neto1 and 2) and C. elegans Neto/Sol-2 have emerged as auxiliary subunits that modulate the gating properties of AMPA/kainate-type channels and their synaptic localization without influencing their delivery to the cell surface [24–29]. Likewise, Drosophila Neto associates with iGluRs in vivo and controls their trafficking and clustering at NMJ synapses without affecting their muscle expression levels [23]. Reduced synaptic iGluRs alter the function of NMJs causing locomotor defects and reduced synaptic transmission [12,30]. Lack of junctional iGluRs also induces a cascade of defects in the assembly and maintenance of postsynaptic specializations [23,30]. For example, Neto- or iGluRs-deprived synapses have reduced accumulation of PSD components, such as p21-activated kinase (PAK), and sparse subsynaptic reticulum (SSR), a structure comprised of stacks of muscle membranes surrounding and stabilizing synaptic boutons [31].
Intriguingly, synapses developing at suboptimal Neto/iGluR levels share a number of morphological and physiological defects with mutants in the BMP signaling, a pathway that controls the NMJ growth and confers synaptic homeostasis [32]. Similar to neto mutants, BMP mutant NMJs have fewer boutons and reduced excitatory junctional potential (EJP) (reviewed in [32]). Furthermore, Neto in complex with type-A receptors promote the phosphorylation and accumulation of the BMP pathway effector Mad at synaptic locations [33]. The BMP-type signaling factors are produced as inactive precursors, with inhibitory prodomains that must be removed by proprotein convertases to generate the active ligands [34]. Furin-type proteases control the limited proteolysis of inactive BMP precursors and directly regulate their activities [35–37]. In many tissues, sequential processing of BMP prodomains modulates the range and signaling activities of BMP ligands [38]. At the Drosophila NMJ additional TGF-β factors regulate the expression of Glass bottom boat (Gbb), a BMP7 homolog required for the BMP retrograde signaling [39,40]. Furin-type proteases activate all these TGF-β-type factors as well as the BMP-1/Tolloid enzymes that augment TGF-β signaling indicating that Furins provide an important means for controlling cellular signaling at the Drosophila NMJ.
Here we report that Neto protein levels are critical for synaptic trafficking and clustering of iGluRs. Excess or reduced Neto protein in the striated muscle induced formation of NMJs with reduced number of synaptic boutons, decreased synaptic iGluRs and diminished neurotransmission. Neto activities are regulated by Furin-mediated proteolysis and removal of an inhibitory prodomain. In the absence of prodomain cleavage, Neto engages the iGluRs but fails to promote their recruitment and stable incorporation at synaptic sites and to initiate postsynaptic differentiation. Since Furins also cleave and activate signaling molecules, such as TGF-β factors, Furins may synchronize the processing of Neto and TGF-β to control synaptic growth.
Similar to neto hypomorphs, RNAi-mediated knockdown of Neto in the striated muscle altered NMJ development (Fig. 1A, B and [23,33]). Interestingly, neto overexpression in the muscle also induced abnormal synapse development. We rescued neto null mutants (neto36) with neto transgenes with various expression levels and found that excess Neto accumulated at NMJ synapses and extrajunctional locations in a dose-dependent manner (Fig. 1A, C). Low to moderate levels of Neto clustered at synaptic sites (i.e. using neto-A9 transgene), but excess Neto (neto-A3, or neto-A1 for the highest level) had predominantly diffuse distribution with fewer individual synaptic puncta and abundant extrasynaptic signals. Similar patterns were found in animals with overexpressed neto transgene (where neto-A1 induced the strongest phenotypes). Excess Neto had detrimental effects on the viability of rescued animals at all stages of development (S1 Fig.). Neto levels also affected NMJ growth. In larvae with either reduced or excess Neto, the number of boutons was decreased although the branching patterns differed: longer branches at reduced Neto and shorter branches at excess Neto (Figs 1A, S1). This suggests that independent signaling pathways control NMJ growth and bouton formation.
To examine the effects of Neto levels on synapse function we recorded excitatory junction potentials (EJPs) and spontaneous miniature potentials (mEJPs, or minis) from muscle 6 of third instar larvae (Fig. 1D-F). In control larvae (Dicer/+: 24B-Gal4/+) minis occurred two times per second on average. This was reduced to 0.3 events per second at Neto-depleted synapses (dicer; 24B>netoRNAi) similar to that observed in neto hypomorphs [23]. Excess Neto showed a reduction in mini frequency, and to a lesser extent in mini amplitude, but only when Neto was expressed at very high levels; larvae with moderate levels of additional Neto had normal mEJPs. The mini frequency appeared particularly sensitive to Neto levels and was significantly reduced in both Neto-depleted and Neto-excess conditions. The reduction in mini frequency and amplitude occurred in muscles with no change in both resting potential and input resistance. The EJP amplitude was similarly sensitive to Neto levels: mild/moderate increase in Neto levels showed no significant change in EJP amplitude, while strong perturbations of the Neto levels (depletion or excess) induced significant reduction in the EJP amplitude.
Although GluRIIC muscle levels were constant in larvae with increased Neto expression (Fig. 1B), the similarities between the NMJ physiological properties at reduced or excess Neto suggest that excess Neto could affect the number and density of postsynaptic iGluRs. Indeed, excess Neto produced a significant decrease of GluRIIC synaptic clusters: the number of synaptic contacts per bouton did not change, but the intensity of the GluRIIC synaptic signals was reduced to 58% ± 12% of the control (Fig. 2A-E). The anti-GluRIIC also labeled extrasynaptic puncta that occasionally accompanied small Neto clusters, but did not co-localize with the large extrajunctional Neto-positive puncta, presumably associated with secretory vesicles (Fig. 2C). In contrast, the synaptic distribution of Bruchpilot (Brp), an active zone scaffold [41], remained unaffected by excess Neto, indicating that Neto specifically regulates the distribution of postsynaptic receptors. The decrease of synaptic iGluRs showed no subtype specificities when Neto was overexpressed in a wild-type background (G14>neto-A1); both GluRIIA and GluRIIB synaptic levels were similarly decreased (to 60% and respectively 54% from control) (Fig. 2E-F). This is consistent with the normal quantal size (or mini amplitude), observed at these NMJs (Fig. 1E) [15,16]. However, when excess Neto was introduced in the neto null background (neto36; G14>neto-A3), the GluRIIA synaptic levels were reduced slightly more than the GluRIIB, to 48% and respectively 62% from control.
Loss of synaptic pMad, the BMP pathway effector, correlated with small NMJs with reduced synaptic release in neto and importin-β11 mutants [33,42]. We found that Neto overexpression also caused attenuation of the synaptic pMad levels likely by decreasing the levels of synaptic type-A receptors (Fig. 3). Reduced synaptic iGluRs together with diminished retrograde BMP signaling could explain the small size of NMJs with excess or reduced Neto levels. However, there were several differences between these NMJs. Unlike neto hypomorph larvae, which showed diminished synaptic localization of multiple synaptic components, such as p21-activated kinase (PAK), Discs large (Dlg), and α-Spectrin [23], excess Neto did not affect the synaptic accumulation of any of these proteins (S2 Fig.). In line with normal Brp, excess Neto did not affect the presynaptic localization of cysteine-string protein (CSP) [43]. Thus, the neto gain-of-function NMJ phenotypes cannot result from insufficient trafficking and recruitment of postsynaptic components. Normal recruitment of Dlg at synaptic locations was also observed when V5- or GFP-tagged Neto variants replaced the endogenous Neto protein (S3 Fig.). Similar to untagged Neto, excess Neto-V5 or Neto-GFP induced smaller NMJs with normal synaptic transmission (Neto-GFP-rescued NMJ shown in S3 Fig.), indicating that the addition of tags did not affect Neto activities and gain-of-function phenotypes (Figs 1, S3 and [23]).
How could excess Neto diminish the synaptic iGluR levels without affecting any other synaptic components tested here? Stable synaptic receptors are thought to be part of large aggregates organized by proteins secreted from the presynaptic compartment [44,45] and further stabilized by postsynaptic scaffolds [46]. Neto may interact with neuron-secreted proteins that trigger iGluRs synaptic clustering and/or with intracellular motors and scaffolds that promote iGluRs trafficking and stabilization at synaptic sites. Excess Neto may engage in unproductive interactions and overwhelm the cellular machineries involved in the trafficking and clustering of iGluRs at synaptic locations. Since Neto does not affect the net levels of receptor subunits in the postsynaptic muscle (Fig. 1B and [23]), then iGluRs are predicted to accumulate at extrajunctional locations at suboptimal Neto levels. Indeed, genetic manipulation of Neto levels triggered a redistribution of iGluR-positive signals from junctional to extrajunctional locations (Fig. 2C and [23]). Moreover, Neto proteins appear to have no roles in the surface delivery of the iGluRs in vertebrate and in C. elegans [25,26] suggesting that reduced or excess Neto levels should induce accumulation of extrajunctional iGluRs at the muscle surface. We tested this prediction by staining the larval fillets in detergent-free protocols with antibodies raised against the extracellular domain of GluRIIC. Under these conditions, extrajunctional GluRIIC staining was barely visible in control, but was very prominent on the muscle of larvae with reduced or excess Neto (Fig. 4A-B). Surface accumulation at extrajunctional locations of GluRIIA was also observed in neto109 hypomorphs [23]. Similar results were obtained in both rescue and overexpression experiments with either neto-A3 or neto-A1 transgenes even though neto-A1 appears to induce a higher Neto expression level (Figs 4, 1). Together, our data indicate that optimal Neto levels are crucial for the recruitment and stabilization of iGluRs at synaptic sites. Similar to vertebrate or C. elegans, perturbations of the Drosophila Neto levels do not appear to affect the surface delivery of iGluRs and instead influence the iGluRs distribution between synaptic and extrasynaptic locations.
Unlike vertebrate or C. elegans Neto, Drosophila Neto contains a long sequence preceding the first CUB domain (CUB1). Full-length Neto is predicted to be a 78 kD protein, yet when expressed in S2 insect cell, Neto runs as two bands: a minor band with relative mobility ~100 kD, and a major band of ~85 kD (Fig. 5A). Truncated Neto variants containing only the extracellular part (Neto-extra) showed bands of ~60 and ~45 kD. Similar pattern was also detected in neto36 null embryos rescued with a neto-V5 transgene.
To examine whether Neto is cleaved, we generated a binary-tagged CUB1 fragment (Myc-CUB1-V5/His, Fig. 5B). This secreted fragment produced three distinct bands corresponding to full length, N-terminal and C-terminal fragments. The C-terminal fragment was purified and analyzed by Edman degradation and mass spectrometry. Three cleavage sites within a region containing tandem repeats of RXXR dibasic motifs, upstream of the CUB1 domain, were identified. The major cleavage site appears to be the R129-Q bond, but R126-S and R123-A bonds could also be cleaved (Fig. 5C). Interestingly, this region is highly conserved in all Drosophila species but not in vertebrate or C. elegans Neto, suggesting that this processing has functional implications for Neto functions in flies. The cleavage sites match the consensus processing sequence for Furin-like proprotein convertases (PC), also known as PACE (Paired basic Amino acid Cleaving Enzyme), which process latent precursor proteins into their biologically active forms [47].
Drosophila genome codes for three Furin-type enzymes: Furin1 (Fur1), Furin2 (Fur2), and Amontillado (Amon). Fur1 and Fur2 were expressed and analyzed in vitro, but their mutants have not been described yet [48,49]. Mutants in amon, encoding the Drosophila homolog of the neuropeptide precursor processing protease PC2, display partial embryonic lethality, defective larval growth, and arrest during the first to second instar larval molt [50,51]. To confirm that Furins are responsible for cleaving Neto we used an RNAi approach [36]. We generated double strand RNA (dsRNA) for each of the three Furin-like coding genes, co-transfected them with Neto expression constructs in S2 cells, and examined the protein products. The efficiency of RNAi treatments was verified by RT-PCR (Fig. 5D). We found that knockdown of Fur1 activities reduced the production of the small, cleaved bands and increased the level of unprocessed form (Fig. 5E, lanes 1 and 2). However, we did not find any difference by knocking down Fur2 or Amon (Fig. 5E, lanes 3 and 4). Combination of all 3 different dsRNAs did not further reduce the proportion of uncleaved Neto forms compared to Fur1 RNAi (Fig. 5E, lanes 1 and 5), indicating that Fur1 is the primary enzyme for cleaving Drosophila Neto in S2 cells.
In flies, fur1 is expressed throughout development in multiple tissues including larval central nervous system and carcass [52]. RNAi-mediated fur1 knockdown in the striated muscle produced NMJs with fewer and smaller boutons, normal Brp synaptic clusters, but significantly diminished levels of synaptic iGluRs (S4 Fig.). While these phenotypes are reminiscent of NMJs with suboptimal Neto, they cannot be solely attributed to reduced Neto activities due to lack of processing. Fur1, like all Furin-type proteases, cleaves and activates multiple developmentally important substrates, including extracellular matrix components and signaling molecules such as TGFβ-type ligands [53]. In fact, stronger RNAi treatments (in the presence of Dicer or at higher rearing temperature) distorted the muscle fibers and induced early larval lethality. A pulse of high temperature (one day at 30°C) also disrupted the muscle structures. Fur1 knockdown also induced significant reduction in GluRIIA and pMad synaptic signals, likely because inefficient activation of precursor TGFβ-type factors, including Gbb (S4 Fig.). Interestingly, down-regulation of fur1 in motor neurons elicited similar NMJ phenotypes, underscoring the complexity of Fur1-dependent activities.
To study the biological relevance of Neto processing by Furin-type proteases we generated a constitutively active Neto variant (CA-Neto), without the prodomain, and a processing mutant Neto (PM-Neto), with an uncleavable prodomain (Fig. 6A). When expressed in S2 cells, Neto-GFP was detected as double bands of expected sizes, mostly processed form. CA-Neto-GFP was found as a single, processed protein, while PM-Neto-GFP was predominantly unprocessed. We noticed that a small fraction of PM-Neto (<15%) was processed presumably by promiscuous proteolysis, which may partly remove the prodomain; however, such cleavage usually occurs at ectopic locations, adding or removing additional residues from the processed product. Further mutations in this conserved region did not completely abolish Neto processing, but could impact the proper function of the adjacent CUB domain.
To examine the subcellular distribution of Neto variants we took advantage of the apical localization of Neto in epithelial tissues. G14-Gal4 drives the expression of UAS transgenes in muscles but also in salivary glands. We found that all Neto variants localized to the luminal side of the salivary gland (apical surface), indicating that prodomain processing does not affect membrane targeting and apical localization of Neto proteins (Fig. 6B). Nor did prodomain processing impact the ability of Neto variants to form complexes with iGluRs in the striated muscle. Similar to Neto, CA-Neto and PM-Neto retained the capacity to pull-down iGluRs from muscle extracts (Fig. 6C). However, PM-Neto was severely impaired in its ability to rescue the neto null mutants, while CA-Neto generally resembled the Neto control (Fig. 6D, E). Very few PM-Neto rescued animals reached the adult stages: these flies did not fly and had locomotor defects.
Similar to the wild-type neto transgenes, moderate levels of CA-Neto rescued the NMJ morphology and iGluRs clustering defects of neto null mutants, while excess CA-Neto generated smaller NMJs with reduced iGluRs synaptic signals (Fig. 7A, B). In contrast, PM-Neto rescued NMJs developed abnormally irrespective of the expression levels. At moderate levels, PM-Neto distributed diffusely and disrupted the synaptic localization of iGluRs, in particular the type-A receptors (Fig. 7A-D, quantified in 7E, F). Animals rescued with high PM-Neto levels died during the early larval stages; the rare third instar escapers did not move and had severely altered NMJs with sparse boutons decorated by irregular Brp-positive aggregates and almost undetectable synaptic GluRIIC puncta (Fig. 7A, B, G). These data suggest that PM-Neto is inadequate for the proper recruitment and stabilization of iGluRs at postsynaptic locations even though PM-Neto appears to bind to GluRIIC in vivo and to enable embryos to hatch into larval stages (Fig. 6C, D). The severity of phenotypes at PM-Neto rescued synapses indicates that prodomain removal is required for iGluRs synaptic clustering, for development of postsynaptic structures, or both.
Perisynaptic Dlg signals flank but do not co-localize with PSD components [54]. At control NMJ, Dlg appeared to surround the Neto-positive puncta (Fig. 8A). The synaptic accumulation of Dlg was severely reduced at PM-Neto rescued NMJs, without any detectable change in the level of Dlg protein in larval muscle. These mutant NMJs were hardly recognizable since both Dlg and Neto synaptic signals were diminished and distributed diffusely among very few boutons. Similar to iGluRs and Dlg, PAK did not accumulate at PM-Neto rescued NMJs (Fig. 8B). In contrast, the assembly of presynaptic components was not affected in PM-Neto rescued synapses: Brp and CSP showed discrete synaptic distributions (Figs 7, 8C). The severe postsynaptic defects at PM-Neto rescued NMJs were not accompanied by cytoskeletal disruption as indicated by normal α-Spectrin distribution (Fig. 8D). Thus, postsynaptic differentiation and organization of PSD structures appear to be specifically affected by Neto processing.
The aberrant postsynaptic differentiation at PM-Neto rescued NMJs was also captured by electron micrographs of larval NMJs. These NMJs had rare boutons with no postsynaptic electron dense structures and no detectable SSR, and surrounded instead by dense ribosome fields or myofibrils (Figs 9A, B-F, and S5). The T-bar structures were often misshaped, collapsed or floating at PM-Neto boutons, suggesting that lack of Neto/iGluRs clustering affects proper assembly and organization of presynaptic structures. Larger T-Bars and synaptic vesicles at PM-Neto-rescued NMJs may reflect a homeostatic compensatory response to reduced postsynaptic receptors. Similar structures were reported in mutants with enhanced presynaptic release [10].
Physiological recordings indicated that the mini frequency was severely reduced at PM-Neto rescued NMJs consistent with drastically reduced synaptic iGluRs (Fig. 9G, H). The mini amplitude was also decreased, likely due to the preferential loss of type-A receptors at these synapses (Figs 7D, 9I). Consistent with the large vesicle seen in electron micrographs we occasionally observed very large minis at PM-Neto rescued NMJs. However, the evoked potentials were normal suggesting a presynaptic compensatory response (Fig. 9J-L). Thus, Neto processing is required for the normal density of postsynaptic iGluRs, but is not essential for triggering a compensatory increase in presynaptic release.
PM-Neto not only failed to cluster and stabilize the iGluRs at postsynaptic locations but it was also unable to support the recruitment of postsynaptic components, formation of PSDs, and stabilization of postsynaptic structures. The postsynaptic differentiation program was simply not initiated at PM-Neto rescued NMJs. Our data are consistent with a model in which Fur1-dependent processing activates Neto and allows it to function to stabilize iGluR complexes at synaptic sites. The prodomain may prevent the formation and/or maintenance of stable Neto/iGluR synaptic aggregates by obstructing Neto-mediated protein interactions. Lack of iGluRs clustering precludes the initiation of postsynaptic differentiation.
Clustering of Drosophila iGluRs at synaptic locations is limited by the essential subunits GluRIIC-E and by the obligatory auxiliary protein Neto [12–14,23]. Reduced Neto levels limit the iGluRs synaptic distribution and formation of postsynaptic specializations, without any change in the net levels of iGluRs or synaptic components [23,33]; excess Neto triggers a selective reduction of the synaptic iGluRs, but not other postsynaptic components, likely by sequestering iGluRs at extrajunctional locations. Neto activities are modulated by Furin-type proteases, which remove the inhibitory prodomain of Neto and enable Neto/iGluRs synaptic clustering and postsynaptic differentiation. Since Furins also process and activate multiple other substrates, including TGF-β ligands and extracellular matrix components, modulation of Neto activities by prodomain processing may serve as a way to coordinate synapse assembly with NMJ growth and development.
The increase as well as the decrease of Neto levels affects the NMJ development, albeit with different consequences. Neto-deprived NMJs have diminished postsynaptic specializations and long branches, spanning over large muscle areas, suggesting that lack of postsynaptic receptors maintains the motor neurons in a growing, exploratory state. By contrast, NMJs with excess Neto are short and have normal accumulation of postsynaptic components. In fact, PAK and Dlg signals are slightly elevated at NMJs with excess Neto compared with control (S2 Fig.). Early accumulation of synaptic Dlg may restrict expansion of these NMJs and produce hypo-innervation. Interestingly, overexpression of Neto in the wild-type background (G14>neto-A1) induced gain-of-function phenotypes slightly milder than when the same transgene replaced the endogenous neto in rescue experiments (compare the last two columns in Fig. 1A). This could be due to the different genetic backgrounds or may indicate additional Neto functions that are missing at neto-A1-rescued NMJs.
Physiological studies also captured the differences between postsynaptic iGluR receptor fields at different Neto levels. Neto-deprived NMJs in neto hypomorphs or RNAi experiments have severely reduced mini frequency consistent with their reduced postsynaptic iGluRs density (Fig. 1C-F and [23,33]). Strong reduction of postsynaptic Neto levels induced a reduction of EJP amplitudes, suggesting that Neto deprivation interferes with the normal homeostatic mechanisms. Similar to iGluRs-deprived synapses, lack of Neto may render these synapses “beyond repair” [12,14]. In contrast, the NMJ physiological parameters appeared more to be resilient to excess Neto since addition of moderate levels of Neto did not affect the mEJP and EJP amplitude. However, high levels of excess Neto (G14>neto-A1) induced a significant decrease of mEJP frequency, consistent with the reduced synaptic and increased extrasynaptic iGluRs observed at these NMJs (Figs 1, 2, 4).
At central glutamatergic synapses in vertebrates, synaptic receptors are cycling into and out of the synapses indicating that synapses behave as donors or acceptors for receptors, and the extrasynaptic receptors function as a reserve pool [2]. At the Drosophila NMJ, the iGluRs are recruited to the nascent synapses from extrajunctional receptor pools, but are stably integrated in synaptic aggregates with very low turnover [22]. In the absence of Neto, or any essential iGluR subunit, the iGluRs are not recruited at synaptic locations [55]. Conversely, excess Neto induces accumulation of iGluR-positive puncta at extrajunctional locations (Figs 2, 4). This is different than overexpression of any of the essential iGluR subunits, which don’t show gain-of-function phenotypes, presumably because other subunits are limiting [11]. Furthermore, the iGluR complexes appear to be on the muscle membrane at suboptimal Neto levels since they are accessible by antibodies in the absence of detergents (Fig. 4 and [23]). Likewise, Neto proteins from worms and mammals appear to have no roles (or very modest ones) in the surface delivery of the iGluRs [25,26]. We speculate that Neto binds iGluRs on the cell surface and engages in extracellular and/or intracellular interactions that enable the recruitment and clustering of iGluRs at synaptic sites. In this scenario, reduced Neto levels are inefficient for the iGluRs synaptic trafficking and clustering, whereas excess Neto may engage in protein interactions that sequester iGluRs at ectopic locations.
At the Drosophila NMJ, Neto activities are regulated by Fur1-dependent limited proteolysis. The removal of Neto prodomain appears to be essential for the stabilization of iGluRs at PSDs. Lack of iGluRs stabilization precludes postsynaptic differentiation although the receptors are functional (Figs 8, 9). Thus, synapse activity does not trigger iGluRs clustering or postsynaptic differentiation; instead, stabilization of iGluRs at synaptic sites initiates the recruitment of PSD components and assembly of postsynaptic structures.
It has been proposed that a neuron secreted molecule triggers clustering of iGluRs at Drosophila NMJ [18,20,56]. Secreted molecule(s) may mediate iGluRs clustering by binding and trapping Neto/iGluR complexes at new synapses. Mind the gap (Mtg) is a neuronal protein reported to organize the synaptic cleft [57]. In mtg null mutant embryos, Neto and iGluRs form aggregates comparable in size with control clusters, but which fail to concentrate at nascent synapses [55]. Unfortunately, we could not detect in vitro interactions between Neto and Mtg. But while the molecular nature of the “trapping” mechanism remains to be determined, our study demonstrates that this process requires the removal of Neto prodomain.
The Neto prodomain does not interfere with targeting and apical localization of Neto, nor does it affect its ability to bind iGluRs and form complexes, but it appears to preclude Neto engagement in protein interactions required for the formation of iGluR synaptic clusters. Neto CUB1 domain interacts with itself, but self-association is not enough to explain the formation of large iGluR aggregates. Prodomains could mediate binding to extracellular factors, such as heparan proteoglycans, fibrillin and perlecan that protect the active molecules and modulate their extracellular distribution [58,59]. Our study does not address a role for Neto prodomain in binding to extracellular molecules that modulate Neto distribution. The prodomains could also function as chaperones that allow proper folding of biologically active molecules, such as TGF-β-type ligands [34]. However, Neto prodomain is unlikely to play a role in the folding and secretion of Neto because CA-Neto is functional and induces NMJ gain-of-function phenotypes similar to excess Neto. Alternatively, the prodomain could maintain Neto in an inactive form, thus limiting clustering and stable incorporation of Neto/iGluR complexes at PSDs. Similar regulation has been described for the Tolloid/BMP-1 family of enzymes: their prodomains must be removed before the catalytic domains could assume active conformations [60]. It is tempting to speculate that the prodomain masks Neto extracellular domain(s) and prevents interactions required for iGluR clustering at PSDs.
Is Neto processing a general step in Neto passage through the secretory pathway or could it actively modulate Neto activity/ availability? To test if processing plays an active role in regulating Neto function we compared the changes in Neto processing in larvae with hunger-induced increase of locomotion [61]. The proportion of processed Neto increased in starved larvae and decreased in fed animals (Fig. 10), indicating that Neto processing indeed changes in response to an increase in locomotion and/or due to starvation. While this analysis cannot distinguish between the two possibilities, Neto processing emerges as an active mechanism to control the level of Neto available for effective iGluRs recruitment at PSDs.
Neto processing/activation phenomenon appears to be highly conserved in insects. Most insects have glutamatergic NMJs, and their genomes encode for Neto proteins with prodomains and Furin minimal sites (R-X-X-R) preceding the first CUB domain. For example, Neto proteins in Apis florea and Apis mellifera share an R-Q-M-R motif at positions equivalent to the Furin site in Drosophila Neto. In all cases, the Furin consensus sites are suboptimal suggesting that processing of insect Neto proteins will be slow and restricted by Furin activities. Furins cleave their substrates mainly in late Golgi, though recent data indicate that Furins also function at the cell surface and in the extracellular space [62]. Interestingly, Fur1 also cleaves and activates TGF-β-type ligands, including Gbb, Maverick and Dawdle, which are secreted from muscle and glia and control NMJ development [37,39,40]. This raises the possibility that Fur1 synchronizes the activation of Neto and TGF-β factors and may serve as a means to coordinate synapse assembly with NMJ growth.
This study does not exclude other mechanisms that may regulate the density of synaptic iGluR, such as local insertion of iGluRs from intracellular vesicles [63]. Nonetheless, our study demonstrates that Neto activation by prodomain processing plays an important role in the regulation of iGluR trafficking and clustering at synapses. Trafficking of Neto itself or Neto/iGluR complexes on the muscle membrane may be further controlled by cellular signals that modulate the intracellular domain of Neto and regulate its coupling with scaffold and motor complexes. In fact, Drosophila neto locus codes for two isoforms generated by alternative splicing that differ in their intracellular domains. Both intracellular domains contain multiple putative phosphorylation sites, raising the possibility of rich modulation of Neto/iGluRs distribution in the striated muscle.
Fly lines were generated by standard germline transformation of pUAST-based plasmids containing various neto constructs (BestGene, Inc). Other stocks used in this study were as follows: neto null and hypomorph alleles, neto36 and respectively neto109 [23], netoRNAi [33], G14-Gal4 and MHC-Gal4 (obtained from C. Goodman, University of California at Berkeley), da-Gal4 (BL-5460), 24B-Gal4 (BL-1716), and elav-Gal4 (BL-8760). For RNAi-mediated knockout we used the following TRiP lines generated by the Transgenic RNAi Project: GluRIIC (P[TRiP.JF01854}attP2), and fur1 (P[TRiP.GL01340] attP40). The control is y1w1118 unless otherwise specified.
For rescue analyses, neto transgenes were introduced into neto36 null mutant background using tissue-specific promoters. Since neto is on the X-chromosome we used only FM7-GFP balanced stocks to eliminate any meiotic non-disjunction event. The F1 progenies were genotyped during late embryogenesis and reared at the indicated temperatures. After 24 hours, crawling larvae were removed, counted, and kept at the same temperatures for further analyses or adult viability testing.
Neto variants were generated using QuikChange site-directed mutagenesis kit (Stratagene) as described previously [23]. CA-Neto has a deletion that joins A51-Q130 and loops out the Neto prodomain. PM-Neto has two point mutations: R123I and R126I. Double-tagged Neto constructs were generated by QuikChange loop-in of various Neto fragments in a previously described AcPA-SP-Myc-V5/His plasmid [64]. This actin promoter/terminator plasmid contains the sequences coding for the Tolloid-related signal peptide, the 5xMyc cassette, a multiple cloning site, followed by the V5 and RGS-6xHis epitopes. All constructs were verified by DNA sequencing.
For RNA interference, PCR primers for Furins that carry the T7 promoter sequence at the 5’ end were designed as previously described [36]. The primers were as follows:
dFur1-F 5’-TAATACGACTCACTATAGGGACGCAAAGATCCTCTGTGGCA;
dFur1-R 5’- TAATACGACTCACTATAGGGACATTGCTCCCGGAACTGC;
dFur2-F 5’- TAATACGACTCACTATAGGGACGCTAGAGGCCAATCCGGAA;
dFur2-R 5’- TAATACGACTCACTATAGGGACCCTTCTCGCCCCAAAAGTG;
Amon-F 5’- TAATACGACTCACTATAGGGACCCACATGGAGCTGGCTGT;
Amon-R 5’- TAATACGACTCACTATAGGGACCCTGACTTTGCCGCCATT.
PCR products were amplified from genomic DNA or S2 cells cDNA. In vitro transcribed dsRNA was produced using the MEGAscript kit (Ambion). RNAi treatment was carried out by transfections of 5 mg/ml of dsRNA into S2 cells.
S2 cells were transfected with indicated constructs and harvested after five days incubation. Total RNA was extracted using TRIZOL reagent (Invitrogen) according to manufacturer's instructions. AccuScript High Fidelity First-Strand cDNA Synthesis Kit (Agilent) was used to generate cDNAs from the extracted total RNAs according to manufacturer’s instructions. PCR reaction for each target gene was executed using the cDNAs as templates with specific primer pairs (above) and β-Actin as a reaction standard (Actin-Forward: 5’-CTGGCACCACACCTTCTACAATG-3’, Actin-Reverse: 5’-GCTTCTCCTTGATGTCACGGAC-3’).
Wandering third instar larvae were dissected as described previously in ice-cooled Ca2+-free HL-3 solution [65,66]. Dissecting larval tissues were fixed in either 4% formaldehyde or Bouin's fixative (Polysciences, Inc.) for 20 min or 5 min respectively. PBS containing 0.5% Triton X-100 was used for washing and antibody reaction. For detergent-free staining, 1X PBS was used. Primary antibodies from Developmental Studies Hybridoma Bank were used at the following dilutions: mouse anti-GluRIIA (MH2B), 1:100; mouse anti-Dlg (4F3), 1:1000; mouse anti-Brp (Nc82), 1:100; mouse anti-CSP (6D6), 1:100; mouse anti-α-spectrin (3A9), 1:100. Other primary antibodies were as follows: rat anti-Neto, 1:1000 [23], rabbit anti-GluRIIB, 1:2000 (a gift from David Featherstone) [67]; rabbit anti-GluRIIC, 1:2000 [33]; rabbit anti-PAK, 1:2000 (a gift from Nicholas Harden) [68]; FITC-, rhodamine-, and Cy5-conjugated goat anti-HRP, 1:1000 (Jackson ImmunoResearch Laboratories, Inc.). Alexa Fluor 488-, Alexa Fluor 568-, and Alexa Fluor 647-conjugated secondary antibodies (Molecular Probes) were used at 1:400. All samples were mounted with ProLong Gold reagent (Invitrogen) and incubated for 24 hours at RT. Confocal images were acquired using Carl Zeiss LSM 780 or 510 laser scanning microscopic system with Plan-Apochromat 63X/1.4 oil DIC objective using ZEN software. Z-stacked images were collected, processed, and analyzed using Imaris X64 (7.6.0, Bitplane) or ImageJ (NIH) software. In each experiment, samples of different genotypes were processed simultaneously and imaged under identical confocal settings. To quantify fluorescence intensities, confocal regions of interest (ROIs) surrounding anti-HRP immunoreactivities were selected and the signals measured individually at NMJs from ten or more different larvae for each genotype (number of samples is indicated in the graph bar). The signal intensities were calculated relative to HRP volume and subsequently normalized to control. For the extrajunctional, cell surface GluRIIC staining, where the GluRIIC positive signals are predominantly in the form of puncta at both Neto-depleted and Neto-excess NMJs, intensities from several size-matched areas of the muscles were collected and averaged using Image J software. The numbers of muscles analyzed per genotype are indicated inside the bars. Quantification of NMJ morphological features was performed at muscle 4 of abdominal segment 4 using the filament tracing function of Imaris software. Boutons were counted manually, while blind to the genotype, using anti-HRP and anti-Dlg staining. Statistical analyses were performed using the Student’s t-test with a two-tailed distribution and a two-sample unequal variance. All graphs represent mean value of all samples of the given genotype ± SEM.
Transiently transfected Drosophila S2 cells were used for producing recombinant proteins as previously described [69]. The S2 cells were maintained in M3 (Shields and Sang M3 insect medium, Sigma) with 1x insect medium supplement (Sigma) and Penicillin/Streptomycin (Sigma), and sub-cultured every 7 days at 2 X 106 cells/ml. For the transfection, dimethyldioctadecyl-ammonium bromide (DDAB) solution (250 μg/ml) was mixed with M3 media at 1:2 ratio and incubated 5 min at RT, then the DNA was added to the DDAB-M3 mixture (1μg of plasmid DNA to 100 μl suspension). The mixture was incubated for 20 min and transfected into S2 cells (100 μl mixture to 2 X 106 cells/ml culture). After five days, the secreted proteins were collected for analysis, and membrane proteins were extracted by homogenizing cells in lysis buffer (50 mM Hepes-NaOH, 150 mM NaCl, 0.2 mM EDTA, 0.5% NP-40, 0.1% SDS, 2mM AEBSF [MP BIO], and protease inhibitor cocktail [Roche]) for 30 min on ice. The lysates were collected by centrifugation at 13,000rpm for 30 min at 4°C, separated by SDS-PAGE on 4%–12% NuPAGE gels (Invitrogen) and transferred onto PVDF membranes (Millipore). Primary antibodies were used at the following dilutions: rat anti-Neto, 1:1000; chicken anti-GFP (Abcam), 1:2000; anti-GluRIIC, 1:1000; anti-tubulin (Sigma), 1:1000. Immune complexes were visualized using secondary antibodies coupled with IR-Dye 700 or IR-Dye 800 followed by scanning with the Odyssey infrared imaging system (Li-Cor Biosciences).
To analyze muscle proteins, wandering third instar larvae were dissected, and the body walls were mechanically homogenized in lysis buffer for 30 min on ice. The lysates were analyzed by Western blotting. For co-immunoprecipitation, the lysates were incubated with rabbit anti-GFP antibody (Invitrogen) for 1 hr at 4°C. Protein A/G UltraLink Resin (50% slurry, Thermo Scientific) was added and incubated overnight at 4°C. The beads were washed with lysis buffer. Proteins were eluted with 1x SDS sample buffer and analyzed by Western blotting.
Secreted and processed Neto fragment (CUB1-V5/His) was purified using His-Trap affinity column equipped with AKTA FPLC system (Pharmacia) and separated by SDS-PAGE. A specific gel band was isolated and analyzed at Microchemistry and Proteomics Analysis Facility, Harvard University.
Wandering third instar larvae were dissected in Jan's saline containing 0.1 mM Ca2+ and processed as previously described [70]. Dissected larvae were fixed in EM fixative (4% p-formaldehyde, 1% glutaraldehyde, 0.1 M sodium cacodylate, and 2 mM MgCl2, pH 7.2) for 20 min at room temperature followed by incubation overnight at 4°C, then washed extensively (0.1 M sodium cacodylate, and 132 mM sucrose, pH 7.2). The samples were processed and analyzed at the Microscopy and Imaging Core Facility, NICHD.
The standard larval body wall muscle preparation first developed by Jan and Jan (1976) was used for electrophysiological recordings [71,72]. Wandering third instar larvae were dissected in physiological saline HL-3 [65], washed, and immersed in HL-3 containing 0.8 mM Ca2+ using a custom microscope stage system [73]. The nerve roots were cut near the exiting site of the ventral nerve cord so that a suction electrode could pick up the motor nerve later. Intracellular recordings were made from muscle 6. Data were used when the input resistance of the muscle was >5 MΩ and the resting membrane potential was between −60 mV and −80 mV for the entire duration of the experiment. The input resistance of the recording microelectrode (backfilled with 3 M KCl) ranged from 20 to 25 MΩ. Muscle synaptic potentials were recorded using Axon Clamp 2B amplifier (Axon Instruments) and pClamp software. Following motor nerve stimulation with a suction electrode (100 μsec, 5 V), evoked EJPs were recorded. Three to five EJPs evoked by low frequency of stimulation (0.1 Hz) were averaged. For mini recordings, TTX (1 μM) was added to prevent evoked release [65]. To calculate mEJP mean amplitudes, 50–100 events from each muscle were measured and averaged using the Mini Analysis program (Synaptosoft). Minis with a slow rise and falling time arising from neighboring electrically coupled muscle cells were excluded from analysis [72,74]. In addition, when comparing mini sizes between preparations, the Kolmogorov-Smirnov test was administrated. Quantal content was calculated by dividing the mean EJP by the mean mEJP after correction of EJP amplitude for nonlinear summation according to the methods described [75,76]. Corrected EJP amplitude = E[Ln[E/(E − recorded EJP)]], where E is the difference between reversal potential and resting potential. The reversal potential used in this correction was 0 mV [75,77]. Data are presented as mean ± SEM, unless otherwise specified; EJP amplitudes and quantal contents after the nonlinear correction are shown. A one-way analysis of variance followed by Tukey's HSD test was used to assess statistically significant differences among the genotypes. Differences were considered significant at p < 0.05.
|
10.1371/journal.pgen.1002864 | The MicroRNA mir-71 Inhibits Calcium Signaling by Targeting the TIR-1/Sarm1 Adaptor Protein to Control Stochastic L/R Neuronal Asymmetry in C. elegans | The Caenorhabditis elegans left and right AWC olfactory neurons communicate to establish stochastic asymmetric identities, AWCON and AWCOFF, by inhibiting a calcium-mediated signaling pathway in the future AWCON cell. NSY-4/claudin-like protein and NSY-5/innexin gap junction protein are the two parallel signals that antagonize the calcium signaling pathway to induce the AWCON fate. However, it is not known how the calcium signaling pathway is downregulated by nsy-4 and nsy-5 in the AWCON cell. Here we identify a microRNA, mir-71, that represses the TIR-1/Sarm1 adaptor protein in the calcium signaling pathway to promote the AWCON identity. Similar to tir-1 loss-of-function mutants, overexpression of mir-71 generates two AWCON neurons. tir-1 expression is downregulated through its 3′ UTR in AWCON, in which mir-71 is expressed at a higher level than in AWCOFF. In addition, mir-71 is sufficient to inhibit tir-1 expression in AWC through the mir-71 complementary site in the tir-1 3′ UTR. Our genetic studies suggest that mir-71 acts downstream of nsy-4 and nsy-5 to promote the AWCON identity in a cell autonomous manner. Furthermore, the stability of mature mir-71 is dependent on nsy-4 and nsy-5. Together, these results provide insight into the mechanism by which nsy-4 and nsy-5 inhibit calcium signaling to establish stochastic asymmetric AWC differentiation.
| Cell identity determination requires a competition between the induction of cell type–specific genes and the suppression of genes that promote an alternative cell type. In the nematode C. elegans, a specific sensory neuron pair communicates to establish stochastic asymmetric identities by inhibiting a calcium signaling pathway in the neuron that becomes an induced identity. However, it is not understood how cell–cell communication inhibits the calcium signaling pathway in the induced neuronal identity. In this study, we identify a microRNA that represses the expression of a key molecule in the calcium signaling pathway to promote the induced neuronal identity. Overexpression of the microRNA causes both neurons of the pair to become the induced identity, similar to the mutants that lose function in the calcium signaling pathway. In addition, the stability of the mature microRNA is dependent on a claudin-like protein and a gap junction protein, the two parallel signals that mediate communication of the neuron pair to promote the induced neuronal identity. Our results provide insight into the mechanism by which cell–cell communication inhibits calcium signaling to establish stochastic asymmetric neuronal differentiation.
| Cell fate determination during development requires both the induction of cell type specific genes and the suppression of genes that promote an alternative cell fate [1]–[4]. For example, both inductive signaling, mediated by an EGFR-Ras-MAPK pathway, and lateral inhibition, mediated by LIN-12/Notch activity and microRNA (miRNA), are required for six multipotential vulval precursor cells to adopt an invariant pattern of fates in C. elegans [5]. Notch signaling-mediated lateral inhibition also plays a crucial role in the neuronal/glial lineage decisions of neural stem cells; as well as the B/T, alphabeta/gammadelta, and CD4/CD8 lineage choices during lymphocyte development [6], [7]. In the Drosophila eye, the kinase Warts and PH-domain containing Melted repress each other's transcription in a bistable feedback loop to regulate the two alternative R8 photoreceptor subtypes expressing Rhodopsin Rh5 or Rh6 [2]. In the C. elegans sensory system, two sets of transcription factors and miRNAs reciprocally repress each other to achieve and stabilize one of the two mutually exclusive ASEL and ASER taste neuronal fates [8]–[10]. Notch signaling acts upstream of the miRNA-controlled bistable feedback loop to regulate ASE asymmetry through a lineage-based mechanism in early embryos [11].
The C. elegans left and right sides of Amphid Wing Cell C (AWC) olfactory neurons specify asymmetric subtypes through a novel mechanism independent of the Notch pathway in late embryogenesis [12]. Like ASE neurons, the two AWC neurons are morphologically symmetrical but take on asymmetric fates, such that the AWCON neuron expresses the chemoreceptor gene str-2 and the contralateral AWCOFF neuron does not [12]–[14]. Asymmetric differentiation of AWC neurons allows the worm to discriminate between different odors [15]. In contrast to reproducible ASE asymmetry, AWC asymmetry is stochastic: 50% of animals express str-2 on the left and the other 50% express it on the right. Ablation of either AWC neuron causes the remaining AWC neuron to become AWCOFF, suggesting that AWCOFF is the default state and the induction of AWCON requires an interaction or competition between the AWC neurons [12]. The axons of the two AWC neurons form chemical synapses with each other; AWC asymmetry is established near the time of AWC synapse formation [16], [17]. In addition, axon guidance mutants are defective in inducing the AWCON state. These results suggest that the synapses could mediate the AWC interaction for asymmetry [12].
nsy-4, encoding a claudin-like tight junction protein, and nsy-5, encoding an innexin gap junction protein, act in parallel to downregulate the calcium-mediated UNC-43 (CaMKII)/TIR-1 (Sarm1)/NSY-1 (MAPKKK) signaling pathway in the future AWCON cell [18], [19]. Both AWCs and non-AWC neurons in the NSY-5 gap junction dependent cell network communicate to participate in signaling that coordinates left-right AWC asymmetry. In addition, non-AWC neurons in the NSY-5 gap junction network are required for the feedback signal that ensures precise AWC asymmetry [18]. Once AWC asymmetry is established in late embryogenesis, both the AWCON and AWCOFF identities are maintained by cGMP signaling, dauer pheromone signaling, and transcriptional repressors [12], [20], [21]. unc-43(CaMKII), tir-1 (Sarm1), and nsy-1 (MAPKKK) are also implicated in the maintenance of AWC asymmetry in the first larval (L1) stage [22]. Although multiple genes were identified to be involved in the establishment and the maintenance of AWC asymmetry (for a review, see [23]), it is still unknown how the calcium-regulated signaling pathway is inhibited by nsy-4 and nsy-5 in the AWCON cell.
The TIR-1/Sarm1 adaptor protein assembles a calcium-signaling complex, UNC-43 (CaMKII)/TIR-1/NSY-1 (ASK1 MAPKKK), at AWC synapses to regulate the default AWCOFF identity [16], thus downregulation of tir-1 expression may represent an efficient mechanism to inhibit calcium signaling in the cell becoming AWCON. In support of this idea, a prior large scale examination of potential miRNA targets indicated that tir-1 and unc-43 may be downregulated by this class of RNAs [24]. Here, we analyze the function of the miRNA mir-71 in stochastic AWC asymmetry by characterizing its role in downregulation of the calcium signaling pathway in the AWCON cell. We show that mir-71 acts downstream of nsy-4/claudin and nsy-5/innexin to promote AWCON in a cell autonomous manner through inhibiting tir-1 expression, in parallel with other processes. We also show that nsy-4 and nsy-5 are required for the stability of mature mir-71. Our results suggest a mechanism for genetic control of AWC asymmetry by nsy-4 and nsy-5 through mir-71-mediated downregulation of calcium signaling.
The calcium-regulated UNC-43 (CaMKII)/TIR-1 (Sarm1)/NSY-1 (ASK1 MAPKKK) signaling pathway suppresses expression of the AWCON gene str-2 in the default AWCOFF cell [12], [16], [25], [26]. To establish AWC asymmetry, the calcium-mediated signaling pathway is suppressed in the future AWCON cell. miRNAs are small non-coding RNAs that are robust in mediating post-transcriptional and/or translational downregulation of target genes [27]. In C. elegans, miRNAs are processed from premature form into mature form by alg-1/alg-2 (encoding the Argonaute proteins) and dcr-1 (encoding the ribonuclease III enzyme Dicer) [28]. Gene expression profiling revealed increased levels of unc-43 and tir-1 in dcr-1 mutants [24], suggesting that unc-43 and tir-1 may be downregulated by miRNAs. Thus, we hypothesized that miRNAs may play a role in downregulation of the UNC-43/TIR-1/NSY-1 signaling pathway in the cell becoming AWCON.
To test this hypothesis, we took a computational approach to identify miRNAs predicted to target the 3′ UTRs of known genes, including unc-2, unc-36, egl-19, unc-43, tir-1, nsy-1, and sek-1, in the AWC calcium signaling pathway. Only the miRNAs that fit the following criteria were selected for further analysis: 1) At least 6 nucleotides in the seed region (position 1–7 or 2–8 at the 5′ end) of a miRNA is perfectly matched to the target 3′ UTR; 2) The seed match between a miRNA and its target 3′ UTR is conserved between C. elegans and a closely related nematode species C. briggsae, since evolutionary conservation between C. elegans and C. briggsae genomes is useful in identifying functionally relevant DNA sequences such as regulatory regions [29], [30]; and 3) A miRNA is predicted by both MicroCosm Targets (formerly miRBase Targets; http://www.ebi.ac.uk/enright-srv/microcosm/htdocs/targets/v5/) [31]–[33] and TargetScan (http://www.targetscan.org/worm_12/) [34]. Based on these criteria, we identified six potential miRNAs (mir-71, mir-72, mir-74, mir-228, mir-248, mir-255) predicted to target unc-2, unc-43, tir-1, nsy-1, and sek-1 (Figure S1A). A subset of these identified miRNA-target pairs were also predicted by other miRNA target prediction programs, including PicTar (http://pictar.mdc-berlin.de/) [35] and mirWIP (http://146.189.76.171/query.php) [36].
Since most miRNAs are not individually essential and have functional redundancy [37]–[40], loss-of-function mutations in a single miRNA may not show a defect in AWC asymmetry. To circumvent potential problems that may be posed by functional redundancy, we took an overexpression approach to determine the role of these six miRNAs in AWC asymmetry. We generated transgenic strains overexpressing individual miRNAs in both AWCs using an odr-3 promoter, expressed strongly in AWC neuron pair and weakly in AWB neuron pair [41]. Wild-type animals have str-2p::GFP (AWCON marker) expression in only one of the two AWC neurons (Figure 1A and 1E). Since loss-of-function mutations in the AWC calcium signaling genes (unc-2, unc-36, unc-43, tir-1, nsy-1, and sek-1) led to str-2p::GFP expression in both AWC neurons (2AWCON phenotype) (Figure 1B and 1E) [12], [16], [25], [26], we proposed that overexpression of the miRNA downregulating one of these calcium signaling genes would also cause a 2AWCON phenotype. We found that mir-71(OE) animals overexpressing mir-71, predicted to target tir-1 and nsy-1, had a strong 2AWCON phenotype (Figure 1C, 1E, and Figure S1B). This result suggests that mir-71 may downregulate the expression of tir-1 and nsy-1 to control the AWCON fate and that mir-71 is sufficient to promote AWCON when overexpressed. However, overexpression of the other five miRNAs individually caused a mixed weak phenotype of 2AWCON and 2AWCOFF (Figure S1B). Since the activity of the nsy-1 3′ UTR in AWC was independent of mir-71(OE) (Figure S2B), we focused on the investigation of the potential role of mir-71 in promoting AWCON through negatively regulating tir-1 expression.
The genetic interaction between mir-71 and tir-1 was characterized by double mutants (Figure 1E). tir-1(ky648) gain-of-function (gf) mutants had two AWCOFF neurons (2AWCOFF phenotype) (Figure 1D and 1E) [22]. We found the tir-1(ky648gf) 2AWCOFF phenotype was significantly reduced in the tir-1(ky648gf); mir-71(OE) double mutants (p<0.001) (Figure 1E). These results support the hypothesis that mir-71 downregulates tir-1 to control the AWCON fate.
To further determine the requirement of mir-71 in AWC asymmetry, we analyzed str-2p::GFP expression in the mir-71(n4115) deletion null allele [40]. mir-71(n4115) mutants displayed wild-type AWC asymmetry (Figure 1E), suggesting that mir-71 may function redundantly with other miRNAs or non-miRNA genes to regulate calcium signaling in AWC asymmetry. In addition to mir-71, mir-248 was also predicted to target tir-1 by three programs (Figure S1A). mir-71 and mir-248 have different predicted target sites in the tir-1 3′ UTR. Since mir-248 mutants are not available, we analyzed the effect of mir-248 overexpression on AWC asymmetry. Unlike the highly penetrant 2AWCON phenotype caused by mir-71 overexpression, mir-248 overexpression generated a mixed weak phenotype of 2AWCON and 2AWCOFF (Figure S1B). To test whether mir-71 and mir-248 have a synergistic effect on AWC symmetry, we made transgenic animals overexpressing both mir-71 and mir-248 in AWCs. The 2AWCON phenotype was not significantly higher in mir-71(OE); mir-248(OE) animals than in mir-71(OE) (data not shown). These results suggest that mir-71 may not act redundantly with mir-248 to regulate tir-1 expression in AWC asymmetry. To knockdown mir-248 expression, we made an anti-mir248 transgene expressing short hairpin RNA (shRNA), consisting of both sense and antisense sequences of mir-248, in AWC. The anti-mir-248 transgene caused an AWC phenotype similar to mir-248(OE) (data not shown), suggesting that the effect of the anti-mir-248 transgene on AWC asymmetry is not through knockdown of mir-248 but mainly due to overexpression of sense mir-248 in the shRNA construct.
Functional redundancy of miRNAs and other regulatory pathways has been suggested by a previous study in the Drosophila eye [42]. To overcome functional redundancy of mir-71, we crossed mir-71(n4115) into sensitized backgrounds including tir-1(ky388), nsy-4(ky616), and unc-76(e911) mutants. tir-1(ky388) is a temperature-sensitive (ts) allele that caused a 2AWCON phenotype in 29% of animals at 15°C (Figure 1E) [16]. The 2AWCON phenotype of tir-1(ky388ts) mutants was significantly suppressed by mir-71(n4115), such that 20% of mir-71(n4115); tir-1(ky388ts) double mutants had a 2AWCON phenotype (p<0.05; Figure 1E). These results further support the hypothesis that mir-71 antagonizes the function of tir-1 in the calcium signaling pathway to promote the AWCON fate.
mir-71 is located within a large intron of the F16A11.3a (ppfr-1) gene, encoding a protein phosphatase 2A regulatory subunit (Figure 1F). It is possible that the 181 bp deletion mutation within the intron of ppfr-1 in mir-71(n4115) mutants may affect ppfr-1 activity leading to suppression of the tir-1(ky388ts) 2AWCON phenotype. To test this possibility, we analyzed AWC phenotypes in ppfr-1(tm2180); tir-1(ky388ts) double mutants. ppfr-1(tm2180) has a 1027 bp deletion removing the first three exons and therefore is a potential null allele (Figure 1F) [43]. The 2AWCON phenotype of ppfr-1(tm2180); tir-1(ky388ts) double mutants was not significantly different from that of tir-1(ky388ts) single mutants (Figure 1E). This result suggest that ppfr-1 is not required for AWC asymmetry and that suppression of the tir-1(ky388ts) 2AWCON phenotype was most likely caused by loss of mir-71 activity in mir-71(n4115) mutants.
The nsy-4 claudin-like gene and the unc-76 axon guidance pathway gene induce the AWCON state by inhibiting the downstream calcium-signaling pathway. Loss-of-function mutations in nsy-4 and unc-76 cause a partially penetrant 2AWCOFF phenotype (Figure 1E) [12], [19]. mir-71(n4115) mutations significantly enhanced the 2AWCOFF phenotype of nsy-4(ky616) and unc-76(e911) mutants (p<0.001). On the other hand, the 2 AWCON phenotype of nsy-4(OE) trasnsgenic animals overexpressing nsy-4 in AWCs was significantly suppressed in nsy-4(OE); mir-71(n4115) double mutants (p<0.001; Figure 1E). These results are consistent with a role of mir-71 function in promoting the AWCON fate, and suggest that mir-71 may act in parallel with other regulatory molecules to antagonize the calcium-regulated signaling pathway to generate the AWCON identity.
The predicted mir-71 target site in the tir-1 3′ UTR is 96 bp downstream of the stop codon; the prediction is strongly supported by four different programs, including MicroCosm Targets, TargetScan, PicTar, and mirWIP (Figure S1A). The nucleotides at position 1–8 in the seed region of mir-71 perfectly match the target site of the tir-1 3′ UTR; the seed match is conserved between C. elegans and C. briggsae (Figure 2A).
To determine whether mir-71 acts directly through the predicted binding site in the tir-1 3′ UTR, we made GFP sensor constructs with the AWC odr-3 promoter and different 3′ UTRs: wild-type tir-1 3′ UTR or the tir-1 3′ UTRmut with mutated mir-71 target site (Figure 2B). Transgenic animals expressing each sensor construct were crossed to mir-71(OE) animals. The GFP intensity of each sensor construct in an individual AWC neuron was normalized to the nucleus-localized TagRFP intensity of the transgene odr-3p::2Xnls-TagRFP::unc-54 3′ UTR in the same cell. The unc-54 3′ UTR does not contain any strongly predicted mir-71 sites. The normalized GFP intensity of each sensor construct was compared between mir-71(OE) animals and their siblings losing the mir-71(OE) transgene in the L1 stage, during which tir-1 is functional for the maintenance of AWC asymmetry [22]. We found that mir-71(OE) animals, compared with wild type, had a significantly reduced normalized expression level of GFP from the tir-1 3′ UTR sensor construct (p<0.005; Figure 2B upper panels). However, the normalized expression level of GFP from the tir-1 3′ UTRmut was not significantly different between wild-type and mir-71(OE) animals (Figure 2B bottom panels). These results suggest that mir-71 directly inhibits gene expression through the predicted target site in the tir-1 3′ UTR. However, we did not observe a significant difference in the GFP expression level from the tir-1 3′ UTR between wild-type animals and mir-71(n4115lf) mutants (Figure S2A). This result suggests potential functional redundancy of mir-71 in the regulation of tir-1 expression.
Interactions between the 5′ and 3′ UTRs have been shown to regulate translation in mammalian cells [44], bacteria [45], and RNA viruses [46]. To determine if the tir-1 5′ UTR plays a role in regulating the inhibitory effect of mir-71 on the tir-1 3′ UTR, we included the tir-1 5′ UTR in the GFP sensor constructs (Figure S3). Similar to the tir-1 3′ UTR sensor constructs without the tir-1 5′ UTR (Figure 2B), the normalized expression level of GFP from the tir-1 5′ UTR/tir-1 3′ UTR sensor construct was significantly decreased in mir-71(OE) animals compared with wild type (p<0.04; Figure S3A). However, the normalized expression level of GFP from the tir-1 5′ UTR/tir-1 3′ UTRmut sensor construct was not significantly different between wild-type and mir-71(OE) animals (Figure S3B). These results suggest that the tir-1 5′ UTR does not affect mir-71(OE)-mediated suppression of gene expression through the tir-1 3′ UTR.
The nsy-1 3′ UTR was also predicted to contain a mir-71 binding site by the four programs used in this study (Figure S1A), but the GFP expression level from the nsy-1 3′ UTR was not significantly different between wild-type and mir-71(OE) animals (Figure S2B). This result suggests that the predicted mir-71 site in the nsy-1 3′ UTR may not be functional in AWC cells, therefore we did not further investigate the regulation of nsy-1 expression by mir-71.
tir-1(OE) animals overexpressing tir-1 in AWC had a 2AWCOFF phenotype [16]. We used the tir-1(OE) 2AWCOFF phenotype as readout to determine if mir-71 acts through the tir-1 3′ UTR to suppress the AWCOFF fate. We made tir-1(OE) sensor constructs by replacing GFP in the GFP sensor constructs (Figure 2B) with tir-1 and crossed transgenic animals expressing each tir-1(OE) sensor construct into mir-71(OE) animals (Figure 2C). The fold change in tir-1(OE) 2AWCOFF phenotype was determined by dividing the 2AWCOFF percentage of tir-1(OE) animals with the 2AWCOFF percentage of their tir-1(OE); mir-71(OE) siblings, which was then normalized to the relative tir-1(OE) transgene copy number determined by qPCR. The higher normalized fold change in tir-1(OE) 2AWCOFF indicates more suppression of 2AWCOFF phenotype by mir-71(OE) in tir-1(OE); mir-71(OE) animals. The normalized fold change in tir-1(OE) 2AWCOFF of tir-1 3′ UTR was significantly higher than that of the tir-1 3′ UTRmut (p = 0.03; Figure 2C). These results suggest that mir-71 suppresses the AWCOFF fate by downregulating tir-1 expression through its 3′ UTR.
To determine if mir-71 is expressed in AWC neurons, we generated transgenic animals expressing YFP under the control of a 2.4 kb promoter upstream of the mir-71 transcript (Figure 1F). The expression of YFP was detected in several head neurons and the body wall muscle in L1 (Figure 3A), which is consistent with previously reported expression pattern of mir-71 [47]–[49]. The mir-71p::YFP transgenic animals were crossed into an odr-1p::DsRed strain, expressing DsRed primarily in AWC and AWB neurons (Figure 3B). YFP was coexpressed with DsRed in AWC and AWB neurons (Figure 3C), suggesting that mir-71 is expressed in these neurons. We found that 52% of animals had visible mir-71p::YFP in both AWC cells, 28% had visible YFP in only AWC left (AWCL), and 20% had visible YFP in only AWC right (AWCR) (Figure 3D). These results suggest that the expression of mir-71, when detected in one of the two AWC neurons, does not have a side bias towards AWCL or AWCR, which is consistent with stochastic choice of the AWCON fate.
We then investigated whether mir-71, when detected in both AWC neurons, has differential expression levels between AWCON and AWCOFF. Transgenic animals expressing mir-71p::GFP, ceh-36p::myr-TagRFP (myristoylated TagRFP marker of AWCON and AWCOFF), and str-2p::2Xnls-TagRFP (nucleus-localized TagRFP marker of AWCON) were generated and analyzed in the L1 stage (Figure 4A, 4A′, 4A″, 4B, 4B′, and 4B″). The ceh-36 promoter is expressed in AWCL, AWCR, ASEL, and ASER [50], [51]. mir-71p::GFP expression was significantly higher in the AWCON cell than in the AWCOFF cell in 71% of the animals (p<0.001; Figure 4C). To confirm this result, we generated transgenic animals expressing mir-71p::NZGFP, odr-3p::CZGFP, and str-2p::2Xnls-TagRFP in which reconstituted GFP (recGFP) [52] expression from two split GFP polypeptides, NZGFP and CZGFP, was restricted mainly in the two AWC cells. Consistent with the mir-71p::GFP result, recGFP expression was significantly higher in the AWCON cell than in the AWCOFF cell in 81% of the animals (p<0.001; Figure S4). Together, these results suggest that mir-71 is expressed at a higher level in the AWCON than in the AWCOFF cell. The higher expression of mir-71 in the AWCON cell is consistent with the role of mir-71 in promoting the AWCON fate.
The suppression of gene expression by mir-71 through the tir-1 3′ UTR (Figure 2B and 2C) and the role of mir-71 in promoting the AWCON fate (Figure 1C and 1E) suggest that gene expression through the tir-1 3′ UTR may be downregulated in the AWCON cell. To investigate this possibility, transgenic animals expressing odr-3p::GFP::tir-1 3′ UTR (GFP reporter of the tir-1 3′ UTR regulation in both AWCs), odr-3p::2Xnls-TagRFP::unc-54 3′ UTR (nucleus-localized TagRFP marker of both AWCON and AWCOFF), and str-2p::myr-mCherry (myristoylated mCherry marker of AWCON) were generated and analyzed in the L1 stage (Figure 4D, 4D′, 4D″, 4E, 4E′, and 4E″). The GFP intensity was normalized to the nucleus-localized TagRFP intensity measured in the same AWC cell to account for variation in focal plane and promoter activity. Normalized GFP intensity was significantly lower in the AWCON cell than in the AWCOFF cell in more than 85% of the animals (p<0.02; Figure 4F). These results suggest that the expression of tir-1 is downregulated in the AWCON cell, consistent with a higher expression level of mir-71 in AWCON and downregulation of tir-1 expression by mir-71.
To determine the site of mir-71 action, mosaic animals in which the two AWC neurons have differential mir-71 activity were used to ask whether mir-71 acts in the future AWCON cell or the future AWCOFF cell. Mosaic animals were generated by random and spontaneous mitotic loss of an unstable transgene expressing the mir-71(OE) construct odr-3p::mir-71 and a mosaic marker odr-1p::DsRed that showed which AWC cells retained the transgene. We specifically looked for the mosaic animals in which only one of the two AWC neurons expressed the mir-71(OE) transgene; this cell was identified by expression of the DsRed marker.
Mosaic analysis was first performed in transgenic lines expressing the mir-71(OE) transgene in a wild-type background. Expression of the mir-71(OE) transgene in both AWC neurons resulted in a 2AWCON phenotype (Figure 5A and 5C). When the mir-71(OE) transgene was retained in only one of the two AWC neurons, the mir-71(OE) AWC neuron became AWCON and wild-type AWC neuron became AWCOFF in the majority of these mosaic animals (p<0.0001; Figure 5B and 5D). This result is consistent with a significant cell-autonomous requirement for mir-71 in the AWCON cell to regulate its identity, which is opposite to the cell autonomous function of tir-1 in regulation of the AWCOFF identity. This result suggests that the AWC cell with higher mir-71 activity can prevent the contralateral AWC cell from becoming AWCON and that mir-71 may play a role in a negative-feedback signal sent from pre-AWCON to pre-AWCOFF. Similar results were obtained from previous mosaic analysis of nsy-4 and nsy-5 [18], [19].
NSY-4 claudin-like protein and NSY-5 gap junction protein are the two parallel signaling systems that antagonize the calcium signaling pathway to specify the AWCON identity [18], [19]. To determine whether mir-71 acts downstream of nsy-4 and nsy-5 to promote AWCON, mosaic analysis was performed with the mir-71(OE) transgene in nsy-4(ky627) and nsy-5(ky634) mutants. Loss-of-function mutations in nsy-4 and nsy-5 caused a 2AWCOFF phenotype (Figure 5E) [18], [19], opposite to the mir-71(OE) 2AWCON phenotype. Overexpression of mir-71 in both AWC neurons significantly suppressed the 2AWCOFF phenotype of nsy-4(ky627) and nsy-5(ky634) mutants. In addition, nsy-4(ky627); mir-71(OE) and nsy-5(ky634); mir-71(OE) animals resembled the mir-71(OE) parent more closely than the nsy-4(ky627) or nsy-5(ky634) parent, but mixed phenotypes were observed (Figure 5E). These results suggest that mir-71 mainly acts at a step downstream of nsy-4 and nsy-5 to promote AWCON. In the majority of the mosaic animals retaining the mir-71(OE) transgene in only one of the two AWC neurons, the mir-71(OE) AWC neuron expressed str-2p::GFP and the other AWC neuron did not (Figure 5F). This significant cell-autonomous requirement for mir-71 in the future AWCON neuron in nsy-4(ky627) and nsy-5(ky634) mutants is the same as in the wild-type background. These results suggest that mir-71 acts cell autonomously downstream of nsy-4 and nsy-5 to promote the AWCON identity.
alg-1 mutants had overaccumulation of premature mir-71 and underaccumulation of mature mir-71, indicating that ALG-1/Argonaute-like protein is required for processing of mir-71 from premature form into the mature form [53]. alg-1(gk214) single mutants had wild-type str-2p::GFP expression. However, alg-1(gk214) significantly suppressed the 2AWCON phenotype of mir-71(OE) and caused a weak 2AWCOFF phenotype in alg-1(gk214);mir-71(OE) animals (p<0.001; Figure 1E). In addition, alg-1(gk214), like mir-71(n4115) mutants, also significantly suppressed the 2AWCON phenotype of tir-1(ky388ts) mutants (p<0.05; Figure 1E). These results suggest that alg-1 is required for mir-71 function in the AWCON cell.
Consistent with previous northern blot analysis [53], we found a significantly reduced level of mature mir-71 in alg-1(gk214) mutants (p<0.001; Figure S5A) using a stem-loop RT-PCR technique designed for specific quantification of mature miRNAs [54]. In addition, mature mir-71 was not detected in mir-71(n4115) mutants (Figure S5B), suggesting that mir-71(n4115) is a null allele. Since mir-71 is expressed broadly in the animal [48], [49] (Figure 3A), we introduced the AWC-expressing transgene odr-3p::mir-71 in mir-71(n4115) mutants and used stem-loop RT-PCR to assay the level of mature mir-71 mainly in AWC cells (Figure S5B).
To determine if the maturation and/or the stability of mir-71 in AWCs is regulated by the signaling molecules that act upstream of tir-1, we assayed the level of mature mir-71 in mir-71(n4115); nsy-4(ky627) double mutants, mir-71(n4115); nsy-5(ky634) double mutants, and mir-71(n4115); unc-36(e251) double mutants containing the AWC mir-71(OE) transgene using stem-loop RT-PCR (Text S1). The level of mature mir-71 was significantly reduced in nsy-4(ky627) (p = 0.015) and nsy-5(ky634) (p<0.0001) mutants compared with control, but was not significantly different between control and unc-36(e251) mutants (Figure S5B). The decreased level of mature mir-71 was not due to reduced transmission rates of the odr-3p::mir-71 transgene (Figure S6A) or downregulation of the odr-3 promoter in nsy-4(ky627) and nsy-5(ky634) mutants (Figure S6B). These results suggest that nsy-4 and nsy-5 are required for the generation and/or the stability of mature mir-71.
To further determine whether nsy-4 and nsy-5 regulate the formation and/or the stability of mature mir-71, we performed stem-loop RT-qPCR to quantify the level of premature and mature mir-71 in mir-71(n4115) mutants, mir-71(n4115); nsy-4(ky627) double mutants, and mir-71(n4115); nsy-5(ky634) double mutants containing the AWC mir-71(OE) transgene. Consistent with stem-loop RT-PCR results (Figure S5B), the abundance of mature mir-71 was significantly decreased in nsy-4(ky627) (p<0.05) and nsy-5(ky634) (p = 0.0003) mutants (Figure 6). However, the level of premature mir-71 was not significantly different between control and nsy-4(ky627) as well as nsy-5(ky634) mutants (Figure 6). These results suggest that the stability, but not the generation, of mature mir-71 is reduced in nsy-4(ky627) and nsy-5(ky634) mutants, and are consistent with a model in which nsy-4 and nsy-5 promotes the stability of mature mir-71 for downregulation of tir-1 in the future AWCON cell (Figure 7).
mir-71 is expressed at a higher level in the AWCON cell than in the AWCOFF cell (Figure 4A–4C), suggesting that mir-71 is differentially regulated at the transcriptional level in the two AWC cells. To determine if nsy-4 and nsy-5 also regulate differential expression levels of mir-71 between the two AWC cells, we crossed the transgene (Figure 4A–4C) containing mir-71p::GFP, ceh-36p::myr-TagRFP, and str-2p::2Xnls-TagRFP into nsy-4(ky627) and nsy-5(ky634) mutants. Since the AWCON marker str-2 is not expressed in nsy-4(ky627) or nsy-5(ky634) mutants, we analyzed and compared the expression levels of mir-71p::GFP between the two AWC cells in the mutants, instead of comparing the expression level between AWCON and AWCOFF (Figure 4A–4C). We found that mir-71 was also differentially expressed between the two AWC cells in nsy-4(ky627) and nsy-5(ky634) mutants (Figure S7), like in wild-type animals. These results suggest that differential regulation of mir-71 transcription in the two AWC cells is not dependent on nsy-4 or nsy-5.
Stochastic cell fate acquisition in the nervous system is a conserved but poorly understood phenomenon [1]. Here, we report that the miRNA mir-71 is part of the pathway that controls stochastic left-right asymmetric differentiation of the C. elegans AWC olfactory neurons through downregulating the expression of tir-1, encoding the TIR-1/Sarm1 adaptor protein in a calcium signaling pathway. In addition, we have linked NSY-4/claudin- and NSY-5/innexin-dependent stability of mature mir-71 to downregulation of calcium signaling in stochastic AWC neuronal asymmetry. Previous studies have identified the role of miRNAs in reproducible, lineage-based asymmetry of the C. elegans ASE taste neuron pair, in which the miRNA expression pattern is largely fixed along the left-right axis [8], [9], [55]. This study provides one of the first insights into miRNA function in stochastic left-right asymmetric neuronal differentiation, in which the miRNA expression pattern is not fixed and is likely regulated by the stochastic signaling event driving random asymmetry.
The seed match between mir-71 and the tir-1 3′ UTR is conserved between C. elegans and C. briggsae. However, the str-2 promoters share little sequence similarity between C. elegans and C. briggsae. The C. elegans str-2 promoter GFP reporter, when expressed in C. briggsae, does not show detectable GFP expression in AWC neurons in embryos, first stage larvae, or adults (data not shown). This result suggests that the transcriptional regulation of str-2 has diverged in C. briggsae.
mir-71 has been implicated in various cell biological and developmental processes including promotion of longevity, resistance to heat and oxidative stress, DNA damage response, control of developmental timing, dauer formation, and recovery from dauer [47], [56]–[60]. However, it is largely unknown how mir-71 functions to regulate these biological processes. RNA interference (RNAi) of tir-1 did not affect C. elegans longevity [61], suggesting that mir-71 may regulate distinct target genes for different functions.
miRNAs are important post-transcriptional and translational regulators of gene expression during development and disease. Several miRNA target prediction algorithms such as MicroCosm Targets, TargetScan, PicTar, and mirWIP provide useful tools with which to identify potential target genes of miRNAs [62]. However, many miRNAs have redundant functions and therefore give subtle or no phenotypes when mutated [37]–[40]. Overexpression approach or phenotypic analysis of miRNA mutants in sensitized genetic backgrounds have been successful in elucidating the role of miRNAs for which null mutants are not available or functional redundancy is a potential problem [5], [8], [38]–[40], [42], [63]–[65]. Using miRNA target prediction programs, we identified mir-71 and five other miRNAs as potential regulators of the calcium-regulated UNC-43 (CaMKII)/TIR-1/NSY-1 (MAPKKK) signaling pathway. Through an overexpression approach and functional analysis of mir-71(n4115) mutants in sensitized genetic backgrounds, we revealed the role of mir-71 in genetic control of the AWCON identity.
miRNAs that share the same sequence identity in their seed regions and could be potentially capable of downregulating the same set of target genes are grouped as members of a family [66]–[69]. Some miRNA family members have been shown to function redundantly and work together to regulate specific developmental processes [37], [38], [70]–[74]. However, many families of miRNAs did not show synthetic phenotypes, indicating that most miRNA families act redundantly with other miRNAs, miRNA families, or non-miRNA genes [38]. Since there is only one mir-71 family member identified, the absence of an AWC phenotype in mir-71(n4115) single mutants suggests that mir-71 may act redundantly with other miRNA family members or non-miRNA genes to regulate calcium signaling in AWC asymmetry. dcr-1, encoding the ribonuclease III enzyme Dicer, is required for processing of premature miRNAs to mature miRNAs [28]. dcr-1(ok247) null mutants had wild-type AWC asymmetry (data not shown). This result suggests that the dcr-1 mutation may cause simultaneous knockdown of several miRNAs (including mir-71) with opposite functions in AWC asymmetry, thereby masking the role of mir-71 and its redundant miRNAs in AWC asymmetry.
The UNC-76 axon guidance molecule and NSY-4 claudin-like protein act to antagonize the calcium-regulated signaling pathway to generate the AWCON identity [12], [19]. We found that mir-71(n4115) mutants significantly suppressed the 2AWCON phenotype of nsy-4(OE) and enhanced the 2AWCOFF phenotype of nsy-4(ky627) and unc-76(e911) mutants. These results suggest an alternative mechanism for functional redundancy of mir-71 in AWC asymmetry. mir-71 may act in parallel with other regulatory pathways downstream of unc-76 and nsy-4 to downregulate the calcium signaling pathway in the AWCON cell. Functional redundancy of miRNAs and other regulatory pathways has been demonstrated by a previous study suggesting that Drosophila miR-7 may act in parallel with a protein-turnover mechanism to downregulate the transcriptional repressor Yan in the fly eye [42].
Our results suggest that mir-71 is regulated at transcriptional and post-transcriptional levels in AWC. At the transcriptional level, mir-71 is expressed at a higher level in the AWCON cell than in the AWCOFF cell. This transcriptional bias of mir-71 is not dependent on NSY-4 claudin-like protein or NSY-5 innexin gap junction protein. The mechanisms that regulate differential expression of mir-71 in the two AWC cells are yet to be elucidated. At the post-transcriptional level, the stability of mature mir-71 is dependent on nsy-4 and nsy-5. It is possible that nsy-4 and nsy-5 may antagonize the miRNA turnover pathway to increase the level of mature mir-71. The C. elegans 5′→3′ exoribonuclease XRN-2 has been implicated in degradation of mature miRNAs released from Argonaute [75]. However, xrn-2(RNAi) animals did not show AWC phenotypes (data not shown), suggesting that the stability of mature mir-71 may be independent of xrn-2.
The TIR-1/Sarm1 adaptor protein assembles a calcium-regulated signaling complex at synaptic regions to regulate the default AWCOFF identity [16]. Downregulation of the TIR-1 adaptor protein by mir-71 and other parallel pathways may represent an efficient mechanism to inhibit calcium signaling in the cell becoming AWCON. Calcium signaling is one of the most common and conserved systems that control a wide range of processes including fertilization, embryonic pattern formation, cell proliferation, cell differentiation, learning and memory, and cell death during development and in adult life [76]. In addition, calcium signaling is implicated in left-right patterning in several tissues of different organisms [77]. It has been shown that negative regulation of calcium signaling by miRNAs is important for normal development and health [78]–[81]. In summary, our study and the studies from other labs demonstrate that downregulation of calcium signaling by miRNAs is one of the important mechanisms for cellular and developmental processes.
Wild-type strains were C. elegans variety Bristol, strain N2. Worm strains were generated and maintained by standard methods [82]. Mutations and integrated transgenes used are as follows: kyIs140 [str-2p::GFP; lin-15(+)] I [12], kyIs323 [str-2p::GFP; ofm-1p::GFP] II [22], oyIs44 [odr-1p::DsRed] V [51], kyIs136 [str-2p::GFP; lin-15(+)] X [12], mir-71(n4115) I [40], nsy-5(ky634) I [18], ppfr-1(tm2180) unc-29(e1072) I (gift from P. Mains, University of Calgary, Canada) [43], rol-6(e187) II, tir-1(ky388ts) III [16], tir-1(ky648gf) III, tir-1(tm3036) III [22], unc-36(e251) III, dcr-1(ok247) III; nsy-4(ky616) IV, nsy-4(ky627) IV [19], unc-43(n498gf) IV, eri-1(mg366 IV), unc-76(e911) V, lin-15b(n744) X, and alg-1(gk214) X.
Transgenes maintained as extrachromosomal arrays include kyEx1127 [odr-3p::nsy-4; myo-3p::DsRed] [18], vyEx149 [odr-3p::mir-71 (25 ng/µl); ofm-1p::DsRed (20 ng/µl)], vyEx187 [mir-71p::YFP (50 ng/µl); elt-2p::CFP (5 ng/µl)], vyEx527, 528 [odr-3p::mir-71 (50 ng/µl); odr-1p::DsRed (12 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx605, 606 [odr-3p::GFP::tir-1 3′ UTR (7.5 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx611, 615 [odr-3p::GFP::unc-54 3′ UTR (7.5 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx647 [odr-3p::GFP::nsy-1 3′ UTR (7.5 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx649, 651 [odr-3p::GFP::tir-1 3′ UTRmut (7.5 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx835, 836, 838 [odr-3p::tir-1::tir-1 3′ UTRmut (70 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx703, 720 [odr-3p::tir-1::tir-1 3′ UTR (70 ng/µl); elt-2p::CFP (7.5 ng/µl)], vyEx905, 907 [odr-3p::mir-74 (50 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx914, 917 [odr-3p::mir-248 (50 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx915, 918 [odr-3p::mir-72 (50 ng/µl); ofm-1::DsRed (30 ng/µl)], vyEx916, 920, 921 [odr-3p::mir-228 (50 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx922, 923, 924 [odr-3p::mir-255 (50 ng/µl); ofm-1::DsRed (30 ng/µl)], vyEx927, 931 [mir-71p::GFP (10 ng/µl); ceh-36p::myr-TagRFP (5 ng/µl); str-2p::2Xnls-TagRFP (25 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx1316, 1317 [mir-71p::NZGFP (30 ng/µl); odr-3p::CZGFP (15 ng/µl); str-2p::2Xnls-TagRFP (25 ng/µl); ofm-1p::DsRed (30 ng/µl)), vyEx1318, 1319 [nsy-5p::mir-248IR (100 ng/µl); odr-1p::DsRed (15 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx1065 [str-2p::myr-mCherry (100 ng/µl); ofm-1p::DsRed (30 ng/µl)], vyEx1097 [odr-3p::2Xnls-TagRFP (40 ng/µl); pRF4(rol-6(su1006) (50 ng/µl)], vyEx1351, 1352 [odr-3p::tir-1 5′ UTR::GFP::tir-1 3′ UTR (15 ng/µl); odr 3p::TagRFP::unc-54 3′UTR (15 ng/µl); elt-2p::CFP (7.5 ng/µl)], and vyEx1353, 1375 [odr-3p::tir-1 5′UTR::GFP::tir-1 3′ UTRmut (15 ng/µl); odr 3p::TagRFP::unc-54 3′ UTR (15 ng/µl); elt-2p::CFP (7.5 ng/µl)].
A 2476 bp PCR fragment of mir-71 promoter was subcloned to make mir-71p::YFP and mir-71p::GFP. mir-71p::NZGFP was made by replacing GFP in mir-71p::GFP with a NZGFP fragment from TU#710 (Addgene) [52]. odr-3p::CZGFP was made by cloning a CZGFP fragment from TU#711 (Addgene) [52] into an odr-3p vector. ceh-36p::myr-TagRFP, in which the 1852 bp ceh-36 promoter drives expression of myristoylated TagRFP, was generated by replacing TagRFP in ceh-36p::TagRFP [83] with myr-TagRFP. odr-3p::2Xnls-TagRFP was made by replacing the str-2 promoter in str-2p::2Xnls-TagRFP [22] with the odr-3 promoter [41]. str-2p::myr-mCherry was generated by replacing GFP in str-2p::GFP [12] with a myr-mCherry fragment. A 94 bp mir-71 PCR fragment was subcloned to make odr-3p::mir-71. A 561 bp PCR fragment of the tir-1 3′ UTR, which represents the average length of the 3′ UTR in the majority of identified tir-1 cDNA clones such as yk1473h08 (www.wormbase.org), was subcloned to make odr-3p::tir-1::tir-1 3′ UTR and odr-3p::GFP::tir-1 3′ UTR. miRNA target prediction algorithms including MicroCosm Targets, PicTar, and mirWIP use 300–590 bp of tir-1 3′ UTR for analysis. The predicted mir-71 binding site, TCTTTC, in the tir-1 3′ UTR was mutated into CAGGCA using QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) to make odr-3p::GFP::tir-1 3′ UTRmut. tir-1a splice form was used for all tir-1 constructs. odr-3p::tir-1 5′ UTR::GFP::tir-1 3′ UTR was made by cloning a 150 bp PCR fragment of tir-1 5′ UTR, amplified from wild-type embryo cDNA, into the odr-3p::GFP::tir-1 3′ UTR. odr-3p::tir-1 5′ UTR::GFP::tir-1 3′ UTRmut was made by replacing GFP::tir-1 3′ UTR in the odr-3p::tir-1 5′ UTR::GFP::tir-1 3′ UTR with GFP::tir-1 3′ UTRmut. To make shRNA anti-mir-248 (mir-248IR), the sense and antisense oligos, each consisting of mir-248 sense (24 nt) and antisense (24 nt) sequences that flank a 12 nt linker (loop) sequence, were designed (SBI System Biosciences) and annealed (IDT) as described. This hairpin construct was subcloned to make nsy-5p::mir-248IR. To generate transgenic strains, DNA constructs were injected into the syncytial gonad of adult worms as previously described [84].
Z-stack images of transgenic animals expressing fluorescent markers were acquired using a Zeiss Axio Imager Z1 microscope equipped with a motorized focus drive and a Zeiss AxioCam MRm CCD digital camera. All animals of each set of experiments had the same exposure time for comparison of fluorescence intensity. The single focal plane with the brightest fluorescence in each AWC cell was selected from the acquired image stack and measured for fluorescence intensity. To measure fluorescence intensity, the outline spline tool in the Zeiss AxioVision Rel 4.7 image analysis software was used to draw around the AWC cell body (Figure 2B; Figure 4A, 4B, 4D, 4E; Figure S4A, S4B; and Figure S6B) or nucleus (Figure 2B, Figure 4D′ and 4E′) from captured images. To measure fluorescence intensity in dim GFP-expressing cells (Figure 4B and Figure S4B), the display contrast and brightness were adjusted to visualize and outline the cells. For each category of animals, images from a minimum of 10 animals were collected and analyzed.
Mosaic analysis was performed as previously described [13], [18], [19], [25]. Transgenic lines expressing the odr-3p::mir-71; odr-1p::DsRed transgene were passed for minimum of six generations before scoring for mosaic animals. The same transgenic lines were crossed into nsy-4(ky627) and nsy-5(ky634) mutants for the analysis.
Three adult hermaphrodites from each tir-1(OE) transgene line were collected in 25 µl of worm lysis buffer (50 mM KCl, 0.01% gelatin, 10 mM Tris-HCl pH 8.3, 0.45% Tween 20, 0.45% NP-40, 2.5 mM MgCl2, 100 µg/ml Proteinase K). Collected worms were then incubated at −80°C for minimum of one hour, 65°C for one hour, and 95°C for 15 minutes. 5 µl of the worm lysate was used for subsequent qPCR with Fast SYBR Green Master Mix (Invitrogen). qPCR reactions were run in triplicate at 95°C for 3 minutes, followed by 45 cycles of 95°C for 30 seconds, 57°C for 30 seconds, and 72°C for 30 seconds on the CFX96 Real-Time PCR Detection System (Bio-Rad). PCR product was scanned for fluorescent signal at the end of each cycle and the C(T) values were obtained using the CFX Manager Software (Bio-Rad). The relative tir-1(OE) transgene copy number was determined using the 2[−Delta Delta C(T)] method as previously described [85] with the actin-related gene, arx-1, as internal control.
Stem-loop RT-qPCR was performed as described [54] to detect and quantify relative expression levels of premature and mature mir-71. The odr-3p::mir-71 transgenes used in genetic mosaic analysis were crossed into various genetic backgrounds. Total RNA samples were isolated from first stage larvae using RNeasy Mini kit (QIAGEN). Reverse transcription (RT) reactions were performed with 1 µg of total RNA, SuperScript III reverse transcriptase (Invitrogen), and RT primer (oligo d(T)18, premature mir-71 stem-loop RT primer, or mature mir-71 stem-loop RT primer). 1 µl of 1∶35 diluted reverse transcription product was used as template for subsequent qPCR reactions with Fast SYBR Green Master Mix (Invitrogen). All PCR reactions were run in triplicate at 95°C for 3 minutes, followed by 45 cycles of 95°C for 30 seconds, 51°C for 30 seconds, and 72°C for 30 seconds on the CFX96 Real-Time PCR Detection System (Bio-Rad). PCR product was scanned for fluorescent signal at the end of each cycle and the C(T) values were obtained using the CFX Manager Software (Bio-Rad). The actin-related gene, arx-1, was used as internal control to normalize variation between samples. Relative expression of premature and mature mir-71 was analyzed using the 2[−Delta Delta C(T)] method as previously described [85]. Relative expression was set to one for mir-71(n4115); odr3p::mir-71 and was normalized accordingly for other samples. Student's t-test was used to calculate statistical significance.
|
10.1371/journal.pgen.1003549 | The Gene Desert Mammary Carcinoma Susceptibility Locus Mcs1a Regulates Nr2f1 Modifying Mammary Epithelial Cell Differentiation and Proliferation | Genome-wide association studies have revealed that many low-penetrance breast cancer susceptibility loci are located in non-protein coding genomic regions; however, few have been characterized. In a comparative genetics approach to model such loci in a rat breast cancer model, we previously identified the mammary carcinoma susceptibility locus Mcs1a. We now localize Mcs1a to a critical interval (277 Kb) within a gene desert. Mcs1a reduces mammary carcinoma multiplicity by 50% and acts in a mammary cell-autonomous manner. We developed a megadeletion mouse model, which lacks 535 Kb of sequence containing the Mcs1a ortholog. Global gene expression analysis by RNA-seq revealed that in the mouse mammary gland, the orphan nuclear receptor gene Nr2f1/Coup-tf1 is regulated by Mcs1a. In resistant Mcs1a congenic rats, as compared with susceptible congenic control rats, we found Nr2f1 transcript levels to be elevated in mammary gland, epithelial cells, and carcinoma samples. Chromatin looping over ∼820 Kb of sequence from the Nr2f1 promoter to a strongly conserved element within the Mcs1a critical interval was identified. This element contains a 14 bp indel polymorphism that affects a human-rat-mouse conserved COUP-TF binding motif and is a functional Mcs1a candidate. In both the rat and mouse models, higher Nr2f1 transcript levels are associated with higher abundance of luminal mammary epithelial cells. In both the mouse mammary gland and a human breast cancer global gene expression data set, we found Nr2f1 transcript levels to be strongly anti-correlated to a gene cluster enriched in cell cycle-related genes. We queried 12 large publicly available human breast cancer gene expression studies and found that the median NR2F1 transcript level is consistently lower in ‘triple-negative’ (ER-PR-HER2-) breast cancers as compared with ‘receptor-positive’ breast cancers. Our data suggest that the non-protein coding locus Mcs1a regulates Nr2f1, which is a candidate modifier of differentiation, proliferation, and mammary cancer risk.
| Most non-Mendelian disease variants identified through genome-wide association studies are low-penetrance, common in the population and located in non-protein coding genomic loci. It is currently unknown how these loci modulate disease risk. Insights in their mechanisms could lead to the development of novel prevention or early intervention strategies. We used comparative genetics to model such loci in a rat model for breast cancer susceptibility. For the Mcs1a locus presented in this paper, we describe its non-protein coding localization and the mechanism through which it affects mammary carcinoma susceptibility, involving transcriptional regulation of the orphan nuclear receptor gene Nr2f1/Coup-tf1 and mammary epithelial cell proliferation/differentiation. In addition, we show that low NR2F1 transcript levels are associated with upregulation of cell cycle-related genes and high histological grade (grade 3, poorly differentiated, highly proliferative) in human breast cancers, including triple-negative therapy-resistant breast cancers. Our findings highlight the orphan nuclear receptor NR2F1 as a novel target for breast cancer prevention and/or intervention strategies. Since COUP-TFs (including NR2F1) are nuclear hormone receptors, whose crystal structure suggests these are ligand controlled, identification of the ligand for NR2FI could provide a potential breast cancer therapeutic.
| An important indicator for breast cancer risk is the family history, suggesting a strong genetic component in breast cancer susceptibility [1]. The heritable portion of a woman's risk to breast cancer consists of numerous risk-increasing and risk-decreasing alleles. Through familial linkage studies in the 1990s, deleterious mutations affecting the coding regions of well-known tumor suppressor genes, i.e. BRCA1 and BRCA2, were found to associate with increased breast cancer risk [2], [3]. Such mutations are rare in the population. More recently, genome-wide association studies (GWAS) have been employed to discover association of common variants with breast cancer susceptibility. GWAS have proven to be successful at uncovering loci harboring low-penetrance breast cancer susceptibility variants [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14]. In the future, the identification of these variants will likely impact population-based risk prediction [15], [16]. Many of the variants are located in non-protein coding areas of the genome, such as promoters, introns and intergenic areas with large genomic regions without known genes, called gene deserts. It is anticipated that these variants are involved in gene regulation as exemplified by breast cancer susceptibility-associated variant rs2981582 that is correlated with FGFR2 transcript levels in breast cancer [17] and normal breast tissue [18]. For other non-protein coding breast cancer-associated variants, their spatiotemporal regulatory function and gene targets are largely undefined. Moreover, mechanisms underlying common breast cancer-associated variants on the level of the mammary gland tissue and mammary epithelial cells (MECs) are unknown.
MEC proliferation and differentiation are strongly interconnected processes for which ample evidence exists that these are involved in the development of breast cancer. Recently, flow cytometry-based approaches have yielded many markers that aid in the understanding of the hierarchical order of MEC differentiation [19]. For mouse mammary epithelial cells (MMECs), luminal and basal/myoepithelial populations were identified based on expression of heat stable antigen (HSA; CD24) and β1 integrin (CD29), or HSA and α6 integrin (CD49f). Specific cells of the basal lineage expressing high levels of CD29 or CD49f and high levels of CD24 were shown to harbor repopulating ability in single cell transplantation assays in mice, suggesting the presence of bipotential mammary stem/progenitor cell activity [20], [21]. Markers for lineage-specific progenitor cells have also been identified in the mouse, including CD61, which enriches for mouse luminal progenitors [22]. Later, additional markers for luminal progenitors have been identified, which include c-kit and ALDH [23], [24]. We have reported a flow cytometry-based approach to identifying the luminal and basal/myoepithelial cell lineages in the rat mammary gland [25]. In the early 1990s, clonogenic rat mammary epithelial cells (RMECs) were found to stain with peanut lectin [26], [27]. There is strong interest in the biology of mammary stem/progenitor cells as these are thought to be the target cells for tumorigenic transformation events, mainly because of their immortality and ability to sire many generations of daughter cells. Interestingly, human germ line mutation of BRCA1 has been shown to stimulate luminal-to-basal tumor formation by affecting the luminal progenitor cell pool and luminal cell fate [28], [29], exemplifying a consequence of the involvement of breast cancer susceptibility variants in MEC differentiation.
It is currently unclear if other (e.g. low-penetrance, non-protein coding) breast cancer-associated variants affect MEC proliferation and differentiation. In order to model breast cancer susceptibility loci in a mammalian organism, we have conducted a rat-human comparative genetics approach. Upon initiation of this approach we selected the rat mammary carcinogenesis model, as the arising mammary carcinomas well reflect specific aspects of human breast adenocarcinoma, i.e. staged progression and ovarian hormone responsiveness. The advantage of the rat-human comparative genetics approach is that the availability of mammalian genetic model organisms aids in the dissection of the mechanisms underlying the susceptibility loci [30]. The rat mammary carcinoma resistance quantitative trait locus (QTL) Mcs1 was identified in the backcross progeny of an intercross between the resistant Copenhagen (Cop) and susceptible Wistar-Furth (WF) parental inbred rat strains [31], [32]. Physical confirmation of this resistance locus was presented in a study using a congenic recombinant inbred line having a large portion of the original Mcs1 QTL from the Cop strain introgressed onto the WF genetic background [33]. Congenic rats harboring a homozygous Mcs1 Cop allele had a 85% reduction of 7,12-dimethylbenz(a)anthracene (DMBA)-induced mammary carcinoma multiplicity as compared with congenic control animals homozygous for the susceptible WF Mcs1 allele. Testing various other congenic lines with smaller Cop Mcs1 portions on the WF genetic background revealed that the initial Mcs1 QTL harbors three modifier loci of mammary carcinoma susceptibility, namely Mcs1a, Mcs1b and Mcs1c [33]. In this study, we describe the congenic fine-mapping of the Mcs1a critical interval to a ∼277 Kb region entirely embedded within a large gene desert on rat chromosome 2. Using a congenic rat mammary gland transplantation assay, we show that the Mcs1a locus controls DMBA-induced mammary carcinoma development in a mammary cell-autonomous manner. While the rat mammary carcinogenesis model has proven value to study certain aspects of breast cancer etiology, complex genome-engineering technology for the rat is still under development. Since Mcs1a shows good evolutionary conservation to human, mouse and other mammalian species, we describe the genetic engineering of a novel megadeletion (MD) model in the mouse. Homozygous MD mice lack a large piece of the gene desert including the region orthologous to the Mcs1a critical interval. Taking advantage of both rodent genetic model systems we found an effect of this non-protein coding locus on MEC proliferation/differentiation and identified the orphan nuclear factor Nr2f1/Coup-tf1 as the Mcs1a target gene. To investigate its translational potential we analyzed NR2F1 transcript levels in available global gene expression data for human breast cancers. We show the correlation of low NR2F1 transcript levels with high-grade and discuss the implication of this finding for human breast cancer. Using this rat-mouse-human comparative genetics approach we identified Nr2f1 as a novel gene target for the development of breast cancer prevention or therapeutic strategies.
We previously showed that the mammary carcinoma resistance allele Mcs1a from the Cop inbred strain when introgressed onto the susceptible WF inbred genetic background reduced DMBA-induced mammary carcinoma multiplicity by ∼50%, as compared with the susceptible congenic control line, not carrying the Cop Mcs1a allele [33]. Using multiple additional congenic lines, we now present further fine-mapping of the interval conferring the reduction in DMBA-induced mammary carcinoma multiplicity phenotype (Figure 1A). The resistant congenic lines have a significantly (P<0.001) lower mammary carcinoma multiplicity as compared with the susceptible congenic control line (WF.Cop; Figure 1B). The susceptible congenic lines have a mammary carcinoma multiplicity not different from the susceptible congenic control line (P>0.2). The resistant congenic lines together with the susceptible congenic lines V5 define the Mcs1a critical interval as a ∼277 Kb genomic region located in a gene desert on rat chromosome 2. The gene desert is flanked by Nr2f1 at the proximal side (Figure 1A) and Arrdc3 at the distal side (outside the window in Figure 1A).
Using resistant congenic lines W4 and W5, susceptible congenic line R5 and the susceptible congenic control line WF.Cop, we tested if Mcs1a also confers resistance to N-methyl-N-nitrosourea (MNU)-induced mammary carcinogenesis. The resistant congenic line W5 showed a decreased mammary carcinoma multiplicity phenotype (P = 0.008) as compared with WF.Cop and the resistant congenic line W4 showed a strong trend (P = 0.08) towards a decreased MNU-induced mammary carcinoma multiplicity. The susceptible congenic line R5 was not different from the WF.Cop line (Figure 1C). Subsequently, carcinoma multiplicities following mammary ductal infusion of retrovirus expressing the activated HER2/neu oncogene were determined [34]. In this assay, the resistant congenic line R3 had a significantly reduced mammary carcinoma multiplicity (P = 0.04) as compared with the susceptible congenic line A4 (Figure 1D). These data show that Cop inbred strain-derived alleles of the Mcs1a locus (that include the smallest critical interval) introgressed on the susceptible genetic background confer resistance to three distinctly acting mammary carcinogenic treatments. These data suggest that the resistance mechanism likely manifests beyond the stage of (carcinogen-specific) cancer initiation.
To ask if Mcs1a acts via a mammary cell-autonomous mechanism, a mammary gland transplantation assay was carried out. Mammary gland tissue from donor animals of the susceptible inbred WF rats or the Mcs1a resistant congenic line Y4 was transplanted into the interscapular white fat pads of recipient animals with the same genotype or F1 animals of an intercross between WF and Y4. A total of 228 transplantations were performed, of which only 2 failed to produce a mammary outgrowth. For the 4 transplant groups (donor to recipient; susceptible to susceptible S:S, susceptible to F1 S:F1, resistant to F1 R:F1, and resistant to resistant R:R) carcinoma development following DMBA exposure was monitored (Table 1). The mammary carcinoma incidence at the transplant site was 41%, 36%, 13%, and 9% for transplant groups S:S, S:F1, R:F1, and R:R, respectively. Logistic regression analysis revealed that donor genotype (P = 0.0024), but not recipient genotype (P = 0.59) was significantly associated with transplant site carcinoma development (Table 1). The interaction between donor and recipient genotype was not significant (P = 0.44) for the dependent variable mammary gland transplant carcinoma susceptibility (Table 1). These data demonstrate that the mammary carcinoma susceptibility phenotype mediated by Mcs1a is transferable by transplantation of the mammary gland, indicating that Mcs1a modulates susceptibility in a mammary cell-autonomous manner.
Considering the evolutionary conserved nature of the locus, we sought to genetically engineer a mouse model for the rat Mcs1a locus. In mouse ES cells, we employed a MICER vector-assisted double targeting strategy to insert (on the same chromosome) loxP sites at either side of the gene desert region orthologous to the rat Mcs1a critical interval that was known at the time of design (Figure 2A). Both targeting steps were checked for proper integration of the MICER construct by Southern blot analysis. The proximally located MICER construct harbors the 3′ half (exons 3–9) of the Hprt gene and the distal construct harbors the 5′ half (exons 1–2) of Hprt that upon proper Cre-lox recombination form a functional Hprt gene. Following Cre-recombinase transfection and Hprt selection, the MD mutation was created and the mouse model was generated through blastocyst injections of karyotypically normal ES cells. After germ line transmission of the mutation, homozygous mutants were obtained and tested for lack of the 535 Kb targeted region in the Mcs1a orthologous gene desert on mouse chromosome 13 (Figure 2A). PCR tests using 4 different primer combinations within the deleted sequence and 2 primer combinations spanning the deletion showed consistent results that the region is indeed deleted. An example of a genotyping gel image is shown in Figure S1. The mutation was transferred through 10 generations of breeding to 2 inbred genetic backgrounds, namely FVB/N (FVB) and C57Bl/6 (B6).
Homozygous MD mice of both genetic backgrounds are viable and litter sizes are normal as compared with wild type (WT) animals, suggesting there is no embryonic or neonatal lethality associated with the mutation. Groups of homozygous MD and WT mice were monitored for obvious phenotypes. There was no difference in body weight up to 1 year of age and life span was not affected during the same period. An obvious phenotype we noticed was delayed eyelid opening of homozygous MD mice on both genetic backgrounds (Figure 2B, shown for FVB). While all WT animals had both eyelids completely open by 17 days of age, this was observed for only 65% of homozygous MD mice. For some animals, the closed eyelid phenotype persisted for months (unpublished data).
It is anticipated that a large portion of the non-protein coding capacity of the genome may be involved in the spatiotemporal regulation of gene expression [35]. For many non-coding elements, the target genes of regulation are unknown. We performed a global gene expression study by RNA-seq on mammary gland RNA samples from MD and WT mice (FVB). First, the quality-filtered reads were mapped to the mouse genome. We focused on reads mapping to the Mcs1a-associated gene desert to check for putative unknown transcripts located within the deleted region. We found only 6 reads from the WT samples aligning to the MD region, suggesting that no highly expressed unknown transcript exists within the region. As expected, no reads from the MD samples aligned to the deleted region.
Next, the reads were mapped to the mouse Ensembl reference set of 82,508 transcripts (annotated to 31,034 genes) using Bowtie [36]. Relative transcript abundance was determined using the RSEM algorithm [37]. For the detection of differential gene expression between MD and WT samples, the edgeR package was used [38]. First, we looked at the levels of transcripts within 2.5 Mb of either side of the Mcs1a-associated gene desert (Figure 2C). The only transcript with significantly different levels between the MD and WT samples was Nr2f1 (P<0.001). This gene is located adjacent to the gene desert at a genomic distance of approximately 800 Kb from the Mcs1a orthologous locus. To verify that Nr2f1 is indeed a target gene, we used TaqMan gene expression assays on additional mammary gland samples. Nr2f1 was found to be downregulated by more than 80% in the MD samples as compared with the WT samples (Figure 2D, P<0.001). We also checked Nr2f1 transcript levels in three other tissues. In thymus and ovary, we found that Nr2f1 transcript levels are greatly reduced (Figure S1, P<0.001), to similar levels as the mammary gland. In the liver, however, Nr2f1 transcript levels were not significantly different between MD and WT samples (Figure S1, P = 0.27), suggesting that there is some tissue-specificity in this regulatory mechanism.
Considering Nr2f1 as the main target of the Mcs1a locus, we also looked at its transcript levels in mammary glands (MG), rat mammary epithelial cells (RMECs) and mammary carcinomas (carc.; induced by DMBA and MNU) from susceptible congenic control (WF.Cop) and Mcs1a resistant congenic rats. The resistance allele was provided by the W4 or W5 congenic lines. Since the W4 and W5 lines did not differ significantly, data from both lines was included. For the RMECs, only data for the W4 resistant congenic line was obtained. The Mcs1a resistance allele was associated with increased Nr2f1 levels in mammary gland (P = 0.01; Figure 3A) and RMEC (P = 0.02; Figure 3B). Both DMBA- and MNU-induced carcinomas from Mcs1a resistant congenic animals had strongly increased Nr2f1 transcript levels, as compared with DMBA- and MNU-induced carcinomas from susceptible control congenic rats (P<0.001; Figure 3C).
As the Mcs1a locus is located at a genomic distance of over 800 Kb from Nr2f1, we asked if a chromatin looping structure exists that would support such long distance regulation. The chromosome conformation capture (3C) assay was developed to detect higher-order chromatin interactions for any locus of interest [39], [40]. To apply this methodology and investigate higher-order chromatin interactions between Mcs1a and Nr2f1, RMECs were fixed using formaldehyde to crosslink proteins and DNA, thus capturing interacting chromatin fragments. Crosslinked chromatin was digested using the BglII restriction enzyme and ligated in a large volume (of 7 ml). The large volume ligation reaction reduces random ligations and favors ligations of crosslinking-captured DNA fragments. Ligation events were detected and quantified using PCR and agarose gel electrophoresis. The PCR detection assay was designed such that the fixed primer was located in the rat Nr2f1 promoter (Figure 3D). The experimental primers were located within the Mcs1a critical interval (Table S2), overlapping with areas of the strongest evolutionary conservation (Figure 3D). Each experimental primer was tested in combination with the fixed primer on the 3C templates and a positive control (BAC-derived) template. The relative interaction frequency of two chromatin fragments represented by a primer pair equals the PCR signal intensity of the RMEC 3C template relative to that of the positive control template. We found an area of multiple BglII restriction fragments with increased relative interaction frequencies above background levels (P<0.05) to exist within the Mcs1a critical interval (Figure 3D). The main peak coincides with genetic elements of the highest evolutionary conservation present in Mcs1a. These findings suggest that a putative regulatory element within Mcs1a forms a higher-order chromatin structure with the Nr2f1 promoter over 820 Kb of genomic sequence (Figure 3E). None of the interactions is significantly different between the susceptible and resistant Mcs1a genotypes, suggesting that the DNA-binding proteins facilitating the interactions do not involve polymorphic sites. To identify genetic variants within and in the vicinity of the looped fragments that may explain Nr2f1 transcript regulation, we resequenced approximately 12.5 Kb of the genomic region involved in the higher-order chromatin structure in the WF and Cop parental inbred strains and found 17 genetic variants (Figure S2; Table S3). For only 1 variant, the resistance (Cop) allele (a 14 bp deletion) is predicted to disrupt a rat-mouse-human-conserved binding motif, namely COUP-TF (V$Coup_01; Figure S2). As the Mcs1a resistance allele is associated with increased Nr2f1/Coup-tf1 expression (Figure S2), it is possible that the 14 bp deletion polymorphism of the Mcs1a resistance allele omits a Nr2f1 self-repressive gene expression modulatory function that acts through the intact COUP-TF binding motif on the susceptible Mcs1a allele. Such regulatory function would have to be investigated in detail in the future.
Since Mcs1a is mammary cell-autonomous, we asked if the locus has an effect on MEC biology, i.e. proliferation and differentiation. First, we tested for differential repopulating ability of RMECs from susceptible congenic control (WF.Cop) and Mcs1a resistant congenic animals. The resistance allele was provided by the W4 or W5 congenic lines. Differential repopulating ability could be indicative of a quantitative or functional difference in the mammary stem cell pool potentially underlying the susceptibility phenotype. Freshly isolated RMECs from Mcs1a resistant congenic animals and susceptible congenic control animals were grafted into the interscapular white fat pads of recipient animals of the same genotype. A dilution series of 250, 500, 1000, 2000, 4000 and 8000 cells was tested (Figure 4A). Six weeks after transplantation, the interscapular fat pads were harvested and scored for presence of mammary ductal structures as previously described [41]. The repopulating ability (determined by the estimated number of cells required to give 50% outgrowth occurrence) of RMECs from the susceptible congenic control animals was found not to be different (P>0.05) than that of the Mcs1a resistant congenic animals of either line W4 or W5. Another statistical approach was taken to seek for a possible difference in outgrowth potential at each cell number individually between the susceptible and resistant (W4 and W5 combined) genotypes. Therefore, Chi-square tests for independent distributions in a 2×2 contingency matrix were conducted. At all cell numbers the outgrowth potential for the resistant genotype was not different (P>0.05) than that of the susceptible genotype. These data suggest that the Mcs1a allele does not affect mammary stem cell activity.
In the next experiment, we tested the colony-forming ability in Matrigel of a purified population of clonogenic RMECs from Mcs1a resistant congenic (line W4) and susceptible congenic control (WF.Cop) animals. This assay tests the proliferating potential of the clonogenic RMEC pool. Freshly isolated single RMECs were antibody-stained and sorted using FACS. Gating strategies were used to exclude hematopoietic cells (CD45) and endothelial cells (CD31). From the RMEC-enriched (CD31–CD45−) population, the luminal cells (CD24hiCD29med) were selected and separated based on staining with anti-CD61 and peanut lectin (PNL; Figure 4B). We chose to focus on the luminal population, since CD61hi luminal cells have previously been identified in the mouse as the luminal progenitor population [22] and PNLhi clonogenic rat cells were previously found to overlap largely with the luminal population [25]. The colony-forming ability of 3 sorted cell fractions was tested, namely luminal CD61hiPNL+, CD61+PNLhi and CD61+PNL+ (Figure S3). From each sorted fraction, 10,000 cells were plated in Matrigel for each well. After 10 days of culturing, the colony-forming ability was determined by counting the spherical mammary colonies (Figure 4C, lower panel). For every sorted sample, we found that the CD61+PNLhi cell fraction had a ∼6-fold (P<0.001) increased colony-forming ability and the CD61hiPNL+ cell fraction a ∼1.6-fold (P = 0.02) decreased colony-forming ability, as compared with the CD61+PNL+ cell fraction (Figure S3), verifying that in rats PNLhi luminal MECs are enriched with progenitor cells. When comparing the susceptible congenic control and the Mcs1a resistant congenic samples, we found that the CD61+PNLhi cell fraction from the susceptible congenic control samples had a more than 2-fold increased colony-forming ability (Figure 4C; P = 0.02). These results indicate that the susceptible congenic control luminal RMEC fractions have increased proliferative capacity as compared with the Mcs1a resistant congenic RMEC fractions.
We also looked for quantitative differences in RMEC differentiation between susceptible congenic control (WF.Cop) and the Mcs1a resistant congenic animals. The resistance allele was provided by the W4 or W5 congenic lines. Fresh RMECs were obtained and stained for FACS analysis, as described previously [25]. The Mcs1a resistant congenic animals have more luminal (P = 0.02) and less basal (P = 0.02) RMECs as compared with susceptible congenic animals (Figure 5A), which significantly shifts the luminal-to-basal ratio by 34% (P = 0.008). The abundance of PNLhi or CD61hi subpopulations (Figure 5B) as well as the mean fluorescence intensities (not shown) of these stainings were not different between susceptible and resistant Mcs1a congenic animals. In addition, we looked at MMEC differentiation in WT and MD mice by FACS analysis, as described previously [42]. Like for the rat, we used antibodies against CD45 and CD31 to exclude hematopoietic and endothelial cells, respectively, and antibodies against CD24 and CD29 to quantify luminal and basal MMEC populations (Figure 5C). Like for Mcs1a congenic resistant and susceptible control rats, we found a shift of 36% in luminal-to-basal ratio between WT and MD mice. MD mice had a less abundant luminal and a trend towards a more abundant basal population (Figure 5C, P<0.001 and P = 0.10 for luminal and basal, respectively). Similar results were obtained for the MD mutation on both genetic backgrounds (FVB, B6). We also performed this analysis based on CD29 and CD61 expression to quantify mature luminal (ML, CD29medCD61−), luminal progenitors (LP, CD29medCD61+) and mammary stem/progenitor cells (MaSc, CD29hiCD61+). As compared with WT mice, the MD mice had a less abundant mature luminal population (P<0.001), a trend towards a less abundant luminal progenitor population (P = 0.09) and no significantly different mammary stem/progenitor cell population (P = 0.26; Figure 5D), suggesting that the Mcs1a-associated gene desert locus primarily affects luminal differentiation. Both MD mice and susceptible congenic control rats have a lower abundance of luminal cells and lower Nr2f1 transcript levels, as compared with WT mice and Mcs1a resistant congenic rats, respectively, suggesting that Nr2f1 modulates luminal MEC differentiation.
Next, we tested if short interfering RNA (siRNA)-mediated knockdown of Nr2f1 transcript levels in cultured human breast cells is capable of directly influencing the cellular differentiation pattern. The human breast cancer cell line MCF7 and breast epithelial cell line MCF10A were transfected with siRNAs against NR2F1 (siNR2F1) and non-targeting control siRNA (siCONTROL). No morphological differences were observed between cells transfected with siNR2F1 or siCONTROL. At 40 hours after transfection, cells were harvested for Nr2f1 expression analysis and stained for FACS analysis with fluorescently labeled antibodies against commonly used markers of MEC differentiation, namely CD24, CD29, CD44 and CD49f. As expected, the siNR2F1-treated cells have an over 2-fold reduction of NR2F1 transcript level, as compared with siCONTROL-treated cells (Figure S4). In both MCF7 and MCF10A we found that NR2F1 knockdown upregulated CD24, as compared with treatment of the cells with non-targeting siRNAs. None of the other markers of differentiation was affected by NR2F1 knockdown (Figure S4), suggesting that Nr2f1 transcript levels have a direct effect on cellular differentiation through upregulation of CD24.
In the global RNA-seq expression analysis, 1,531 genes were found to be differentially expressed (DE) between the mammary gland samples from MD and WT (FVB) mice (Table S4). Of these, 412 genes have annotated 1-1-1 mouse-rat-human orthologues. Nr2f1 is listed in the top 10 genes with the lowest P-value and is the top of the list of genes with 1-1-1 mouse-rat-human orthologues (Table S4). We applied a gene expression correlation clustering analysis using the 412 DE genes with 1-1-1 orthologues. The DE genes mainly clustered into three groups and Nr2f1 is found in the first group (Figure 6A). To functionally annotate the groups of correlated genes, two online gene ontology (GO) category enrichment calculation tools were used, namely the Gene Ontology enRIchment anaLysis and visuaLizAtion tool (GOrilla) and the Database for Annotation Visualization and Integrated Discovery (DAVID) [43], [44]. The Nr2f1 containing group is weakly enriched for genes related to cell migration, the extracellular matrix and innate immunity/inflammation (Table S5). The second group of strongly correlated genes was found not to correlate or anti-correlate with groups 1 and 3. This group was enriched for genes related muscle contractile function (Table S5). Group 3 is anti-correlated to the Nr2f1-containing group and is strongly enriched in genes related to the cell cycle, proliferation and DNA-damage response (Table S5). These results implicate that reduced Nr2f1 transcript levels in the MD mammary gland is associated with an increased expression of cell cycle-related genes, which may render the mammary gland in a more proliferative state.
From the publicly available breast cancer global gene expression study GSE3494 containing data for 243 breast cancers [45], we selected 412 human probe sets from the Affy U133a array that are annotated to correspond to the 412 mouse DE genes. By performing the same correlation clustering analysis, an expression correlation pattern was identified to consist of 3 groups (Figure 6B). Group 1, again, is the NR2F1-containing group, weakly enriched with genes involved in developmental processes such as brain segmentation and morphogenesis (Table S6). Group 2 is now a large group that can be split up into subgroup 2a and 2b. Group 2 as a whole and subgroup 2b are strongly enriched with muscle contraction-related genes, whereas subgroup 2a is weakly enriched for muscle protein- and extracellular-related genes and is considered to be a mix between group1 and 2. Similar to the mouse mammary gland correlation analysis, group 3 is very strongly enriched for cell cycle/proliferation and DNA-damage response-related genes and is anti-correlated to the NR2F1-containing cluster (Table S6), suggesting that also in human breast cancer low NR2F1 transcript levels are associated with an increase in cell cycle-related gene expression.
Next, we looked if the similarities in gene expression patterns between the mouse MD mammary gland and the human breast cancer data set hold up within the genome-wide data sets. From the GSE3494 global gene expression study, we selected 9,828 human probe sets from the Affy U133a array that are annotated to have 1-1-1 mouse-rat-human orthologues. First, we checked for similarities between the human and mouse data sets in gene lists anti-correlated to Nr2f1 transcript levels. From the list of 126 human genes in the clustered group of genes anti-correlated to NR2F1 transcript levels (Group 3, Table S6), 51 and 64 genes were found to be present in the top 100 and 200 anti-correlated genes to the Nr2f1 transcript level in the mouse study, respectively. The probability that 51 or 64 of the anti-correlated genes (Group 3) would be in the top 100 or 200 anti-correlated from all 9,828 genes by chance would be <10−74 or <10−78, respectively, suggesting high similarity in genes anti-correlated to NR2F1/Nr2f1 between human breast cancer and the mouse MD mammary gland. Similar analysis for the 37 genes in the NR2F1-containing cluster (Group 1, Table S6) yielded 3 and 5 genes in the top 100 and 200 genes correlated to the Nr2f1 transcript level, which translates to a probability of this occurring randomly of 0.000571 and 0.000106, much higher than the probabilities for the anti-correlated genes.
We also functionally explored the genes most correlated and anti-correlated to NR2F1/Nr2f1 in both the human and mouse data set, regardless of the occurrence of the genes in the mouse DE gene-based cluster analysis. From the 59 genes with strongest anti-correlation (r<−0.3) to NR2F1 in the human breast cancer data set, 41 (69%) were also found to be among the strongest anti-correlated (r<−0.3) to Nr2f1 in the mouse MG data set, which is 2-fold enrichment in comparison to strongly anti-correlated genes (r<−0.3) to Nr2f1 in the entire mouse data set (34%). These 41 genes are found to be strongly enriched with cell cycle/proliferation-related genes (Table S7). From the 297 genes with strongest correlation (r>0.3) to NR2F1 in the human breast cancer data set, 64 (22%) were also found to be the strongest correlated (r>0.3) to Nr2f1 in the mouse MG data set, which is 1.3-fold enrichment in comparison to equally strongly correlated genes (r>0.3) to Nr2f1 in the entire mouse data set (17%). These 64 genes are found to be enriched with genes involved in a wide variety of processes, including extracellular matrix and developmental processes, as well as signaling pathways (Table S7). These analysis suggest that the human breast cancer and the MD mouse MG gene expression data sets are particularly similar in genes anti-correlated with NR2F1/Nr2f1, which are strongly enriched with genes involved in cell cycle/proliferation.
Proliferation gene signatures have been explored for usage as prognostic markers in breast tumor expression studies [46]. Upregulation of such genes in breast cancer is generally indicative of poor prognosis [47], [48]. In a study to identify a novel gene list for “breast cancer intrinsic” subtype classification, a 20-gene proliferation signature was found to form one of the predictive modules [49]. Of these 20 genes, we found 14 to have human-rat-mouse orthologues of which 10 were DE in the mouse MG RNA-seq study with all 10 genes upregulated in the MD samples. The probability of selecting by chance 10 of these 14 proliferation signature genes into the 412 mouse DE gene set out of a total of 9,828 genes is lower than 10−11. This result suggests that the MD mammary gland gene expression profile (with low Nr2f1 transcript levels) shows signs of a proliferative environment. This result is in accordance with the result from the Matrigel assay that indicated an increased colony-forming ability for selected cells from the susceptible mammary gland (with lower Nr2f1 transcript levels) as compared to those from the resistant congenic mammary gland.
Since Nr2f1 is implicated in MEC proliferation and differentiation in mice and rats, we asked if NR2F1 transcript levels correlate with clinical features of human breast cancer. The Oncomine database encompasses a comprehensive listing of breast cancer gene expression studies, including available clinical information on the samples [50]. When including 12 studies with 120+ samples for each study we found that the average of the median NR2F1 transcript levels reduces with increased histological grade (Figure 7A). Histological grade 3 tumors are more proliferative and more poorly differentiated than grade 1 or 2 tumors. The therapy-resistant and most aggressive form of breast cancer, based on hormone receptor status is the ‘triple-negative’ class of tumors that are more likely to be of grade 3 when resected. Median NR2F1 transcript levels were found to be lower in triple-negative breast cancers, as compared with ‘receptor positive’ (non-triple-negative) breast cancers (Figure 7A). In accordance with this finding, estrogen receptor (ER)-negative and progesterone receptor (PR)-negative breast cancers had lower median transcript levels of NR2F1 as compared with their positive counterparts (Figure 7A). Notably, NR2F1 transcript levels were found to be lower in human epidermal growth factor receptor 2 (HER2)-negative breast cancers (that mostly are ER-positive), as compared with the more aggressive HER2-positive breast cancers (Figure 7B).
Finally, we asked if Nr2f1 transcript level anti-correlated with histological grade within both ER-positive/ER-negative and within both HER2-positive/HER2-negative breast cancer subtypes. For 2 of the 12 previously mentioned breast cancer gene expression studies we obtained the raw data from the gene expression omnibus (GEO; GSE3494 and GSE5460). The GSE3494 data set consists of microarray gene expression data for ER-positive and ER-negative breast tumors including histological grade 2 and 3 tumors in both ER classes and the GSE5460 data set consists of data for HER2-positive and HER2-negative breast tumors including histological grade 2 and 3 tumors in both HER2 classes. We found that NR2F1 transcript levels are significantly lower in grade 3 tumors as compared with grade 2 tumors in both ER classes (Figure 7B, left panel), as well as in both HER2 classes (Figure 7B, right panel). Additionally, in the GSE5460 data set the grade 3 HER2-positive breast cancers were found to have higher NR2F1 transcript levels as compared with the grade 3 HER2-negative breast cancers (Figure 7B, right panel), suggesting a regulatory effect of HER2 amplification on NR2F1 transcript levels. In summary, these analyses indicate that low transcript levels of NR2F1 are strongly associated with high histological grade, poorly differentiated, highly proliferative breast cancers, including therapy-resistant ‘triple-negative’ breast cancer.
The inherited portion of breast cancer susceptibility is complex and likely involves numerous genetic factors [51], [52]. With the results from genome-wide association studies (GWAS) it became clear that the genetic landscape of breast cancer susceptibility largely consists of low-penetrance alleles that are common in the population [53]. Many such alleles are located in non-protein coding regions of the genome, including in gene deserts, such as the one on human chromosomal band 8q24 [5]. Mechanisms underlying the genetic associations are largely unknown. It is generally hypothesized that non-protein coding variants could modulate disease processes through the regulation of gene expression.
Like for human breast cancer susceptibility, many loci associated with rat mammary cancer susceptibility have been discovered over the last decade [54]. Multiple of these QTL have been identified by our laboratory through linkage analysis and fine-mapping using congenic recombinant lines [32], [33], [55], [56], [57]. For some of the rat loci, common alleles associated with breast cancer susceptibility in the human orthologous loci were found [58], [59]. One major advantage of such rat-human comparative genetics approach is the availability of a highly relevant genetic mammalian model system for mechanistic studies [30]. Understanding how loci affect breast cancer susceptibility will be informative in the design of preventative or early intervention strategies that would be applicable to many women at risk.
In this study, we identified a 277 Kb critical interval for the previously discovered Mcs1a resistance allele that is derived from the Cop inbred rat strain [33]. The allele when introgressed on the susceptible genetic background from the WF inbred rat strain modulates mammary carcinoma multiplicity by approximately 50%. The protective effect of the allele works against three distinctly acting carcinogenic treatments, indicating that the mechanism modulates mammary carcinogenesis beyond a carcinogen-specific initiation stage. Since the susceptibility or resistance phenotype of the WF or Cop Mcs1a allele, respectively, is transferrable in a mammary gland transplantation/carcinogenesis study, we concluded that the locus modifies mammary carcinoma development in a mammary cell-autonomous manner. This is in contrast to the Wistar-Kyoto (WKy) inbred strain-derived Mcs5a resistance locus, for which we previously published a non-mammary cell-autonomous mode of mammary carcinoma development modulation using a similar transplantation/carcinogenesis assay [60].
Markedly, the location of the 277 Kb critical interval lays within a 3 Mb gene desert, which classifies the Mcs1a allele as non-protein coding. To identify the gene targets for regulation by the Mcs1a associated non-protein coding region, we developed a mouse model lacking a 535 Kb gene desert region orthologous to Mcs1a. The mouse model organism was chosen, as a large deletion (i.e. megadeletion; MD) engineering resource by means of MICER vectors was readily available. A MICER vector-assisted large deletion mouse model had aided before in discovering the gene targets of regulation by a non-protein coding region associated with coronary artery disease [61]. At the time we developed our MD mouse model, targeted rat genetic manipulation technologies were still under development. With the current maturation of zinc-finger nuclease-mediated genome editing technology [62], [63], a similar MD approach will be applicable to the rat model organism in the near future. Using RNA-seq we characterized mammary gland gene expression of MD and WT mice. Of genes surrounding the gene desert within 2.5 Mb, we only found the transcript level of the orphan nuclear receptor gene Nr2f1/Coup-tf1 to be strongly reduced upon deletion of the non-protein coding region. In addition, in the global RNA-seq gene expression analysis, Nr2f1 had the lowest P-value of all differentially expressed genes with annotated 1-1-1 mouse-rat-human orthologs. This gene is located at a genomic distance of over 800 Kb from the MD mutation, suggesting the presence of a strong Nr2f1 distal enhancer in the deleted region. Nr2f1 transcript levels were found to be downregulated in whole mammary gland, RMEC and mammary carcinoma samples from susceptible congenic controls as compared with Mcs1a resistant congenic rats. These results identify Nr2f1 as a strong candidate breast cancer susceptibility gene whose increased mammary transcript levels are associated with resistance to mammary carcinoma development. It is worth mentioning that the difference in Nr2f1 transcript levels between the susceptible and resistant Mcs1a alleles are more substantial in tumors as compared with untransformed cell types of the mammary gland. A plausible explanation for this observation could be that Nr2f1 transcript levels in an unidentified progenitor (or perhaps cancer-initiating) RMEC population may show similar substantial differences, which could be masked by other cell types present in the whole mammary gland or RMEC samples. The presence of a higher-order chromatin structure connecting the Nr2f1 promoter with a strongly conserved element within Mcs1a supports the long-range (∼820 Kb) regulatory potential of the Mcs1a locus. It should also be noted that the 3C assay was biased towards elements with the strongest evolutionary conservation (through fishes). Potentially interesting interacting elements in less conserved sequences may have been overlooked. Because the intensity of the chromatin loop is not affected by the Mcs1a alleles, but Nr2f1 transcript levels are, we hypothesize that the proteins involved in the higher-order chromatin interaction may be not be the same factors regulating Nr2f1. Resequencing of the interacting region in the Cop and WF parental inbred strains revealed the presence of 17 polymorphisms. Only one polymorphism, a 14 bp deletion in the Cop strain, affects a human-rat-mouse conserved binding motif, which is a COUP-TF binding site. Since the resistance (Cop) allele is associated with increased Nr2f1 (Coup-tf1) transcript levels we hypothesize that the 14 bp deletion removes a repressive autoregulatory module of Nr2f1 communicating with its own promoter. Since the MD mutation of the entire locus profoundly downregulates Nr2f1 transcript levels, the entire region is acting as a strong enhancer. Thus, we propose that in the susceptible strain harboring the WF allele with the intact COUP-TF binding motif, the repressive autoregulatory mechanism may modulate Nr2f1 transcription in the context of the activity of the enhancer. A germ-line mutation in the orthologous binding motif is not known to exist in mice or humans. Other variants outside of the conserved element (perhaps located in closer genomic proximity to the NR2F1 promoter) may confer similar NR2F1 regulation and thus potentially associate with breast cancer risk.
By taking advantage of the available congenic rat and genetic mouse models, we focused on dissecting the mechanisms underlying the non-protein coding Mcs1a locus on the organismal level. Because of the mammary cell-autonomous mechanism and change in mammary Nr2f1 transcript levels, we looked for differences in MEC biology between susceptible and resistant Mcs1a congenic rats. We found in a limiting dilution RMEC transplantation assay for repopulation ability that mammary stem cell activity is not affected by Mcs1a. Next, we tested the proliferation potential of a specific clonogenic population of RMECs. Clonogenic RMECs have previously been described to stain brightly with peanut lectin (PNL) and to be mostly located within the luminal population [25], [26]. In this study, we show that the luminal clonogenic RMEC population with colony-forming ability in Matrigel is indeed marked by bright PNL staining and not by high CD61 expression, which was shown to enrich for luminal progenitor cells with colony-forming ability in the mouse mammary gland [22]. In later studies, c-kit and ALDH have been identified as more specific markers for luminal progenitor cell populations, illustrating the heterogenity of the luminal MEC population [23], [24]. In the future, these markers can be tested on RMECs in combination with PNL staining to pinpoint the rat luminal progenitor population, provided good antibodies are available for the rat. Interestingly, the colony-forming ability of the luminal PNLhi population was found to be reduced in animals carrying the Mcs1a resistance allele, as compared with susceptible controls. As determined by multiparameter FACS analysis of freshly isolated RMECs, the abundance of PNLhi and luminal PNLhi cells among CD31–CD45− RMECs was not significantly different. Hence, we concluded that the proliferation potential of the colony-forming luminal population is affected by Mcs1a. The FACS analysis, however, did reveal a RMEC differentiation phenotype associated with Mcs1a. Resistant congenic animals have a higher abundance of luminal and lower abundance of basal RMECs, as compared with susceptible congenic control animals. Basal RMEC are mainly characterized by high β1-integrin (CD29) expression and loss of β1-integrin in the mouse mammary gland impairs mammary cancer development [64], suggesting that lower abundance of basal RMECs in the resistant Mcs1a congenics may contribute to lower mammary carcinoma susceptibility. Interestingly, in both the Mcs1a congenic rat model and genetically engineered mouse model, higher expression of Nr2f1 in the mammary gland is associated with higher abundance of luminal MMECs, identifying Nr2f1 as a candidate MEC differentiation gene modulating luminal cell fate. Since in the MD mouse model the abundance of mature luminal cells was significantly affected, and the abundance of luminal progenitors and basal cells were not, we propose that the Mcs1a-associated locus may impact luminal cell fate through activities in the luminal progenitors. Several other genes have been shown to regulate luminal cell fate mainly through activities in luminal progenitors. Downregulation of Gata-3 and upregulation of FoxM1 have been demonstrated to lead to impaired luminal cell differentiation [22], [65]. Interestingly, we found the transcript level of FoxM1 significantly upregulated in the MD mammary gland samples, whereas the Gata3 transcript level was not affected (Table S4). FoxM1 is predicted not to have a COUP-TF binding motif in its vicinity, thus the exact relationship of FoxM1 and Nr2f1 transcript levels remains to be investigated.
NR2F1 is an orphan nuclear receptor of the steroid hormone receptor superfamily [66]. Homodimers of NR2F1 bind the DR1 (direct repeats with 1 spacer) motif with the highest affinity [67]. NR2F1 is thought to act as a transcriptional repressor [68], [69], but can activate target genes as well [70], [71]. Nr2f1 has been previously recognized as an important factor in the development of the mouse nervous system [72], [73], [74] and the inner ear [75], [76]. Interestingly, ectopic Nr2f1 expression in the developing telencephalon and knockdown of Nr2f1 in primary neurospheres have been shown to result in defect neuronal cell fate determination [77], [78], suggesting that Nr2f1 may function as a neuronal as well as a MEC differentiation gene. The MD mice generated in this study have normal bodyweight, lifespan, and startle response to finger flicking above the cage (to test for hearing loss), but do display a delayed eyelid opening phenotype. Delayed eyelid opening could be indicative of an eye development defect or an eyelid epidermal apoptotic defect [79]. Interestingly, Nr2f1 has previously been implicated in eye development and was found to be highly expressed in progenitor cells of the developing eye [80], suggesting that the delayed eyelid opening phenotype in the MD mice is likely due to aberrant Nr2f1 expression in the differentiating eye progenitors.
There is a moderate amount of evidence that NR2F1 is involved in breast cancer, mainly through its cross-talk activities with the ER- [81], the arylhydrocarbon receptor- [82], and/or retinoic acid-mediated signaling [83]. Again, several studies emphasize NR2F1's dual role as a transcriptional repressor and activator, depending on the promoter its acting on and the cellular context, i.e. presence of other nuclear factors such as ER [84], [85]. We show in this paper that in all large human breast cancer gene expression studies examined, triple-negative (aggressive, therapy-resistant) and histological grade 3 (poorly differentiated, highly proliferative) breast cancers display lower NR2F1 transcript levels as compared with ‘receptor-positive’ and histological grade 1/2 breast cancers, respectively. This observation is in accordance with the mouse and rat MEC differentiation and mouse mammary gland gene expression studies that show reduced Nr2f1 transcript levels associated with less luminal differentiation and a more proliferative epithelial environment. Recently, NR2F1 was presented in a breast cancer dormancy gene signature as a gene upregulated in dormant cells [86]. Notably, in the same study, MCF7 cells with siRNA-mediated depletion of NR2F1, when injected into the mammary fat pad of immunosuppressed mice resulted in increased growth as compared with negative control siRNA-treated MCF7 cells [86]. Again, consistent with our findings, this result provides functional evidence that lower NR2F1 transcript levels increase the proliferative potential of breast cancer cells in an in vivo model system. In addition, we found in the MCF10A and MCF7 cell lines that siRNA-mediated reduction of NR2F1 transcript levels results in increased expression of CD24. CD24 has previously been implicated in breast cancer, for example as a marker for the breast cancer-initiating cell population [87]. Ectopic expression of CD24 in breast cancer cells has been shown to result in increased proliferation, as well as cell motility and invasion [88] and the expression of CD24 in breast carcinomas has been associated with poor prognosis [89].
It should be noted that according to available HER2 status classification in the human breast cancer gene expression studies, the more aggressive HER2-positive breast cancers (also associated with poorer clinical outcome) were found to express higher NR2F1 transcript levels, as compared with HER2-negative breast cancers (that are mostly ER-positive and less aggressive). In addition, we show for the GSE5460 gene expression data set that within both HER2 classes, histological grade 3 tumors have lower NR2F1 transcript levels as compared with grade 2 tumors. In this data set, the grade 3 tumors from the HER2-positive class have higher NR2F1 transcript levels as compared with grade 3 tumors from the HER2-negative class. We speculate that over expression of HER2 and subsequent stimulation of downstream signaling pathways increases NR2F1 transcript levels. The cellular effects of higher NR2F1 transcript levels may be very different for the amplified and unamplified HER2 backgrounds.
NR2F1 is located on human chromosomal band 5q15. Interestingly, a hotspot for copy number alterations (CNA) in breast cancer maps to chromosomal arm 5q [90], [91], with deletions most frequently occurring at 5q11-5q34 [92]. These CNA have been associated with high histological grade, basal-like tumors, p53-mutation status, triple-negative tumors, and tumors from BRCA1 carriers [90], [91], [93]. A recently published study describing comprehensive molecular portraits of human breast tumors identified the 5q deletion hotspot as a large trans-eQTL, as the expression levels of hundreds of genes across the genome are associated with occurrence of 5q deletions [93]. Interestingly, this study found the associated genes to enrich in GO categories involved in cell cycle processes, FoxM1 transcriptional regulation and proliferation, many of which are also found in our MD and WT RNA-seq study to be anti-correlated to Nr2f1 transcript levels. Placement of the Mcs1a-orthologous gene desert and NR2F1 within the deletion hotspot suggests that NR2F1 may play a role in deregulation of a fraction of these cell cycle-related genes associated with the triple-negative breast cancer-specific 5q deletions.
In summary, we describe the genetic dissection of the gene desert breast cancer susceptibility locus Mcs1a. We hypothesize that the resistance allele (from the Cop strain) carrying a truncated, potential transcriptionally suppressive COUP-TF (autoregulatory) binding motif, leads to increased Nr2f1 transcript levels in the mammary gland, which increases luminal RMEC differentiation and creates a less proliferative, more differentiated mammary epithelium with decreased mammary carcinoma susceptibility (Figure 8). We present NR2F1 as a strong candidate breast cancer susceptibility gene and MEC differentiation gene. In addition to a potential role in breast cancer susceptibility, we propose that reduced NR2F1 transcript levels associated with the human breast cancer 5q chromosomal deletions play a role in high-grade, poorly differentiated, proliferative breast cancer, including therapy-resistant triple-negative breast cancer. The human non-coding region orthologous to Mcs1a as well as NR2F1 are located on chromosomal band 5q in the region of frequent deletion. Raising NR2F1 transcript levels, or enhancing NR2F1's activities has great potential as a strategy to aid in breast cancer prevention or breast cancer intervention, including for triple-negative breast cancer (and possibly excluding HER2-positive breast cancer). As a steroid hormone receptor family member, NR2F1 potentially is an attractive therapeutic target. To our current knowledge NR2F1 is still an orphan nuclear receptor, which means a ligand has not been identified yet. Based on strong amino-acid conservation of the NR2F1 ligand binding domain (96%) with that of NR2F2, the crystal structure of the NR2F2 ligand binding domain suggests that NR2F1 may also be activated by retinoic acid through coactivator recruitment-based release of its autorepressed conformation [94]. Identification of ligand-mediated activator mechanisms for NR2F1 is important to begin to exploit its therapeutic potential in the near future.
All animal protocols were approved by the University of Wisconsin School of Medicine and Public Health Animal Care and Use Committee.
The congenic rat lines were established and maintained in an AAALAC-approved facility as previously published [57]. Congenics are defined as genetic lines developed on a Wistar-Furth (WF; susceptible) genetic background and carrying the selected Copenhagen (Cop; resistant) Mcs1a alleles in homozygous fashion [33]. Resistant congenic lines with decreased susceptibility phenotypes (Q, R3, V4, W4, Y4, W5) carry Cop alleles that define the Mcs1a locus critical interval. The susceptible congenic lines (P5, V5, R5, A4, Y3) are WF-homozygous at the newly defined Mcs1a locus. The susceptible congenic control line (WF.Cop) was derived from congenic line W5 and is WF-homozygous for all Mcs1 loci. The primer sequences for the genetic markers polymorphic between the WF and Cop inbred parental stains that are used to define the congenic lines are listed in Table S1.
Female rats, 7–8 weeks of age, were either orally gavaged with 7,12-dimethylbenz(a)anthracene (DMBA) at 65 mg/kg of body weight, injected intraperitoneally with N-nitroso-N-methylurea (MNU) at 50 mg/kg of body weight, or subjected to mammary ductal infusion of replication-defective retrovirus expressing the activated HER2/neu oncogene (HER2/neu) at a concentration of 1×105 Colony Forming Units (CFU)/ml [34]. To obtain multiplicities, mammary carcinomas >3×3 mm were counted at 15 weeks post-DMBA, 15 weeks post-MNU, and 7 weeks post-HER2/neu treatment. Multiplicity data were statistically analyzed using Mann-Whitney nonparametric tests.
All mice are maintained in an AAALAC-approved facility. The megadeletion mice were produced in collaboration with the University of Wisconsin Biotechnology Center Transgenic Animal Facility (http://www.biotech.wisc.edu/facilities/transgenicanimal). MICER clones (MHPP256h04; MHPN5k06; Sanger Institute, UK) were obtained and the vectors were purified. To ensure proper directionality of the construct upon insertion, the genomic insert from vector MHPP256h04 was flipped by AscI digestion and religation. The vectors were prepared by creating a gap in the genomic insert, as efficient targeting using MICER vectors relies on the embryonic stem (ES) cell's gap repair mechanism [95]. AB2.2 (HPRT deficient) ES cells from the 129/SvEv strain were transfected by electroporation in the presence of a linear gap-containing MICER vector. The first electroporation was done in the presence of the (flipped) gap-containing MHPP256h04 vector. Puromycin-resistant ES cell clones were expanded and checked for proper targeting by Southern blot analysis (Figure S1). Following karyotyping, karyotypically normal clones were selected and expanded for targeting with the second MICER vector. After electroporation in the presence of the gap-containing MHPN5k06 vector, neomycin-resistant clones were expanded and checked for proper targeting by Southern blot analysis. A karyotypically normal doubly targeted ES cell clone was expanded. To excise the 535 Kb region (megadeletion, MD) between the loxP sites present in both inserted vectors, a Cre-recombinase expressing vector (pTurbo-Cre) was introduced through electroporation. As proper Cre-loxP recombination restores the functional HPRT gene [95], hypoxanthine/aminopterin/thymidine (HAT) resistant clones were expanded and checked for recombination by PCR using 2 primer combinations spanning the deletion (mmMICERdel:TGTCTAGAGCTTGGGCTGCAG mmMICERdel:AGACAGAATGCTATGCAACCT and Del2F:CATGGACTAATTATGGACAGG Del2R:CTCCTTCATCACATCTCGAGC). Karyotypically normal ES cell clones were monodispersed and microinjected into C57Bl/6 blastocysts to produce chimeric founders. After germ line establishment, the MD mutation was introgressed onto the FVB/N and C57Bl/6 inbred genetic backgrounds for >10 generations. Deletion of the 535 Kb region was further verified by PCR using 4 primer combinations within the deleted sequence (mmdelNeg:TGGACTTGATGTGCTCCTTG mmdelNeg:TGCCATCAATGAGTTTGAGG, 2F:AAGTGAAAGATGCTGACATTTCC 2R:AAGACTGAATTCTTGCCACTCAC, 3F:GGGAGCCATTTATCACAGTCCTA 3R:GACCCTCACAAAAGCTGGTTTA, 4F:ACACATTTGGAGATGCAAACAG 4R:CACAAAAGTCACCTAAAAGGATCA). Examples of genotyping images for primers mmMICERdel and mmdelNeg are shown in Figure S1.
Females from the susceptible (S) inbred strain WF, the resistant (R) congenic line Y4, and WFxY4 or Y4×WF intercross (F1) were used. Donor mammary glands with lymph nodes (LN) excised (both abdominal and adjacent inguinal glands) from 30–35 day old females were finely minced over ice and divided into four equal volumes. One volume was transplanted onto the interscapular white fat pad of each 30–35 day old recipient (1 donor/4 recipients). Three weeks after transplantation, all recipients were treated with DMBA. At 15 weeks post-DMBA, interscapular fat pads were examined for carcinoma development. In addition, each fat pad was whole mounted and stained with aluminum carmine to verify mammary gland establishment. As only 15 out of 228 rats developed multiple carcinomas in the transplant sites, the mammary carcinoma incidence data were analyzed as a binary response by logistic regression. The four transplant groups (S:S, S:F1, R:F1, R:R) form a 2×2 factorial design with donor and recipient genotypes as the main effects. Standard logistic regression was applied to the binary response data with two main effects and an interaction term.
Female rats (11 weeks of age) of the susceptible congenic line WF.Cop (susc.) and resistant congenic lines W4 and W5 (res.), or female MD and WT mice (9 weeks of age; FVB) were used as tissue donors. RNA was extracted from snap-frozen mammary gland and mammary carcinoma tissues, from fresh RMEC samples, or from siRNA-treated human cell line samples. To synthesize cDNA from 800 ng of TURBO-free DNaseI-treated total RNA, the reverse transcriptase Superscript II kit (Invitrogen) was used according to manufacturer's directions. Quantitative real-time PCR was used to quantify transcript levels. TaqMan quantitative PCR primers and probes were ordered as premade assays (ABI/Applied Biosystems): rat Nr2f1 Rn01489978_m1 (FAM), mouse Nr2f1 Mm01354342_m1 (FAM), human NR2F1 Hs00818842_m1 (FAM), rat ActB Rn00667869_m1 (VIC, endogenous control), mouse ActB Mm00607939_s1 (VIC, endogenous control) and human GAPDH Hs03929097_g1(VIC, endogenous control). Reactions were run as described previously [96]. Quantities of transcripts were measured by comparison of Ct values with a standard curve calculated from serial dilutions made from reverse transcriptase reactions that contained 2 µg of total RNA. Sample measurements are an average of three or four replicates within 0.5 Ct value. Sample measurements were normalized by dividing the gene specific transcript quantity over the endogenous control quantity. For each sample, the ratio was scaled to the average ratio of the control group from the same experiment, which are the susceptible congenic control group (rat), the WT group (mouse) or the siCONTROL-treated group (human cell lines). Data were analyzed using Mann-Whitney nonparametric tests.
RNA was extracted from snap-frozen mammary gland tissue from MD and WT mice (FVB) using the MagMax-96 Total RNA isolation kit (Ambion) according to manufacturer's directions. RNA samples were checked for integrity using Agilent 2100 Bioanalyzer. Two RNA samples were pooled using 5 µg of each for 10 µg of total RNA per library preparation sample. Sample preparation and next-generation sequencing was done at the University of Wisconsin Biotechnology Center Gene Expression Center. Sample preparation was done using the Rev.D mRNA sample preparation kit (Illumina), according to the manufacturer's recommendations. Samples were run on the Illumina GAIIx. Reads that made the quality cut-off were aligned to the mouse Ensembl set of 82,508 transcripts (http://genome.ucsc.edu) using Bowtie (v0.12.3; http://bowtie-bio.sourceforge.net/index.shtml) with the following settings: -f –v 1 −3 0 –a –m 100. Transcript levels were estimated using the RSEM algorithm [37]. Differential expression between the MD and WT samples was determined using edgeR [38]. Correlation analysis of gene expression was done on the 1,531 DE mouse genes and 412 DE mouse genes with 1-1-1 mouse-rat-human Affy (U133a probe set) orthologues. The Pearson correlation between the mean-centered RSEM tau expression values was calculated and visualized in R using the gplots library. Similar correlation analysis was done on the 412 human orthologous genes for which microarray data (Affy U133a) was available from a human breast cancer gene expression study. For this analysis, the GSE3494 dataset was downloaded from the Gene Expression Omnibus (GEO; www.ncbi.nlm.nih.gov/geo). The raw data was normalized using the RMA approach in R [97]. The online functional annotation tools GOrilla (http://cbl-gorilla.cs.technion.ac.il) and DAVID (http://david.abcc.ncifcrf.gov) were used to find Gene Ontology (GO) categories and biological processes enriched with DE genes [43], [44]. For the 412 DE mouse-rat-human orthologs genes analysis, the full mouse-rat-human orthologues gene list (9,828 genes, Table S5, S6) was used as the background list.
Templates were prepared from isolated RMECs from 6 animals of the susceptible congenic control line WF.Cop and 6 animals from the Mcs1a resistant congenic line, as described previously [96]. The restriction enzyme of choice was BglII. The fixed primer was chosen to be located in the predicted promoter of the Nr2f1 gene. The experimental primers were chosen to be located within the Mcs1a critical interval, biased towards regions of evolutionary sequence conservation (Figure 3d). Primer sequences are listed in Table S2. The relative interaction frequency for each experimental primer in combination with the fixed primer was determined for each sample as the average of 3 replicate measurements divided by the average of a positive control (BAC-derived) template run in the same PCR plate [96]. A non-parametric Mann-Whitney test was performed to test for differences between genotypes. Since no genotype differences were detected, data for the Mcs1a profile were taken from all samples. Next, non-parametric Krukal-Wallis tests were performed to test for increased (P<0.05) interaction frequency of a primer pair above the background levels. The interaction frequencies of peak fragments were tested against interaction frequencies below two background cut-offs, namely 0.05 and 0.1.
Genomic DNA was isolated from spleens of inbred WF and Cop rats using phenol and chloroform extractions. A total of 10 ng of genomic DNA was used in each PCR reaction. Primers were designed using Primer3 software and are listed in Table S3. Successful PCR reactions as verified by agarose gel electrophoresis were diluted 6 times. Of this dilution, 1 µl was used in a BigDye sequencing reaction in a total volume of 15 µl, according to the vendor's (Applied Biosystems) specifications with the exception that we used 0.7 µl BigDye enzyme mix. Sequencing reactions were cleaned-up using the CleanSeq kit (Agencourt) and submitted for sequencing through the UW Biotechnology DNA Sequencing Facility. Reads were visualized using 4Peaks software (Meckentosj) and scanned for mutations using BLAT (UCSC Genome Browser).
Rat and mouse mammary epithelial cells (RMECs and MMECs) were prepared from LN-excised abdominal and adjacent inguinal mammary glands, as described previously [25], [42]. For phenotyping RMECs and MMECs, staining was done using the low-volume staining method to reduce antibody costs [42]. To stain the single RMECs, antibodies against rat CD49f (Santa Cruz), CD24, CD29, CD31, CD45, and CD61 (BD Biosciences) or peanut lectin (Sigma) were used. To stain MMECs, antibodies against mouse CD24, CD29, CD31, CD45, CD49f, and CD61 (BD Biosciences) were used. Live cells were gated based on FSC, SSC, and Hoechst staining for 2n–4n DNA content. Single cells were gated using forward scatter (FSC) and side scatter (SSC) width. For phenotyping, the stained samples were acquired on a BD LSR II flow cytometer equipped with 4 lasers (multi-line UV, 405 nm, 488 nm and 633 nm). The data were collected as fcs3 files using FACS Diva software and analyzed using FlowJo software (Treestar Inc). Data files obtained from cell samples stained with single antibodies and control unstained cell samples were used for compensation. Data on percentages of cells in various gated populations or mean fluorescence intensities of entire populations were exported and statistically analyzed using Student's t-test.
For RMEC sorting, 50–70 million single cells were stained with anti-rat CD24, CD29, CD31, CD45, CD61 and peanut lectin at a concentration recommended by the vendor's specifications. Sorting was done on a BD FACSAria flow cytometer equipped with 5 lasers (multi-line UV, 405 nm, 488 nm, 540 nm and 640 nm). Cells were collected in 50% FBS.
We have previously shown that the clonogenic cells assayed in our transplant system are capable of dividing and differentiating into morphologically and functionally normal mammary parenchyma [41]. For this RMEC transplantation assay, donor and recipients from the susceptible congenic control line WF.Cop and Mcs1a resistant congenic lines W4 and W5 were used. The final dilutions of single RMECs were mixed with an equal volume of a 50% brain homogenate, which was extracted from the donor rats. Aliquots of 40 µl of the mixture containing a known number of cells were then injected into the interscapular fat pad of recipient animals of the same genotype. The frequency of viable clonogenic stem-like cells in a cell suspension was quantitated using a limiting dilution assay as previously described [98], [99]. In each rat, two sites were used for grafting. For each cell dilution, between 8 and 32 sites were transplanted per genotype. Recipient rats were sacrificed 6 weeks after mammary cell grafting, and the fat pads injection sites were removed, fixed, stained, and examined for the growth of mammary tissue. The percentages of transplant sites with a mammary outgrowth were then plotted against the number of cells injected per site. The data were fit to the transplantation model of Porter et al. [100] and according to the statistical methodology for this model there was no significant difference between any of the 3 groups in the estimated number of cells required to give 50% outgrowth occurrence. For display purposes, the data for transplant sites from congenic lines W4 and W5 were combined and plotted in Figure 4A as a single line for the resistant genotype. A second statistical approach was taken to detect a possible difference in outgrowths at each cell number individually between the susceptible and resistant (W4 and W5 combined) genotypes. Therefore, Chi-square tests for independent distributions in a 2×2 contingency matrix were conducted for comparing susceptible to resistant for each cell number. The P-values were adjusted for multiple comparisons.
For the matrigel assay, single sorted RMEC suspensions containing 10,000 cells were spun 450× g for 5 minutes at 4°C. Supernatant was discarded, the cell pellet was resuspended in 100 uL phenol red free Matrigel (BD Biosciences) and immediately plated in 12- or 24-well plates while on ice. Plates were placed at 37°C (with 5% CO2) for 30–60 minutes to allow gelling process. Mammary Epithelial Cell Medium (PromoCell) with 5% bovine calf serum and antibiotics was then added to the wells containing the sorted cells in matrigel. Fresh media was provided on days 2 and 5. On day 10, Matrigel containing RMEC colonies was fixed with 2% paraformaldehyde in phosphate buffer pH 7.4 (PB) for 30 minutes at 37°C followed by staining with 0.5% methylene blue in PB for 5 minutes at 37°C. Colonies were counted using a microscope. Count data were normalized to the average colony count of the susceptible congenic control line run in the same experiment and was statistically analyzed using Student's t-test.
MCF10A and MCF7 cell lines were obtained (ATCC) and cultivated according to the manufacturer's recommendations. Transient transfections were done in 24-well plates using the Lipofectamin2000 reagent (Invitrogen). Short interfering RNAs (NR2F1 SMARTpool and Non-targeting pool) were obtained (Dharmacon) and used at a final concentration of 125 nM in the transfection media. The transfection media was washed off the cells after 6 hours. The cells were cultured for 40 additional hours before harvesting for FACS and expression analysis.
The Oncomine database (www.oncomine.org) was queried using the following filters: Gene: NR2F1, Cancer Type: Breast carcinoma, Sample Type: Clinical Specimen. Only studies with 120+ samples were considered. If available, the median levels of NR2F1 for the clinical parameters, histological grade, ER-status, PR-status, HER2-status, TN-status were entered in Excel. To be able to compare clinical parameters across studies, the median levels for each clinical parameter in a study were normalized by the median level for the entire study. The average of the normalized median values was plotted. For statistical analysis, the non-parametric Kruskal-Wallis test was used.
To ask if lower NR2F1 levels are associated with histological grade 3 in both ER-positive and ER-negative breast cancers or HER2-positive and HER2-negative breast cancers, the RMA normalized NR2F1 probe set levels (probe ID 209505_at) from the GSE3494 and GSE5460 studies were used, respectively. To match the Y-axis scale in panel A, the RMA normalized values were expressed relative to the median level for the entire study. The values were statistically compared between groups using the non-parametric Mann-Whitney U test.
|
10.1371/journal.pgen.1003368 | Long Noncoding RNA MALAT1 Controls Cell Cycle Progression by Regulating the Expression of Oncogenic Transcription Factor B-MYB | The long noncoding MALAT1 RNA is upregulated in cancer tissues and its elevated expression is associated with hyper-proliferation, but the underlying mechanism is poorly understood. We demonstrate that MALAT1 levels are regulated during normal cell cycle progression. Genome-wide transcriptome analyses in normal human diploid fibroblasts reveal that MALAT1 modulates the expression of cell cycle genes and is required for G1/S and mitotic progression. Depletion of MALAT1 leads to activation of p53 and its target genes. The cell cycle defects observed in MALAT1-depleted cells are sensitive to p53 levels, indicating that p53 is a major downstream mediator of MALAT1 activity. Furthermore, MALAT1-depleted cells display reduced expression of B-MYB (Mybl2), an oncogenic transcription factor involved in G2/M progression, due to altered binding of splicing factors on B-MYB pre-mRNA and aberrant alternative splicing. In human cells, MALAT1 promotes cellular proliferation by modulating the expression and/or pre-mRNA processing of cell cycle–regulated transcription factors. These findings provide mechanistic insights on the role of MALAT1 in regulating cellular proliferation.
| The mammalian genome encodes large number of long non protein-coding RNAs (lncRNAs). These lncRNAs are suggested to regulate key biological processes (including cellular proliferation and differentiation), and aberrant expression of these is associated with cancer. However, only a few of these lncRNAs have been functionally validated in biological or disease processes. MALAT1, an abundant nuclear-retained lncRNA, is overexpressed in several cancers, and its elevated expression has been associated with hyper-proliferation and metastasis. However, the underlying mechanism behind this deregulation and its association with cancer is poorly understood. Here, we establish the role of MALAT1 in the cell cycle pathway and propose the molecular mechanism of its function during normal cell cycle progression. MALAT1 RNA levels are differentially regulated and critical for normal cell cycle progression. Depletion of MALAT1 results in cell cycle arrest with significantly reduced cellular proliferation, simultaneously leading to activation of p53 and its target genes. Further, the accurate levels of MALAT1 in the cell are extremely crucial for expression and activity of the oncogenic transcription factor B-MYB, which is involved in G2/M progression. Our data indicates that the cancer-associated MALAT1 RNA regulates cellular proliferation by modulating the expression and/or pre-mRNA processing of cell cycle–regulated transcription factors.
| The eukaryotic genome harbors a large number of noncoding RNAs, which include small and long noncoding RNAs (lncRNAs) [1], [2], [3], [4]. Small ncRNAs such as microRNAs regulate the expression of target genes at the level of translation or mRNA stability, whereas piwi-interacting RNAs (piRNA) have been linked to transcriptional gene silencing of retrotransposons and other repeat-containing genetic elements [5], [6], [7], [8]. In addition to the class of well-studied small ncRNAs, lncRNAs, which are noncoding transcripts that are >200 nucleotides in length, have recently emerged as important molecules in several cellular processes [4], [9], [10]. The human genome encodes ∼15,000–17,000 potential lncRNAs. However, the function of less than 2% of the human lncRNAs is clearly elucidated [11], [12]. LncRNAs are involved in several crucial functions: they can act as a scaffold to keep several proteins tethered to a specific cellular compartment, act as a guide to recruit proteins to a specific chromatin site or influence local chromatin architecture [3], [4], [13], [14], [15]. For instance, the X-chromosome encoded Xist lncRNA coats the inactive X-chromosome (Xi) in female mammals and facilitates the recruitment of chromatin modifiers to the Xi [16], [17]. Similarly, the imprinted Air lncRNA interacts with the histone-methyl transferase G9a and recruits it to epigenetically silence Slc22a3 [18], [19], [20]. Further, nuclear-retained lncRNAs act as structural components of specific sub-nuclear domains [21], [22], [23], [24]. LncRNAs are also known to regulate transcription and RNA processing events and also serve as precursors for small RNAs [14], [15].
Several lncRNAs exhibit temporal and spatial expression patterns or their expression is restricted to particular tissue or cell types or cell cycle stages, indicating vital and diverse biological roles of lncRNAs [4], [25], [26], [27], [28]. A recent study demonstrated cell cycle-regulated expression of several of the lncRNAs in mammalian cells [29]. The authors identified >200 lncRNAs encoded in close proximity to more than 50 protein-coding genes involved in cell cycle, including cyclins and cyclin-dependent kinases. During cell cycle progression, the levels of both lncRNAs and the nearby cell cycle gene mRNAs displayed dynamic fluctuations. However, the expression of these lncRNAs did not correlate either positively or negatively with the expression of the nearby cell cycle genes [29]. This implies that although some lncRNAs and mRNAs could be regulated in concert, they may not necessarily regulate each other. On the other hand, several other studies have demonstrated the involvement of lncRNAs in regulating the expression of cell cycle genes in cis [30], [31], [32], [33]. The ANRIL lncRNA is located upstream of the tumor suppressor locus encoding p16INK4A and p15INK4B, mutations or depletion of ANRIL results in the loss of p16INK4A and p15INK4B repression [30], [31], [32]. Similarly, an lncRNA transcribed from the 5′regulatory region of Cyclin D1 (CCND1) recruits TLS, an RNA-binding protein, to the CCND1 gene in response to DNA damage, and results in the transcription repression of CCND1 [33].
A large number of lncRNAs display deregulated expression in human cancer samples and are regulated by oncogenic or tumor suppressor pathways [28], [34], [35], [36], [37], [38]. The HOTAIR lncRNA, which is known to regulate the expression of HOX gene clusters, is highly induced in breast cancer samples and its elevated expression has been correlated with metastasis and death [39]. Recent studies have also demonstrated the involvement of lncRNAs in the p53 gene regulatory pathway [4]. For example, lincRNA-p21 is activated by p53 and serves as a repressor in the p53-dependent transcriptional network [40]. DNA damage induces the expression of another lncRNA, PANDA in a p53-dependent manner. PANDA is transcribed from the p21 (CDKN1A) promoter, and it negatively regulates the expression of pro-apoptotic genes upon DNA damage, thereby controlling apoptosis [29]. Finally MEG3, an imprinted lncRNA induces accumulation of p53 by negatively regulating MDM2 expression [41].
The lncRNA MALAT1 is upregulated in several solid tumors and its differential expression is linked with cancer metastasis and recurrence [15], [28], [42]. MALAT1 is a highly abundant nucleus-restricted RNA that localizes to nuclear speckles, a sub-nuclear domain suggested to coordinate RNA polymerase II transcription, pre-mRNA splicing and mRNA export [43], [44], [45]. MALAT1 interacts with several pre-mRNA splicing factors including serine arginine dipeptide-containing SR family splicing factors [46], [47], [48], [49], [50], [51]. Furthermore, MALAT1 modulates the cellular distribution and activity of SR splicing factors thereby influencing alternative splicing of pre-mRNAs [50]. By utilizing such a mechanism, cells could alter the local concentration of a particular splicing factor upon a specific external signal or during specific stages of the cell cycle.
Previous studies have shown that transient overexpression of MALAT1 enhanced cellular proliferation in cell lines and tumor formation in nude mice, while depletion of MALAT1 in tumor cells reduced tumorigenicity [52], [53]. A recent study suggested the involvement of MALAT1 in regulating the E2F1 transcription factor activity, which is a crucial determinant of cell cycle progression and tumorigenesis [54]. These results indicate that MALAT1 has a pro-proliferative function; however, the mechanism has yet to be identified. In the present study, we examined the role of MALAT1 in cell cycle progression. We demonstrate that in human cells, MALAT1 levels are regulated during the cell division cycle. The differential levels of MALAT1 during specific cell cycle stages influence the expression of genes involved in cell cycle progression. Furthermore, MALAT1 depletion in normal human diploid fibroblasts (HDFs) induces DNA-damage response and results in the activation of p53 and its target genes. The cell cycle defects observed in MALAT1-depleted cells are sensitive to the cellular levels of p53, indicating that p53 is an important effector of MALAT1 function. Finally, we establish that the pro-proliferative role of MALAT1 is accomplished by its involvement in regulating the expression and/or pre-mRNA processing of oncogenic transcription factors, especially those that control mitotic progression.
In order to understand the role of MALAT1 in cellular proliferation, we synchronized human osteosarcoma cells (U2OS) in specific cell cycle stages (Figure 1A) and examined the levels of MALAT1 in each stage of the cell cycle. Quantitative RT-PCR (qRT-PCR) results revealed cell cycle-dependent expression of MALAT1, with low levels during G1 and G2 and high levels during G1/S and mitosis (M) (Figure 1A). Similar cell cycle regulation of MALAT1 was also observed in WI-38 human diploid lung fibroblasts (HDFs; Figure S1A). We have previously proposed that MALAT1 modulates pre-mRNA splicing by titrating the cellular levels of SR splicing factors [15], [50]. This prompted us to also examine the levels of SR proteins during various phases of cell cycle. SRSF1 levels remained unaltered during cell cycle (Figure S1B), indicating that the changes in MALAT1 levels during the cell cycle could fine tune the association of SR proteins with pre-mRNAs and thereby modulate cell cycle-specific alternative splicing.
To gain insight into the functional relevance of cell cycle-regulated expression of MALAT1, we examined the effects of MALAT1 depletion on cell cycle progression in HDFs that have a finite life span (WI-38 and IMR-90 cells). BrdU-PI and PI flow cytometry data revealed that MALAT1-depleted cells show reduced replication (S-phase) (Figure 1B and Figure S1C). MALAT1-depleted cells also showed an increase in the G1 (scr, 59%: AS1, 72%: AS2, 68%) and a marginal increase in G2/M (scr, 13%: AS1, 18%: AS2, 19%) population (Figure 1B). BrdU pulse labeling as well as proliferation assays confirmed the reduced proliferation of HDFs upon MALAT1 depletion (Figure 1C and 1D). MALAT1-depleted HDFs also showed increased expression of genes that are associated with cell cycle arrest (Figure 1E). The tumor suppressor p53 (TP53) and the cdk inhibitors p21 (a transcriptional target of p53) and p27 were upregulated upon MALAT1 depletion (Figure 1E and 1F). Interestingly, MALAT1-depleted cells also showed increased levels of γH2AX, indicative of double stranded DNA damage response (Figure 1F). Cell cycle arrest phenotypes, including G1 arrest and a significant reduction in S-phase were observed in WI-38 (Figure 1B and Figure S1D) and IMR-90 cells (data not shown) that were depleted of MALAT1 using either multiple independent sets of DNA antisense oligonucleotides or double-stranded siRNAs. These results suggest that MALAT1 is required for cell proliferation of HDFs.
A significant population of the MALAT1-depleted normal fibroblasts showed changes in the cellular morphology with cells appearing flat and displaying ‘fried egg’ morphology, reminiscent of senescent cells (Figure S1E). MALAT1-depleted cells showed enhanced β-galactosidase (β-gal) staining indicative of cellular senescence (Figure 1Ga–b). Furthermore, human fibroblasts that were depleted of MALAT1 (both by DNA antisense oligonucleotides and siRNAs) showed a gene expression signature characteristic of senescent cells (Figure 1H and Figure S1F). Recent studies have demonstrated specific loss of nuclear lamina associated lamin B1 in senescent cells [55], [56]. We also observed reduced lamin B1 mRNA and protein levels in MALAT1-depleted cells (Figure 1F and 1H). Finally, human lung fibroblasts that had undergone replicative senescence showed reduced levels of MALAT1 compared to actively proliferating cells (Figure 1I). Our results demonstrate that MALAT1 levels are cell cycle regulated and depletion of MALAT1 in HDFs results in proliferation defects with a population of cells undergoing senescence.
In order to understand the molecular mechanism underlying the cell cycle arrest upon MALAT1 depletion, we looked at changes in gene expression by performing microarrays from control or MALAT1-depleted HDFs (WI-38) using two independent antisense oligonucleotides (AS1 & AS2) against MALAT1. WI-38 cells were treated with a control (scr) or MALAT1-specific antisense oligonucleotides for 24 hr and then re-transfected with these oligonucleotides for another 24 hr. Total RNA was isolated 24 hr after the second antisense treatment and subjected to microarray analysis. Analyses of the microarray data from triplicate samples revealed that a common set of 413 mRNAs showed ≥2.5-fold reduced abundance in MALAT1-depleted cells (Tables S1, S2, S3). Silencing MALAT1 also increased the expression (≥2.5-fold) of ∼390 genes (Table S4). To gain an insight into the pathways activated by MALAT1, we performed Gene Ontology (GO) analyses of the downregulated genes and found that cell cycle was the most affected biological process in MALAT1-depleted cells (Figure 2A). Using qRT-PCR, we validated the changes in the cellular levels of a large number of mRNAs (∼150) (Figure 2B–2D, Figure S2Aa–c and S2Bc). The genes that were downregulated in MALAT1-depleted cells were broadly classified into several subgroups. A large number of the genes encoding proteins involved in G1/S transition and S-phase progression displayed reduced expression upon silencing of MALAT1 (Figure 2B), corroborating the cell cycle arrest phenotype. The other sub-group of genes, whose expression was severely affected, included those encoding proteins responsible for mitotic progression (Figure 2C). In addition, the expression of genes encoding pre-mRNA processing factors (hnRNPs [hnRNP A, L & K]) and chromatin modifiers (HP1α [CBX5], TOP2A, TOP2B, SMC2 & SMC4) was also compromised in MALAT1-depleted cells (Figure 2D and Figure S2Ab–c). Depletion of MALAT1 using siRNAs also showed similar changes in the expression of genes encoding cell cycle regulatory proteins (Figure S2Ad). Immunoblot analyses further corroborated the qRT-PCR data (Figure 2E). However, the levels of some of the encoded proteins (e.g., BUB3) were comparable between control and MALAT1-depleted cells, even though mRNA levels were somewhat reduced in MALAT1-depleted cells (Figure 2E and Figure S2Aa). Based on our microarray results, it was evident that several E2F target mRNAs were downregulated upon MALAT1 depletion. In this regard, a recent study documented the role for MALAT1 in controlling E2F1 function [54]. Surprisingly, a few of the bona fide E2F1 target gene mRNA levels remained unaltered in MALAT1-depleted cells (PCNA, Figure S2Ae), suggesting that MALAT1 downregulation in fibroblasts did not completely disrupt E2F1 activity.
GO analyses indicated that the p53 signaling pathway was the second most represented biofunctional pathway that was activated upon MALAT1 depletion (Figure S2Bb). This was further confirmed by the observation that MALAT1-depleted cells showed increased expression of p53 and several of its target genes (p21 [CDKN1A], GADD45A, 45B, TP53INP1; Figure 1E, 1F and Figure S2Bc).
Finally, we demonstrated that the exogenously expressed full-length MALAT1 could rescue the expression of several of the cell cycle genes in the absence of endogenous human MALAT1 (Figure S2Bd), supporting the involvement of MALAT1 in regulating the expression of these genes in human fibroblasts.
MALAT1 is upregulated during the G1/S phase of cell cycle and its depletion showed reduced expression of genes involved in the G1/S transition. This prompted us to examine whether MALAT1 plays a role in G1/S progression. To test this, HDFs were synchronized in G0 (quiescence) by serum starvation and released from quiescence for various time points (0 hr, 24 hr and 36 hr) in the presence or absence of MALAT1 (Figure 3A). qRT-PCR analysis revealed that MALAT1 was significantly depleted at each of the time points analyzed (Figure S3). Flow cytometry data showed that control (scr) oligo-treated cells displayed normal cell cycle progression upon addition of serum (24 and 36 hr after release) (Figure 3B). However, MALAT1-depleted cells (MALAT1-AS1 & -AS2) did not respond to serum and showed defects in S-phase entry with cells accumulated with 2C DNA content (Figure 3B). In vivo BrdU incorporation assays also revealed significantly reduced proliferation in MALAT1-depleted cells, indicating cell cycle progression defects (Figure 3C). Finally, in scr-oligo-treated G0 cells, addition of serum induced the expression of genes involved in G1/S transition and S-phase progression, whereas MALAT1-depleted cells failed to activate most of these genes (Figure 3D and 3E). These results suggest that in HDFs, depletion of MALAT1 specifically at G0 prevents the progression of cells into S phase. Our flow cytometry data could not differentiate whether the MALAT1 depleted cells were arrested in G0 or G1 phase of the cell cycle. However, the absence of ORC1, an integral component of the origin recognition complex for DNA replication that is expressed during G1 phase [57], strongly suggests that the cells remained arrested in G0 upon MALAT1 depletion (Figure 3D and 3E).
MALAT1-depleted HDFs showed a reduction in S-phase cells with a concomitant increase in G1. However, HeLa cells, upon MALAT1 depletion (either using DNA antisense oligonucleotides or siRNAs) showed prominent G2/M arrest with nuclear breakdown phenotype, primarily due to defects in chromosome segregation and spindle assembly (Figure 4A, Figure S4A–S4C). These defects could be partially rescued by the exogenously expressed mouse Malat1, indicating that MALAT1 is involved in mitotic progression (Figure S4Da–b). To determine whether MALAT1 depletion in HeLa cells results in S phase defects (similar to HDFs), we synchronized HeLa cells in mitosis, and released them in presence or absence of MALAT1 and examined the cell cycle progression. We could not arrest HeLa cells in G0 by serum starvation, consistent with the absence of a quiescent state in HeLa cells. Therefore, we synchronized them in prometaphase by nocodazole treatment, transfected with control or MALAT1-specific antisense oligonucleotides and released them for different time points (12, 15 & 18 hrs release) (Figure S4Ea–c). Flow cytometry analyses revealed that both control and MALAT1-depleted HeLa cells showed normal S-phase progression (Figure S4Ea). BrdU incorporation analyses in control and MALAT1-depleted HeLa cells also corroborated the flow data (Figure S4Ec). These results indicate that unlike in normal HDFs, depletion of MALAT1 in HeLa cells did not result in S phase arrest. Since MALAT1-depleted HDFs and HeLa cells showed different phenotypes, we examined the effect of MALAT1 depletion in different cell lines. Indeed, we observed cell line- or cell type-specific responses upon MALAT1 knockdown (Figure S4F). In general, human fibroblasts with a finite life span showed proliferation defects (WI-38, IMR-90 cells), whereas cancer or immortalized cell lines displayed a wide spectrum of abnormalities upon MALAT1 depletion. Surprisingly, MALAT1-depleted HepG2 cells (hepatocarcinoma) did not show any obvious phenotype even though we achieved similar levels of MALAT1 knockdown in these cells. Similarly, depletion of Malat1 in mouse primary (mouse embryonic fibroblasts, MEFs) and transformed fibroblasts (NIH3T3) did not reveal any phenotype (also see [58] (Figure S4F).
Furthermore, detailed analyses revealed that upon MALAT1 depletion, cell lines containing low p53 and/or p16INK4A (CDKN2A or p16) activity, including HeLa and U2OS [59], [60] did not display obvious S-phase defects but continued to show severe mitotic abnormalities. We reasoned that depletion of MALAT1 in cells with low p53/p16 activity might not activate the G1/S or intra-S phase checkpoints. In the absence of intact p53, MALAT1-depleted cells, despite having damaged DNA, progressed through G1/S and S phase but eventually arrested in mitosis [50]. To test this possibility, we depleted MALAT1 in WI-38 and WI-38-VA13 subline 2RA cells (WI-38 cells transformed using SV40 T-antigen and as a result have reduced p53 and p16 activity) [61] and examined the cellular phenotype. In contrast to WI-38 cells, MALAT1-depleted WI-38-VA13 cells displayed dramatic mitotic abnormalities and nuclear breakdown phenotype (Figure S4G). Loss-of-function analyses using siRNAs/antisense oligonucleotides against p53, p16 and MALAT1 indicate that WI-38 cells that were co-depleted of MALAT1 along with p53 or p53 and p16 did not show defects in S-phase entry but displayed breakdown of nuclei and mitotic defects (Figure 4Ba–b, Figure S4H). We also examined the alterations in the expression of genes involved in G1/S or S-phase progression in MALAT1 and p53 co-depleted WI-38 cells (Figure 4C). Cells depleted of MALAT1 alone showed reduced expression of several cell cycle genes. However, cells depleted with p53 alone or p53 along with MALAT1 showed normal expression of genes involved in G1/S transition (Figure 4C). We also examined the serum-stimulated response of genes involved in G1/S or S-phase progression in HDFs that were depleted of MALAT1 alone or were co-depleted of MALAT1 along with p53. Unlike the MALAT1 alone-depleted cells, both p53 alone or p53 + MALAT1 co-depleted cells responded to serum by stimulating the expression of genes involved in cell cycle progression (Figure 4Da–b, Figure S4I). These results indicate that functional p53 is vital to evoke G1 or S phase arrest in MALAT1-depleted cells.
An earlier study reported the involvement of MALAT1 in regulating the activity of E2F1 transcription factor by modulating the PC2 polycomb protein-mediated sumoylation of E2F1 [54]. Both p53 pathway and E2F1-mediated transcriptional activity are interconnected and reduction in E2F1 activity or activation of p53 can elicit similar responses, including cell cycle arrest. To determine which event (reduced activity of E2F1 or activation of p53) is primarily responsible for the cell cycle arrest observed in MALAT1-depleted HDFs, we conducted a time course analysis and assessed p53 and E2F1 activity at different time points after MALAT1 depletion (Figure 5). HDFs showed increased p53 activity within 12 hrs of MALAT1 depletion, as observed by increased p53 levels as well as enhanced expression of p53 response genes, including p21 and GADD45a (Figure 5Aa–c, 5B). Interestingly, cells also showed increased levels of γH2AX within 12 hrs of MALAT1 depletion, suggesting double stranded DNA damage response (DDR) (Figure 5B). These results indicate that induction of p53 upon MALAT1 depletion could be a consequence of double-stranded DNA damage response (DDR). We also examined changes in the expression of E2F1 and its target genes at different time points after MALAT1 depletion. Control and MALAT1-depleted HDFs displayed comparable E2F activity within 12 hrs of MALAT1 knockdown (Figure 5Ad–e, Figure S5Aa–b). However, MALAT1-depleted cells showed reduced levels of E2F1 mRNA and other E2F1 transcription targets after 24 and 48 hr of MALAT1 depletion (Figure 5Ad–e, Figure S5Aa–b). Moreover, control and MALAT1-depleted cells showed similar levels of phosphorylated retinoblastoma (Rb) protein within 12 hrs of MALAT1 silencing. Interestingly, HDFs showed reduced Rb phosphorylation after 24 hr of MALAT1 depletion. Rb is a key factor that controls cell proliferation by regulating progression through the restriction point in G1-phase of the cell cycle. Rb by associating with E2F family proteins negatively regulates E2F transcriptional activity [62]. Cell cycle-dependent hyper-phosphorylation of Rb by cyclin-dependent kinases (CDKs) prevents the association of Rb with E2Fs, allowing cell cycle progression. In MALAT1-depleted cells, p53 accumulation precedes the loss of Rb-phosphorylation. This suggests that the loss of E2F activity upon MALAT1 depletion occurs downstream of p53 activation and is perhaps a consequence of p53-dependent checkpoint activation. How DDR is evoked and how p53 is activated upon MALAT1-depletion remains to be determined. At an earlier time point (12 hr), control and MALAT1-depleted cells showed comparable levels of HDM2/MDM2, a negative regulator of p53 (Figure 5B). However, MALAT1-depleted cells showed increased levels of HDM2 at 24 and 48 hr time points, demonstrating p53-negative feed back loop mechanism.
To determine the importance of p53 activation in the reduced E2F activity and cell cycle arrest observed upon MALAT1 depletion, we compared E2F activity in isogenic wild-type (WT) and p53 −/− HCT116 cells in the presence or absence of MALAT1. Similar to what was observed in HDFs, MALAT1-depleted HCT116-WT cells showed reduced levels of E2F1-target mRNAs (Figure 5Ca–d, Figure S5Ba–c). However, HCT116-p53 −/− cells showed similar levels of E2F-transcribed mRNA in control and MALAT1-depleted cells (Figure 5Ca–d, Figure S5Ba–c). Based on these results, we conclude that a functional p53 is essential for the reduced E2F activity and proliferation defects observed in MALAT1-depleted human cells. It is not clear how MALAT1 influences p53 activity. Transient overexpression of MALAT1 in HDFs did not alter p53 levels (Figure S5Ca–b), indicating that p53 activation in MALAT1-depleted cells could be a part of specific stress response, including DDR.
The mitotic defects observed in MALAT1-depleted cells suggested that either MALAT1 has a role in mitosis via regulating the expression/pre-mRNA processing of genes that are involved in mitosis, or the defect observed in p53-deficient cells is a result of damaged DNA or stalled replication forks. Interestingly, mitosis was scored as the most significant canonical pathway of the cell cycle genes that were downregulated in MALAT1-depleted fibroblasts (Figure S6A and S6B). Also, MALAT1 levels were higher during mitosis compared to G2, further supporting the potential involvement of MALAT1 in mitotic progression (Figure 1A). To further examine the potential involvement of MALAT1 in mitotic progression, HeLa (Figure 6A–6C, Figure S6C) or WI-38 (Figure 6D–6E, Figure S6D–S6F) cells were synchronized in G1/S and were released to S-phase in presence of control and MALAT1 antisense oligonucleotides. qRT-PCR results revealed that in G1/S-arrested cells, the levels of MALAT1 were comparable between control (scr) and MALAT1 antisense oligo-treated cells (Figure S6C, scr aphidi. 0 hr rel vs AS1 aphidi. 0 hr rel). However, 12 and 24 hr after G1/S release, cells showed efficient knockdown of MALAT1 (Figure S6C, scr 12 or 24 hr aphidi. rel vs AS1 12 or 24 hr aphidi. rel), indicating that MALAT1 was depleted only in cells beyond G1/S. Flow cytometry analyses showed that the scr-oligo-treated cells showed normal cell cycle progression (Figure 6B). By contrast, while MALAT1-depleted cells showed normal progression in the first 12 hr of release, a significant population of them subsequently accumulated in mitosis by 24 hr after release (Figure 6B, G2/M cells 21% in control vs 37% in MALAT1-depleted cells). Microscopic analyses of the 24 hr-release cells revealed a higher percentage of broken nuclei in MALAT1-depleted cells, indicative of mitotic defects (Figure 6Ca–b). Similar results were also observed in MALAT1-depleted WI-38 cells (Figure S6D–S6F). We also compared the changes in expression of genes involved in mitosis in the 24 hr post G1/S release cells in control (scr-24 hr rel) versus MALAT1-depleted (AS1-24 hr rel) WI-38 cells. In comparison to control cells (scr-24 hr rel), MALAT1-depleted cells (AS1-24 hr rel) showed a consistent reduction in the RNA and protein levels of several of the mitotic genes analyzed (Figure 6D and 6E). Altogether, our results suggest that in human cells, MALAT1 plays a crucial role in mitotic progression that is independent of its involvement in G1 to S transition.
We previously demonstrated that in cancer cells, MALAT1 modulates alternative splicing of pre-mRNAs and regulates the cellular activity of SR splicing factors [50]. In order to determine whether the aberrant expression of cell cycle genes in MALAT1-depleted HDFs were due to alterations in alternative splicing, we analyzed the genome-wide changes in alternative splicing using exon microarrays in control and MALAT1-depleted HDFs (using two independent antisense oligonucleotides against MALAT1). Approximately ∼15% of genes showed changes in alternative splicing (cassette exon inclusion or exclusion) between control and MALAT1-depleted cells (Table S5), which is consistent with our previous findings in HeLa cells [50]. Surprisingly, GO analyses of the mRNAs with altered alternative splicing revealed that cell cycle was not among the top biological processes affected in MALAT1-depleted cells (Figure 7A). Indeed, we observed changes in alternative splicing of a few of the key mitotic regulators, including CENPE (Figure 7B). A large proportion of the cell cycle genes (involved in both G1/S transition and M phase) whose expression was reduced in MALAT1-depleted cells did not show any change in alternative splicing of their pre-mRNAs, implying that the expression of these genes was regulated at the level of transcription or mRNA stability. We reasoned that MALAT1 could influence the expression of key cell cycle regulators (transcription activator or repressor), and that would in turn result in the alteration of transcription of a large number of downstream cell cycle genes. We consistently observed alternative splicing changes for one such transcription regulator B-MYB in MALAT1-depleted cells (Figure 7B). B-MYB is a transcription factor that is required for the transcription of a large number of genes involved in mitotic progression [63], [64], [65]. To determine the involvement of MALAT1 in the SR splicing factor-mediated alternative splicing of B-MYB pre-mRNA, we examined the interaction of SRSF1 with B-MYB exons in the presence or absence of MALAT1. Ribonucleoprotein immunoprecipitation (RNA-IP or RIP) analysis using SRSF1 antibody followed by qRT-PCR revealed the increased association of SRSF1 to B-MYB exons in MALAT1-depleted WI-38 and HeLa cells (Figure 7C and Figure S7A). In addition to B-MYB, SRSF1 displayed increased association with CENPE exons in MALAT1-depleted cells. Both B-MYB and CENPE transcripts displayed aberrant alternative splicing in MALAT1-depleted cells (Figure 7B–7C). We had previously reported changes in alternative splicing of MGEA6 in MALAT1-depleted HeLa cells [50] but not in WI-38 cells (Figure 7B), indicating that MALAT1 could influence alternative splicing in a cell type specific manner. Interestingly, SRSF1 showed increased association to the alternatively spliced exon in MGEA6 only in MALAT1-depleted HeLa cell extracts (Figure S7A) and not in WI-38 cells (Figure 7C), indicating a positive correlation between changes in exon inclusion and increased association of SRSF1 to specific exonic regions upon MALAT1 depletion. Finally, HeLa cells overexpressing SRSF1 showed increased association of SRSF1 to mRNAs, including B-MYB and CENPE (Figure S7B). These results suggest that reduced cellular levels of MALAT1 enhanced the binding affinity of SRSF1 to specific pre-mRNAs, including B-MYB and CENPE pre-mRNA and influenced alternative splicing. We and others have previously reported that MALAT1-depleted cancer cells show increased levels (including the dephosphorylated pool) of SRSF1 [50], [66]. Therefore, the increased association of SRSF1 on B-MYB and CENPE exons observed in MALAT1-depleted cells could be due to the increased total SRSF1 pool. However, unlike HeLa cells, MALAT1-depleted HDFs (WI-38) and mammary epithelial cells (MCF7) did not show any increase in the total or dephosphorylated pool of SRSF1 (Figure S7Ca–c), indicating that other factor/s along with MALAT1 could contribute to changes in SR protein levels observed in MALAT1-depleted HeLa cells. Interestingly, WI-38 cells in which we co-depleted p53 and MALAT1 showed an increased pool of dephosphorylated SRSF1, further indicating the cooperation between MALAT1 and p53 (Figure S7Cd). Taken together, these results suggest that the changes in alternative splicing of pre-mRNAs observed upon MALAT1 depletion is not entirely due to alterations in the total cellular levels of SR proteins but could be due to alterations in the affinity of SR proteins to specific pre-mRNAs.
Recent studies have demonstrated the role of B-MYB in the recruitment of the transcription factor FOXM1 at the promoter of mitotic genes in G2 phase and their transcription as B-MYB-depleted cells fail to recruit FOXM1 to the mitotic gene promoters and showed reduced mitotic gene expression and mitotic defects [64], [67]. Depletion of B-MYB is also known to destabilize FOXM1 transcripts [64]. MALAT1-depleted HDFs showed reduced levels of B-MYB mRNA and protein (Figure 7D and 7E) with a concomitant reduction in FOXM1 transcript levels (Figure S2Ab). B-MYB is transcriptionally induced during G1/S and S-phase of the cell cycle; therefore, reduced levels of B-MYB in MALAT1-depleted fibroblasts could be due to the defects in G1/S and S-phase progression [63], [64]. To determine this, we examined the levels of B-MYB in control and MALAT1-depleted HeLa cells since MALAT1-depletion in HeLa cells did not show defects in S-phase entry (Figure S4E). We observed reduced levels of B-MYB in MALAT1-depleted HeLa cells, indicating that the reduction in B-MYB levels upon MALAT1-depletion is not a reflection of defects in S-phase (Figure S7D). We have also confirmed this by conducting a time course experiment where HeLa cells were synchronized in G1/S, treated with control and MALAT1-specific antisense oligonucleotides and released so that the cells proceed to S and G2 phase. The levels of B-MYB mRNA were subsequently determined at various time points (Figure 7F and Figure S7E). In control cells, B-MYB mRNA level was increased during 2, 4, 6 hr post G1/S-release and showed a reduction at 8 hr post release from G1/S (Figure 7F) [64], [68]. However, MALAT1-depleted post G1/S cells, though showed normal S-phase progression, failed to activate B-MYB and continued to show reduced levels of B-MYB (Figure 7F and Figure S7E). These results reveal that cellular levels of MALAT1 control the expression and/or RNA processing of B-MYB.
B-MYB is an E2F1 target gene. It is possible that the changes in B-MYB expression observed in MALAT1-depleted cells could be due to reduced E2F1 activity. To determine this possibility, we compared the levels of B-MYB mRNA in HCT116 WT and p53 −/− cells in the presence or absence of MALAT1. Both WT and p53 −/− cells showed reduction in B-MYB mRNA levels upon MALAT1 depletion (Figure 7G). This was in contrast to other E2F target genes, whose expression remained unaltered in MALAT1-depleted p53 −/− cells (Figure 5Ca–d, Figure S5Ba–c). Furthermore, exogenously expressed E2F1 could not rescue B-MYB mRNA levels in MALAT1-depleted HDFs, indicating that the reduced B-MYB levels in MALAT1-depleted cells is not due to alterations in E2F1 activity (Figure S7F). Based on these results, we conclude that MALAT1 regulates B-MYB mRNA levels by controlling specific post-transcriptional events.
Pathway analyses indicate that the majority of mitotic genes downregulated in MALAT1-depleted fibroblasts were activated by FOXM1 and that B-MYB regulated this process (Figure 7H). Based on our results, we hypothesize that MALAT1 modulates the FOXM1-mediated transcription of mitotic genes by controlling the expression of B-MYB. To test this model, we conducted a rescue experiment where we transiently expressed B-MYB cDNA in MALAT1-depleted fibroblasts and examined the expression of mitotic genes. Cells transiently expressing B-MYB rescued the expression of FOXM1 and several of its targets (Figure 7I). These data demonstrate that B-MYB acts downstream of MALAT1 and the altered expression of a large number of mitotic genes in MALAT1-depleted fibroblasts are due to reduced levels of B-MYB. To determine whether the cellular levels of MALAT1 truly influence B-MYB expression, we overexpressed MALAT1 in fibroblasts and examined the relative changes in B-MYB mRNA levels. MALAT1-overexpressing cells showed an increase in the levels of B-MYB mRNA (Figure 7J and Figure S7G). At the same time, MALAT1 overexpression did not alter the levels of several other genes that are involved in cell proliferation (including E2F1 and PCNA), indicating that the cellular levels of MALAT1 specifically influence the expression of B-MYB.
The human genome contains a large number of lncRNAs that are dynamically expressed in tissue-, differentiation- and cell type-specific patterns. LncRNAs influence the expression and intracellular distribution of specific proteins and also recruit factors of the chromatin-modifying complexes to specific chromatin sites [4], [9], [10], [13], [15]. Aberrant expression of several lncRNAs is associated with cancer [28], [35], [37], [69]. In the present study, we examined the involvement of the abundant nuclear-retained lncRNA MALAT1 in cell cycle progression. MALAT1 is overexpressed in several cancer types, including lung, breast, colon and hepatocarcinoma, and overexpression of MALAT1 in various cell lines enhanced cell proliferation whereas in nude mice, increased levels of MALAT1 promoted tumor formation [28], [42], [52], [70], [71], [72], [73]. Recent studies have also demonstrated that depletion of MALAT1 impairs proliferative and invasive properties of cancer cells [52], [53], [70], [73]. These results imply that MALAT1 acts as a pro-proliferation gene. However, it was not clear how MALAT1 controls cell proliferation.
In human cells, MALAT1 interacts with several pre-mRNA splicing factors, including SR splicing factors, and MALAT1 influences alternative splicing of pre-mRNAs by regulating the distribution and activity of SR splicing factors. In general, SR splicing factors are constitutively expressed in all cell types and cell cycle stages. Despite this fact, alternative splicing is regulated in a cell type- or cell cycle-specific manner [74], [75]. Based on our recent observations, we propose that differential expression of MALAT1 in different cell types or cell cycle phases, acts as a ‘molecular sponge’ to titrate the cellular pool of SR splicing factors. This in turn creates a gradient of functionally competent splicing factors in the cell ultimately controlling alternative splicing. In support of this model, we demonstrated a cell cycle-regulated expression of MALAT1. Cellular levels of MALAT1 were high during G1/S and M; two important periods of the cell cycle where a cell either commits to enter DNA replication phase or undergoes cell division (M). Depletion of MALAT1 in human diploid cells compromised G1-to-S transition and mitotic progression. Transcriptome microarray in MALAT1-depleted human cells revealed downregulation of genes involved in G1/S and mitotic progression, further supporting the involvement of MALAT1 in cell cycle progression.
A recent study proposed a role for MALAT1 in cell cycle progression through its involvement in regulating E2F1 activity [54]. The authors suggested that MALAT1 modulates the interaction of unmethylated Pc2, a polycomb protein, with E2F1 in serum-activated cells, and such an association facilitates E2F1 SUMOylation, leading to the activation of serum-induced genes [54]. Further, MALAT1-depleted cells showed proliferation defects and an inability to activate E2F target genes upon the addition of serum. Based on these results, the authors proposed that the proliferation defects observed in MALAT1-depleted cells are due to the inactivation of E2F1 [54]. Similar to the previous study, we also observed cell cycle arrest and reduced expression of E2F target genes upon MALAT1 depletion only in specific cell lines (example includes HDFs). In these cells, MALAT1 depletion also resulted in the activation of p53 and its downstream target genes, including p21. It is well established that p53 induction results in cell cycle arrest, cellular senescence or apoptosis [76]. We speculate that the S-phase defects and activated cellular senescence observed in MALAT1-depleted fibroblasts is due to p53 activation and not due to E2F1 inactivation. Our conclusion is based on the observation that HeLa, U2OS, or WI-38-VA13, cells that display weak p53, p16 and Rb activity and retain normal E2F1 function (HeLa due to HPV18 infection, U2OS due to increased HDM2/MDM2 levels and WI-38-VA13 due to SV40 T-antigen) [59], [60], did not show G1 or G1/S arrest upon MALAT1 depletion. Additionally, the G1 arrest in MALAT1-depleted diploid fibroblasts could be successfully rescued by simultaneous depletion of p53. Furthermore, MALAT1 and p53 co-depleted HDFs and MALAT1-depleted HCT116 p53 −/− cells expressed E2F target genes involved in G1/S or S phase progression, indicating normal E2F activity. Finally time course study in MALAT1-depleted HDFs clearly demonstrated that p53 is activated prior to Rb dephosphorylation and E2F inactivation as is evident by the reduced expression of E2F target genes. These results strongly support our model that in normal diploid human lung fibroblasts, the G1- and S-phase defects observed upon MALAT1 depletion is primarily due to p53-mediated checkpoint activation. It remains to be determined how an lncRNA like MALAT1 influences p53 signaling pathway. It is possible that MALAT1 and p53 share a synthetically lethal relationship, where the silencing of one gene (MALAT1) is lethal in the context of the mutation in the second gene (p53). Such a synthetic lethal relationship has been previously reported for cdk2 and MYCN [77]. In this case, depletion of cdk2-induced apoptosis in neuroblastoma cells that overexpressed MYCN, whereas cdk2-depleted neuroblastoma cells with normal MYCN expression did not show any phenotype [77]. It is interesting to note that recent studies have demonstrated the role of other lncRNAs as the potential mediators of p53 signaling network [29], [40], [41].
Our study has demonstrated a role for MALAT1 in mitotic progression that is independent of its involvement in G1/S progression. Chromosome segregation defects were observed in cells even when MALAT1 was depleted in post G1/S stage of the cell cycle, indicating that the mitotic phenotype is not a consequence of defects in G1/S transition. MALAT1-depleted cells showed defects in the expression of a large number of genes involved in mitotic progression. At the same time, only a limited number of mitotic genes, including CENPE and B-MYB, displayed defects in alternative splicing upon MALAT1 depletion. CENPE is a kinesin-like motor protein that is specifically expressed during the G2 phase of cell cycle [78]. It localizes to kinetochores in pro-metaphase cells and plays a crucial role in both chromosome segregation and spindle elongation [78], [79], [80]. Altered splicing and reduced expression of CENPE in MALAT1-depleted cells could contribute to the mitotic defects. On the other hand, B-MYB is a transcription factor with known cell cycle control functions [63]. It is one of the dominant gene signatures for highly proliferative cells observed across multiple tumor types and is highly overexpressed in several cancers [81], [82], [83]. Recent studies demonstrated that during S-phase, B-MYB associated with the multiprotein MuvB complex and localized to the promoters of genes that are expressed during mitosis [64]. Further, the B-MYB-MuvB complex efficiently recruited FOXM1 transcription factor to the mitotic gene promoters and activated their transcription [64]. MALAT1-depleted cells showed a reduction in the cellular levels of B-MYB and its mitotic targets. Finally, exogenously expressed B-MYB rescued the expression of several of the mitotic genes in MALAT1-depleted cells. Based on these results, we suggest that the reduced expression of B-MYB in MALAT1-depleted cells is the major effector leading to aberrant mitotic gene expression. Interestingly, recently published ChiP-seq data reveals that B-MYB binds to the MALAT1 promoter and could potentially regulate MALAT1 expression, indicating that MALAT1 and B-MYB could be part of a positive regulatory loop [64].
In addition to the reduction in B-MYB and CENPE mRNA levels, MALAT1-depleted cells also showed changes in alternative splicing of B-MYB and CENPE transcripts and increased binding of SRSF1 to the exons of these transcripts. The change in alternative splicing of B-MYB and CENPE observed in MALAT1-depleted cells phenocopies that observed in cells overexpressing SRSF1. It is possible that the dynamic changes in MALAT1 levels during the cell cycle titrate the intracellular pool of SR proteins and its association with pre-mRNAs, which in turn influence alternative splicing, stability and expression of specific mRNAs, including B-MYB and CENPE. A recent study revealed differential binding of U2AF65 with MALAT1 in cells when the chromatin structure was perturbed [51]. Global increase in histone hyperacetylation decreased the recruitment of U2AF65 to pre-mRNAs, coincident with an increase in its binding to MALAT1 lncRNA [51]. This further supports the role that MALAT1 acts as a ‘molecular sponge’ to sequester the free pool of splicing factors in the nucleoplasm.
We and others have recently reported that the in vivo MALAT1 knockout (KO) mouse is viable and fertile and MEFs from the knock out (KO) mouse did not show any defects in alternative splicing and SR protein activity, indicating that MALAT1 is largely dispensable in mice [58], [84], [85]. The cell type- or organism-specific phenotype observed upon depletion of a particular gene is not specific to MALAT1, as earlier studies had reported similar results for other lncRNAs and protein-coding genes. In human cell lines, the HOTAIR lncRNA transcribed from the HOX C cluster inhibits transcription from HOX D cluster by guiding the recruitment of histone modifiers to specific chromatin in HOX D region [39], [86], [87]. However, a mouse in which the region of HOX C cluster spanning the entire Hotair was deleted, showed normal viability and did not show any defects in the HOX D cluster transcription and/or chromatin modifications [88]. Cyclin-dependent kinase 2 (cdk2), a kinase that along with cyclin E is known to play an important role in cell proliferation and G1/S transition, is another classical example of a gene that exerts a cell type- or cell line-specific phenotype upon its depletion. The in vivo cdk2 knock out mouse is viable and does not show any obvious defects in normal mitotic cell division cycles [89], [90]. However, depletion of cdk2 showed severe S-phase defects in several human cell lines (HeLa [cervical carcinoma], NCI-H1299 [non-small cell lung carcinoma], A375 & SKMEL5 [both melanoma]) [91], [92], [93], [94] but not in others (SW480 and HT-29 [both colon carcinoma]) [93], [95]. Similar to cdk2, in vivo KO mice for several other genes involved in cell proliferation (cyclin E, Cdk4, cyclin D) clearly lacked phenotypes, but showed cell proliferation defects when they were depleted in human cell lines. Finally, cyclin A2 KO mouse is embryonic lethal, as cyclin A2 is required for proliferation in hematopoietic and embryonic stem cells but is not required for cell cycle progression in mouse embryonic fibroblasts, further supporting the cell type-specific role of cell cycle regulators [96]. MALAT1 is involved in a cell type- or tissue type-specific function; accordingly, its depletion shows defects only in specific cells from those tissue lineages. In this context it is known that in mouse, MALAT1 shows a tissue-specific and developmental state-specific expression [43], [58]. Further, the genes whose expression is known to be controlled by MALAT1 could function in a cell type- or tissue type-specific manner. For example, B-MYB knockdown in fibroblasts induces cell cycle arrest and apoptosis [97], [98]. However, B-MYB is found to be dispensable for the proliferation of glioblastoma cells [65].
We have shown that in human cells MALAT1 positively regulates the expression of the oncogenic transcription factor B-MYB. It is interesting to note that the aberrant expression of MALAT1 has also been observed in several cancers [42], [72], [99]; whether such altered expression is a cause or an effect of carcinogenesis needs to be determined. Based on our results, we propose that abnormal expression of MALAT1 in specific cell types or tissues results in aberrant alternative splicing leading to misexpression of genes that are involved in cell cycle progression and/or cell death, thereby contributing to tumor progression.
HeLa, U2OS, HepG2, WT-MEFs were grown in DMEM containing high glucose, supplemented with penicillin-streptomycin and 10% fetal bovine serum (FBS) (Hyclone, Logan, UT). WI-38, WI-38-VA13 and IMR-90 cells were grown in DMEM containing high glucose + 10% FBS, and 1% non-essential amino acid (NEA). RKO cells were grown in MEM containing high glucose + 10% FBS (Hyclone). HCT116-WT and p53 −/− cells were grown in McCoy's 5A medium + 10%FBS. NIH-3T3 cells were grown in DMEM containing high glucose supplemented with 10% Bovine calf serum (BCS).
Phosphorothioate internucleosidic linkage-modified DNA antisense oligonucleotides were used to deplete human MALAT1. The oligonucleotides were transfected to cells two times (48 hr) within a gap of 24 hr, at a final concentration of 100 nM, using Lipofectamine RNAimax reagent as per the manufacturer's instructions (Invitrogen, USA). Double-stranded siRNAs (Sigma-Genosys, USA) [100] were also used to deplete MALAT1 from cells at a final concentration of 40–50 nM. Depletion of p53 and p16 were performed by smart-pool siRNAs (Dharmacon, Thermoscientific, USA) at 10 nM final concentration.
For Microarray Analysis, WI-38 cells were reverse transfected in triplicate with either control (scrambled antisense oligo) or two independent MALAT1 antisense oligonucleotides at a final concentration of 100 nM using Lipofectamine RNAiMax (Invitrogen, USA). Total RNA isolated 48 hr post-transfection, was amplified, labeled and hybridized to Illumina arrays (Refseq-8). Raw hybridization intensity data were log-transformed and normalized to yield Z-ratios, which in turn were used to calculate a Z-ratio value for each gene. The Z-ratio was calculated as the difference between the observed gene Z-ratios for the experimental and the control comparisons, divided by the standard deviation associated with the distribution of these differences [101]. Z-ratio absolute values ≥2.5 were chosen as cut-off values, defining increased and decreased expression, respectively. The complete microarray data is available at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE44240.
For exon arrays to determine changes in alternative splicing, total RNA (100 ng) was reverse transcribed and amplified using Ambion WT expression kit following the manufacturer's suggested protocol. Sense strand cDNA was fragmented and labeled using Affymetrix WT terminal labeling kit. Duplicate samples from MALAT1 antisense oligonucleotides transfected cells or a control oligonucleotide were hybridized to Affymetrix human Exon ST 1.0 GeneChip in Affymetrix hybridization oven at 45°C, 60 rpm for 16 hr. The arrays were washed and stained on Affymetrix Fluidics Station 450 and scanned on Affymetrix GeneChip scanner 3000 7G. Data were collected using Affymetrix AGCC software. Statistical and clustering analysis was performed with Partek Genomics Suite software using RMA normalization algorithm. Differentially expressed genes and alternatively splicing events were identified with ANOVA analysis. Genes that are up- or downregulated more than 2-fold and with a p<0.001 were considered significant. Significant genes were analyzed for enrichment for pathways using DAVID bioinformatics database (http://david.abcc.ncifcrf.gov/) and Ingenuity Pathway Analysis software. The exon array data is available at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE43830.
For the validation of the Microarray data, RNA was reverse transcribed into cDNA using Multiscribe reverse transcriptase. qRT-PCR was performed using primers for genes that showed ≥2.5-fold change in the Microarray data.
Relative levels of gene expression were normalized to GAPDH or 7SK RNA. In the RNA-IP experiments, the relative quantities of IP samples were normalized by individual inputs respectively. Results are represented as mean ± SEM of three independent experiments. Comparisons were performed using two-tailed paired Student t-test. *p<0.05, **p<0.01 and ***p<0.0001. In the microarray analyses, relative quantities of gene expression change were normalized by z-ratio between experimental cell group and control cell group. Comparisons were performed using z-test with SAM protocols, plus ANOVA filtering for the sample groups. Cutoff were made for the significant genes with |z ratio|> = 2.5, z-test *p<0.05, ANOVA p-value< = 0.05 , false discovery rate < = 0.30, as well as the average gene expression values in each comparison groups were not negative. Further functional analyses were performed by using Parameterized Gene Set Enrichment (PAGE) algorithm with ALL of the genes on the array. **p<0.01 and ***p<0.0001.
For labeling of S-phase cells, BrdU was added in mid-log phase cells at a final concentration of 50 µM and incubated for 1 hr at 37°C. Cells were harvested and washed with PBS + 1% BSA. Cells were further resuspended in chilled 0.9% NaCl at a cell density of 2×106 cells/ml and fixed in equal volume of ethanol and incubated at −20°C for 1 hr. Further, ethanol was removed and the cells were resuspended and incubated in 2N HCl + 0.5% Triton X-100 solution for 30 min at RT followed by washing in 0.1 M Sodium tetraborate solution. Finally, the cells were resuspended in PBS/0.5%Tween-20 + 1% BSA and incubated with anti-BrdU antibody for 1 hr at RT. The cells were washed again with the same buffer and incubated in PBS + PI/Triton X-100 at 37°C for 15 min. Cells were analyzed on a flow cytometer.
Please see Protocol S1 and Table S6 for additional materials and methods.
|
10.1371/journal.pgen.1000690 | A Pol V–Mediated Silencing, Independent of RNA–Directed DNA Methylation, Applies to 5S rDNA | The plant-specific RNA polymerases Pol IV and Pol V are essential to RNA–directed DNA methylation (RdDM), which also requires activities from RDR2 (RNA–Dependent RNA Polymerase 2), DCL3 (Dicer-Like 3), AGO4 (Argonaute), and DRM2 (Domains Rearranged Methyltransferase 2). RdDM is dedicated to the methylation of target sequences which include transposable elements, regulatory regions of several protein-coding genes, and 5S rRNA–encoding DNA (rDNA) arrays. In this paper, we have studied the expression of the 5S-210 transcript, a marker of silencing release at 5S RNA genes, to show a differential impact of RNA polymerases IV and V on 5S rDNA arrays during early development of the plant. Using a combination of molecular and cytological assays, we show that Pol IV, RDR2, DRM2, and Pol V, actors of the RdDM, are required to maintain a transcriptional silencing of 5S RNA genes at chromosomes 4 and 5. Moreover, we have shown a derepression associated to chromatin decondensation specific to the 5S array from chromosome 4 and restricted to the Pol V–loss of function. In conclusion, our results highlight a new role for Pol V on 5S rDNA, which is RdDM–independent and comes specifically at chromosome 4, in addition to the RdDM pathway.
| In plant genomes, the RNA–directed DNA methylation (RdDM) process induces de novo methylation of cytosines at repeated sequences. The RNA polymerases Pol IV and Pol V are two key components of the RdDM pathway. Pol IV acts with RDR2 (RNA–dependent RNA polymerase 2) and DCL3 (Dicer-Like protein 3) to generate short interfering RNAs (siRNAs). Pol V, in a partnership including AGO4 (Argonaute4) and DRM2 (Domains Rearranged Methyltransferase 2), drives DNA methylation at the targeted sequence. Changes in 5S (ribosomal DNA) rDNA methylation, 5S rDNA chromatin compaction, and 5S siRNA accumulation in Pol IV/V mutants have been reported. However, 5S rDNA arrays were considered together. In the present study, we observed an overexpression of the atypic 5S-210 transcript, restricted to the 5S rDNA array from chomosome 4. This derepression is specific to the Pol V–loss of function (and not to Pol IV) and comes in addition to the RdDM pathway. The Pol V–loss of function induces also the chromatin decondensation of the derepressed 5S locus at chomosome 4. Our results highlight a new role for Pol V which, suprisingly, appears to be Pol IV– and RdDM–independent.
| The plant-specific RNA polymerases firstly named Pol IVa and Pol IVb and now referred as Pol IV and Pol V [1], contribute to siRNA production and are essential to RNA-directed DNA methylation (RdDM) [2]–[6]. The revised nomenclature denotes the largest subunits of Pol IV and Pol V as NRPD1 and NRPE1. Both Pol IV and Pol V share the second largest subunit NRPD2. In current models of RdDM silencing pathway [7]–[10], Pol IV is speculated to produce single-stranded RNA transcripts from heterochromatic repeated regions. These transcripts, converted onto double-stranded RNAs by RDR2 (RNA-DEPENDENT RNA POLYMERASE 2), are processed into siRNA duplexes by DCL3 (DICER-LIKE 3) [11]–[14]. The resulting siRNAs are methylated at 2′ hydroxyl groups of 3′-terminal nucleotides by HEN1 (HUA ENHANCER 1) prior to loading into AGO4/RISC complex [15]. Recently, Pol V and DRD1 (DEFECTIVE IN RNA-DIRECTED DNA METHYLATION 1) were found to be required to mediate production of non-coding transcripts which are necessary for steps downstream of siRNA biogenesis [1]. Recent evidence suggests that siRNAs/AGO4 complexes bind to Pol V transcripts, ultimately guiding the de novo DNA methyltransferase DRM2 [16] and histone modifying complexes to the target loci [17],[18].
Targets of the Pol IV/Pol V-dependent RdDM include transposable elements, regulatory regions of several protein-coding genes and 5S rRNA-encoding DNA (rDNA) arrays [2]–[5], [19]–[23].
We [24] and others [4],[5],[11] have reported changes in 5S rDNA methylation, 5S rDNA chromatin compaction and 5S siRNA accumulation in Pol IV/V mutants. However, in these reports, 5S rDNA arrays, which have separate functions and locations (For a review, [25]) were considered together.
Arabidopsis thaliana contains approximately 1000 copies of 5S RNA genes per haploid genome. 5S rDNA is arranged in tandem arrays [26] located within the pericentromeric heterochromatin of chromosomes 3, 4 and 5 in the Columbia accession [27],[28]. Only 5S-repeat clusters located on chromosomes 4 and 5 are transcribed by Pol III to produce the 120 nucleotide (nt) transcripts which are integrated into ribosomes [29]. Nevertheless, it is not yet clear what proportion of these 5S genes is active at any one time.
Indeed, both active 5S-repeat clusters contain transcribed and repressed 5S RNA genes in WT plants. Previous study revealed that in adult wild-type plants, only «major» 5S RNA genes were expressed whereas «minor» genes, which diverge from «major» ones at only one or several positions, are repressed [30]. In addition to «major» and «minor» 5S RNA species, we have previously identified an atypical 5S RNA (5S-210) composed of genic and intergenic regions which is a marker of silencing release at 5S RNA genes [31]. The presence of a chromosome-specific T stretch identifies the chromosome origin of 5S-210 transcripts [29].
To determine the individual contribution of Pol IV and Pol V in the transcriptional silencing and heterochromatic state of each 5S array, we assayed 5S-210 transcript expression, chromatin compaction and DNA methylation during early development. We have shown that Pol IV, Pol V and several actors of the RdDM, contribute to maintain a transcriptional silencing of 5S RNA genes at chromosomes 4 and 5. Moreover, we showed an additional Pol V activity, Pol IV- and RdDM-independent which drives silencing of 5S rDNA specifically at chromosome 4. The large silencing release observed at chromosome 4 in NRPE1 and NRPE5a (Pol V) mutants, is accompanied by a decompaction of the corresponding 5S rDNA locus, as well as decompaction of NOR loci.
We previously identified a 5S transcript, 210 bases-long (5S-210) which is a marker of silencing release at 5S RNA genes. This 5S-210 transcript homologous to the 120 nt genic region and 90 nt from the adjacent intergenic region [31] contains the sequence of the chromosome-specific T-stretch which identifies its 5S array-origin (Figure 1).
In order to define the impact of Pol IV (formerly Pol IVa) and Pol V (formerly Pol IVb) on 5S RNA genes silencing, we analysed 5S-210 accumulation by RT-PCR experiments in WT, nrpd1 (mutant of the largest subunit of Pol IV), nrpd2 (mutant of the common subunit of Pol IV and Pol V), nrpe1 (mutant of the largest subunit of Pol V) and nrpe5a (mutant of a new Pol V-specific subunit; [32]) plants. RDR2 (RNA-DEPENDENT RNA POLYMERASE 2), DCL3 (DICER-LIKE 3), AGO4 (ARGONAUTE 4), DRM2 (DOMAINS REARRANGED METHYLTRANSFERASE 2), HEN1 (HUA ENHANCER 1) and HDA6 (HISTONE DEACETYLASE 6) involved with Pol IV and POL V in the production of siRNAs and the associated DNA methylation and histone modifications, were also tested.
5S-210 transcripts overaccumulated by a factor between 2 and 2,5 in nrpe1, nrpd2 and nrpe5a compared to WT plants. On the contrary, nrpd1, rdr2, dcl3, hen1, ago4, drm2 and sil1 (mutant allele of HDA6) plants accumulate similar 5S-210 transcripts quantities than WT (Figure 2A and 2B).
The release of silencing in nrpe1 and nrpe5a indicates that Pol V is involved in the quantitative regulation of 5S-210 expression. The results obtained with nrpd2 confirm these observations. The absence of silencing release in nrpd1, rdr2, dcl3, ago4, drm2, hen1 and sil1 shows that they have no influence, at the quantitative level, on 5S-210 RNA transcription.
5S rDNA arrays are located within the pericentromeric heterochromatin of chromosomes 3, 4 and 5. We assessed whether this Pol V-mediated silencing operates on all 5S arrays or operates selectively. Using the T stretch signature [29], we analysed the origin of the 5S-210 transcripts in Pol IV/Pol V mutants and in two mutants of proteins involved in respectively upstream and downstream RdDM steps i.e. RDR2 and DRM2. Sequencing of RT-PCR products revealed that 5S-210 transcripts only originate from the transcriptionally active 5S-repeat clusters located on chromosomes 4 and 5. In WT conditions, the 5S array from chromosome 5 contributes for 70%, the 5S array from chromosome 4 contributing for the remaining 30% of the 5S transcripts (Figure 2C). The proportions and quantities are not significantly different in nrpd1, rdr2 and drm2 compared to WT. In nrpe1 and nrpd2, the overaccumulation of 5S-210 transcripts results only from the silencing release of the 5S array from chromosome 4, since the quantity of 5S-210 transcripts provided by chromosome 5 is unchanged (Figure 2C).
These results refine our previous conclusions indicating that the silencing release observed in nrpe1 concerns the 5S array from chromosome 4; the results obtained with nrpd2 confirm these observations. Therefore, the Pol V-mediated repression operates on the 5S array from chromosome 4.
In order to determine whether chromatin decompaction is associated to the release of silencing observed at chromosome 4, we performed FISH experiments with both 5S rDNA and 45S rDNA probes on WT, nrpd1, nrpe1 and nrpd2 plants. The chromosome 4 is the only chromosome to carry both rDNA species. 5S rDNA from chromosome 4 colocalizes with 45S rDNA signals in almost all nuclei [33]. 5S signals outside chromosome 4, i.e. coming from chromosomes 3 and 5 were considered together (Figure 3A and 3B).
FISH analysis revealed that 36% of the WT nuclei contain one or two decondensed 5S signals at chromosome 4. The proportions are similar in WT and nrpd1 mutant whereas a significant larger proportion (59%) of the nrpe1 and (77%) of nrpd2 nuclei harbor decondensed 5S signals at chromosome 4. From these analyses, we conclude that 5S arrays from chromosome 4 are decondensed in nrpe1, and these results are confirmed with nrpd2 observations, whereas nrpd1 has no visible effect on this compaction. Therefore, NRPE1/Pol V has the ability to act on the chromatin of 5S rDNA from chromosome 4 in a NRPD1/Pol IV- independent manner.
We observed no significant variation in the proportion of nuclei with decompacted 5S signals at chromosomes 3 and 5 in WT, nrpd1, nrpe1 and nrpd2 plants (Figure 3B). These results show that nrpd1 and nrpe1 mutations have no visible effect on chromatin compaction of 5S arrays from chromosomes 3 and 5. From these analyses, we conclude there is a correlation between the large silencing release and the decompaction of 5S rDNA at chromosome 4 in nrpe1 mutant.
A question arises from these results: how and why is the 5S rDNA array from chromosome 4 concerned by this particular Pol V regulation? The main difference between 5S arrays from chromosomes 4 and 5 is the close proximity of NOR4 (Nucleolar Organizing Region) with the 5S array from chromosome 4. Therefore, we hypothesized that both rDNA arrays (5S and 45S) might be co-regulated and be the site of common decompaction events.
We analyzed the NOR condensation in WT, nrpd1, nrpe1 and nrpd2 mutants (Figure 3A and 3C). There are four NORs in diploid A. thaliana, but they tend to coalesce and in agreement with Pontes et al. [34] we detected two to four NORs in 79% of WT nuclei and more than four signals in 21% of them. Similar results were obtained in nrpd1 nuclei. Contrarily, more than four NORs FISH signals were observed in 37% and 39% of nrpe1 and nrpd2 nuclei reflecting a significant increase of NOR decompaction compared to WT and nrpd1 nuclei.
These results show that no detectable chromatin decompaction is observed for NOR loci and 5S array from chromosome 4 in nrpd1 mutant whereas a concomitant decompaction event is observed for 5S rDNA from chromosome 4 and 45S loci in nrpe1 and nrpd2 mutants. This might illustrate a common regulation, Pol V-mediated, at 45S and adjacent 5S rDNA.
The quantitative 5S RNA derepression associated with chromatin decompaction is unambiguously limited to the 5S array from chromosome 4 in PolV-loss of function mutants. However, previous results have unequivocally shown that 5S rDNA is hypomethylated in NRPD1, RDR2 and DRM2 mutants of the RdDM pathway [4],[5],[11],[24],[35]. 5S rDNA loci consist of both active and heterogenous-silent copies of the 5S RNA gene. We previously showed that the release of silencing of 5S RNA genes illustrated by the increase of the proportion of “heterogenous” 5S transcripts (i.e. containing some mutations in the genic region) can occur without changes of the 5S RNAs quantity [30],[36].
We therefore decided to analyze the heterogeneity of 5S-210 transcripts produced by 5S arrays from chromosomes 4 and 5 to test whether NRPD1, RDR2, DRM2 and NRPE1 act on each 5S array. As shown Table 1, the proportion of heterogenous 5S RNA from chromosomes 4 and 5 is enhanced in nrpd1, nrpe1, nrpd2, rdr2 and drm2 compared to the WT. It demonstrates the impact of Pol IV, RDR2, DRM2 and Pol V on the 5S array from both chromosomes.
These results refine our previous conclusions indicating that silencing of 5S RNA genes from chromosome 5 is controlled by Pol IV, RDR2, DRM2 and Pol V at the qualitative level. Their mutation has an equivalent effect i.e. a derepression of heterogenous 5S RNA genes is observed without increasing the total amount of 5S-210 RNA. Silencing of 5S RNA genes from chromosome 4 is controlled by Pol IV, RDR2, DRM2 and Pol V at the qualitative level, and Pol V exerts an additional role acting at the quantitative level.
The results show that Pol IV, RDR2, DRM2 and Pol V act in the RdDM pathway, to maintain the repression of heterogenous 5S RNA genes from chomosomes 4 and 5. They also show a specific and additional role of Pol V, Pol IV-, RDR2- and DRM2- independent and therefore RdDM-independent, on 5S array from chromosome 4. There is therefore a differential Pol V impact on 5S arrays from chromosomes 4 and 5.
To confirm that Pol V-additional activity at chromosome 4 is RdDM-independent, we assayed cytosine methylation in asymmetric sequence context (CHH), which largely results from RdDM. Indeed, if the Pol V-additional activity is RdDM-independent, a lower DNA methylation in nrpe1 and nrpd2 compared to RdDM mutants is not expected [3],[37],[38]. Digestion with NlaIII for which the same restriction site is present in the intergenic spacer of every 5S rDNA unit of chromosomes 4 and 5 and PCR amplification with primers hybridizing to the chromosome-specific T-stretch were performed. As shown Figure 4, there is a reduction of CHH (CAT for NlaIII) methylation in nrpd1, nrpe1, nrpd2, rdr2 and drm2 mutants compared to the WT for both 5S rDNA arrays. Moreover, the same reduction of methylation is observed at chromosome 4 in all the mutants. These results reveal that derepression and decompaction of the 5S array at chromosome 4 in nrpe1 is not associated with specific changes of 5S rDNA asymmetric methylation. They also show that the similar reduction of methylation observed at both arrays results from the loss of RdDM pathway.
Previous results have shown that 5S rDNA is subject to a variety of overlapping regulation pathways, such as the limiting amount of TFIIIA (TRANSCRIPTION FACTOR IIIA; 5S rDNA specific) [30], the methylation-independent MOM1 pathway [31], the DDM1/MET1- pathway [30],[39], as well as the Pol IV/Pol V RdDM [4],[24],[39]. Although the impact of the Polymerases IV and V on 5S rDNA was previously demonstrated on the basis of DNA hypomethylation, decrease of 5S small RNA accumulation and chromatin decompaction [4],[24],[39], the relative impact of Pol IV and Pol V, their potential selective action on the different 5S arrays and the consequence of their mutation on 5S rDNA silencing of each 5S array were unknown.
In this paper, we have shown that silencing of 5S RNA genes from chromosomes 4 and 5 is controlled at the qualitative level by RdDM including Pol IV, RDR2, DRM2 and Pol V activities. Loss of this silencing pathway leads to the derepression of heterogenous 5S RNA genes, without an increase of total 5S RNA amount and without a detectable chromatin decompaction. These results are consistent with the previously reported reduction or elimination of 5S siRNAs (the 1003 siRNA) as well as with the 5S rDNA hypomethylation observed in nrpd1, rdr2, drm2, nrpe1, ago4 or dcl3 mutants [4],[5],[11],[24],[40]. Our results demonstrate that each of the two 5S arrays is a target of RdDM. Derepression of heterogenous 5S RNA genes at chromosomes 4 and 5 in mutants of the RdDM pathway is associated with reduction of asymmetric methylation at each array.
In addition to the RdDM process common to 5S arrays from chromosomes 4 and 5, an additional Pol V activity specifically applies to chromosome 4. Higher amounts of 5S-210 transcripts from chromosome 4 are observed in nrpe1 and nrpe5a, two Pol V-specific subunits, correlating with a specific decompaction of this 5S rDNA array in nrpe1 and nrpd2. On the contrary, similar amounts of transcripts were obtained in WT and all the tested RdDM mutants, and chromatin decompaction is absent in nrpd1. Therefore, the additional role of Pol V on chromosome 4 is Pol IV- and RdDM- independent. Moreover, the Pol V activity observed specifically at chromosome 4 is not associated with changes of 5S rDNA asymmetric methylation, in agreement with the similar global 5S rDNA methylation observed in nrpd1 and nrpe1 (symmetric and asymmetric methylation; [5]. It suggests that Pol V complex might recruit other repressive epigenetic marks than DNA methylation and probably also other than H3K9me2, since 5S rDNA is not decompacted in met1 mutants [36], despite a decrease of H3K9me2 [42]. H3K27me2, a repressive mark methylation-independent, which labels 5S rDNA [41], is a potential candidate.
The release of silencing of 5S RNA genes that we observed in RdDM mutants without changes of the 5S RNAs quantity might be surprising. However, we previously observed this phenomenon in several mutants [36]. Moreover, in S. cerevisiae the number of active rRNA genes can change more than twofold without changing steady-state rRNA transcript levels [43]. On the contrary, higher amounts of 5S-210 transcripts from chromosome 4 are observed in nrpe1and nrpe5a, suggesting that the large 5S rDNA decompaction enhances the transcription i.e. it might facilitate the access to transcription factors and/or the reinitiation process.
The needs for stoichiometric amounts of 5S and 45S rRNA implies that co-regulation events apply to 5S and 45S rDNA. We have shown that NOR loci and 5S array from chromosome 4 are decompacted in nrpe1 and nrpd2 mutants illustrating a common regulation process. This Pol V activity, NRPD1 and RdDM-independent, suggests that an alternative pathway i.e. without NRPD1 and RdDM partners, exists to target 5S and 45S rDNA loci.
Our results open the question about the interest for the plant to have a 5S array whose regulation is more dependent on Pol V. There is a need of the plant kingdom for rapid, reversible changes in gene expression, to respond to growth demands or environmental changes. Some of the rDNA repeats may be specifically targeted for silencing as a mechanism to modulate or fine-tune total cellular rDNA gene activity.
In conclusion, our results provide new insights on Pol V function in 5S rDNA regulation. On the basis of previous results [4],[5],[11],[24] it was supposed that RdDM (including Pol IV, RDR2 and Pol V activities) was responsible for both silencing and compaction of 5S RNA genes. The current study confirms the participation of both Pol IV and Pol V in the 5S RNA gene silencing of each array, and clearly shows that 5S RNA silencing and compaction are not necessarily linked. It demonstrates that 5S rDNA decompaction is due to the sole Pol V- loss of function, and not to that of both Pol IV and Pol V. Finally, it shows that Pol V has two different activities on the same 5S array, one in the RdDM pathway and the other one, RdDM-independent and chromatin-based which is not associated with methylation changes in early development.
Arabidopsis thaliana nrpd1a-1 (nrpd1), nrpd1b-1 (nrpe1), nrpd2a-1 (nrpd2), rdr2-1, dcl3-1, drm2-2 and the corresponding wild type young plantlets were from Columbia ecotype. nrpd1a-1, nrpd1b-1, rdr2-1, dcl3-1 seeds were obtained from Dr T. Lagrange (University of Perpignan, France). nrpd2a-1 and drm2-2 seeds were obtained from the Arabidopsis Biological Resource Center (Stock # SALK 095689 and 150863 respectively). sil1, ago4-1 and WT plants were in Landsberg erecta background. Seeds of WT, sil1 and ago4-1 plants were obtained from the NASC (stock numbers NW20 for Ler-0, N1894 for sil1 and N6364 for ago4-1). nrpe5a-1 and WT WS seeds were obtained from Dr T. Lagrange. After synchronization 2 days at 4°C, seeds were grown on a germination medium (MS Salt [Duchefa Biochemie] supplemented with 3% sucrose and 0.8% BactoAgar) in a growth chamber using a 16 h light (120 µE.m−2.sec−1)/8 h dark regime at 23°C. Plantlets at 4 days post-germination were used.
Prior to use, tissues were fixed in ethanol/acetic (3∶1) solution. Probes were labeled by PCR using gene specific primers with biotin-16-UTP (Roche) or digoxigenin-11-UTP (Roche). FISH experiments were performed according to Mathieu et al. (2003). Biotin-labeled (5S rDNA) and digoxigenin-labeled (25S rDNA) probes were used. Avidin conjugated with Texas Red (1∶500; Vector Laboratories) followed by goat anti-avidin conjugated with biotin (1∶100; Vector Laboratories) and avidin–Texas Red (1∶500) were used for the detection of the biotin-labeled probe; mouse anti-digoxigenin (1∶125; Roche) followed by rabbit anti-mouse fluorescein isothiocyanate (FITC) (1∶500; Sigma) and Alexa 488-conjugated goat-anti-rabbit (Molecular Probes) were used for the detection of the digoxigenin-labeled probe. Before microscopic analysis, nuclei were stained with DAPI (4′, 6-diamidino-2-phenylindole).
For microscopic analysis, an epifluorescence Imager Z1 microscope (Zeiss) with an Axiocam MRm camera (Zeiss) was used. Fluorescence images for each fluorochrome were captured separately through the appropriate excitation filters. The images were pseudocolored, merged and processed with the Adobe Photoshop software (Adobe Systems). 45 to 62 nuclei were analyzed for each genotype.
Compaction of 5S arrays from chromosomes 4 and chromosomes 3+5 were considered separately. Each group of 5S array (4 or 3+5) was considered as decompacted when at least one signal was decondensed. The number of NOR signals was analyzed in 45 to 62 nuclei for each genotype.
Proportion of 5S-210 transcripts from chromosomes 4 and 5 and percentage of heterogenous 5S-210 transcripts were compared with Fisher's exact test for a 2×2 contingency table. The probabilities were calculated from a one-tailed test. Statistical analysis of 5S-210 transcripts amounts were performed using the nonparametric Mann-Whitney U-test with mean values comparison. For statistical analyses of 5S rDNA and NOR compaction, a comparison of proportions Z-test was used. The probabilities were calculated from a one-tailed test. Interval confidence (IC) was calculated for each proportion with a confidence level of 99%.
Aliquot of 1 µg of total RNA was treated with DNA-free™ Kit (Ambion) and 100 ng of DNase-treated total RNA was used as input in semi-quantitative RT-PCR reactions using the OneStep RT-PCR Kit (Qiagen). Controls were performed without reverse transcription step to detect contaminating DNA. Amplification of ACTIN2 RNA was used as an internal control and to normalize RNA amounts. Detection of 5S-210 and ACTIN2 transcripts was performed in the same reaction tube. Amplification conditions: 50°C for 30 min (reverse transcription step); 95°C for 15 min (reverse transcriptase inactivation step); 30 cycles [95°C for 30 s; 51°C for 30 s; 72°C for 45 s]; 72°C for 10 min. 5S-210 transcripts were amplified using primers RTPCR5S1 (5′-GGATGCGATCATACCAG-3′) and 5SUNIV2 (5′-CGAAAAGGTATCACATGCC-3′). ACTIN2 transcripts were amplified using primers ACT2-F and ACT2-R according to Vaillant et al. [36].
Amounts of amplicons were estimated using a Versadoc coupled to the QuantityOne software (Biorad).
PCR products were subcloned in the pGem-T easy plasmid using the pGem-T vector system (Promega). Sequencing was performed using the CEQ 2000 Dye terminator cycle sequencer (Beckman). Computer sequence analysis was performed with the Clustawl program (www.infobiogen.fr).
Genomic DNA was extracted from seedlings according to the cetyltrimethylammonium bromide (CTAB) method [44]. 200 ng of DNA was digested with NlaIII, which recognizes the sequence CATG and is inhibited by methylation of the cytosine, overnight at 37°C. PCR amplification was subsequently done on 20 ng of digested and undigested DNA using the following primers: CATCCCTC(T)17 specific for chromosome 4, CATCCCTCTTTTATGTTTAACC specific for chromosome 5 and TCGAAAACAATGCTTGAACAAG used for both arrays. The specificity of this amplification was tested on YACs containing respectively the 5S array from chromosome 4 (YAC 9D3) and from chromosome 5 (YAC 6A1) [45].
Primers ACT2-F and ACT2-R Vaillant et al. [36] were used to amplify ACTIN2. ACTIN2 amplification was used to control equal templates concentration.
Total digestion was controlled with APETALA1 gene (accession AT1G69120.1) which contains 2 non methylated NlaIII sites. Primers: 2F: TTTGGTTGGTTCAGATTTTGTTTCG and 2R: CCAAGAATCAGTGGAGTATTCGAAG were used for PCR amplification.
Amplification conditions: 20 cycles [95°C for 30 s; 60°C for 30 s; 72°C for 30 s]; 72°C for 10 min for 5S rDNA. ACTIN2: 28 cycles [95°C for 30 s; 55°C for 30 s; 72°C for 30 s]; 72°C for 10 min. APETALA1: 25 cycles [95°C for 30 s; 58°C for 30 s; 72°C for 30 s]; 72°C for 10 min. 5 to 8 experiments were performed.
Amounts of amplicons were estimated using a Versadoc coupled to the QuantityOne software (Biorad). The DNA methylation level was calculated with the ratio: Amount of amplicon in digested DNA/Amount of amplicon in non-digested DNA.
|
10.1371/journal.pcbi.1004353 | Modeling Inhibitory Interneurons in Efficient Sensory Coding Models | There is still much unknown regarding the computational role of inhibitory cells in the sensory cortex. While modeling studies could potentially shed light on the critical role played by inhibition in cortical computation, there is a gap between the simplicity of many models of sensory coding and the biological complexity of the inhibitory subpopulation. In particular, many models do not respect that inhibition must be implemented in a separate subpopulation, with those inhibitory interneurons having a diversity of tuning properties and characteristic E/I cell ratios. In this study we demonstrate a computational framework for implementing inhibition in dynamical systems models that better respects these biophysical observations about inhibitory interneurons. The main approach leverages recent work related to decomposing matrices into low-rank and sparse components via convex optimization, and explicitly exploits the fact that models and input statistics often have low-dimensional structure that can be exploited for efficient implementations. While this approach is applicable to a wide range of sensory coding models (including a family of models based on Bayesian inference in a linear generative model), for concreteness we demonstrate the approach on a network implementing sparse coding. We show that the resulting implementation stays faithful to the original coding goals while using inhibitory interneurons that are much more biophysically plausible.
| Cortical function is a result of coordinated interactions between excitatory and inhibitory neural populations. In previous theoretical models of sensory systems, inhibitory neurons are often ignored or modeled too simplistically to contribute to understanding their role in cortical computation. In biophysical reality, inhibition is implemented with interneurons that have different characteristics from the population of excitatory cells. In this study, we propose a computational approach for including inhibition in theoretical models of neural coding in a way that respects several of these important characteristics, such as the relative number of inhibitory cells and the diversity of their response properties. The main idea is that the significant structure of the sensory world is reflected in very structured models of sensory coding, which can then be exploited in the implementation of the model using modern computational techniques. We demonstrate this approach on one specific model of sensory coding (called “sparse coding”) that has been successful at modeling other aspects of sensory cortex.
| The diverse inhibitory interneuron population in cortex has been increasingly recognized as an important component in shaping cortical activity [1]. However, it remains unclear in many settings how the inhibitory circuit specifically contributes to the neural code. While theoretical and simulation investigations of proposed neural coding models could be extremely valuable for providing insight into the role of inhibition, many current high-level functional and mechanistic models do not include inhibitory cell populations that approach the biophysical complexity seen in nature.
Though the main ideas likely extend to other areas, for concreteness we will focus the present discussion on the primary visual cortex (V1). In V1, visual information is encoded using a rich interconnected network of excitatory principal cells and inhibitory cells, and different coding functions appear to be implemented by distinct inhibitory populations [2, 3]. Though V1 has been extensively studied through experiment and modeling, there are often significant discrepancies between what is known about biophysical sources of inhibition and how inhibitory influences are instantiated in a model. For example, in previous high-level functional coding models (e.g. in [4–6], with the exception of [7] as discussed later), neural activity is often treated as a signed quantity without explicitly distinguishing between excitatory and inhibitory cell types. On the other hand, while state-of-the-art large scale mechanistic models (e.g. [8]) typically include a distinct inhibitory population, these types of models often use a single recurrent connectivity pattern (e.g., weights that decrease with spatial separation). This approach results in interneurons with uniform physiological properties and without the complex tuning diversity observed in inhibitory interneurons.
For theoretical and simulation studies to illuminate the role of inhibition in neural coding, it is imperative that coding models begin to incorporate experimental observations regarding the distinct properties of excitatory cells and inhibitory interneurons. Specifically, to realistically investigate the role of inhibition in neural coding, models should incorporate at least three major properties while staying faithful to the coding rule and other desirable properties (e.g., robustness):
Inhibitory and excitatory interactions arise from distinct cell types, and synapses from an inhibitory cell cannot have excitatory influences on postsynaptic cells and vice versa (Dale’s law [9]); Excitatory neurons generally outnumber inhibitory interneurons, with E/I ratios recently estimated to be in the range 7:1 to 6:1 (apparently preserved across animals [10, 11]); and The interneuron population has diverse tuning properties [12], including to varying degrees both orientation tuned and untuned interneurons in cat [13] and rodent V1 (reviewed in [14]).
The main contribution of this paper is to demonstrate a systematic computational method for effectively incorporating these biophysical interneuron properties into dynamical systems implementing neural coding models. In our proposed approach we exploit the fact that in many cases of interest, the total required inhibition is highly structured due to the relationship between the coding model and the statistics of the inputs being encoded. Similar to efficient coding hypotheses that postulate compact representations of sensory stimuli, the structure of the sensory statistics and the coding model can also be used to implement the required inhibition with a parsimonious computational structure. Specifically, we propose to reformulate the connectivity matrix to respect Dale’s law and exploit the inhibition structure in a matrix factorization to minimize the number of inhibitory interneurons. Furthermore, we leverage recent results from the applied mathematics community on advanced matrix factorizations to develop an approach that demonstrates the observed diversity of orientation tuning properties in inhibitory interneurons.
The end result of this approach is a network implementation that is functionally equivalent to the original model, but which has an interneuron population that better respects the three major biophysical properties ignored by many current coding models. In addition to this primary goal of providing a recipe for including inhibitory interneurons into coding models, this approach also suggests possible functional interpretations of some biophysical properties of the interneuron population. In particular, we propose that while Dale’s law may reflect a physical constraint of individual cells, in contrast the E/I ratio can be viewed as an emergent characteristic of a population implementation that maximizes efficiency by minimizing the number of interneurons and thus maintenance costs. In addition, we demonstrate that the orientation tuning diversity in the inhibitory population can arise from differential connectivity patterns between the excitatory and inhibitory cells.
In a recurrent network implementing a neural coding model, each node in the network is generally driven by both exogenous inputs (i.e., bottom-up inputs due to the stimulus or top-down feedback) and lateral connections from other cells in the same network. These lateral connections are often described in terms of a connectivity matrix G, where the element [G]m, n describes synaptic strength from the nth neuron to the mth neuron. While G can take many forms, the structure is governed by the coding model and the statistics of the stimuli being encoded.
To illustrate how G arises for a family of commonly-used coding models, we consider the Bayesian inference paradigm that has found increasing support as a framework for studying neural coding [15]. While there are many ways to develop a neural coding model based on the ideas of optimal inference, one of the most common approaches is to assume a generative model where the sensory scene is composed of a linear combination of basic features (i.e., causes) that must be inferred. Specifically, a linear generative model for vision proposes that an image patch s ∈ ℝN (i.e., an N-pixel image patch) can be approximately written as a linear superposition of M dictionary elements {ϕi} representing basic visual features (i.e., there are M principal cells):
s = ∑ i = 1 M a i ϕ i + n = Φ a + n , (1)
where the coefficients for each feature are {ai}, n represents a noise source, and the N × M matrix Φ consists of one dictionary element on each column. These dictionary elements are often interpreted as the receptive fields (RFs) of a principal cell, such as spiny stellate cells or pyramidal cells.
Given the dictionary Φ and the stimulus s, the coefficients a in the linear generative model (taken to be principal cell activities, such as instantaneous firing rates) can be found by maximum a posteriori (MAP) estimation. Assuming Gaussian noise and a prior distribution P(a), the MAP estimate is found by minimizing the negative log of the posterior:
E ( a ) = 1 2 ‖ s - Φ a ‖ 2 2 - λ log P ( a ) , (2)
where λ is a scalar capturing the model SNR. When the prior distribution is log-concave (as are many common distributions including the exponential family [16]), the inference can be achieved by simple descent methods. The simplest dynamical system for this coding strategy would be a network implementing gradient descent with population dynamics given by
τ a ˙ = Φ T s - G a + λ ∇ log P ( a ) ,
where τ is the system time constant and the M × M recurrent weight (connectivity) matrix is given by G = ΦTΦ. G can be interpreted as a recurrent matrix because its off-diagonal terms capture the influence between cell activities. In particular when we assume that the prior is independent, i.e. log P(a) = ∑i log P(ai), as is common in efficient coding models, G captures all the recurrent influence. Note that any dynamical system involving a derivative of an energy function such as Eq (2) will contain a recurrent matrix G of this form.
While the most obvious implementation of the network would use a single interneuron for each entry of G (connecting two cells), there are many implementations that would result in a functionally equivalent coding rule. For example, one of the approaches we will utilize is to model the connectivity between the interneurons and principal cells using a matrix factorization:
G = U Σ V T
where the VT matrix captures the synaptic connections onto a set of interneurons from the principal cells, the U matrix captures the synaptic connections from these interneurons back onto the network of principal cells, and Σ is a diagonal matrix representing the independent gains/sensitivity of each interneuron.
As a concrete relevant example, we will demonstrate the proposed approach in the context of a dynamical system implementing a sparse coding model of V1, where a population of cells encodes a stimulus at a given time using as few active units as possible. The sparse coding model (combined with unsupervised learning using the statistics of natural images) has been shown to be sufficient to explain the emergence of V1 classical and nonclassical response properties [17–19], potentially has many benefits for sensory systems [20–23], and is consistent with many recent electrophysiology experiments [24–26]. The sparse coding model has been implemented in networks that have varying degrees of biophysical plausibility (e.g., [18, 27–30]), though this model has rarely been implemented with distinct inhibitory neural populations (excepting [7], discussed later).
The sparse coding model can be viewed as a special case of inference in the linear generative model described above with
E ( a ) = 1 2 ‖ s - Φ a ‖ 2 2 + λ ‖ a ‖ 1 , (3)
where ‖ a ‖ 1 = ∑ i = 1 M ∣ a i ∣, corresponding to a Laplacian prior with zero-mean. We base our discussion on a dynamical system proposed in [27] that uses neurally plausible computational primitives to implement sparse coding. This system has strong convergence guarantees [31, 32], can implement many variations of the sparse coding hypothesis [33], and is implementable in neuromorphic architectures [29, 34, 35]. Specifically, the system dynamics for this sparse coding model are:
u ˙ ( t ) = 1 τ [ Φ T s - u ( t ) - ( G - I ) a ( t ) ] a ( t ) = T λ ( u ( t ) ) , (4)
where I is the identity matrix, u are internal state variables for each node (e.g., membrane potentials), G = ΦTΦ governs the connectivity between nodes, and Tλ(⋅) is the soft thresholding function. Note that despite not using steepest descent on Eq (3), this network model still has recurrent connections described by the connectivity matrix G = ΦTΦ. In the simulations in this study, the dictionary Φ is pre-adapted to the statistics of the natural scene with a standard unsupervised learning method, resulting in Gabor wavelet-like kernels that resemble V1 classical receptive fields [17].
This dynamical system model requires influences between cells that are described by the matrix G, but it is agnostic about the network mechanism that implements these interactions. Specifically, the model as described in [27] does not incorporate a separate population of inhibitory interneurons with any non-trivial interneuron structure, and this naïve description would only imply a point-to-point connection between all pairs of cells in the network as illustrated in Fig 1. This model is therefore unhelpful in its current form for understanding the coding properties of the inhibitory population.
This sparse coding network will serve as a concrete demonstration of the proposed strategy to incorporate more biophysically realistic inhibitory interneurons. The example network we use has 2048 excitatory neurons and has the same parameters as in a previous work [19] (see Materials and Methods).
As a first step towards a biologically realistic interneuron population encoding model, we show that Dale’s law can be respected in the model by decomposing the recurrent connectivity matrix G into matrices representing excitatory and inhibitory interactions. Specifically, the recurrent connectivity matrix G can be decomposed into inhibitory and excitatory effects:
G = G + + G - = G Inhib + G Excite , (5)
where G+ are the positive elements of the matrix (representing the inhibitory recurrent connections) and G− are the negative elements (representing excitatory recurrent connections).
While GExcite can be implemented by direct synapses between excitatory principal cells, the inhibitory component GInhib requires inhibitory interneurons between the relevant principal cells. To capture these disynaptic connections, we factor the inhibitory matrix into two matrices: GInhib = UVT. For a simple stylized illustration, the network in Fig 1 shows an example implementation with
GInhib=(00wI1,E30)︸U(wE1,I1wE2,I100)︸VT, (6)
and
G Excite = ( 0 0 0 0 0 0 0 0 0 0 0 0 0 0 w E 4 , E 3 0 ) . (7)
Using the approach above, we can derive a network implementation that is equivalent to the dynamical system instantiating the desired neural coding rule but that also has inhibitory cell properties that can be varied by the choice of factorization for GInhib. For a simple concrete example, we can achieve the same encoding as Eq (4) while incorporating an inhibitory population by using the decomposition:
G Inhib = I G + (8)
where I is the identity and plays the role of U; G+ as defined in Eq (5) plays the role of VT. The resulting network is shown in Fig 2A. While this approach does utilize distinct excitatory and inhibitory sub-populations, it still requires M inhibitory neurons (i.e., one for each principal cell) and all inhibitory cells in this implementation have the same orientation tuning properties as the excitatory cells (see Supporting Information S1 Text “RFs of inhibitory cells in the direct implementation”). While this may introduce orientation tuning diversity due to the orientation tunings of the excitatory population, the diversity is distributed uniformly [36] instead of a bimodal dichotomy observed in the inhibitory population [37].
In areas such as V1, the principal excitatory cells are presumed to form the explicit representation of the stimulus that is communicated to higher cortical areas while inhibitory neurons are presumed to play a more localized computational role within a circuit. Using limited physical resources, there are many desirable properties for the stimulus representation: informational efficiency matched to scene statistics [38], stability to small stimulus changes [4], and simple downstream decoding [39]. The principal cell population in V1 appears to be substantially overcomplete (i.e., in both cats and primates, the estimated ratio between the output fibers and the input fibers ranges from 25:1 to 50:1 [40]), which is a feature adopted in some coding models because it can help achieve these desirable properties [40].
In contrast, if inhibitory neurons only need to achieve a computational goal for the circuit without requiring these same stimulus coding properties, there is no need for an overcomplete inhibitory population. In fact, the system could exploit this structure to use the fewest number of inhibitory cells possible to avoid incurring unnecessary cell maintenance costs [41]. In contrast to the direct model of Fig 2A, this approach would require interneurons that communicate simultaneously with a population of excitatory neurons rather than a single excitatory neuron. As an aside, we note that the reasoning above suggests that the inhibitory population should be overcomplete in systems where these neurons do form the explicit stimulus representation. Indeed, this is proposed in a theory of olfactory bulb encoding where granule cell interneurons form the olfactory representation and are an overcomplete population [42].
A natural question to ask is, what is the minimum number of inhibitory cells required to implement the influences specified by the matrix G? Said mathematically, what choice of factorization results in the fewest number of inhibitory cells, corresponding to the number of columns of U and V? In many cases of interest, the connectivity matrix G is likely to be low-rank (i.e. M > rank(G)), providing an opportunity to achieve an efficient implementation of the interneuron population by “compressing” the recurrent connectivity to its most essential components. There are two different causes of low-rank structure in G for the types of models considered in this study. First, an overcomplete representation of the principal cells implies directly that G is low-rank (i.e., M > N ≥ rank(Φ) = rank(ΦTΦ) = rank(G)). Second, natural images are highly structured, meaning that image patches have fewer “degrees of freedom” than the number of photoreceptors N being used to transduce the image (i.e. N > rank(Φ) = rank(G)) [43, 44]. This high level of input redundancy means that the connectivity structure implementing this coding rule also has structure that can lead to a simplified implementation. Taking both of these aspects together, models that encode stimuli with low-dimensional structure using an overcomplete code could expect to efficiently implement the encoding rule with highly-structured, low-rank connectivity matrix G.
In detail, these two sources of low-rank structure can be exploited to achieve the same coding function of Eq (4) with fewer interneurons than a direct implementation of Eq (8). The original description in Eq (4) of G as a Gramian matrix gives rise to the following decomposition of the recurrent matrix:
G = Φ T Φ = ( Φ + + Φ - ) T ( Φ + + Φ - ) = Φ + T Φ + + Φ - T Φ - ︸ G Inhib + Φ + T Φ - + Φ - T Φ + ︸ G Excite , (9)
shown in Fig 2B. Assuming first that we only take advantage of an overcomplete representation (i.e. the Φ matrix has more columns than rows because M > N), the resulting E/I ratio is M:N and requires (potentially many) fewer inhibitory cells than excitatory cells. However, this implementation does not produce the diversity of tuning properties observed in V1 interneurons, which can be either orientation tuned or non-orientation tuned (with no apparent structure) [37]. In fact, when using sparse dot stimuli to map out the RFs [17] of these interneurons, the resulting RFs have a dot-shaped structure (Fig 2B) inconsistent with cortical observations (see Supporting Information S1 Text “RFs of inhibitory cells in the Gramian decomposition” for discussion relating this RF shape to the network structure).
Further assuming that we exploit the fact that G encodes redundant structure in natural scenes, the recurrent connectivity can be represented by an even lower dimensional decomposition than Eq (9). This can be achieved by seeking a lowest-rank (i.e., fewest number of interneurons) recurrent matrix that is also a good approximation to G (noting that up to this point we have only examined strategies that exactly solve the original encoding problem). Written mathematically, this approximation is:
L = arg min L rank ( L ) , s.t. ‖ G - L ‖ F ≤ ϵ (10)
where ‖⋅‖F is the Frobenius norm. This is equivalent to solving:
L = arg min L ‖ G - L ‖ F , s.t. rank ( L ) ≤ r
with a suitable choice of r and ϵ. The solution to this problem can be found by the truncated singular value decomposition (SVD), known commonly as Principal Component Analysis (PCA). Note that in our case the truncated singular values are equivalent to the truncated eigenvalues because G is symmetric semi-positive definite. Specifically, we can decompose
G ≈ L = U Σ V T = ( U + + U - ) Σ ( V + T + V - T ) = [ U + Σ V + T + ( - U - ) Σ ( - V - T ) ︸ G Inhib ] + [ U - Σ V + T + U + Σ V - T ︸ G Excite ] , (11)
where U and V are truncated singular vectors with orthogonal columns and implement the recurrent synaptic weights (see the Discussion section for the biological plausibility of assuming orthogonal connectivity); Σ is a positive diagonal matrix truncated from the full SVD and implements the interneuron gain (see Materials and Methods).
The resulting inhibitory population receives dense, low-rank connections from the principal cells with synaptic weights defined by V + T (i.e., each row representing synapses convergent onto a single interneuron) as illustrated in Fig 2C. Note that another group of low-rank inhibitory cells with different detailed connectivity is defined by − V − T, but the qualitative characteristics of these cells are similar to those defined by V + T. Both groups in this population have a gain modulation defined by the diagonals of Σ, followed by projection back to the principal cells with synaptic weights defined by U+ and −U− (i.e., each row represents synapses convergent onto a single principal cell).
In our example sparse coding network, this implementation only requires 220 interneurons to capture about 99% of the variance in G, representing a significant savings compared to 2048 and 256 interneurons required in Eqs (8) and (9) respectively. However, the resulting interneurons are again not orientation tuned, lacking the diversity observed in V1 interneurons (Fig 2C). In the Supporting Information S1 Text “RF of inhibitory cells in low-rank decomposition”, we show that the receptive fields of this population approximate the principal components of Φ in a generative linear model and are thus untuned.
Inhibitory neurons are diverse. There are at least two populations with either tuned or untuned orientation selectivity [37]. At the same time, different inhibitory neurons connect to the excitatory population with different frequencies [45]. Could the diverse connectivity contribute to the differences in tuning? It is indeed conceivable that inhibitory neurons densely connected to the excitatory population combine inputs from different sources, and as a result have a broader selectivity. Conversely, inhibitory neurons connecting more sparsely and locally with the excitatory population might be more selective to the stimulus.
To test the hypothesis that tuning diversity could arise from differential connectivity, we decompose the recurrent connectivity matrix into two distinct matrices L and S G = L + S , (12)
where L is a dense matrix and S is a sparse matrix capturing relatively few inhibitory influences in G. To also respect the E/I cell ratio constraint, we would like L to be low-rank in particular so that a condensed representation could be achieved using SVD as demonstrated in the previous section.
Recently the applied mathematics community has developed a principled algorithmic approach known as Robust PCA (RPCA) [46–48] that effectively solves this decomposition problem. In this approach, a sparse matrix S that models “outliers” (having a disproportionate effect on the rank of G) is included so that the remainder L has a lower rank than G.
In the context of our study, an unstructured sparse connectivity matrix can result in a relatively large number of interneurons because there can be a large number of columns containing at least one non-zero value. To maintain the small number of interneurons, we also want the sparse matrix to be row or column-sparse (see for example the connectivity represented in Eq (6)). To achieve this, we used an adaptive version of RPCA (ARPCA) [49] to decompose the recurrent connectivity matrix G = ΦTΦ into a low-rank matrix L and a column-sparse matrix S by solving the following convex optimization program iteratively:
L , S = arg min L , S ‖ L ‖ * + ‖ Λ S ‖ 1 , subject to G = L + S , (13)
where ‖⋅‖* is the nuclear norm (i.e., the sum of absolute values of eigenvalues) to encourage L to have low rank, ‖⋅‖1 is the ℓ1-norm (i.e., the sum of absolute values of the vectorized matrix) to encourage sparsity, and Λ is a diagonal weighting matrix updated at each iteration to encourage column sparsity in S. The update rule for Λ is given by
Λ i , i = β ‖ S ( i ) ‖ 1 + γ , (14)
where S(i) is the ith column of S, and β and γ control the speed of adaptation. At each iteration, the columns of S with smaller entries are assigned larger values of λ, thus encouraging the values in that column to become even smaller and eventually approach zero. The end effect is that the algorithm converges to a decomposition where only a few columns in S are non-zero (see Materials and Methods for details). We note that the RPCA formulation in Eq (13) is a natural extension to SVD in Eq (10): instead of constraining the power in G−L (via the Frobenius norm), we now model this difference using a structured matrix S.
After convergence, as before we perform a singular value decomposition (SVD) on the low-rank matrix L = UΣVT. To respect Dale’s law we separate out the excitatory and inhibitory influence similar to Eq (9) in each matrix:
G = L + S = U Σ V T + S = ( U + + U - ) Σ ( V + T + V - T ) + ( S + + S - ) = [ U + Σ V + T + ( - U - ) Σ ( - V - T ) + S + ︸ G Inhib ] + [ U - Σ V + T + U + Σ V - T + S - ︸ G Excite ] . (15)
With this decomposition, the recurrent matrix can be rewritten with separate excitatory and inhibitory recurrent interactions. In the sparse coding model example described earlier (Eq (4)), the equivalent network dynamics are:
u˙(t)=1τ[ ΦTs︸feed-forward−(U+ΣV+T+(−U−)Σ(−V−T)︸low-rank+S+D︸sparse)︸recurrent inhibitorya(t)+(I−GExcite)︸recurrent excitatorya(t)−u(t) ], (16)
where D is a diagonal matrix with 0s and 1s on the diagonal and represents the synaptic weights on the sparsely-connected interneurons made by the principal cells (Fig 3).
With a parameter choice that strikes a balance between sparseness and low rank (see Materials and Methods), the E/I cell ratio in the model network is also close to the observed ratio. Specifically, with 2048 principal cells and 320 inhibitory interneurons (220 in the low rank population and 100 in the sparse population), the model network has an E/I cell ratio of 6.4:1.
Decomposing the connectivity matrix in this manner results in two distinct populations of inhibitory interneurons with a relative size controlled by the magnitude of the average weights in the matrix Λ. The first subpopulation (exemplified by the inhibitory cell I1 in Fig 3) originates from the low-rank connectivity matrix L, and has properties described in the previous section. The second subpopulation (exemplified by I2 in Fig 3) originates from the sparse connectivity matrix S. This population receives one-to-one (i.e. sparse) connections with unit weights defined by the diagonal matrix D from the principal cells, and projects back to the principal cells with weights defined by S. Because S is column-sparse, the rows in D that correspond to the zero columns in S can be set to 0 without altering the recurrent influence. Said another way, we can eliminate the zero rows of the D matrix and the zero columns of S, meaning that only a few interneurons are required in this subpopulation (Fig 3).
This model network accurately solves the sparse coding inference problem (Eq (3)), despite using only the top principal components of L in the approximation to the recurrent matrix. This is shown in Fig 4, where we compare the original network (i.e., the idealized implementation of Eq (4) that is not biophysically plausible) with the approximation described above in the metrics of interest. Specifically, for a number of grating test patches we plot the final value of the energy function (i.e., the quantity to be minimized in Eq (3)), along with the individual quantities relevant to the objective: the sparsity of the final answer (measured by the number of active coefficients ‖a‖0) and the relatve ℓ2 error for the input image (‖s−Φa‖2/‖s‖2). As demonstrated in Fig 4, the approximation achieves performance very similar to the original (mean relative error of energy approximation 0.008±0.001). We note specifically that in both the approximated and the original network, the activity is very sparse—only up to 5% of all 2048 neurons are active.
Interestingly, the sparse and low-rank interneuron populations in RPCA show the same kind of diverse orientation tuning as V1 inhibitory cells in vivo. The low-rank inhibitory population has RFs that are mostly untuned (Fig 5A and 5C; orientation tuning mapped using a grating stimulus centered in the middle of the visual field), comparable to the untuned inhibitory neurons observed in cats [37] (Fig 5B). The sparse inhibitory population has RFs that resemble the primary cell RFs in Φ and are orientation tuned (Fig 5D and 5F; orientation tuning mapped using a grating stimulus centered on the RF of the interneuron), comparable to the tuned inhibitory neurons observed in cats [37] (Fig 5E). This tuning dichotomy is expected from the difference in connectivity: the orientation-tuned inhibitory RFs arise from orientation-selective inputs from single principal cells (i.e., sparse synaptic connections), whereas untuned RFs arise from dense synaptic inputs from many principal cells of different tunings.
For simplicity in the above model we treat the interneurons as instantaneous linear units (Eq (16)). To make the model more biologically realistic, we can incorporate the same first-order dynamics (i.e., leaky integration) used by the principal cells into the interneurons. Specifically, the full population dynamics can be written as:
u ˙ ( t ) = 1 τ [ Φ T s - ( U + a I, L1 ( t ) + ( - U - ) a I, L2 ( t ) + S + a I, S ( t ) ) + ( I - G Excite ) a ( t ) - u ( t ) ] a ( t ) = T λ ( u ( t ) ) a ˙ I, L1 ( t ) = 1 τ ( Σ V + T a ( t ) - a I, L1 ( t ) ) a ˙ I, L2 ( t ) = 1 τ ( Σ ( - V - T ) a ( t ) - a I, L2 ( t ) ) a ˙ I, S ( t ) = 1 τ ( D a ( t ) - a I, S ( t ) ) , (17)
where aI, L1(t) and aI, L2(t) are the dynamic responses of the two low rank interneuron populations and aI, S(t) is the response of the sparse population. Here we assume that inhibitory neurons have the same time constant as the principal cells. As shown in Fig 6, the model defined in Eq (17) still accurately solves the original sparse coding problem (mean relative error of energy approximation 0.029±0.004). Note that due to the added dynamics, the new dynamical system needs more numerical integration steps to converge (all systems run for 100 steps in Fig 6 vs. 25 steps in Fig 4, resulting in some differences in the sparsity-rMSE tradeoff).
The main contribution of this study is a biologically plausible computational framework for including inhibitory interneurons in efficient dynamical system models of neural coding based on ideas from matrix factorization and convex optimization. From the demonstrated results, we conclude that techniques such as low-rank plus sparse decomposition can be used to find implementations of a recurrent connectivity matrix that produce equivalent population dynamics while using an inhibitory structure that matches many biophysical properties, including respecting Dale’s law, known E/I cell ratios, and diversity of orientation tuning properties. In our example of a network implementing sparse coding, the resulting representation is nearly as accurate as the idealized coding model while being much more faithful to the biophysics of the inhibitory population. Because the proposed approach only depends on the structure of the recurrent matrix (which may be common among many energy based models, including many other derivatives of sparse coding [33]), we expect that the results will be applicable to many dynamical systems implementing neural coding models.
Our approach suggests that the excitatory to inhibitory cell ratio in V1 is an emergent property of interneurons implementing efficient visual coding in a resource-conserving way. Specifically, in our model a comparatively small number of interneurons efficiently route the inhibitory influence by taking advantage of the overcomplete and low-rank (redundant) structure in the recurrent connectivity pattern. We have further demonstrated that the tuning diversity of interneurons could arise from differential connectivity with the excitatory population—a prediction that could be tested experimentally.
Recently a few studies explicitly introduced inhibitory interneuron populations into high-level functional encoding models. Lochmann et al. [50] developed a generative model that demonstrates contextual effects in sensory coding and includes a population of inhibitory neurons. These inhibitory cells contribute to efficient perceptual inference through input targeted divisive inhibition. However, this model only works with binary one dimensional inputs and the inhibitory connectivity pattern predicted by this model presently lacks anatomical support at the cortical level. Therefore, its connection with realistic visual encoding remains unclear.
In a more recent work, Boerlin et al. [51] illustrated a way to include a separate inhibitory population in an efficient coding spiking network that estimates the state of an arbitrary linear dynamical system. While providing a spiking model for the inhibitory cells, their approach did not investigate the issues of excitatory-inhibitory cell ratio and tuning diversity. It should also be noted that the Gram recurrent matrix in our model also occurs in their model (their Eq 10). It is therefore possible that our approach could be applied in their scenario.
Another recent study [7] has developed a spiking sparse coding network based on [28] that incorporates a population of inhibitory cells with connectivity weights adapted to natural scenes. Similar to the results of our study, the work in [7] has found that a relatively small number of inhibitory cells are sufficient to provide recurrent competition required for sparse coding. In contrast, the present study formulates a framework for including biologically plausible inhibitory interneurons in a wide range of models in a way that can potentially be proven equivalent computationally to the original model objective (e.g., Eq (2)). Furthermore, the present work captures the observed tuning diversity of inhibitory interneurons in V1. We note that the work in [7] does use a more biophysically realistic learning rule, whereas the present paper uses a global convex optimization approach on a fixed connectivity matrix that may have been established through a learning process.
Our model gives several experimentally verifiable predictions of interneuron properties that we detail in this section. We also note that while biologically plausible, there are limitations with the current model (see the Caveats section later).
First of all, our model predicts the existence of two distinct connectivity patterns between inhibitory interneurons and principal cells: the recurrent connections between principal cells and the low-rank interneurons are dense while the recurrent connections between the principal cells and the sparse interneurons are selective. According to these patterns, a likely biological correlate for the low-rank interneurons in mice is the fast-spiking parvalbumin-expressing (PV) interneurons, which receive dense synaptic inputs from nearby pyramidal cells of diverse selectivities [52], and project densely back to neighboring pyramidal cells [53]. Interestingly, as predicted by our model, the PV neurons indeed have broader selectivity than principal cells [54]. Similarly in cats, a subgroup of fast-spiking interneurons were found to have broader tunings than other interneurons [55]. Note that this broad selectivity means that the interneuron population derived from the low-rank component will use a dense code (i.e., most cells participating for most stimuli) even in coding rules such as the sparse coding example used in this work.
It is less clear what biological correspondence is most appropriate for the sparse interneuron population arising in the model. One candidate is the irregular firing cannabinoid receptor-expressing (CB1+) neurons, which have been shown to be more sparsely connected to the principal cells than the PV neurons [45]. However it is unclear what selectivity properties these neurons have in the visual cortex. Another candidate is the somatostatin expressing (SOM) neurons, which are orientation selective and have weaker response [54], similar to the sparse population in our model. If they indeed correspond to the sparse population in our model, we predict that these neurons receive sparser connections from the principal cells compared to the PV neurons (this however might differ from layer to layer, as evidenced by a recent study in L2/3 [56]).
In addition to general connectivity patterns, our model also provides predictions on the distribution of inhibitory synaptic weights in V1. As shown in Fig 7A, we observe a near log-normal distribution of the inhibitory synaptic weights when using a dictionary adapted to the statistics of natural scenes. Compared to a standard log-normal distribution however, the model distribution has a longer tail towards the smaller values as visible from the Q-Q plot (Fig 7B). Note that while the heavy tail is significant, in fact only a small part of the distribution deviates from log-normal (below the -2.33 quantile—corresponding to 1% of the cumulative density). Compounded with the difficulty of measuring from weak synapses, we anticipate that this tail would be hard to capture from experimental measurements. We note that there was a previous study [57] demonstrating a log-normal distribution between excitatory neurons, but we are unaware of similar findings for inhibitory cells. It should be noted that this model distribution is in agreement with the prediction of a previous model of spiking sparse coding [28]. Whether this is true in physiology requires further experimental validation.
In discussing the recurrent connections in the network of Fig 3, we concentrate mostly on the inhibitory connections represented by the GInhib term. The excitatory influences are assumed to be implemented by direct excitatory-excitatory connections represented by the connectivity matrix I−GExcite. The identity matrix I is assumed to be implemented by an independent mechanism that results in self-excitation. Biologically, there are at least three ways this self-excitation could be achieved: through “autapses” [58] (although most of these self-connections were observed in inhibitory cells); through excitatory interneurons that connect back to the principal cells; or through dendritic back-propagation [59].
We note that some of the biological features of inhibitory circuits modeled in this work are still controversial among physiology studies. For example, although Dale’s law is a generally accepted operating principle, it was recently suggested that neurons can segregate neural transmitters to different synapses [60]. As another example, the diversity of tuning properties and the functional roles of inhibitory interneurons are still controversial. Most studies on this topic were conducted in rodents (the study we compared our simulation to [37] in the Results being a notable exception), with few implications for primates and leaving substantial uncertainty even in mouse neocortex [14]. For example, it is still unclear whether PV interneurons have a diversity of tuning properties [61] or are mostly broadly tuned [62]. In addition, in our simulation the recurrent inhibition sharpens the orientation tuning of the principal cells [19]; in physiology, there are conflicting accounts of whether this is the case [2, 3]. In summary, the modeling results here should be considered as a demonstration of the capability of a theoretical model to reproduce a variety of detailed biological phenomenon, not as support for any specific anatomical inhibitory circuit structures and functions.
There are several biological details of the inhibitory population that the current model does not capture. First, the non orientation-tuned inhibitory interneurons in cat primary visual cortex have complex cell characteristics such as overlapping ON/OFF receptive fields (Fig 5B). To capture such features, a coding model involving complex cells may be necessary. Second, the current model does not attempt to capture the prevalent electrical and chemical interconnections between inhibitory interneurons in the cortex [1, 63]. These recurrent connections can potentially be incorporated by allowing off-diagonal entries in the gain matrix Σ. Third, we have treated inhibitory interneurons as continuous-time units with instantaneous dynamics (Eq (16)) or with first-order dynamics (Eq (17)). In reality, interneurons emit spikes and have diverse temporal dynamics involving short-term plasticity [64]. A previous work from our group [35] showed that the non-spiking sparse coding network (without a separate inhibitory population) can be equivalently implemented by a network of integrate and fire cells. While we would expect a spiking network with a similar connectivity pattern as we have demonstrated would exhibit similar kind of interneuron properties, it is unclear without further analysis whether using more biologically realistic spiking neurons would affect the overall dynamics. Finally, though it is well-known that thalamic inputs innervate inhibitory interneurons constituting feedforward inhibition [1], the model discussed in the main text does not include a detailed model of this feedforward component. However, we argue in the Supporting Information S1 Text “Feedforward inhibition” section that the cell ratio and orientation tuning properties could be modeled in a similar manner as the recurrent network.
It is known that neural network models with different parameters may share the same input-output functionality [65]. Similarly, there are other model configurations (i.e. inhibitory connection patterns) not considered in this work that could implement the same coding functionality. For one example, in the Supporting Information S1 Text “Global inhibition” section we consider the example of global inhibition structures. In this case, while very few inhibitory cells are needed, only non orientation-tuned inhibitory cells can be modeled.
A remaining question is whether the proposed decomposition can be learned in a biologically plausible way. While it is out of the scope of the current study, we do expect the orthonormal low-rank connectivity matrices to be learnable in a biologically plausible fashion. Indeed, with Sanger’s learning rule—a classical unsupervised learning method for feedforward neural networks that can be implemented locally—the network weights converge to orthonormal eigenvectors of the input [66]. Note that while the orthonormality emerges automatically from the learning rule, we are not suggesting that the singular vectors are the only plausible weights in the interneuron network. For example, performing a linear transform (e.g. a rotation) in the low-rank principal subspace gives rise to an alternative decomposition that maintains the cell ratio and tuning properties we have modeled. This alternative implementation may in fact have additional computationally benefits. For example, a linear transform equalizes the gain distribution in the SVD and potentially improves the robustness of the network against noise.
Eq (13) is a convex optimization problem that can be solved efficiently through numerical optimization techniques. In this study we solve this optimization problem through an adaptive version of Alternating Direction Method of Multipliers (ADMM), a robust dual ascent method [67]. Specifically, the inner loop of the algorithm finds the optimal L and S given a choice of Λ by alternating between a primal update that achieves (augmented) Lagrangian minimization and a dual update. The outer loop updates Λ according to Eq (14). See [49] for details of the algorithm.
We start with a model network of 2048 principal neurons with receptive fields adapted to 16 × 16 natural image patches using sparse coding [17]. The principal cell activities are interpreted as the sparse coefficients of a dynamical system implementing sparse coding (a in Eq (4) constrained to be positive) [27] with a threshold value λ = 0.1, as was done previously in [19].
In the proposed implementation, the required number of inhibitory interneurons is governed by the rank of L and the number of non-zero columns in S. To achieve a biophysically accurate small E/I cell ratio, we would like both the rank of L and the number of non-zero columns of S to be small. However, these are two competing requirements whose tradeoff depends on the parameters in Eqs (13) and (14). Indeed, making L lower rank necessarily makes S less column-sparse. To find a compromise solution, we chose the following set of parameters: the initial diagonal of Λ is 0.038; α = 2.5; β = 0.01. After convergence, we chose to keep 110 cells (implementing top 110 eigenvalues in L) in each of the two low-rank inhibitory populations with a total of 220 cells so that 99% of the variance in L was retained. We also used 220 interneurons in the SVD implementation to facilitate comparison between the models.
|
10.1371/journal.pcbi.1004520 | Established Microbial Colonies Can Survive Type VI Secretion Assault | Type VI secretion (T6S) is a cell-to-cell injection system that can be used as a microbial weapon. T6S kills vulnerable cells, and is present in close to 25% of sequenced Gram-negative bacteria. To examine the ecological role of T6S among bacteria, we competed self-immune T6S+ cells and T6S-sensitive cells in simulated range expansions. As killing takes place only at the interface between sensitive and T6S+ strains, while growth takes place everywhere, sufficiently large domains of sensitive cells can achieve net growth in the face of attack. Indeed T6S-sensitive cells can often outgrow their T6S+ competitors. We validated these findings through in vivo competition experiments between T6S+ Vibrio cholerae and T6S-sensitive Escherichia coli. We found that E. coli can survive and even dominate so long as they have an adequate opportunity to form microcolonies at the outset of the competition. Finally, in simulated competitions between two equivalent and mutually sensitive T6S+ strains, the more numerous strain has an advantage that increases with the T6S attack rate. We conclude that sufficiently large domains of T6S-sensitive individuals can survive attack and potentially outcompete self-immune T6S+ bacteria.
| Type VI secretion (T6S) is a cell-to-cell injection system that can be used as a microbial weapon. T6S kills vulnerable cells, and is present in a significant fraction of bacteria. Given the tactical advantage conferred by T6S, the system’s lack of universality suggests limits to its effectiveness relative to its costs. In our study, we use theory and experiments to identify the limits of T6S as a cell-to-cell weapon. We find that cell birth inside an existing colony can offset cell death due to T6S killing at the colony’s edge, helping sufficiently large (“established”) groups of sensitive cells to survive. T6S has been extensively studied because of its implications in both disease and inter-microbial competition. The present study is the first to identify the practical limits of T6S as a killing mechanism.
| Microbes employ a staggering range of extracellular tools to engineer their immediate environment [1–6]. Very often, that environment is defined by the multitude of other cells in close proximity. These neighbors pose both a threat and an opportunity, and represent an important target for manipulation [7–10].
The Type VI secretion system (T6SS) is a mechanism for direct cell-to-cell manipulation through the translocation of effector proteins. The T6SS consists of a helical sheath, surrounding an inner tube with associated effectors, and a baseplate attached to the bacterial cell wall (Fig 1a) [11, 12]. The T6SS is functionally close to the contractile phage tail, with which it shares evolutionary origins [13–17]. When triggered, the sheath contracts rapidly, pushing the effector through a specialized pore and into a neighboring cell [18–22].
Specialized T6SSs can directly damage both prokaryotic and eukaryotic target cells through the translocation of toxic proteins directly into the target cell. T6SSs are observed to cause death via numerous mechanisms in both bacteria and eukaryotes (Fig 1b; S1 Video) [13, 18, 23–28]. In fact, many species have developed multiple, specialized T6SSs [26]; for example, Burkholderia thailandensis has five separate T6SSs, which allow it to attack both prokaryotic and eukaryotic cells [29]. T6SSs are present in approximately 25% of the Gram-negative genomes studied by Boyer and colleagues [30]. Antibacterial T6SSs appear to be found with cognate immunity proteins in every case [26]. Given this tactical advantage, one might expect T6S to be even more widespread. The lack of universality of the T6SS suggests that there are limits to its utility relative to its costs.
To address the question of T6S’s utility, we focused on the case of cell-to-cell killing between bacteria. We explored this scenario through the use of individual-based models (IBMs; also called “agent-based models”). IBMs simulate the behavior of many, possibly different individuals each of which obeys rules that dictate the individual’s behavior as a function of its immediate environment. IBMs are a common tool in ecology, and have been widely used in the study of spatially explicit biological processes. Examples at the multicellular scale include the evolution of cancer, the spread of disease, and the dispersal of plants [31–38]; IBMs are also used to study dynamics at the subcellular scale [39]. More generally, IBMs have been used to address a wide range of questions concerning cooperation and conflict, of which T6S strategy can be viewed as an example [40–45].
In this study, we develop a series of IBMs. The first competes self-immune T6S+ and sensitive individuals in a range expansion, analogous to a surface colony (2D) or a biofilm (3D). We find that cell growth from the inside of a sufficiently large (or “established”) domain can offset cell death at the interface between a T6S-sensitive strain and a self-immune T6S attacker. Consequently, given a sufficiently large domain, T6S-sensitive strains can survive T6S attack. The sensitive strain does not require a growth advantage to survive; in fact, the sensitive strain can resist elimination even with a slower growth rate. Given even a small growth advantage, the T6S-sensitive strain can outcompete a self-immune T6S+ competitor. In a variant on the original model, we also find that moderate nutrient limitation has a negligible effect on competition outcomes.
We validated these findings through in vivo competition experiments between T6S+ Vibrio cholerae and T6S-sensitive Escherichia coli. In these 2D plate assays, E. coli can form microcolonies that survive, provided the initial local density of V. cholerae is not too high. Along similar lines, simulated competitions between mutually sensitive T6S+ strains (strains that are self-immune but sensitive to one another) reveal that the initially more numerous strain benefits most from higher attack rates. We conclude with a discussion of the ecological impact of T6SSs.
Escherichia coli MG1655 GentR (LacZ+) was competed against Vibrio cholerae str. 2740–80 (LacZ-), similarly to what was described previously [19]. E. coli and V. cholerae were each grown from frozen stocks in Luria-Bertani broth (LB), supplemented with the appropriate antibiotic, shaking overnight at 37°C and 200 rpm. The cells were washed twice with LB before being diluted to an OD600nm of 0.5. To confirm that the initial number of viable cells were comparable among the competition assays, the colony forming units (CFUs) were determined by serially diluting the washed and diluted V. cholerae and E. coli cultures 10-fold in 96-well plates in triplicate. Thereafter, 5 μL of each dilution were spotted on an LB agar plate (LA).
For the competition assays, the cultures were mixed in a 1:1 ratio, which was then serially diluted 3-fold in a 96-well plate. For selected dilutions 5 μL were spotted on a LA/IPTG 100 μM/X-Gal 40 μg/mL plate in duplicate. The competition plates were incubated at 37°C overnight. To determine the E. coli to V. cholerae ratios resulting from the competition assays, the CFUs of both strains were determined for each spot. This was achieved by excising the spots from the competition assay plates and resuspendig the bacteria in 1 mL LB by vigorously vortexing for at least 15 sec. These suspensions were serially diluted 10-fold in 96-well plates and 5 μL of each dilution were spotted on LA plates supplemented with the appropriate antibiotic. The CFU plates where either incubated at 37°C overnight or at lower temperatures until colonies were visible. Images of the plates were taken on a white light transilluminator. Timelapse movies of the competition assay were obtained by preparing the competition assay plates and the pre-competition CFU plates as described before, except that the competition mixtures were only spotted once. The competition assay plate was incubated at 37°C on a white light transilluminator while taking an image every 10 min over 24 h using a Nikon D5200. The contrast, brightness and white balance of the images were adjusted using Adobe Photoshop CS5. The same settings were applied to all timelapse images. Thereafter the images were further processed and converted to a video using Fiji [46].
The growth rate determination was carried out under the same conditions as the killing assay. The same cultures (OD600nm = 0.5) were individually spotted on LA plates and incubated at 37°C. Every hour the CFU was determined from a spot of each strain, as described for the endpoint killing assay. The growth rate was then derived from the parameters of the fit of an exponential curve. For the E. coli MG1655 GentR overnight cultures and selective CFU plates the growth medium was supplemented with 15 μg /mL Gentamicin, whereas for V. cholerae str. 2740–80 50 μg /mL Streptomycin was added.
Imaging of a competition between E. coli and V. cholerae VipA-msfGFP strains was performed under conditions similar to those used previously for imaging of T6SS activity in V. cholerae [17]. Strains were grown to OD600nm ≈ 1 and mixed at a 1:1 ratio on an LB 1% agarose pad. Imaging started after 10–20 min and was performed at 37°C for the indicated number of frames and at the indicated frame rate.
Computer models were implemented using Nanoverse 0.x, a prototype of our freely available individual-based modeling platform [47]. In Nanoverse, individual agents (e.g. cells) occupy spaces on a regular lattice. In every step of a simulation, one or more individuals perform a series of behaviors; if multiple individuals act simultaneously, the events are resolved in random order.
Two types of individual cells are included in the simulations (Fig 2a and 2b): self-immune T6S+ (“T6S+”) cells, shown in red, and sensitive T6S- (“sensitive”) cells, shown in blue. (Self-sensitive T6S+ strains “self-destruct” rapidly in simulations, and indeed have not been observed in nature.) Every cell has an associated probability of cell division per step of the simulation. The T6S+ division rate αt is taken as the (inverse) time unit of the system and is set equal to 1. The sensitive division rate αs is generally set higher than αt, as only T6S+ cells pay the cost of maintaing the T6S. Upon cell division, a copy of the dividing cell is placed in a vacant space adjacent to the dividing cell (Fig 2a). If no vacancies exist adjacent to the dividing cell, nearby cells are pushed out of the way to make room (S1 Text).
Each T6S+ cell has a fixed rate γ of initiating an attack (Fig 2b). The attack is then resolved according to an individual-based rule: attack exactly one randomly chosen nearest neighbor if there is one; otherwise do nothing. If the attack targets a sensitive cell or a cell of a different T6S+ strain, the target dies and its lattice site becomes unoccupied; T6S+ cells are immune to attack by cells of the same T6S+ strain, as observed experimentally [26]. The overall rate of events is controlled by the simulation timestep multiplier, λ (S1 Text).
To determine the effect of T6S on multi-species population dynamics, we simulated a competition between T6S+ and sensitive strains during a range expansion. The simulations begin with a well-mixed, fully occupied circular inoculum of approximately 500 individuals (S1 Text). For 2D simulations on a triangular lattice, the starting population is 469 individuals (i.e. inoculum radius r0 = 12).
The T6S+ division rate is chosen as the unit of time, αt = 1. The three other parameters are the sensitive strain growth rate αs, the initial sensitive strain fraction, and the attack rate γ. (In simulations in which there are no T6S+ cells, the unit of time is αs = 1.) The attack rate γ and the sensitive strain growth rate αs are found to offset one another as discussed below. The parameter space was extensively explored. Fig 2 shows parameters chosen to emphasize the effect of varying the attack rate γ and the initial sensitive strain fraction. Specifically, we fixed the sensitive strain growth rate as αs = 4 and varied γ and the sensitive fraction.
When the attack rate is low (γ = 5), sensitive cells can ultimately dominate even when the sensitive strain fraction starts as only a 10% minority (Fig 2c, S2 Video). Initially, the sensitive population declines as isolated individuals are attacked and killed. Eventually, only a small number of surviving sensitive domains remain, concentrated along the periphery of the colony. However, because sensitive cells grow faster than T6S+ cells, these domains begin to outgrow the T6S+ strain, eventually leading to a majority sensitive population. By contrast, at high attack rate (γ = 15) and an initial 10% sensitive strain fraction all sensitive individuals are rapidly eliminated (Fig 2d). When the initial sensitive strain fraction is increased to 50%, a larger number of sensitive cells begin near to one another, accelerating the formation of sensitive domains; the early formation of these domains helps the sensitive strain to survive and eventually dominate the T6S+ strain, even under a high rate of attack (Fig 2e and 2f).
An analysis of multiple, independent simulations (Fig 2g) shows that sensitive populations decline and then recover when both the attack rate and initial sensitive strain fraction are low (upper left), or when both are high (lower right). During the period of decline, isolated sensitive cells are eliminated while clusters of sensitive cells enjoy a degree of protection from attack. The monotonic increase of the sensitive population fraction in the most favorable conditions—high initial sensitive strain fraction, low attack rate (lower left)—results from the early formation of sensitive domains, whereas adverse conditions—low initial sensitive strain fraction, high attack rate (upper right)—preclude sensitive domain formation and lead to elimination of the sensitive strain.
Since T6S-mediated killing can take place only at the interface between T6S+ and sensitive strains, we hypothesized that the net growth rate of the sensitive strain depends on the difference between the area or volume of a sensitive domain and the extent of the interface between the strains. To identify the dependence of this relationship on attack rate and relative growth rates, we studied a simple sensitive domain model (Fig 3a and 3b). The 2D simulations begin with a fully-occupied, homogeneous circular sensitive inoculum. As in the competition model, all individuals are capable of cell division. As before, the model assumes that interior cells can push other cells toward the surface of the colony to make room for their daughter cells (S1 Text). To simulate attack, individuals at the outer periphery are subject to being killed at a rate γ ˜, essentially equivalent to embedding the sensitive domain in a larger T6S+ domain.
To explore the transition from sensitive strain collapse to sensitive strain growth observed in Fig 2, we varied the sensitive strain domain radius while holding constant the “attack” rate γ ˜ = 8, retaining the αs = 4 growth rate from the earlier competitions. Most sensitive strain domains with starting radius r0 ≤ 5 shrank toward zero, while larger domains survived (S3 Video). We then varied the sensitive strain growth rate, allowing it to fall below αs = 1. Strikingly, the minimum sensitive strain domain radius required for survival depends inversely on the relative sensitive strain growth rate, implying that a sufficiently large sensitive strain domain can resist displacement by even a faster-growing T6S+ attacker (Fig 3c).
We can readily estimate the critical population size n* above which a sensitive strain domain is expected to enjoy a net positive growth rate. Above this value, a sensitive domain would not shrink as a result of T6S+ competition, although it could, depending on conditions, represent an increasingly small fraction of total population. Eq 1 represents a theoretical “worst-case” scenario for a domain of sensitive cells, in which they are completely surrounded by an infinite domain of T6S+ cells. The key observation is that the rate of killing is proportional to the length of the interface between strains, while the rate of sensitive strain population growth is proportional to the sensitive population. For a population size n in 2D, the size of the interface is simply the circumference of the circle. Hence,
d n d t = α s n - 2 γ ˜ ( π n ) 12. (1)
Solving Eq 1 for n at dn/dt = 0, i.e. at the unstable fixed point between increasing and decreasing n, we find that
n * = 4 γ ˜ 2 π α s 2 , (2)
which is shown as a dotted line on Fig 3c. The slight divergence at high radius between the predicted and simulated values is the result of accumulated simulation error (S1 Text). The finding suggests that, even at this theoretical limit of maximal contact with T6S+ competitors, a sensitive domain can persist for long times.
Fig 3d shows simulation results for dn/dt plotted against the prediction from Eq 1. The rate of change of sensitive strain population was measured periodically in simulations with initial domain radii from r0 = 3 to r0 = 12. Attack rates ranged from γ ˜ = 0 to γ ˜ = 14; sensitive strain growth rates ranged from αs = 1 to αs = 4. The simulations show excellent agreement with the predicted dynamics (R2 > .98), despite deviations of the sensitive domain from a pure circle arising both from the lattice structure and from the stochasticity of the simulations. Similar results are obtained for a corresponding relationship in 1D and 3D (S2 Text).
The simulations described so far assume an unlimited supply of nutrients. To determine the effect of nutrient depletion on T6S population growth and competition, we developed a variant of the IBM that incorporates local depletion of nutrients. Even very limited nutrient concentrations still lead to exponential growth during range expansions, resulting in growth and competition dynamics that are nearly identical to those of the unlimited-nutrient case (S3 Text).
To validate our simulation results, we inoculated 2.5 μL each of of LacZ- T6S+ V. cholerae and LacZ+ T6S- E. coli onto X-Gal plates at various dilutions (see “Materials and Methods”). We compared the outcomes of these experiments with simulations for which the growth rates of sensitive and T6S+ cells were matched to those of E. coli and V. cholerae, respectively. In a preliminary estimate, E. coli was observed to grow slightly faster than V. cholerae (2.19 h−1 vs 2.05 h−1), so this difference was also used in the simulations. The simulation attack rate was set to γ = 5, which yielded a rough parallel with the experimental images. These simulations were run until the colony had doubled in radius.
Fig 4a–4d compare the experimental and simulated competitions, with initial inoculum concentrations decreasing 9-fold with each successive panel. As the inoculum becomes more dilute, single-species domains become larger. Simultaneously, E. coli become more numerous (Fig 4f; S4 Video). In a micrograph of the experimental competition, large domains of E. coli are observed to grow, while smaller domains undergo proportionately higher cell death (Fig 4e). S5 Video suggests that these E. coli domains persist stably after 24h. In the simulations, the final sensitive population is seen to increase as initial inoculum density decreases. This is due to the formation of large sensitive domains prior to initial T6S+ encounter, leading to increased sensitive strain survival.
Interestingly, in the low-resolution images, a darkened region is observed along the interspecies interfaces, but not at same-species microcolony interfaces. We infer that the darkened zones represent an accumulation of E. coli lysates, due to the continual renewal of the interspecies front by cell division within the bulk.
So far, we have considered competition between T6S+ and sensitive bacteria. We next investigated whether being T6S+ could help in the case of invasion by a T6S+ competitor. To answer this question, we simulated a competition between two T6S+ strains during a range expansion. Each strain can kill the other, but is immune to self-attack. Each strain has the same attack rate γ and cell division rate αt = 1. Fig 5 shows two T6S+ strains (yellow and red) that were allowed to compete during a range expansion from n0 = 469 (r0 = 12) to a final population of nf = 4690. The relative success of the invasion was measured by comparing the initial yellow (minority) fraction to the final yellow fraction.
In the presence of attack, the minority population is quickly eliminated (Fig 5a). By contrast, in the absence of attack the minority fraction remains roughly constant throughout the course of the range expansion (Fig 5b, S6 Video). As the attack rate increases, the initial minority fraction needed for survival asymptotically approaches 50% (Fig 5c). Note that for equal initial numbers of red and yellow cells, attack leads to spontaneous segregation from a well-mixed inoculum, with higher attack rates leading to faster and more thorough sectoring (S6 Video). Equivalent competitions in 1D and 3D led to analogous results (S4 and S5 Figs). These results imply that T6S+ is useful for defending established populations against invasion.
Gram-negative bacteria can employ T6S to kill competitors, yet the system is not found universally among these bacteria. To better understand the conditions favoring T6S, we modeled a competition between T6S+ and sensitive strains. In a range expansion from a well-mixed inoculum, we found that the sensitive cells can survive in the presence of T6S+ competitors by forming compact domains that protect interior cells from attack. To test these results, we competed T6S+ V. cholerae and T6S-, sensitive E. coli in an analogous range expansion. We observed that E. coli outcompeted V. cholerae, so long as the E. coli had the opportunity to form compact domains. Finally, we found that in a model competition between two equivalent T6S+ strains the more numerous strain always drove the minority to extinction.
It is informative to compare the current model to related model systems. For example, in a Lotka-Volterra model, a prey species grows in the absence of predation, and a predator grows faster in the presence of prey [48]; such systems have also been generalized to lattices [49]. By contrast, T6S+ does not grow faster as a result of killing, but potentially occupies more of the habitat. In this sense, the current model is more closely analogous to colicin dynamics in E. coli [50, 51]. Chao and Levin [52] observed that a colicin-producing strain of E. coli dominated a sensitive strain on soft agar by creating a zone of inhibition around itself, preventing the sensitive cells from exploiting the habitat. Colicin dynamics have also been studied using an IBM based on contact-mediated killing [53]. The colicin IBM differs from our T6S model in two respects: in [53], agents can only divide into adjacent vacancies, and sensitive cells have a strict growth advantage. The colicin model predicts that either species can dominate, with dominance depending primarily on parameter choices. By contrast, in the current study, initial colony size determines the survivorship of sensitive cells at all parameter values. The difference comes from the fact that in our model for T6S-mediated competition, interior sensitive cells are protected from killing by the outermost layer of cells. Such a “refuge” effect has previously been studied in the context of predator-prey dynamics, where density-driven sheltering is observed to destabilize predator-prey ratios relative to a well mixed model [54].
Our model employs a number of simplifying assumptions. Most importantly, cells are represented as agents on a regular lattice, and cells divide stochastically. While cell shape can affect the details of colony morphology during range expansions, it does not seem to affect the qualitative population dynamics [55]; indeed, lattice population dynamics have been shown to be consistent with the dynamics of real microbial populations [56]. The similarity of our observations in 1D, 2D, and 3D further suggests that our results are not sensitive to cellular geometry. Similarly only overall growth rates, rather than the detailed timing of cell divisions, are important for long-term population dynamics [55].
It has been hypothesized that nutrient depletion may introduce a substantial advantage for T6S+ strains. In practice, cells at the interior of a natural community face nutrient and oxygen depletion [57]. Does this limitation result in a different competitive outcome? In a simple model of nutrient depletion, we found that a moderately nutrient-limited environment leads to dynamics extremely similar to those in the absence of limitation (S3 Text). This is because exponential growth ensures that only a very small fraction of the population occupies a fully depleted zone (S7 Fig). Thus, our preliminary results suggest that the effects of nutrient depletion on cell growth do not qualitatively alter the population dynamics arising from T6S-mediated competitions. Under special circumstances, such as burrowing invasions of a nutrient-depleted biofilm, T6S-mediated cell lysis could provide a significant nutrient benefit beyond the direct benefit of killing competitor cells. Typically, this effect would be limited, as the nutrient benefit would be divided among both T6S+ species and their prey. In an entirely nutrient-depleted environment, though, actively growing invaders would have an early growth advantage over previously quiescent resident cells.
In determining the ecological role of T6S, the costs of maintaining a T6SS must be taken into consideration. The T6SS requires the expression of 13 core genes, the assembly and disassembly of the baseplate structure and sheath, and the production of the secreted effectors [19, 27, 30]. Immunity to T6S requires the maintenance of a complementary immunity protein, and may require additional modifications to the attacker’s peptidoglycan [26]. Selective use of T6S can mitigate these costs by reducing the frequency of wasteful attacks. To this end, bacteria have evolved a variety of T6SS regulatory schemes, including quorum-sensing and retaliation.
Quorum sensing can reduce wasteful attacks by repressing T6S until it is likely to provide a benefit [21, 58]. For example, QS regulates expression of T6SS in V. cholerae [59]. Interestingly, expression of T6SS and natural competence is induced by high cell density and growth on chitinous surfaces, which suggests a role of T6SS in horizontal gene transfer [60]. In addition, the V. cholerae QS signal integrates both species-specific and multigeneric signals [61], which means that the presence of competitors could also activate V. cholerae’s T6SS. However, reflecting the diversity of T6S roles, T6S is not always upregulated in response to high density. In P. aeruginosa, there are three T6SSs; species-specific QS signals LasR and MvfR activate two of these T6SSs, but repress the third [62].
Like quorum sensing, “retaliatory” T6S attack can prevent attack until a hostile cell is encountered. For example, P. aeruginosa is observed to engage in retaliatory T6S attack [27, 63, 64]. This ‘tit-for-tat’ strategy could limit wasteful T6S+ interactions within clonal populations, as well as facilitating coexistence within productive consortia. Notably, P. aeruginosa also attacks its target repeatedly; by eliminating wasteful attacks, retaliators are also free to employ a more concerted (and damaging) series of attacks [27].
In considering the ecological role of T6S, it is instructive to consider an analogous system found in marine invertebrates. Members of the phylum Cnidaria, which includes corals, hydrae, and jellyfish, possess an explosive cell called a nematoycte containing a harpoon-like projectile [65]. Upon detonation, the effector is propelled with extreme force (up to 40,000g) into a target, leading to paralysis and death [66]. Among corals, nematocytes are used interspecifically to compete for habitat access. High attack rates are most commonly observed among slower-growing species, where nematocytes are used to defend against encroachment [67]. Our results suggest that, like nematocytes, T6S can also offset a growth rate disadvantage. The full breadth of its ecological role, however, is only beginning to come into focus.
|
10.1371/journal.pgen.0030121 | Plasticity of Fission Yeast CENP-A Chromatin Driven by Relative Levels of Histone H3 and H4 | The histone H3 variant CENP-A assembles into chromatin exclusively at centromeres. The process of CENP-A chromatin assembly is epigenetically regulated. Fission yeast centromeres are composed of a central kinetochore domain on which CENP-A chromatin is assembled, and this is flanked by heterochromatin. Marker genes are silenced when placed within kinetochore or heterochromatin domains. It is not known if fission yeast CENP-ACnp1 chromatin is confined to specific sequences or whether histone H3 is actively excluded. Here, we show that fission yeast CENP-ACnp1 can assemble on noncentromeric DNA when it is inserted within the central kinetochore domain, suggesting that in fission yeast CENP-ACnp1 chromatin assembly is driven by the context of a sequence rather than the underlying DNA sequence itself. Silencing in the central domain is correlated with the amount of CENP-ACnp1 associated with the marker gene and is also affected by the relative level of histone H3. Our analyses indicate that kinetochore integrity is dependent on maintaining the normal ratio of H3 and H4. Excess H3 competes with CENP-ACnp1 for assembly into central domain chromatin, resulting in less CENP-ACnp1 and other kinetochore proteins at centromeres causing defective kinetochore function, which is manifest as aberrant mitotic chromosome segregation. Alterations in the levels of H3 relative to H4 and CENP-ACnp1 influence the extent of DNA at centromeres that is packaged in CENP-ACnp1 chromatin and the composition of this chromatin. Thus, CENP-ACnp1 chromatin assembly in fission yeast exhibits plasticity with respect to the underlying sequences and is sensitive to the levels of CENP-ACnp1 and other core histones.
| The DNA of all genomes is organized into chromosomes that are packaged in chromatin in which DNA is wrapped around nucleosomes composed of the histones H2A, H2B, H3, and H4. Centromeres are the specialized regions on chromosomes that attach them to spindle microtubules, and this process is required to allow each daughter cell to receive one copy of each chromosome after they have duplicated. Centromere regions are distinguished from other parts of the chromosome by the incorporation of the histone H3 variant CENP-A instead of histone H3 into specialized nucleosomes. This CENP-A chromatin allows the machinery (the kinetochore) responsible for attaching chromosomes to microtubules to assemble. Fission yeast centromeres contain a central domain where CENP-A is prevalent. This study shows that CENP-A can associate with “foreign” DNA placed in the central domain. Therefore, CENP-A appears to associate with any DNA placed in this environment, independent of its DNA sequence. Increasing the relative level of H3 allows H3 to be assembled in place of CENP-A in this critical central domain and results in defective centromere/kinetochore function and chromosome segregation. This study highlights the plasticity of centromeric chromatin.
| In most eukaryotes, chromosomes contain a centromere that occupies a single locus. The centromere acts as the site for assembly of the kinetochore that mediates the attachment of chromosomes to spindle microtubules and orchestrates their equational segregation to daughter nuclei at mitosis. In many organisms, long tandem arrays of repetitive satellite DNA, such as alpha-satellite DNA in humans, are found at each centromere [1,2].
Chromosomal DNA is packaged in chromatin composed of nucleosomes containing the four core histones H2A, H2B, H3, and H4. Histone variants can play specific roles in the regulation of gene expression. For example, H3.3 replaces H3 in regions of active transcription [3–5]. Intriguingly, kinetochores contain a specific form of chromatin in which canonical histone H3 is replaced by the centromere-specific histone H3 variant known generally as CENP-A [1,2,6,7].
CENP-A is essential for the assembly of a functional kinetochore and as such must represent a key component in establishing and/or maintaining the site of kinetochore assembly at the centromere [8–12]. Although CENP-A proteins are found at centromeres in all organisms, there appears to be no specific conserved sequence that ensures the assembly of CENP-A chromatin [1,2,7,13]. Indeed, the deposition of CENP-A appears to be malleable since inactivated human centromeres lack CENP-A, even though they retain alpha-satellite repeats [14]. In addition, neocentromeres occasionally arise on chromosomal DNA that lacks any similarity to alpha-satellite repeats, and CENP-A can associate with noncentromeric sequences included in human artificial chromosomes [14–19]. Similarly, in Drosophila, noncentromeric DNA can acquire the ability to assemble and propagate kinetochore proteins [20–22]. These and other observations suggest that the site of CENP-A chromatin assembly is epigenetically regulated and propagated [1,2,13].
Fission yeast centromeres are reminiscent of those of metazoa in that they contain repetitive elements (outer repeats, otr) that flank the central domain (see Figure 1). The central domain is composed of inner repeats (imr) surrounding the central core (cnt) [23–25]. Noncoding transcripts arising from the outer repeat provide a substrate for RNA interference (RNAi) that directs methylation of histone H3 on lysine 9 and the assembly of silent chromatin (reviewed in [26]). Within the central domain most histone H3 is replaced by the centromere-specific H3 variant CENP-ACnp1 to form the unusual chromatin that occupies most of the 10–12 kb comprising imr and cnt [11,23,27–29]. Consistent with the notion that CENP-ACnp1 chromatin is a signature of kinetochore activity, kinetochore-specific proteins are confined to the central domain [11,29–35]. At each centromere a cluster of tRNA genes demarcates the two distinct chromatin domains: outer repeat silent heterochromatin and kinetochore chromatin [23,36]. Deletion of one of the two tRNA genes from one side of a centromere allows heterochromatin to infiltrate the central domain [37]. The fact that all three fission yeast centromeres have a common organization of DNA elements suggests that the assembly of heterochromatin and CENP-ACnp1 chromatin domains could be strictly governed by sequence.
Marker genes placed within outer repeat chromatin are silenced, and this requires RNAi components, Clr4 histone H3K9 methyltransferase and Swi6 (homologue of HP1) [36,38]. Genes are also silenced in the central domain, but this silencing is strongly dependent on kinetochore integrity rather than RNAi-mediated heterochromatin [31,35,36,39]. Mutations in several kinetochore proteins, including CENP-ACnp1, alleviate silencing, specifically in the central domain, and affect the association of CENP-ACnp1 with the central domain and/or the unusual chromatin found within the central domain [11,29,35,36].
Dissection of fission yeast centromeric DNA showed that outer repeat and central domain sequences are required for the assembly of a kinetochore that supports mitotic segregation [23,24,40]. In Drosophila and humans, centromeres can arise at sites apparently lacking particular features at the primary DNA sequence level. It is not known whether the similarity between fission yeast and metazoan centromeres extends to the ability of CENP-A chromatin to assemble on noncentromeric sequences or if CENP-A assembly in this organism more closely resembles the situation in budding yeast in which CENP-A associates at a specific sequence.
Here, we investigate the ability of fission yeast CENP-ACnp1 chromatin to assemble on noncentromeric sequences. We determine whether the amount of CENP-ACnp1 incorporation directly correlates with silencing, kinetochore assembly, and kinetochore function. Altering the relative ratios between H3, H4, and CENP-ACnp1 influences the assembly of CENP-ACnp1 chromatin, the recruitment of other kinetochore proteins, and the fidelity of chromosome segregation. Surprisingly, there is no impediment to depositing H3 in the central domain. Thus, our observations indicate that the relative levels of histones are crucial for the correct formation of CENP-ACnp1 chromatin.
The central domain of centromere 1 (cen1) is composed of the cnt1 element, a portion of which is shared with cen3, and the cen1-specific imr1 repeats that are virtually identical in sequence on the left and right sides [23]. The outer repeats flanking the central domain are also found at cen2 and cen3, totaling 17–18 copies at centromeres [23,25].
The distribution of endogenous CENP-ACnp1 across cen1 was assessed using anti-CENP-ACnp1 antiserum for chromatin immunoprecipitation (ChIP), confirming that endogenous CENP-ACnp1 associates with the central domain (cnt1 and imr1), but not with flanking outer repeats (Figure S1). To determine whether CENP-ACnp1 chromatin forms on noncentromeric sequences in fission yeast, its association with a ura4+ gene inserted at different sites in cen1 was assessed using ChIP (Figure 1A). The use of ura4+ insertions also provides specificity in ChIP for cen1, especially for duplicated cnt1 and imr1 regions and the more repetitive outer repeats. The ura4-DS/E allele (279-bp deletion) at its euchromatic locus provides an internal control for quantification. CENP-ACnp1 shows no association with ura4+ in the outer repeats of cen1 (sites 1 and 2). No association of CENP-ACnp1 is seen at a euchromatic control site R.int-cnt1:ura4+ (R: between open reading frames SPBC342.01 and .02), even though in this case the ura4+ gene is also flanked by 1.7 and 1.6 kb of DNA from the central domain of cen1 (Figure S9). However, just outside the tRNAala and tRNAglu genes that demarcate the edge of the central domain, a 4-fold enrichment of CENP-ACnp1 is detected (site 3). A similar level of CENP-ACnp1 is detected inside the first tRNAala gene (site 4), but much higher levels (20- to 50-fold enrichment) of CENP-ACnp1 are associated with ura4+ in the core of the central domain of cen1 (site 5 and 6). This confirms and extends previous analyses utilizing C-terminally 3 × HA tagged CENP-A [11] and is similar to the pattern of association seen for Mis6-HA across cen1 [36].
These observations suggest that fission yeast CENP-ACnp1 can associate with noncentromeric DNA inserted within the central domain. However, it is possible that the strong enrichment of ura4+ in ChIPs is simply due to the association of CENP-ACnp1 with adjacent centromeric DNA sequences in the IPs. To test this rigorously we used two strains in which 1.7 kb or 4.7 kb of DNA bearing the ura4+ gene was inserted at site 6 in the middle of the central domain of cen1, cnt1:ura4+, and cnt1:bigura4+, respectively (Figure 1B). The sonicated chromatin used for anti-CENP-ACnp1 ChIP was less than 800 bp (Figure 1C). The region within ura4+ monitored by PCR is 0.9 kb and 0.5 kb (cnt1:ura4+) or 2.2 kb and 2.2 kb (cnt1:bigura4+) away from centromeric sequences, and thus, any enrichment of ura4+ indicates association of CENP-ACnp1 with ura4+ DNA. Using this stringent assay, we observe that CENP-ACnp1 assembles on ura4+ DNA in cnt1:ura4+ (19–35× enrichment) and even associates with the middle of cnt1:bigura4+ which is at least 2 kb from any endogenous centromeric sequences (6–12× enrichment). Thus, we conclude that CENP-ACnp1 can assemble on noncentromeric sequences.
The R.int-cnt1:ura4+ control (R) consists of the ura4+ gene flanked by 1.7 and 1.6 kb of central domain DNA [39], yet no association of CENP-ACnp1 is seen at this location. Thus, central domain sequences alone are not sufficient to induce CENP-ACnp1 assembly when placed on a chromosome arm. This is consistent with previous analyses showing that plasmids containing just central domain DNA do not exhibit centromere function or features associated with functional centromeres [23,24,27]. In addition to ura4+, we have also detected CENP-ACnp1 on ade6+ when inserted in the central domain of centromeres (Figure S1). Thus, the deposition of CENP-ACnp1 in Schizosaccharomyces pombe does not require specific underlying DNA sequences and can probably assemble on any noncentromeric DNA sequence provided that it is placed in the context of a functional centromere.
The ChIP analyses above indicate that there is more CENP-ACnp1 on cnt1:ura4+ than on cnt1:bigura4+. Both ura4+ insertions at site 6 are silenced, as indicated by growth on counter-selective 5-fluoro-orotic-acid (FOA) plates. However, the cnt1:ura4+ insertion exhibits more growth on FOA plates indicating that it is more strongly silenced than the larger cnt1:bigura4+ insertion (Figure 1D). In support of this, more ura4+ mRNA is detected in cnt1:bigura4+ cells compared to cnt1:ura4+ cells by reverse transcriptase (RT)-PCR (Figure 1E).
Consistent with the idea that silencing is due to spreading of CENP-ACnp1 chromatin onto ura4+ gene insertions, cnp1 mutants alleviated silencing of ura4+ within the central domain (less growth on FOA; Figure 2A) and less CENP-ACnp1 was detectable by ChIP on the central domain (Figure 2B); this is supported by previous analyses [11,35,41]. The most severe cnp1 alleles showed the greatest reductions in CENP-ACnp1 levels: cnp1–1(L87Q) [11] > cnp1–76 (T74M) > cnp1–87 (E92K) > cnp1–169 (V52A) (previously referred to as sim2 mutants; [35]). This suggests that the production of defective CENP-ACnp1 results in less CENP-ACnp1 chromatin in the central domain, and that this chromatin state is more compatible with ura4+ gene expression. ChIP with antibodies recognizing the C-terminus of histone H3 indicates that this is accompanied by increased incorporation of histone H3 into the central domain chromatin in all cnp1 mutants (Figure 2B), and this concurs with analyses of other mutants affecting CENP-ACnp1 deposition [41,42]. Silencing also correlates with chromatin composition in the centromere insertions: more H3 is associated with cnt1:bigura4+ than with the more silent cnt1:ura4+ (Figure 2C). Interestingly, in Drosophila, H3 can take the place of CENP-ACID when CENP-ACID levels are reduced [43].
In addition, cnp1 mutants are sensitive to excess H3, displaying a correlation between allele severity and sensitivity to different H3 levels (Figures 2D and S2). Moreover, overexpression of histone H4 suppressed the temperature sensitivity of cnp1 mutants, with higher levels of H4 being required to suppress more severe alleles (Figures 2E and S2), as reported for cnp1–1 [42]. These genetic interactions suggest that additional H4 may assist in incorporation of mutant CENP-ACnp1 into the central core by facilitating the formation of more H4/CENP-ACnp1 heteromers. On the other hand, excess H3 would exacerbate the phenotype by competing with CENP-ACnp1 for H4 from the available pool, further favoring H3/H4 deposition into central domain chromatin.
Together, these data imply that assembly of CENP-ACnp1 chromatin on DNA inserted in the central domain is incompatible with gene expression and that when CENP-ACnp1 is defective, histone H3 takes its place, allowing greater expression.
Since defective CENP-ACnp1 alleviates silencing, we determined whether, conversely, overexpression of CENP-ACnp1 causes an increase in central domain silencing. Unlike budding yeast, fission yeast are able to tolerate overexpression of CENP-ACnp1, to an extent that the normally undetectable CENP-ACnp1 is detectable by western analysis [44] (Figure S3). Overexpression of CENP-ACnp1 (from the attenuated nmt1 promoter in prep81x) in a wild-type strain bearing cnt1:ura4+ or cnt1:bigura4+ increased the level of silencing, as indicated by increased growth on FOA (Figure 3A), and more CENP-ACnp1 was detected on cnt1:ura4+ and cnt1:bigura4+ (Figure 3B). This suggests that the ura4+ sequence inserted in the central domain is not saturated for CENP-ACnp1. There was also a detectable increase in the levels of CENP-ACnp1 on endogenous central domain sequences (Figure 3B).
Most CENP-ACnp1 is normally confined to the central domain, which is flanked on both sides by tRNAala and tRNAglu genes, separated by 349 bp [37]. When CENP-ACnp1 was overexpressed, increased silencing (more growth on FOA) occurred at site 4 (ura4+ between the tRNAala and tRNAglu genes; Figure 3C), and a subtle increase in CENP-ACnp1 levels was reproducibly detected on ura4+ at this site. However, in strains overexpressing CENP-ACnp1, additional CENP-ACnp1 could not be detected in the heterochromatin domain beyond the tRNAala and tRNAglu genes (unpublished data). Moreover, there was no apparent effect on the extent of the CENP-ACnp1 domain in a strain lacking heterochromatin on the outer repeats (clr4Δ; unpublished data). These data suggest that CENP-ACnp1 chromatin is confined to the central domain but that overexpression of CENP-ACnp1 allows some expansion in the extent of the CENP-ACnp1 domain beyond the proximal tRNAala gene into an inserted ura4+ gene.
As the degree of central domain silencing correlates with increased H3 levels in both cnp1 mutants and cnt1:bigura4+ versus cnt1:ura4+, we investigated the effects of simply overexpressing H3 on central domain silencing and composition. Additional histone H3 was expressed from the nmt1 promoter (prep3X-H3) in strains harboring cnt1:ura4+ or cnt1:bigura4+ (Figure 4A). Quantitative western analyses indicate that 2- to 3-fold more H3 is present and that the levels of CENP-ACnp1 are unaffected (Figures S4 and S5). H3 overexpression resulted in increased ura4+ expression from cnt1:ura4+ as shown by loss of growth on FOA. This was accompanied by increased H3 on cnt1:ura4+ and cnt1:bigura4+ (Figure 4B) and a concomitant decrease in CENP-ACnp1 levels (Figure 4C).
The nmt1 promoter used in the above experiments is active throughout the cell cycle whereas histone gene expression is normally induced from endogenous promoters in early S phase [11,45]. To rule out the possibility that the observed effects were due to constitutive expression of H3 from the nmt1 promoter, we altered the ratio of H3:H4 genes while retaining the normal endogenous promoter elements. In fission yeast, three pairs of H3 and H4 genes are transcribed divergently from a putative common regulator at three distinct loci (H3.1/H4.1, H3.2/H4.2, and H3.3/H4.3) [46]. We completely deleted the H3.1/H4.1 pair and either the H3.2 gene, resulting in a H3:H4 gene ratio of 1:2 (H4 > H3), or the H4.2 gene, resulting in a H3:H4 gene ratio of 2:1 (H3 > H4) [47] (Figure S6). Increased H3 relative to H4 (H3 > H4) alleviated silencing of both cnt1:ura4+ and cnt1:bigura4+, resulting in failure to grow on counter-selective FOA plates (Figure 5A), indicating a complete inability to silence ura4+ expression. Conversely, excess H4 (H4 > H3) enhanced silencing of cnt1:ura4+ as indicated by reduced growth on plates lacking uracil. These phenotypic effects were confirmed by RT-PCR analysis (Figure 5B); more ura4 mRNA is produced from cnt1:ura4+ in the H3 > H4 strain (although we were unable to detect a further increase in the comparatively high levels of ura4 mRNA in cells containing cnt1:bigura4+). These analyses also indicate that central domain silencing is fully intact in the control strain (H3 = H4; H3:H4 ratio 2:2) used in these experiments (Figure 5A and 5B; compare to wild type; H3:H4 ratio 3:3). The strain also displays robust ChIP of CENP-ACnp1 on endogenous centromeric sequences and ura4+ sequences inserted therein (e.g., Figure 5D). Thus (as with cnp1 mutants), disturbing the H3:H4 balance so that more H3 is expressed alleviates silencing in the central domain, while additional H4 enhances silencing.
To determine the impact of additional H3 on centromere integrity, we assessed CENP-ACnp1 association with centromeres by ChIP and immunolocalization. Cells with a histone imbalance in favor of H3 (H3 > H4) were confirmed to have increased H3 levels on cnt1:ura4+ and cnt1:bigura4+ by ChIP (Figure 5C). CENP-ACnp1 levels were greatly reduced on cnt1:ura4+ and cnt1:bigura4+ in cells with excess H3 (Figure 5D), yet the cellular levels of CENP-ACnp1 are unaffected (Figure S5). Lower levels of CENP-ACnp1 are normally detected on the ura4+ region of cnt1:bigura4+ relative to cnt1:ura4+, and excess histone H3 caused a further decrease in the level of CENP-ACnp1 associated with the middle of the cnt1:ura4+ and a consistent but less dramatic effect on cnt1:bigura4+ (Figure 5D). We also observed a reduction of CENP-ACnp1 levels on endogenous centromeric sequences in strains with excess H3 (Figure 5D). In strains with an excess of H4 relative to H3 (H4 > H3), more CENP-ACnp1 was detected on cnt1:bigura4+, although the effects on endogenous centromeric sequences and cnt1:ura4+ were more variable (Figure 5D and unpublished data).
Chromatin immunoprecipitation of a protein reports the population average for its association with particular DNA sequences. To assess CENP-ACnp1 levels in individual cells, immunolocalization was performed. In fission yeast, the three centromeres cluster adjacent to the spindle pole body (SPB) in interphase and this was used as a marker for centromere position. In both H3 = H4 and H4 > H3 strains a clear CENP-ACnp1 signal (red) was detected next to the SPB (green) in all cells (Figure 6A). However, in H3 > H4 cells, the CENP-ACnp1 signal was weaker or undetectable and displayed variation between cells. The majority of cells had much weaker staining than H3 = H4 (54% versus 7%), and very few displayed bright/very bright staining (9% versus 93% in H3 = H4). In addition, the CENP-ACnp1 signal in H4 > H3 cells was quantified and found to be of greater intensity than in H3 = H4 cells (2.5-fold brighter on average in the H4 > H3 cells compared to the H3 = H4 cells; details of quantification methods described in Figure S7 and Materials and Methods). Thus, the immunolocalization data confirm the ChIP analyses and indicate that the level of CENP-ACnp1 at centromeres in H3 > H4 cells exhibits variation within the population.
These observations are consistent with a scenario in which excess CENP-ACnp1 allows more CENP-ACnp1/H4 chromatin assembly in the central domain, while elevated H3 levels permit more H3/H4 nucleosomes to occupy the central domain at the expense of CENP-ACnp1. It also appears that an increase in the available H4 pool allows CENP-ACnp1 to compete more effectively with H3 for incorporation in the centromere.
Altering the ratios of H3:H4:CENP-ACnp1 have clear effects on central domain chromatin, but what are the consequences for chromosome segregation? It is possible that a functional kinetochore can be assembled despite alterations in the proportion of CENP-ACnp1/H3 in the central domain. To determine whether chromosome segregation is aberrant in strains with altered histone ratios, fixed cells were stained for chromosomal DNA (DAPI: red) and anti-α-tubulin to identify cells with a mitotic spindle (green) (Figure 6B). The H3 > H4 strain exhibited significantly higher rates of chromosome segregation defects in anaphase (lagging chromosomes and uneven segregation) compared to the H4 > H3 and control (H3 = H4) strains (16.6%, 0%, and 1.4%, respectively, see Figure S8). Together these data suggest that excess H3 interferes with the localization of CENP-ACnp1 at centromeres resulting in defective kinetochore function during mitosis.
It could be argued that disturbing histone ratios in the cell is likely to have pleiotropic effects that lead to chromosome missegregation. For instance, mutants that affect silencing and function of centromeric outer repeat chromatin display high frequencies of lagging chromosomes. However, alleviation of outer repeat silencing was not observed when H3 was overexpressed or in H3 > H4 strains (unpublished data). In addition, prep81x-Cnp1 rescued defective growth of the H3 > H4 strain, allowing growth at 32 °C (Figure 5E), suggesting that the primary defect in H3 > H4 cells is in central domain function. As expected, the defective growth of the H3 > H4 strain was also rescued by H4 overexpression (Figure 5E).
Mutations in cnp1 or genes encoding kinetochore proteins alleviate silencing in the central domain [31,35,36]. Previous analyses suggest that the kinetochore proteins Mis6 and Sim4 can associate with cnt1:ura4+ [35,36]. To examine this further, ChIP was used to determine if three kinetochore proteins, CENP-CCnp3-GFP, Mal2-GFP, and Sim4-GFP, are also associated with the middle of cnt1:ura4+ and cnt1:bigura4+. Under conditions where chromatin was extensively sheared, all three kinetochore proteins were found to associate with the middle of the ura4+ gene when it was inserted within cnt1 at site 6 (both cnt1:ura4+ and to a lesser degree for cnt1:bigura4+), but not when it was flanked by 1.7 and 1.6 kb of central domain DNA at a euchromatic site (R: R.int-cnt1:ura4+) (Figure 7A).
Thus, not only CENP-ACnp1, but three other kinetochore proteins tested associate with noncentromeric DNA when inserted in the context of a functional fission yeast centromere.
Elevated levels of CENP-ACnp1 or histone H3 have opposite effects, respectively leading to more and less deposition of CENP-ACnp1 on cnt1:ura4+ and cnt1:bigura4+. Since CENP-ACnp1 is required for kinetochore assembly, we tested whether the levels of Mal2-GFP, Sim4-GFP, and CENP-CCnp3-GFP on cnt1:ura4+ and cnt1:bigura4+ are also affected in strains expressing additional CENP-ACnp1 or histone H3 (Figure 7B). Cells expressing additional CENP-ACnp1 reproducibly showed more CENP-CCnp3-GFP, Mal2-GFP, and Sim4-GFP associated with ura4+ in the central core, especially for cnt1:bigura4+. Conversely, in cells expressing more histone H3, less of these three kinetochore proteins were detected on cnt1:ura4+. This indicates that adjusting the levels of CENP-ACnp1 and H3 not only alters the extent and density of CENP-ACnp1 chromatin at centromeres but also influences the recruitment of other kinetochore proteins, resulting in defective chromosome segregation.
We have found that CENP-ACnp1 in fission yeast can associate with noncentromeric sequences provided that they are placed in an environment with the required contextual cues for CENP-ACnp1 chromatin assembly. Our analyses show that excess CENP-ACnp1 (or H4) allows the deposition of more CENP-ACnp1 at the expense of H3 while an excess of H3 allows the deposition of more H3 in place of CENP-ACnp1 and leads to defective chromosome segregation. These analyses strongly support the concept that deposition of CENP-ACnp1 in the central kinetochore domain exhibits surprising plasticity in that it can be disturbed or enforced simply by changes in the ratios of H3, H4, and CENP-ACnp1. These observations indicate that maintenance of the unique chromatin composition of the central domain is vital in ensuring proper kinetochore assembly and function. Thus, both the functional state of CENP-ACnp1 and its density are important for centromere function.
In metazoa, the assembly of CENP-A chromatin and kinetochores is plastic [1,2,13]. In humans, CENP-A and kinetochores are normally associated with a subset of the centromeric alpha-satellite repeats. However, CENP-A can assemble at locations on chromosome arms lacking alpha-satellite DNA, resulting in the formation of neocentromeres [14]. In Drosophila, experiments with truncated minichromosomes derived from the X chromosome demonstrate that CENP-ACID and kinetochore proteins can spread into chromosomal regions where they do not normally associate and then act as a functional neocentromere when the X centromere is subsequently deleted [20]. In addition, overexpression of CENP-ACID in Drosophila cells can attract other kinetochore proteins and even direct microtubule association at noncentromeric locations [22].
The similarity in the organization of centromeric DNA at all fission yeast centromeres suggested that, like Saccharomyces cerevisiae, the assembly of CENP-ACnp1 chromatin might be more DNA sequence–dependent and thus less pliable than in metazoa [23,25]. The identification of Ams2, a GATA-like DNA-binding factor that affects CENP-ACnp1 deposition, supported this possibility [42]. However, here we have shown that, as observed in Drosophila and human cells, fission yeast CENP-ACnp1 can spread into and associate with additional sequences inserted within the central domain. Thus, provided the right contextual cues exist, fission yeast CENP-ACnp1 chromatin can potentially associate with any DNA sequence. The establishment and continued loading of CENP-A in the central domain may be dependent on the central domain sequences themselves, but once it has been initiated, our analyses indicate that it can engulf inserted noncentromeric DNA. In addition, the association of CENP-ACnp1 with exogenous sequences in the central domain is accompanied by the recruitment of other known kinetochore proteins. It is also possible that the association of CENP-A with ura4 DNA in the central domain might reflect some higher order structure within the centromere.
In S. cerevisiae, additional CENP-ACse4 is degraded and only nondegradable mutant protein can be overexpressed [44]. In fission yeast there appears to be no inherent difficulty in overexpressing CENP-ACnp1 since increased CENP-ACnp1 is detected by western analysis upon overexpression (Figure S3); and this is the cause of increased CENP-ACnp1 incorporation into the central domain. Conversely, elevated levels of H3 lead to more H3 and less CENP-ACnp1 in central domain chromatin. In addition, mutations in CENP-ACnp1 are antagonized by H3 but rescued by H4 overexpression, and defects exhibited by H3 > H4 cells are suppressed by additional CENP-ACnp1. This indicates that perturbation in the relative levels of CENP-ACnp1, H3, and H4 can adversely affect the composition of CENP-ACnp1 chromatin assembled at centromeres, affecting the relative density of CENP-ACnp1/H4 to H3/H4 nucleosomes. Disturbing the normal CENP-A:H3:H4 ratio clearly results in defective kinetochore function and chromosome segregation. Consistent with this, decreased CENP-ACnp1 at centromeres in cells lacking Ams2 is suppressed or antagonized by overexpression of H4 and H3, respectively [41,42].
In fission yeast, it is likely that H3 and H4 mRNAs are coordinately expressed early in S phase from a common regulatory element residing between their divergent promoters, whereas CENP-ACnp1 mRNA is expressed from late M, peaking prior to S phase [11,41,45,46]. This suggests that normally CENP-ACnp1 is available before maximal H3 and H4 expression. As fission yeast centromeres are replicated early in S phase [48,49], this difference in expression timing may enhance the effective concentration of CENP-ACnp1 relative to H3 at the start of S phase, and thus allow it to compete more effectively with the initially low levels of newly produced H3 for assembly with new H4 (Figure 8). Our data suggest that altering the ratio of H3:H4 either by altering the normal gene number from equivalency (3:3 or 2:2) to 2:1, or overexpressing H3, provides more H3 to titrate out the pool of H4 available for assembly with CENP-ACnp1 into chromatin. Conversely, excess H4 relative to H3 (H3:H4 1:2) increases the pool of H4 available for assembly into chromatin with CENP-ACnp1, leading to more CENP-ACnp1 at centromeres. These data suggest that the relative amounts of H3, H4, and CENP-A are normally delicately balanced in the cell to allow normal CENP-ACnp1 chromatin and kinetochore assembly. Changes in this balance, the timing, or amount of expression can lead to increased or decreased CENP-ACnp1 chromatin at centromeres (Figure 8). Precise regulation of histone levels, e.g., coregulation of H3 and H4, is likely to be important in other organisms (including human) where CENP-A overexpression has been detected in the tissue from colorectal tumors [50]. In S. cerevisiae, elevated levels of histone H3 and H4 cause defective chromosome segregation [51]; it is possible that this is due to defects in deposition of CENP-ACse4 at centromeres.
The extent of sequence occupied by CENP-ACnp1 chromatin appears to be flexible and increases in response to higher levels of CENP-ACnp1. The tRNA genes flanking the central domain act as a barrier preventing heterochromatin seeping into the kinetochore domain [37]. Here, we detect more CENP-ACnp1 on ura4+ between the tRNAala and tRNAglu when CENP-ACnp1 is overexpressed but not distal to tRNAala in the heterochromatin domain (Figure 3C); thus, CENP-ACnp1 chromatin is still mainly confined to the central domain.
The mechanism of CENP-A assembly into centromeric chromatin is not known. In mammalian cells, CENP-A is synthesized in G2 and thereby separated from bulk histone synthesis [52,53]. Such analyses suggested that CENP-A chromatin assembly is uncoupled from replication. One model suggests that the CENP-A nucleosomes are randomly segregated at S phase and are subsequently recognized by the CENP-A chromatin assembly machinery resulting in the deposition of neighboring new CENP-A during interphase [2,52,53]. In fission yeast, in addition to a replication-independent pathway, there appears to be a replication-coupled pathway that allows CENP-ACnp1 chromatin assembly [11,41]. Deposition during S phase is coupled with the expression of new histones (dependent on Ams2) while G2 assembly requires Mis6 [11,41,42]. Higher levels of H3 interfere with CENP-ACnp1 incorporation at centromeres, suggesting that excess H3 can overwhelm the normal assembly pathways to predominate in the central domain. Surprisingly, there is no inherent impediment to deposition of H3 in the central domain. If a mechanism exists to prevent the over-incorporation of H3 in the central domain, it is easily overwhelmed. It is possible that H3 is incorporated during S phase but can be subsequently replaced by CENP-ACnp1 via a replication-independent mechanism operating throughout the cell cycle. We envisage a scenario in which excess H3 results in overloading of the central domain with H3 nucleosomes, which in turn interferes with the recognition of central domain chromatin by activities that evict H3 and replace it with CENP-ACnp1. Thus, the balance between H3, H4, and CENP-ACnp1 levels is critical to the incorporation of new CENP-ACnp1 nucleosome and kinetochore function.
S. pombe strains are listed in Table S1. Standard procedures were used for growth and genetic manipulations [54]. To construct FY4638 (cnt1:bigura4+), a 4.7-kb fragment of DNA containing the ura4+ gene (1.7 kb) flanked by 1.3 kb and 1.7 kb of DNA from the ade6+ locus on the left and right side, respectively [36], was introduced at cnt1 by transforming strain FY319 (cnt1:ade6+) [39]. Correct integrants (cnt1:ade6-ura4+-ade6) were identified by growth on plates lacking uracil and red color on limiting adenine and confirmed by PCR.
Crosses were performed using existing strains containing altered numbers of histone H3/H4 genes: FY3569 (H3:H4, 2:2), FY4753/4754 (H3:H4, 2:3), and FY4755/4756 (H3:H4, 3:2) [47] (see Table S1) to obtain the imbalanced strains FY7488 (H3:H4, 1:2) and FY7450 (H3:H4, 2:1). The imbalanced strains FY4813 (H3:H4, 2:3) and FY4816 (H3:H4, 3:2) were obtained by replacing the ura4+ gene from FY4753 (H3:H4, 2:3) and FY4755 (H3:H4, 3:2) by a 2.2-kb LEU2Sc fragment amplified using M32 and M35 primers (see Table S2). To obtain the imbalanced strains FY7370 (H3:H4, 1:2) and FY7372 (H3:H4, 2:1), FY3569 was crossed to FY4813 or FY4816, respectively.
cnp3+ was C-terminally tagged in the genome with green fluorescent protein using a PCR-based method [55].
Primers used in this study are listed in Table S2.
Histone H3.2, histone H4.2, and cnp1+ open reading frames were amplified using the following primers: H3.2-XhoI-RI, H3.2–3-Xho-Bam, H4.2–5-Xho-Bam, H4.2–3-Xho-Bam, cnp1–5-XhoI, and cnp1–3-Xho-Bam and inserted into BamHI/XhoI digested pREP plasmids. For overexpression studies, pRep81X-Cnp1, pRep3X-H3, and pRep3X-H4 were used.
Total RNA was prepared from strains grown in YES at 25 °C and RT-PCR performed as described [56]. Primers WA41 and WA42 (Table S2) were used for amplification and quantitation performed as described [35].
ChIP was performed as described [57] except for the following modifications: For Cnp1 and H3C ChIPs, cells were fixed 1% PFA for 20 min at room temperature. For GFP ChIPs, cells were grown at 25 °C, incubated 2 h at 18 °C, and then fixed for 30 min at 18 °C with 3% PFA. Cells were spheroplasted at 1 × 108 cells/ml in PEMS + 0.4 mg/ml zymolyase-100T (MP Biomedicals, http://www.mpbio.com) for 40 min at 36 °C. Cells were washed twice in PEMS and cell pellets frozen at 80 °C. The chromatin was sheared using either a MSE Soniprep 150 (SANYO, http://sanyo.com) sonicator (3 times 17 s, maximum amplitude) or the Bioruptor (Diagenode, http://www.diagenode.com) sonicator (20 min, 30 s ON and 30 s OFF at “High” [200 W] position). The extent of the shearing was checked either by ethidium bromide on a 1.7% agarose gel or by Southern blot (using ura4+ probe). The sonicated chromatin used for ChIP was less than 800 bp. 10 μl of α-Cnp1 antiserum [58], 2–4 μl of α-H3C antibody (Abcam, ab1791; http://www.abcam.com), and 1.5 μl of α-GFP antibody (Invitrogen, A-11122; http://www.invitrogen.com) were used in ChIPs. Multiplex PCR analysis was performed as described [31]. PCR products were quantified as described for RT-PCR. For the input PCR, the cnt, imr, and otr values were normalized to the fbp value, giving the “input ratio.” Enrichment of cnt (WA26-WA27 primers pair, F7cnt1-R9cnt1 primers pair, and F10cnt1-R12cnt1 primers pair, Table S2), imr (WA28-WA29 primers pair, imrEf-imrEr primers pair, or imrC2f-imrC2r primers pair, Table S2) and otr (WA31-WA32 primers pair and CenF-CenR primers pair, Table S2) bands in the ChIPs was calculated relative to the fbp (WA33-WA34 primers pair, Table S2) band and then corrected for the ratio obtained in the input PCR. ChIPs performed on strains with ura4+ insertions at centromere 1 were analyzed by PCR as described [56]. ChIP experiments were performed two to five times; representative examples are presented.
Immunolocalization was performed as described [59]. Cells were grown at 32 °C and fixed for 5–10 min in 3.7% freshly prepared formaldehyde for staining with α-Cnp1 and α-Sad1 (provided by I. Hagan) antibodies or fixed for 10–15 min in 3.7% formaldehyde, 0.05% glutaraldehyde for immunolabeling of microtubules. The following antibodies were used: sheep α-Cnp1 antiserum (1:300), rabbit α-Sad1 (1:50), mouse TAT1 α-tubulin tissue culture supernatant (from I. Hagan and K. Gull) (1:15). Alexa Fluor 594 (Invitrogen, A11016) or Alexa Fluor 488 (Invitrogen, A11029 or A21441) conjugated secondary antibodies were used at 1:1,000. Microscopy was performed as described [59] using a Zeiss Imaging 2 microscope (Zeiss, http://www.zeiss.com). Image acquisition was controlled using Metamorph software (Universal Imaging Corporation, http://www.moleculardevices.com).
Comparison of Cnp1 signal intensity in H3 = H4 (2:2) versus H4 > H3 was performed using the following procedure: Sad1 staining served as an internal control for staining efficiency and a marker for the approximate position of the centromeres (which cluster adjacent to the SPB). Image capture and analysis was performed using Metamorph software. Random fields of cells were captured using identical exposures for all samples (2 s for Sad1 and 0.5 s for Cnp1). Only G2 cells in which the Sad1 (SPB) signal was in focus were included in quantification of Sad1/Cnp1 spots. A circular region of interest (ROI) of 0.77 μm diameter was drawn around these Sad1 spots. In addition, ten ROIs per field were randomly placed on cells to measure background signal. All ROIs were then transferred to the Cnp1 image. If necessary, the position of ROIs was altered slightly (Sad1 and Cnp1 signals are adjacent or overlapping). Total signal intensity was measured for each ROI. Mean background intensity was calculated for each channel and subtracted from the mean of Sad1 or Cnp1 spot intensities as appropriate. The mean corrected Cnp1 intensity was divided by the mean corrected Sad1 intensity to normalize for efficiency of staining, giving a final value for Cnp1 staining intensity. To determine the relative Cnp1 staining intensity, the value for the H4 > H3 strain was divided by the value for the H3 = H4 strain. This method was, however, determined not to be appropriate for measurement of relative Cnp1 signal intensity in H3 > H4 cells. This was because in many cells it was not possible to place the Cnp1 ROI with confidence due to very low intensity or undetectable Cnp1 staining at centromeres. Therefore, the Cnp1 signals were simply categorized instead. Images were obtained as above, and then larger ROI circles of 1.16 μm diameter were drawn around Sad1 spots that fit the criteria described above. The ROIs were then transferred to the Cnp1 images. If a Cnp1 spot could be identified within this ROI, it was classified as “very bright,” “bright,” or “faint” according to the maximum intensity within the spot. If no spot could be distinguished above background, it was recorded as “undetectable.” |
10.1371/journal.pcbi.1003755 | Identifying Selection in the Within-Host Evolution of Influenza Using Viral Sequence Data | The within-host evolution of influenza is a vital component of its epidemiology. A question of particular interest is the role that selection plays in shaping the viral population over the course of a single infection. We here describe a method to measure selection acting upon the influenza virus within an individual host, based upon time-resolved genome sequence data from an infection. Analysing sequence data from a transmission study conducted in pigs, describing part of the haemagglutinin gene (HA1) of an influenza virus, we find signatures of non-neutrality in six of a total of sixteen infections. We find evidence for both positive and negative selection acting upon specific alleles, while in three cases, the data suggest the presence of time-dependent selection. In one infection we observe what is potentially a specific immune response against the virus; a non-synonymous mutation in an epitope region of the virus is found to be under initially positive, then strongly negative selection. Crucially, given the lack of homologous recombination in influenza, our method accounts for linkage disequilibrium between nucleotides at different positions in the haemagglutinin gene, allowing for the analysis of populations in which multiple mutations are present at any given time. Our approach offers a new insight into the dynamics of influenza infection, providing a detailed characterisation of the forces that underlie viral evolution.
| The evolution of the influenza virus is of great importance for human health. Through evolution, current influenza viruses develop the ability to infect people who have been vaccinated against earlier strains. New strains of influenza that infect birds and pigs could evolve to infect and spread between people, causing a global pandemic. The influenza virus lives within a human or animal host, so that viral evolution happens within, or in the spread between, individuals. As such, what happens to the virus during the course of an infection is a question of great interest. We here describe a statistical method that uses viral genome sequence data to measure how evolution affects the influenza virus within a single host. Studying data from infections transmitted between pigs, we find evidence for evolutionary adaptation in six of sixteen animals for which data were available. In one case, an immune response mounted by a pig against the virus is apparent. Our method provides a statistical framework for using sequence data to study viral evolution on very short timescales, enabling new research into within-host viral evolution.
| The overall risk to human health posed by the novel H7N9 influenza virus [1], while potentially severe, is as yet unknown [2], [3]. Pandemic influenza is a zoonosis [4], and as such any new pandemic may be expected to arise through a two-step process [5], [6], the virus first gaining the ability to cause sporadic, localised infections in humans until, after a second transition, emerging into a global pandemic. Each of these steps are evolutionary in nature, being characterised in turn by the adaptation of a virus to be able to infect a human host, and the development of increased transmissibility between hosts. In the nH7N9 strain, the first of these steps has already taken place, including the acquisition of mutations responsible for human-specific receptor binding [7]. Progression to a global epidemic, therefore, depends upon the evolution of increased transmissibility of the virus, a phenotypic change which can only occur while the virus grows in a host environment. As is true for other viral species [8], understanding the intra-host evolution of influenza is an important task.
A vast array of mathematical modelling approaches have been directed at the questions of influenza infection, transmission, and evolution [9]. Of particular relevance to this study are models which track the dynamics of a single infection. Based upon observed changes in viral titre over time, inferences have been made of many important properties of infection, including the reproductive number for cellular infection, the timescale and numbers of viruses produced during the infection of a cell, and the impact upon the viral population of both innate and adaptive immune responses [10]–[14]. Considering data of intracellular RNA levels, the fine detail of viral replication within a cell has been described [15]. Evolutionary models of competition between viral strains have clarified the relationship between selection for growth and transmission effects, and the dynamics of immune escape [16]–[18].
In the cases above, the viral population was either modelled as a population of identical individuals, or as a set of distinct classes of virus, characterised by differing immune escape or transmission properties. Building upon these approaches, a genetic classification of viruses was used to model H5N1 influenza evolution [19]; the fitness of a virus was defined according to the presence or absence of a set of mutations. Here we divide the viral population in a similar manner, expressing the fitness of a virus as a function of its genetic composition. However, rather than analysing the consequences of a proposed fitness landscape, we here infer how selection was actually at work based upon observed genetic sequence data.
In chronic infections such as HIV, time-resolved sequence data from individual hosts is readily available [20]. However, the course of an influenza infection, even in an immunocompromised host [21], is relatively short. As such, time-resolved genetic data is rare, the main examples having been collected from experimentally-infected animal populations [22], [23]. In this work, we consider data from one such study, examining the evolution of H1N1 influenza within individuals in a swine population [24], [25].
The basic principle of our method is to learn the role of selection acting upon a viral population by means of a maximum likelihood method. We adopt a coarse-grained quasispecies model (cf. [26]) to describe the evolution of the viral population, in which viruses are classified according to the nucleotides (here denoted alleles) present at a limited number of positions (or loci) in their genomic sequence. In this model, evolution proceeds deterministically, contingent only upon the initial state of the population, and the role of selection for or against specific alleles. By considering the consequences for the population dynamics of different proposed models of selection, and comparing these to the observed evolution of the system, we estimate how selection was at work.
The low rate of recombination within RNA segments of influenza [27], [28], combined with a high viral mutation rate, leads to complex evolutionary dynamics, with the fate of mutations being strongly affected by genetic hitchhiking and clonal interference [29]–[31]. As such, discerning the effects of selection requires that interactions between alleles at different loci are taken into account [32]. Here this is achieved by considering the frequencies of haplotypes, sets of sequences with specific alleles at specific loci (e.g. allele C at locus i and allele T at locus j).
In our model, the viral population can be described at potentially any genomic resolution, keeping track of the population in terms of haplotyes spanning arbitrary numbers of loci. However, higher-locus models are more computationally demanding. As such, we first apply a filtering process to cut out loci at which alleles do not show statistical evidence of having evolved under selection. For each polymorphic locus, we use a single-locus model of evolution to find alleles that appear to evolve in a non-neutral behaviour, changing in frequency over time. Change in the frequency of an allele may occur as the direct result of selection, or due to linkage disequilibrium with a selected allele, or alleles, at other loci. As such, to distinguish between these cases, wherever apparent non-neutrality is observed at more than one locus, we apply a multi-locus model of haplotype frequency change to the data. This model explicitly accounts for interactions between alleles at different loci, and is used to identify the maximum likelihood explanation for the changes observed in the sequence data.
As has been noted elsewhere, the use of viral sequence data to understand population structures requires substantial care (e.g. [33], [34]). Selective amplification of sequences, or general sequencing bias, can produce a misleading picture of a population as a whole. PCR-induced recombination can lead to false measurements of linkage disequilibrium between alleles at different loci. We discuss the potential impact of each of these factors upon our results.
Viral sequence data collected from a previous transmission experiment [25] were analysed. An overview of the structure of this experiment is shown in Figure 1. The chain of infection was propagated by a process of housing pairs of uninfected pigs with pairs of infected pigs, the previously-infected pigs being removed after transmission had occurred. Throughout the experiment, samples were collected from pigs using nasal swabs, with viral sequences being amplified via RT-PCR and Sanger sequenced. Viral sequences were collected from the majority of the pigs; for 16 of the 24 pigs involved in the experiment, data was collected at more than one time-point, an essential prerequisite for our method. For the samples collected in these animals the depth of sequencing varied from 6 to 81 sequences (mean 51) from a pig at a given time-point, with data being collected at up to five time-points across the course of an infection. Limited transmission of variants was observed between individual infections.
In our analysis, non-neutral behaviour was identified in six populations. In general, signs of selection were relatively rare. While very many individual mutations were observed in the population as a whole, most of the substantial changes in allele frequency occurred at a small number of sites (e.g. Figure 2). As such, eighteen alleles in the dataset were identified as being potentially non-neutral. Interference effects between alleles were found to be of importance; of these eighteen alleles, a total of nine were identified as being genuinely under selection, changes in frequency at the other nine being explicable in terms of linkage disequilibrium with other selected alleles. In the populations identified to be non-neutral, a variety of forms of selection were found, including evidence for time-dependent selection, and for selection acting simultaneously at more than one locus (a selection of inferred trajectories are shown in Figure 3; further inferences are presented in Supporting Figure S1). Our multi-locus model discriminated between cases where multiple alleles changed in frequency under independent selection, and cases where selection acting upon one allele led to substantial changes in the frequency of others (Table 1).
In Pig104, strong evidence [35] was found for negative selection acting against the G → A mutation in locus 114, with an inferred selection coefficient of −1.6 per 12 hours (h). Such a magnitude of selection is relatively large; by comparison, an allele at frequency 50% with a selection coefficient of −1 per 12 h would decrease to 12% frequency after one day and to less than 2% after 2 days. The mutation under selection in this case is synonymous, such that the observation of strongly deleterious selection is perhaps a surprising one. While, using our method, no statistical evidence for selection upon this allele was identified in other pigs, the same polymorphism was found in data collected at the earliest time point for pigs 115 and 116, but not at subsequent time-points, consistent with a hypothesis of negative selection for this nucleotide across all viral populations.
In Pig109 strong evidence was found for positive selection upon at least two of three alleles; in favour of the G → A polymorphism at locus 553, the A → G polymorphism at locus 696, or the G → A polymorphism at locus 914. Fixation of all three of these mutations occurred between two samples, and models with any single one of these mutations as the selected allele performed similarly well, giving estimated selection coefficients between 3.0 and 3.1 per 12 h for the selected allele. Joint consideration of four-locus haplotype frequencies provided evidence that at least two of these mutations were independently under selection. The most likely model had coefficients of 2.8 per 12 h at each of the loci 696 and 914. However, the difference between two-locus additive models was small, and models in which any two of the three polymorphisms were under selection performed similarly well (Supporting Table S1). An interesting feature of this result is that the pairs of mutant alleles inferred to be under selection are highly linked, the mutant alleles at loci 696 and 914 appearing only jointly on a sequence, and never in isolation. The inference that selection is acting at two loci, rather than at only one locus, arises from the effect of mutation in the model; this result is explored more fully in Supporting Information. We note that, while the polymorphism at locus 696 is synonymous, those at 553 and 914 are non-synonymous in character, corresponding to the mutations D185N and S305N (the former being contained within the Ca2 epitope region [36]).
In Pig115 weak evidence was found for positive selection in favour of the G → A polymorphism at the locus 188, with an inferred selection coefficient of 1.2 per 12 h. This polymorphism is non-synonymous, representing the amino acid substitution G63E. Bootstrapping of this result against inferences from sequence data that had been randomised in time largely supported this inference; from a total of 200 sets of randomised sequence data, a stronger signal in favour of a model of constant selection was identified in only eight cases. Details of the bootstrapping of all results are given in Supporting Text S1 and in Supporting Figure S2.
In Pig405, strong evidence was found for positive selection acting upon the G →A polymorphism at locus 844, with a selection coefficient of 0.4 per 12 h, along with simultaneous, time-dependent selection acting upon the A → G polymorphism at locus 553. Selection at this second locus was inferred to be initially positive, with mean strength 0.9 per 12 h during the first time-interval, weakly negative during the second time interval, with mean strength −0.1 per 12 h, then finally strongly negative, of mean magnitude greater than −2 per 12 h for the final time interval. Each of these polymorphisms are non-synonymous (corresponding to the mutations V282I and N185D respectively; the mutation at locus 553 is identical to that observed in Pig109, albeit in the reverse direction). Identification of time-dependent selection acting upon the latter, epitope mutation is of particular interest, raising the possibility that this corresponds to an adaptive immune response by the host to the virus. In this population the magnitude of the time-dependent selection inferred for the final time-point was large and negative, but hard to identify with precision. This arises from a time-dependent model of selection being coupled with an observed allele frequency of zero at the final time-point. Excluding the influence of allele frequencies at other loci, the data in such a case can lead to an inference of arbitrarily strong negative selection; the time resolution at which data are collected imposes a limit on the magnitude of selection that can correctly be inferred [37].
In Pig410 we identified weak evidence for time-dependent selection acting upon the synonymous C → T mutation at locus 447; in this case, a bootstrapping calculation produced a stronger signal of selection than that for the real data in only three out of 200 cases (Supporting Figure S2). Time-dependent selection was also identified in Pig412, where strong evidence was found for time-dependent selection acting upon the synonymous G → A mutation at locus 696, with further weak evidence for negative selection acting upon the synonymous A → G mutation at locus 48. Under the multi-locus model, a selection coefficient of 1.8 was identified at locus 696 for the first time interval. The inferred strength of selection at this locus for the second, final time interval was imprecise, but very large and negative; the value of −22.8 per 12 h reported in Table 1 again being caused by an observed frequency of zero at the final time-point.
Alleles at which selection was inferred were distributed across the HA protein (Supporting Figure S3). Significant changes in allele frequency were identified in more than one infection at five different loci (447, 553, 696, 824 and 844). Of these, selection was inferred to act at the loci 696 and 844 in more than one infection. This repetition of mutations may be explained by the design of the experiment; selection is most likely to be observed when polymorphisms exist at non-negligible frequency in the population, while polymorphisms at higher frequencies are more likely to be transmitted between infections.
Under an initial scan for potentially non-neutral alleles, very weak evidence for selection was identified in the data from Pig113 at the three loci 447, 824 and 844. However, under the full multi-locus model, a neutral model of evolution was finally preferred. As we discuss further in Supporting Text S1, our evolutionary model is more conservative in identifying selection in cases where multiple loci are considered simultaneously.
We have here described a novel approach to understanding the within-host evolution of the influenza virus, based upon sequences collected at subsequent times within a single infection. Our method combines a quasispecies model of viral evolution with a hierarchical set of potential models of selection, identifying the evolutionary scenario which best explains the observed sequence data. A crucial component of our model is its accounting for linkage disequilibrium between alleles at different loci; while a single-locus model is sufficient for cases in which only one mutation in a gene changes in frequency [38], the observation of more than one simultaneous change in allele frequency within a non-recombinant gene demands a more sophisticated analysis.
Our approach to inferring selection differs substantially from the calculation of dN/dS [25], not least in considering data at the haplotype frequency level. While in earlier work dN/dS has been applied to sequences collected across viral populations from all observed infections, we allow for the landscape of selection acting upon the virus to vary between animals, or potentially to change within a single animal over time. The results of our analysis also differ; while significant dN/dS ratios were identified at the codon positions 204 and 257, we did not find evidence of selection for alleles at either of these loci. We note that, over short time-scales, difficulties may arise in using numbers of synonymous and non-synonymous mutations to infer selection. While this approach is of great value when applied to diverged sequences, such as those collected from homologous genes in different species [39], its application to sequences from a single population gives results that may be harder to interpret [40], [41].
Our approach to within-host viral evolution is rooted in the interpretation of viral sequence data, collected at multiple times from single infections. By modelling evolution, it is possible to assess the consequences for a viral population of hypothetical fitness landscapes (e.g. [16]). If it is known that a mutation fixes with given probability in a given timescale, the requisite fitness advantage conferred by that mutation can be learnt [42], [43]. However, obtaining a detailed picture of within-host viral evolution requires the use of time-resolved sequencing, describing the population at multiple time points. Our method provides a systematic approach to inferring selection; while the set of potential fitness models is very large [44], we build upwards from a neutral model to increasing complexity, as guided by the data. Keeping data central to our approach means that we may miss the influence of certain fitness effects; sufficient data may not be available to infer the complete picture of how evolution is at work. However, our hierarchical approach means that, given accurate data describing a population, we should not generate false inferences of the presence of selection.
Analysing the data, we identified selection acting upon both synonymous and non-synonymous mutations. Weak selection acting upon synonymous mutations has been identified for codon usage in influenza [45] and against mutations that disrupt RNA structure in HIV [46], although the magnitude of selection inferred here is significantly higher than in either case. While inferring the presence of selection, our method cannot match occurrences of selection to specific biological mechanisms; further data would generally be required to do this.
One result for which a biological mechanism may be proposed is in the viral population of Pig405, where we identified variable, and decreasing selection acting upon a non-synonymous mutation in the Ca2 epitope region, potentially as a result of a specific immune response. For this mutation the timing of the onset of strong negative selection, in the fourth day after exposure to the virus, is earlier than the five days before detection of an adaptive response reported for an H3N2 influenza infection in mice [47]. Further to this, modelling studies have associated the innate immune response with an initial decline in viral load, the adaptive response leading to final clearance of the virus [13]. Here, no drop in viral titre was seen at the time of inferred negative selection, with clearance occurring eight days after infection [24]. Again, further data would be required to produce a more specific conclusion; combined data of viral sequence and immune response would lead to greater understanding of systems such as this.
Our evolutionary model assumes that the viral population is genetically well-mixed in the host, and that it evolves in a deterministic manner, both with respect to mutation, and to selection. The first of these assumptions asserts that each sample of viruses collected from the pig is representative of the viral population in the animal at the time. This would not be true if, for example, the viral population was split into diverse subsets, with selection acting in very different ways in each. Study of these effects was not possible given the data studied here.
Our assumption of deterministic evolution is based on the underlying viral population being large in number, that is, large enough that and are significantly greater than 1, where N is the number of viruses in an animal, μ is the mutation rate per locus, and σ is the magnitude of selection [48]. Considering selection, the lowest resolution at which we report selection, of 0.1 per 12 h, is, accounting for two rounds of replication in the lifetime of an infected cell [10], [15], equivalent to a fitness difference of 0.05 per generation. As such, this part of the assumption holds if N is substantially larger than 20 viruses. Considering mutation, the criterion that is stricter than that for selection (where μ is of order 10−5 [49], [50]), requiring N to be substantially larger than 105. In influenza, models of replication in a single cell suggest that of the order of 104 virions are produced within each cell [51], while in the samples from which viruses were sequenced, a viral load of between 30 and 5500 particles per µl [24] was measured; once an infection has progressed to the point where viral sequencing is possible, the population is very likely large enough for this to be fulfilled. In the earliest stages of an infection, stochastic mutational behaviour could potentially lead to an incorrect inference of the initial variant frequencies within the population; however, these values are not used to draw any biological conclusions about the system.
Horizontal transmission between co-housed animals was not incorporated into the model; we believe this was unlikely to have greatly influenced the collected data. If the viral populations in the two simultaneously infected pigs were substantially different in composition, transmission of viruses from one animal to the other might alter the composition of the viral population in the second animal. However, the viral populations in this experiment were not sufficiently different in sequence to be able to distinguish superinfection from the growth of de novo mutations. Further, while the viral titre implicated in transmission is unknown, we believe that the incoming titre is likely to be substantially smaller than the pre-existing number of viruses in the second infected animal.
A second assumption in our study is that the collected sequence data are relatively accurate. That is, we assert that the sequences obtained from the sample are representative of the sample itself. The basis of our inference upon data means that the accuracy of the data is vital for obtaining useful results. For example, in addition to raw allele counts, our approach makes explicit use of linkages between mutations. Our method allows for the possibility of generic error in the sequencing process, and fully accounts for the statistical noise inherent to a finite data sample. However, there are systematic data biases that may also affect the results obtained. For example, PCR-induced recombination has the potential to alter the observed frequencies of multi-locus haplotypes [52], [53]. Testing for such an effect, by fitting an exponential model to the observed absolute linkage disequilibrium between pairs of alleles, we found no evidence for such recombination, no decay in this statistic being observed with increasing distance between alleles (Supporting Figure S4).
Sequencing bias also has an effect on whether or not a mutation is recognised as being under selection. Mutations that are preferentially identified by a sequencing method would appear in the sample at higher frequencies, such that changes in their frequencies were amplified, leading to a greater chance that such mutations were found to be under selection. For this dataset, a consistent sequencing method was used to process all of the samples; we therefore assumed sequencing bias to be consistent between samples, such that observed changes in allele frequency were caused either by the finite sampling process, or by a process of mutation and selection. Estimating the extent of sequencing bias in the observed sequences is difficult, the sequences themselves representing the only information about the real viral population. Counting the mutations observed in the data showed a high transition:transversion ratio of 9.7 (Supporting Figure S5). This is broadly consistent with values observed for other RNA viral populations [54], [55], albeit that measurements of this ratio in influenza have previously been based upon global, rather than within-host, populations [56]. Biased sampling, whether occurring via the collection of a biological sample that is unrepresentative of the whole population, or as a result of the subsequent PCR amplification, also has the potential to affect our inference. We have here assumed that the data is an unbiased sample of the real population.
Our inferences are partially limited by the use of sequences describing only the HA1 region of the influenza virus. While our inferences of deviation from neutrality in a population are not affected by alleles elsewhere in the virus, the attribution of selection to given alleles may be affected by unobserved polymorphisms in the HA2 region of influenza, or if reassortment were limited (though see [57]), with alleles in other viral segments. The potential influence of selection acting upon polymorphisms that have not been observed is of greatest relevance to the cases of apparently time-dependent selection; constant selection acting upon interfering mutations causes time-dependent selection effects [32]. One example is the case of Pig412 where initially positive, then negative selection is inferred. In this infection, many haplotypes which are observed at the intermediate time point are no longer seen in the final time point; this pattern is consistent either with a switch in the direction of selection acting upon the synonymous mutation at locus 696, as was inferred, or with very strong positive selection acting upon an unobserved mutation on the consensus haplotype causing a selective sweep later in the observation. Such a scenario is much less likely in the case of Pig405, where the haplotype containing the allele inferred to be under negative selection is outcompeted in the final time interval by four other haplotypes, including that of the initial consensus.
We have here described a framework for the inference of selection acting upon a viral population within an individual host, based upon time-resolved sequence data. Within-host selection is of importance for the future evolution of the H7N9 influenza virus, and for understanding the epidemiology of other influenza strains. During an epidemic, both within-host growth, and the transmission of viruses, are important, and potentially competing factors; a mutation which is beneficial for within-host growth may prove deleterious for transmission and vice versa. While we have here considered only the first of these factors, our method could easily be used to infer the role of selection for transmission, given specific conditions. First of all, substantial continuity would be required between the native and the transmitted populations, such that changes in allele frequencies before and after transmission were primarily the result of selection; severe bottlenecking would distort the population structure. Secondly, clarity would be required about the source of each infection; in the experiments considered, where an infection begins with an unknown mixture of viruses from two other individuals, the role of selection in transmission cannot be evaluated. Transmission events in the data analysed here have been discussed elsewhere [58]. In more straightforward cases, where transmission occurs between known individuals, and where continuity between viral populations is more evident (e.g. [59]), use of our method to infer selection acting across transmission events is likely to be achievable.
The collection of sequence data describing the within-host evolution of influenza is at present, relatively rare, although we anticipate that improvements in sequencing technology will make such data increasingly accessible. Increased collection of sequence data from patients, and from evolutionary experiments, will greatly add to our understanding of viral infection. Our approach increases the value of such work, characterising in detail the forces that underlie within-host viral evolution.
Quasispecies theory [26] provides a deterministic description of the evolution of mutation-prone, self-replicating organisms; this framework has profoundly influenced studies of RNA viral evolution [60]–[63]. To describe the evolutionary dynamics of the influenza virus within an individual host we apply a coarse-grained quasispecies model, in which the viral population is described as haplotypes spanning a limited set of loci, rather than as complete viral sequences. Specifically, we represent the viral population as a frequency vector , defined at discrete times , and comprised of elements , where is the fraction of sequences in the population with the haplotype ; that is, with the nucleotides at a subset of loci in the viral genome.
To model mutation between haplotypes, we assumed a constant rate of mutation, μ, between any two specific nucleotides at a given locus, the probability of mutation from haplotype to haplotype in a single generation being given by (1)where is the Hamming distance between the two haplotype sequences.
Selection was accounted for by ascribing to each haplotype the (potentially time-dependent) selection coefficient . The effect of selection on the haplotype frequency between times and was thus defined by the function : (2)where . Considering the evolution of influenza, we supposed time-points to be spaced at 12-hour intervals, roughly approximating the time required for a round of intracellular growth within a cell [10]. Within such a round of growth, each virus undergoes two rounds of replication, modelled as having equal mutation rates, with the parameter representing an overall rate of mutation per nucleotide per generation of 10−5 [49], [50]. Selection was assumed to act upon the viral population once it has exited the cell, giving the relation (3)where is the matrix consisting of elements , modelling a single round of replication. The behaviour of the system is thus specified in a deterministic manner by the selection parameters , and by the initial state of the system, given by the elements of the vector .
We note that, while sequence data was collected at known times throughout the course of each infection, the precise moment at which each infection began is unknown. Here, we assumed to be precisely 24 hours before the first observed set of sequence data from the infection. While the uncertainty in this value has consequences for the accuracy of the elements of the inferred vector , no conclusions were finally drawn from these values.
An inference of selection was carried out by comparing maximum likelihood values obtained under a hierarchical series of models, each specifying the parameters and . The coarse-grained quasispecies model can be expressed in terms of haplotypes of arbitrary length. We describe the general model below.
In order to test for the influence of PCR-induced recombination upon the dataset, we calculated a measure of linkage disequilibrium between loci. For each pair of polymorphic loci in the dataset, we calculated the value , equal to the absolute linkage disequilibrium between these loci, normalised by the maximum potential linkage disequilibrium given the allele frequencies in question (8)where the labels 0 and 1 represent the consensus and most common minor alleles at each locus, represents the frequency at time of the allele at locus , and represents the frequency at time of the haplotype at loci and . Values of were compared for loci at different positions in the sequence, fitting a model of the form, for all points for which , where is the sequence distance between loci and . Here a greater negative value of would indicate that a higher mean rate of recombination in the viral sequences occurred during the sequencing process.
A test of the ability of the method to discriminate between selected and non-selected alleles, and to correctly infer the magnitude of selection acting upon a locus, was performed by running analyses for simulated data. For simulated populations with a single allele under selection, a correlation coefficient of of more than 0.95 was found between real and inferred selection coefficients, with an equivalent correlation of 0.91 for simulated systems with two alleles under selection. Further details are given in Supporting Text S1 and Supporting Figures S6 and S7.
|
10.1371/journal.pgen.1005850 | Exome Sequencing of Uterine Leiomyosarcomas Identifies Frequent Mutations in TP53, ATRX, and MED12 | Uterine leiomyosarcomas (ULMSs) are aggressive smooth muscle tumors associated with poor clinical outcome. Despite previous cytogenetic and molecular studies, their molecular background has remained elusive. To examine somatic variation in ULMS, we performed exome sequencing on 19 tumors. Altogether, 43 genes were mutated in at least two ULMSs. Most frequently mutated genes included tumor protein P53 (TP53; 6/19; 33%), alpha thalassemia/mental retardation syndrome X-linked (ATRX; 5/19; 26%), and mediator complex subunit 12 (MED12; 4/19; 21%). Unlike ATRX mutations, both TP53 and MED12 alterations have repeatedly been associated with ULMSs. All the observed ATRX alterations were either nonsense or frameshift mutations. ATRX protein levels were reliably analyzed by immunohistochemistry in altogether 44 ULMSs, and the majority of tumors (23/44; 52%) showed clearly reduced expression. Loss of ATRX expression has been associated with alternative lengthening of telomeres (ALT), and thus the telomere length was analyzed with telomere-specific fluorescence in situ hybridization. The ALT phenotype was confirmed in all ULMSs showing diminished ATRX expression. Exome data also revealed one nonsense mutation in death-domain associated protein (DAXX), another gene previously associated with ALT, and the tumor showed ALT positivity. In conclusion, exome sequencing revealed that TP53, ATRX, and MED12 are frequently mutated in ULMSs. ALT phenotype was commonly seen in tumors, indicating that ATR inhibitors, which were recently suggested as possible new drugs for ATRX-deficient tumors, could provide a potential novel therapeutic option for ULMS.
| Uterine leiomyosarcomas are rare, malignant smooth muscle tumors with a poor 5-year survival and high recurrence rate. They account for 1–2% of all uterine malignancies with an estimated incidence of 0.4/100,000 women per year. The symptoms and signs of this tumor type widely overlap with those of common benign uterine leiomyomas, making early diagnosis of uterine leiomyosarcomas difficult. Currently, the diagnosis of these tumors is often incidental and postoperative. Despite previous cytogenetic and molecular studies, their molecular background has remained elusive. Identification of novel molecular genetic characteristics in uterine leiomyosarcomas is clinically relevant to further improve the diagnosis and prognosis of the patients. Here, we performed exome sequencing on 19 tumors, revealing frequent mutations in TP53, ATRX, and MED12. The discovery of frequent inactivating ATRX mutations provides a potential novel therapeutic target for uterine leiomyosarcomas.
| Uterine leiomyosarcoma (ULMS) is a rare, highly malignant tumor that originates from the smooth muscle layer of the uterus, the myometrium. It is the most common subtype of uterine sarcoma and accounts for 1–2% of all uterine malignancies with an estimated incidence of 0.4/100,000 women per year [1,2]. The majority of ULMSs occur in women over 50 years of age typically causing symptoms such as abnormal vaginal bleeding, palpable pelvic mass, and abdominal pain. These symptoms greatly resemble those of common benign uterine leiomyoma, making early diagnosis of ULMS difficult. Surgical resection is the primary treatment option, while the use of adjuvant therapies varies widely. ULMS show low sensitivity to both chemotherapy and radiation therapy [3,4]. In most cases, the diagnosis is made histologically after the surgery, and even then, the clinical course of ULMS is difficult to predict. Currently, the most prominent prognostic factors include stage, age, and tumor size [5–7]. The 5-year overall survival has remained <50% due to a high recurrence rate (53–71%) and metastatic capacity [6,8].
Most ULMSs are aneuploid with both complex numerical and structural chromosomal aberrations [9]. Albeit no consistent structural aberrations have been identified, abnormalities affecting chromosomal regions 1p, 10q, 13q, and 14q have been observed in multiple cases [10]. So far, only a few genes have been associated with this tumor type, including tumor protein P53 (TP53), RB1, MDM2, CDKN2A, and KIT [9,11]. These are all common cancer genes not specific for smooth muscle malignancies and the exact molecular mechanisms underlying ULMS tumorigenesis remain elusive.
During the last decade, next-generation sequencing technologies have increasingly provided genome-wide data on somatic landscapes in various cancer types enabling the discovery of novel cancer genes and mechanisms with important prognostic and therapeutic implications [12]. Here, we performed exome sequencing on 19 ULMSs to further elucidate the molecular etiology of these tumors, identifying frequent mutations in TP53, alpha thalassemia/mental retardation syndrome X-linked (ATRX), and mediator complex subunit 12 (MED12). This is the first description of high-throughput sequencing on ULMSs.
We performed exome sequencing on genomic DNA of 19 formalin-fixed paraffin-embedded (FFPE) ULMSs. The average coverage of captured exonic regions reached a mean depth of 21x and 92% of the captured regions had a minimum coverage of four reads (S1 Table). After filtering the exome sequencing data, we observed a mean of 373 somatic mutations per tumor (range 240–779). The majority of mutations in each tumor specimen represented single-nucleotide variations (∼88%; range 81–95%), while deletions accounted for ∼9% (range 4–15%) and insertions ∼3% (range 1–7%) (S1 Table). Two tumors, LMS49 and LMS51, harbored more mutations than other ULMSs, but the mutation spectrum followed the common pattern.
In the exome sequencing data analysis, we focused on genes that were mutated in at least two tumors. This resulted in a list of 43 genes (S2 Table). The majority of these genes (37/43; 86%) were mutated in two tumors, while six genes, TP53, ATRX, MED12, fibrous sheath interacting protein 2 (FSIP2), ATP-binding cassette, sub-family A (ABC1), member 13 (ABCA13), and ankyrin repeat domain 26 (ANKRD26), had mutations in three or more tumors (Fig 1). The most frequently mutated gene was TP53, which was mutated in six tumors (6/19; 32%) (S1 Fig). Two mutations were nonsense mutations creating a premature stop-codon and four were missense alterations; all missense changes were predicted pathogenic by two independent in silico tools (S2 Table). All the observed TP53 mutations have been reported as somatic mutations in the COSMIC-database.
The second most commonly mutated gene was ATRX, which was mutated in five tumors (5/19; 26%) (S1 Fig). The total number of mutations was six as one tumor (LMS71) contained two distinct mutations. All mutations were either nonsense mutations or small frameshift insertions or deletions, and were thus predicted to result in a truncated protein product. As ATRX mutations have been associated with alternative lengthening of telomeres (ALT), we specifically searched the exome sequencing data for possible mutations in death-domain associated protein (DAXX), as also these mutations have been associated with the ALT phenotype [13,14]. Indeed, one ULMS (LMS61) had a mutation in DAXX. This mutation was a nonsense mutation (Glu650Stop) most likely leading to a truncated or unstable protein product.
Four mutations (4/19; 21%) were observed in MED12 (S1 Fig). All these were missense changes affecting amino acids Gly44 (3 mutations) or Leu36 (1 mutation), which have previously been reported as mutational hotspots in uterine leiomyomas [15]. All mutations were predicted to have a deleterious effect on protein function (S2 Table). Neither MED12, TP53, nor ATRX mutations were mutually exclusive (Fig 1).
Alterations in FSIP2 (4/19; 21%), ABCA13 (3/19; 16%), and ANKRD26 (3/19; 16%) all represented missense changes that scattered along the gene lengths. Two tumors had the same Met487Ile substitution in ANKRD26. One alteration (Gln581Leu) in FSIP2 and all changes in ABCA13 were predicted pathogenic by both Polyphen-2 and SIFT, whereas none of the other variants were predicted damaging by both in silico tools.
We evaluated the protein expression levels of TP53, ATRX, and DAXX in the 19 exome-sequenced ULMSs by immunohistochemistry and validated the results in a larger set of 33 additional tumors (S2 Fig and S3 Table). DAXX immunostaining was successful in all 52 tumors and interpretable results for TP53 and ATRX were obtained from 50 and 44 tumors (50/52, 96%; 44/52, 85%). Aberrant TP53 expression was observed in 33 out of 50 ULMSs (66%) (S3 Table). Twenty-three out of 44 successfully analyzed ULMSs (52%) showed loss of nuclear ATRX expression, including all immunohistochemically successful ATRX mutation-positive tumors. Clearly diminished DAXX expression was present in only one ULMS (1/52, 2%) (Fig 2A and 2C): a tumor with the nonsense mutation (S3 Table).
Telomere-specific fluorescence in situ hybridization (FISH) was carried out to assess the potential effect of ATRX and DAXX mutations on telomere length (Fig 2B and 2D). Twelve out of 19 exome-sequenced ULMSs (63%) were ALT-positive (S3 Table). This included four out of five ATRX mutation-positive tumors (80%) as well as the one DAXX mutation-positive tumor. Also seven out of 13 cases (54%) without detectable ATRX or DAXX mutations showed ALT positivity. Loss of ATRX or DAXX expression seems to correlate very well with the ALT phenotype.
Kaplan-Meier survival curves were generated to study the association between TP53 and ATRX expression and overall survival time. The median overall survival time for all patients was 65 months (95% confidence interval 31.7–98.3 months). Only the number of Stage I tumors was large enough for the analyses. Neither aberrant TP53 or ATRX expression associated with poor survival (P = 0.261, P = 0.127) (Fig 3). Of note, TP53 and ATRX expression statuses correlated with each other (P = 0.005).
In this study, we examined somatic variation in 19 ULMSs by exome sequencing. We focused on genes, which were mutated in at least two tumors; altogether 43 such genes were identified. The most frequently mutated genes included TP53, ATRX, MED12, FSIP2, ABCA13, and ANKRD26. TP53 was the most commonly mutated gene with 32% of the tumors harboring mutations. Alterations in TP53 have been previously implicated in leiomyosarcomas and suggested to play a role in leiomyosarcoma pathogenesis [16–18]. In this study, most mutations (67%) were missense changes located in exons 4–8. This is in line with previous studies, where the majority of mutations have been missense mutations in exons 5–8, the most highly conserved region of the gene [9,17,19]. These mutations are known to alter the protein structure, inhibit its tumor suppressor function, and result in its prolonged half-life. Immunohistochemical analysis including 50 ULMSs confirmed altered expression in the majority of tumors, highlighting the role of TP53 in ULMS development.
ATRX was the second most frequently mutated gene with mutations observed in five tumors (26%). All mutations were either nonsense or frameshift alterations most likely leading to a truncated protein product. Loss of ATRX expression has been reported in leiomyosarcomas of various sites [20–22] and a recent meeting abstract on ULMSs reported genomic alterations of this gene in 32% (8/25) of the studied tumors, supporting our findings [23]. We successfully analyzed ATRX protein levels in 44 ULMSs and showed that 52% of the tumors, including all reliably analyzed mutation-positive lesions, had clearly reduced expression. ATRX encodes a transcriptional regulator that contains an ATPase/helicase domain, and is thus a member of the SWI/SNF family of chromatin remodelling proteins. Loss of ATRX expression has been associated with ALT [13,24], which prompted us to analyze the telomeres with telomere-specific FISH. The ALT phenotype was confirmed in all ULMSs with diminished ATRX expression. Some exome-sequenced tumors with reduced ATRX expression and ALT positivity did not show ATRX mutations, suggesting that there are regulatory or larger structural alterations undetectable by exome sequencing, or that the quality of FFPE samples was inadequate to reveal the underlying mutation. Interestingly, the only ULMS with two ATRX mutations did not show ALT.
ATRX is known to functionally cooperate with DAXX and DAXX mutations have been associated with ALT [13,14]. We therefore scrutinized the exome data for possible DAXX mutations. One nonsense mutation was identified, and FISH confirmed the ALT phenotype. Overall, these results show that ALT is very common in ULMS and that in addition to ATRX, also DAXX mutations contribute to the phenotype. Importantly, ALT was recently suggested to render cancer cells hypersensitive to ATR inhibitors [25]. These inhibitors might provide a novel treatment for ULMS, in which chemotherapeutic options have thus far been limited.
MED12 was mutated in four ULMSs (21%). All mutations were in exon 2, which is a known mutational hotspot in MED12. These mutations were first observed in uterine leiomyomas [15], and subsequently they have been identified in other tumor types [26]. Previous screening studies have reported recurrent MED12 mutations also in ULMS with similar frequencies as observed here [26]. It may be that a subset of ULMSs arises through a leiomyoma precursor, or alternatively MED12 mutations may provide growth advantage to ULMSs. MED12 is part of a multi-protein complex Mediator, which plays a key role in global transcription regulation in eukaryotic cells [27]. Based on our results, MED12 mutations can co-occur with TP53 and ATRX mutations.
FSIP2, ABCA13, and ANKRD26 were mutated in at least three tumors and additional 37 genes had mutations in two tumors. Most alterations were missense changes and gave either neutral or controversial results in in silico predictions. The possible role of these genes in ULMS development cannot be directly assessed as in addition of providing growth advantage to the cell, the observed alterations may represent rare germline polymorphisms or passenger mutations with no functional significance.
Although aberrant expression of both TP53 and ATRX in Stage I ULMSs, the only group of tumors large enough for the analyses, did not associate with poor overall survival, a trend toward poorer survival was seen in the patients. The limited number of samples in the survival analyses and the observation that expression statuses were associated with each other makes it difficult to draw conclusions regarding prognostic implications of TP53 or ATRX expression levels. In general, TP53 alterations are the most common genetic changes in human cancers and they are particularly associated with an aggressive phenotype. Recently, loss of ATRX expression was associated with poor clinical outcome in ULMS [21,22]. Larger sample series with information on both TP53 and ATRX are required to confirm these findings.
ULMSs are rare and aggressive cancers. In most cases the diagnosis is made only at surgery, and many patients thus present with an advanced disease. Here, we have utilized exome sequencing and identified several recurrently mutated genes, including TP53, ATRX, and MED12. While MED12 mutations are the most common alterations in benign conventional leiomyomas, TP53 or ATRX mutations have not been observed in these tumors. Specifically, identification of inactivating ATRX mutations and their association with the ALT phenotype in the substantial proportion of tumors may be translatable into clinical practice should the suggested effect of ATR inhibitors prove effective.
This study was approved by the appropriate ethics review board of Hospital District of Helsinki and Uusimaa, Finland (408/13/03/03/2009).
Fifty-two archival FFPE ULMS tissue samples were derived from the Department of Pathology, Hospital District of Helsinki and Uusimaa, Finland, according to Finnish laws and regulations by permission of the director of the respective health care unit. These specimens represented diagnostic ULMS samples collected during surgery in 1985–2013. Simultaneously with the sample collection, clinical data were obtained for these cases (Table 1) after which the samples were anonymized for the study. Nineteen ULMSs (diagnosis 2003–2013) entered exome sequencing, while the remaining 33 tumors were available on a tissue microarray for immunohistochemistry.
Before exome sequencing, hematoxylin-eosin-stained sections from each specimen were re-evaluated by a pathologist (RB) and verified as ULMSs according to the WHO criteria [28]. Tumor percentage was ≥90% in all samples.
Genomic DNA was extracted with a standard phenol-chloroform method. Sample libraries were prepared using NEBNext DNA Library Prep Reagent Set for Illumina (New England Biolabs Ltd. catalog# E6000) and subjected to exome capture with NimbleGen SeqCap EZ System (Roche NimbleGen). Paired-end short read sequencing was performed with HiSeq 2000 (Illumina Inc.) at Karolinska Institutet, Sweden.
Read mapping and somatic variant calling were carried out as previously described [29]. Additionally, single duplicate reads were removed with an in-house script.
Exome data was analyzed with an in-house analysis and visualization tool RikuRator. The requirements to call a variant included a minimum coverage of six reads and the mutated allele to be present in at least 20% of the reads. The Genome Analysis Toolkit (GATK) quality score of variants was required to be 25 or above. Both exonic regions and sequences within three base pairs of the exon-intron boundaries were included in the study. Synonymous changes, variants present in the dbSNP database (release 138), and variants in Exome Aggregation Consortium Server with a frequency over 0.1% were disregarded. To remove other potential germline variants, the exome data was filtered against data from 2315 Finnish controls (93 individuals from the 1000 Genomes Project, 1941 individuals from The Sequencing Initiative Suomi (SISu) (http://www.sisu.fimm.fi), and 281 in-house control exomes or genomes). Recently, it has been estimated that about 400 control samples remove germline variation (single-nucleotide variants and indels) from a tumor sample at least as efficiently as the matched normal sample [30]. Lastly, all the remaining variants were individually visualized with Rikurator to exclude those only present in the same direction reads as likely artifacts. The functional effects of the variants were predicted by two independent in silico tools: SIFT (http://sift.jcvi.org/) and Polyphen-2 (http://genetics.bwh.harvard.edu/pph2/).
All candidate variants in genes mutated in at least three tumors were validated by direct sequencing. Oligonucleotide primers were designed with Primer3Plus software (http://www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi) (S4 Table). PCR products were sequenced directly utilizing Big Dye Terminator v.3.1 sequencing chemistry (Applied Biosystems) on an ABI3730 Automatic DNA Sequencer.
TP53, ATRX, and DAXX immunolabeling was performed on FFPE sections of all 19 exome-sequenced ULMSs and on a tissue microarray containing 33 additional tumors. For TP53, immunostaining was performed as previously described [31]. For ATRX and DAXX, heat-induced antigen retrieval was carried out in a microwave using citrate buffer (pH 6.0) for 20 min. Endogenous peroxidase blocking was followed by overnight incubation with the primary antibody at 4°C (anti-ATRX 1:500 dilution, Sigma-Aldrich catalog# HPA001906; anti-DAXX 1:500 dilution, Sigma-Aldrich catalog# HPA008736). The primary antibody was detected with DAB Plus Substrate System (Thermo Fisher Scientific catalog# TA-060-HDX). Immunohistochemical scoring was assessed by a pathologist (RB). Only nuclear labeling of the proteins was evaluated. The loss of nuclear staining in tumor cells together with retained expression in non-neoplastic cells (endothelial or inflammatory cells) was considered loss of expression. The scoring was done without knowledge of the clinical outcome data.
Detection of large, abnormally intense, intra-nuclear telomere DNA aggregates via telomere-specific FISH is considered the most sensitive and specific marker for identifying ALT phenotype in fixed tissue samples [13].
FFPE sections were deparaffinized at room temperature with xylene (3x10 min) and 100% EtOH (2x10 min) and air-dried. Subsequently, the slides were rinsed in phosphate-buffered saline (PBS) at 37°C (2x5 min) followed by RNAse A treatment (Sigma-Aldrich catalog# R4642) at 37°C for an hour. After a series of washes at room temperature with saline-sodium citrate (pH 7.0; 3x5 min) and deionized water (2x5 min), the slides were digested with Digest All 3-pepsin (Invitrogen/Life Technologies catalog# 00–3009) at 37°C for 10 min and rinsed with PBS at room temperature (2x5 min). Next, the slides were dehydrated and hybridized with Cy3-labeled peptide nucleic acid (PNA) probe (Panagene Inc. catalog# F1006-5). The denaturation took place at 85°C for 10 min and hybridization in dark at room temperature for an hour. Post-hybridization washes with saline-sodium citrate/0.1% Tween-20 (pH 7.0; 2x10 min) at 55°C and at room temperature for 10 min were followed by nuclear counterstaining with DAPI. The slides were imaged with a Zeiss Axio Imager epifluorescence microscope and image acquisition took place through Hamamatsu Orca Flash 4.0 LT camera and Zen software. The assessment of FISH slides was carried out independently by two authors (NM, MA). ULMSs were classified as ALT-positive if ≥5% of 300 assessed neoplastic cells displayed ALT-associated, abnormally bright telomeric DNA aggregates. In all cases, regions of necrosis and overlapping cells difficult to interpret were excluded from consideration.
Statistical analyses were performed using SPSS statistical software for Windows version 22.0 (SPSS Inc.). Here, survival was defined as overall survival time from the time of diagnosis. Survival curves related to TP53 and ATRX expression were generated using the Kaplan–Meier method, and median survival times with 95% confidence intervals were given. Comparison of survival curves between normal and aberrant expression was performed using the log-rank test. P-value <0.05 was considered statistically significant. Association between TP53 and ATRX expression statuses was evaluated using cross tabulation and Fisher’s exact test.
|
10.1371/journal.pntd.0006773 | Molecular characterization and phylogenetic analysis of dengue viruses imported into Taiwan during 2011-2016 | A total of 1,596 laboratory-confirmed imported dengue cases were identified in Taiwan during 2011–2016. Most of the imported cases arrived from Southeast Asia as well as the Indian subcontinent, the Pacific region, Latin America, Australia and Africa. Phylogenetic analyses of the complete envelope protein gene sequences from 784 imported dengue virus (DENV) isolates were conducted, and the results suggest that the DENV-1 genotype I and DENV-2 Cosmopolitan genotype comprise the predominant serotype/genotype of DENV strains circulating in Southeast Asia. The DENV-1 genotype III, DENV-3 genotype III and DENV-4 genotype I and II strains were found to be newly emerging in several Southeast Asian countries. Our results also showed that geographical restrictions of DENV-1 genotype I, DENV-1 genotype III and DENV-2 Cosmopolitan genotype are becoming blurred, indicating the extensive introductions and continuous expansions of DENV strains between nations in Southeast Asia. In this study, we present the geographic distribution and dynamic transmission of DENV strains circulating in Southeast Asian countries. In addition, we demonstrated local dengue epidemics caused by several imported DENV strains in Taiwan during 2011–2016.
| Dengue is the most prevalent mosquito-borne viral disease in the world. The expansion of dengue viruses to different parts of the world has been accelerated by the increase in worldwide travel and trade. In this study, we present the results of a laboratory-based dengue surveillance in Taiwan during 2011–2016. A total of 1,596 laboratory-confirmed imported dengue cases were identified. The travelers were infected in 29 countries in Southeast Asia, the Indian subcontinent, the Pacific region, Latin America, Australia and Africa. Phylogenetic analyses of the envelope gene sequences of 784 imported dengue virus isolates suggest that the DENV-1 genotype I and DENV-2 Cosmopolitan genotype comprise the predominant serotype/genotype DENV strains circulating in Southeast Asia. Our results also showed that geographical restrictions of some of the DENV genotypes are becoming blurred, indicating the extensive introductions and continuous expansions of DENV strains between countries in Southeast Asia. In addition, we demonstrated dengue outbreaks in Taiwan caused by viruses imported from Asia and the Americas. The DENV envelope gene sequences from this study will contribute to a better understanding of the genetic evolution, dynamic transmission and global expansion of dengue viruses.
| Dengue is the most prevalent mosquito-borne viral infection of humans in tropical and subtropical regions of the world [1]. In recent decades, the incidence of dengue has grown dramatically; approximately half of the world’s population is now at risk [2]. An estimated 390 million dengue infections occur annually, of which 96 million dengue infections manifest clinically [3, 4]. Dengue virus (DENV) belongs to the genus Flavivirus in the family Flaviviridae. The DENV genome consists of a single-stranded, positive-sense RNA, which is of approximately 10,700 nucleotides and contains a long open reading frame that encodes three structural proteins (capsid [C], premembrane/membrane [prM] and envelope [E] proteins) and seven nonstructural (NS) proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) [5, 6].There are 4 genetically and antigenically distinct DENV serotypes (DENV-1 to DENV–4) that cause dengue. DENVs are transmitted to humans through the bite of an infected female Aedes mosquito [7]. Dengue disease can manifest as mild dengue fever or the more severe and potentially fatal dengue hemorrhagic fever or dengue shock syndrome [8, 9].
Dengue is endemic to most countries in Southeast Asia, the Western Pacific region and the Americas, with a very high morbidity rate and disease burden [4, 10]. A large number of cases are reported each year, and all four DENV serotypes currently circulate in hyperendemic countries. The rapid expansion of DENV strains to different parts of the world has been accelerated by the increase in worldwide travel and trade. Studies on DENV infection in travelers can thus provide useful information on the geographic distribution and global movement of DENV [11–14]
Taiwan is an island off the southeastern coast of mainland China in the western Pacific Ocean. The island straddles the Tropic of Cancer, giving it a warm tropical-subtropical climate. Aedes albopictus is found throughout Taiwan, whereas Ae. aegypti is distributed in the south [15]. Dengue is not considered endemic in Taiwan; thus, the close commercial links and air travel between Taiwan and other countries are responsible for the constant importation of multiple DENVs and the outbreaks that occur each year [16, 17]. To reduce the introduction of DENV strains into Taiwan and prevent local epidemics, both passive and active surveillance of DENV infections have been implemented in Taiwan. We previously reported the molecular characterization of DENV strains imported into Taiwan during 2003–2010 [18, 19]. The results provided information on the geographic distribution and dynamic transmission of DENV strains in Southeast Asian countries. In this study, we continued to perform laboratory-based surveillance and provide essential information on the molecular epidemiology of DENV strains circulating in Southeast Asian countries during 2011–2016.
Dengue is a reportable infectious disease in Taiwan, and suspected cases must be reported within 24 hours of clinical diagnosis. To provide effective surveillance, both passive (the hospital-based reporting system) and active (such as fever screening at airports, self-reporting, and expanded screening for contacts of confirmed cases) surveillance systems were implemented by the central and local health departments in Taiwan. Human serum samples of suspected dengue cases were submitted to the Centers for Disease Control, Taiwan (Taiwan CDC), for confirmation of DENV infection. The human serum samples used in this study were derived from confirmed dengue cases submitted to the Taiwan CDC during 2011–2016. All samples analyzed were anonymized. The study protocol was reviewed and approved by the Taiwan CDC Institutional Review Board (IRB 104121). The informed consent requirement was waived by the board. An imported dengue case was defined as a laboratory-confirmed dengue case with travel history to endemic countries within 14 days before the date of onset of dengue. An indigenous case was recorded when no overseas travel was indicated.
DENV infection was defined as a febrile illness associated with the detection of DENV RNA by reverse transcription-polymerase chain reaction (RT-PCR), isolation of DENV by cell culture, detection of DENV nonstructural protein 1 (NS1) antigen, or a seroconversion or at least a four-fold increase in the titer of IgM or IgG antibodies against DENV in paired acute and convalescent serum samples tested by capture IgM and IgG enzyme-linked immunosorbent assays (ELISA). Isolation of DENV was performed using a mosquito cell line (clone C6/36 of Ae. albopictus cells) as previously described [18]. Briefly, for each acute-phase serum sample, 50 μL of the sample diluted at ratios of 1:20, 1:40, 1:80, and 1:160 with RPMI 1640 medium (Gibco/BRL, Life Technologies, Auckland, New Zealand) containing 1% fetal calf serum was added to a 96-well microtiter plate. Then, 1x105 cells/100 μL/well of C6/36 were added to the microtiter plate and incubated for 7 days at 28°C. Cells were harvested, and infection was confirmed by immunofluorescence assay using dengue serotype-specific monoclonal antibodies, including 5F3-1 (DENV-1 specific, ATCC HB-47), 3H5-1 (DENV-2 specific, ATCC HB-46), 5D4-11 (DENV-3-specific, ATCC HB-49) and 1H10-6 (DENV-4-specific, ATCC HB-48). The viruses were subcultured in C6/36 cells and harvested for nucleotide sequencing after the first or second passage. Isolated viruses were identified using the nomenclature of serotype/country of origin/strain/year of isolation. To detect and differentiate DENV serotypes in acute-phase samples, we performed one-step, SYBR Green I-based, real-time RT-PCR (QuantiTect SYBR Green RT-PCR kit, Qiagen, Hilden, Germany) using the LightCycler 96 Real-Time PCR System (Roche Diagnostics, Mannheim Germany). Real-time RT-PCR was performed using two sets of consensus primers, one primer set targeting a region of the nonstructural protein 5 (NS5) genes to detect all of the flaviviruses and the other primer set targeting a region of the C gene to detect all of the DENV serotypes. The DENV serotypes of the positive samples were then confirmed by DENV serotyping using four sets of serotype-specific primers targeting the C gene [20]. A commercial DENV NS1 Ag strip rapid test kit (Bio-Rad Laboratories, Marnes La Coquette, France) and SD Dengue NS1 Ag test (Standard Diagnostics, Inc. Kyonggi-do, Korea) were used to detect the DENV NS1 antigen in serum samples. Envelope (E)/Membrane (M)-specific capture IgM and IgG ELISA were used to detect DENV-specific IgM and IgG antibodies as previously described [21].
Viral RNA was extracted from acute-phase serum samples or the culture supernatant of C6/36 cells infected with each of the isolated DENV strains using a QIAamp Viral RNA Mini Kit (QIAGEN, Hilden, Germany). Primers used for amplification and sequencing of C, prM and E gene sequences of DENVs were described previously [17] [18]. The RT-PCR reaction was carried out with the SuperScript III One-Step RT-PCR system with Platinum Taq High Fidelity (Invitrogen). The cDNA synthesis step was performed at 55°C for 30 min; PCR at 94°C for 2 min; 40 cycles of 94°C for 15 sec, 50°C for 30 sec, and 68°C for 1 min; and prolonged elongation at 68°C for 5 min. PCR products were purified using a Qiagen QIA quick Gel Extraction Kit (QIAGEN). Nucleotide sequences were determined by an automated DNA sequencing kit and an ABI Prism 3730XL DNA sequencer (Applied Biosystems, Foster City, CA) according to the manufacturer’s protocols. Overlapping nucleotide sequences were combined for analysis and edited with the Lasergene software package (DNASTAR Inc, Madison, WI). Nucleotide sequences of the complete E gene of the DENV strains described in this study were submitted to GenBank with the following accession numbers: 334 DENV-1 strains (KT175076-KT175078, KT175082-KT175101, KT175103-KT175110, KU365900, KY496854, KY496855, and MG894671-MG894970), 234 DENV-2 strains (KT175111-KT175140, KU365901, and MG894971-MG895173), 133 DENV-3 strains (KP176703-KP176710, KP175715, MG895174-MG895297), and 99 DENV-4 strains (MG895298-MG895396). All the strain identifiers and their accession numbers are shown in the S1 Table.
The nucleotide sequences of the complete E gene of 784 imported and 16 epidemic strains in Taiwan in combination with sequences of epidemic strains from Southeast Asian countries and various global reference strains of different genotypes available from GenBank were analyzed. In addition, sequences representing the most closely related to the epidemic strains in Taiwan obtained using BLAST were selected for phylogenetic analyses. Sequences of DENV strains were aligned, edited and analyzed using Clustal W software [22]. The phylogenetic analysis was performed using MEGA version 7 (http://www.megasoftware.net/) [23]. To construct the phylogenetic trees, the maximum likelihood method using the general time reversible as a substitution model and the neighbor-joining method using the maximum composite likelihood as a substitution model were utilized. The reliability of the analysis was evaluated by a bootstrap test with 1,000 replications. Sequences of D2/New Guinea/NGC/1944 strain/M29095, D2/Senegal/DAKHD10674/1970/AF231720, D1/USA/Hawaii/1945 strain/AF425619 and D2/New Guinea/NGC/1944 strain/M29095, were used as outgroups to root the tree of the DENV-1, DENV-2, DENV-3 and DENV-4 strains, respectively.
A total of 1,596 laboratory-confirmed imported dengue cases (both visitors to Taiwan and local returning residents) were identified in Taiwan during 2011–2016. Among them, 703 cases (44.0%) were identified by fever screening at airports (Table 1) and most (>90%) of these cases were in their viremic stages with positive real-time RT-PCR and negative IgM and IgG results. Most cases arrived from Southeast Asian countries, with Indonesia (24.8%, 396 cases), the Philippines (19.2%, 306 cases), Malaysia (14.2%, 226 cases), Thailand (12.0%, 192 cases), and Vietnam (12.0%, 191 cases) being the most frequent country sources of importation. Cases were also imported from other Asian countries (16.6%, 265 cases, including Myanmar, Singapore, Cambodia, India, China, Bangladesh, Maldives, Sri Lanka, Laos, Saudi Arabia, and Japan), the Pacific region (0.7%, 11 cases, including Palau, Papua New Guinea, Nauru, Fiji, Solomon Islands, Tuvalu, and French Polynesia), Australia (0.1%, 2 cases), Latin America (0.4%, 7 cases, including Brazil, Costa Rica, and Saint Lucia), and Africa (0.1%, 2 cases, one from South Africa and the other from Kenya). Comparing the numbers of imported dengue cases between 2003–2010 and 2011–2016, we found there is an increasing trend of imported cases from the Philippines, Malaysia and Singapore during the study period. Fig 1 shows the country sources of importation of DENVs in Taiwan during 2011–2016. Fig 2 shows the number of imported dengue cases in Taiwan during 2003–2016.
From the 1,596 imported dengue cases, 380, 303, 171 and 114 cases were determined to be infected with DENV-1, DENV-2, DENV-3, and DENV-4, respectively. Table 1 summarizes serotype and genotype distributions of imported DENV strains from 29 countries. The serotype distributions of DENV strains imported from the most common Southeast Asian countries each year during 2003–2016 are shown in Fig 3. Yearly changes in serotype distribution were observed, and all four serotypes of DENV were found to circulate in each of these countries during 2011–2016. The number of imported dengue cases from Malaysia increased sharply during 2014–2016, and the main serotypes were DENV-1 and DENV-2. The main serotype of imported DENV strains from Vietnam shifted from DENV-1 during 2007–2010 to DENV-2 during 2012 and 2015 and then back to DENV-1 during 2016. The number of imported dengue cases from Singapore increased significantly during 2013–2016, and the main serotypes in recent years have been DENV-1 and DENV-2. A relatively high number of imported cases was observed from Myanmar in 2015, and DENV-1, DENV-2 and DENV-4 were the main serotypes. All 4 serotypes of DENV were found to cocirculate in Cambodia during 2015–2016.
Among the 1,596 imported dengue cases, 784 DENV strains were isolated from acute-phase serum samples of patients infected in 23 countries (Table 1). Phylogenetic analyses of the E gene sequences of imported DENV strains were conducted to determine the genotype and genetic relationship of these viral strains. The designations of DENV genotypes are based on the classification of A-Nuegoonpipat et al. [24], Twiddy et al. [25], Lanciotti et al. [26], and Klunthong et al. [27] for the DENV-1, DENV-2, DENV-3 and DENV-4 strains, respectively. The genotype distributions of the DENV-1 to DENV-4 strains imported from the 8 most common Southeast Asian countries during 2003–2016 are shown in Figs 4–7, respectively.
Fig 4 shows genotype distributions of imported DENV-1 strains. The DENV-1 strains obtained from Asia and the Pacific can be classified into three genotypes (I, II and III). Genotype I contains the majority of strains from Asia, while genotype II comprises a smaller set of Asian and Pacific strains. Genotype III contains viruses from a wide geographic area [24]. Genotype I of DENV-1 was the predominant genotype imported from Southeast Asian countries, including Indonesia, Malaysia, Vietnam, Thailand, Myanmar and Cambodia. Before 2006, genotype II of imported DENV-1 was the main genotype in Indonesia; however, since 2007, the main genotype has shifted to genotype I. Genotype II was the predominant genotype of imported DENV-1 from the Philippines. The numbers of genotype III of imported DENV-1 strains from Malaysia and Singapore increased during 2013–2014.
Fig 5 shows genotype distributions of imported DENV-2 strains. The DENV-2 strains can be classified into six genotypes. The Cosmopolitan genotype has a wide geographic distribution. The Asian genotype 1 and 2 contain viruses from Asia, and the Asian/American genotype comprises viruses from Southeast Asia and Latin America. The American genotype consists of viruses from Latin America and older isolates collected from Indian subcontinent and the Pacific, and the Sylvatic genotype contains sylvatic strains from Asia and Africa [25]. The Cosmopolitan genotype was the predominant genotype of imported DENV-2 strains from Indonesia, the Philippines, Malaysia, and Singapore, and Asian genotype 1 was the main genotype of imported DENV-2 strains from Vietnam, Thailand, Myanmar and Cambodia.
Fig 6 shows genotype distributions of imported DENV-3 strains. The DENV-3 strains can be classified into four genotypes. Genotype I consists of viruses from Indonesia, Malaysia, the Philippines, the Pacific islands and Australia. Genotype II contains viruses from Southeast Asia. Genotype III has a wide geographical distribution which includes Asia, Africa and Latin America. Genotype IV consists of viruses from Puerto Rico and the 1965 Tahiti virus isolates [26]. Genotype I was the predominant genotype of imported DENV-3 strains from Indonesia and the Philippines. In Malaysia, the number of imported genotype III strains increased between 2015 and 2016. Genotype II was the main genotype of imported DENV from Vietnam, Myanmar and Cambodia. In Thailand, the main detected genotype has shifted from genotype II to genotype III in recent years.
Fig 7 shows genotype distributions of imported DENV-4 strains. The DENV-4 strains were separated into four genotypes. Genotype I contains viruses from Asia. Genotype II consists of viruses from Asia, the Pacific and Latin America. Genotype III contains viruses from Thailand and genotype IV contains sylvatic strains from Malaysia [27]. Genotype I was the predominant genotype of imported DENV-4 from Vietnam, Thailand, Myanmar and Cambodia, whereas Genotype II was the main genotype from Indonesia and Malaysia. In the Philippines, the main detected genotype shifted from genotype I during 2003–2009 to genotype II during 2010–2016.
We first made trees for all E gene sequences of the imported and epidemic strains in Taiwan in combination with sequences of epidemic strains from Southeast Asian countries and various global reference strains of different genotypes available from GenBank. In addition, sequences representing the most closely related to the epidemic strains in Taiwan obtained using BLAST were selected for phylogenetic analyses. The results are shown in S1 Fig–S4 Fig for DENV-1 to DENV-4, respectively. The representative E gene sequences based on country source of importation and date of sample collection, were selected to build the trees in Figs 8–11.
Except for the Philippines, most of the DENV-1 strains isolated from imported cases from Southeast Asian countries belonged to genotype I (Table 1 and Fig 8). Imported DENV-1 genotype I strains from Indonesia and Malaysia showed a high degree of genetic diversity and strains in different lineages that were co-circulating in these countries. Some of the strains from Thailand, Singapore, Myanmar, Laos and China were clustered with strains from Indonesia and Malaysia. Viral strains from Cambodia were closely related to viruses from Vietnam and Thailand, whereas viral strains from Myanmar and Sri Lanka were clustered with viruses from Thailand. Genotype II contained imported viral strains from the Philippines, Malaysia and Indonesia. Genotype III contained imported viral strains from diverse geographical regions, including Asia (Singapore, Bangladesh, Malaysia, Maldives, Thailand, China and India) and the Americas (USA and Costa Rica). It is interesting to note that the strains for DENV-1 tend to be less geographically clustered than in the other serotypes.
The DENV-2 strains isolated from imported cases during 2011–2016 fell into two genotypes, the Cosmopolitan genotype and Asian genotype 1 (Table 1 and Fig 9). The Cosmopolitan genotype strains from imported cases can be divided into three clusters. Cluster 1 contains viral strains from Malaysia, Singapore and Indonesia. Some of the imported viral strains from Thailand, Maldives, Vietnam and China also fell into this cluster. Cluster 2 contains imported viral strains from the Philippines, Tuvalu and Palau. Cluster 3 contains imported strains from India, Saudi Arabia and Kenya. A strain from Thailand and two strains from Vietnam were also found to cluster with strains from India. Asian genotype 1 contains viral strains from Thailand, Vietnam, Cambodia, Lao, and Myanmar. Imported strains from Malaysia also fell into this genotype. No Asian/American genotype and Asian genotype 2 strains were found among imported cases during 2011–2016.
The DENV-3 strains isolated from imported cases fell into three genotypes, genotype I, II and III (Table 1 and Fig 10). Genotype I can be divided into two clusters: one contains viral strains from Indonesia, Malaysia, Singapore and Solomon Islands, and the other contains viral strains from the Philippines. Genotype II contains imported strains from Vietnam, Thailand, Cambodia, and Laos. Genotype III contains viral strains from diverse geographical localities, including India, Singapore, Malaysia, Thailand, Vietnam and Cambodia.
The DENV-4 strains isolated from imported cases fell into two genotypes, genotype I and II (Table 1 and Fig 11). Genotype I contains two major clusters: one cluster contains viral strains from the Philippines, and the other contains viral strains from Vietnam, Thailand, Myanmar and Cambodia. In 2016, an imported strain from the Maldives also fell into this cluster and was closely related to virus strains from Sri Lanka. Genotype II contains imported viral strains from the Philippines, Indonesia, Malaysia and Singapore. Imported viral strains from Papua New Guinea and Brazil also belonged to this genotype.
Table 2 lists the major dengue outbreaks and epidemic DENV strains circulating in Taiwan during 2011–2016. Our results showed that a DENV-1 strain (D1/Taiwan/700TN1109a/2011) caused outbreaks in southern Taiwan during 2011–2013. This strain belongs to genotype III of DENV-1 and is closely related to viral strains from the Americas. This is the first time that an American DENV strain caused an epidemic in Taiwan. In 2011, the other 3 epidemic strains, DENV-1 (D1/Taiwan/111TP1110a/2011), DENV-2 (D2/Taiwan/802KH1108c/2011) and DENV-3 (D3/Taiwan/811KH1109a/2011), caused outbreaks in Taipei City, Kaohsiung City and Penghu County. These strains were likely introduced from Myanmar and Vietnam. In 2012, in addition to the epidemic strain D1/Taiwan/700TN1109a/2011, the other four DENV strains (D1/Taiwan/234NP1209a/2012, D2/Taiwan/802KH1208a/2012, D3/Taiwan/832KH1210a/2012 and D4/Taiwan/811KH1207a/2012) caused outbreaks in New Taipei City and Kaohsiung City. These strains were likely introduced from Cambodia, Indonesia, Thailand and the Philippines. In 2013, in addition to the epidemic strain D1/Taiwan/700TN1109a/2011, which was transmitted to Pingtung County, the other three strains (D2/Taiwan/920PT1306a/2013, D2/Taiwan/900PT1308a/2013 and D3/Taiwan/932PT1305b/2013) caused outbreaks in southern Taiwan. These strains were likely introduced from Indonesia. During 2014–2015, there was a large outbreak caused by a DENV-1 strain (D1/Taiwan/806KH1405a/2014) in southern Taiwan; this epidemic strain belonged to genotype I and is closely related to virus strains from Indonesia. In 2014, a DENV-2 strain (D2/Taiwan/807KH1411a/2014) also caused a small outbreak in Kaohsiung City. In 2015, a DENV-2 strain (D2/Taiwan/704TN1505a/2015) caused a large outbreak in Tainan City and later in Kaohsiung City, this strain belonged to the Cosmopolitan genotype and is closely related to strains from Indonesia. From January to April 2016, a total of 372 indigenous cases were identified. These cases represented the last wave of the 2015 outbreak in southern Taiwan. In 2016, there were only 8 indigenous cases identified between May and December in Taiwan. A DENV-1 strain (D1/Taiwan/114TP1611a/2016) caused a small outbreak in Taipei City in November 2016; this strain belonged to Genotype II and is closely related to strains from the Philippines. (Figs 8–11)
A total of 1,596 laboratory-confirmed imported dengue cases were identified in Taiwan during 2011–2016, most of which arrived from Asian countries and other regions, including the Pacific Islands, Australia, Africa and the Americas. Among them, 92.7% of cases (1,480 cases) arrived from the eight most common country sources of importation: Indonesia, the Philippines, Malaysia, Thailand, Vietnam, Myanmar, Singapore and Cambodia. An analysis of imported DENV strains from these countries showed changes in serotype distributions during the study period. As expected, all 4 serotypes of DENV were found to cocirculate in each of the most common country sources of importation during 2011–2016.
A total of 784 DENV strains, namely, 329 DENV-1, 227 DENV-2, 130 DENV-3, and 98 DENV-4 strains, were isolated during 2011–2016. Phylogenetic analyses of E gene sequences of imported DENV strains suggested that genotype I of DENV-1 and the Cosmopolitan genotype of DENV-2 were the predominant DENV strains imported from Southeast Asian countries during 2011–2016. Notably, genotype III of the DENV-1 strain was found to newly emerge in Malaysia, Vietnam, Thailand and Singapore. In addition, genotype III of the DENV-3 strain also emerged in Malaysia, Thailand and Singapore. However, Asian genotype 2 and the Asian/American genotype of the DENV-2 strain were not found in imported cases from Southeast Asian countries in the last decade, suggesting a low prevalence of these two genotypes in this region.
Previous studies have shown that DENV-1 genotype I was the predominant DENV genotype circulating in Southeast Asian countries [18, 19, 28–32]. In our study, the numbers of imported DENV-1 genotype I strains increased sharply in Indonesia and Malaysia and became the predominant genotype in the last decade. Phylogenetic analysis of E gene sequences of imported DENV-1 genotype I strains from Indonesia, Malaysia, Thailand and Vietnam showed a high degree of genetic diversity. Interestingly, we found that most of the DENV-1 genotype I strains did not segregate into a distinct clade in each country but that viral strains from Indonesia, Malaysia and a few strains from Singapore, Thailand, and Laos were clustered together. In addition, most of the imported DENV-1 genotype I strains from Thailand, Myanmar, Cambodia, China and Sri Lanka formed another cluster [33–35]. A DENV-1 genotype I strain (D1/Laos/1508aTw/2015) isolated from a case imported from Laos in 2015 is closely related to virus strains from Malaysia. Although only a few DENV strains have been isolated from imported cases from Laos in Taiwan during 2003–2016, genotype distribution of these imported DENV strains within each serotype is consistent with the results of a study by Castonguay-Vanier et al. [36]. Genotype III of DENV-1 strains imported from China, Vietnam, Malaysia, Singapore and Bangladesh were clustered together. The results suggest a close genetic relationship and frequent transmission of DENV-1 among Southeast Asian countries and may reflect frequent trade and travel between these countries [37, 38].
Genotype distribution of DENV-2 in Southeast Asian countries remains largely unchanged in the last decade. However, it is interesting to note that the number of DENV-2 Cosmopolitan genotype strains imported from Vietnam and Thailand increased and that Asian genotype I strains were found in Malaysia, indicating that these two genotypes of DENV-2 have expanded into new territories. Imported DENV-2 strains from the Maldives during 2015–2016 belonged to the Cosmopolitan genotype and are closely related to virus strains from Malaysia and Singapore. A DENV-2 Cosmopolitan genotype strain (D2/Palau/1612aTw/2016) isolated from a case imported from Palau is closely related to virus strains from the Philippines, Papua New Guinea and Fiji, suggesting cocirculation of these virus strains among these countries.
Except for Malaysia and Thailand, the genotype distribution of DENV-3 strains imported from Southeast Asian countries remains largely unchanged. It is interesting to note that the genotype of DENV-3 strains from Malaysia shifted from genotype I to genotype III. In addition, the genotype of DENV-3 strains from Thailand shifted from genotype II to genotype III in recent years. Recent studies have shown that the DENV-3 genotype III strains are emerging in Asian countries [39–41]. In this study, the number of DENV-3 genotype III strains has been increasing in Malaysia and Thailand, and these strains were clustered together with strains from Cambodia, Vietnam and Singapore.
Although a relatively low prevalence of DENV-4 strains was found in Southeast Asia [42], there was an increase in the numbers of DENV-4 strains imported from the Philippines, Myanmar and Cambodia, during 2011–2016. A DENV-4 genotype II strain (D4/Papua New Guinea/1608aTw/2016) isolated from a case imported from Papua New Guinea in 2016 is closely related to virus strains from Indonesia. Interestingly, we found that two DENV-1 genotype I strains (D1/Maldives/1604aTw/2016 and D1/Maldives/1608aTw/2016), a DENV-2 Cosmopolitan genotype strain (D2/Maldives/1612aTw/2016) and a DENV-4 genotype I strain (D4/Maldives/1605aTw/2016) were cocirculating in the Maldives in 2016, indicating multiple introductions of DENV strains into the Maldives in the recent dengue epidemic.
We previously reported that the genetic relationship and genotype distribution of DENV depend largely on the geographical location [18, 19]; however, our study demonstrates that geographical restrictions of DENV genotypes are becoming blurred. For example, DENV-1 genotype I, DENV-1 genotype III and DENV-2 Cosmopolitan genotype strains from different geographical locations or countries were clustered together, indicating the extensive introductions and continuous expansions of DENV strains between nations in Southeast Asia.
Due to the small sample size of imported cases from country sources of importation in Taiwan, the potential limitations of our study include: (1) Tourists/travelers may primarily travel to specific locations within a country, and thus, viruses circulating within other regions may not be represented often in our dataset; (2) Potentially changing patterns in travel (i.e. the number of travelers between Taiwan and country sources of importation and the purpose of travel) may affect the numbers of imported dengue cases and the representativeness of our analyses; (3) Because of the limited epidemiological data from country sources of importation, it is unclear whether the serotype/genotype dynamics described in this study are representative of the epidemiological situation within country.
Because air travel has become increasingly popular and convenient, DENV strains will be transmitted by travelers to other countries, sometimes leading to extensive outbreaks. During 2011–2016, several dengue outbreaks occurred in Taiwan, and most epidemic strains were introduced from neighboring Southeast Asian countries, including Indonesia, the Philippines, Vietnam and Myanmar (Table 2). In accordance with our previous studies, the patterns of imported DENV strains observed among the travelers are connected to the overall patterns of the dengue dynamics in Taiwan [19, 43]. It is worth noting that an epidemic DENV-1 strain (D1/Taiwan/700TN1109a/2011), which caused the dengue outbreak in southern Taiwan for three consecutive years (2011–2013), is closely related to virus strains from Central America. This is the first time that a dengue outbreak in Taiwan was caused by an American strain and suggests not only that imported DENV strains from neighboring Southeast Asian countries cause local outbreaks but also that viral strains introduced from the Americas or other continents may establish transmission chains within Taiwan.
In this study, we conducted molecular epidemiological analyses to monitor DENV serotype and genotype distributions and dynamic movements in Southeast Asian countries. The DENV strains isolated from imported dengue cases and the availability of a DENV genome sequence database can provide essential information on the global expansion and genetic evolution of DENV, which is useful for disease surveillance, laboratory diagnoses, pathogenesis investigation and vaccine development. Our results also indicate that it is important to reinforce active surveillance and travel and border health measures for dengue prevention and control in Taiwan.
|
10.1371/journal.pntd.0003411 | Regulation of Leishmania (L.) amazonensis Protein Expression by Host T Cell Dependent Responses: Differential Expression of Oligopeptidase B, Tryparedoxin Peroxidase and HSP70 Isoforms in Amastigotes Isolated from BALB/c and BALB/c Nude Mice | Leishmaniasis is an important disease that affects 12 million people in 88 countries, with 2 million new cases every year. Leishmania amazonensis is an important agent in Brazil, leading to clinical forms varying from localized (LCL) to diffuse cutaneous leishmaniasis (DCL). One interesting issue rarely analyzed is how host immune response affects Leishmania phenotype and virulence. Aiming to study the effect of host immune system on Leishmania proteins we compared proteomes of amastigotes isolated from BALB/c and BALB/c nude mice. The athymic nude mice may resemble patients with diffuse cutaneous leishmaniasis, considered T-cell hyposensitive or anergic to Leishmania´s antigens. This work is the first to compare modifications in amastigotes’ proteomes driven by host immune response. Among the 44 differentially expressed spots, there were proteins related to oxidative/nitrosative stress and proteases. Some correspond to known Leishmania virulence factors such as OPB and tryparedoxin peroxidase. Specific isoforms of these two proteins were increased in parasites from nude mice, suggesting that T cells probably restrain their posttranslational modifications in BALB/c mice. On the other hand, an isoform of HSP70 was increased in amastigotes from BALB/c mice. We believe our study may allow identification of potential virulence factors and ways of regulating their expression.
| Leishmaniasis is an important disease that affects 12 million people in 88 countries. Leishmania amazonensis is an important agent of leishmaniasis in Brazil, leading mainly to localized (LCL) and sometimes to diffuse cutaneous leishmaniasis (DCL), depending on the host immune response to infection. We believe that host immune response affects not only the clinical form and survival of Leishmania, but also the phenotype and virulence of the parasite. To analyze the effects of the host on Leishmania phenotype, we compared protein expression (proteome) of parasites isolated from wild type mice and from mice lacking T cells. We identified some protein isoforms differentially expressed, which may further be studied as potential virulence factors.
| Leishmaniasis is an important disease that affects 12 million people in 88 different countries in Europe, Africa, Asia and America, and 2 million new cases are reported every year (WHO 2004; [1,2]. There are different forms of tegumentary and visceral leishmaniasis, that depend on the Leishmania species and on the genetic/immunologic status of the host, all transmitted to man by the bite of naturally infected species of phlebotomine sand flies [3].
In Brazil, Leishmania braziliensis and Leishmania amazonensis are considered the main pathogenic species causing human tegumentary leishmaniasis [4]. The human L. amazonensis infection may lead to different clinical forms, varying from the localized cutaneous leishmaniasis (LCL), with moderate cellular hypersensitivity, to the diffuse cutaneous leishmaniasis (DCL), frequently associated to anergy to parasite’s antigens [4,5].
The murine model has been commonly used to analyze several aspects of Leishmania infection such as the virulence of different parasite species [3,6] and how different mouse strains respond to the same Leishmania species [7,8,9,10]. The infection of mice by Leishmania major has been the most commonly used model, and allowed the definition of resistant and susceptible lineages such as C57BL/6 and BALB/c, which mount Th1 and Th2 responses, respectively [11,12]. In infections by L. amazonensis the dichotomy of susceptible and resistant mice is not evident. In fact, most lineages are susceptible to this Leishmania species [3,13] and develop a mixed Th1-Th2 response to the parasite, producing IL-4 and IFNγ [6,11]. However, some differences can be observed in the progression and size of lesions according to the strain [7,8]. The low and mixed Th1/ Th2 responses seen in L. amazonensis-infected mice are similar to those observed in human infections [4], validating the biological relevance of these mouse models to study the human disease [14]. The response to Leishmania infection in athymic nude mice, however, has not been thoroughly analyzed. Nude mice of C57BL/6 background have been shown not to develop lesions when infected by L. amazonensis, and had an expected reduced influx of T cells and monocytes in the site of infection [15]. No similar analysis has been performed to date in BALB/c nude mice. These observations suggest that immunopathology of Leshmania infections should be better characterized.
One interesting issue difficult to study in human infections and rarely analyzed in mouse model is how the host immune response affects Leishmania phenotype and virulence. One example of modulation already described is phosphatidylserine (PS) exposure in L. amazonensis amastigotes. The display of PS in the external membrane is an apoptotic feature that leads to parasite intracellular survival due to inhibition of macrophage inflammatory response [16,17]. It has been shown, that the host immune response modulates PS exposure by L. amazonensis amastigotes so that parasites derived from the more susceptible BALB/c mice display more PS than parasites derived from less susceptible C57BL/6 mice [17]. Accordingly, PS exposure was positively correlated with clinical parameters of the human infection (number of lesions and time of disease) and with characteristics of the experimental infection such as macrophage infection and anti-inflammatory cytokine induction [18].
Other amastigote molecules besides PS are certainly modulated by the host immune response. Since Leishmania and other trypanosomatids lack a conventional network of transcription factors and most genes are constitutively transcribed [19,20], most changes in Leishmania phenotypes are better studied in terms of proteins [21].
The analysis of cell proteomics is an efficient method to compare protein profiles. Most studies of Leishmania proteomes compared abundance or post-translational modifications (specially phosphorylation) of proteins in amastigotes and promastigotes of the same Leishmania species [20,22,23,24,25,26], parasites sensitive and resistant to drugs [27,28,29,30], and proteins from different Leishmania species [31,32]. Some works also analyzed immunogenic proteins [22,33,34] and secreted proteins [34,35,36,37]. Due to the difficulty to obtain robust amounts of virulent amastigotes from infected animals for in vitro analysis, most works have used axenic amastigotes for proteomics analysis [23,26,35]. However, comparison of proteomes from lesion derived amastigotes and axenic amastigotes have shown important differences among them [38,39].
Aiming to evaluate the effect of the host immune system on protein expression in Leishmania, we analyzed proteomics of mouse lesion-derived amastigotes employing the protein separation by two-dimensional electrophoresis with fluorescent labeling (DIGE) and protein identification by mass spectrometry (MALDI-ToF/ToF) approach. We compared the proteomes of amastigotes isolated from BALB/c and BALB/c nude mice. The immune system of nude mice, in which T lymphocytes are nearly absent [40], may shed some light on the immune system of patients bearing the diffuse cutaneous leishmaniasis, who are anergic individuals [4].
Promastigotes of Leishmania amazonensis LV79 strain (MPRO/BR/72/M 1841-LV-79) were cultured at 24°C in Warren medium with 10% FCS. Parasites were subcultured every 7 days for 2x106/mL.
Four to 8-week-old female BALB/c and BALB/c nude female mice were bred under specific- pathogen free conditions at the Isogenic Mouse Facility of the Parasitology Department, University of São Paulo, Brazil. Mice were infected in one of the hind footpads with 2 x 106 stationary-phage promastigotes of L.amazonensis strain LV79 (MPRO/BR/72/M 1841-LV-79). Footpad thickness was measured weekly using a caliper.
All animals were used according to the Brazilian College of Animal Experimentation (CONEP) guidelines, and the protocols were approved by the Institutional Animal Care and Use Committee (CEUA) of the University of São Paulo (protocol number 001/2009).
Thirteen weeks after infection the animals were sacrificed and the lesions were removed under sterile conditions. Amastigote isolation was performed as described by Wanderley et al., 2006. Briefly, lesions were minced and homogenized in 5ml PBS using a tissue grinder (Thomas Scientific). After centrifugation at 50 x g for 10 min at 4°C, the supernatant was recovered and centrifuged at 1450 x g for 17 min at 4°C. Supernatant was then removed and the pellet was washed three times with PBS followed by centrifugations at 1450 x g for 17 min at 4°C. After 3h incubation under rotation at room temperature to liberate endocytic membranes (8), amastigotes were further centrifuged, resuspended in 2mL of erythrocyte lysis buffer (155mM NH4Cl, 10mM KHCO3, 1mM EDTA, pH7,4) and incubated for 2min in ice.
This method has been previously validated for isolation of amastigotes free from macrophage contaminants (Balanco et al., 2001).
For lysis parasites were washed twice in PBS, resupended at 109 cells/300μl in PBS+Proteoblock 1x (Fermentas) and lysed by 8 cycles of freeze thaw in liquid nitrogen-42°C. Soluble proteins were obtained after centrifugation at 12.000xg for 3 minutes, concentrated for ~5mg/ml using Microcon 5K (Millipore) and quantified by Bradford (Biorad).
50 μg of extracts (adjusted to pH 8–9) of amastigotes from BALB/c or BALB/c nude were labeled by “minimal labeling” with 1ul (400pmol) of N-hydroxysuccinimidyl-ester-derivates of the cyanine dyes Cy3 or Cy5 (GE Healthcare) for 30min on ice. The reaction was quenched with 1ul of 10 mM lysine for 10 min on ice. 50 μg of a pool of all samples was similarly labeled with Cy2 (GE Healthcare) as an internal standard. The three differently labeled extracts were pooled and incubated with Immobiline DryStrips pIs 4–7 (linear, 13cm, GE Healthcare) in the presence of 7M urea, 2M tiourea, 15mM TrisHCl, 2% CHAPS, 0,5% IPG 4–7 and 3μl of DeStreak reagent for 16h. This procedure was performed for the preparation of analytical gels. Furthermore, for subsequent identification of proteins, a preparative gel was also performed, applying a sample containing 500μg of protein from the Pool (450 ug of unlabeled protein and 50 ug labeled protein with Cy2) to the IPG strip by rehydration. Isoelectric focusing was performed at: 300V for 4h, 500V until 0,5kVh, 1000V until 0,8kVh, 8000V until 18,7kVh, 300V for 2h, at a maximum current of 50μA/strip. Focused IPG strips were incubated for 15 min in equilibration solution (75 mM Tris-HCl, pH 8.8, 6 M urea, 30% glycerol, 2% SDS) with 10mg/mL DTT and then the proteins were alkylated for further 15 min in equilibration solution containing 25mg/mL iodoacetamide. Strips were transferred to 12% SDS-PAGE gel and eletrophoresis was performed at 30mA for 30min and 60mA for the remaining time. Directly after the second dimension, the fluorescent gels were scanned. The preparative gel was fixed in a solution containing 40% methanol and 10% acetic acid, and then stained with Deep Purple stainer according to the manufacturer’s instructions (GE Healthcare). The gels were kept in a solution containing 1% citric acid.
For 2D Western blots, 120μg of extracts were used in each 7cm pI 4–7 Strip (GE healthcare). Strips were rehydrated in DeStreak solution containing one of the samples for 16h. Focusing was performed as recommended for these strips and equilibration and alkylation were done as described above. Second dimension was done in 12% SDS-PAGE gel and Western blot was performed as described below.
The gels containing samples labeled with fluorophores were scanned using "Typhoon 9410 Variable Mode Imager” (GE Healthcare), with the following parameters: Cy2, 488nm excitation and 520nm-BP 40nm emission filters; Cy3, 532nm excitation and 580nm-BP 40nm emission filters; Cy5, 633nm excitation and 670nm-BP 40nm emission filters. For Deep Purple-stained gels, 532nm excitation and 610nm-BP 30nm emission filters were used. The gels were scanned with resolution 100 micra and the sensitivity ranged from 450 to 550 PTM.
The analysis of differential protein expression using the DIGE technique was performed using the program "DeCyder Differential Analysis (GE Healthcare). The volume of each spot was normalized in relation to the total volume of spots selected for that labeling (sample), and the gels were normalized together using the image of the pool of samples labeled with Cy2. Statistical analysis, through the One Way Anova and the Student’s test, were performed to compare protein expression in different samples. Spots were considered to be differentially expressed if p<0.05. We considered a protein to be differentially expressed if one or more of the associated spots were differentially expressed.
Selected spots were collected automatically using Ettan Workstation (GE Healthcare), transferred to 96 well plates and kept at -20°C until shipping. Samples were analyzed at Institut Pasteur of Montevideo. Peptide mass fingerprinting was carried out by in-gel trypsin treatment (Sequencing-grade Promega) overnight at 37°C. Peptides were extracted from the gels using 60% acetonitrile in 0.2% TFA, concentrated by vacuum drying and desalted using C18 reverse phase micro- columns (OMIX Pippete tips, Varian). Peptide elution from micro-column was performed directly into the mass spectrometer sample plate with 3 μl of matrix solution α-cyano-4-hydroxycinnamic acid in 60% aqueous acetonitrile containing 0.2% TFA). Mass spectra of digestion mixtures were acquired in a 4800 MALDI-TOF/TOF instrument (Applied Biosystems) in reflector mode and were externally calibrated using a mixture of peptide standards (Applied Biosystems). Collision-induced dissociation MS/MS experiments of selected peptides were performed. Proteins were identified by NCBInr database searching with peptide m/z values using the MASCOT program and using the following search parameters: monoisotopic mass tolerance, 0.05 Da; fragment mass tolerance, 0.25 Da; methionine oxidation, as possible modifications; and one missed tryptic cleavage allowed.
BALB/c and BALB/c nude female mice were infected with Leishmania amazonensis as described above. Thirteen weeks after infection, mice were euthanized and spleens, popliteal lymph nodes, and paws were harvested for single cell suspension preparations and flow cytometry analysis. Uninfected animals with similar ages were used as controls. All tissues were mechanically dissociated in MTH (Mouse Tonic Hanks: 1x HBSS, 15 mM HEPES pH 7.4, 0.5 U/ml DNase I, 5% fetal bovine serum). Splenocytes were treated with hypotonic buffer for red cell lysis. After washing and counting, 106 cells from each tissue were aliquoted, incubated with Fc Block (BD Biosciences, San Jose, CA) for 10 min in ¼ of the final staining volume. Antibodies were added to blocked cell suspensions and incubated for 20 min, on ice. After washing in MTH, cells were fixed in 4% formaldehyde in PBS. Antibodies used for cell labeling were: FITC-conjugated anti-CD11c, Alexa 647-conjugated anti-CD8, PE-conjugated anti-CD4, biotin-conjugated anti-CD19, APC-conjugated anti-CD3, PE-conjugated anti-Gr1 (BD Biosciences, San Jose, CA), FITC-conjugated anti-CD11b (R&D Systems, Minneapolis, MN), PECy5.5-conjugated anti-F4/80, PECy5.5-conjugated anti-rat IgG (eBiosciences, San Diego, CA). We also used Alexa 647-conjugated streptavidin (LifeTechnologies, former Molecular Probes, Carlsbad, CA). Cell staining was performed to identify mainly myeloid populations (CD11b, Gr1, F4/80, CD3) and mainly lymphoid populations (CD4, CD8, CD19, CD11c). Cells were analyzed in a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA), where at least 30,000 events per sample were acquired. Data was analyzed with the FlowJo software (TreeStar, Ashland, OR).
10ug of soluble amastigote proteins were separated in 12% acrylamide gels and transferred to nitrocellulose membranes (GE healthcare) using a semidry system (GE healthcare). Membranes were incubated in PBS with 5% milk and 0.1% Tween 20 for one hour and with primary antibodies (anti-SHP70, anti-OPB or anti-TXNPx) in PBS with 2.5% milk and 0.1% Tween 20 for two hours. Three washing steps with PBS 1x 0.1% Tween 20 for 10min were performed and followed by incubation with secondary antibodies diluted in PBS with 2.5% milk and 0.1% Tween for one hour. Membranes were washed twice in PBS 1x 0.1% Tween 20 and once in PBS, 10min each. After incubation with ECL Prime Western Blotting Detection Reagent (GE healthcare) for five minutes, membranes were exposed to X-Ray films. Images (in TIF files) were analyzed using ImageJ software and the results were normalized to GAPDH band intensities.
Footpads from mice infected for 13 weeks were fixed in paraformaldehyde 4% at 4°C for 18h and after washing and dehydration in ethanol were infiltrated with xylene and paraffin. 4μm paraffin sections were used for immunohistochemistry. After deparaffinization and blocking with 5% BSA in PBS for 30 minutes, endogenous peroxidase was blocked in 0,1% sodium azide, 3% H2O2 in methanol for 30 minutes. After incubation with the primary antibodies in PBS 2% (w/v) BSA for 18h at 4°C, the samples were washed in PBS and incubated with secondary peroxidase-conjugated antibodies for two hours, followed by washings in PBS. They were next incubated with DAB (DAKO, Glostrup, Denmark), counterstained with hematoxylin, dehydrated and diafanized, and mounted with Permout (Sigma).
Intensities of DIGE spots of amastigotes from BALB/c and nude mice were compared using t test of DeCyder BVA Software. Western blot ratios of samples from BALB/c and nude mice were also compared using t test.
Statistical analyses of FACs data was done using ANOVA followed by Tukey´s multiple comparison test for all comparisons except for lymphoid and dendritic cells in footpads, where t test was employed. * = p<0.05.
Mice were infected in the left footpads and lesion size was estimated by measuring left footpad thickness weekly. As shown in Fig. 1, footpad thickness increased earlier in BALB/c than in BALB/c nude mice and was larger in the former than in the latter during the thirteen weeks of monitoring.
The difference between BALB/c and BALB/c nude footpad thickness does not seem to reflect parasite numbers and the establishment of infection after 13 weeks. In fact, quantification of parasite loads in the footpads showed slightly higher numbers of amastigotes in BALB/c nude lesions (mean values of 1.51x108 and 4.94 x 108 parasites for BALB/c and nude mice, respectively), although the difference was not statistically significant.
To better characterize the difference between the wild type and the athymic mice, we compared immune cell populations in spleens, draining lymph nodes (popliteal) and footpads in infected and non-infected (control) mice. Using flow cytometry, we quantified the percentages of the following leukocyte populations: T lymphocytes as CD3+CD4+ or CD3+CD8+ cells, monocytes as CD11b+ (Mac-1) cells, macrophages as CD11b+F4/80+, polymorphonuclear cells (PMNs), mainly neutrophils, as CD11b+Gr1+ cells, DCs as CD11b+CD11c+ or CD11b+CD11c+CD8+, and finally B lymphocytes as CD19+ cells. The frequency of the above populations in spleen, lymph nodes and footpads in control and infected BALB/c and nude mice are shown in Fig. 2.
The frequencies of CD4 and CD8 T cells were higher in spleens and lymph nodes of control BALB/c than in nude mice, as expected (Fig. 2A and B). CD4 corresponds to 21% and 0,8% of the cells in spleens of BALB/c and nude, respectively, and CD8 to 12 and 0.2%. In lymph nodes of BALB/c and nude CD4 corresponds to 50% and 0.8% of the cells, respectively, and CD8 to 16 and 0.2%. The number of cells in uninfected footpads was very low and insufficient for the labeling of all markers. We therefore analyzed only myeloid cells in these tissues (Fig. 2C), so that we could compare to nude mice, which display mainly myeloid cells.
Fig. 2A shows that infection did not significantly change the frequency of CD4 and CD8 T cells, as well as B lymphocytes, in the spleen. Monocytes (CD11b+) and polymorphonuclear cells (Gr1+) were more abundant in nude mice spleens before and after infection (CD11b around 22% and Gr1 around 10% in nude, versus 8 and 3% in BALB/c), possibly due to the low numbers of T cells. Their proportions did not change after infection. Macrophages had similar abundances in BALB/c and nude mice spleens, before and after infection.
In popliteal lymph nodes from BALB/c infected mice, there was a significant decrease in the frequency of CD4 T cells (from 50 to 25%, Fig. 2B). Infection significantly increased B cells in BALB/c lymph nodes (from 20 to 55%), but infected nude showed higher proportions of these cells (75%) than wild type mice. This result is expected, since we are working with percentages of total cells and nude mice virtually lack T lymphocytes. Dendritic cells CD8+ and CD8- were more frequent in control nude lymph nodes (0.8 and 2.3%, respectively, versus 0.3 and 0.8% in BALB/c), and DC CD8+ population decreased significantly (to 0.1%) in these mice after infection. As observed for spleens, monocytes were more common in infected nude than in BALB/c lymph nodes (6.4 versus 1.7%), but differently from spleen, polymorphonuclear cells increased in BALB/c mice after infection and became significantly more abundant than in infected nude mice (2.3 versus 0.4%). Macrophages were more frequent in uninfected nude than in BALB/c mice (47 versus 1.6%), but decreased significantly (to 4.9%) in the athymic lineage after infection.
In infected footpads the wild type and athymic mice had similar proportions of CD8+ T cells and dendritic cells (Fig. 2C). B cells and CD4+ T cells, on the other hand, were less abundant in BALB/c nude lesions (0.2 and 0.4% versus 1.6 and 1.4% in BALB/c, respectively). Footpads showed recruitment of some cell populations after infection, although a more complete analysis was hampered by the low number of cells recovered from control tissue, as already mentioned. BALB/c had a decrease in the proportion of monocytes after infection, while nude mice had increased frequency of polymorphonuclear cells and a decrease in macrophages in the lesions (Fig. 2C).
Soluble proteins from amastigotes isolated from BALB/c and BALB/c nude mice lesions were analyzed by DIGE. Fig. 3 shows a representative image of one experiment showing labeling of a pool of all samples (3A) and differential labeling of samples isolated from the two mouse strains (3B). In Fig. 3A spots corresponding to isoforms differentially expressed in the two samples are highlighted and represented in 3D images.
In all gels labeling of Cy3 and Cy5 was normalized with the Cy2 labeled pool of all samples (shown in Fig. 3A). Spots detected by the software were manually adjusted to exclude artifacts. The reproducibility and technical accuracy was verified by comparison of labeling of the pool of all samples with Cy3 and Cy5. Considering a cut-off of two fold for differential expression, 99,20% of the 1944 spots included in the analysis were similarly labeled by the two dyes. This cut-off was therefore employed for the experimental comparisons.
The DeCyder DIA module detected and matched 2100–2300 spots in each gel, that after manual validation were reduced to about 1700 spots. The DeCyder BVA module was then employed to compare the differentially expressed spots out from a total of 1178 spots that were matched considering the three experiments. According to our data set, amastigotes isolated from the two mouse strains share over 96% of the protein expression profile (p<0.05). The differences that can be attributed to the presence of T cell dependent responses are linked to only 3.4% (40 spots) of the proteins, which have decreased (18 spots) or increased (22 spots) abundance in BALB/c nude derived amastigotes. These spots were selected for identification by MS analysis. This analysis was performed with spots collected from preparative gels containing a pool of the protein samples from the three experiments. Due to the low abundance of most of these proteins and incompleteness of Leishmania protein databases, only 21 spots yielded protein identifications. These spots are listed in Table 1.
Spot number (ordered based on molecular weights), experimental molecular weight (MW gel) and isoeletric point (pI gel), p value obtained after T-test (T-test), ratio of intensity between amastigotes from nude/BALB/c (N/B), gene identification (gi) of the Hit, protein identification and its respective MW.
Among the differentially expressed spots comparing amastigotes from BALB/c and BALB/c nude mice we observed many proteins associated with oxidative/nitrosative stress (trypanothione reductase, peroxidoxin, tryparedoxin peroxidase, tryparedoxin, heat shock proteins) or proteins with protease/peptidase activity (oligopeptidase B, metallo-peptidase). Among them there are some known virulence factors of Leishmania such as oligopeptidase B [41,42] and tryparedoxin peroxidase [39,43,44]. Four isoforms of oligopeptidase B and four isoforms of tryparedoxin peroxidase were differentially expressed (Table 1 and Fig. 3A); all of them overexpressed in nude mice derived parasites. Isoforms of proteins such as HSPs 70 and 83 were less abundant in nude derived parasites (Table 1 and Fig. 3A) and are known to have chaperone function upon increased temperature and oxidative stress in differentiating and proliferating amastigotes [23]. Different isoforms of alpha and beta-tubulin were more abundant in nude-derived amastigotes. Interestingly, these cytosketetal proteins were also shown to be more expressed in antimonial resistant L. braziliensis and L. infantum [30].
We next evaluated the expression of OPB, TXNPx and HSP70 in five pairs of BALB/c and BALB/c nude derived amastigotes by Western blot. Figs. 4–6 show Western blot images and densitometric quantifications for OPB, HSP70 and tryparedoxin peroxidase, respectively.
As can be observed, amastigotes from the two mouse strains show similar expression of OPB, TXNPx and HSP70. OPB and TXNPx had four isoforms overexpressed in nude-derived parasites and HSP70 one isoform decreased in nude-derived amastigotes. The lack of correspondence between 2D-DIGE and conventional Western blots highlights the importance of analyzing isoforms of a protein individually.
To confirm the differential abundance of isoforms of TXNPx in parasites isolated from BALB/c and nude lesions, we performed 2D Western blots for this protein in a pair of samples. Three isoforms were identified in both samples, with pIs of 5.22 5.39 5.65 (Fig. 7A). Since the total amount of TXNPx was shown to be similar in the two types of extract by conventional Western blot (Fig. 6), we considered the sum of the three spots similar in the two samples and calculated the relative abundance of each isoform. The data shown in Fig. 7B indicate higher abundance of isoforms with pIs 5.2 and 5.39 in amastigotes isolated from BALB/c nude, confirming DIGE findings (Table 1). The same membranes were incubated with anti-phospho S, T, Y antibodies, and the observed labeling of the three isoforms (Fig. 7C) indicate that there are phosphorylated in at least one of these residues. Besides analyzing OPB, TXNPx and HSP70 expression in parasite extracts from lesions (Figs. 4, 5, 6)), we analyzed the abundance of the three proteins in sections of lesions from infected BALB/c and BALB/c nude mice. As shown in Fig. 8, all proteins are visible in amastigotes inside macrophages of both mice, and no labeling is observed in non-infected footpads. These results indicate that the proteins are specifically expressed by the parasites in both hosts.
Interaction between pathogens and their hosts derives from long term co-existence. Intracelular pathogens such as Leishmania respond to an aggressive microenvironment in the host creating evasion mechanisms that guarantee their survival. To further explore the adaptation between Leishmania and its host, we have studied the protein expression profile of parasites harvested from footpad infections in immunocompetent BALB/c mice, and immunodeficient BALB/c nude mice. Our results indicate that there may be a correlation between the cellular immune response of the host and specific protein isoforms of the parasite.
Footpad thickness increased in BALB/c and in BALB/c infected mice, but with a delay in the athymic animals. This difference is probably due to the observed and already described low number of T cells in the nude mice [15], leading to a compromised IFNγ production, inflammation in the site of infection and parasite clearance. Lesion progression patterns similar to our wild type mice were observed in BALB/c infected with other L. amazonensis strains [11,45,46].
Comparison of cellular compositions in lesions and lymphoid organs of mice indicate that L. amazonensis amastigotes in infected BALB/c and BALB/c nude footpads face different cells and consequently a distinct inflammatory milieu. As expected, lymphoid organs and lesions of nude mice had very low percentages of T lymphocytes. The lack of T cell responses impairs effector responses against the parasite, leading to uncontrolled Leishmania growth in the footpads of nude mice. On the other hand, lack of T cell responses may also lead to uncontrolled innate inflammation, as observed in HIV patients with very low CD4 T cell levels [47,48]. In our experimental model, we observed changes in the frequency of leukocytes in peripheral lymphoid organs of naïve and infected mice, mainly in the lesion draining lymph nodes. Of particular interest, we observed a significant decrease in the frequency of CD4 T cells and increase of B cells and Gr1 cells in BALB/c mice. In nude mice, there was a significant decreased in myeloid populations CD11cCD8 dendritic cells and macrophages, probably due to cell death induced by the known high parasite loads in these cells [49,50]. Interestingly, in footpads we observed a decrease of CD11b cells of BALB/c infected mice and macrophages in nude mice. However, there was a robust increase in the granulocyte population in infected nude mice, which could be indicative of an inflammatory response against the parasite.
The different pattern of cells and cytokines in the lesions of BALB/c and BALB/c nude mice resulted in 3.4% differential expression of soluble proteins in amastigotes. About half of the proteins that were significantly differentially expressed in the three experiments were identified, and many of them were related to oxidative/nitrosative stress or had protease/peptidase activity. Specific isoforms of trypanothione reductase, peroxidoxin, cytoplasmic tryparedoxin peroxidase (different isoforms), oligopeptidase B (different isoforms), alanine aminotransferase, metallo-peptidase, and small GTP-binding protein Rab1 were increased in BALB/c nude derived amastigotes. On the other hand, isoforms of heat shock 70 kDa protein, heat shock 83 kDa protein and a smaller form of cytosolic tryparedoxin were decreased in BALB/c nude derived parasites.
Infection of macrophages leads to production of cytotoxic oxidants, and Leishmania must be able to detoxify these agents to survive and proliferate inside the cell. Similarly to the other trypanosomatids but differently from other eukaryotes and prokaryotes, Leishmania has a trypanothione redox system instead of the more ubiquitous glutathione/glutathione reductase (GR) system [51]. Trypanothione participates in the detoxification of hydroperoxides, metals and drugs and in the synthesis of DNA precursors. The molecule is used as a donor of electrons and reduces the hydroperoxides generated by macrophages during infection [52]. Tryparedoxin peroxidase (TXNPx) catalyzes the detoxification reaction, and trypanothione reductase regenerates the reduced dithiol state of trypanothione necessary for all these reactions [51]. Recent data showed that trypanothione reductase, tryparedoxin peroxidase and peroxidoxin were 1.2 to 2 fold in Leishmania donovani promastigotes under oxidative and/or nitrosative stress in vitro [43,53]. TXNPx also participates in oxidative resistance in L. donovani [43], in L. infantum [54] and in L. amazonensis [55], and increases infection of L. donovani and survival in the presence of antimonials [43]. Accordingly, splenic amastigotes of L. donovani express higher levels of the enzyme than axenic amastigotes and are more resistant to H2O2 [39], and a more virulent strain of L. donovani expressed more of two specific cTXNPx isoforms than a less virulent strain [56]. High levels of cTXNPx were observed in L. donovani isolates [57], L. braziliensis and L.chagasi lines [30] unresponsive to antimony, in L. amazonensis resistant to arsenite [55] and in metastatic L. guyanensis [58]. In our study, four TXNPx isoforms were overexpressed in nude derived parasites. It is important to analyze whether the differentially expressed isoforms are functional and whether the enzyme activity is modulated in amastigotes by the host immune system.
Isoforms of Oligopeptidase B (OPB) were also identified as overexpressed in nude-derived amastigotes. This enzyme is a serine peptidase restricted to bacteria, plants and trypanosomatids that hydrolyses peptides of up to 30 amino acids after basic residues, especially arginine [59]. OPB has been considered a virulence factor in trypanosomatids, including Leishmania [41,42,60]. In fact, promastigotes of L. major deficient in OPB did not differentiate normally to metacyclic form, showed reduced infection and survival in macrophages in vitro [41] and a significant delay in lesion development in vivo [42]. OPB has already been described in L.(L.) major [41,42,61], L. braziliensis [62], L. donovani [26] and L. amazonensis [63]. No previous study has described isoforms of OPB, and we have no clues about their impact on the enzyme function.
Differently from OPB and the trypanothione related proteins mentioned above, some proteins were less expressed in parasites from athymic nude mice, such as heat shock 70 and heat shock 83 kDa proteins. HSP70 and HSP83 are evolutionarily conserved, constitutively transcribed and regulated at post-transcriptional level [64]. HSPs protect cells against different types of stimuli that can cause cell damage. HSP70 assists in protein translation and translocation across membranes, avoids aggregation of damaged proteins and reactivate denatured proteins. HSP70 may protect from toxic environmental conditions by cooperating with other stress-induced proteins to prevent heat-induced denaturation prior to protein aggregation and by suppressing programmed cell death that would be triggered by the activation of specific kinases [65]. The control of HSP70 activity in Leishmania is regulated not only by the protein abundance, but also by phosphorylation at specific residues [23]. When expressed in response to stress encountered in mammalian host, HSPs are likely to confer protection to the parasite and to play a crucial role in their survival [66]. In fact, HSP83 increased in response to heat shock and in the initial hours of promastigote- amastigote differentiation in L. infantum [64] and was shown to control differentiation in L. donovani [67], and HSP70 and HSP83 levels are higher in L. donovani amastigotes compared to promastigotes [24]. Both HSP 70 and 83 were overexpressed in L. infantum and L. braziliensis resistant to antimonials [30]. HSP70 has been shown to be increased in L. infantum under a heat shock or sub lethal oxidative stress, and the overexpression of this HSP conferred increased resistance to H2O2 [65]. Similarly, virulent promastigotes of L. donovani exposed to NO showed appreciable increase in relative synthesis of HSPs 83, 70 and 65 [66], and a more virulent strain of L. donovani had higher abundances of three isoforms of HSP70 [56]. Different combinations of oxidative and nitrosative stresses increased in 1.3 to 1.8 fold the expression of 13 heat shock proteins (HSPs) in L. donovani, including HSP70 and 83 [53]. In agreement with the induction of HSPs in stress conditions, our data shows that one isoform of HSP70 was less abundant in nude derived parasites. It remains to be shown whether the increase in this isoform confers survival advantages to amastigotes, or if this isoform is a less active form of HSP70. The sequencing of L. amazonensis genome has recently shown that this species has a higher number of genes containing HSP70 domain compared to other Leishmania [68]. This information reinforces the importance of studying this gene in the context of L.amazonensis infection and host cell interaction.
The increased expression of specific trypanothione reductase, cytoplasmic tryparedoxin peroxidase and oligopeptidase B isoforms in amastigotes from nude mice suggest that T cells or T cell-derived mediators and cellular interactions are associated with post-translational modifications of these proteins in BALB/c infected footpads. Another possibility is that the lack of T cell suppression in nude mice could lead to higher production of reactive oxygen and/or nitrogen species by macrophages, that could stimulate the modification of trypanothione-associated enzymes, as already described [53]. Analysis of infected footpads indicated higher levels of iNOS mRNA in BALB/s than nude mice (Velasquez et al., manuscript in preparation). Comparison of ROS should also be done to better define the stress conditions faced by the parasite in the two mice. Interestingly, we found that isoforms of HSPs 70 and 83 had the opposite regulation, being under expressed in nude-derived parasites. As discussed by others [65], we believe that Leishmania have redundant mechanisms for surviving oxidative stress. Accordingly, we postulate that different factors must stimulate the production of TXNPx and OPB isoforms in nude-derived amastigotes (or repress them in BALB/c-derived parasites) and induce the modifications of HSP in BALB/c parasites.
Conventional Western blot experiments did not show differences in the expression of OPB, TXNPx and HSP70 between BALB/c and BALB/c nude-derived amastigotes. Since we showed that these three proteins had several isoforms, the differences noted for specific isoforms were probably compensated when the sum of all isoforms was analyzed in conventional Western blots. These results suggest that post-translational modifications (PTMs) but not total levels of the three proteins are modulated by the presence of T cells and cytokines in mice lesions. The most well characterized PTMs in Leishmania are phosphorylations, while less information is available on pathways and roles of methylations, acetylations and glycosylations [26]. The isoforms of these three proteins differentially expressed in the amastigotes may correspond to one or more of these PTMs, which may lead to a more or a less active form of the protein that may affect parasite survival and lesion progression. 2D Western blot for TXNPx confirmed the higher abundance of two of the four isoforms identified in DIGE experiments in amastigotes from nude mice. The other two were not labeled by the antibody, possibly due to lower abundances. These three isoforms were also recognized by anti-phospho threonine, tyrosine and serine antibodies, suggesting that phosphorylation is the PTM process that generated the different isoforms of the enzime.Different isoforms of HSP70 and TXNPx were also differentially regulated in L. donovani strains with different virulences, but the significance of these findings and the PTM involved is still unknown [56].
Since infected nude mice could partially reproduce the immune response of a patient with diffuse cutaneous leishmaniasis, it is important to analyze the activities and roles of the protein isoforms over expressed in this mouse strain. It is also important to search for the specific stimuli that drive the post-translational modifications of HSP70, OPB and TXNPx. We are currently attempting to generate L. amazonensis clones over expressing TXNPx or OPB to evaluate the contribution of these proteins to infection by this important parasite species.
|
10.1371/journal.pntd.0003476 | Intrachromosomal Amplification, Locus Deletion and Point Mutation in the Aquaglyceroporin AQP1 Gene in Antimony Resistant Leishmania (Viannia) guyanensis | Antimony resistance complicates the treatment of infections caused by the parasite Leishmania.
Using next generation sequencing, we sequenced the genome of four independent Leishmania guyanensis antimony-resistant (SbR) mutants and found different chromosomal alterations including aneuploidy, intrachromosomal gene amplification and gene deletion. A segment covering 30 genes on chromosome 19 was amplified intrachromosomally in three of the four mutants. The gene coding for the multidrug resistance associated protein A involved in antimony resistance was also amplified in the four mutants, most likely through chromosomal translocation. All mutants also displayed a reduced accumulation of antimony mainly due to genomic alterations at the level of the subtelomeric region of chromosome 31 harboring the gene coding for the aquaglyceroporin 1 (LgAQP1). Resistance involved the loss of LgAQP1 through subtelomeric deletions in three mutants. Interestingly, the fourth mutant harbored a single G133D point mutation in LgAQP1 whose role in resistance was functionality confirmed through drug sensitivity and antimony accumulation assays. In contrast to the Leishmania subspecies that resort to extrachromosomal amplification, the Viannia strains studied here used intrachromosomal amplification and locus deletion.
This is the first report of a naturally occurred point mutation in AQP1 in antimony resistant parasites.
| Drug resistance remains a major concern in leishmaniasis chemotherapy, a neglected tropical disease that causes 60,000 deaths around the world annually. To better understand the molecular mechanisms behind drug resistance, we selected L. guyanensis parasites resistant to antimony, the first-line drug against this disease in many countries. Through whole-genome sequencing we found variations in the copy number of chromosomes in addition to gene amplification and gene deletion events in antimony-resistant parasites. A marker previously related to antimony resistance, the gene coding for multidrug resistant protein A, was found to be amplified. Transport studies revealed a reduced antimony accumulation in resistant parasites that we correlated with the deletion of the gene coding for the aquaglyceroporin 1 (AQP1) responsible for antimony uptake in Leishmania. Additionally, a point mutation in AQP1 was found to be associated with antimony resistance. These findings may contribute to the development of new chemotherapy strategies against leishmaniasis.
| Leishmaniasis defines a spectrum of infectious diseases caused by protozoan parasites belonging to the genus Leishmania that are transmitted to mammals via the bite of sandflies. Leishmaniasis are neglected tropical diseases that could potentially affect ~ 350 million people in 98 countries [1]. Clinical manifestations differ widely depending on the host immune response and the Leishmania species responsible for infection and vary from visceral leishmaniasis—VL to cutaneous leishmaniasis—CL [2]. The clinical manifestations of CL can further vary from localized ulcerative skin lesions to destructive mucosal inflammation (mucocutaneous leishmaniasis—MCL), the latter being mostly associated with infections caused by the Viannia subgenus in South America [3–6].
No vaccine is available against leishmaniasis and chemotherapy thus represents the main strategy for the treatment of all forms of the disease [7]. Despite the introduction of paromomycin [8], amphotericin B [9] and miltefosine [10] in the anti-Leishmania arsenal, pentavalent antimony (SbV)-derived compounds have been used for more than 65 years and are still the first-line of treatment against leishmaniasis in many countries [11]. Drug combinations, short therapeutic schemes and single drug doses are solutions currently debated to avoid drug resistance, one of the major drawback against leishmaniasis especially in the case of antimony [12]. Antimonial resistance has first emerged against VL in India [13] but cases of treatment failure involving species from the Viannia subgenus have since been reported in Brazil [14,15], Peru [16] and Colombia [17]. Drug susceptibility screenings also supported the notion that antimony resistant L. (Viannia) parasites can develop in the field [18,19].
Antimony is most active against Leishmania in its trivalent form (SbIII) which is produced through the reduction of pentavalent antimony (SbV) possibly within the macrophage hosts [20] but also within Leishmania [21,22]. SbIII is then passively transported into Leishmania cells through aquaglyceroporin 1 (AQP1), a porin also allowing the transport of water, glycerol, urea, dihydroxyacetone, methylglyoxal and polyols [23]. SbIII is indeed a chemical mimic of natural AQP1 substrates, having a similar conformation and charge as glycerol [24]. AQP1 plays an important role in volume regulation and osmotaxis in Leishmania [25] and its reduced expression is associated with SbIII resistance [26]. On the other hand, re-sensitization is achieved when AQP1 is overexpressed in resistant parasites deficient for AQP1 [26,27]. Targeted mutagenesis of L. major AQP1 demonstrated a role for residues Glu125 and Ala163 located at the extracellular loop in SbIII susceptibility [28].
While several molecular mechanisms leading to antimony resistance in Leishmania have been described, resistance remains only partly understood and most likely constitutes a multifactorial process [29]. Next generation sequencing has been used to produce several L. donovani genomes and revealed genomic alterations and plasticity that correlated with antimony resistance [30,31]. Gene amplification is also frequently observed in both laboratory-raised or field isolates resistant to antimony, in which circular or linear extrachromosomal DNA are formed by homologous recombination and annealing of direct or inverted repeated sequences, respectively [32]. A well-studied example of such amplification is the gene coding for the multidrug resistance associated protein A (MRPA) which is frequently amplified as part of circular amplicons originating from chromosome 23 in SbIII-resistant strains [33–35] and whose role in resistance involves the intravesicular sequestration of Sb-thiol conjugates in SbIII-resistant Leishmania [36].
Our understanding of drug resistance mechanisms come from the analysis of parasites belonging to Leishmania subgenus and little is known about the mechanisms leading to antimony resistance in the Viannia group, with the exception of few recent studies that highlighted previously observed alterations [37,38]. In this study, whole-genome sequencing was performed in laboratory-selected antimony resistant (SbR) Leishmania (Viannia) guyanensis mutants aiming at the dissection of molecular mechanisms of SbIII resistance in Leishmania (Viannia) parasites.
Leishmania (Viannia) guyanensis (MHOM/BR/1975/M4147) promastigotes were axenically maintained in minimum essential culture medium (α-MEM) (Gibco, Invitrogen, Grand Island, NY, USA) at pH 7.0 supplemented with 10% (v/v) heat inactivated fetal bovine serum (Wisent Inc., St-Jean-Baptiste, QC, CA), 100 mg mL-1 kanamycin, 50 mg mL-1 ampicillin, 2 mM L-glutamine, 5 mg mL-1 hemin, 5 mM biopterin, (Sigma-Aldrich, St Louis, MO, USA) and incubated at 25°C. Four L. guyanensis SbIII-resistant mutants (LgSbIII650.1 to LgSbIII650.4) were independently selected from WT L. guyanensis in 25 cm2 flasks containing 5 mL of α-MEM in the presence of increasing SbIII concentrations. Potassium antimonyl tartrate (Sigma-Aldrich, St Louis, MO, USA) was used as the source of SbIII. The stepwise drug selection ranged from 80 μM up to 650 μM of SbIII. Last-level SbR mutants were grown in absence of drug pressure for 26 passages to revert resistance. In addition, two independent L. guyanensis SbIII-resistant mutants (LgSbIII.1/2013 and LgSbIII.2/2013) were selected by SbIII increments (resistant to 80, 160, 240, 325 or 650 μM SbIII) and maintained in culture. For drug susceptibility assay, 106 parasites mL-1 in mid-log phase growth were seeded in 24-wells cell culture plates containing 1.5 mL of α-MEM, incubated under gentle agitation at 25°C during 72 h in presence or absence of several concentrations of drug. Growth was monitored daily by measuring absorbance at 600 nm to obtain the Sb sensitivity profile [39].
Sequencing libraries were produced from 50 ng of phenol-extracted/ethanol-precipitated genomic DNA by using the Nextera DNA sample preparation kit (Illumina Inc, San Diego, CA, USA) according to manufacturer instructions. Genome sequences were determined by Illumina HiSeq1000 101-nucleotides paired-end reads. Sequencing reads were aligned to a Leishmania (Viannia) braziliensis (MHOM/BR/1975/M2904) reference genome (TriTrypDB version 6.0) [40] using the software package Burrows-Wheeler Alignment [41]. The maximum number of mismatches was 4, the seed length was 32 and 2 mismatches were allowed within the seed. The detection of single nucleotide polymorphisms was performed using SAMtools (version 0.1.18), bcftools (distributed with SAMtools) and vcfutils.pl (distributed with SAMtools) [42]. Putative SNPs detected by whole genome sequencing were verified by conventional PCR amplification and DNA sequencing. Sequencing data are available at the EMBL-EBI European Nucleotide Archive (http://www.ebi.ac.uk/ena) under study accession number PRJEB6114 with samples ERS434587, ERS434588, ERS434589, ERS434590 and ERS434591 corresponding to L. guyanensis WT, LgSbIII650.1, LgSbIII650.2, LgSbIII650.3 and LgSbIII650.4, respectively.
First-strand cDNA was synthesized from 5 μg of total RNA using Oligo dT12–18 and SuperScript II RNase H-Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) according to the manufacturer protocol. Equal amounts of cDNA were run in triplicate and amplified in 25 μL reactions containing 1 x iQ SYBR Green Supermix (Bio-Rad, Hercules, CA, USA), 100 nM forward and reverse primers and 100 ng of cDNA target. Reactions were carried out using a rotator thermocycler Rotor Gene (RG 3000, Corbett Research, San Francisco, USA). Mixtures were initially incubated at 95°C for 5 min and then cycled 30 times at 95°C, 60°C and 72°C for 15 s. No-template controls were used as recommended. Three technical and biological replicates were established for each reaction. The relative amount of PCR products generated from each primer set was determined based on the cycle threshold—Ct value and the amplification efficiencies. Data were analyzed using the comparative 2-ΔΔCt method. Gene expression levels were normalized to constitutively expressed mRNA encoding glyceraldehyde-3-phosphate dehydrogenase (GAPDH, LbrM.30.2950). Primers for targeted genes and internal gene expression control GAPDH were designed using PrimerQuest (http://www.idtdna.com/Primerquest/Home/Index) and sequences are listed in S1 Table.
Molecular karyotype was obtained from L. guyanensis WT and Sb-resistant mutants by separation of chromosomes by pulse field electrophoresis [43]. 108 mid-log phase parasites were embedded in low melting point agarose blocks, digested with proteinase K and electrophoresed in a contour clamped homogenous electric field apparatus (CHEF Mapper, Bio-Rad, Hercules, CA, USA). The blocks were mounted in 1% agarose gel and electrophoresed in 0.5x Tris-Borate-EDTA running buffer at 5 V cm-1 with 120° separation angle at 14°C during 30 h. A range of 150 to 1500 kb was applied for a wide chromosomal separation, resolving most of Leishmania chromosomes in a single molecular karyotype gel. Saccharomyces cerevisiae chromosomes were used as DNA size marker (Bio-Rad, Hercules, CA, USA). For Southern blots, genomic DNA was isolated using DNAzol reagent (Life Technologies, Carlsbad, CA, USA) following manufacturer’s instructions and digested with the PstI restriction enzyme (New England Biolabs Inc, Ipswich, MA, USA). Digested genomic DNA or PFGE-derived molecular karyotype were transferred by capillarity onto nylon membranes (Hybond-N+, Amersham Pharmacia Biotehc, Sunnyvale, CA, USA) and cross-linked with UV light. The blots were hybridized with [α-32P]dCTP labeled DNA probes according to standard protocols [44]. Primers used for southern blot probes are listed in S1 Table. Densitometric quantification of southern blot-derived bands was performed using Image J version 1.48a.
The gene LgAQP1 (GenBank accession numbers KJ623262 and KJ623263) was amplified from genomic DNA of L. guyanensis WT and LgSbIII650.4 using primers containing 5’ XbaI and 3’ HindIII restriction sites, followed by cloning in pGEM T-easy (Promega, Madison, WI, USA). The WT AQP1 and its LgSbIII650.4 variant were subcloned into the pSP72αZEOα expression vector, a derivative of pSP72αNEOα [45] in which the gene neomycin phosphotransferase (NEO) was replaced by the bleomycin-binding protein gene (ZEO) conferring resistance to zeocin [46]. To validate the expression of episomal LgAQP1, a green fluorescent protein (GFP)-tagged construct was made using a PCR fusion-based strategy as previously described [47] using primers listed in S1 Table. The GFP gene was amplified using the pSP72αNEOαGFP vector as template. The LgAQP1-GFP fusions were cloned into pGEM T-easy and subcloned into the XbaI/HindIII sites of pSP72αZEOα, resulting in the pSP72αZEOαLgAQP1WTGFP or pSP72αZEOαLgAQP1(G133D)GFP constructs that were transfected by electroporation as previously reported [45]. Transfected parasites were preselected in the presence of 500 μg mL-1 of Zeocin Selection Reagent (Life Technologies, Carlsbad, CA, USA) and after 24 h, selection of transfectants was carried out in presence of 1 mg mL-1 of Zeocin Selection Reagent. Before transfection, all constructs were confirmed by DNA sequencing.
Total and membrane protein fractions were extracted from Leishmania as previously described [48]. Briefly, parasites were centrifuged and washed three times with ice-cold Hepes-NaCl at 3000 rpm for 5 minutes. The pellet was resuspended in a lysis buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 1.5 mM MgCl2, 1 mM DTT) and homogenized by vortexing after addition of proteases inhibitors cocktail (1 mg mL-1 leupeptin, 2 μg mL-1 aprotinine, 5 mM EDTA). The lysate was then incubated on ice for 15 min, followed by three cycles of freeze (-80°C) and thaw (37°C). The supernatant was recovered after centrifugation at 15000 rpm for 30 min at 4°C. From there, supernatant containing membrane fractions was kept at -80°C. Proteins were then extracted from membranes fractions using solubilisation buffer (50 mM Tris-HCl pH 8, 150 mM NH4Cl, 2 mM MgCl2, CHAPS 1%) by incubation on ice for 30 min. 50 μg of proteins were run on 10% acrylamide gel and transferred electrically onto nitrocellulose membrane (Bio-Rad, Hercules, CA, USA). The blots were blocked overnight in PBS (1 X), Tween (0.1%), Milk (5%). Membranes were incubated overnight at 4°C with a GFP monoclonal antibody (Roche, Basel, Switzerland) and an α-tubulin monoclonal antibody (Life Technologies, Carlsbad, CA, USA) diluted 1:1000 in PBS-Tween-Milk solution. Membranes were then washed three times for 5 min in PBS-Tween and incubated 1 h with horseradish peroxidase-conjugated goat anti-mouse IgG (Thermo Fisher Scientific Inc, Waltham, MA, USA) diluted 1:10000 in PBS-Tween. Membranes were washed again three times and incubated with Immobilon western chemiluminescent HRP substrate (Millipore, Billerica, MA, USA).
Antimony accumulation measurements were carried out based on previous studies [49,50]. Briefly, 108 mid-log phase Leishmania promastigotes were washed and resuspended in 1 mL of Hepes/NaCl/Glucose buffer (20 mM HEPES, 0.15 M NaCl, 10 mM glucose, pH 7.2) followed by incubation with 540 μM SbIII at 25°C as previously described [38]. One hour of Sb incubation was chosen to compare differences in Sb accumulation among the conditions evaluated. Drug accumulation was stopped by incubating cells on ice followed by three washes with ice-cold Hepes/NaCl/Glucose buffer. Parasites were centrifuged at 1800 g during 5 min at 4°C and the dried pellet was digested in 100 μL of 65% HNO3 (Merck, Darmstadt, Germany) before Sb was quantified by graphite furnace electrothermal atomic absorption spectrometry using an AAnalyst 600/800 spectrometer (Perkin Elmer, Waltham, MA, USA). Blank matrix was established by measuring Sb traces in Sb-unexposed HN03-digested cells. Blank absorbance values were subtracted as background. Intracellular Sb content was normalized by number of cells. Sb-resistant Leishmania mutants were maintained at least 2 passages without drug pressure prior to the transport assay to avoid contaminations.
EC50 values were calculated by non-linear regression when applied, data were analyzed by Student’s t test or analysis of variance (ANOVA) followed by correction performed using Bonferroni’s multiple comparison test. A p value ≤ 0.05 were considered statistically significant. Statistical analyses were carried out using the software GraphPad Prism version 5.0 (GraphPad Software Inc., La Jolla, CA, USA).
Four independent SbR L. guyanensis mutants (LgSbIII650.1, LgSbIII650.2, LgSbIII650.3 and LgSbIII650.4) were obtained in vitro by stepwise SbIII selection. While L. guyanensis wild-type (WT) parasites presented an EC50 of 53.72 μM, the four SbR mutants were resistant to at least 1 mM of SbIII, representing a resistance index (RI) of more than 18 times (Table 1). The resistance phenotype of every mutant remained stable even after 26 passages in absence of antimony, at which point parasites still presented SbIII EC50 values superior to 1 mM (Table 1).
Whole-genome sequencing was conducted on the four independent L. guyanensis SbR lines as well as on the isogenic L. guyanensis M4147 WT line by Illumina next-generation sequencing. For all strains, this produced genome assemblies of 31 Mb with a coverage depth of at least 50 fold. Copy number variations (CNVs) associated with resistance were identified by comparing the coverage of uniquely mapped reads between L. guyanensis SbR mutants and the WT line as part of small non-overlapping genomic windows (5 kb) along the chromosomes (normalized for the total number of uniquely-mapped reads for each strain) [51]. This enabled the observation of CNVs at the level of entire chromosomes (aneuploidy) and at specific genomic loci (amplification/deletion).
Several cases of supernumerary chromosomes were observed in the SbR mutants (Figs. 1 and S1). Most of these had log2 SbR/WT read ratios close to 0.5 indicating a gain of about 1.5 chromosome copies compared to WT parasites. Parasites from the Leishmania Viannia subgenus are distinct from other Leishmania species in harboring predominantly trisomic genomes [52] and this should thus represent a gain of one allele compared to WT parasites (going from 3 to 4 chromosome copies). Most supernumerary chromosomes were not shared by the mutants however; chromosome 13 was consistently increased in all SbR mutants, and chromosomes 11 and 25 were increased in three of the four mutants (LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3) (S1 Fig.). Chromosome losses were also observed in the SbR mutants and these were consistent with the loss of one allele (S1 Fig.). Interestingly, CNVs calculated from read depth coverage often led to a cumulative ploidy not matching with a clear-cut number of chromosomes but instead to intermediate log2 SbR/WT values (S1 Fig.). This was observed for both chromosome gains and losses and suggests differences in chromosome-level CNVs between individual cells within the population, a phenomenon known as mosaic aneuploidy [53]. Overall, mutant LgSbIII650.4 was more divergent and displayed the highest level of chromosome-level CNVs compared to the three other mutants (Fig. 1).
Normalized read depth coverage allowed the identification of amplified and deleted genomic loci in the L. guyanensis SbR mutants. These are characterized by punctuated series of genomic windows one beside the other whose normalized read coverage varies compared to the WT baseline [51], as observed for chromosomes 19 and 23 in more than one mutant (S1 Fig.). For chromosome 19, a subtelomeric region of 87.5 kb covering 30 genes (LbrM.19.0010 to LbrM.19.0300) appeared to be amplified in mutants LgSbIII650.1, LgSbIII650.2, and LgSbIII650.3 based on normalized read counts (Fig. 2A) and this amplification was confirmed by the hybridization of Southern blots with three distinct probes along the chromosome (Fig. 2A and 2B). Probes derived from genes LbrM.19.0270 and LbrM.19.0280 (all gene IDs reported in this work are based on the closest L. braziliensis genome used for alignments of L. guyanensis sequencing reads) located within the amplified region of chromosome 19 yielded 1.6 to 1.9 fold-increase hybridization intensities for mutants LgSbIII650.1, LgSbIII650.2, and LgSbIII650.3 compared to WT cells after normalization with LbrM.19.1070 used as an internal control for DNA loading (Fig. 2B). Consistent with the NGS data, mutant LgSbIII650.4 had band intensities equivalent to WT for both LbrM.19.0270 and LbrM.19.0280 (Fig. 2B). Interestingly, for mutants LgSbIII650.1, LgSbIII650.2, and LgSbIII650.3 harboring the subtelomeric amplification on chromosome 19, hybridization of chromosomes separated by PFGE with probes derived from genes LbrM.19.0270 and LbrM.19.0280 revealed a unique band (corresponding to chromosome 19) supporting intrachromosomal duplication of a specific region rather than extrachromosomal elements.
For chromosome 23, the amplified region was much larger than for chromosome 19 and covered 480–495 kb in mutants LgSbIII650.1, LgSbIII650.2, and LgSbIII650.3 (Fig. 3A), starting from one subtelomeric end and encompassing the locus coding for the well-established Sb resistance gene MRPA (LbrM.23.0280) (Fig. 3A and 3B). In mutant LgSbIII650.4, the increased in read length covers (almost) the entire length of the chromosome (Fig. 3A), suggesting an increased in ploidy. Southern blots hybridization of Pst1-digested genomic DNA revealed an up to 1.7 fold increased intensity for a MRPA-derived probe in the LgSbIII650.1–4 mutants after normalization with GAPDH signals used as DNA loading control (Fig. 3B). This is consistent with NGS data that revealed a 1.4–1.7 increased reads counts in the mutants compared to WT parasites (Fig. 3A). Intriguingly, PFGE-derived Southern blots hybridized with MRPA and Lbr.23.1000, two genes comprised in the 480–495 kb region amplified in mutants LgSbIII650.1 to LgSbIII650.3, presented a signal at 785 kb corresponding to chromosome 23 but also an additional signal at around 1.1 Mb (Fig. 3C and 3D). This 1.1 Mb band did not hybridize with LbrM.23.1660, a probe outside of the 480–495 kb amplified region (Fig. 3E). It is unclear how a region of chromosome 23 found its way to this chromosome. It is unlikely that it presents a linear amplicon as we never observed such large extrachromosomal elements [54,55] and the hybridization intensity (Fig. 3C and 3D) would suggest that this region has translocated into only one of the two homologous recipient chromosome in mutants LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3 (see also S2 Fig.).
Quantitative real time PCR validated that DNA amplification on chromosome 19 and 23 translated into increased mRNA levels (Fig. 4). The four independent mutants presented twice-more mRNA levels for MRPA compared to WT (Fig. 4A) while genes on chromosome 19 were upregulated in mutants LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3, but not in LgSbIII650.4 (Fig. 4B), confirming what was previously observed at the genomic level.
A fine scale analysis of sequencing coverage revealed that a subtelomeric deletion occurred on chromosome 31 in three of the four SbR L. guyanensis mutants (S1 Fig. and Fig. 5A). The deleted region covered around 25 kb in mutants LgSbIII650.1 and LgSbIII650.3 and 27 kb in mutant LgSbIII650.2 (Fig. 5A). In all three mutants the deleted region harbored the gene coding for the aquaglyceroporin AQP1 (LbrM.31.0020) known to be associated with antimony uptake in Leishmania [27]. Interestingly, sequencing reads could still be detected within 5 kb of the end of chromosome 31 in the mutants that presented AQP1 deletion (Fig. 5A) which could suggest telomere seeding in response to the loss of a terminal part of chromosome 31 (Figs. 5A and S1). The subtelomeric deletion in LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3 was confirmed by hybridization of Southern blots using probes located within and outside the deleted region. As expected, no signal was detected from gene LbrM.31.0010 to LbrM.31.0070 in mutants LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3 while LgSbIII650.4 and WT parasites presented a clear AQP1 signal (Fig. 5B). Conversely, hybridization signals were detected for every strain when the blots were probed with gene LbrM.31.0100 located outside the deleted regions or with the GAPDH gene located on a distinct chromosome (Fig. 5B). Since AQP1 was not deleted in mutant LgSbIII650.4, qRT-PCR assays were carried out in order to infer about any possible regulation of AQP1 expression in this mutant. However, AQP1 mRNA levels were similar in WT and in the LgSbIII650.4 mutant growing in presence of SbIII or in its absence for 26 passages (Fig. 5C). As expected, AQP1 expression was not detected by qRT-PCR in any of the three other mutants (Fig. 5C).
To better understand the kinetics of AQP1 deletion and its implication on growth fitness in the presence and absence of drug pressure, we selected two new series of SbR L. guyanensis mutants by five SbIII increments until reaching a final concentration of 650 μM. These series were named LgSbIII.1/2013 and LgSbIII.2/2013 (Table 1). At the first selection step (80 μM) AQP1 remained unaltered in both cell lines (Fig. 6A). When SbIII selection was increased, the AQP1 gene remained intact in the LgSbIII.1/2013 series but was lost in mutant LgSbIII160.2/2013 already at 160 μM (Fig. 6A). The amount of AQP1 mRNA was consistent with the copy number of the gene (Fig. 6B). Growth curves of LgSbIII.1/2013 and LgSbIII.2/2013 mutants revealed an advantage associated with the loss of AQP1 when parasites were cultivated in the presence of SbIII. Indeed, the LgSbIII.2/2013 mutant without AQP1 grew faster under SbIII selection (up to 325 μM) than LgSbIII.1/2013 mutants (Fig. 6D to 6F) presenting intact AQP1 copies (Figs. 6A and S3). This growth advantage of LgSbIII.2/2013 was not observed when parasites were cultures in drug free medium (S4 Fig.).
While the loss of AQP1 allows for a faster acquisition of resistance (Fig. 6), mutant LgSbIII650.4 had intact AQP1 copy number (Fig. 5B) and expression levels (Fig. 5C). Antimony accumulation experiments were thus performed with the L. guyanensis mutants. Quantification of intracellular antimony in L. guyanensis revealed, as expected, a very low accumulation of metalloid in the L. guyanensis SbR mutants in which AQP1 was deleted when compared to WT parasites (Fig. 7). Surprisingly, we also observed low accumulation in LgSbIII650.4 (Fig. 7). We hypothesized that AQP1 in LgSbIII650.4 may be mutated and this was confirmed by sequencing the gene, which revealed a single nucleotide polymorphism (SNP) at AQP1 position 398 in LgSbIII650.4, substituting a guanine for an adenine (S5 Fig.). This missense mutation in AQP1 translated into the replacement of a glycine (Gly) residue by an aspartic acid (Asp) at position 133 (G133D) of the protein in LgSbIII650.4 (S6 Fig.). Glycine 133 is putatively located in the third transmembrane domain in LgAQP1 (Fig. 8A) and is conserved among several Leishmania species and also in the Plasmodium falciparum AQP (PfAQP) (Fig. 8B).
To functionally validate the contribution of the AQP1 G133D mutation in antimony resistance in L. guyanensis, GFP-tagged version of AQP1WT and AQP1G133D were episomally maintained in LgSbIII650.2, which is naturally disrupted for AQP1 (Fig. 5B). Hybridization of Western blots with an antibody directed against GFP yielded the expected 50 kDa band for the fusion protein and confirmed the overexpression of the fusion protein in the respective LgSbIII650.2 transfectants (Fig. 9). The overexpression of LgAQP1WT substantially sensitized LgSbIII650.2 to SbIII whose EC50 dropped from more than 1000 μM in the mock-transfected control, to 29 μM in the presence of the WT AQP1 allele (Table 1). On the other hand, LgSbIII650.2 transfected with an AQP1 version harboring the G133D mutation remained as resistant as the mock-transfected control (Table 1). The presence of GFP did not interfere with the function of AQP1, as the tagged version of WT AQP1 was equally potent as an untagged version of the protein at sensitizing the LgSbIII650.2 mutant to SbIII (Table 1). The WT version of AQP1 but not its mutated version also restored SbIII sensitivity when transfected in LgSbIII650.2 (Table 1). The G133D AQP1 also failed to alter SbIII EC50 when overexpressed in a L. guyanensis WT background (Table 1).
Next generation sequencing has been a useful approach for studying drug resistance in Leishmania parasites for detecting both point mutations and changes in copy number of genes [30,31,52,56–58]. A frequent mechanism of drug resistance is gene amplification of specific regions that happens at the levels of repeated sequences that abound in the Leishmania genomes [55]. Changes in copy number can extend also to whole chromosomes [30,33,59] and it has been argued that tolerance of such chromosomal CNVs may be beneficial under stress conditions as in the presence of drug pressure [60]. Finally the individual parasites within a population may have different specific genes amplified [55] and may have different ploidy of specific chromosomes [53,61]. The NGS technology was useful to detect several ploidy changes in Leishmania species and here we have tested it with the Viannia subgenus. Normalized read depth coverage identified chromosomes in our SbR L. guyanensis mutants whose ploidy was altered compared to WT parasites (Fig. 1). Recurrent changes are often strong candidates for linking a phenotype to a genotype and it is salient to point out that no single chromosome ploidy was identical between the 4 mutants (Fig. 1). The link between aneuploidy and drug resistance might therefore be circumstantial, but antimony resistance is a complex and multifactorial process [29] and, in this context, studying the cellular consequences of aneuploidy might still provide novel insights on drug resistance in Leishmania.
Sequence reads corresponding to a subtelomeric region of chromosome 19 were higher in three mutants out of four (Fig. 2A). This was confirmed by Southern blots, but chromosome sized gels did not support the possibility that this region was amplified as part of an extrachromosomal element and instead consisted in an intrachromosomal duplication (Fig. 2C and 2D). While representing a rare event in Leishmania compared to extrachromosomal amplification, this has already been observed while attempting to inactivate the essential gene GSH1 in Leishmania [62] or in L. major cells resistant to antimony in which an intrachromosomal amplification of a subtelomeric region of chromosome 34 was observed [31]. Species belonging to the Viannia subgenus have previously been reported to display a limited capacity to generate/maintain extrachromosomal DNA [63,64] which is consistent with the intrachromosomal amplifications observed here. Nonetheless, episome transfection is possible in Viannia (see Table 1) [65] and in some studies gene amplification was observed in Viannia parasites [38,66].
Read depth coverage also revealed large regions of chromosome 23 encompassing the MRPA resistance locus that were amplified in the four resistant mutants (Fig. 3A), a feature confirmed by hybridization of Southern blots (Fig. 3B). Karyotype analyses by PFGE revealed a new band of ~ 1 Mb in three mutants and an apparent change in chromosome ploidy for the fourth mutant (Fig. 3C and 3D). The exact mechanism of formation of the 1 Mb band is unknown but a fragment of 495 kb derived from chromosome 23 must have been rearranged since the MRPA and LbrM.23.1000 probes are 495 kb apart (Fig. 3C and 3D). The hybridization signal to the novel ~ 1 Mb band in LgSbIII650.1, LgSbIII650.2 and LgSbIII650.3 is less intense then the native chromosome 23 signal at ~ 800 kb, suggesting that the duplication of the MRPA containing fragment may have happened in only part of the population. Because of the short reads linked to Illumina sequencing, it is not helpful in determining how a portion of chromosome 23 has duplicated into a larger chromosome. One possibility would involve translocation (possibility subtelomeric) from one chromosome to another. While gene amplification (extrachromosomal or intrachromosomal) usually occurs at the level of repeated sequences [55], we have reported rare mechanistic events leading to gene amplification [31] and further studies are required to explain how these MRPA amplifications are produced. These results are consistent with the data shown above for the subtelomeric region of chromosome 19 where in Viannia, in comparison to Leishmania, increase in copy number is mediated by mechanisms that do not involve extrachromosomal amplification.
A terminal deletion of ~ 20 kb of seven genes on chromosome 31 including the gene coding for AQP1 was observed in three mutants (Fig. 5A and B). AQP1 is considered the major route of entry of trivalent antimony in Leishmania [27] and its overexpression leads to SbIII hypersensitivity [26,27]. Downregulation of AQP1 has been observed in both laboratory-raised and clinical Leishmania parasites resistant to antimony [26,31,67,68] and constitutes a potentially useful biomarker for antimony resistance. Deletion of AQP1 is also a major contributor to SbR in L. guyanensis because episomal overexpression of a WT AQP1 allele was sufficient to restore Sb sensitivity (Table 1) and accumulation (Fig. 7) in every mutant tested. The deletion of the gene also suggests that AQP1 is not essential in L. guyanensis. Chromosome 31 is polyploid in all Leishmania spp. tested [30,52,60] and the 20 kb region was deleted from all chromosome copies (Fig. 5B). Terminal deletions from 67 to 205 kb covering the AQP1 locus on chromosome 31 were also recently observed in SbR L. major for which the break points occurred at the level of inverted repeated sequences [31]. In these cases however, there was still one intact copy of AQP1. In contrast, L. guyanensis SbR mutants presented deleted regions of only 20.7 to 22.7 kb on chromosome 31 without any sign of inverted repeated sequences. The terminal deletion could thus be driven by micro-homologies with telomere-associated repeated sequences [31,69] or through a double strand break followed by a terminal healing process driven by telomere seeding [70], although any of these mechanisms need to be ascertain.
In one mutant, the AQP1 gene was not deleted but transport experiments indicated that there was no accumulation of SbIII (Fig. 7). This prompted us to sequence AQP1 and to demonstrate for the first time that a point mutation in AQP1 (G133D) can also be a novel resistance mechanism. Mutational analysis on LmAQP1 had already revealed that residues located at C-loop Ala163 and Glu152 (equivalent to LgAQP1 residues Val167 and Glu156, respectively—Fig. 8A) are involved in metalloid uptake and reduced permeability to antimony [28,71]. LmAQP1 is also post-transcriptionally regulated by a mitogen activated protein kinase 2-mediated phosphorylation at Thr197 (LgAQP1 Thr201—Fig. 8A) modulating SbIII uptake and sensitivity [72]. The absence of a crystal structure for Leishmania AQP1 precludes hypothesizing about the precise role of G133D in resistance. The lack of antibodies and the low level of GFP fluorescence have not allowed to test whether the G133D mutation could also impact the subcellular localization of AQP1. Interestingly, the homolog of LgAQP1 in the related protozoan Trypanosoma brucei (TbAQP2) was also linked to resistance to the arsenical-based compound melarsoprol (arsenic is a metalloid chemically related to antimony) [73,74]. Melarsoprol-resistant T. brucei were shown to have lost TbAQP2 or to harbor a nonfunctional chimera derived from recombination events between TbAQP2 and TbAQP3 [75] which show similarity with the mutants studied here (deletion or point mutations). Given the important role of aquaglyceroporins in volume and osmotaxis regulation [25], how null mutants compensate these functions is still an open question and further studies will be required for understanding their physiological adaptations.
Every L. guyanensis SbR mutants had a defect in antimony accumulation. Lower accumulation can be achieved either through decreased uptake or increased efflux. An additional transport mechanisms leading to resistance would be drug sequestration mediated by the intracellular ABC protein MRPA [36]. The lack of a functional AQP1 will lead to reduced uptake. Possibly minimal amounts of antimony can enter by other routes and overexpression of MRPA can lead to drug sequestration. Alternatively, the contribution of MRPA to resistance may be more important in early selection steps when all AQP1 copies (chromosome 31 harbouring AQP1 is polyploidy in every Leishmania species) have not yet been inactivated/deleted. It is salient to point out that alterations in expression of MRPA and AQP1 has been described in antimony-resistant natural isolates of L. donovani [29,67,68] and L. tropica [76], suggesting a concomitant antimony sequestering and decreased uptake [67,76]. These results are also consistent with antimony resistant L. amazonensis mutants selected in vitro [38]. The loss of AQP1 appears to be dominant in our current mutants since providing the mutants with a functional version of AQP1 enables complete re-sensitization to SbIII.
The present study highlighted that similar markers are involved in resistance in Leishmania and Viannia subgenus but that gene amplification differs with mostly extrachromosoal amplicons in Leishmania and intrachromosomal ones in Viannia. A new resistance mechanism corresponding to a point mutation in AQP1 was also discovered and this will allow further testing of the role of AQP1 in resistance.
|
10.1371/journal.pntd.0006829 | Rodent control to fight Lassa fever: Evaluation and lessons learned from a 4-year study in Upper Guinea | Lassa fever is a viral haemorrhagic fever caused by an arenavirus. The disease is endemic in West African countries, including Guinea. The rodents Mastomys natalensis and Mastomys erythroleucus have been identified as Lassa virus reservoirs in Guinea. In the absence of a vaccine, rodent control and human behavioural changes are the only options to prevent Lassa fever in highly endemic areas. We performed a 4 year intervention based on chemical rodent control, utilizing anticoagulant rodenticides in 3 villages and evaluating the rodent abundance before and after treatment. Three additional villages were investigated as controls. Analyses to assess the effectiveness of the intervention, bait consumption and rodent dynamics were performed. Anthropological investigations accompanied the intervention to integrate local understandings of human–rodent cohabitation and rodent control intervention. Patterns of bait consumption showed a peak at days 5–7 and no consumption at days 28–30. There was no difference between Bromadiolone and Difenacoum bait consumption. The main rodent species found in the houses was M. natalensis. The abundance of M. natalensis, as measured by the trapping success, varied between 3.6 and 16.7% before treatment and decreased significantly to 1–2% after treatment. Individuals in treated villages welcomed the intervention and trapping because mice are generally regarded as a nuisance. Immediate benefits from controlling rodents included protection of food and belongings. Before the intervention, local awareness of Lassa fever was non-existent. Despite their appreciation for the intervention, local individuals noted its limits and the need for complementary actions. Our results demonstrate that chemical treatment provides an effective tool to control local rodent populations and can serve as part of an effective, holistic approach combining rodent trapping, use of local rodenticides, environmental hygiene, house repairs and rodent-proof storage. These actions should be developed in collaboration with local stakeholders and communities.
| In the absence of a Lassa fever vaccine, rodent control is the primary prevention option. An effective rodent control intervention must understand human behaviour towards the rodent such as: human–rodent interactions, cohabitation, and local rodent control measures. We conducted a rodent control intervention at community level in a Lassa Virus endemic area in Upper Guinea (Guinea) accompanied by an anthropological study on people’s perceptions and recommendations on the intervention. Based on our results we seek to broaden the rodent control intervention by including environmental hygiene, house repairs and rodent-proof storage. Chemical treatment has proven effective for rodent control but other factors involved in human-rodent interactions should also be addressed. Our findings highlight the need for Lassa fever prevention and rodent control initiatives to work in collaboration with communities and undertake a holistic approach towards rodent control.
| Lassa fever is a viral haemorrhagic fever caused by an arenavirus, which was first discovered in Nigeria in 1969 [1, 2]. The disease is endemic in West African countries, including Sierra Leone, Liberia, Guinea, southern Mali, northern Cote d’Ivoire and Nigeria [3–6]. Lassa fever recently emerged in Benin and Togo [7–9] and affects between 200,000 and 300,000 people per year with up to 5,000 to 10,000 deaths annually [10]. The mortality rate is low (1–2%) in communities in endemic areas but can be as high as 50% among hospitalized patients during outbreaks [11, 12]. In general, most cases remain asymptomatic [13]. In Guinea, an epidemiological survey in human populations showed a high seroprevalence, up to 40%, in the south of the country near the border with Sierra Leone [14]. Acute cases have been reported in regional hospitals [15].
For many years, the Natal multimammate mouse, Mastomys natalensis, was considered to be the sole reservoir of the virus [16, 17]; however, a recent study in Guinea and Nigeria showed that two other species, the Guinea multimammate mouse, Mastomys erythroleucus, and the African wood mouse, Hylomyscus pamfi, can also serve as reservoirs [18, 19]. Humans can be infected by touching objects contaminated with rodent urine, breathing aerosolized particles, being bitten by rodents or consuming rodents [20–24]. Human-to-human transmission occurs in the community and in health care settings [10, 25, 26].
Treatment options are limited, with Ribavirin given early in the course of disease to improve survival in patients with Lassa fever [27]. In the absence of a vaccine, rodent control and human behavioural changes are currently the only options available to prevent Lassa fever in highly endemic areas. In Sierra Leone in 1983, a team from the United States Centers for Disease Control and Prevention reported an experimental approach to reduce the incidence of Lassa fever by controlling the rodent population [28]. The experiment lasted 5 weeks and was spatially focused in a single township. The authors did not find any change in disease incidence between people living in houses with the intervention (trapping) and those in houses without the intervention. It is probable that the experiment was too short to adequately measure an effect on the disease incidence. Furthermore, treatment of only a few houses within a single village led to a rapid re-infestation from neighbouring houses. In light of these limitations, our aim was to perform rodent control on a larger scale and for a longer duration. To that end, we investigated a dozen villages in Faranah, in rural Upper Guinea, where Lassa Virus (LASV) is widely distributed [29]. Previous studies of rodent dynamics have shown that M. natalensis rodents aggregate in houses during the dry season and disperse into gardens and surrounding fields in the rainy season, where they forage in cultivated areas before the harvest [30]. We therefore planned rodent control interventions inside houses during the dry season only, which starts in November and finishes in April [30].
Studies on rat poison use and availability in Africa are rare, but some examples of acute poison or anticoagulant use and supply in Tanzania [31], South Africa [32, 33] and Sierra Leone show effectiveness in controlling rodent population for short periods of time. These examples also indicate that people primarily use anticoagulants and acute poisons (zinc phosphide) because they are easily accessible, cheap and less labour intensive than trapping. In some cases, the anti-inflammatory drug Indomethacin [34, 35] may also be used instead of rodenticide. Whatever the substance used, the effects of rodent control are short-lived.
Despite the noise and the loss of crops, people often adjust to living alongside the rodents.
As in other studies on rodent control acceptability in Africa [33] we aimed to assess first the feasibility and acceptability of community rodent control activities before extending to an holistic approach including environmental sanitation, house repair and rodent proof containers.
To evaluate the feasibility of rodent control in the Faranah region, we performed an intervention based on chemical rodent treatment in 3 villages, evaluating the rodent abundance before and after treatment. Three additional villages were investigated as controls for comparison with the treatment villages. This article discusses the rodent diversity, rodent abundance and sociocultural factors affecting the feasibility, efficacy and acceptability of chemical rodent control. Based on our experiences during 4 consecutive years in Upper Guinea, we discuss ideas for advancing sustainable rodent control.
Six villages were chosen in the surrounding area of Faranah; 3 to serve as controls and 3 to perform the intervention. The choice of villages was based on their remote location from a paved road, a size not exceeding 1000 inhabitants, less than 45 minutes driving time from Faranah and the presence of LASV (see map in Fichet-Calvet et al. 2016). We first sampled rodents in 10 villages in November-December 2013. Of the 10 villages, 9 were positive, with a range of 1 to 10 LASV-positive M. natalensis rodents in each village. The villages were classified as either high or low prevalence and therefore allocated randomly to control and treated groups, leading to 3 villages in the control group with the following prevalence rates: 20.6% (7/34) in Sokourala, 19.6% (10/51) in Damania, and 2.1% (1/46) in Sonkonia. The 3 villages in the treated group had the following prevalence rates: 20.0% (8/40) in Dalafilani, 17.8% (8/45) in Yarawalia, and 3.8% (2/52) in Brissa. Thus, two villages with a high prevalence and one village with a low prevalence were included in each category (control versus treatment).
We planned to perform the treatment intervention during the dry season (November-April), when the rodents are expected to aggregate inside. We employed a treatment using anticoagulant rodenticide baits, which we distributed in baiting stations (Coral, Ensystex Europe) to all open houses of the village. Rodent control is more effective when it is managed at the collective level rather than at the individual level [36]. Two baiting stations were distributed in each room, totalling 300–600 stations per village, according to their size.
During the first three years, we purchased the anticoagulants locally available in Conakry. This treatment was a mixture of wheat and Bromadiolone, labelled at 0.01%, sold in small sealed bags (S1 Fig). In these baits, Bromadiolone was titrated at 30 ppm, which corresponds to a concentration of 0.003% (V. Lattard, pers. com. according to [37]). The concentration was therefore 3 times lower than that claimed on the label. During the last year, we used Difenacoum at 0.005% mixed with cereals and paraffin as bait in cubes weighing 50 g (Rodenthor bloc, Ensystex Europe). The duration of chemical treatment was 10 days during the first 2 years and 30 days during the last 2 years (S1 Table). In the villages with treatment, dead rodents found outside of their burrows were collected by the team and cremated in a special hole outside the village. This hole was also used to burn the contaminated waste produced by necropsies during the routine sampling of rodents for LASV testing. For comparison with the captured rodents, dead rodents were numbered and identified during years 3 and 4.
To verify whether rodents were eating the bait, we evaluated the consumption during the whole process of treatment during years 3 and 4. On day 1, the bait stations were weighed empty and filled with 50 g of bait before being set. On day 2, each station was weighed and the value was recorded (S2 Fig). The difference between day 1 and day 2 indicated the daily consumption for each bait station. The values were thereafter summed for all the stations set in the village.
Checking and baiting were performed each day during the first 5 days and then every 2 or 3 days between days 6 and 30. The reason for the more frequent checking at the beginning and less frequent checking at the end of the process was the decrease in the local rodent population due to the anticoagulant rodenticide activity. Typically, the highest rodent mortality occurs between 3 and 10 days after the first anticoagulant ingestion [38].
To measure the local rodent abundance, we set 120 traps along a transect crossing the villages. Two Sherman live traps (Sherman Live Trap Co., Tallahassee, FL, USA) were set per room over 3 consecutive nights. The traps were checked by team members each morning, and the captured animals were immediately necropsied in situ according to a BSL3 procedure [39, 40]. The animals were morphologically identified, and several biopsies were collected for further ecological and virological analysis. Morphological identification was facilitated from knowledge gained during 14 years of work with small mammal species in this area. Previous studies using both morphological and molecular identifications in the same geographical area have shown that M. natalensis was predominantly living in houses while M. erythroleucus was never found in houses [41].
The abundance was estimated by using the trapping success for each 3-night trapping session (Σ trapped rodents during 3 days/360 trap-nights x 100). A trapping session was performed twice in each treated village, i.e., before and after treatment, and once in the control villages. In total, we analysed 14,394 trap nights. Data are available in DOI 10.6084/m9.figshare.5545267.
We applied several techniques to collect local perception regarding cohabitation with rodents and the intervention after the first treatment and throughout the 4 years. Qualitative investigations included focus group discussions with groups of women and men separately in each village (11), in-depth interviews (39), informal discussions and participant observation and photographs.
Focus group discussions are used to collect views on a particular topic from specific groups of people with similar experiences. These discussions usually include approximately ten individuals and two moderators, one who asks questions and stimulates the discussion and another who takes notes. We used this technique to explore peoples’ views on the project activities and on why rodents live with them.
In-depth interviews were conducted individually with people who could provide detailed information on the research topic.
We conducted informal discussions with the team and people living in the villages regarding the intervention. Photographs documented the domestic space, the distribution of the houses, and building materials, among other inputs. In addition, we distributed a short quantitative questionnaire in August 2016 to evaluate the acceptance of our project by the villagers. We selected people in each village according to their availability, knowledge and willingness to participate [42]. Oral consent was obtained before distribution of the questionnaire. In total, 203 people were questioned regarding their views of the project and specifically whether rodents had disappeared from their houses and for how long this absence had persisted.
All the discussions were facilitated by a translator in Malinke, Djallonke and French. We obtained permission from the local health authorities (Directeur Régional de la Santé, Directeur Préfectoral de la Santé) in Faranah and in each village from the local chief, elders and youth leaders before starting the research activities.
During the Ebola Virus Disease (EVD) epidemic, lasting from March 2014 to January 2016, we were not able to follow the original planned intervention. People refused the trapping in some villages due to the fear and mistrust generated by the EVD epidemic. In year 2 for example, a trapping delay of 6 months meant that we did not have time to perform 1 trapping session in 6 villages before treatment, 1 trapping session in 6 villages after treatment, nor elimination in 3 villages before the end of dry season in April. Furthermore, 2 of 3 control villages refused to adhere to the experiment. In year 3, one control village was still reluctant to participate. In years 2 to 4, we therefore reduced pre- and post-treatment trapping in treated villages only, and kept a single trapping in control villages where possible. Details are presented in a supplementary document (S1 Text).
A comparison of the daily consumption between the Bromadiolone grains and the Difenacoum blocks, and also between villages was evaluated through a linear model (“lm” tool in R software, R Development Core Team, 2017), with the quantity as a response variable, and molecule, village and days as explanatory variables. Rodent dynamics were analysed using the linear mixed effects model (“lme” tool in R software), with the abundance of M. natalensis as a response variable, and treatment as an explanatory variable coded as a fixed effect. The village and the year were also entered in the model as random variables. The abundance of M. natalensis was here estimated through the number of trapped animals since the denominator of TS was always the same, i.e. 360 trapping nights. To analyse the possible trend of rodent abundance between year 1 and year 4, a third model was implemented in R, with the abundance of M. natalensis as the response variable, and type of village (treated vs control) and year (year1, year4) as explanatory variables in a 2-ways interaction. To evaluate the relationship between rodent abundance and bait consumption, a simple regression was performed with 6 points (3 villages x 2 years).
In both year 3 and year 4, the bait consumption at the beginning of the baiting period was low and then increased, reaching a maximum at day 5 or day 7 depending on the year (Fig 1). Subsequently, bait consumption gradually decreased to zero after days 25 or 28, depending on the year.
The bait consumption was similar regardless of the molecule (p = 0.06, df = 72), Bromadiolone (sum = 46.440; median = 0.940; IC95% = 0.723.8–1.340.2 kg) or Difenacoum (sum = 39.140; median = 0.930; IC95% = 0.682–1.057 kg). Village effect was evident, with a bait consumption higher in Brissa, than in Yarawalia (p = 0.007, df = 72, estimate = -287.7) and Dalafilani (p = 0.006, df = 72, estimate = -292.7).
The villages in our study were primarily inhabited by M. natalensis, which represented 94% (1047/1114) of the captured animals (Table 1). The dynamics of M. natalensis population changes in treated villages are shown in Fig 2, where the values of trapping success (TS) are plotted according to the time schedule, either before or after treatment.
After treatment, the population was significantly lower than before (p<0.0001, df = 11). After 3 years of treatment, the treated villages had a significant lower abundance than the control villages (significant interaction year 4 x treated villages, p = 0.03, df = 8, estimate = -19.7, in Fig 3).
A simple regression between abundance and bait consumption during year 3 and 4 gives a significant coefficient of correlation (r2 = 0.94, df = 4, p = 0.001). The highest abundance observed in Brissa corresponded to the highest bait consumption.
The most frequent building type is a round Sudanese-style mud hut with thatch roof; people use sleeping huts or rooms as storehouses; kitchens are adjacent buildings which follow the same construction style. Owners of concrete and metal-roofed buildings, which have several rooms, typically reserve one room as a store to protect harvest and seeds from fires (Fig 4).
Granaries are rare in the region; many people stopped using granaries because they were unable to prevent fires and thefts.
Women can preserve condiments, dry fish and leftover food in wooden boxes. Small quantities of rice remain freely available in rooms, especially in women’s rooms, for weekly family feeding. Before consumption, husk-rice is boiled, dried on the ground several times and pounded. During the drying process, rice can be transferred from the ground outside the house onto the floor inside the house or kitchen several times until the rice is dry.
Rodent-proof containers for post-harvest storage are not available in the region beyond plastic bags and plastic and metallic containers with and without lids, which are also used to store water, local handmade condiments and other crops.
Consequently, in certain villages, people highlighted the need for building food storage and making available rodent-proof containers in parallel with poisoning or trapping.
Rodents are considered a nuisance because of their effects on food stocks and personal properties. Individuals believed that it would be impossible to kill ‘all’ rodents. For them, they live in rural areas, surrounded by fields and vegetation, and are therefore in permanent cohabitation with rodents. Persons living in the periphery, bordering the bush, complained that rodents come back sooner to these houses than to those in the middle of the village.
Locally, preventive measures taken against rodents are very limited: acute poisons available on the market are used by individual owners in kitchens, stores and rooms before harvesting crops, when people are annoyed by rodent noise or when mice damage their belongings. Several people use Indomethacin, instead of poison because they want to prevent small children and domestic animals from accidental intoxication with poison. Adults may have cats, and the presence of cats was reported to result in fewer rodents in some houses. Children perform rodent trapping by hand when they find a nest in the house or have dogs with them to hunt rodents in the fields.
The consumption curves of the two compounds, Bromadiolone and Difenacoum, showed that both are effective and that after 28 days, no rodents were feeding. This complete lack of consumption at the end of the operation also showed that there was no resistance phenomenon. A resistance phenomenon is visible when the consumption curve makes a plateau, even peaks again after 15–20 days, due to resistant individuals, which continue to eat the bait. Despite a difference of amplitude between the two curves (Bromadiolone vs Difenacoum), the pattern is similar and quantities of bait were equally eaten at the end of the process when the 3 villages are taken into account. Only one village (Brissa) showed a higher bait consumption, which corresponds to a larger rodent population in years 3 and 4. The peak of Bromadiolone consumption observed during day 4–6 may be due to the behaviour of the rodents, which transport the wheat seeds to their burrows [43]. Paraffin blocks used in year 4 are generally recognized to have a lower palatability and consumption rate than whole cereals. The post-treatment trapping showed that the efficacy was similar between the 2 baits.
Information collected from villagers indicated that the rodents returned very quickly. Similar observations were done by the population in Sierra Leone after using poison [44]. Increasing the treatment duration to 30 days caused the rodent population, according to local individuals’ observations, to remain low for a longer period of time, with treatment effects persisting for approximately 2 to 3 months. This period would correspond to a return to the carrying capacity of a M. natalensis population if 90% were initially eliminated (calculation in [45]). Two months correspond also to the mean period for recovery of R. rattus populations after removal trapping in villages located in Uganda [46]. Assessment of trapping success after treatment was below 2% (years 3 and 4), which is comparable to an elimination performed with continuous trapping during one month [47].
As a consequence, the populations of M. natalensis oscillated from one year to the next, mimicking seasonal variations. The decline observed in our experiment, occurred however during the dry season when M. natalensis populations were expected to be abundant indoors, as shown by sampling during Year 1 in control villages after treatment (Fig 3) or in longitudinal studies performed in 2 other villages in the same region (Fichet-Calvet et al 2007). In this study, the decline is therefore due to our experiment and not to a normal seasonal effect. A comparison of rodent abundance between year 1 and year 4 showed a slight decrease in the treated villages. This finding needs to be confirmed in a long-term study. Year 2 had a slightly different pattern, as the abundance was surprisingly low before treatment. The trapping before treatment in year 2 was conducted in March, not in November as usually performed. This shift towards the end of the dry season may explain the altered abundance of rodents indoors because they had begun to disperse outdoors [30]. However, the difference might also be due to human behaviour. We observed that villagers in Dalafilani used rodenticides and increased the number of cats between 2 trapping sessions, which may explain the similar trapping success rates measured after treatment in year 1 and before treatment in year 2. This intensification of rodent control by villagers influenced the subsequent rodent abundance.
After 4 years, however, we conclude that once-yearly treatment is not sufficient to maintain the low population abundance of M. natalensis. There are several reasons for the return of rodents 2 to 3 months after the end of operations: 1) the high proliferation of this species, whose mean litter size is 9.2 (3–14 in [48]; 2) the survival of a few animals in the fields surrounding the houses, which allows recolonization of the human habitat; 3) several closed houses in which we were not able to enter and deposit of the poisoned baits, which may have served as shelters from which the rodents could recolonize houses in the surrounding areas; 4) several very attractive foods available for rodents; 5) the porosity of the walls and roofs, allowing rodents to enter very easily; and 6) the low prevalence of predators, such as cats and dogs. Points 1) and 2) would be more difficult to change because animals’ inherent biological traits are not very easily influenced. Experiments concerning fertility control in rodents are rare because chemosterilization is difficult to practice in the field [49, 50]. The surrounding fields are also the natural optimal habitat for M. natalensis, and it seems unrealistic to remove them from this habitat. Recent studies on M. natalensis movements in different microhabitats in the villages confirm the need to expand rodent control measures outside the house to nearby field and gardens [51]. However, points 3) to 6) could be modified more easily in collaboration with the residents taking advantage of the knowledge we are generating together.
Based on our findings we present in the following lines some ideas for improvement and work towards a sustainable and holistic rodent control intervention.
Based on these findings and the acceptability of rodent control activities at community level, we aim to promote, in coordination with health and agricultural authorities, a more holistic approach [56], including rodent trapping and poisoning, environmental hygiene, personal hygiene, house repairs and rodent-proof storage. The present scenario creates the potential to develop a research-based project and design a collective "one-health" action [57] for rodent management and Lassa fever control.
|
10.1371/journal.pgen.1007735 | Genotype to phenotype: Diet-by-mitochondrial DNA haplotype interactions drive metabolic flexibility and organismal fitness | Diet may be modified seasonally or by biogeographic, demographic or cultural shifts. It can differentially influence mitochondrial bioenergetics, retrograde signalling to the nuclear genome, and anterograde signalling to mitochondria. All these interactions have the potential to alter the frequencies of mtDNA haplotypes (mitotypes) in nature and may impact human health. In a model laboratory system, we fed four diets varying in Protein: Carbohydrate (P:C) ratio (1:2, 1:4, 1:8 and 1:16 P:C) to four homoplasmic Drosophila melanogaster mitotypes (nuclear genome standardised) and assayed their frequency in population cages. When fed a high protein 1:2 P:C diet, the frequency of flies harbouring Alstonville mtDNA increased. In contrast, when fed the high carbohydrate 1:16 P:C food the incidence of flies harbouring Dahomey mtDNA increased. This result, driven by differences in larval development, was generalisable to the replacement of the laboratory diet with fruits having high and low P:C ratios, perturbation of the nuclear genome and changes to the microbiome. Structural modelling and cellular assays suggested a V161L mutation in the ND4 subunit of complex I of Dahomey mtDNA was mildly deleterious, reduced mitochondrial functions, increased oxidative stress and resulted in an increase in larval development time on the 1:2 P:C diet. The 1:16 P:C diet triggered a cascade of changes in both mitotypes. In Dahomey larvae, increased feeding fuelled increased β-oxidation and the partial bypass of the complex I mutation. Conversely, Alstonville larvae upregulated genes involved with oxidative phosphorylation, increased glycogen metabolism and they were more physically active. We hypothesise that the increased physical activity diverted energy from growth and cell division and thereby slowed development. These data further question the use of mtDNA as an assumed neutral marker in evolutionary and population genetic studies. Moreover, if humans respond similarly, we posit that individuals with specific mtDNA variations may differentially metabolise carbohydrates, which has implications for a variety of diseases including cardiovascular disease, obesity, and perhaps Parkinson’s Disease.
| The detection and quantitation of mtDNA polymorphisms in populations and across whole habitats continues to be used as a central investigatory tool in evolutionary genetics. But, the approach is laden with assumptions about selection that are rarely examined. We present a series of studies that traverse the genotype to the phenotype. The studies were designed to experimentally test the interaction between diet and mitotype in Drosophila flies and provide a mechanism by which selection occurs. We start with population cage studies that include four laboratory diets and four mitotypes. We then directly compete two mitotypes (Alstonville and Dahomey) on a high protein and a high carbohydrate diet and show a flip in their relative fitness that is driven by differences in larval development. Next, we identify a single naturally-occurring point mutation, which drives the cage results. We track the ripple effects up to the level of the organelle (mitochondria), through the labyrinth of metabolic pathways and on to the phenotype. Notably, when flies were fed the high carbohydrate diet, energy metabolism was extensively remodelled in both mitotypes causing increased physical activity in Alstonville flies. These data invite an extensive experimental re-evaluation of the assumption that mtDNA inescapably evolves in a manner consistent with a strictly neutral equilibrium model. It also motivates investigation of genotype-specific dietary manipulation as an integrative treatment of human disorders involving mitochondrial metabolism and offers the potential for future therapeutic strategies.
| Diet and an organism’s genes contribute towards its phenotype and impact a range of scientific disciplines that span from the more fundamental disciplines of evolutionary biology and quantitative genetics to the more medically applied fields of nutrigenomics and pharmacogenomics. In nature, the dietary macronutrient balance is a strong selective force within and among populations. The relative proportions of macronutrients in food can fluctuate seasonally, vary when species colonise new habitats and can influence the frequency of alleles in populations [1, 2]. It is well documented that nutritional responses vary with genotype [3–8] and it has been convincingly argued that the human genome is maladapted to our 21st century diet [9]. Dietary modification is an established treatment for certain diseases including cardiovascular disease, diabetes, and obesity [10], yet, we still have an incomplete knowledge of how genetic variants that modulate susceptibility to disease are influenced by exogenous factors. This study explores the potential for diet to differentially influence mitochondrial function and the organismal health and fitness of Drosophila melanogaster flies harbouring distinct mtDNA types (mitotypes).
Protein and carbohydrate are the two primary energy-yielding macronutrients in fly food, and their ratio has been shown to have profound impacts on various aspects of physiology, behaviour, and biochemistry [11, 12]. In adult females of D. melanogaster Canton S, a 1:2 Protein: Carbohydrate (P:C) ratio of food yielded the highest egg-laying rate and a 1:16 P:C ratio maximised survival [13]. Aw et al. [14] and Towarnicki and Ballard [15] demonstrated a more complex scenario whereby diet interacted with Drosophila mitotype and with other environmental factors such as temperature. For adults, Aw et al. [14] reported sex-specific influences of mitotype and diet on mitochondrial functions and physiological traits in males harbouring the Alstonville and Japan mitotypes. In larvae, Towarnicki and Ballard [15] manipulated food and temperature to study the development of the Alstonville and Dahomey mitotypes. We observed that larvae harbouring the latter mitotype developed more slowly than the former when fed a high protein diet at all temperatures, but more quickly when fed the high carbohydrate diet at higher temperatures. These studies did not determine the magnitude of selection at an organismal level or differentiate the relative importance of the interactions in larvae and adult stages, nor did they provide a mechanism of action. Toward these goals, we constructed laboratory diets that differed in their P:C ratios (1:2, 1:4, 1:8 and 1:16 P:C) and also tested natural fruits that differed in their P:C ratio.
Laboratory population cage studies are a sensitive method to test for selection in Drosophila and the frequency of each genotype type in cages is taken as an indicator of fitness [16, 17]. Previous cage studies have provided evidence that distinct mitotypes can influence the frequency of flies in the laboratory, but none of these studies manipulated the diet [17–19]. The cage study paradigm used here does not involve flies breeding until termination of fecundity or lifespan; instead, it enforces a short window for flies to lay eggs. As a consequence, repeatable changes in the frequency of genotypes are caused by differences in immature development time and the fitness of young adults during the period that larval fat body remains [20]. Other experimental methods that have been utilised to estimate fitness of flies harbouring different mitotypes include in vivo competition and assaying mitotype frequencies of wild-caught animals [e.g., 21, 22, 23]. Ma and O’Farrell [21] created fly lines with multiple mitotypes and utilised the uniparental mode of inheritance in mitochondria to test for selection. They observed that non-coding differences in the origin of replication region could cause the frequency of individuals harbouring a genome with a detrimental mutation to increase, but then lead to population death after several generations. The mechanism for this is still unknown. Thermal selection has been proposed to shape the pattern of mtDNA variation in eastern Australian D. melanogaster, but no experimental information has been provided on which mutation(s) may be driving these data [22, 24]. Here, we chose the population cage paradigm for its high sensitivity and have quantified the frequencies of four globally sourced D. melanogaster mitotypes (Alstonville, Dahomey, Japan and w1118) fed our four P:C diets. We then directly compete two mitotypes fed two diets.
Mechanistically, provisioning of dietary macronutrients to mitochondria may be influenced by genetic variations that influence the activity of the electron transport system, organelle retrograde signalling to the nuclear genome, anterograde signalling to the mitochondrion and epigenetic modifications [12, 25]. These variations may result from mtDNA mutations, mito-nuclear interactions and nuclear-encoded differences [7, 22, 26–28]. Mitochondria produce energy by utilising electrons harvested from oxidisable dietary substrates and O2 to build up a proton-motive force by pumping protons from the mitochondrial matrix into the intermembrane space. The subsequent backflow of protons to the matrix across complex V (ATP synthase) of the inner membrane drives the synthesis of ATP. Functional differences in mitochondrial energy production influence evolutionarily important physiological and organismal traits. In Drosophila, these traits include development time and egg production, and in humans, they include inherited disease and the decline in mitochondrial function with advancing age [25, 29–31]. Here, we identified functionally significant differences between mitotypes by carefully controlling the nuclear genetic background, modelling quaternary and secondary structures, conducting multiple independent in vitro assays, adding electron transport system inhibitors to the diets, and assaying independently collected mitotypes [32–35].
Is it possible that a given mtDNA mutation could be slightly deleterious in one environment but advantageous in another? If a mtDNA mutation is functionally deleterious, and linked mutations are neutral or nearly neutral, current models predict that the mitotype will have a selective disadvantage, causing it to decline in frequency in nature and population cage studies. Slightly deleterious mutations have been reported in Drosophila [26, 36–38], purifying selection has been demonstrated in the mouse female germline [39, 40], and deleterious mtDNA mutations are well-known in humans [41–43]. However, as dietary stress increases, genotype-specific mitochondrial responses may trigger flexible and broad cytosolic and nuclear reactions that have collectively been termed mitohormesis [44]. Remarkably, rather than being harmful, these changes caused by low levels of stress can result in a reconfiguration of metabolism, which in turn can enable increased production of ATP, increased evolutionary potential, and decreased susceptibility to disease [12, 45]. Again, the mechanisms for this are not well understood. Various mechanisms by which stressed mitochondria may signal outward to the cytosol and the nucleus have been proposed. These include regulation of ATP levels, altering mitochondrial membrane potential to allow recruitment and assembly of signalling molecules, and the production of reactive oxygen species (ROS). These are, however, not the only available pathways in the mitochondrial repertoire [44]. For instance, calcium signalling from the endoplasmic reticulum likely influences a multitude of mitochondrial functions [46, 47]. Above a genotype-specific threshold, increasing the level of a specific stressor is expected to be deleterious and disease-causing, with the distribution of the response determined by the “norm of reaction” [48]. The norm of reaction describes the pattern of phenotypic expression across a range of environments and may be entirely different for two mitotypes. To investigate the possibility of functional compensation through mitohormesis, here, we conducted transcriptomics and metabolomics studies. We then experientially examined the mechanisms involved by manipulating dietary sugars and inhibiting specific metabolic pathways.
Our series of studies show that a diet by mitotype interaction mediated the remodelling of carbohydrate metabolism in two Drosophila mitotypes. When fed the high protein 1:2 P:C diet, the slightly deleterious ND4 mutation in complex I of Dahomey mtDNA caused the mitotype to have a selective disadvantage compared to those harbouring Alstonville mtDNA. Complex I is the primary entry point for electrons into the mitochondrial electron transport system and is a site of electron leak to oxygen and the generation of ROS [49]. In contrast, when fed the high carbohydrate 1:16 P:C diet, mitotypes differentially remodelled energy metabolism, and this resulted in an evolutionary advantage to Dahomey. Were the same mechanisms found to occur in humans, the enhanced lipogenesis in individuals with slightly deleterious complex I mutations could make them more susceptible to obesity when eating a high carbohydrate diet, however, for those individuals with a predisposition to Parkinson’s disease, which has been linked to defects in lipogenesis [50], this diet could delay onset or rate of decline.
Unravelling the influence of diet on DNA variations is a challenge with broad evolutionary, health care and disease implications [12, 51], yet we still have an incomplete knowledge of the mechanisms involved. In this series of studies, we tested the interaction between diet and mitotype in Drosophila to determine the presence and mechanism of selection. We tracked mitotype specific effects up to the level of the mitochondrion, through the morass of metabolic pathways and on to the phenotype. We concluded that differential provisioning of macronutrients to mitochondria harbouring distinct mitotypes led to phenotypic changes in food consumption, starvation resistance, and movement, as oxidative phosphorylation and β-oxidation of fatty acids were differentially regulated.
To test the hypothesis that the fitness of mitotypes can be differentially influenced by diet [12], we fed four diets varying in P:C ratio to four Drosophila mitotypes and assayed their frequency in population cages over 12 generations. The nuclear genome was standardised to w1118 and the microbiome controlled each generation by adding a ground homogenate of laboratory males. Given random mating of Drosophila harbouring distinct mitotypes [18], population cage studies are a sensitive method to detect positive selection [17, 18].
We investigated whether differences in immature development or adult fitness best described the observed mitotype frequencies on the four diets. Demonstrating that natural selection acts on mitochondrial genes is now firmly established [e.g., 14, 28, 52–56], but the specific life history stages and exogenous conditions through which mtDNA variations benefit the organism have rarely been experimentally identified. When all else is equal, reduced immature development time is beneficial in nature as it reduces exposure to predators and limited food supply [57]. It is also advantageous in population cages if a higher proportion of females of a specific mitotype develop into adults and more eggs are laid. The fitness of young females is experimentally determined by assaying fecundity and fertility. Female fecundity is sensitive to dietary changes and is experimentally measured as the number of eggs laid [13, 14]. Fertility is a central determinant of an animals inclusive fitness and is quantified as the number of offspring, per female [58].
We conducted three additional cage studies to corroborate the hypothesis that the fitness of the Alstonville and Dahomey mitotypes was differentially influenced by diet [12]. In the first set of cage studies, we permute the diet to determine whether the mitotype specific responses are generalisable. Here, we include the 1:2 P:C diet for generations 1–4, swapped to the 1:16 P:C diet for generations 4–20, and then returned to the 1:2 P:C diet for generations 20–26. In a second set of cages, we include fruits with ~1:2 and ~1:16 PC ratios. Fruits have previously been used to validate laboratory diets as they effectively control for artificial differences in amino acids, lipids, and micronutrients [63]. We include passionfruit (~1:2 P:C) and banana (~1:16 P:C). In the third set of cages, we compete the two D. melanogaster mitotypes independently against Drosophila simulans (Wolbachia uninfected with the siIII mitotype collected from Kenya [64]). These species are sympatric through large parts of their range and compete for similar resources.
We assay immature development time and test for reproducibility, permute the nuclear genome, replace the laboratory diets with natural fruits and include the microbiome from orchard fed flies. To test for reproducibility, development time was assayed at ~6-month intervals. Mito-nuclear interactions have been shown to influence a range of molecular and organismal traits in insects, crustaceans, fish, and mice [7, 65–68]. Here, we substituted the w1118 nuclear genome with Oregon R and with Canton S using balancer chromosomes and then conducted five generations of backcrossing before our experiments [7]. The w1118 nuclear genome diverged from the wild caught Oregon R line in 1984, and they have been separated for more than 800 generations. The Canton S line was collected before 1916 in Canton, Ohio [69]. To corroborate the cage studies that included fruit, we test the development times of the mitotypes fed passionfruit and banana. Host-associated microbiota can impact metabolism and gene expression at cellular and organism-level scales [70, 71]. Adair et al. [72] quantified the bacterial communities associated with natural populations of D. melanogaster and found microbes were predominantly of two to three taxa. Here, we focus on levels of Acetobacter and Lactobacillus as they dominated the microbial communities in our populations.
To predict whether a nonsynonymous change, an RNA mutation, or variation in A+T repeat number was likely to be functionally significant we generated quaternary and secondary structure models and then assayed repeat number variation [79–82]. There are three nonsynonymous difference between Alstonville and Dahomey mtDNA [66] and we modelled each complex harbouring a change—complex I (V161L, ND4 subunit), complex IV (D40N, COIII subunit) and complex V (M185I, ATP6 subunit) [79–81]. There are also three rRNA differences (two srRNA and one lrRNA) [83] and 52 A+T-rich region variations [84] (S1 Table). Towarnicki and Ballard [15] mapped the two srRNA mutations on the human mitoribosome and proposed that they are unlikely to influence mitochondrial function [83, 85] so they are not considered here. Selection has rarely been shown to act on the mitochondrial A+T rich or control region [but see, 21, 27], and no differences were identified in secondary structures or the central T-stretch between the Alstonville and Dahomey mitotypes [15]. However, differences in the number of repeats have been recently shown to influence mitochondrial functions [86].
We tested hypotheses generated from the modelling by extracting mitochondria and assaying organelle function, independent of cellular interaction. These cellular assays included electron transport system complex activity assays, in vitro mitochondrial oxygen consumption, Western blots and native protein gels. Activity was included because it has previously been employed to corroborate the influence of a mtDNA mutation [87, 88]. We predicted that functionally significant mutations would reduce the activity of the complex. In vitro mitochondrial oxygen consumption is increasingly recognised as a fundamental measure of mitochondrial function [89, 90] and we assayed the rate from extracted mitochondria using complex I and II substrates [91]. Complex I substrates assay the combined mitochondrial functions of complexes I, III and IV, while complex II substrates assay the collective functions of complexes II, III and IV. Western blots were used to measure expression of complex I and complex V and native protein gels to determine native protein masses of complex I and its protein–protein complexes.
Chemical impairment of complex I reproduces the observed flip in development rates. Here, complex I inhibitors were added to the diet to create phenocopies in Alstonville of the Dahomey ND4 mutation. Goldschmidt [100] coined the term “phenocopy” to describe morphological alterations in Drosophila that could be induced by the imposition of stress during development. Thus, a phenocopy is produced environmentally and shows features characteristic of a genotype other than its own. Chemically induced phenocopies in Drosophila are well studied with the production of eyeless mutants by feeding food containing borate [101] and production of bithorax mutants by treating embryos with diethyl ether [102]. We added rotenone to the diet to phenocopy the ND4 mutation in Dahomey because it inhibits electron transfer from the iron-sulphur centres in complex I, leading to a partial blockade of oxidative phosphorylation with reduced synthesis of ATP [103]. We then quantify the rate of development. If the slightly deleterious V161L ND4 mutation in Dahomey was driving the differences in development time (Fig 1B), we predicted that Alstonville larvae fed food containing rotenone (the phenocopy) would develop more slowly than untreated larvae on the 1:2 P:C diet, but faster on the 1:16 P:C food. In contrast, we predicted that Dahomey larvae fed rotenone would develop more slowly when fed both diets as the complex I dysfunction would be the combined effects of the mutation and the inhibitor. We then tested the generality of the rotenone result with paraquat. Paraquat is a common herbicide that has been proposed to cause mitochondrial dysfunction by complex I toxicity following lipid peroxidation of the mitochondrial inner membrane [104].
Next, we tested whether dietary addition of rotenone influenced complex I activity, superoxide dismutase (SOD) activity, and larval dry weight. Complex I activity in larvae fed the standard diets was measured in Study 3 (Fig 4A). It was included here to test whether the ND4 mutation and the dietary addition of rotenone had similar effects on the complex. SOD constitutes the first line of defence in the antioxidant enzyme network [105, 106], is the primary scavenger of the ROS superoxide [107], and total activity was assayed. Larval weight was assayed as an organismal trait that can influence development time [reviewed in 108] and patterns of adult reproductive investment [109].
To test whether the flip in immature development time was generalisable to a second pair of mitotypes, one of which harboured the V161L ND4 mutation, we compared the immature development times of flies harbouring Madang (Papua New Guinea) and Victoria Falls (Zimbabwe, Africa) mtDNA [84]. Madang mtDNA has the same ND4, and lrRNA mutations and differs from Dahomey by 27 A+T rich region mutations (S1 Table). Victoria Falls does not harbour either the ND4 or the lrRNA mutations. It has three nonsynonymous (ND2, ATP6 and COIII), two sRNA and 49 A+T rich region differences from Alstonville (S1 Table). For experimentation, both mitotypes were harboured in the w1118 nuclear genetic background and the microbiome was controlled.
To gain mechanistic insight into the processes underpinning the flip in the development times of the mitotypes we include transcriptomics and metabolomics studies. Next-generation RNA sequencing has permitted the mapping of transcribed regions of the genomes of a variety of organisms [120–122]. Studies of Drosophila reveal a transcriptome of high complexity that is expressed in discrete, tissue- and condition-specific mRNA and ncRNA transcript isoforms [120]. This enables a dynamic ensemble of transcript isoforms that gives rise to substantial diversity. Recently, Crofton and colleagues [123] asked whether D. melanogaster mothers who experience poor nutrition during their own development change their gene product contribution to the egg. They find an increase in transcripts for transport and localization of macromolecules and for the electron transport chain. In this study flies were raised for at least two generations on instant Drosophila food. Eggs were then transferred to each diet and a standard microbiome added after 2 d. A limitation of the technique is that not all transcripts currently have a known function.
Metabolomic profiling provides an additional layer of knowledge for the most complete representation of the phenotype of the animal, revealing the combined contributions of gene expression, enzyme activity, and environmental context [124]. Here, we include gas chromatography-mass spectroscopy (GC/MS), which is capable of measuring small molecules with a mass <500 Da. One constraint of the GC/MS method for metabolomics studies is that distinct molecules may have similar retention times and it is necessary to validate results with standards [125].
In this section, we explore the disadvantage to Dahomey on the 1:2 P:C diet and begin to investigate the mitohormetic responses of the mitotypes fed the 1:16 P:C diet. We assay basal ROS production and expression of two Glutathione S-transferase (GST) genes because SOD activity was higher in Dahomey than Alstonville larvae and higher in mitotypes fed the 1:2 P:C diet (Fig 5C). Basal mitochondrial ROS gives the levels produced at the resting state and are an indicator of mitochondrial coupling efficiency in respiration [143]. ROS production and detoxification are tightly balanced, and numerous stress response mechanisms have evolved [144]. GSTs are a large supergene family of an ancient detoxifying enzyme and respond to endogenous and exogenous substrates through glutathione conjugation [145]. Transcriptomic data showed that Dahomey larvae fed the 1:2 P:C diet exhibited an elevation in cytochrome P450 metabolism (Fig 7A), and had higher expression levels of GstE1 and GstE5 (S2A Table). Here, we perform quantitative reverse transcription PCR (RT-qPCR) to confirm that the genes that were identified in the original RNA-seq were also altered as expected.
Next, mtDNA copy number and levels of ATP were assayed. Copy number is regulated by ROS in yeast [146], is positively linked to levels of ATP [147], and is crucial for maintaining cellular energy supplies [147, 148]. In Drosophila, mtDNA copy number is proposed to impact the organismal phenotype by influencing the respiratory membrane and the efficiency of oxidative phosphorylation [86]. ATP production has been shown to influence many cellular processes and evolutionary important physiological parameters including development rates [29, 149].
We posited that the polyol pathway was mechanistically involved in the mitohometic response in Dahomey larvae due to the elevation of sorbitol levels in the metabolomics data and predicted that including dietary sugars in the pathway would be beneficial. If true, we hypothesised that the addition of the polyol pathway inhibitor Epalrestat would mitigate the net benefit. We then assayed the number of flies eclosing in 3 d, quantified the expression of Notch (N) and Cyclic-AMP response element binding protein B (CrebB), and determined food consumption. In the non-disease context the polyol pathway is essential for cellular osmoregulation but, in the context of diabetes, it is associated with tissue-damage during hyperglycaemia [151]. In the pathway, glucose is reduced to sorbitol, via the action of the enzyme aldose reductase, and then oxidized to fructose. O-fucose and O-glucose are essential for normal Notch signalling [152] and their levels are regulated by derivatives of the polyol pathway including fructose, sorbitol and mannose, while xylose negatively regulates signalling [153, 154]. Notch regulates the cAMP responsive element binding protein (CREB) [155, 156], and experientially blocking CREB activity in Drosophila fat body has been shown to increase food intake [157]. N and CrebB were differentially expressed in the transcriptomics data (S2B Table).
The polyol pathway does not produce ATP so could not adequately account for the similarities in ATP levels between the mitotypes fed the 1:16 P:C diet. Here, we test the hypothesis that rates of β-oxidation differed between the mitotypes and add Etomoxir to the diet. β-oxidation of fatty acids generates NADH and FADH2 and thereby partially bypasses complex I of the electron transport system [91]. Etomoxir inhibits entry of long-chain fatty acids into the mitochondrion via the carnitine shuttle and we predicted its addition would result in loss of the selective advantage to Dahomey. We then quantified development, triglyceride levels, expression of elongase F (eloF) and brummer (bmm), β-oxidation activity, acetyl-coA enzyme activity, NAD+/NADH ratio and starvation survival. Metabolomic data showed high levels of stearic and palmitic acid in Dahomey larvae so we assayed triglycerides. To test for increased lipogenesis, we assayed the expression of eloF and bmm. elofF is a female-biased elongase involved in long-chain hydrocarbon biosynthesis [158]. bmm is a lipase which promotes fat mobilisation and is responsible for channelling fatty acids toward β-oxidation [159]. Both, elofF and bmm were differentially expressed in the transcriptomics data (S2B Table). β-oxidation was directly quantified using 14C-labelled palmitic acid. Acetyl-CoA was measured because the breakdown of carbohydrate influences its levels. NAD+ is required for fatty acid metabolism and the NAD+/NADH ratio was assayed. Starvation resistance was tested as a significant organismal trait [160]. When a larva is not feeding, energy can only come from the metabolism of existing resources [161], which occurs when fruits are small, when food quality declines and also in a fluctuating environment [162].
Transcriptomic data discussed in Study 5 actively support the result that a general increase in mitochondrial gene expression is part of rewiring in Alstonville on the 1:16 P:C diet. Furthermore, we hypothesised that glucose-6-phosphate was differentially metabolized in Alstonville due to the observed elevation in gluconate. Glucose 6-phosphate can be converted to store glycogen through the action of glycogen synthase and so we assayed levels of glycogen. Glycogen synthase and insulin-like receptor (Inr) are elevated in Alstonville (S2 Table). Glycogen is a primary source of energy for adult muscle function [170, 171] and the ubiquitous activation of Inr has previously been shown to cause larvae to feed less and to wander off the food [172]. Therefore, we assay development time and movement.
Glucose 6-phosphate is also metabolized by the pentose phosphate pathway and D-Gluconate can be phosphorylated to 6-phospho-D-gluconate to enter the oxidative phase of the pathway [173]. Here we quantified the expression of Zwischenferment (Zw) and assayed glucose-6-phosphate dehydrogenase (G6PD) activity. Zw was differentially expressed in the transcriptomics data (S2B Table). Zw catalyses the oxidation of glucose-6-phosphate (G6P) to 6-phosphogluconate. G6PD is the rate-limiting enzyme of the pentose phosphate pathway [174, 175]. We then assayed one aspect of insulin signalling. The insulin/insulin-like growth factor signalling pathway controls a wide variety of biological processes in metazoans [176] and stimulates glucose metabolism via the pentose phosphate pathway in Drosophila cells [177]. The most upstream central players in this pathway are members of the insulin-like peptide (ILP) family, which includes insulin and insulin-like growth factors in mammals [178], as well as multiple ILPs in worms and insects [179]. ILPs are regulated by nutritional status and Insulin-like peptide 2 (Ilp2) is essential for maintaining normoglycemia [180]. We assayed Ilp2 to corroborate the results from the transcriptomics data (S2B Table). Here, we replaced sucrose (control) with gluconate as the dietary sugar, but did not include any blockers because we considered this the wild-type pathway on the 1:16 P:C diet.
Over the past decade, it has become clear that diet is an evolutionary force that has immediate implications for our understanding of health and disease. Here, we provide substantial evidence to suggest that a single mtDNA encoded nonsynonymous mutation can differently influence the regulation of dietary metabolites and have significant phenotypic consequences. When fed the 1:2 P:C diet, Alstonville larvae had a relative advantage as the V161L ND4 mutation in Dahomey caused an increase in ROS production, which resulted in oxidative stress and a decrease in mitochondrial functions leading to reduced mtDNA copy number and ATP levels.
When fed the 1:16 P:C diet, Dahomey larvae had the relative advantage with multiple linked pathways working in a synergistic mitohormetic response that enabled larvae to eat more and develop more quickly. The remodelled pathways in Dahomey included upregulation of the polyol pathway, which fed back to increase food consumption and fuelled increased β-oxidation of fatty acids. Each cycle of β-oxidation results in the donation of electrons to the quinone pool downstream of complex I in the electron transport system, thereby bypassing the V161L, ND4 subunit mutation [189]. This process maintains levels of the quinone pool, which has been shown to be functionally important [190]. In Alstonville, mitochondrial gene expression was higher, glycogen metabolism increased and larvae were more active. We postulate that the greater physical movement in Alstonville larvae on the 1:16 P:C diet caused a reallocation of ATP away from cell division and growth, thereby slowing development. ATP drives many cellular processes and constrains development rates [29, 149]. An alternative explanation is that upregulation of Notch and/or FOXO signalling in Dahomey may be responsible for driving mitotype-specific differences in development [127, 128].
These data further question whether mtDNA can be assumed to accurately reflect species or population-level demographic processes when the dietary protein to carbohydrate ratio varies over time or space. It is now well documented that purifying selection affects the variability of mtDNA encoded genes, and the purging of deleterious variants will result in the removal of linked variants through background selection. In humans, deleterious mtDNA mutations are well-known [41–43], and evidence for a profound effect of accumulated mutations on men’s health has been reported [191]. Purifying selection has been demonstrated in the female mouse germline [39, 40] and in Drosophila slightly deleterious mutations have been reported [26, 36–38]. Evidence of positive selection on mitogenomes has been reported [27, 52], but to our knowledge, no specific mutation has been experimentally shown to result in an evolutionary advantage. Our observation that distinct mitotypes reached high frequency when fed different macronutrient ratios in population cages suggests that diet may also be a strong selective force in nature. Here, we advocate future studies test for selection on mtDNA within and among naturally occurring populations where macronutrients change over time and space.
The influence of diet is extensively studied in the literature but few studies investigate genotype-by-diet interactions and fewer still that have unravelled the underlying mechanisms [3–7, 51, 143, 192]. One prediction from these data is that experimentally increasing the P:C ratio (i.e., 1:20 P:C) may further increase the dietary-induced metabolic stress and cause increased mortality in larvae harbouring Dahomey mtDNA. Conversely, development time in Alstonville larvae may decrease if the polyol pathway is upregulated. Experimentally, such a dietary perturbation would be outside the range of P:C ratios encountered by Drosophila in nature, but perhaps would reflect the human genomes clash with modern life and the vending machine.
Most common mtDNA mutations are thought to be deleterious and involved in a variety mitochondriopathies and complex diseases like diabetes, cardiovascular disease, gastrointestinal disorders, skin disorders and elevated blood pressure [e.g. 193, 194–197]. Further, the accumulation of somatic mtDNA mutations likely influences primary cancers and the ageing process [198, 199]. The data presented here suggest that it is also possible that slightly deleterious mtDNA mutations may confer an advantage in certain situations. Our data, therefore, support matching an individual’s diet to their mitotype as an approach to treating mitochondriopathies, complex diseases or even for optimising health in non-disease populations. For example, were the same mechanisms found to occur in humans the enhanced lipogenesis in individuals with mild complex I mutations could make them more venerable to obesity when eating a high-carbohydrate diet, yet less susceptible to Parkinson’s disease, which has been linked to defects in lipogenesis [200].
Here, RNA and metabolites were extracted from female third instar wandering larvae sourced from the side of the bottle that had developed on 1:2 and 1:16 P:C diets [108, 208]. In uncrowded conditions, on a fixed light/dark regimen, larval wandering is highly synchronous and begins some 24 h before pupation (at 25° C) [108].
To test specific hypotheses, we replaced sucrose as the dietary sugar. The 1:16 P:C diet was prepared without the addition of sucrose. Then, 200 ml of food was combined with 1.87 g of either sucrose (Sigma S0389) as the control, sorbitol (Sigma S1876), fructose (Sigma F0127), mannose (Sigma M6020), fucose (Sigma F2252) or xylose (Sigma X3877). Each new diet was poured into 8 bottles. Equal amounts of eggs harbouring Alstonville or Dahomey mtDNA were added to each food and microbiome was added after 2 d. Flies were kept at 23° C, 50% humidity on a 12 h light/dark cycle. Emerging adult female flies were counted over 3 d, and percentage eclosion of each mitotype was calculated (dx.doi.org/10.17504/protocols.io.rtfd6jn).
For inhibitors, freshly prepared aldose reductase (polyol pathway) inhibitor (Epalrestat, Sigma SML0527) and carnitine palmitoyltransferase-1 inhibitor (Etomoxir, Sigma E1905) were solubilised in water to make a 5 mM stock. The stock solutions were added to the 1:16 P:C diet to final concentrations of 25 μM Epalrestat, 12.5 μM Etomoxir. Methodology followed that described above for 2-Deoxy-D-Glucose.
Unless otherwise stated, all data are biological replicates and statistically analysed by ANOVA followed by Student’s t-tests to determine difference (JMP software 12, SAS Institute, NC, USA). Biological replicates are parallel measurements of biologically distinct samples. Where the numbers of Dahomey larvae eclosing in 3 d was compared between dietary sugars (sorbitol, fructose, mannose, fucose, xylose and gluconate) and the control diet with the diets supplemented with an inhibitor (Epalrestat, and Etomoxir)we conducted Dunnett’s tests. Data were checked for normality using a Shapiro-Wilks W test and outliers removed before statistical analyses using box plots. Values that were greater than ± 1.5 interquartile range were categorised as an outlier and excluded from the data set. No statistical methods were used to predetermine sample size.
|
10.1371/journal.pntd.0006993 | Vector competence of biting midges and mosquitoes for Shuni virus | Shuni virus (SHUV) is an orthobunyavirus that belongs to the Simbu serogroup. SHUV was isolated from diverse species of domesticated animals and wildlife, and is associated with neurological disease, abortions, and congenital malformations. Recently, SHUV caused outbreaks among ruminants in Israel, representing the first incursions outside the African continent. The isolation of SHUV from a febrile child in Nigeria and seroprevalence among veterinarians in South Africa suggests that the virus may have zoonotic potential as well. The high pathogenicity, extremely broad tropism, potential transmission via both biting midges and mosquitoes, and zoonotic features warrants prioritization of SHUV for further research. Additional knowledge is essential to accurately determine the risk for animal and human health, and to assess the risk of future epizootics and epidemics. To gain first insights into the potential involvement of arthropod vectors in SHUV transmission, we have investigated the ability of SHUV to infect and disseminate in laboratory-reared biting midges and mosquitoes.
Culicoides nubeculosus, C. sonorensis, Culex pipiens pipiens, and Aedes aegypti were orally exposed to SHUV by providing an infectious blood meal. Biting midges showed high infection rates of approximately 40–60%, whereas infection rates of mosquitoes were lower than 2%. SHUV successfully disseminated in both species of biting midges, but no evidence of transmission in orally exposed mosquitoes was found.
The results of this study show that different species of Culicoides biting midges are susceptible to infection and dissemination of SHUV, whereas the two mosquito species tested were found not to be susceptible.
| Arthropod-borne (arbo)viruses are notorious for causing unpredictable and large-scale epidemics and epizootics. Apart from viruses such as West Nile virus and Rift Valley fever virus that are well known to have a significant impact on human and animal health, many arboviruses remain neglected. Shuni virus (SHUV) is a neglected virus with zoonotic potential that was recently associated with severe disease in livestock and wildlife. Isolations of SHUV from field-collected biting midges and mosquitoes suggests that SHUV may be transmitted by these insects. Laboratory-reared biting midge species (Culicoides nubeculosus and C. sonorensis) and mosquito species (Culex pipiens pipiens and Aedes aegypti), that are known to transmit other arboviruses, were exposed to SHUV via an infectious blood meal. SHUV was able to successfully disseminate in both biting midge species, whereas no evidence of infection or transmission in both mosquito species was found. Our results show that SHUV infects and disseminates in two different Culicoides species, suggesting that these insects could play an important role in the disease transmission cycle.
| Arthropod-borne (arbo)viruses continue to pose a threat to human and animal health [1, 2]. In particular the order Bunyavirales comprises emerging pathogens such as Crimean-Congo haemorrhagic fever virus (CCHFV) and Rift Valley fever virus (RVFV) [3, 4]. The World Health Organization (WHO) has included both CCHFV and RVFV to the “Blueprint” list of ten prioritized viruses likely to cause future epidemics and for which insufficient countermeasures are available [5]. In the veterinary field, prioritized viral diseases of animals, including RVFV, are notifiable to the World Organization for Animal Health (Office International des Epizooties, OIE). Apart from pathogens that are recognised as major threats by WHO and OIE, many have remained largely neglected. Before the turn of the century, West Nile virus, chikungunya virus, and Zika virus were among these neglected viruses until they reminded us how fast arboviruses can spread in immunologically naïve populations [2]. Although these outbreaks came as a surprise, in hindsight, smaller outbreaks in previously unaffected areas could have been recognised as warning signs.
Shuni virus (SHUV; family Peribunyaviridae, genus Orthobunyavirus, Simbu serogroup) recently emerged in two very distant areas of the world [6]. SHUV was isolated for the first time from a slaughtered cow in the 1960s in Nigeria [7]. During subsequent years, the virus was isolated on several occasions from domestic animals including cattle, sheep, goats, and horses [7–10], from wild animals including crocodiles and rhinoceros [10], and from field-collected Culicoides biting midges and mosquitoes [8, 11, 12]. More recently, SHUV was associated with malformed ruminants in Israel [13, 14]. Emergence of SHUV in areas outside Sub-Saharan Africa shows the potential of this virus to spread to new areas, and increases the risk for SHUV outbreaks in bordering territories such as Europe. Isolation of SHUV from a febrile child and detection of antibodies in 3.9% of serum samples from veterinarians in South Africa shows that SHUV can infect humans as well, although its ability to cause human disease is still uncertain [7, 15, 16].
Proper risk assessments rely on accurate knowledge of disease transmission cycles. Arbovirus transmission cycles can only become established when competent vectors and susceptible hosts encounter under suitable climatic conditions. Although SHUV has been isolated from pools of field-collected Culicoides biting midges and mosquitoes [7, 11, 12], the role of both insect groups as actual vectors remains to be confirmed. Detection of virus in field-collected insects is not sufficient to prove their ability to transmit the virus. Arboviruses need to overcome several barriers (i.e. midgut and salivary gland barriers) inside their vector, before they can be transmitted [17, 18]. In addition to virus isolation from field-collected vectors, laboratory studies are therefore needed to experimentally test the ability of blood-feeding insects to become infected with, maintain, and successfully transmit arboviruses (i.e., vector competence) [19]. To gain insights into the potential of Culicoides biting midges and mosquitoes to function as vectors of SHUV, we studied the susceptibility of four main arbovirus vector species (Culicoides nubeculosus and C. sonorensis biting midges, and Culex pipiens biotype pipiens and Aedes aegypti mosquitoes) for SHUV.
African green monkey kidney cells (Vero E6; ATCC CRL-1586) were cultured in Eagle’s minimum essential medium (Gibco, Carlsbad, CA, United States) supplemented with 5% fetal bovine serum (FBS; Gibco), 1% non-essential amino acids (Gibco), 1% L-glutamine (Gibco), and 1% antibiotic/antimycotic (Gibco). Cells were cultured as monolayers and maintained at 37°C with 5% CO2.
Vero E6 cells that were used in biting midge and mosquito infection experiments in the biosafety level 3 (BSL3) facility were cultured in Dulbecco's modified Eagle medium (Gibco) supplemented with 10% FBS, penicillin (100 U/ml; Sigma-Aldrich, Saint Louis, MO, United States), and streptomycin (100 μg/ml; Sigma-Aldrich). Prior to infections in the BSL3 facility, Vero E6 cells were seeded in 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid-buffered DMEM medium (HEPES-DMEM; Gibco) supplemented with 10% FBS, penicillin (100 U/ml), and streptomycin (100 μg/ml), fungizone (50 μg/ml; Invitrogen, Carlsbad, United States), and gentamycin (50 μg/ml; Gibco).
C6/36 cells (ATCC CRL-1660), derived from Ae. albopictus mosquitoes, were cultured in Leibovitz-15 (L-15) growth medium (Sigma-Aldrich) supplemented with 10% FBS, 2% Tryptose Phosphate Broth (Gibco), 1% non-essential amino acids solution, and 1% antibiotic/antimycotic. Cells were cultured as monolayers and incubated at 28°C in absence of CO2.
KC cells, derived from embryos of colonized C. sonorensis biting midges [20], were cultured as monolayers in modified Schneider’s Drosophila medium (Lonza, Basel, Switzerland) with 15% FBS, and 1% antibiotic/antimycotic at 28°C in absence of CO2.
SHUV (strain An10107, P2 Vero, 1980) was kindly provided by the World Reference Center for Emerging Viruses and Arboviruses (WRCEVA). The virus was originally isolated from the blood of a slaughtered cow in 1966 in Nigeria by inoculation of neonatal mice, and passaged twice in Vero cells [21]. The passage 3 (P3) stock was generated by inoculation of Vero E6 cells with the P2 stock at a multiplicity of infection (MOI) of 0.001. The supernatant was harvested at 6 days post inoculation, centrifuged, and stored in aliquots at -80°C. The P4 stock was generated by inoculating Vero E6 cells at MOI 0.01 using the P3 stock. At this MOI, full cytopathic effect (CPE) was present at 3 days post infection. Virus titers were determined using endpoint dilution assays (EPDA) on Vero E6 cells [22]. Titers were calculated using the Spearman-Kärber algorithm and expressed as 50% tissue culture infective dose (TCID50) [23, 24]. The virus detection and titration procedure was validated using a SHUV-specific reverse transcriptase quantitative PCR (RT-qPCR; S1 Supporting Information).
Cells were seeded in T25 cell culture flasks at densities of 7.5 × 105 (Vero E6), 1.5 × 106 (C6/36), or 2.5 × 106 (KC cells) per flask in 10 ml complete medium. After overnight incubation, the flasks were inoculated with SHUV at an MOI of 0.01 (P4 stock). The MOI calculation for each cell line was based on the virus titer that was determined on Vero E6 cells. One hour after inoculation, the medium was removed and replaced with fresh medium. At time points 0 (sample taken directly after medium replacement), 24, 48, and 72 h post infection, 200 μl samples were taken and stored at -80°C for later analysis. For each cell line, virus titers were determined in triplicate per time point by EPDA using Vero E6 cells, which showed distinct CPE [22].
Culicoides nubeculosus were kindly provided by The Pirbright Institute, Pirbright laboratories, United Kingdom, in 2012 [25], and were maintained at 23°C with 16:8 light:dark cycle and 60% relative humidity. Culicoides sonorensis were kindly provided by the Arthropod-Borne Animal Diseases Research Laboratory, USDA-ARS (courtesy of Dr. Barbara Drolet) in 2017 [26], and were maintained at 25°C with 16:8 light:dark cycle and 70% relative humidity. Similar rearing protocols were used for both biting midge species. Eggs were transferred to square larval holding trays (C. nubeculosus: 25 x 25 x 8 cm, Kartell, Noviglio, Italy; C. sonorensis: 19 x 19 x 20 cm, Jokey, Wipperfürth, Germany) with filter wool (Europet Bernina International, Gemert-Bakel, The Netherlands) attached with double-sided tape to the bottom. Trays were filled with tap water, a few millilitres of rearing water in which larvae had completed their life cycle, and two drops of Liquifry No.1 (Interpet, Dorking, United Kingdom). Larvae were fed with a 1:1:1 mixture of bovine liver powder (MP biomedicals, Irvine, CA, US), ground rabbit food (Pets Place, Ede, The Netherlands), and ground koi food (Tetra, Melle, Germany). Culicoides nubeculosus larvae were additionally fed with nutrient broth No. 2 (Oxoid, Hampshire, UK). Pupae were transferred to plastic buckets (diameter: 12.2 cm, height: 12.2 cm; Jokey) and closed with netting on the top through which the biting midges could feed. Emerged adults were provided with 6% glucose solution ad libitum. Cow blood (Carus, Wageningen, The Netherlands) was provided through a Parafilm M membrane using the Hemotek PS5 feeding system (Discovery Workshops, Lancashire, United Kingdom) for egg production.
The Cx. pipiens pipiens colony was established in the laboratory from egg rafts collected in the field in The Netherlands during August 2016. Egg rafts were individually hatched in tubes. Pools of approximately 10 first instar larvae were identified to the biotype level using real-time PCR [27]. The colony was started by grouping larvae from 93 egg rafts identified as the pipiens biotype. Mosquitoes were maintained at 23°C with 16:8 light:dark cycle and 60% relative humidity [28, 29]. Adult mosquitoes were kept in Bugdorm-1 rearing cages and maintained on 6% glucose solution ad libitum. Cow blood or chicken blood (Kemperkip, Uden, The Netherlands) was collected in BC Vacutainer lithium heparin-coated blood collection tubes (Becton Dickinson, Breda, The Netherlands), and stored at 4°C. Blood was provided through a Parafilm M membrane using the Hemotek PS5 feeding system for egg production. Egg rafts were transferred to square larval holding trays (25 x 25 x 8 cm, Kartell) filled with tap water and two drops of Liquifry No. 1. Hatched larvae were fed with a 1:1:1 mixture of bovine liver powder, ground rabbit food, and ground koi food. Pupae were collected every 2 days and placed in Bugdorm-1 insect rearing cages.
Aedes aegypti mosquitoes from the Rockefeller strain (Bayer AG, Monheim, Germany) were used in all experiments. The mosquito colony was maintained as described before [30]. In short, mosquitoes were maintained at 27°C with 12:12 light:dark cycle and 70% relative humidity. Adult mosquitoes were kept in Bugdorm-1 rearing cages and maintained on 6% glucose solution ad libitum. Human blood (Sanquin Blood Supply Foundation, Nijmegen, The Netherlands) was provided through a Parafilm M membrane using the Hemotek PS5 feeding system for egg production. Eggs were transferred to transparent square larval holding trays (19 x 19 x 20 cm, Jokey), filled for approximately one-third with tap water and three drops of Liquifry No. 1. Hatched larvae were fed with Tetramin Baby fish food (Tetra). Larval trays were closed with fine-meshed netting, to allow adult mosquitoes to emerge inside larval trays. Twice a week, adults were aspirated from the larval trays and collected in Bugdorm-1 insect rearing cages.
Groups of adult C. nubeculosus (1–7 days old), C. sonorensis (1–11 days old), Cx. p. pipiens (4–20 days old), and Ae. aegypti (4–7 days old) were transferred to plastic buckets (diameter: 12.2 cm, height: 12.2 cm; Jokey) and closed with netting before being taken to the BSL3 facility. Culex p. pipiens mosquitoes were kept on water for 3 days, whereas the other species were maintained on 6% glucose solution until being offered an infectious blood meal. SHUV P3 stock with a mean titer of 3.0 x 106 TCID50/ml was mixed 1:1 with cow blood. The used cow blood was tested negative for Schmallenberg virus (SBV) antibodies, to prevent cross-neutralisation with SHUV. The infectious blood meal was provided through a Parafilm M membrane using the Hemotek PS5 feeding system, under dark conditions at 24°C and 70% relative humidity.
After 1 h, insects were anesthetized with 100% CO2 and kept on a CO2-pad to select fully engorged females. For each species, five fully engorged females were directly stored at -80°C for each replicate. These samples were used to determine the ingested amounts of SHUV for each species. All remaining and fully engorged females were placed back into buckets with a maximum group size of 110 individuals per species per bucket. All insects were provided with 6% glucose solution via a soaked ball of cotton wool on top of the netting ad libitum. Culicoides sonorensis and Ae. aegypti were kept at 28°C for 10 days, whereas C. nubeculosus and Cx. p. pipiens were kept at 25°C for 10 days. These temperatures were selected for optimal replication of the virus, and to reflect differences in the rearing temperature for each species. Three replicate experiments of C. nubeculosus (N1 = 84, N2 = 82, N3 = 77, Ntotal = 243), C. sonorensis (N1 = 9, N2 = 9, N3 = 30, Ntotal = 48), and Cx. p. pipiens (N1 = 89, N2 = 57, N3 = 65, Ntotal = 211) were carried out, and two replicate experiments of Ae. aegypti (N1 = 72, N2 = 77, Ntotal = 149). During each replicate, biting midges and mosquitoes were fed in parallel with the same infectious blood meal.
Adult female Cx. p. pipiens (3–9 days old) and Ae. aegypti (4–6 days old) mosquitoes were injected with SHUV into the thorax to investigate the role of mosquito barriers on dissemination of SHUV. Mosquitoes were anesthetized with 100% CO2 and positioned on the CO2-pad. Female mosquitoes were intrathoracically injected with 69 nl of SHUV (P3 stock with a titer of 3.0 x 106 TCID50/ml) using a Drummond Nanoject II Auto-Nanoliter injector (Drummond Scientific, Broomall, Unites States). Injected Cx. p. pipiens were maintained at 25°C and injected Ae. aegypti were maintained at 28°C. Mosquitoes were incubated for 10 days at the respective temperatures, and had access to 6% glucose solution ad libitum. Injections were done during a single replicate experiment for Cx. p. pipiens (N = 50) and Ae. aegypti (N = 50).
After 10 days of incubation at the respective incubation temperatures, samples from surviving biting midges and mosquitoes were collected. Biting midges were anesthetized with 100% CO2 and transferred individually to 1.5 ml Safe-Seal micro tubes (Sarstedt, Nümbrecht, Germany) containing 0.5 mm zirconium beads (Next Advance, Averill Park, NY, United States). For a selection of C. nubeculosus (N = 77) and C. sonorensis (N = 30) from one replicate experiment, heads were removed from bodies and separately stored in tubes. All samples were stored at -80°C until further processing.
Mosquitoes were anesthetized with 100% CO2 to remove legs and wings. Mosquito saliva was then collected by inserting the proboscis into a 200 μl yellow pipette tip (Greiner Bio-One) containing 5 μl of a 1:1 solution of 50% glucose solution and FBS. The saliva sample was transferred to a 1.5 ml micro tube containing 55 μl of fully supplemented HEPES-DMEM medium. Mosquito bodies were individually stored in 1.5 ml Safe-Seal micro tubes containing 0.5 mm zirconium beads.
Frozen biting midge and mosquito tissues were homogenized for 2 min at maximum speed (setting 10) in the Bullet Blender Storm (Next advance), centrifuged for 30 seconds at 14,500 rpm in the Eppendorf minispin plus (Eppendorf, Hamburg, Germany), and suspended in 100 μl of fully supplemented HEPES-DMEM medium. After addition of the medium, samples were blended again for 2 min at maximum speed, and centrifuged for 2 min at 14,500 rpm. Mosquito saliva samples were thawed at RT and vortexed before further use. In total 30 μl of each body or saliva sample was inoculated on a monolayer of Vero E6 cells in a 96 wells plate. SHUV stock or infectious blood mixture was included as positive control and wells to which no sample was added were included as negative controls. After 2–3 h the inoculum was removed and replaced by 100 μl of fully supplemented HEPES-DMEM medium. Wells were scored for virus induced CPE at 3 and 7 days post inoculation, with full CPE being observed at the latter time point. Afterwards, virus titers for positive samples of biting midge bodies and heads, as well as mosquito bodies and saliva were determined with single EPDA on Vero E6 cells [30]. Virus titers were determined using the Reed & Muench algorithm [31]. A subset of samples was validated by RT-qPCR, to confirm that observed CPE was induced by SHUV (S1 Supporting Information).
Infection rate (virus-infected whole body) and dissemination efficiency (virus-infected head) were determined for biting midges, whereas infection rate (virus-infected whole body) and transmission efficiency (virus-infected saliva) were determined for mosquitoes. Infection rate, dissemination efficiency, and transmission efficiency were calculated, respectively, by dividing the number of females with virus-infected bodies (infection), virus-infected heads (dissemination), or virus-infected saliva (transmission) by the total number of females tested in the respective treatment and that survived the incubation period. The values were subsequently expressed as percentages by multiplying with 100. Two biting midge samples of which only the head was virus-positive, but not the body, were considered to be uninfected.
Mammalian, mosquito, and midge cells were inoculated with SHUV to gain insight into the replicative fitness of this virus and strain in different host cell types. The results show that SHUV is capable to produce progeny in all three cell types (Fig 1 and S1 Data). Of note, a strong CPE was observed in the Vero E6 cells upon infection whereas no CPE was observed in the insect cell lines. Therefore, Vero E6 cells were used to determine titers by EPDA.
To evaluate the susceptibility of two species of biting midges (C. nubeculosus and C. sonorensis) for SHUV, groups of individuals of both species were orally exposed to an infectious blood meal with a mean SHUV titer of 3.0 x 106 TCID50/ml. SHUV titers of ingested blood were determined for a selection of 10 fully engorged females for each species, that were directly stored at -80°C after feeding. Both species ingested low amounts of SHUV that were below the detection limit of the endpoint dilution assay of 103 TCID50/ml.
Infection rates were also determined after 10 days of incubation at temperatures of 25°C (C. nubeculosus and Cx. p. pipiens) or 28°C (C. sonorensis and Ae. aegypti; Fig 2 and S2 Data). Both biting midge species showed high infection rates of 44% for C. nubeculosus (N = 243), and 60% for C. sonorensis (N = 48; Fig 2A). SHUV replicated to median titers of 2.4 x 103 TCID50/ml in body samples of C. nubeculosus and 1.1 x 104 TCID50/ml in body samples of C. sonorensis (Fig 2E). For one replicate experiment, heads were separated from the bodies and tested for presence of SHUV to assess whether the virus successfully passed from the midgut to the haemocoel, indicative of dissemination throughout the body. Dissemination efficiencies were 18% (N = 77) for C. nubeculosus and 10% (N = 30) for C. sonorensis (Fig 2C). In all virus-positive heads that induced CPE, SHUV titers were lower than 103 TCID50/ml. Because only very low amounts of SHUV were detected in biting midge heads, the actual percentage of disseminated infections might be higher. A subset of the samples was additionally tested by RT-qPCR to confirm that CPE was induced by SHUV (S1 Supporting Information). The relatively high infection rates and dissemination efficiencies observed in this study and the absence of a salivary glands barrier in biting midges as shown in previous studies [17, 32], suggests that both C. nubeculosus and C. sonorensis have the potential to transmit SHUV.
SHUV was previously isolated from field-collected mosquitoes [8]. Therefore, we determined vector competence for two mosquito species (Cx. p. pipiens and Ae. aegypti) which are important vectors for several arboviruses [22, 28, 30]. SHUV titers of ingested blood were determined for a selection of 10 fully engorged female mosquitoes that were directly stored at -80°C after feeding on an infectious blood meal with a SHUV titer of 3.0 x 106 TCID50/ml. Similar to results obtained with the biting midges, the amounts of SHUV ingested by both mosquito species was less than 103 TCID50/ml.
No SHUV infection was observed in the Cx. p. pipiens mosquitoes (N = 211) following oral exposure, whereas infection rates of 2% were found for orally exposed Ae. aegypti mosquitoes (N = 149; Fig 2B). SHUV replicated to median titers of 6.3 x 103 TCID50/ml in body samples of Ae. aegypti (Fig 2F), which was comparable to titers found in biting midges. No SHUV was detected in any of the saliva samples taken from either Cx. p. pipiens or Ae. aegypti (Fig 2D). Thus, SHUV was able to successfully infect a small proportion of Ae. aegypti mosquitoes but not Cx. p. pipiens, and no evidence was found for transmission of SHUV by mosquitoes.
The very low infection rates of mosquitoes triggered further investigation into potential mosquito barriers against SHUV infection. To this end, Cx. p. pipiens and Ae. aegypti mosquitoes were intrathoracically injected with SHUV, to bypass the potential midgut barrier. Direct injection of SHUV into the thorax resulted in high infection rates of 70% for Cx. p. pipiens (N = 50), and 100% for Ae. aegypti (N = 50; Fig 3A). Transmission efficiency of 32% (N = 50) was found for Cx. p. pipiens and 8% (N = 50) for Ae. aegypti (Fig 3B). Interestingly, although infection rates of Cx. p. pipiens were below 100%, we found a relatively high transmission efficiency. This may indicate a relatively weaker salivary gland barrier in Cx. p. pipiens compared to Ae. aegypti mosquitoes that had 100% infection rate, but relatively low transmission efficiency.
To gain more insight in replication of SHUV, virus titers were determined for virus-infected mosquito body and saliva samples. Titers of virus-infected Cx. p. pipiens body samples were almost all below the detection limit of 103 TCID50/ml of the endpoint dilution assay (Fig 3C). This indicates that even when SHUV is injected into the thorax, there is no productive virus replication. In contrast, we found median titers of 7.1 x 104 TCID50/ml for virus-infected Ae. aegypti body samples. This shows that SHUV is able to successfully replicate in Ae. aegypti when the midgut barrier is bypassed. In the majority of mosquito saliva samples, SHUV titers were less than 103 TCID50/ml (Fig 3D). Taken together, SHUV is able to disseminate in mosquitoes, but both the midgut and salivary glands form a barrier for SHUV.
SHUV was previously isolated from field-collected pools of Culicoides biting midges and from mosquitoes, but their involvement in SHUV transmission remained to be confirmed [8, 11, 12]. Here, we show for the first time that SHUV is able to infect and replicate in biting midges as well as in mosquitoes, but only the biting midge species evaluated in the present study can be considered highly susceptible to infection.
Both C. nubeculosus and C. sonorensis showed high infection rates of 44% and 60% when incubated for 10 days at 25°C and 28°C, respectively. It has been demonstrated that a salivary gland barrier is absent for Orbiviruses and Schmallenberg virus in biting midges [17, 32]. This knowledge, in combination with evidence of successful dissemination of SHUV to the heads indicates that the biting midge species evaluated in the present study are likely competent vectors of SHUV. Importantly, the finding that SHUV replicates efficiently in two biting midge species from a different geographic background suggests that various species of Culicoides may function as vectors of SHUV.
SHUV infection and replication in biting midges seems more efficient compared to other biting midge-borne viruses such as SBV and bluetongue virus (BTV), which generally show infection rates up to 30% [32–36]. Both SBV and BTV have caused sudden and large-scale epizootics in Europe, with devastating consequences for the livestock sector [37, 38]. The relatively high susceptibility and efficiency of replication in biting midges, and recent spread of SHUV to areas outside Sub-Saharan Africa [13], should therefore be interpreted as a warning for its epizootic potential.
In contrast to the high infection rates in biting midges, only few orally exposed Ae. aegypti mosquitoes became infected with SHUV during 10 days of incubation at 28°C. In addition, no evidence of successful dissemination to the salivary glands of the two mosquito species was found. SHUV replication and transmission (8%) was observed when the virus was directly injected into the thorax of Ae. aegypti mosquitoes. This indicates that both the midgut infection barrier and the salivary gland barrier prevent infection and subsequent transmission of SHUV by Ae. aegypti mosquitoes. Of the Cx. p. pipiens mosquitoes that were orally exposed to SHUV, none became infected during 10 days of incubation at 25°C. Moreover, replication of SHUV was low in Cx. p. pipiens, as evidenced by low titers when it was directly injected into the thorax. However, a relatively high percentage of mosquito saliva samples contained SHUV. We therefore conclude that the midgut barrier is the main barrier that prevents infection of Cx. p. pipiens with SHUV. Our findings are in line with an earlier study on the closely-related SBV, which showed no evidence for involvement of Cx. pipiens in virus transmission, although SBV was able to infect Cx. pipiens mosquitoes [39]. However, as Cx. theileri has been identified as a vector of several other bunyaviruses, this mosquito may also be a possible vector of SHUV [40, 41]. Thus, vector competence studies with additional mosquito species collected from the field are needed to fully understand the possible role of mosquitoes in natural transmission cycles of SHUV.
In this study, we determined infection, dissemination, and transmission of SHUV by infectivity assays and virus titers by EPDA (i.e. assays based on inoculation of samples on Vero cells which are then screened for CPE). Such infectivity assays and EPDAs have the advantage of detecting infectious virus particles, whereas other methods like qPCR that quantify genome equivalents, may include defective virus particles and thereby not accurately represent infectious virus. Of note, observed CPE in the infectivity assays and EPDAs was found to invariably correspond with SHUV RNA as determined by RT-qPCR (S1 Supporting Information).
Recent outbreaks of SBV and BTV exemplified the tremendous impact of midge-borne viruses on animal health [37, 38]. Our study demonstrates highly efficient infection, replication, and dissemination of SHUV in two biting midge species (C. nubeculosus and C. sonorensis). However, conclusive evidence for SHUV transmission by biting midges should be provided by experiments with infected biting midges and susceptible mammals, although these kind of experiments are costly and complex. We cannot exclude that results obtained with laboratory-reared vectors are different from those obtained with field-collected vectors. Therefore, future studies should test vector competence of field-collected Culicoides biting midge and mosquito species exposed to different quantities of SHUV, to more accurately predict the risk of SHUV transmission in specific areas. These experiments in combination with behavioural and ecological research will contribute to our understanding of the transmission cycle of SHUV.
|
10.1371/journal.pgen.1000528 | Comprehensive Linkage and Association Analyses Identify Haplotype, Near to the TNFSF15 Gene, Significantly Associated with Spondyloarthritis | Spondyloarthritis (SpA) is a chronic inflammatory disorder with a strong genetic predisposition dominated by the role of HLA-B27. However, the contribution of other genes to the disease susceptibility has been clearly demonstrated. We previously reported significant evidence of linkage of SpA to chromosome 9q31–34. The current study aimed to characterize this locus, named SPA2. First, we performed a fine linkage mapping of SPA2 (24 cM) with 28 microsatellite markers in 149 multiplex families, which allowed us to reduce the area of investigation to an 18 cM (13 Mb) locus delimited by the markers D9S279 and D9S112. Second, we constructed a linkage disequilibrium (LD) map of this region with 1,536 tag single-nucleotide polymorphisms (SNPs) in 136 families (263 patients). The association was assessed using a transmission disequilibrium test. One tag SNP, rs4979459, yielded a significant P-value (4.9×10−5). Third, we performed an extension association study with rs4979459 and 30 surrounding SNPs in LD with it, in 287 families (668 patients), and in a sample of 139 cases and 163 controls. Strong association was observed in both familial and case/control datasets for several SNPs. In the replication study, carried with 8 SNPs in an independent sample of 232 cases and 149 controls, one SNP, rs6478105, yielded a nominal P-value<3×10−2. Pooled case/control study (371 cases and 312 controls) as well as combined analysis of extension and replication data showed very significant association (P<5×10−4) for 6 of the 8 latter markers (rs7849556, rs10817669, rs10759734, rs6478105, rs10982396, and rs10733612). Finally, haplotype association investigations identified a strongly associated haplotype (P<8.8×10−5) consisting of these 6 SNPs and located in the direct vicinity of the TNFSF15 gene. In conclusion, we have identified within the SPA2 locus a haplotype strongly associated with predisposition to SpA which is located near to TNFSF15, one of the major candidate genes in this region.
| Spondyloarthritis (SpA) is a common variety of articular inflammatory disorder characterized by axial and/or peripheral arthritis, frequently associated with extra-articular manifestations such as psoriasis, uveitis, and inflammatory bowel diseases (ulcerative colitis or Crohn's disease (CD)). SpA is a complex disorder with high heritability. The MHC class I HLA-B27 allele is a very strong risk factor for its development, but other genetic factors located outside the MHC also play a role in disease susceptibility. By a previous whole-genome linkage investigation, we have demonstrated that a region located on the chromosome 9q31–34 was involved in SpA susceptibility. The present study aimed to further characterize this locus. Using a stepwise linkage and association approach, we identified a haplotype spanning 6 single-nucleotide polymorphisms strongly associated with SpA and located in a genomic region paralogous to the MHC, near to the TNFSF15 gene. Interestingly, polymorphisms of this gene have previously been shown to be associated with CD. This original finding offers a new research track for the understanding of SpA pathophysiology, which is still poorly understood, as well as new hope for diagnostic and therapeutic innovation.
| Spondyloarthritis (SpA) is one of the most frequent varieties of articular inflammatory disorders with an estimated prevalence of 0.3% in the western European adult population [1]. It is characterized by a predominant axial skeleton inflammation, by a frequent occurrence of enthesitis and peripheral arthritis, and also by a high rate of extra-articular features, the most characteristic of which are acute anterior uveitis, psoriasis, and inflammatory bowel diseases (such as ulcerative colitis or Crohn's disease (CD)) [2]. Depending on its clinical features, SpA is classically subdivided into the following subsets: ankylosing spondylitis (AS), which is the prototypical form characterized by predominant axial skeletal involvement and advanced radiographic sacroiliitis, psoriatic arthritis (PsA), arthritis associated with inflammatory bowel disease (AIBD), reactive arthritis (ReA), and undifferentiated SpA (uSpA). Familial aggregation among these conditions has been well established. Notably, we have previously shown, by analyzing a large number of pedigrees with multiple cases of SpA, that all subtypes are likely to be determined by a core set of predisposing factors and may therefore be studied together in genetic studies [3]–[5].
The HLA-B27 allele is the first genetic factor which was demonstrated to be associated with AS [6],[7] and other SpA [4],[5],[8]. Although about 80% of Caucasian patients are HLA-B27 positive, as compared to only 6–8% in the general population, the exact mechanism for this association remains poorly understood [9].
Family and twin studies have demonstrated additional non-MHC susceptibility regions elsewhere in the genome [10]. For example, concordance rates for HLA-B27 positive monozygotic twins are twice as high as the concordance rates for HLA-B27 positive dizygotic twins [10]. Furthermore, the involvement of genetic factors arising from outside the HLA region is suggested by the large non-HLA component of the relative recurrence risk for the SpA estimated in sib-pairs (λnon-HLA). Indeed, if the overall relative recurrence risk in sibling (λs) has been estimated to be 40 [11], estimates of the λs component attributable to the HLA region (λHLA), based on previous affected sib-pairs linkage analyses, ranges from 5.2 to 6.25 [12],[13]. Variants in several genes such as the IL-1 family gene cluster [14],[15], IL-23R [16], and ARTS1/ERAP1 [16], have recently been reported to be associated with AS based on a candidate-gene approach [14],[15] or a non-synonymous single-nucleotide polymorphisms (SNPs) genome-wide association study [16].
Our team has previously reported results of the first genome-wide linkage screen and its extension study performed in SpA [13]. Overall, 893 individuals from 120 multiplex families (families with several patients) comprising 336 affected relative pairs have been genotyped in this study. Non parametric multipoint linkage analysis of the whole dataset yielded evidence for significant linkage to the chromosomal region 9q31–34 (NPLmax = 4.87, P = 2×10−5). This locus overlapped with one of those identified by the genome-wide linkage screen performed in AS by a group from Oxford [12]. We named this new susceptibility location SPA2, in reference to the MHC locus, which we considered as SPA1 [13]. SPA2 encompasses a 23.95 cM region (17.44 Mb) containing 85 genes and predicted coding sequences as well as 110 pseudogenes. This locus is very appealing with regard to SpA susceptibility. First of all, it is one of three genomic regions paralogous to the MHC, which is the major SpA susceptibility region [17],[18]. Furthermore, it is syntenic to the Pgis2 susceptibility locus mapped in a murine model of SpA [19]. Within its borders SPA2 contains both the TNFSF15 gene found to be associated with CD a condition belonging to the SpA spectrum [20],[21], and the TRAF1-C5 locus associated with rheumatoid arthritis another inflammatory rheumatic disease [22],[23].
The goal of the present study was to identify variants associated with the disease and located in the SPA2 locus. Using a comprehensive four-step linkage and association study in a total of 287 families including 668 affected individuals, followed by an independent case/control analysis (2 samples including a total of 371 cases and 312 controls), we identified a strongly associated six-SNPs haplotype, located at 28.6 kb from the TNFSF15 gene.
The initial step of our study aimed to refine the linkage signal in the 23.95 cM (17.44 Mb) SPA2 locus. To realise this investigation we selected a fine-grained set of 28 microsatellite markers (more than one marker per cM). These markers were genotyped in 149 independent multiplex families (including the 120 families studied in our initial genome-screen) [13] (Figure 1B) consisting of 1,065 individuals including 458 affected with SpA (Figure 1A, Table 1).
Non parametric multipoint linkage analysis allowed us to identify two prominent linkage peaks yielding a significant Zlr value>2.91 (nominal P<1.79×10−3) corresponding to a P<0.05 after correction for multiple testing (Table 2, Figure 2A). The highest peak of linkage was found for the marker D9S1824 at 120.1 cM from the p-telomere (Zlr = 3.20; nominal P = 6.94×10−4). At this stage of the study it was not possible to discriminate between these two peaks, thus we decided to pursue our investigations in the 13.1 Mb region surrounding them between D9S279 and D9S112 (Figure 2A).
In the second part of our study we performed a linkage disequilibrium (LD) mapping of the 13.1 Mb region selected after the linkage fine mapping, using a family-based association test. We employed a tag SNP strategy that consisted of genotyping a set of 1,536 markers, extensively representative of genetic variability at the chromosomal region, in a sample of 136 families (Table 1). The sample was composed of 36 families with the highest linkage values in the selected 13.1 Mb region, and 100 novel families never tested before for either linkage or association (Figure 1). Among the 1,536 tag SNPs genotyped, 1,489 (96.9%) were suitable for family-based association testing. The remaining 47 tag SNPs were discarded from the analysis for genotyping failure or lack of polymorphism across the sample. Association analysis was performed using a transmission disequilibrium test (TDT) adapted for families larger than trios, and suitable for testing of association in the area of known linkage [24]. Considering the number of tag SNPs tested, applying crude Bonferroni correction would set the nominal P-value corresponding to a global type I error of 5% at 3.4×10−5. However such threshold is overly conservative, since some of the 1,489 tag SNPs presented a weak level of LD and were therefore not totally independent. To date, there is no consensus on the best method to take into account the non independence between SNPs. A method such as that proposed by Nyholt [25], would set the nominal 5% threshold at P = 5.56×10−5. Alternatively, accepting a corrected global type I error of 7.5% with Bonferroni would set the nominal threshold at P = 5×10−5. Using such criteria, one single intergenic tag SNP, rs4979459, was found to be significantly associated, with SpA (P = 4.9×10−5, Figure 2A, Table S1). Suggestive association was also found for several additional markers in LD with rs4979459 (Figure 2B, Table S1). The evidence of association for these SNPs was also supported by the comparison of observed and expected distributions of association P-values (Figure 3A). The distribution of observed P-values was very suggestively skewed from the null distribution with 23 SNPs having P<0.01, versus 14 expected under the null hypothesis. All SNPs demonstrating significant or suggestive association were located within an 80 kb LD block downstream from the TNFSF15 gene (Figure 2A and 2C). There was no other region in the SPA2 locus presenting significant or even suggestive association with SpA. Notably the TRAF1-C5 locus previously identified as associated with rheumatoid arthritis did not show any association with SpA in our investigation (Figure 2A).
To refine the association signal identified by the LD mapping stage we genotyped the tag SNP rs4979459 and an additional panel of 30 surrounding SNPs in an extended sample of 287 families comprising 668 SpA patients (including the 149 families of the linkage fine mapping, the 100 families added for the LD mapping, and 38 additional families (Table 1, Figure 1B)), as well as in an independent set of 139 cases and 163 controls (Figure 1A and 1B). Association was investigated by the TDT described above for the family sample and by an allelic chi-square test for the case/control set. In keeping with the LD mapping stage, we used the Bonferroni correction for multiple testing. The nominal P-values to achieve global type I errors of 5% and 7.5% significance were 1.61×10−3 and 2.42×10−3 respectively. Of note, the Nyholt correction set the nominal 5% threshold at P = 3.93×10−3.
In the family-based association study, the 5% Bonferroni corrected significance threshold was reached for 3 SNPs: rs10817669, rs10739427, and rs10759734, and the 7.5% significance threshold for 2 additional SNPs: rs10733612, and rs7849556 (Table 3, Table S2). Several other markers including the tag SNP rs4979459 yielded low P-values (Table S2). For all these SNPs the major allele was overtransmitted to affected children.
In the case/control study two markers, in strong LD with each other, reached a Bonferroni corrected 0.005 significance threshold: rs6478105 and rs10982396 (Table 3, Figure 4). Odds ratios (ORs)<1 were observed for both of them, indicating that the major allele was more frequent in cases than in controls. Other markers yielded non-significant low P-values (Table S3).
The evidence of association for these SNPs was also supported by the comparison of the observed and expected distributions of P-values for association (Figure 3B). The distribution of observed P-values was very suggestively skewed from the null distribution with 7 SNPs having P<0.01 in the family-based study and 2 in the case/control study, versus 0 expected under the null hypothesis.
All the markers significantly associated with SpA in all of our association studies belong to the same LD block (Figure 4).
The 5 SNPs located within the TNFSF15 gene, which were previously found to be associated with CD [20],[21], did not reach a level of significant association in both the family-based design and the case/control study (Table S2 and Table S3).
To replicate our results in an independent sample, we genotyped a set of eight SNPs consisting of the most strongly associated markers in the foregoing extension study (rs7849556, rs10817669, rs10759734, rs6478105, rs1982396, and rs10733612), the tag SNP (rs4979459) and one SNP located in the neighboring TNFSF15 gene (rs4246905) in an additional independent set of 232 cases and 149 controls (Figure 1, Table 3). Association was assessed using a chi-square test. After correction for multiple testing, one SNP, rs6478105, reached a suggestive association level (nominal P = 0.029), the Bonferroni corrected for multiple testing thresholds being 0.0063 (5%) and 0.0094 (7.5%).
Of note, OR<1 was observed for all eight SNPs. This trend was similar to that observed in the whole extension step of the study, for both family-based and case/control approaches (see above), suggesting that even if the P-value for significance was not reached here, the direction of association was consistent between both studies for all markers.
The entire case/control sample, comprising the “extension” and the “replication” part (371 cases and 312 controls) was actually an independent set, as compared to the family sample (Figure 1). Thus, it made sense to perform an association analysis on this “pooled” sample. Results of this analysis are displayed in the Table 3. Significant association level was reached for two of the eight tested SNPs: rs6478105 (P = 3×10−5; OR = 0.5) and rs10982396 (P = 2×10−4; OR = 0.53). Furthermore, three other SNPs presented suggestive association with low P-values (rs7849556, rs10759734, and rs10733612). Notably, HLA-B27 conditioned exploratory analyses showed exactly the same trend of allelic distribution between cases and controls, suggesting that the detected association signal was independent of the presence of HLA-B27 (Table S4 and Table S5).
We also performed combined analysis of the genotyping issued from both the family extension sample and either the pooled case/control set or the extension case/control set for the same eight SNPs. The tests were performed using the Cochran-Mantel-Haenszel method [26] (Table 3 and Table S6, respectively). These investigations led us to confirm the strong association with SpA of the whole LD block containing these SNPs (Figure 4).
Of note the non-synonymous SNP, rs4246905, which changes a histidine into arginine, located in the fourth exon of TNFSF15 gene and previously identified as strongly associated with CD [20],[21] reached a suggestive P-value of less than 0.01 in both combined analyses. Nonetheless, this was by far the weakest association of the eight tested markers.
The six SNPs presenting the lowest P-values after the combined analysis were located in the same strong LD block (Figure 4). It is known that in the presence of multiple tightly linked markers a haplotype test may be more powerful to detect association.
Association results for the haplotypes comprising these six SNPs are displayed in the Table 4, together with their estimated frequencies.
The family-based haplotype association analysis identified the most frequent allele H1 as overtransmitted to affected children with a high significance level (Table 4; P = 8.81×10−6), the other three alleles being undertransmitted. The significance threshold corrected for the multiplicity of tested haplotypes and extrapolated with Bonferroni method was set to 0.013.
The case/control haplotype association analysis conversely showed that the rare haplotype H2 was very significantly more frequent in controls than in cases (Table 4; P = 8.75×10−5), with all the other haplotypes being more frequent in cases than in controls. Case/control HLA-B27 conditioned analyses showed that the frequency of H2 haplotype was significantly decreased in patients, independently of the presence of HLA-B27 (Table S4 and Table S5).
A significant omnibus haplotype association was detected in family-based sample (P = 4.5×10−5 at 4 degrees-of-freedom (df)), as well as in the case/control one (P = 7.0×10−4 at 3 df).
We report for the first time an association between several SNPs determining a haplotype near the TNFSF15 gene (9q32) and SpA. This was achieved by a comprehensive study with several linkage and association steps.
The basis of our investigations aimed to narrow down from our previous whole genome linkage screen the susceptibility region for SpA on chromosome 9q31–34 called SPA2 [13]. Since initial linkage in this locus was spread over a long distance of 23.95 cM, our first attempt was to refine it using a fine-grained set of 28 microsatellite markers in an extended set of families. The results revealed two areas of statistically significant linkage, with the highest linkage peak located on the marker D9S1824 at 120.1 cM (115.9 Mb) from the p-telomere, only 1.3 cM apart from D9S1776, which corresponded to the linkage peak in our former screen. The second statistically significant area was located near a suggestive linkage peak reported in AS (D9S1682, P = 6×10−4) [12], supporting the validity of linkage between this region and SpA. Since non parametric linkage analyses usually do not characterize linkage localization with high precision [27], we decided to continue our investigations on the 13.1 Mb region between markers D9S279 and D9S112 comprising both significantly linked areas.
To identify whether one or several loci of the refined SPA2 region were associated with SpA we performed a family-based dense LD mapping of this 13.1 Mb area, using a tag SNP strategy. Our sample set was enriched in families with strong evidence of linkage, in order to minimize the risk of false negative results. After correction for multiple testing, significant association was observed for one single tag SNP, rs4979459 (P = 4×10−5).
Several arguments support the assumption that this is a true positive finding. Firstly, suggestive association was also found for several additional markers in LD with rs4979459, indicating that the identified significant association was unlikely to be explained by systematic genotyping error. Secondly, the use of a family-based design rules out the possible confounding effect of population stratification. Finally, rs4979459 was located in the direct vicinity of our highest linkage peak, between the microsatellite markers D9S279 and D9S1855.
None of the genes having a counterpart in the MHC, i.e. those that are theoretically good candidates for disease susceptibility, were shown to be associated with the disease in this study. Notably, the TRAF1-C5 locus which lies at 122.7 Mb from the p-telomere and was recently described as associated with rheumatoid arthritis [22],[23], did not show any association with SpA in our study.
To refine the association pointed out by our LD map, we performed an extension study with 31 SNPs. Characteristically, the initial strong association observed with rs4979459, was not as strong in the extension stage, suggesting that rs4979459 is not the causal variant. Its effect was probably overestimated in the LD mapping step, since initial positive reports tend to overestimate found effects, while subsequent studies regress to the true value [28]. Moreover, SpA is a complex disease with known genetic heterogeneity, and therefore enriching the LD mapping family set with strongly linked families is also likely to have contributed to this overestimation.
After extension and replication steps of our study, strong association was seen in the family sample as well as in the entire case/control set with SNPs located in the strong 40.3 kb LD block showed in the Figure 4. The markers that were found to be associated by these two approaches were not exactly the same. Several reasons could readily explain such an apparent discrepancy. First of all, the alleles of rs6478105 and rs10982396, which were found to be associated with SpA in the pooled case/control sample, were highly frequent (>0.89). Therefore there were a relatively modest number of informative families for these markers in the family-based study, since at least one parent has to be heterozygous to render the family suitable for association analysis. On the other hand, the power of our case/control study was much lower to demonstrate association with rs10817669 (the SNP significantly associated in the family-based study (OR of 0.82)), than with rs6478105 and rs10982396 (OR of 0.50, and 0.53, respectively).
Nevertheless, even if the significance level was not reached in every study for every marker (probably attributable to a lack of power), we can see that in three totally independent samples trends for all SNPs were exactly the same: the frequent allele of each marker was overtransmitted to affected children and noticeably more present in cases rather than in controls. Moreover the results of combined analysis of family and case/control samples also strongly support the association with all six SNPs, composing the 40.3 kb LD block. Finally, our haplotype investigations also confirmed this trend and showed that haplotypes composed of markers of this block were very significantly associated with the disease, for both family-based and case/control investigations.
The high risk haplotype (H1) identified in our study was too frequent to explain the strong linkage signal detected in SPA2 [29], and was likely only a surrogate for the causal variant(s). As the LD block containing this haplotype is located 28.6 kb from the TNFSF15 gene, one of the best candidate genes in the region, in particular because of its implication in CD [20],[21], it made sense to test this gene directly. However, our LD mapping stage data did not implicate the TNFSF15 “strictly genic” region (introns and exons). Indeed, no tag SNP association was identified in this region and none of the five SNPs described as associated with CD were associated with SpA in our analyses. A full re-sequencing of all exons, 1st and 3rd introns, and additional intronic boundaries of the TNFSF15 gene in several patients and controls did not reveal unmatched variations either (data not shown).
We were also well aware of the limits presented by our extension/replication approach, which was focused only on the “LD mapping” of an association peak. Thus, we tested by a classic candidate-gene approach several genes located within the region surrounding the linkage peaks. We first performed variants screening in a group of independent SpA patients from families presenting a high linkage signal within the studied region, and unrelated controls. The most suggestive polymorphisms were subsequently genotyped. The association with the disease was therefore either assessed with TDT in a sample of independent trios or with a chi-square test in a large case/control sample. In this way, we tested the implication of Tenascine-Cytoactine gene (TNC), coding for an extracellular matrix glycoprotein and presenting a paralogous counterpart in the MHC class III locus. No association was found between polymorphisms in this gene and SpA [30]. We also performed a more systematic candidate-gene approach, testing among others Tumor necrosis factor ligand superfamily member 8 - CD30 ligand (TNFSF8), Zinc finger protein 618 (ZNF618), Kinesin-like protein (KIF12), Alpha-1-acid glycoprotein 1 Precursor (ORM1), Alpha-1-acid glycoprotein 2 Precursor (ORM2) and alpha-1-microglobulin/bikunin precursor (AMBP) genes. The implication in the disease of polymorphisms tested within these genes was excluded by this approach, corroborating the results of our LD mapping, presented in this article (Zinovieva E et al. manuscript in preparation).
Despite the fact that SNPs within the TNFSF15 coding region were excluded by our combined tag SNP and candidate-gene approach, it is still possible that polymorphisms identified by the H1 haplotype play a role in the regulation of this gene. TNFSF15 belongs to the TNF superfamily of genes, otherwise implicated in SpA [31] and is specifically implicated in gut inflammation, a frequent SpA manifestation [32],[33]. Its product, TNFSF15, also called TLA1, is implicated in the modulation of T-helper 17 lymphocytes activation [32], the number of which has been shown to be increased in patients with SpA [34]. Finally a recent study reported that polymorphisms within TNFSF15 to be associated with CD are playing a role in the transcriptional regulation of the gene [35]. This report compounds with our hypothesis, since our findings could be consistent with an indirect role of TNFSF15 in SpA. However, whether the causal variant(s) tagged by the six SNPs haplotype is(are) related to TNFSF15 function and/or regulation or to any other gene in the SPA2 region will require further investigations.
This study was approved by the institutional ethics committee of Cochin Hospital (Paris, France) and of Ambroise Paré Hospital (Boulogne-Billancourt, France), and written informed consent was obtained from each participant.
Caucasian families consisting of one or several cases of SpA and additional parents were recruited throughout France by the “Groupe Français d'Etude Génétique des Spondylarthropathies” (GFEGS). In case/control panels, independent cases were recruited through the Rheumatology clinic of Ambroise Paré Hospital (Boulogne-Billancourt), or through the national self-help patients' organization: “Association Française des Spondylarthritiques”. Independent controls were obtained from the “Centre d'Etude du Polymorphisme Humain”, or were recruited as healthy spouses of cases with either no children or no affected children.
The phenotypic description of patients from familial and case/control samples is shown in Table 5. The diagnosis of SpA was made according to the classification criteria of Amor et al [36] and/or the European Spondylarthropathy Study Group (ESSG) [37]. Within the group of SpA, AS was diagnosed according to the modified New York criteria [38]. Regarding extra-articular manifestations, the diagnosis of psoriasis required the presence of typical lesions and/or a clinical diagnosis established by a dermatologist. The diagnosis of anterior uveitis required examination by an ophthalmologist. Inflammatory bowel disease diagnosis (including CD and ulcerative colitis) was based on endoscopic and histological examination of the gut. ReA was diagnosed according to the criteria published by Willkens [39]. Finally, uSpA was diagnosed when SpA criteria were fulfilled, without any of the foregoing diagnosis.
Genomic DNA was extracted from peripheral blood using standard methods. HLA-B typing was routinely performed by a polymerase chain reaction (PCR) - based sequence-specific method [40]. For individuals already typed as positive for HLA-B27, retyping was not routinely performed.
For familial studies, Mendelian inheritance inconsistencies were identified with the PEDCHECK program [48]. PLINK program [49] was used to assess the deviation from Hardy-Weinberg equilibrium in unrelated subjects. Family pairwise distributions among first and second degree relative pairs were accounted for with the PEDSTATS program [50].
For the fine mapping linkage study, allele frequencies were estimated using MENDEL software [51]. Evidence for linkage was assessed using Zlr statistic [52] based on Spairs with the exponential model and multipoint identity-by-descent computation using ALLEGRO program [53]. P-values were computed on the basis of large-sample theory; the distribution of Zlr statistic approximates a standard normal random variable under the null hypothesis [52],[53]. Whole fine map significance was extrapolated using Bonferroni correction for 28 tests.
Family-based allelic single-locus association analyses (1,536 tag SNP in the 136 families of the LD mapping and 31 SNPs in the 287 families of the extension study (Figure 1)) were carried out using FBAT [54]. FBAT is a flexible program appropriate for analyses of family data larger than trios, allowing association tests that are robust to population cofounds in the case where parental data are missing and/or other offspring are included in the analysis [55]. We specified the option to calculate the variance empirically (“-e” option) in order to provide valid tests of association in the presence of linkage [24]. The global significance threshold for each set of SNPs was assessed using Bonferroni correction.
It has been shown that in some situations (such as when r2 factor between the risk variant and a particular multi-SNP haplotype is very strong) haplotypes may provide more information for association than corresponding single-locus tests [56]. HBAT is an elaboration of FBAT that allows family-based association tests of haplotypes, even when the phasing is ambiguous. Family-based haplotype-specific association in the presence of known linkage was assessed with the “hbat -e” option of FBAT program, for a set of pre-selected tightly linked markers. This option allows one to perform two types of haplotype tests. In the first type, each haplotype allele is tested for association against all the others using a one degree-of-freedom (df) test. Significance of these tests must be extrapolated with a multiple testing correction; here we used the Bonferroni method. The second type of haplotype tests is a global multiallelic test with several df. In this case, there is no need to correct for multiple testing.
Case/control association studies were carried out using the standard chi-square test comparing allelic frequencies between cases and controls and giving asymptotic P-values, implemented in PLINK package [49]. Allele frequencies, ORs, and their 95% confidence intervals were also estimated using this software. When needed, the adjustment for multiple testing was performed using the Bonferroni correction.
The quantile-quantile (Q-Q) plots (Figure 3) were constructed by ranking the sets of association P-values from the largest to the smallest and plotting them against their expected values. Under the null hypothesis the expected P-value for the ith SNP is i/n, where n is the total number of tested markers.
Haplotype-specific association was assessed in case/control samples for a set of pre-selected tightly linked SNPs (the same as for the family investigation described above) with the “hap-assoc” option of PLINK. This procedure takes into account the uncertainty of haplotype phase and performs both one df chi-square haplotype-specific tests, which significance must be extrapolated with a multiple testing correction, here Bonferroni correction, and an omnibus association statistic considering all the haplotypes.
The explorative combined analysis of the whole association data from the extension and replication studies (1 family sample and 2 case/control samples; Figure 1) was performed with the “dfam” option of PLINK. This particular test implements a sib-TDT [57] for nuclear families, to include sibships without parents as well as unrelated individuals and assesses the association via a clustered-analysis using the Cochran-Mantel-Haenszel test [26]. It does not take into account the presence of linkage in the region.
LD plots were constructed using HAPLOVIEW program [58].
The URLs for data presented herein are as follows:
* Association Française de Spondylarthritiques (AFS): http://www.spondylarthrite.org/
* Centre d'Etude du Polymorphisme Humain (CEPH): http://www.cephb.fr/
* University of California Santa Cruz (UCSC) public database: http://genome.ucsc.edu/
* French National Genotyping Center (CNG): http://www.cng.fr/
* Detailed protocol for the Illumina BeadChips GoldenGate assay: http://icom.illumina.com/General/pdf/LinkageIV/GOLDENGATE_ASSAY_FINAL.pdf
* Ensembl database: http://www.ensembl.org/index.html
* NCBI dbSNP database: http://www.ncbi.nlm.nih.gov/projects/SNP/
* HapMap database: http://www.hapmap.org/
* PEDCHECK program: http://watson.hgen.pitt.edu/register/docs/pedcheck.html
* MENDEL program: http://www.genetics.ucla.edu/software/mendel
* ALLEGRO program: http://www.decode.com/software/
* FBAT program: http://www.biostat.harvard.edu/~fbat/fbat.htm
* PEDSTATS program: http://www.sph.umich.edu/csg/abecasis/PedStats/index.html
* PLINK program: http://pngu.mgh.harvard.edu/~purcell/plink/
* HAPLOVIEW program: http://www.broad.mit.edu/mpg/haploview/
|
10.1371/journal.ppat.1005840 | Progesterone-Based Therapy Protects Against Influenza by Promoting Lung Repair and Recovery in Females | Over 100 million women use progesterone therapies worldwide. Despite having immunomodulatory and repair properties, their effects on the outcome of viral diseases outside of the reproductive tract have not been evaluated. Administration of exogenous progesterone (at concentrations that mimic the luteal phase) to progesterone-depleted adult female mice conferred protection from both lethal and sublethal influenza A virus (IAV) infection. Progesterone treatment altered the inflammatory environment of the lungs, but had no effects on viral load. Progesterone treatment promoted faster recovery by increasing TGF-β, IL-6, IL-22, numbers of regulatory Th17 cells expressing CD39, and cellular proliferation, reducing protein leakage into the airway, improving pulmonary function, and upregulating the epidermal growth factor amphiregulin (AREG) in the lungs. Administration of rAREG to progesterone-depleted females promoted pulmonary repair and improved the outcome of IAV infection. Progesterone-treatment of AREG-deficient females could not restore protection, indicating that progesterone-mediated induction of AREG caused repair in the lungs and accelerated recovery from IAV infection. Repair and production of AREG by damaged respiratory epithelial cell cultures in vitro was increased by progesterone. Our results illustrate that progesterone is a critical host factor mediating production of AREG by epithelial cells and pulmonary tissue repair following infection, which has important implications for women’s health.
| Worldwide, the use of hormonal contraceptives is on the rise as a primary intervention for improving women’s health outcomes through reduced maternal mortality and increased childhood survival. There are many hormone contraceptive formulations, all of which contain some form of progesterone. Although the effects of hormone contraceptives and progesterone, specifically, have been evaluated in the context of infections of the reproductive tract, the effects of progesterone at other mucosal sites, including the respiratory tract have not been systematically evaluated. We have made the novel observation that administration of progesterone to female mice depleted of progesterone confers protection against both lethal and sublethal influenza A virus infection. In particular, progesterone reduces pulmonary inflammation, improves lung function, repairs the damaged lung epithelium, and promotes faster recovery following influenza A virus infection. Progesterone causes protection against severe outcome from influenza by inducing production of the epidermal growth factor, amphiregulin, by respiratory epithelial cells. This study provides insight into a novel mechanistic role of progesterone in the lungs and illustrates that sex hormone exposure, including through the use of hormonal contraceptives, has significant health effects beyond the reproductive tract.
| Hormonal contraceptives are listed as an essential medication by the World Health Organization (WHO)[1] because of the profound benefits these compounds can have on women’s health outcomes, including decreased rates of maternal mortality and improved perinatal outcomes and child survival, by widening the intervals between pregnancies [2]. Hormonal contraceptive formulations vary, but all contain some form of progesterone (P4) either alone or in combination with estrogen. There are currently over 100 million young adult women on P4-based contraceptives worldwide [3], with the WHO projecting that over 800 million women will be using contraceptives, including P4-based contraceptives, by 2030 [2]. Despite the staggering numbers of women taking these compounds, very few studies evaluate the impact of contraceptives on responses to infection or vaccination, especially in non-sexually transmitted diseases.
Natural P4, produced by the ovaries during reproductive cycles, or synthetic P4 analogues found in contraceptives, signal primarily through progesterone receptors present on many cells in the body, including immune cells (e.g., NK cells, macrophages, dendritic cells (DCs), and T cells) as well as non-immune cells, such as epithelial cells, endothelial cells, and neuronal cells [4, 5]. Human, animal, and in vitro studies show that P4 can alter the immune environment and promote homeostasis by decreasing inflammation and inducing anti-inflammatory responses. For example, in the presence of P4, macrophages and DCs have a lower state of activation, produce higher levels of anti-inflammatory cytokines, such as IL-10, and produce lower amounts of proinflammatory cytokines, such as IL-1β and TNF-α, as compared with placebo treated cells [6, 7]. When either mice or cord blood cells from humans are treated with P4, the percentages of Foxp3+ regulatory T cells (Tregs) increase [8, 9]. Although the immunomodulatory effects of P4-based therapies in the form of contraception have been studied in the context of sexually transmitted infections, including HIV and herpes simplex virus [10–12], the impact of P4 on the outcome of viral infectious diseases outside of the reproductive tract has not been considered in either humans or animal models.
Influenza A viruses (IAVs) primarily infect respiratory epithelial cells and induce the production of proinflammatory cytokines and chemokines that recruit immune cells, causing a local proinflammatory environment [13]. Infiltration and activation of CD4+ and CD8+ T cells, while necessary for the clearance of IAVs [13–15], can trigger inflammation and lead to tissue damage and severe outcomes from IAV infection [16]. Protection requires a balance between inflammatory responses generated to control virus replication and eliminate virus-infected cells with responses that mediate the repair of damaged areas of the lung. Repair involves a complex interplay among many cell types, cytokines, chemokines, growth factors, and extracellular matrix proteins that remodel tissue after acute injury, such as IAV infection [17]. Amphiregulin (AREG) is an epidermal growth factor that has emerged as a significant mediator of tissue repair at mucosal sites, including the lungs [18, 19], gastrointestinal tract [20, 21], and reproductive tract [22, 23]. Many immune cells produce AREG, but epithelial cells are the principle producer of AREG following inflammation or tissue injury [24]. If P4 can downregulate inflammatory immune responses and promote regulatory or tissue repair responses, then this hormone, at concentrations that reflect the luteal phase of the reproductive cycle, may improve the outcome of IAV infection.
Epidemiological and experimental evidence suggest that young adult females suffer a worse outcome than males following IAV infection, which in mice is associated with infection-induced suppression of reproductive hormones and excessive inflammatory immune responses in females [25–27]. In addition to influenza, young adult females suffer a worse outcome than males from several autoimmune diseases, including multiple sclerosis [28]. Paradoxically, a growing body of literature reveals that exogenous treatment of females (both humans and mice) with either estrogens or P4 limits inflammation and protects against infectious and autoimmune diseases by decreasing inflammation and promoting repair [25, 29–31]. In this series of studies, we show that treatment with sustained physiological doses of P4 protects females against IAV by reducing inflammation and improving pulmonary function, primarily through upregulation of AREG in epithelial cells. The observation that P4 regulates the cellular and molecular mediators of tissue repair at a mucosal site outside of the reproductive tract to restore tissue homeostasis after infection or injury has broad implications for women’s health.
To analyze the effects of P4 on morbidity and mortality in female mice, we depleted P4 by removing the ovaries and replaced P4 with subcutaneous pellets that delivered a continuous dose of physiological levels of P4 over the course of 21 days. Mice were subsequently mock-infected or infected with a dose of IAV (PR8) that is uniformly lethal for P4-depleted mice. Circulating levels of P4 and uterine horn mass, a biomarker of circulating P4 levels [32], were assessed over the course of infection to confirm the continuous effects of hormone replacement. Exogenous replacement of P4 significantly increased and sustained plasma P4 concentrations within the normal physiological range [33] throughout the duration of the study. Both mock- and IAV-infected females treated with exogenous P4 had higher circulating concentrations of P4, greater uterine horn mass, and higher expression of progesterone receptors (Prs) in the lungs than either mock or IAV-infected females treated with placebo throughout the 21 days (Fig 1A and 1B; P<0.05).
During the course of IAV infection, treatment of female mice with P4 mitigated the effects of infection on morbidity and mortality (Fig 1C and 1D; P<0.05), with the average day of death being later for females treated with P4 (11.14±1.0 days post-infection [dpi]) as compared to placebo-treated females (9.5±0.6 dpi) (P<0.05). Progesterone treatment did not alter virus titers over the course of the first week of infection as compared to placebo treatment (Fig 1E), suggesting that P4 did not render females more resistant to IAV infection. To test whether P4 improved survival during IAV infection by making females more tolerant to the negative consequences of infection on host health, we analyzed the interaction between virus titers and body temperature during peak disease (7dpi) [34]. Females treated with P4 suffered less hypothermia relative to their pulmonary viral load than the placebo-treated females, suggesting that P4 made females more tolerant of IAV infection (Fig 1F; P<0.05). To test the hypothesis that P4 may increase tolerance by reducing inflammation and damage in the lung, pulmonary tissue was evaluated for vasculitis, bronchiolitis, alveolitis, and edema. In mock-infected animals, P4 alone did not result in changes in any of the parameters examined (Fig 1G [panels 1 and 2]). Seven days post-infection with IAV, treatment with P4 decreased vasculitis (Fig 1G [panels 3 and 4] and 1H) and edema (Fig 1G [panels 5 and 6] and 1H) as compared to the placebo-treated mice (P<0.05). Progesterone improved the outcome of lethal IAV infection by limiting lung inflammation and damage, but not virus replication.
Virus-specific CD8+ T cells are necessary for clearance of IAV but can also contribute to immunopathology [35, 36]. Although the total numbers of CD8+ T cells increased in all females following IAV infection, the total number of CD8+ T cells, the number of IAV-specific CD8+ T cells, and the production of IFN-γ and TNF-α by virus-specific CD8+ T cells in the lungs did not differ between P4- and placebo-treated females (Table 1). These data indicate that P4 did not affect the cell-mediated antiviral immune response during acute IAV infection.
IAV infection is characterized by the induction of a cytokine storm and excessive immunopathology, which leads to tissue damage [37]. Damage to the lung endothelium and/or epithelium results in vascular leakage into the air spaces, and can be quantified by measuring protein concentration in bronchoalveolar lavage (BAL) fluid. Consistent with the histopathological findings of increased vasculitis and edema (Fig 1H) following lethal IAV infection, treatment of females with P4 decreased the total amount of protein contained in the BAL as compared to placebo-treated mice (Fig 2A; P<0.05). Among infected females, treatment with P4 also increased cellular proliferation (as measured by Ki67 expression) in the lungs as compared to placebo treatment during peak disease (7dpi) (Fig 2B and 2C; P<0.05). Analysis of the expression of Ki67 in the different areas of the lungs revealed greater proliferation in several regions of the lungs, but was most pronounced in the epithelial cells lining the airways during IAV infection in P4-treated mice (Fig 2C). The epidermal growth factor, AREG, promotes proliferation of epithelial cells and protects mice from excessive pathology during IAV infection [18, 19]. Analysis of AREG expression during peak disease (7 dpi) revealed increased mRNA expression, as well as AREG protein in the bronchioles, but not the alveoli, in the lungs of P4-treated mice as compared to placebo-treated mice infected with IAV (Fig 2D–2F, P<0.05).
Progesterone treatment altered inflammation during IAV infection (Fig 1G and 1H) and induced a repair environment through cellular proliferation and restoration of barrier integrity (Fig 2A–2C). To further characterize the effect of P4 on inflammatory responses to IAV, a panel of 13 cytokines and chemokines was analyzed in the supernatant of whole lung homogenates. As expected, following infection with IAV, pulmonary concentrations of IL-1β, TNF-α, IFN-γ, and IL-12p70 were significantly increased during the first week of infection in all females, regardless of P4 treatment (S1 Table; P<0.05). P4 treatment decreased pulmonary production of the alarmins IL-13 and IL-33 as compared with placebo treatment during IAV infection (S1 Table; P<0.05). The only two cytokines that were significantly increased in P4-treated females compared with placebo-treated females during IAV infection were IL-6 and TGF-β (Fig 3A and 3B; P<0.05). P4 treatment of IAV-infected mice had no effect on the other canonical regulatory protein, IL-10, as compared to placebo treatment (S1 Table).
Production of TGF-β and IL-6 increases differentiation of Th17 cells. Th17 cells promote repair of the gut epithelium [38] and may be similarly involved in orchestrating repair of the pulmonary epithelium. To test this hypothesis, populations of CD4+ T cells from mock- and IAV-infected mice were enumerated during peak disease (7 dpi). There was no effect of P4 treatment on total numbers of CD4+ T cells, Th1, Th2, or Treg cells in the lungs at 7 dpi (Table 1). In contrast, P4 treatment increased the total number of Th17 cells in the lungs during IAV infection as compared with placebo treatment (Fig 3C; P<0.05). The cytokine IL-23 is necessary for maintenance of Th17 cells and the expression of Il23 mRNA in the lungs was increased in P4- compared with placebo-treated females (Fig 3D; P<0.05). Th17 cells exert their tissue reparative effects by increasing the production of IL-22 [39]. The expression of Il22 mRNA in the lungs was greater in P4- than placebo-treated females during IAV infection (Fig 3E; P<0.05). Finally, one surface marker on Th17 cells that is associated with reducing inflammation (i.e., regulatory or suppressive Th17 cells) is the ectonucleotidase CD39 (ref. [40, 41]). The percentage of Th17 cells that expressed CD39 was significantly increased in P4-treated as compared to placebo-treated females during IAV infection (Fig 3F; P<0.05). These data indicate that P4 alters the inflammatory milieu of the lungs by promoting a repair environment in IAV-infected female mice, with increased numbers of regulatory Th17 cells, elevated expression of Il22, and upregulated expression of Areg during lethal IAV infection.
To further evaluate the role of P4 in lung repair and recovery from IAV infection, P4- and placebo-treated female mice were infected with a less pathogenic IAV strain, ma2009, at a dose (0.4mLD50) that allowed for monitoring of the mice over a longer duration of time. Similar to lethal IAV infection, P4-treated females infected with sublethal IAV showed less hypothermia (Fig 4A; P<0.05) and reduced clinical disease (Fig 4B; P<0.05) as compared to placebo-treated females. Analysis of pulmonary virus titers confirmed that P4 did not alter virus titers or clearance of infectious virus over the course of IAV infection (Fig 4C). To determine if P4 reduced cell death due to IAV infection, LDH levels in the BAL fluid were quantified. Cellular damage during IAV infection was not altered by treatment with P4 as compared with placebo (Fig 4D). Lung sections were evaluated for markers of inflammation and damage during the recovery (14 dpi) and post-recovery (25 dpi) phases of IAV infection. At 14 dpi, but not at 25 dpi, treatment of IAV-infected female mice with P4 decreased the percentage of lesioned areas, alveolitis, edema, and cumulative inflammation as compared to placebo-treated mice (Fig 4E–4H, P<0.05). Treatment with P4 significantly increased Ki67 expression in pulmonary cells during the recovery phase (14 dpi) of IAV infection as compared with placebo treatment (Fig 4I; P<0.05). Based on the observation that P4 treatment promoted lung repair in IAV-infected female mice, we evaluated the impact of P4 on overall lung physiology during (14 dpi) and after (25 dpi) recovery from sublethal IAV infection by assessing markers of pulmonary function. Lung diffusing capacity (DFCO), lung tissue compliance (Crs), and resistance (Rrs) returned to baseline faster in P4- than placebo-treated mice infected with IAV (Fig 4J–4L, P<0.05). Treatment of female mice with P4 reduces inflammation and promotes faster recovery from sublethal IAV infection.
Progesterone increased pulmonary AREG expression during lethal IAV infection (Fig 2D–2F) and increased AREG expression is associated with an improved outcome from lethal IAV infection [18, 19]. In our sublethal IAV model, we were able to measure pulmonary expression and production of AREG over a longer duration of time to establish the effects of P4 on the kinetics of AREG production in females. P4-treatment induced a 30–70 fold greater induction of Areg mRNA and higher concentrations of AREG protein in the lungs as compared with placebo treatment over the course of IAV infection (Fig 5A and 5B; P<0.05). Peak production of AREG occurred at 9 dpi (Fig 5B), which corresponded with peak disease (Fig 4A and 4B) during sublethal IAV infection. To test the hypothesis that reduced AREG production in P4-depleted females caused a more severe outcome from IAV, we treated P4-depleted female mice with recombinant AREG (rAREG) during the course of IAV infection. Treatment of P4-depleted mice with rAREG resulted in AREG levels that were comparable to those of P4-treated mice at 14 dpi (Fig 5C; P<0.05). Treatment of P4-depleted females with rAREG significantly improved the recovery from IAV infection (Fig 5D and 5E; P<0.05), with reduced inflammation (Fig 5F and 5G; P<0.05) and improved pulmonary function, including lung diffusing capacity (DFCO), lung compliance (Crs), and resistance (Rrs), to levels similar to that of P4-treated females (Fig 5H–5J; P<0.05). These data suggest that the protective effects of P4 on IAV disease may be mediated by an upregulation of AREG.
The contribution of AREG to P4-mediated protection from IAV infection was further determined by using mice that lacked the expression of a functional Areg gene [42]. Deletion of the Areg gene in female mice (Areg-/-) reversed the protective effects of P4 on the outcome of IAV infection (Fig 6A and 6B; P<0.05). This was accompanied by increased inflammation in P4-treated Areg-/- as compared with WT female mice (Fig 6C and 6D; P<0.05). Improvement of pulmonary function in the presence of P4, as measured by lung diffusing capacity (DFCO), compliance (Crs), and resistance (Rrs), was also reversed in IAV-infected Areg-/- mice as compared with WT mice treated with P4 (Fig 6E–6G; P<0.05). Taken together, these data indicate that P4 treatment of IAV-infected female mice promotes a pulmonary repair environment and restoration of lung function through the induction of AREG.
Treatment with P4 induces higher expression of AREG in the lungs of sublethal IAV-infected females, particularly in the epithelial cells lining the larger airways, as compared with placebo-treatment (Fig 7A and 7B; P<0.05). To assess the contribution of P4 treatment to the repair of damaged respiratory epithelia, we used an in vitro model system in which primary, differentiated mouse tracheal epithelial cell (mTECs) cultures were mechanically injured. The mTECs express the progesterone receptor (Pr), which was upregulated in the presence of P4 (Fig 7C; P<0.05). Repair of the epithelial cell layer was measured over time to identify the return of the transepithelial resistance (TER) to baseline. Following injury, cultures of mTECs treated with P4 returned to baseline TER faster than vehicle-treated cultures (Fig 7D; P<0.05). During injury, mTEC cultures treated with P4 produced more AREG mRNA and protein than vehicle-treated mTECs cultures (Fig 7E and 7F; P<0.05). These data illustrate that P4 improves pulmonary repair and function by increasing AREG production and wound repair in epithelial cells.
Hosts have evolved several mechanisms for overcoming viral infections, such as the induction of antiviral defenses that increase resistance to infection, or the activation of regulatory and repair responses that increase tolerance to the negative consequences of infection. In the present study, P4 significantly protected females during IAV infection by altering inflammation, improving pulmonary function, and promoting a pulmonary repair environment, which resulted in an earlier recovery. The protective effects of P4 were primarily mediated by the induction of AREG during both lethal and sublethal infections. Progesterone did not increase resistance to infection in females as demonstrated by the lack of an effect of P4 treatment on virus titers, clearance of infectious virus, numbers of Th1 cells, and CD8+ T cell activity in lungs. Instead, P4 reduced the detrimental consequences of IAV infection in females by increasing their tolerance to infection. Several host immunological factors, including TGF-β, Tregs, and regulatory populations of CD39+ Th17 cells, are associated with maintaining the balance between protective and pathological immune responses during IAV infection. Although P4 treatment had no effect on the numbers of Tregs in the lungs during IAV infection, concentrations of TGF-β and IL-6, the expression of Il23 and Il22, the number of Th17 cells, as well as the proportion of Th17 cells expressing CD39, were increased. Regulatory Th17 cells express the ectonucleotidase CD39 and are associated with repair following inflammation and infection [40, 41]. Th17 cells also promote epithelial cell proliferation and repair in the gut, primarily through the induction of IL-22 [38]. Consequently, treatment of females with P4 increased IL-22, a cytokine that has been shown to mediate regeneration of lung epithelial cells following IAV infection [43]. Whether the P4-induced increase in regulatory Th17 cells and IL-22 promotes cellular proliferation and repair of the lung epithelium during IAV infection by increasing AREG production requires consideration. Because P4 directly induced AREG production in respiratory epithelial cells in vitro, P4-induced AREG production may occur independent of the reparative effects of regulatory Th17 cells in the lungs during IAV infection.
Progesterone induces repair of epithelial cells in the endometrium and myelin fibers in the central nervous system [44, 45]. This repair of myelin fibers by P4 [46] is one factor mediating how this reproductive hormone mitigates the progression of multiple sclerosis [29]. Our data show that P4 promotes proliferation of pulmonary cells, including epithelial cells, and pulmonary tissue repair. The reparative effects of P4 in the reproductive tract are caused by the induction of AREG, which promotes epithelial remodeling in mammary and uterine tissues [22, 23]. In the respiratory tract, AREG is involved in pulmonary tissue remodeling and repair during lung injury, asthma, and infection [18, 19, 21, 47, 48]. Although Areg-gene deficient mice show few abnormalities under homeostatic conditions [42], their ability to resolve inflammation or infection is severely impaired [20, 21]. During IAV infection, administration of rAREG protects mice from severe IAV-mediated morbidity by decreasing hypothermia, improving pulmonary function, and decreasing protein leakage into the airways [18, 19]. The data presented are the first report of P4 induction of AREG outside of the reproductive tract and in the context of infection. The effect of other reproductive hormones on AREG expression, including differential expression between males and females, warrants further study.
AREG is produced primarily by epithelial cells [49], but type 2 innate lymphoid cells (ILC2) and Tregs have also been shown to produce AREG during IAV infection and contribute to the repair during resolution of infection [18, 19, 49, 50]. Because each of these cell type express progesterone receptors [5, 51], each is a potential producer of AREG in response to P4 treatment. Our in vivo and in vitro data suggest that respiratory epithelial cells are a predominant source of P4-induced AREG. Following IAV infection, AREG expression was predominantly localized to the bronchiolar epithelial cells, and P4 treatment of isolated mTECs increased AREG production following mechanical damage. Furthermore, P4-treatment did not activate markers of ILC2s, including IL-13 and IL-33 production, or increase numbers of Tregs in the lungs during infection, suggesting that the induction of AREG in response to P4 may not be occurring in these immune cell populations.
Recovery following IAV infection is generally defined as a return of body temperature or body mass back to homeostatic levels [52]. In this study, however, we showed that pulmonary pathology and impaired pulmonary function persisted after measures of overall health, including hypothermia and clinical disease, returned to baseline. Furthermore, the impact of IAV infection was observed long after infectious virus had been cleared from the lungs. Recovery following IAV infection extended beyond 21 dpi and should be defined not only by reduced morbidity, but also by restored pulmonary function, both of which were expedited by P4 treatment in females.
Progesterone concentrations fluctuate naturally during the female life span, with moderate concentrations during the menstrual cycle, high concentrations during pregnancy, and low concentrations following menopause. Progesterone is also used exogenously by over 100 million women worldwide in P4-based hormonal contraceptives, by post-menopausal women taking hormonal replacement therapy, and by both men and women in the treatment of cancer, osteoporosis, and brain injury [3, 53]. Prior to this study, the health consequences of P4-based therapies in acute respiratory infection had not been characterized. We have demonstrated that AREG, which is a significant factor that induces tissue repair and recovery from infectious diseases, is regulated by P4 during both lethal and sublethal IAV infection. The data presented provide critical mechanistic information about how P4 and possibly synthetic P4 analogues affect women’s health outside of the reproductive tract. Contraceptives that contain P4 are listed as an essential medication by the WHO, being a safe and effective method for improving health outcomes in women, including those living with HIV [1]. During outbreaks of infectious diseases that harm pregnant women and their fetuses (e.g., the current Zika outbreak), the WHO recommends increased use of hormonal contraceptives, which according to our data could have additional beneficial consequences on the outcome of other infectious diseases.
All experiments were performed in compliance with the standards outlined in the National Research Council’s Guide to the Care and Use of Laboratory Animals. The animal protocol (M015H236) was reviewed and approved by the Johns Hopkins University Animal Care and Use Committee. All efforts were made to minimize animal suffering.
Adult (7–8 weeks old) female C57BL/6 mice were purchased from NCI Frederick. Areg+/- (C57BL/6 129 Sv) mice were kindly provided by Dr. Marco Conti (University of California San Francisco) and bred to obtain Areg-/- and Areg+/+ female littermates. Mice were housed 5 per microisolator cages under standard BSL-2 housing condition with food and water ad libitum.
At 8–12 weeks of age, mice were anesthetized with an intramuscular injection of ketamine (80 mg/kg) and xylazine (8 mg/kg) cocktail and ovaries were removed bilaterally as previously described [25]. All animals were given two weeks to recover prior to infection. Recombinant amphiregulin (10μg; R&D) was administered intraperitoneally every other day using saline as the vehicle.
Ovariectomized (ovx) mice were assigned to receive subcutaneous implants of placebo (-P4) or 15 mg progesterone (+P4) 21-day release pellets (Innovative Research of America) prior to IAV inoculation. Circulating concentrations of P4 were assessed from plasma using ether extraction and radiolabelled immunoassay, with P4 antibody (MP Biomedicals) and tracer 3H-P4 (American Radiolabeled). Uterine horns were removed at several time-points upon euthanasia of mice and wet weight was quantified as a bioassay for P4.
Mouse-adapted influenza A viruses, A/Puerto Rico/8/34 (PR8; H1N1) provided by Dr. Maryna Eichelberger at the Food and Drug Administration (FDA) and A/California/04/09 (ma2009; H1N1) generated by Dr. Andrew Pekosz from a published sequence [54], were used in these studies. Mice were anesthetized and inoculated intranasally with 30 μl of DMEM (mock) or H1N1 virus (1.78 50% mouse lethal dose (MLD50) for PR8 and 0.4 MLD50 for ma2009). Clinical disease scores for IAV-infected mice were based on four parameters, with one point given for each of the following: dyspnea, piloerection, hunched posture and absence of an escape response. For virus quantification, log10 dilutions of lung homogenates (starting at 10−1) were plated onto a monolayer of MDCK cells in replicates of 6 for 4–6 days. Cells were stained with naphthol blue black (Sigma Aldrich) and scored for cytopathic effects. The 50% tissue culture infectious dose (TCID50) was calculated according to the Reed-Muench method.
Snap-frozen lung tissue was homogenized in DMEM supplemented with 1% penicillin/streptomycin and 1% L-glutamine (Invitrogen) and centrifuged to remove cellular debris. Supernatants were harvested to measure IL-1β, TGF-β, IL-4, IL-5, IL-13, IL-17, IL-33, and AREG by ELISA (R&D Systems and BD Biosciences) and CCL-2, IL-12(p70), TNF-α, IFN-γ, IL-6 and IL-10 with the mouse inflammation cytometric bead array (BD Biosciences) according to the manufacturer’s protocols.
Snap-frozen lung tissue or mTECs were homogenized in TRIzol and RNA was purified by chloroform extraction. RNA concentration and purity was measured using a NanoDrop (ThermoFisher Scientific). The RNA concentration in each sample was standardized to 1 μg using RNAse-free water. Reverse transcription was carried out using the iScript cDNA synthesis kit (Biorad) according to the manufacturer’s protocol. Pre-designed Il23 (Mm.PT.58.10594618.g), Il22 (NM_016971.2), Areg (Mm.PT58.31037760), Gapdh (Mm.PT.39a.1) and Pr (Mm.PT.58.10254276) PrimeTime Primers were purchased from Integrated DNA Technologies. Semi-quantitative RT-PCR was performed in 96-well optical reaction plates using the SsoFast EvaGreen Supermix (Biorad) on the StepOnePlus RT-PCR system (Applied Biosystems). Gene expression was normalized to Gapdh and mock-infected samples or wells with no injury using the ΔΔCt method.
Lungs were excised and single-cell suspensions were generated following red blood cell lysis. Total viable cells were determined using a hemocytometer and trypan blue (Invitrogen) exclusion and resuspended at 1x106 cells/ml in RPMI 1640 (Cellgro) supplemented with 10% FBS (Fisher Scientific) and 1% penicillin/streptomycin. For IAV-specific T cells enumeration, cells were cultured for 5h with IAV peptide antigen (CD8:NP366-374, or CD4: HA211-255, NP311-325, respectively) (ProImmune) in media containing Brefeldin A (GolgiPlug, BD) The viability of cells was determined by fixable Live/Dead violet viability dye (Invitrogen) and Fc receptors were blocked using anti-CD16/32A. The T cell populations were stained with the following antibodies: PerCP-Cy5.5 conjugated anti-CD4 (RM4-5)A, PerCP-Cy5.5 conjugated anti-CD8 (53–6.7)A, FITC conjugated anti-CD25 (7D4)A, PE conjugated DbNP366-374 tetramer (NIH Tetramer Core Facility), FITC conjugated anti-CD4 (RM4-5)B, APC conjugated anti-CD3 (17A2B, and PerCP-eFluor 710 conjugated anti-CD39 (24DMS1)B. Intracellular staining with PE conjugated anti-TNF-α(MP6-XT22)A, FITC conjugated anti-IFN-γ (XMG1.2)A, PE conjugated anti-IL-4 (11B11)A, and PE conjugated anti-IL-17 (TC11-1810)A, was performed following permeabilization and fixation with Cytofix/Cytoperm and Perm/Wash bufferA. Intracellular staining with PE-conjugated Foxp3 (MF23)A was performed following fixation and permeabilization with a Foxp3 staining buffer setA. Data were acquired using a FACS Calibur (Cellquest Software) and analyzed using FlowJo (Tree Star, Inc.). Total cell counts were determined by multiplying each live cell population percentage by the total live cell counts acquired prior to staining by trypan blue exclusion counts on a hemocytometer. All reagents were purchased from BD BiosciencesA or eBioscienceB unless stated otherwise.
Lungs were inflated, fixed in Z-fix (Anatech), embedded in paraffin, cut into 5μm sections, and mounted on glass slides. Slides were stained with hematoxylin and eosin (H&E) and used to evaluate lung inflammation. Histopathological scoring was performed by a single blinded veterinary pathologist on a scale from 0–3 (0, no inflammation; 1, mild inflammation; 2, moderate inflammation; and 3, severe inflammation) for the following parameters: bronchiolitis, alveolitis, vasculitis, perivasculitis, necrosis, consolidation, and edema [55, 56]. The sum of these parameters represents the cumulative inflammation score. The percentage of lesioned areas within each tissue section was also evaluated. Histopathological slides were deparafinized with xylene and rehydrated in graded ethanol. Heat-induced antigen retrieval with citrate buffer was performed and slides were blocked with 10% normal serum prior to overnight primary antibody incubation. For Ki67, rabbit anti-Ki67 (1/200; Abcam) was used, detected with the EXPOSE rabbit specific HRP/DAB detection kit (Abcam), counterstained with Hematoxylin and slides were mounted using Permount (Fisher). For immunofluorescence, anti-AREG (1/100; R&D) and anti-β-tubulin IV (1/100; BioGenex) were used and detected with appropriate secondary antibodies (1/400) conjugated to AF-555 (Thermo) and AF488 (Molecular probes). Slides were then treated against autofluorescence using 0.3% Sudan Black B (Sigma) in 70% ethanol and mounted using anti-fade medium containing DAPI (ProLong Gold from Cell Signaling Techonology). Images were taken using a Nikon Eclipse E800 (for H&E and Ki67 stains) or a Zeiss AxioImager M2 (for immunofluorescence) and analyzed using ImageJ (NIH).
Mice were euthanized by cervical dislocation and the lungs were lavaged twice with 0.5ml of a 0.9% saline solution. Bronchoalveolar lavage (BAL) fluid was centrifuged at 500g for 10 minutes to remove cells and debris and the supernatant was collected to quantify total protein leakage into the airway using a BCA assay (Pierce). Cell lysis and damage was analyzed from BAL fluid by measuring lactate dehydrogenase leakage using an LDH assay kit (Sigma).
Lung Diffusing Capacity (DFCO) quantifies the ability of the lung to exchange gas, which is its primary function. Diffusing capacity is simple and quick to measure in humans and mice, and it decreases with nearly all lung pathologies, including viral infections. At the selected time points, a cohort of mice was anesthetized via an IP injection of ketamine–xylazine (100 mg/kg–10 mg/kg), and then an 18-g stub needle was secured in the trachea. 0.8 mL of a gas mixture containing 0.3% neon, 0.3% CO in room air was quickly injected into the lungs, held for 9 s, then quickly withdrawn. This post breathold sample was then injected into a desktop gas chromatograph (Inficon, Micro GC model 3000A) to measure the concentrations of Ne and CO. The DFCO in mice is analogous to the DLCO in humans, and is calculated as 1−(CO9/COc)/(Ne9/Nec), where subscripts c and 9 refer to the calibration gas injected and the gas from the 9 s exhaled sample. DFCO is thus a dimensionless variable which varies between 0 and 1, and is used to detect the loss and recovery of lung function after the viral infections used in this study [57].
Lung mechanics: After the DFCO is measured, the tracheostomy cannula was then connected to a Flexivent system (Scireq). Ventilation was accomplished at a rate of 150 breaths/minute and a tidal volume of 10 ml/kg with a PEEP of 3 cm H2O. A deep inspiration to 30 cmH2O was done, and 1 minute later the respiratory resistance (Rrs) and compliance (Crs) were measured [58]. Increased resistance reflects increased difficulty in dynamically moving air into the lung and decreased compliance reflects increased difficulty in expanding the lung parenchyma.
For mTEC cultures, tracheas were obtained from 7–9 week old C56BL/6 female mice, digested overnight in 0.3% pronase, and enriched by depleting fibroblasts as previously described [59, 60]. The mTECs were cultured at a density of 2.22x105 cells/ml on collagen-coated 24-well transwell plates for 7 days (i.e., until the cultures reached a transepithelial resistance above 1000 Ω· cm2) and apical medium was removed to create an air-liquid interface for 14 days to induce differentiation as described previously [60]. Cells were pre-treated for 24 h with basolateral media containing vehicle (100% ethanol) or 100nM P4 (Sigma), and injured by scratching the cell layer with a 10ul XL pipette tip, or left uninjured, and loose cells were removed by washing with media. Transepithelial cell resistance (TER) was measured prior to injury, immediately after, and every 12h for 48 h by adding 100μl of warm TEC basic media to the apical chamber. New media with vehicle or P4 was added every 24h. Every 12h, basolateral media was sampled and analyzed for AREG expression by ELISA (R&D) according to the manufacturer’s protocol. Cells were harvested in Trizol every 12h and analyzed by RT-PCR as described above.
A power and sample size calculation was used to confirm group sizes for a power of 0.8 and contributes to differential sample sizes for some dependent measures. Repeat measures were analyzed with a multivariate analysis of variance (MANOVA) followed by planned comparisons. Discrete measures were analyzed with T-tests or two-way ANOVA followed by the Tukey method for pairwise multiple comparisons. Survival was analyzed using a Kaplan Meyer survival curve followed by a log-rank test. Mean differences were considered statistically significant if P<0.05.
|
10.1371/journal.pcbi.1003419 | Combinatorial Modeling of Chromatin Features Quantitatively Predicts DNA Replication Timing in Drosophila | In metazoans, each cell type follows a characteristic, spatio-temporally regulated DNA replication program. Histone modifications (HMs) and chromatin binding proteins (CBPs) are fundamental for a faithful progression and completion of this process. However, no individual HM is strictly indispensable for origin function, suggesting that HMs may act combinatorially in analogy to the histone code hypothesis for transcriptional regulation. In contrast to gene expression however, the relationship between combinations of chromatin features and DNA replication timing has not yet been demonstrated. Here, by exploiting a comprehensive data collection consisting of 95 CBPs and HMs we investigated their combinatorial potential for the prediction of DNA replication timing in Drosophila using quantitative statistical models. We found that while combinations of CBPs exhibit moderate predictive power for replication timing, pairwise interactions between HMs lead to accurate predictions genome-wide that can be locally further improved by CBPs. Independent feature importance and model analyses led us to derive a simplified, biologically interpretable model of the relationship between chromatin landscape and replication timing reaching 80% of the full model accuracy using six model terms. Finally, we show that pairwise combinations of HMs are able to predict differential DNA replication timing across different cell types. All in all, our work provides support to the existence of combinatorial HM patterns for DNA replication and reveal cell-type independent key elements thereof, whose experimental investigation might contribute to elucidate the regulatory mode of this fundamental cellular process.
| Before a cell divides, its genome must be faithfully duplicated to ensure that the daughter cell receives an exact copy of the parental genetic material. However, this process requires disruption of chromatin, the combination of DNA and histone proteins, whose structure and function have to be readily restored afterwards. This is achieved through a nuclear process known as DNA replication, which represents the basis for biological inheritance. In eukaryotes, genome replication starts from distinct genomic locations termed replication origins. Origins fire in a temporally regulated, cell-type dependent manner and timing of DNA replication is therefore the result of this concerted origin activation. However, replication timing is not encoded in the genome and its regulatory mode remains to a large degree unresolved. Here, we systematically study the relationship between chromatin, represented by histone modifications and chromatin binding proteins, and DNA replication timing. We report combinatorial histone modification patterns exhibiting regulatory potential for this process and we characterize those elements that might contribute to further elucidate the regulatory mode of this fundamental cellular process.
| In eukaryotes, DNA replication is regulated both in time and space and initiates at multiple origins along the genome [1]. When averaged over a cell population, each genomic region shows reproducible replication timing in S-phase [2], [3]. The timing of replication is a mitotically stable cell-type specific feature of chromosomes [4] that was recently legitimated as an epigenetic feature [5]. For example, many tissue specific genes that are subject to developmental regulation are early replicating in their tissue of expression but rather late replicating in other tissues. Conversely, housekeeping genes expressed in almost all tissues are replicated in the first half of the S-phase [6], [7].
From an epigenetic point of view DNA replication constitutes a periodic window of both risk and opportunity. On one hand, established chromatin patterns of genome regulation are challenged by their disruption at the time of replication [8]. On the other hand, the same process paves the way for epigenetic changes and hence adaptation of cells to new cues. Our current understanding of the molecular mechanisms underlying eukaryotic DNA replication is the result of decades of experimental work that exploited model organisms as diverse as budding yeast, Xenopus laevis and Drosophila melanogaster [9]. Very recent work shed light on basic principles that regulate DNA replication timing at a global level [10]–[13]. Genome-wide profiling of DNA replication timing substantially contributed to these findings and a number of replication timing profiles are now available for different organisms and cell lines [4], [14]–[16]. The concurrent release of genome-wide profiles of histone modifications (HMs) and chromatin binding proteins (CBPs) through large scale genomic projects such as modENCODE and ENCODE represents a timely opportunity to systematically investigate the connection between replication timing and chromatin landscape. To date, chromatin feature levels have been individually correlated genome-wide to replication timing in different organisms [4], [15], [17], [18] and this studied extended single-locus-based observations to a genome-wide scale. Particularly, it is now accepted that euchromatin, gene dense, transcriptionally active regions of the genome preferentially replicate in early S-phase, as opposed to constitutive heterochromatin, repetitive, transcriptionally inactive regions that remain condensed throughout the cell cycle [1]. However, the observation that gene expression requires to be averaged over chromatin domains to strongly correlate with their replication timing [2], [19], suggested that this domain-like organization of replication timing might be regulated through higher-order chromatin structure [17], [20]. This, in turn, contributed to the development of qualitative models in which the chromosome accessibility of a domain affects its replication timing [2], [20].
Recent work linked HMs and CBPs levels to gene expression by means of quantitative statistical models [21]–[25], singling out a small number of HMs predicting the transcriptional output with high accuracy. However, HMs and CBPs also play a pivotal role in ensuring faithful completion of the DNA replication program [26]–[30]. As no individual HM has been found to be essential for origin function to date, it is likely that HMs act combinatorially in regulating DNA replication timing. Indeed, the view of chromatin as a platform for the assembly of different protein complexes in conjunction with the combinatorial nature of HMs led to the formulation of the hotly debated histone code hypothesis, in which specific combinations of HMs determine unique biological outputs [31]–[34]. Although proposed as a regulatory mechanism of chromatin-templated processes and well investigated for transcriptional regulation, this concept has to our knowledge not yet been demonstrated for DNA replication. Seminal work by Eaton et al. [15] tightened the link between chromatin features and DNA replication timing by showing that clusters of chromatin features are predictive for early origin activity and changes thereof in Drosophila. Here, we set out to systematically characterize this link and investigate the combinatorial relevance of chromatin features in predicting replication timing. Using a comprehensive data collection encompassing 95 HMs and CBPs profiled by the modENCODE project or independent studies in Drosophila cell lines, we asked the following five questions: i) Is there a quantitative relationship between HMs and CBPs levels and DNA replication timing? ii) Do these features act combinatorially and if yes, do HMs and CBPs convey redundant or distinct information? iii) Which features contribute the most in this relationship? iv) Do these rules apply genome-wide? v) Can these rules be generalized to various cell types? We addressed these points using Lasso (Least Absolute Shrinkage and Selection Operator), an L1-norm regularized linear model [35]. We systematically analyzed the predictive power of different subsets of chromatin features and combinatorial schemes thereof, applied feature importance analyses to obtain a simplified, biologically interpretable model and revealed cell-type independent combinations of chromatin features potentially impacting origin firing and likely to be conserved across species.
Recent studies reported moderate correlations between single chromatin features and DNA replication timing [15], [18], [36], [37]. However, these analyses were based on a rather limited number of genome-wide profiles. Here, we considered a genome-wide replication timing profile generated by [15] using tiling arrays and investigated the individual predictive power of a comprehensive set of 95 chromatin features (30 HMs - more precisely 28 HMs and 2 histone variants hereinafter collectively referred to as HMs - and 65 CBPs) profiled in Drosophila S2 cells using ChIP-chip or ChIP-Seq and generated by modENCODE [38] or independent studies. The goal of our study is to predict the replication timing across the Drosophila S2 genome. To this purpose, as the precise genomic coordinates of replication origins remain rather elusive in metazoans, we first considered a set of 7552 unique promoters (see Methods) for model learning. Several studies reported that replication initiation sites are associated with transcriptional units [14], [36], [39] and share common sequence motifs thereof [36]. In addition, the majority of ORC binding sites overlap with transcription start sites (TSSs) in Drosophila [40]. Feature levels and replication timing were therefore estimated for each promoter in a 1 kb window centered on its TSS (see Methods). As we integrated data sets generated by different laboratories and platforms, we first hierarchically clustered chromatin feature profiles at promoters and verified that feature levels reflected known biological associations between CBPs and HMs (Supplementary Figure S1). Then, for each feature we fitted cross-validated univariate linear regression models to analyze its predictive power on promoter-proximal replication timing. Our results confirm that individual features are rather poor predictors of replication timing (Supplementary Figure S2). Single HMs are on average significantly more predictive than individual CBPs (, two-sided t-test), but only few of them, i.e. H4K8ac, H3K36me1, H3K18ac, H4K5ac, H3K4me1, can predict replication timing with an accuracy (Pearson's correlation coefficient, hereinafter PCC or ρ) of . As previously shown [15], histone variants H2Av and H3.3 are positively correlated with replication timing. In addition, we found that levels of H4K5ac are predictive for early replication and that levels of H4K20me1, total H4 and linker histone H1, are individually predictive for late replication (Supplementary Figure S2). These results support the current view in which levels of acetylated and mono-methylated histones, localizing within euchromatin and marking accessible chromatin, are predictive for early replication, in contrast to levels of heterochromatic marks. Among CBPs, RNA Pol II (Pol II) and chromatin remodelers (such as ISWI, NURF and GAF) were previously shown to correlate with early replication timing in Drosophila [15]. Besides confirming these observations, our analysis highlights two CBPs, i.e. the chaperone protein Hsp90 and the ATP-dependent chromatin-remodeling factor dMi2, as top-ranked features predictive for early replication. The latter is involved in rapid nucleosome turnover, a distinguishing feature of origins of replication and promoters [41], and has been very recently implicated in regulation of higher-order chromatin structures and local decondensation of chromatin in Drosophila [42]. Hsp90 is involved in a number of chromatin processes [43]. Particularly, chromatin-associated Hsp90 is widespread genome-wide, where it binds to the TSSs of Pol II paused genes [44]. Our finding suggests that Hsp90 might be involved in regulating the timing of replication origin firing via a transcriptional-dependent or independent mechanism. However, experimental work will be required to detail this mechanism and to exclude an indirect role of Hsp90 as a marker of accessible chromatin. All in all, the limited predictive power of single features led us to hypothesize the existence of a combinatorial interplay between chromatin features enabling an accurate description of their relationship with replication timing. In the next sections, we test this hypothesis.
Quantitative modeling of the relationship between chromatin features and DNA replication timing requires testing of combinatorial patterns of chromatin features. In this high dimensional space, over-fitting represents a significant risk and therefore model regularization and cross-validation are required to effectively minimize it. Thus, our analysis is based on the statistical model Lasso (Least Absolute Shrinkage and Selection Operator, see Methods for details) [35], [45], a regularized linear model that penalizes model complexity through an L1 norm penalty. As a consequence, Lasso coefficients are sparse and feature selection is performed implicitly, facilitating model interpretation [35], [46]. Regularized regression methods have been previously employed to discover transcription factor binding motifs [47] and Lasso was very recently applied to predict RNA expression and promoter-proximal pausing from CBPs profiles [48].
Figure 1A illustrates our modeling framework. First, unique promoters were randomly partitioned into training and test sets (see Methods). Lasso models were then trained with ten-fold cross validation on the training set. To this purpose, the training set was randomly partitioned into ten subsets of equal size. Then, at each round of cross validation one subset was used in turn as validation set, while the model was learnt on the remaining nine subsets. The resulting ten models were averaged to obtain the cross-validated model. Prediction accuracy was evaluated on the test set and defined as the PCC between measured and predicted replication timing.
We started by analyzing the combinatorial predictive power of CBPs. When CBP levels were jointly considered, the model achieved an accuracy of (Figure 1B and Supplementary Figure S3A, B). Although significantly higher than the predictive power of any individual protein, this value is still modest. Thus, we investigated whether the addition of multiplicative interaction terms, in the form of second-order interactions, could raise the predictive power of CBPs. We found that allowing for pairwise interactions between CBPs significantly improved the model accuracy (, Figure 1C and Supplementary Figure S3A), suggesting that CBPs might combinatorially contribute to the regulation of replication timing. Notably, the higher predictive power of the latter model is not a mere consequence of an increased complexity as consideration of third-order interactions led to predictions that did not correlate any better with measured replication timing (ρ = 0.61, Supplementary Figure S3A, C). Taken together, these results indicate that CBPs and their pairwise interactions can account for a moderate yet substantial fraction, approximately 35%, of the variation in replication timing.
We next analyzed the relationship between HM levels and DNA replication timing. As for CBPs, we combined HMs using Lasso. The prediction accuracy achieved with HMs (ρ = 0.61, Figure 1B and Supplementary Figure S3A,D) is significantly higher than what we previously obtained with CBPs and as for the latter, significantly higher than the predictive power of any individual feature. As the histone code hypothesis postulated that HMs act combinatorially in regulating chromatin processes, with a one-to-one mapping between HM combinations and biological outcomes [31]–[33], we tested whether considering multiplicative second-order interactions between HMs could further increase the accuracy of the previous model. Inclusion of these combinations significantly raised the model accuracy from ρ = 0.61 to ρ = 0.69 (Figure 1B,D), suggesting that a combinatorial interplay between pairs of HMs might modulate DNA replication timing in Drosophila. This result suggested us to test whether more complex combinations, in the form of multiplicative third-order interactions between HMs, could bear even more predictive power than pairwise interactions. On the same line as for CBPs, we found that the prediction accuracy did not significantly increase (ρ = 0.69, Supplementary Figure S3A,E) solely as a consequence of a higher model complexity. Although this result implies that combinatorial patterns of HMs exhibit low complexity, in line with observations in vivo pertaining gene expression regulation [34], a very recent computational analysis showed that a simple histone code, based on modification at two histone residues, may suffice to generate a number of different circuits featuring heritable bistability [49].
In summary, we showed that HMs and their pairwise interactions are more predictive for replication timing than the corresponding terms involving CBPs. This result suggested to analyze the joint predictive power of CBPs and HMs and to test their redundancy for replication timing predictions.
To test whether CBPs and HMs convey redundant information on replication timing, we trained a Lasso model by jointly considering these two sets of features. We found that predictions based on combinations of CBPs and HMs exhibit a significantly lower cross-validated mean squared error (MSE) than the models trained on CBPs or HMs alone (Figure 1B) and thereby outperformed (ρ = 0.67, Supplementary Figure S4A) the accuracy of models solely based on CBPs (, Supplementary Figure S3B) or HMs (, Supplementary Figure S3D). However, this result indicates a partial redundancy between CBPs and HMs, which was further supported by a simple analysis of model residuals. As residuals are differences between measured and estimated DNA replication timing, they can be seen as information about replication timing that can not be explained by the model. Thus, we first considered the residuals of the model trained on CBP levels. Then, we tested whether HMs exhibit any predictive power for these residuals. Under the hypothesis that CBPs and HMs convey redundant information on replication timing, no correlation between model predictions and residuals is expected. Conversely, we found that HM levels can predict replication timing residuals with a highly significant yet moderate accuracy (). A similar result, despite a lower predictive power (), was obtained when CBPs were used to predict the residuals of the model trained solely on HMs. These results suggested us to investigate whether the introduction of CBPs could comparably raise the predictive power of second-order interactions between HMs. Indeed, CBPs in conjunction with pairwise interactions of HMs led to a model able to predict replication timing with higher accuracy (, Figure 1E) than HMs alone (, Figure 1D) and significantly reduced the cross-validated MSE as compared to the latter (Figure 1B). Finally, we tested whether allowing multiplicative cross-interactions between HMs and CBPs could further increase our ability to predict replication timing. However, despite a large increase in complexity this model did not outperform the previous one simply based on CBPs and interactions between HMs (, Supplementary Figure S4B), confirming once again that in our framework prediction accuracies are not a sheer consequence of the number of features. Similarly, further extension of the model by inclusion of RNA-Seq-based gene expression levels from [50] or multivariate Hidden Markov Model-based chromatin states [51], [52] from modENCODE [53] did not significantly improve prediction accuracy ( and data not shown). Taken together, these results indicate that CBPs and HMs are able to explain slightly more than 50% of the variation in DNA replication timing and suggest that these two sets of features contain partially complementary information that, when jointly captured, enable accurate predictions.
Here, we consider the Lasso model based on CBPs and pairwise interactions between HMs, we analyze feature importance and identify simplified models able to achieve a substantial fraction of the full model accuracy using few chromatin features. Although a measure of feature importance is not directly available for Lasso, different approaches can be employed to overcome this issue. First, the geometric constraints imposed to Lasso solutions result in an implicit feature selection [45]. This process depends on the extent of the regularization applied to the model, tuned by the parameter λ. The stronger the regularization, i.e. the higher λ, the smaller the number of selected features (see Methods for details). Consequently, there exists an entire set of Lasso models along the λ-path (i.e. the sequence of values of λ used to fit the model) each characterized by different model coefficients. Figure 2A shows the model coefficient curves along this path. Searching for simplified models is equivalent to identify those models with few non-zero coefficients and relatively high accuracy along the λ-path. Therefore, we considered all models reaching at least 70% of the full model accuracy and identified a first simplified model solely based on four terms involving four histone modifications, i.e. H3K36me1, H4K8ac, H2BUb and H3K79me1, able to reach an accuracy of , namely 76% of the full model accuracy. H3K36me1 and H4K8ac are predictive for early replication whereas pairwise interactions between H2BUb and H3K36me1 or H3K79me1 are predictive for late replication (Figure 2A). Interestingly, H3K36me1 exhibits opposite effects depending on whether it is considered alone or through its interaction with H2BUb, suggesting a context-dependent role of this modification.
If a group of features is characterized by high pairwise correlations, Lasso tends to arbitrarily select only one representative feature from the group [54]. However when not a single, but several models are fit on resampled data, feature selection frequencies can be used to estimate variable importance. Features indispensable to achieve high prediction accuracy will be selected with high frequency whereas redundant features or representatives of group effects will dilute their selection probabilities. Hence, to test whether the HMs identified above are indispensable for accurate predictions or rather representatives of functional groups of HMs, we estimated feature selection probabilities using bootstrap-Lasso [48]. In this method, data points are repeatedly sampled with replacement (bootstrap) to generate data sets used to train a full set of Lasso models along a fixed λ-path. The selection probability of each feature can then be estimated by considering the normalized frequency of non-zero coefficients (see Methods). Our bootstrap-Lasso analysis indicates that H3K36me1 and H4K8ac, followed by Hsp90, are selected with high probability and predictive for early replication timing (Figure 2B). Conversely, three terms involving H2BUb, namely the modification alone and its interaction with H3K36me1 and H3K79me1, are characterized by high selection probabilities and predictive for late replication. These results indicate that the previously identified simplified model was based on indispensable features and led us to test whether the addition of Hsp90 and H2BUb could further raise its predictive power. Indeed, we found that the inclusion of these two features substantially raised the prediction accuracy of the simplified model to , thus reaching 80% of the full model accuracy. The overall significance of H3K36me1, H4K8ac, H2BUb and Hsp90 in predicting replication timing was further substantiated using a bootstrap-based approach in which these features were individually excluded from the model fit (see Methods, Figure 2C). Furthermore, since the Hsp90 profile was generated by ChIP-Seq, we technically excluded that this feature was selected solely based on sequencing depth as Hsp90 was neither among highest coverage features (Supplementary Figure S5A) nor a strong correlation between coverage and individual predictive power of ChIP-Seq-derived chromatin features emerged in our analysis (, Supplementary Figure S5B).
Finally, we independently sought for simplified models using exhaustive model search as proposed in [21], [22]. To this purpose, we considered all possible combinations of two, three and four chromatin features and used each combination to train a multivariate linear regression model (see Methods). Prediction accuracies were recorded for a total of 3 188 010 models (Supplementary Figure S6). The Bayesian Information Criterion (BIC) was used to account for model complexity and monotonically decreased as more features were introduced, indicating that including up to four features is beneficial for prediction accuracy (Supplementary Figure S6A) irrespective of model complexity. Notice that we could not generate models with five or more features as the number of k-features models (n) grows with the binomial coefficient , i.e. for . However, we determined top two-features (H3K36me1, H4K8ac, ), top-three features (Hsp90, H2BK5ac, H4K8ac, ) and top-four features (H2BUb, H3K36me1, H3K36me3, Hsp90, ) models. Although these combinations differ slightly from the ones determined via bootstrap-Lasso, all features therein belong to at least one top-ranked simplified model. Moreover, by analyzing the frequency of appearance of chromatin features in four-features simplified models reaching at least 60% of the full model accuracy, we found that features constituting the bootstrap-Lasso simplified model were clearly overrepresented in the feature appearance profile (Figure 2D).
Collectively, these results highlight key combinations of HMs and interactions thereof that harbor most of the information about replication timing. These combinations are indispensable to achieve faithful predictions and likely to reflect regulatory principles conserved across species. Particularly, monomethylation of H3K36 by the yeast Set2 methyltransferase has been shown to regulate the time of Cdc45 association with origins. Cdc45 is recruited to replication origins at the time of initiation and this binding event is delayed in Set2 mutants, suggesting a direct involvement of H3K36me1 in replication initiation [55]. Histone hyper-acetylation marks active origins of the Drosophila chorion loci [28] and H4K8ac colocalizes with ORC at these developmentally regulated genomic regions. Chorion origin activity can be altered by tethering of the histone deacetylase Rpd3 or of the acetyltransferase Chameau (the ortholog of human MYST2/HBO1), which reduces and increases origin firing, respectively [28]. In addition, recent work indicated histone hypoacetylation as a requirement for maintaining late replication timing of constitutive heterochromatin [56], supporting a view in which histone acetylation levels modulate origin activity. The Ubiquitination of H2B by the ubiquitin ligase Bre1 plays multifaceted, transcriptional dependent as well as independent roles at chromatin. The mark is mostly euchromatic and has been shown to be required for efficient transmethylation of H3 at positions K4 and K79 [57]. Very recent work implicated H2BUb1 in yeast DNA replication [58], where the mark promotes nucleosome assembly and their stability behind advancing replication forks. Although our results may seem to contradict these findings, the impact of a variable on replication timing can be uncoupled from its role during the DNA replication process per se. Interestingly, H2BUb1 was shown to modulate the overall chromatin structure by inducing nucleosome stability and mediating chromatin compaction, in contrast to its supposed role in opening up chromatin [59]. Nucleosome stabilization, in turn, can result in transcriptional repression and a global increase of H2Bub1 levels has been shown to impede cell growth in yeast [59]. Thus, we propose a negative effect of H2BUb on replication timing of euchromatin, where H2BUb enriched regions are characterized by reduced accessibility and more stable nucleosomes. In addition, the two interaction terms involving H2BUb, namely H2BUb:H3K36me1 and H2BUb:H3K79me1, suggest a hierarchy whereby nucleosome stability exerts a dominant effect over the presence of activating marks. Alternatively, these pairwise interactions might indicate a role of Bre1-Set2 and Bre1-Dot1 complexes in delaying euchromatic origin firing. Finally, as the H2BUb antibody used to generate the H2BUb ChIP-chip profile is not specific for mono-ubiquitination, polyubiquitylation of H2B might also be responsible for the inferred effect of H2BUb on replication timing. In yeast, extensive H2B polyubiquitylation occurs with at least two distinct modes, Bre1-dependent and independent, suggesting distinct, yet not elucidated, biological functions [57].
To assess whether the combinations of chromatin features learnt at promoters allow accurate prediction of the genome-wide replication timing profile, we segmented the Drosophila genome in 10 kb bins and computed feature levels therein (see Methods). Then, we used the Lasso model based on CBPs and pairwise interactions between HMs trained at promoters to evaluate its accuracy in predicting the whole genome replication timing profile of S2 cells. Interestingly, we found that promoter-proximal combinations of chromatin features enable accurate genome-wide predictions (, Figure 3A), with comparable prediction accuracies between individual chromosome arms (, Supplementary Figure S7). These values are comparable to the accuracy obtained in promoter regions (, Figure 1E). Consistently with these results, the bootstrap-Lasso simplified model was able to predict the whole genome replication timing profile of S2 cells with an accuracy of (Supplementary Figure S8), the same value exhibited at promoters. These results indicate that combinations of chromatin features with regulatory potential for replication timing can be generalized to the whole genome and are therefore not confined to promoter-proximal regions.
Given the good agreement between experimentally determined and inferred values, we visually compared measurements and predictions as a function of their genomic position, as shown in Figure 3B,C for 6 Mb and 12 Mb of chromosomes 3R and 3L, respectively. This visualization allows us to further evaluate model predictions. First, although Lasso does not account for the spatial organization of DNA replication timing and of HMs, yielding predictions that are more noisy than measured values, the overall structure of the measured replication timing profile was faithfully recapitulated by the inferred one. Second, denoising of predicted values using adaptive smoothing (see Methods) further increased the correlation between measured and predicted values genome-wide () and made their similarity even more striking (Figure 3B,C). This result does not strongly depend on the degree of smoothing, as predicted profiles smoothed at 20 (), 40 () and 80 () kb resolution similarly correlate with measured replication timing values. Third, early-to-late and late-to-early transition zones, which coincide with boundary elements separating distinct chromosomal domains such as those flanking the late replicating Drosophila Bithorax complex [60] (Figure 3B, yellow rectangle), were accurately inferred by the model. However, we systematically investigated whether the prediction accuracy was uniform across different classes of genomic regions or whether some regions could be predicted with higher accuracy than others. To this purpose, we identified timing transition regions (TTRs) and replication domains using a circular binary segmentation algorithm (see [4] and Methods for details), and determined gene dense and poor regions using a two-state Hidden Markov Model (see Methods). We found that prediction accuracies were higher in replication domains () and gene poor regions () than in TTRs () and gene dense regions (), respectively (Supplementary Figure S9). The reduced prediction accuracy at TTRs might depend on the fact that these regions are devoid of replication origins and other chromatin features, and result from passive unidirectional replication fork movement [17]. In contrast, as the Drosophila genome is rather compact, it is plausible that feature averaging in gene dense bins partially reduces prediction accuracy in these regions. Alternatively, as our model globally underestimated early replication timing peaks and as gene density positively correlates with replication timing [36], a subset of chromatin determinants of early origin firing might not yet be part of the profiled CBPs and HMs and remains to be elucidated. Since CBPs are generally characterized by narrower peaks as compared to HMs and hence contribute more locally to replication timing predictions, it is likely that the missing features will correspond to CBPs exhibiting preferential binding to open chromatin.
We have shown that combinatorial modeling of chromatin features can accurately predict DNA replication timing in S2 cells. However, the chromatin landscape varies between cell types, and similarly, replication timing is a cell-type specific epigenetic feature [5]. Previous work from Eaton et al. [15] showed that clusters of chromatin features are predictive for changes in early origin strength across cell types. Thus, we focused on promoters and asked whether differences in the chromatin landscape between two cell types can explain the corresponding differences in their replication timing. Besides for S2 cells, genome-wide DNA replication timing and chromatin feature profiles are available for Drosophila Bg3 and Kc cell lines from modENCODE. The replication timing profiles of these two cell lines are highly correlated to each other and with the replication timing of S2 cells ( at promoters, Supplementary Figure S10). As the number of HMs profiled in both S2 and Bg3 is larger than those in common between S2 and Kc, we considered 21 HMs that were profiled in the former two cell lines (termed CHMs, in Common Histone Modifications, and listed in Methods) for further analyses. First, we assessed the predictive power of CHMs on replication timing at promoters in S2 and Bg3 cells, respectively. For each cell line, we trained a Lasso model based on the corresponding levels of CHMs and their pairwise interactions and obtained fairly accurate predictions in both cell types ( and in S2 and Bg3 cells, respectively; Supplementary Figure S11). These models are cell line specific as their accuracy in predicting unmatched replication timing profiles is significantly lower than the one achieved on the matched profile (data not shown). We next investigated whether differences in replication timing between S2 and Bg3 cells can be predicted from differences in CHMs (ΔCHMs) levels between these two cell lines. Therefore, we used ΔCHMs (S2-Bg3) and their pairwise interactions to predict differential replication timing (S2-Bg3, see Methods) and found that the model was able to achieve a prediction accuracy of (Figure 4A). Although this result indicates that inferring differences in replication timing is more challenging than inferring the timing per se, differences in HMs levels bear a fair predictive power on differential replication timing. Next, we investigated feature importance in predicting differential replication and estimated feature selection probabilities using bootstrap-Lasso as described before. We found that H3K18ac, H3K36me1 and its interactions with H3K27me3, H3K4me1 and H3K36me3, as well as H3K79me1 are selected with high probability and predictive for earlier replication in S2 than Bg3 cells (positive differences, Figure 4B). On the other hand, H3K9me2 and its interaction with H3K4me1, along with H2BUb levels, are stable predictors for later replication timing values in S2 than Bg3 cells (negative differences, Figure 4B). Overall, this analysis revealed that cell-type-specific differences in HMs are more predictive for differences in replication timing than cell-type-specific differences in interactions between HMs.
Finally, we narrowed our attention to differentially replicating promoters (DRPs) between S2 and Bg3 cell lines and asked whether the Lasso model trained on CHMs levels in S2 cells (Supplementary Figure S11A) can predict the replication timing of DRPs in Bg3 cells. The set of DRPs was defined using three different fold change cutoffs at the high end of the overall fold changes in replication timing (Figure 4C, see Methods). Notably, we found that the model based on S2 data was able to predict the replication timing of DRPs in Bg3 cells with high accuracy for all three cutoffs (, Figure 4C and Supplementary Figure S12). Prediction accuracies did not vary significantly upon further increase of the cutoff. Taken together, these results indicate that combinations of HMs allow a general, cell-type independent description of the relationship between replication timing and chromatin.
We systematically investigated the relevance of combinatorial HM patterns for DNA replication timing in Drosophila using Lasso. Developed on linear combinations of chromatin features from a comprehensive collection of HMs and CBPs profiles, our model quantitatively predicts replication timing with high accuracy genome-wide and across cell types. Our results show that combinations of HMs and their pairwise interactions are key in achieving accurate predictions, suggesting that combinatorial HM patters might indeed contribute to the regulation of DNA replication timing. However, it is important to notice that our data and analysis do not allow us to infer causality. Therefore, our description of the relationship between chromatin features and replication timing is a correlative one. In addition, there is a remaining 48% of variation in DNA replication timing that is not explained by our model. Accurate estimates of the maximal fraction of the observed variation in replication timing that could theoretically be explained by the model - e.g. following the recent approach proposed by [61] - were not possible in our framework due to lack of biological replicates for a subset of features and would have nevertheless been challenged by data integration across different laboratories and platforms. Unexplained variation can be possibly due to missing key features, presence of nonlinearities in the modeled relationship and existence of additional factors other than CBPs and HMs, such as the chromatin architecture, contributing to replication timing regulation. Although it is plausible that key determinants of DNA replication timing have not yet been profiled, it is unlikely that this aspect alone could entirely fill the gap. Since the regulatory mode of replication timing has not yet been fully elucidated, we hypothesized that a nonlinear relationship between chromatin landscape and replication timing could explain, at least partially, the remaining variation in DNA replication timing. We tested this hypothesis by using multivariate adaptive regression splines (MARS) [62], a flexible non-parametric regression technique based on piecewise linear basis functions which can also be adopted to estimate feature importance. However, MARS prediction accuracies were comparable to Lasso irrespective of model complexity (Supplementary Table S1), indicating that the relationship between chromatin feature levels and replication timing is well modeled by a linear function. For consistency, performances of the Lasso and MARS fits were also tested and confirmed using a second, independently generated, genome-wide replication timing profile in S2 cells [37] (Supplementary Table S1).
Through feature importance analyses, we identified a minimal set of six terms whose prediction accuracy reaches 80% of the full model accuracy. Remarkably, all elements within this set were selected by the MARS fit, with H4K8ac, H3K36me1, H2BUb:H3K36me1, H2BUb and Hsp90 indicated as the five most important terms. Besides demonstrating the necessity of these features to achieve high prediction accuracy, our results contribute experimentally testable, putative elements of a combinatorial HM pattern for DNA replication. In addition, availability of genome-wide profiles for these features in the same human or mouse cell line will enable to assess whether their predictive power is conserved across species. Finally, experimental investigation of our simplified model terms might unravel the mechanistic basis of their connection to DNA replication, and thereby, shed light on the regulatory mode of this fundamental cellular process.
Genome-wide replication timing profiles of Drosophila S2 and Bg3 cell lines (GEO accession numbers GSE17280 and GSE17281, respectively) were generated by Eaton et al. [15] using Agilent tiling arrays. Normalized smoothed M-values were used for the analysis. ChIP-Seq profiles of CBPs and HMs in S2 cells were downloaded as raw data in sra format from the Short Read Archive (SRA) or fetched from the Gene Expression Omnibus (GEO). Matched input datasets were downloaded where available. ChIP-chip profiles were downloaded from the modENCODE [38] data warehouse. Normalized smoothed M-values as provided by modENCODE were used for the analysis. If a feature was profiled more than once, only one profile was considered by taking into account antibodies characterization and technological platforms and by prioritizing deep sequencing based profiles. Pairwise Pearson's correlations ρ of feature signals at promoters were computed between the selected profile and all possible alternatives and typically . A list of the datasets included in the analysis is provided in Supplementary Tables S2 and S3 for CBPs and HMs, respectively.
Chromosome arms 2L, 2R, 3L and 3R were considered for the analysis. Chromosomes 4 and X were excluded due to special chromatin characteristics. Specifically, the single male X chromosome is hyperacetylated on H4K16 [63] and completes replication significantly earlier than the autosomes in male cell lines [37] whereas the fourth chromosome is predominantly heterochromatic and exhibits a high-transposon density [65]. For the prediction of DNA replication timing of promoters, Ensembl gene annotations were downloaded from biomart (www.biomart.org, genome assembly BDGP 5.12) using the R package biomaRt [64]. Promoter regions were defined as 1 kb windows centered on unique transcription start sites (TSS) in order to limit ambiguous assignment of chromatin feature signals to promoters. We defined a TSS as unique if no other TSS was annotated within the 1 kb genomic region flanking the TSS, regardless of the strand. A total of 7552 unique promoters was then considered for the analysis. For the prediction of genome-wide replication timing, the Drosophila genome was segmented into bins of width 10 kb. A total of 9663 bins was used for the analysis.
ChIP-Seq data in sra format were first converted to fastq format using the NCBI Short Read Archive Toolkit and subsequently aligned to the Flybase Drosophila melanogaster dm3 reference genome assembly r5.22 using Bowtie 0.12.8 with parameters [-n 2 -k 1, –best and -M 100]. Matched input datasets were aligned using the same parameters. The alignment output was converted from SAM to BAM format using SAMtools 0.1.18 and BAM files were imported in R using Rsamtools (Morgan, M. and Pagès, H., Rsamtools: Binary alignment (BAM), variant call (BCF), or tabix file import, R package version 1.8.6). Feature signals in both promoters and genomic bins were estimated as follows. Given a sample dataset S and an input dataset I the feature enrichment M of S relative to I within a given region of interest R was computed using available D replicates as follows. Let and be the library size of S and I, respectively and p an integer pseudocount used to avoid undefined values in logarithmic transformations ( in this analysis). Then, define . Finally, let and be the number of short reads entirely aligning within R for sample and input datasets, respectively. For each replicate d we then computed:and defined the feature enrichment as . For ChIP-chip datasets, the feature signal was computed as mean smoothed M-value within R. Similarly, the replication timing of R was computed as the average replication timing value of probesets mapping entirely within R.
Hierarchical clustering of chromatin features at promoters was performed using a correlation-based dissimilarity measure between feature signals at promoters. Given two profiles and , their dissimilarity d was computed as , where ρ denotes the Pearson's correlation coefficient.
The relationship between DNA replication timing and chromatin features was modeled using Lasso. Briefly, let be the dependent variable (DNA replication timing), be the enrichment matrix where is the number of promoters and the number of independent variables (chromatin features and when considered, their interaction terms) and the -th linear model coefficient associated to the -th independent variable. The Lasso parameters are then estimated as:where the first term corresponds to the residual sum of squares commonly minimized by multiple linear regression models and where the second term is the Lasso penalty that is tuned by the regularization parameter .
To fit the model, the set of 7552 unique promoters was randomly partitioned into two sets (5000 promoters) and (2552 promoters). The model was trained on with ten-fold cross validation. The cross-validated mean squared error (CV-MSE) as a function of was used to inspect the model fit. The value of minimizing the CV-MSE was used to predict the replication timing of the test set . The Pearson correlation coefficient between measured and predicted continuous replication timing values on was used to determine the model accuracy.
Simplified models were obtained using three different approaches: i) By analyzing the coefficients of the Lasso model based on CBPs and interactions between HMs along the -path used to fit the model. Only models leading to a prediction accuracy of at least 75% of the prediction accuracy achieved by the full model were considered; ii) By performing stability analysis of model coefficients (see below); iii) By generating all possible combinations of two (4 371), three (134 044) and four (3 049 501) features and training a multiple linear regression model based on each combination following the same procedure described above for the Lasso model fit. The Bayesian Information Criterion was used to account for model complexity and assess whether increasing the number of features was still beneficial for the model fit.
Stability analysis of model coefficients was performed essentially as described in [48]. Feature selection probabilities (normalized frequencies of non-zero coefficients) were computed using bootstrap-Lasso. Briefly, a Lasso model based on CBP levels and interactions between HMs was trained with ten-fold cross validation using all 7552 unique promoters. The values of the regularization parameter yielding an empty model () and an almost full model () were used to construct a sequence of 100 values ranging from to with constant ratios between consecutive elements. This sequence was then used to fit 100 Lasso models on 100 bootstrap samples of 7552 promoters. Model coefficients were stored for each value of and for each fitted model. Finally, for each chromatin feature the number of non-zero coefficients was summed and normalized to the total number of recorded coefficients. Normalized values represent the estimated selection probabilities.
The overall significance of H3K36me1, H4K8ac, H2BUb and Hsp90 in predicting replication timing was estimated using a bootstrap-based approach. For each feature, 100 bootstrap samples of 7552 promoters were generated and partitioned into and as above. Each sample was used to train a Lasso model based on CBPs and pairwise interactions of HMs but lacking all model terms involving the selected feature with ten-fold cross validation. Model accuracies (PCC) on were recorded and compared to the accuracies achieved by full models trained on the same bootstrap samples using a two-sided Wilcoxon rank sum test.
Adaptive smoothing of predicted genome-wide replication timing profile was performed using a maximum overlap discrete wavelet transform (MODWT). In pratice, we used the R package waveslim (Whitcher,B., waveslim: Basic wavelet routines for one-, two- and three-dimensional signal processing, R package version 1.7.1) with la8 wavelet filter, J = 2 and reflecting boundaries.
Timing transition regions (TTRs) were identified using the circular binary segmentation algorithm implemented in the R package DNAcopy (Venkatraman,E.S. and Olshen,A., DNAcopy: DNA copy number data analysis, R package version 1.32.0) according to [4]. Replication timing of the 9663 Drosophila genomic bins at 10 kb resolution was provided as input and a 30 kb window centered on each identified domain boundary was used to define a TTR. Visual inspection of the segmented replication timing profile was performed and verified accurate recognition of TTRs.
Gene density along the Drosophila genome was computed using Ensembl gene annotations at a 10 kb resolution and used to classify gene poor and gene dense regions by learning a two-state Hidden Markov Model (HMM). The HMM was fit using the Baum-Welch algorithm implemented in the R package RHmm (Taramasco,O. and Bauer,S., RHmm: Hidden Markov Models simulations and estimations, R package version 2.0.3) and the optimal hidden states sequence was computed using the Viterbi algorithm.
A subset of 21 HMs (termed in Common Histone Modifications, CHMs) that have been profiled by modENCODE in both S2 and Bg3 cell lines was considered. This set includes the following features: H1, H2BUb, H3K18ac, H3K23ac, H3K27ac, H3K27me2, H3K27me3, H3K36me1, H3K36me3, H3K4me1, H3K4me3, H3K79me1, H3K79me2, H3K9acS10P, H3K9ac, H3K9me1, H3K9me2, H3K9me3, H4K16ac, H4K20me and H4. Feature scoring and computation of DNA replication timing at unique promoters were performed as described above. The predictive power of CHMs on replication timing of S2 and Bg3 promoters was evaluated using Lasso models based on second-order interactions between CHMs. For each cell line, a model was trained on with ten-fold cross validation using the corresponding CHMs levels. Model accuracy (PCC) was determined on .
To test whether differential CHMs between cell lines can predict differential replication timing (S2-Bg3, Δt), we computed differences in CHMs levels between S2 and Bg3 cells (ΔCHMs) and used them to predict Δt through a Lasso model with pairwise interactions. To predict the replication timing of differentially replicating promoters (DRPs) in Bg3 cells, we defined DRPs based on log fold change differences between S2 and Bg3 promoters using three increasing cutoff values (0.8, 0.9 and 1.0). The Lasso model introduced above to evaluate the predictive power of CHMs in S2 cells was then applied to infer the replication timing of DRPs in Bg3 cells.3
All analyses were performed using R 3.0.0 (R Core Team, R: A Language and Environment for Statistical Computing, http://www.R-project.org). Custom R scripts are available from https://github.com/FedericoComoglio/ToR.
|
10.1371/journal.pntd.0002475 | Aquaporin 2 Mutations in Trypanosoma brucei gambiense Field Isolates Correlate with Decreased Susceptibility to Pentamidine and Melarsoprol | The predominant mechanism of drug resistance in African trypanosomes is decreased drug uptake due to loss-of-function mutations in the genes for the transporters that mediate drug import. The role of transporters as determinants of drug susceptibility is well documented from laboratory-selected Trypanosoma brucei mutants. But clinical isolates, especially of T. b. gambiense, are less amenable to experimental investigation since they do not readily grow in culture without prior adaptation. Here we analyze a selected panel of 16 T. brucei ssp. field isolates that (i) have been adapted to axenic in vitro cultivation and (ii) mostly stem from treatment-refractory cases. For each isolate, we quantify the sensitivity to melarsoprol, pentamidine, and diminazene, and sequence the genomic loci of the transporter genes TbAT1 and TbAQP2. The former encodes the well-characterized aminopurine permease P2 which transports several trypanocides including melarsoprol, pentamidine, and diminazene. We find that diminazene-resistant field isolates of T. b. brucei and T. b. rhodesiense carry the same set of point mutations in TbAT1 that was previously described from lab mutants. Aquaglyceroporin 2 has only recently been identified as a second transporter involved in melarsoprol/pentamidine cross-resistance. Here we describe two different kinds of TbAQP2 mutations found in T. b. gambiense field isolates: simple loss of TbAQP2, or loss of wild-type TbAQP2 allele combined with the formation of a novel type of TbAQP2/3 chimera. The identified mutant T. b. gambiense are 40- to 50-fold less sensitive to pentamidine and 3- to 5-times less sensitive to melarsoprol than the reference isolates. We thus demonstrate for the first time that rearrangements of the TbAQP2/TbAQP3 locus accompanied by TbAQP2 gene loss also occur in the field, and that the T. b. gambiense carrying such mutations correlate with a significantly reduced susceptibility to pentamidine and melarsoprol.
| Human African Trypanosomiasis, or sleeping sickness, is a fatal disease restricted to sub-Saharan Africa, caused by Trypanosoma brucei gambiense and T. b. rhodesiense. The treatment relies on chemotherapy exclusively. Drug resistance in T. brucei was investigated mainly in laboratory-selected lines and found to be linked to mutations in transporters. The adenosine transporter TbAT1 and the aquaglyceroporin TbAQP2 have been implicated in sensitivity to melarsoprol and pentamidine. Mutations in these transporters rendered trypanosomes less susceptible to either drug. Here we analyze T. brucei isolates from the field, focusing on isolates from patients where melarsoprol treatment has failed. We genotype those isolates to test for mutations in TbAQP2 or TbAT1, and phenotype for sensitivity to pentamidine and melarsoprol. Six T. b. gambiense isolates were found to carry mutations in TbAQP2. These isolates stemmed from relapse patients and exhibited significantly reduced sensitivity to pentamidine and melarsoprol as determined in cell culture. These findings indicate that mutations in TbAQP2 are present in the field, correlate with loss of sensitivity to pentamidine and melarsoprol, and might be responsible for melarsoprol treatment failures.
| The chemotherapy of human African trypanosomiasis (HAT, also known as sleeping sickness) currently relies on suramin or pentamidine for the first, haemolymphatic stage and on melarsoprol or eflornithine/nifurtimox combination therapy (NECT) for the second stage, when the trypanosomes have invaded the central nervous system (CNS) [1]. All five drugs have unfavorable pharmacokinetics and adverse effects. Melarsoprol is particularly toxic, causing severe encephalopathies in over 5% of the treated patients [2]. And yet, melarsoprol is the only treatment for late-stage T. b. rhodesiense infections. New and safer drugs are at various stages of (pre)clinical development, thanks largely to the Drugs for Neglected Diseases initiative (www.dndi.org). Two molecules that have successfully passed clinical Phase I trials are now being tested in patients: the nitroimidazole fexinidazole [3], [4] and the benzoxaborole SCYX-7158 [5], [6]. Both are orally available and cure 2nd stage T. b. brucei infections in a mouse model [7]. However, until new drugs for HAT are on the market, the current ones – problematic as they are – need to be used in a sustainable way. This requires an understanding of the mechanisms of drug resistance.
The mechanisms of drug resistance in African trypanosomes have been studied in the lab for over 100 years [8]. Two observations were made recurrently, namely (i) reduced drug uptake by drug resistant trypanosomes [9]–[14] and (ii) cross-resistance between melarsoprol and pentamidine [15], [16]. Both phenomena were attributed to the fact that melarsoprol and pentamidine are taken up by trypanosomes via the same transporters, which appeared to be lacking in drug-resistant mutants. The first transporter identified was called P2 since it was one of two purine nucleoside transporters identified [17], [18]. It is encoded by the gene TbAT1 for adenine/adenosine transporter 1 [19]. Homozygous genetic deletion of TbAT1 in bloodstream-form T. b. brucei resulted in pentamidine and melarsoprol cross-resistance, albeit only by a factor of about 2.5 [20]. This weak phenotype, together with the fact that the TbAT1−/− mutants still exhibited saturable drug import [21], indicated that further transporters are involved in melarsoprol-pentamidine cross-resistance [16], [21], [22]. One such transporter was recently identified, the aquaglyceroporin TbAQP2 [23], [24]. Aquaporins and aquaglyceroporins belong to the major intrinsic protein (MIP) family and form channels that facilitate transmembrane transport of water and small non-ionic solutes such as glycerol and urea [25]. The three aquaporins of T. brucei (TbAQP1-3) are thought to physiologically function as osmoregulators and are involved in glycerol transport [26]. Aquaporins were described to mediate uptake of arsenite in mammalian cells [27] and in Leishmania, and loss of aquaporin function was implicated in heavy metal resistance [28]. Homozygous genetic deletion of TbAQP2 in bloodstream-form T. b. brucei increased the IC50 towards melarsoprol and pentamidine by about 2- and 15- fold, respectively [24]. Moreover, a T. b. brucei lab mutant selected for high-level pentamidine resistance [21] carried a chimeric TbAQP2 gene, where 272 nucleotides had been replaced by the corresponding sequence from a neighboring, very similar gene TbAQP3 [24]. Differences in the TbAQP2/TbAQP3 tandem locus on chromosome 10 were also observed between the reference genome sequences of T. b. gambiense DAL972 [29] and T. b. brucei TREU927 [23], [30]. They possess identical versions of TbAQP2 but differ in TbAQP3 [31]. More recent field isolates of T. brucei ssp. have so far not been genotyped regarding their TbAQP2/TbAQP3 locus.
The genotypic status of TbAT1, located proximal to a telomere on chromosome 5 [32], has been more intensely investigated. Point mutations in TbAT1 were described, both in selected lab strains and in clinical T. brucei ssp. isolates, which rendered the gene non-functional when expressed in yeast [19]. The occurrence of these mutations correlated to a certain degree with melarsoprol treatment failure in 2nd stage T. b. gambiense HAT patients [33]–[36]. However, the relationship between polymorphisms in TbAT1, drug susceptibility, and treatment failure in patients is not fully resolved as the TbAT1 mutant T. b. gambiense were not analyzed phenotypically. Such investigations are notoriously difficult since clinical T. b. gambiense isolates are hard to obtain (given the inaccessibility of HAT foci and the poor success rate of isolation and adaptation in rodents) and cannot readily be propagated in axenic culture. Here we concentrate on clinical T. brucei ssp. isolates from drug refractory cases that have been adapted to axenic in vitro cultivation, aiming to investigate whether mutations at the known melarsoprol and pentamidine transporter loci also occur in the field – and if so, whether such mutations are accompanied by loss of drug susceptibility.
The 16 analyzed isolates are described in Table 1 (origin) and Table 2 (clinical outcome). For more details on the recent isolates from the DRC please refer to Table S4 of Pyana et al (2011) [37]. All have previously been adapted to axenic cultivation. T. b. brucei and T. b. rhodesiense isolates were cultured in minimum essential medium (MEM) with Earle's salts with the addition of 0.2 mM 2-mercaptoethanol, 1 mM Na-pyruvate, 0.5 mM hypoxanthine, and 15% heat-inactivated horse serum as described by Baltz et al (1985) [38]. T. b. gambiense strains were cultured in IMDM medium supplemented according to Hirumi and Hirumi (1989) [39], plus 0.2 mM 2-mercaptoethanol, 15% heat-inactivated fetal calf serum and 5% human serum. The cultures were maintained under a humidified 5% CO2 atmosphere at 37°C and were subpassaged 3 times a week to ensure growth in the exponential (log) phase.
Drug sensitivity was determined with the Alamar blue assay as described by Räz et al (1997) [40], using the redox-sensitive dye resazurin as an indicator of cell number and viability. The trypanosomes were cultivated in 96-well microtiter plates in serial dilutions of drugs for 70 h. 10 ul of resazurin (125 ug/ml (Sigma) dissolved in PBS pH 7.2) was added to each well. The plates were further incubated for 2–4 hours for T. b. rhodesiense and T. b. brucei, and 6–8 hours for T. b. gambiense, before being read with a SpectraMax Gemini XS microplate fluorescence scanner (Molecular Devices) at an excitation wavelength of 536 nm and an emission wavelength of 588 nm. IC50 values were calculated by non-linear regression to a sigmoidal inhibition curve using SoftMax Pro software (V. 5.2). The IC50 values given in Table 2 are averages ± standard deviation of at least 3 independent assays (n = 3–12), each determined in duplicate. Melarsoprol (Sanofi-Aventis) was obtained from WHO. Pentamidine isothionate and diminazene aceturate were purchased from Sigma.
Genomic DNA was isolated from 10 ml dense trypanosome cultures. The cells were spun down and the pellets resuspended in 300 µl 10 mM TrisHCl pH 8, 1 mM EDTA and 3 µl 10% SDS was added before incubating for 10–15 min at 55°C. After 5 min incubation 3 µl of pronase mix (20 mg/ml, Sigma) was added to increase the stability of the extracted DNA. 90 µl of ice cold 5 M potassium acetate was added and the mixture was incubated for 5 min on ice. After spinning down for 5 minutes at max speed in a microfuge, the supernatant was transferred to a new tube and DNA was precipitated in 2–2.5 volumes of absolute ethanol, washed in 70% ethanol and dissolved in 20 µl ddH2O. PCR was performed with Taq polymerase (Solis BioDyne, Estonia); the primers and annealing temperatures are summarized in Table S1. PCR products were run on a 0.8% agarose gel and purified on a silica membrane column (Nucleospin gel and PCR clean up, Macherey Nagel, Germany). The purified PCR products were directly sequenced (Microsynth, Switzerland or GATC, Germany) with the same primers as used for PCR amplification. Only the TbAQP2/TbAQP3 locus of T. b. gambiense K03048 produced two PCR products, which were cloned in pCR2.1-TOPO (Invitrogen). The assembled sequences were submitted to GenBank; accession numbers are listed in Table S2.
To be able to compare – and possibly correlate – genotype and phenotype of T. brucei ssp., we assembled a set of 16 isolates that had been adapted to axenic in vitro cultivation as blood-stream forms. These included 5 recent T. b. gambiense isolates from the Democratic Republic of the Congo (DRC), 2 older isolates from the Republic of Côte d'Ivoire and one isolate from South Sudan, which were all isolated from patients who had relapsed after melarsoprol chemotherapy. Other T. b. gambiense isolates from the DRC, northwestern Uganda, and Liberia were from patients who were successfully treated with melarsoprol or the treatment outcome is unknown. T. b. gambiense STIB 930 is a fully drug-susceptible lab strain that was used as a reference strain. We further included the field isolates T. b. brucei STIB 940, T. b. brucei STIB 950 and T. b. rhodesiense STIB 871, which are multidrug-resistant to isometamidium, diminazene and tubercidin. The fully drug-susceptible reference strain T. b. rhodesiense STIB 900 was included as a reference. The different isolates and their origin are summarized in Table 1. All isolates were genotyped regarding TbAQP2 and TbAT1.
When the TbAQP2/TbAQP3 genomic locus was amplified by PCR from the 16 T. brucei ssp. isolates, all the recent T. b. gambiense isolates from the DRC (40 AT, 45 BT, 130 BT, 349 BT and 349 AT) exhibited a smaller band than expected for the wild-type locus. Direct sequencing of the PCR product in each of the five isolates revealed only one gene at the locus: a chimeric version of TbAQP2 and TbAQP3. The first 813 bp of the open reading frame perfectly matched TbAQP2 while the remaining 126 bp derived from TbAQP3 (Figure 1C). These 126 bp perfectly matched to TbAQP3 of T. b. rhodesiense STIB 900 but this exact sequence is not found in the published genome of T. b. gambiense DAL 972. Note that the present TbAQP2-TbAQP3 chimeric gene (Figure 1C) differs from the one described by Baker et al. from a pentamidine-selected T. b. brucei lab mutant (Figure 1B; [24]). T. b. gambiense K03048 from the South Sudan also gave rise to an abnormal pattern upon PCR amplification of the TbAQP2/TbAQP3 locus from genomic DNA: a distinctly smaller double band instead of the expected product, indicative of heterozygosity. The smaller band contained the upstream region of TbAQP2 followed by the open reading frame of TbAQP3 while the TbAQP2 open reading frame was missing (Figure 1D). The larger band contained a TbAQP2/3 chimera similar to that encountered in the T. b. gambiense isolates of the DRC (Figure 1C). Point mutations in TbAQP2 were encountered in the multidrug-resistant field isolates T. b. brucei STIB 940, T. b. brucei STIB 950 and T. b. rhodesiense STIB 871, all of which had the same 4 SNPs in TbAQP2 compared to the T. b. brucei 927 reference gene (Tb927.10.14170), leading to the amino acid change threonine159 to alanine (Figure 1E). However, the same 4 SNPs also occurred in our drug-susceptible reference strain T. b. rhodesiense STIB 900, so they are not likely to be involved in the mdr phenotype [41], [42] of these isolates. All other isolates analyzed had a wild-type copy of TbAQP2. The identified sequence polymorphisms are summarized in Table 2, GenBank accession numbers are in Table S2.
All of the 12 analyzed T. b. gambiense isolates were identical in TbAT1 sequence to the reference STIB 930 as well as to the genome strain DAL972. The previously described TbAT1R allele [19], [33] was found in the 3 mdr lines T. b. brucei STIB 940, T. b. brucei STIB 950 and T. b. rhodesiense STIB 871. TbAT1R carries 5 coding and 4 silent mutations and a codon deletion as compared to the reference sequence (STIB 900), and the resultant protein appeared to be non-functional when expressed in Saccharomyces cerevisiae [19] or re-expressed in a tbat1 null T. b. brucei (De Koning, unpublished results). The remainder of the isolates did not possess mutations in TbAT1 when compared to the respective reference isolate. The GenBank accession numbers of all the sequences are in Table S2.
Drug sensitivities of the bloodstream-forms of all isolates were determined in vitro regarding melarsoprol, pentamidine, and diminazene. The five T. b. gambiense that possessed the chimeric TbAQP2/3 gene (45 BT, 130 BT, 349 BT, 349 AT, 40 AT), as well as K03048 which carries a deletion of TbAQP2 in one allele, in addition to one chimeric TbAQP2/3 allele, all showed a similar drug sensitivity profile with markedly increased IC50 values towards pentamidine and, to a lesser extent, also melarsoprol (Figure 2). IC50 values were in the range of 70–92 nM for pentamidine and 22–42 nM for melarsoprol (Table 2); compared to the median of the four drug sensitive T. b. gambiense lines STIB 930, STIB 891, STIB 756 and ITMAP 141267, this corresponds to a 40- to 52-fold decrease in susceptibility to pentamidine and a 2.8- to 5.3-fold decrease for melarsoprol. The higher IC50 values of the isolates that carried a mutation in TbAQP2 (n = 6) compared to the remainder (n = 10) were statistically significant both with respect to pentamidine (p = 0.0002, two-tailed Mann-Whitney test) and melarsoprol (p = 0.0047); no association was observed regarding TbAQP2 status and sensitivity to diminazene. However, the isolates that carried the known resistance allele TbAT1R (i.e. STIB 940, STIB 950 and STIB 871) exhibited strongly increased IC50 values to diminazene (p = 0.01, two-tailed Mann-Whitney test) but not to pentamidine (Figure 2, Table 2). T. b. brucei STIB 950 also had an elevated IC50 against melarsoprol (Figure 2), but over all three TbAT1R isolates there was no significant effect on melarsoprol susceptibility.
Across all 16 T. brucei isolates, pentamidine sensitivity positively correlated with that to melarsoprol (Spearman's rank correlation coefficient of 0.67, p = 0.005) while there was no correlation between the two structurally related diamidines, pentamidine and diminazene (Figure 2).
It is an intriguing phenomenon with African trypanosomes that drug resistance is predominantly linked to reduced drug import, typically arising from loss of function mutation of a non-essential transporter [12], . Here we investigated the aminopurine transporter TbAT1 and the aquaglyceroporin TbAQP2, two proteins known to be involved in uptake of – and susceptibility to – melarsoprol and diamidines in bloodstream-form T. brucei. While there is evidence for a link between TbAT1 mutations and melarsoprol treatment failure in the field [33]–[36], the more recently identified gene TbAQP2 has so far not been analyzed in a clinical setting. TbAQP2 is dispensable for growth in culture [24] and partial gene replacement of TbAQP2 with TbAQP3 was observed in a pentamidine-selected T. b. brucei lab mutant [24] that displayed reduced infectivity to rodents [21]. However, it was unknown whether similar mutations also occur in the field, as they might bear a fitness cost in patients or during transmission by the tsetse fly. Concentrating on a panel of clinical T. brucei ssp. isolates that (i) derived from treatment-refractory cases and (ii) had been adapted to axenic in vitro culture, we have genotyped their TbAT1 and TbAQP2 loci, and phenotyped their in vitro sensitivity towards melarsoprol, pentamidine and diminazene. Our aim was to explore whether TbAQP2 mutations occur in the field and if so, whether mutant isolates exhibit reduced drug susceptibility.
Five of the analyzed T. b. gambiense isolates, all from melarsoprol relapse patients of Dipumba Hospital in Mbuji-Mayi, DRC, carried only one gene at the TbAQP2/TbAQP3 tandem locus, an unprecedented TbAQP2/3 chimera. The high degree of sequence similarity between TbAPQ2 and TbAQP3 allows for homologous recombination between the two genes, leading to chimerization and gene loss. TbAQP2 has a unique selectivity filter with unusual NSA/NPS motifs instead of the characteristic NPA/NPA that occur in the vast majority of MIP family members [43] including TbAQP1 and TbAQP3 [24]. The published, pentamidine-resistant T. b. brucei lab mutant possessed a TbAQP2/3 chimera whose C-terminal filter triplet was from TbAQP3, suggesting that the unusual NPS triplet may be involved in pentamidine transport. However, the presently described pentamidine-resistant T. b. gambiense isolates carry a TbAQP2/3 chimera encoding a predicted protein with both selectivity filter triplets from TbAQP2. We hypothesize that the TbAQP2/3 chimera observed in the T. b. gambiense isolates fails to contribute to pentamidine and melarsoprol susceptibility despite having the proposed selectivity filter residues of TbAQP2. Functional expression of the chimeric gene in tbaqp2 null cells will be necessary to test this hypothesis.
The occurrence of rearrangements at the TbAQP2/TbAQP3 locus correlated with reduced susceptibility to pentamidine and, to a lesser extent, melarsoprol. Thus field isolates also exhibit the well known cross-resistance between melarsoprol and pentamidine 15,16,31, while no cross-resistance was observed to diminazene aceturate. This is in agreement with TbAT1 being the primary uptake route for diminazene [44], [45] and consistent with results obtained using TbAQP2−/− cells, which showed no resistance to the rigid diamidines diminazene or DB75 [24], as opposed to pentamidine which has a highly flexible structure. It is also noteworthy that T. b. rhodesiense STIB 871 and T. b. brucei STIB 940 are susceptible to melarsoprol and pentamidine in vitro although both carry the TbAT1r allele. Loss of TbAT1 function has been described without mutations in the open reading frame of the gene [32]. However, since in the present study all isolates with a ‘wild-type’ TbAT1 ORF were fully susceptible to diminazene, we conclude that they possess a functional TbAT1 (i.e. P2) transporter. Trypanosoma congolense and T. vivax appear to lack an AT1 orthologue [46], therefore diminazene transport and resistance must have a different mechanism in these livestock parasites.
The plasma levels of pentamidine in treated patients peak about 1 hour after injection and vary extensively from 0.42 µM to 13 µM, while the mean elimination half-life after multiple applications is approximately 12 days [47]. Thus, since pentamidine is very potent, even a 50-fold increase in IC50 of pentamidine as observed here for the T. b. gambiense isolates with mutations in TbAQP2, is unlikely to jeopardize the success of treatment. With melarsoprol, however, the obtainable drug levels are more critical. Only 1–2% of the maximal plasma levels are seen in the CSF [48], and a 5-fold reduced sensitivity to melarsoprol might allow trypanosomes to survive in the CSF during melarsoprol therapy. Thus mutations in TbAQP2 might indeed be responsible for melarsoprol treatment failures with T. b. gambiense. However, two of the T. b. gambiense isolates from relapse patients (DAL 870R and DAL 898 R) were sensitive to melarsoprol and pentamidine, and they possessed wild-type copies of TbAT1 and TbAQP2, indicating that factors other than drug resistance can contribute to treatment failures. Larger sample sizes will be required to test the significance of TbAQP2 for successful treatment. We show here for the first time that a TbAQP2/3 chimera as well as loss of TbAQP2 occurs in T. b. gambiense clinical isolates, and that the presence of such rearrangements at the TbAQP2/TbAQP3 locus is accompanied by a 40- to 50-fold loss in pentamidine sensitivity and a 3- to 5-fold loss in melarsoprol sensitivity. We recommend genotyping of the TbAQP2/TbAQP3 locus to be integrated into larger field trials such as clinical studies with drug candidates.
|
10.1371/journal.ppat.1005833 | The RNA Binding Specificity of Human APOBEC3 Proteins Resembles That of HIV-1 Nucleocapsid | The APOBEC3 (A3) cytidine deaminases are antiretroviral proteins, whose targets include human immunodeficiency virus type-1 (HIV-1). Their incorporation into viral particles is critical for antiviral activity and is driven by interactions with the RNA molecules that are packaged into virions. However, it is unclear whether A3 proteins preferentially target RNA molecules that are destined to be packaged and if so, how. Using cross-linking immunoprecipitation sequencing (CLIP-seq), we determined the RNA binding preferences of the A3F, A3G and A3H proteins. We found that A3 proteins bind preferentially to RNA segments with particular properties, both in cells and in virions. Specifically, A3 proteins target RNA sequences that are G-rich and/or A-rich and are not scanned by ribosomes during translation. Comparative analyses of HIV-1 Gag, nucleocapsid (NC) and A3 RNA binding to HIV-1 RNA in cells and virions revealed the striking finding that A3 proteins partially mimic the RNA binding specificity of the HIV-1 NC protein. These findings suggest a model for A3 incorporation into HIV-1 virions in which an NC-like RNA binding specificity is determined by nucleotide composition rather than sequence. This model reconciles the promiscuity of A3 RNA binding that has been observed in previous studies with a presumed advantage that would accompany selective binding to RNAs that are destined to be packaged into virions.
| Cellular intrinsic immunity constitutes a key defense against infection by viruses. The APOBEC3 (A3) family of cytidine deaminases are intrinsic immune proteins that can hypermutate and destabilize retroviral genomes. For A3 proteins to exert their antiviral activity, they must be incorporated into nascent virions. Although A3 RNA binding activity has been shown to be critical for virion incorporation, the mechanism by which packaged RNA molecules are targeted by A3 proteins has been unclear. Therefore, we employed a cross-linking and deep sequencing strategy to analyze the targets of A3 proteins in HIV-1 infected cells and purified virions. We found that A3 proteins preferentially bind to particular types of RNA sequence in HIV-1 and cellular RNAs. In particular, we found that A3 proteins bind to G-rich and A-rich RNA sequences, a property that is reminiscent of that exhibited by the HIV-1 Gag protein. Further analyses revealed that A3 proteins preferentially bind the same sequences as the nucleocapsid domain of the HIV-1 Gag protein in viral RNA and cellular 7SL RNA. These results suggest a model of A3 incorporation into virions that that is based on mimicry of the RNA binding specificity of the HIV-1 nucleocapid protein.
| APOBEC3 (A3) proteins are a family of germline-encoded proteins that inhibit the replication of a broad range of viruses and retroelements (reviewed in [1, 2]). A3 proteins exert their antiretroviral activity largely through their deoxycytosine deaminase activity, i.e. modification of dC-to-dU in single-stranded DNA retroviral reverse transcription intermediates, resulting in dG-to-dA hypermutation of the viral genome and error catastrophe [3–6]. In addition to inflicting genetic damage, alternative deamination-independent antiretroviral mechanisms have also been reported [7–11].
In primates, there are seven members of the A3 protein family that are categorized by their possession of either a single zinc (Z)-containing deaminase domain (A3A, A3C and A3H), or two Z domains (A3B, A3D, A3F and A3G) (reviewed in [12]). The antiviral activity of some A3 proteins is antagonized by most lentiviruses, including human immunodeficiency virus type-1 (HIV-1) through the action of the virion infectivity factor (Vif) protein. Vif targets A3 proteins for polyubiquitination and subsequent proteasomal degradation through the recruitment of core binding factor-β (CBF-β) and an E3 ubiquitin ligase complex, comprised of cullin 5, elongin B/C, and Rbx2 [13–19]. In the absence of a functional Vif protein, some A3 proteins are efficiently incorporated into progeny HIV-1 virions, enabling them to exert their antiviral effects during subsequent infection of a target cell.
Packaging of A3 proteins into HIV-1 virions depends on the nucleocapsid (NC) region of the viral Gag polyprotein and its associated RNA [20–25]. In the case of A3G, a large pocket within the A3G amino-terminal domain has been shown to contain residues that are critical for RNA binding, efficient particle incorporation and restriction [26–30]. These observations have led to the conclusion that A3 interacts with the NC region of Gag, indirectly, in an RNA-dependent manner for incorporation into virions.
Studies that have aimed to determine the identity of the RNA that is targeted by A3 proteins for incorporation into virions have yielded a variety of conclusions. One study indicated that viral RNA is targeted by A3 proteins [31], while another proposed that A3 proteins target 7SL [32], a cellular RNA that is normally part of the signal recognition particle (SRP) ribonucleoprotein complex but is enriched in retrovirus particles for unknown reasons [33–36]. Other reports have indicated that both cellular and viral RNA can be targeted by A3 proteins [21, 23]. This latter notion has been supported by a recent study in which incorporation of A3F and A3G into virions could be driven by diverse RNA molecules [37]. Overall, these studies lead to a model in which A3 proteins are promiscuous, non-specific RNA binding proteins and are able to efficiently infiltrate nascent HIV-1 virions by binding to unoccupied regions of nearly any RNA in an infected cell.
Notably, if it is true that A3-RNA interaction is completely non-selective, then cellular RNAs would compete with viral RNAs for A3 binding. Because cellular RNAs are present in large excess over HIV-1 RNAs in infected cells, their presence should inhibit A3 incorporation into HIV-1 virions to a significant degree. Indeed, such a scenario would require that A3 proteins be associated with a large fraction of the RNA molecules in the cell in order to be incorporated into a correspondingly large fraction of virions. While some RNA binding proteins associate with RNA in this completely non-specific manner, many other RNA binding proteins bind to their target RNA molecules through the recognition of specific RNA sequences and/or structures (reviewed in [38]), using discrete, structured RNA binding domains (reviewed in [39, 40]). The RNA binding activity of A3G has been mapped to a small number of amino acids [26–30], suggesting the existence of a discrete RNA binding site, and hinting at a degree of RNA binding selectivity. However, the structural basis for RNA recognition by A3 proteins remains undefined. These considerations make it intuitively surprising that A3 proteins would bind to their target RNA molecules without recognizing specific RNA sequences or elements to at least some degree. Moreover, it is unclear how potentially competitive Gag and NC RNA binding in cells and virions would affect A3 binding to viral and cellular RNAs during virion genesis.
For these reasons, we undertook a detailed study of the interactions between A3 proteins and RNA. Recent advances in ribonomic technologies, such as cross-linking immunoprecipitation coupled to next generation sequencing (CLIP-seq (reviewed in [41]) have enabled high resolution mapping of protein-RNA interactions. We employed CLIP-seq to determine how A3 proteins that have potent antiviral activity and a different complement of Z domains (A3F, A3G and A3H) interact with RNA in cells and virions. While we confirmed that A3 proteins bind to several classes of RNAs in infected and uninfected cells and virions, we found that HIV-1 RNA was bound preferentially over cellular RNA in infected cells. Notably, like HIV-1 Gag and NC, we found that A3 proteins target RNA sequences that are G-rich and A-rich. Comparative analyses of Gag, NC and A3 binding in cells and immature and mature virions revealed that A3 proteins target sequences that are also preferred binding sites for the NC protein in the viral genome and the cellular 7SL RNA. Thus, these data suggest a model in which A3 incorporation into HIV-1 virions is facilitated by its ability to preferentially bind G- and A- rich RNA sequences, partly mimicking properties of the NC domain of the HIV-1 Gag protein [42]. This model reconciles the apparent promiscuity of A3 RNA binding with its demonstrated ability to be incorporated into virions in the presence of a vast excess of potentially distracting cellular RNA molecules.
We employed CLIP-seq techniques [42–44] to determine the RNA binding specificity of three unique members of the A3 protein family that exhibit potent anti-HIV-1 activity (A3F, A3G and A3H) (Fig 1). We first generated HEK 293T cell lines that stably express amino-terminally HA-tagged A3F and A3G and carboxyl-terminally HA-tagged A3H proteins. Then, cells were mock infected or infected with vesicular stomatitis virus G (VSV-G)-pseudotyped HIV-1NL4-3 ΔVif and cultured in the presence of the ribonucleoside analog 4-thiouridine (4SU). Cells and purified virions were UV-irradiated, lysed, and digested with ribonuclease A. Thereafter, A3-RNA complexes were immunoprecipitated using an anti-HA antibody, 5'-end labeled with γ-32P-ATP and detected by autoradiography and Western blotting. As expected, A3F, A3G and A3H from cells and purified virions were cross-linked to RNA (Fig 2A). A3F was immunoprecipitated with reduced efficiency compared to A3G and A3H, resulting in fewer RNA molecules that were cross-linked to A3F for further analyses.
RNA oligonucleotides that were cross-linked to A3 were released by proteinase K digestion, purified, sequentially ligated to 5' and 3' adapters and converted to cDNA. After PCR amplification and next-generation sequencing of the resulting cDNA library, reads were subsequently mapped to the HIV-1 genome. The read density, a measure of the incidence of A3 binding to a specific RNA sequence on the HIV-1NL4-3 genome was determined by plotting the frequency with which each individual nucleotide in the viral genome was detected in mapped reads from CLIP libraries derived from cells and virions.
In cells, binding of A3F, A3G and A3H was observed at numerous sites throughout the viral genome. Discrete regions of the viral genome that yielded a high frequency of A3-bound reads were proximal to and interspersed with regions with low numbers of reads (Fig 2B and S1 Fig). The frequency of A3-binding events in the 3' half of the HIV-1NL4-3 RNA genome was higher than in the 5' half (Fig 2B). This finding potentially reflects the abundance of viral transcripts containing the 3' half of the viral genome, as it is represented in both spliced and unspliced viral mRNAs transcripts, while the 5' half is only present in unspliced transcripts. Correlation analyses revealed very clear reproducibility in the viral RNA sites occupied preferentially by each A3 protein in independent biological replicates, supporting the notion that A3 proteins bind preferentially to particular RNA sequences (Fig 2B and S1 Fig).
In virions, the read density peaks did not exhibit a bias towards the 3' half of the genome and instead were distributed across the entire length of the viral RNA (Fig 2B and S1 Fig). This likely reflects the uniform availability of viral sequences across the genome, as the full-length unspliced genomic RNA is selectively packaged into virions. As in cells, binding of A3F, A3G and A3H to viral RNA in mature virions reproducibly occurred at discrete sites (Fig 2B), with read density peaks containing high frequencies of reads interspersed with regions containing low frequencies of reads.
We were interested in how the RNA binding specificity of A3 proteins would be similar or different in a natural target of HIV-1, and so we performed A3F and A3G CLIP-seq experiments in MT4 cells as a representative T-cell line. Importantly, CLIP-seq experiments performed in a T- cell line (MT4) resulted in a similar distribution of binding sites on HIV-1 RNA indicating that cellular context does not greatly affect the HIV-1 RNA binding specificity of A3F and A3G (S2 Fig). Furthermore, using different crosslinking nucleotides (4SU or 6SG) resulted in a similar distribution of A3G and A3H binding sites on viral RNA (S3A and S3B Fig) indicating that the selective binding of A3 proteins to discrete sequence elements was not an artifact of the use of 4SU as the cross-linking nucleotide.
To investigate potential similarities and differences in A3:RNA binding that might exist among the three A3 proteins in infected cells versus mature virions, we performed correlation analyses of RNA binding sites (Fig 3A and S4A Fig). Despite the fact that RNA sequences are differentially represented in cells versus virions, statistically significant correlation between RNA binding preferences in cells and mature virions was observed for each A3 protein, signifying that the overall binding specificity is consistent in the two environments, although some discrepancies were observed. The binding patterns of A3F, A3G and A3H to viral RNA revealed some differences in their intrinsic RNA binding specificities. For example, A3H had an increased preference for binding the R-U5 region of the viral RNA compared to A3F and A3G in cells (Fig 2B). Nevertheless, pair-wise comparisons of A3F, A3G and A3H binding sites in the HIV-1 genome in both cells and mature virions (Fig 3B and S4B Fig) showed statistically significant correlation of binding preferences among the A3 proteins. However, correlation functions for A3G or A3F versus A3H had less statistical support than did the A3F versus A3G comparison (Fig 3B). These findings imply that although the overall patterns of RNA binding of A3F, A3G and A3H are similar, A3H has subtle differences in RNA binding specificity compared to A3F and A3G, concordant with the fact that A3H (containing a single Z3 domain) is divergent from A3F and A3G (which contain Z1/Z1 and Z1/Z2 domains, respectively).
While the A3 proteins bound to many sites on the viral genome, the aforementioned results suggested a degree of specificity in A3-RNA binding. To investigate RNA target specificity of A3 proteins we undertook a detailed inspection of RNA types and sequences that were most frequently bound by A3F, A3G and A3H in uninfected cells, infected cells and in purified virions. We aligned the raw reads derived from A3 CLIP-seq experiments done using infected cells to both the human and viral genomes and found that 2.5%–4.1% of the total reads were HIV-1 derived, whereas 97.5–95.9% were from cellular RNA (Fig 4A). In comparison, RNA sequencing (RNA-seq) libraries from identically infected cells contained only 0.3%–0.5% of reads derived from HIV-1, and ~99% were from cellular RNA (Fig 4A). Thus, viral RNA sequences appeared to be selectively bound (~5 to 12-fold more frequently than their occurrence) by A3 proteins in infected cells.
To investigate the basis for this apparent specificity, A3 CLIP reads were aligned to the human genome and the HIV-1 genome and analyzed using PARalyzer [45]. This approach defines a cluster, or preferred binding site, based on the incidence of a minimum number of overlapping reads that are proximal to dT-to-dC nucleotide substitutions that occur at the cross-linking site in 4SU-based CLIP assays. We counted the number of reads associated with each cluster, and each analysis was performed at least twice on independent biological replicate datasets. In uninfected cells, the majority of reads associated with clusters were mRNA-derived (~78–86%) with a minority of reads within clusters mapping to other cellular RNA types including tRNA, 7SL RNA, miRNA, and rRNA (Fig 4B). In HIV-1 infected cells, the majority of the cluster-associated reads were also messenger RNA (~74–78%) with reads derived from viral RNA representing ~4–8% of the total cluster-associated reads (Fig 4B). Indeed, viral RNA was the most frequently bound single RNA species in infected cells for each of the A3 proteins. In purified virions, the majority of cluster-associated reads were viral (~77–88%) with a minority of reads mapping to the cellular RNA that is present in HIV-1 virions (Fig 4B). In virions, 7SL RNAs represented 6.2% and 13.4% of reads bound by A3G and A3H respectively, whereas only 0.6% of the reads bound to A3F were from 7SL RNA in purified virions (Fig 4B). We also conducted CLIP-seq experiments using 6SG in A3G and A3H infected cells. This analysis resulted in a similar identification and classification of RNA binding sites, albeit with an even greater apparent selectivity of A3 proteins for viral RNA (S3C Fig). The greater number of G than U nucleotides in the HIV-1 genome may contribute to this slight discrepancy between 6SG and 4SU-based CLIP experiments.
As mRNAs constituted the preferred target of A3 proteins, compared to other types of cellular RNAs, we analyzed the binding pattern of A3 proteins to the 10 most frequently A3 bound mRNAs (S5, S6, S7 and S8 Figs). Comparisons of the read densities across the length of these mRNA sequences in the CLIP and RNA-seq experiments did not correlate, indicating that A3 proteins preferentially target specific portions of these transcripts, rather than binding evenly along the length of the transcript. Indeed, each of the A3 proteins exhibited a strong preference for binding to the 3'UTR sequences of these mRNAs, which likely reflects the lack of available sequences for binding in regions of the mRNAs that are actively translated by ribosomes (S5, S6, S7 and S8 Figs).
These results clearly showed that A3 proteins have diverse RNA targets, but that binding is not indiscriminate. In cells, mRNAs represent the bulk of A3-bound RNA, with a clear preference for 3' untranslated sequences. Moreover, A3 proteins preferentially target viral RNA sequences in infected cells. These findings indicate that A3 proteins are incorporated into HIV-1 virions predominantly via interactions with viral RNA that are preferred over interactions with cellular mRNAs or other RNAs. However, it was unclear how viral RNA might be selected by A3 proteins from a diverse pool of RNAs in the cell. Note that some HIV-1 RNA species, including the full length packaged viral genome, have long 3'UTRs which might contribute to their apparent selection for binding by the A3 proteins (Fig 4A and 4B). An additional possibility was that A3 proteins might have selectivity for certain nucleotides or motifs, that occur with increased frequency in viral RNAs.
To investigate this possibility, we isolated the 100 sites (clusters) in a complex RNA source (cellular mRNAs) that were most frequently bound by A3F, A3G and A3H in cells. This analysis selected clusters representing RNA regions of 200–900 nucleotides in length. Thus, they likely represent concentrations of several A3 protein binding sites rather than individual A3 binding sites. Next, we determined the nucleotide composition of these clusters, or collections of binding sites. This analysis revealed that the clusters had a striking propensity to be G-rich and/or A- rich (mean G-, A-, C- and U- content of ~35%, ~28%, ~17% and ~20%, respectively) compared to mean G-, A-, C- and U- content of ~24%, ~36%, ~18% and ~22%, of the HIV-1NL4-3 genome, and ~23%, ~25%, ~22% and ~30%, respectively, for human RNA as determined by our RNA-seq analysis) (Fig 5A). We also determined the nucleotide composition of short clusters (<50 nucleotides) where A3 protein binding had most frequently occurred, and were more likely to represent small numbers, or individual A3 binding sites. Similarly, in the case of A3F and A3G, we found these clusters to be G- and A-rich (mean G-, A-, C- and U- content of ~31%, ~30%, ~16% and ~23%, respectively), while A3H bound clusters were G-rich, but not as A-rich as A3F or A3G bound clusters (mean G-, A-, C- and U- content of ~32%, ~25%, ~18% and ~24%, respectively) (Fig 5A).
Additionally, we used cERMIT [46] to examine sequence motifs that occurred most frequently in A3-bound RNA clusters in uninfected cells, infected cells and immature and mature purified virions (Fig 5B). In this analysis, G-rich sequence motifs were most often identified in A3-bound clusters in uninfected cells. Conversely, in infected cells, the most frequently identified motifs in A3F and A3G bound RNAs were A-rich in addition to being G-rich. The unusual A-rich nature of the HIV-1 genome, and the apparent selectivity of A3 proteins for viral RNA could be responsible for this effect. However, motifs in A3H bound RNAs were mostly G-rich in both infected and uninfected cells. This finding along with the finding that short A3H binding clusters had a higher propensity to be G-rich rather than A-rich suggests that A3H might have a higher preference for G-rich rather than A-rich RNA as compared to A3F and A3G. Nevertheless, in purified immature and mature virions, A3-bound clusters were found to be G/A-rich, a result that is undoubtedly influenced by the dominance of the A-rich viral RNA in virions. Preferred A3 protein binding sites were correspondingly U- and C-poor, implying that G/A-rich sequences overall represent preferred binding sites for A3 proteins in infected cells.
We noticed that the apparent preference of A3 proteins for G-rich and A-rich RNA sequences was reminiscent of the two modes of RNA binding exhibited by the HIV-1 Gag and NC proteins, before, during and after HIV-1 particle assembly [42]. Therefore, to investigate the relationship between A3 and Gag binding specificities, data obtained in A3 and Gag CLIP experiments [42] were subjected to correlation analysis (Figs 6 and 7, S9 and S10 Figs). The RNA binding profile of A3 proteins and Gag on the HIV-1 genome in infected cells exhibited no statistically significant correlation (Figs 6A and S9A). However, we noticed a marked reduction in A3 binding at sites of high Gag occupancy, including the 5' leader and Rev Response Element (RRE). This finding suggested that A3 might be occluded from these sites as a consequence of high Gag occupancy, or alternatively as a consequence of a high degree of RNA secondary structure at these sites. Although correlation between Gag and A3 binding to viral RNA in cells was not statistically significant, we did observe similar binding profiles in regions of the viral genome that are thought to have less secondary RNA structure (e.g. nucleotides 4000–6000, S9A Fig). The lack of overall statistical significance is likely be due to the fact that the dominant Gag binding signal to viral RNA in cells occurs in highly structured regions of RNA, including psi (Ψ) and the RRE while A3 has been reported to exhibit preference for single-stranded RNA [47].
We next compared the RNA binding profile of A3F, A3G and A3H to the HIV-1 genome with that of NC and Gag in mature and immature virions, respectively, using NC and Gag CLIP-seq data [42]. Remarkably, statistically significant and, in many areas, visually obvious correlation between A3-RNA and NC-RNA binding was observed in mature virions (Figs 6B, S9B). This was especially true for A3G and A3H, indicating that A3 and NC proteins have a similar RNA binding specificity in mature HIV-1 virions. In immature (protease deficient) virions, A3 proteins exhibited a similar pattern of binding to viral RNA as they did mature virions, albeit with some regions in which clear differences were apparent (Fig 7A and 7B, S10A Fig). The discrepancies suggest that some degree of change in A3 RNA binding specificity or target RNA availability occurs during virion maturation. Curiously, A3F RNA binding exhibited higher degree of correlation in mature versus immature virions than did A3G- and A3H-RNA binding (Fig 7B, S10A Fig).
In immature virions, the patterns of A3G and A3H protein binding to the viral genome did not correlate with the pattern of Gag binding (Fig 7C and S10B Fig), in contrast to the situation in mature virions (Figs 6B, S9B). However, binding of A3F and Gag to the viral genome in immature virions were weakly correlated (Fig 7C and S10B Fig). While the intrinsic RNA binding specificity of A3 proteins is not expected to change during virion genesis, the manner in which Gag competes with A3 proteins for sites on the viral RNA is expected to change, as Gag-RNA binding specificity changes dramatically during virion genesis [42]. These findings suggest that A3F better mimics the RNA binding specificity of HIV-1 Gag in immature virions, or is better able to compete with Gag for RNA binding sites than are A3G or A3H.
The aforementioned observations indicated that some degree of change in the RNA binding profile of A3 protein to viral RNA occurs during virion genesis, likely as a consequence of RNA representation, and competition with immature Gag during viron genesis. Ultimately, however, the A3 proteins and NC demonstrate an apparently similar viral RNA binding specificity in mature virions. To investigate whether these changes occurred with a different RNA target, we examined 7SL1, a cellular RNA that is ordinarily part of the SRP ribonucleoprotein complex but is highly enriched in the virions, through binding the HIV-1 Gag and NC proteins. 7SL is also bound by A3G and A3H proteins in infected cells and in virions [32–35, 48]. Specifically, we compared the A3G read densities with Gag and NC read densities on the 7SL1 RNA in infected cells, immature virions and mature virions, respectively (Fig 8). In cells, there was little correlation between A3 and Gag binding on the 7SL1 RNA. Indeed, A3 binding was observed in regions that had relatively few Gag binding events and vice versa. Strikingly, however, as Gag assembled into immature and then mature virions there was a progressive harmonization of binding patterns, and there was clear correlation between A3G and Gag or NC RNA binding in both immature and particularly mature virions. This finding reinforces the notion that A3 and HIV-1 NC proteins have apparently similar RNA binding specificities, using 2 different target RNAs, and suggests that competition by other proteins (including perhaps SRP proteins that bind 7SL proteins in cells) may affect RNA target availability and the apparent propensity of A3 proteins to bind to particular RNA elements in cells.
The RNA binding activity of A3 proteins is essential for incorporation into HIV-1 virions and restriction of virus replication. Our data confirm previous studies in that A3 proteins are promiscuous in the RNA molecules that they bind. Importantly however, our findings also argue that A3 proteins exhibit significant RNA binding selectivity that could facilitate their incorporation into nascent HIV-1 virions and thus contribute to their antiviral activity.
The majority of molecules that are bound by A3 proteins in cells were diverse messenger RNAs, while other cellular RNAs including 7SL RNA, miRNA, tRNA, rRNA were bound less frequently. The promiscuity of A3 protein binding is consistent with previous observations that A3G forms high molecular mass ribonucleoprotein complexes in cells that contain numerous cellular RNAs and RNA binding proteins [49] including poly(A)-binding proteins (PABPs), YB-1, RNA helicases, ribosomal proteins, and Staufen1 [50]. A3G has been found to localize to mRNA processing (P) bodies, cytoplasmic compartments that are responsible for the degradation and storage of non-translating messenger RNAs [51]. Although P-body-association does not appear to be necessary for antiviral activity [52, 53], RNA binding could obviously influence, or be influenced by, localization to these RNA-rich cellular structures.
Although RNA binding by A3 proteins is promiscuous, our data show that A3 proteins preferentially bind HIV-1 viral RNA in infected cells. Moreover, A3F, A3G and A3H proteins bound, with similar, reproducible, non-uniform distributions along the length of the HIV-1 genome. The RNA sequences that are preferentially bound by A3 proteins, and regions of the viral genome that are preferentially targeted could be governed by several factors including: (i) RNA abundance (ii) RNA binding specificity inherent to the A3 proteins (iii) secondary RNA structures present in RNAs, that might differ in cells and in virions and (iv) competition by other RNA binding proteins (most notably Gag). Like all lentiviruses, HIV-1 genomic RNA has a characteristic A-rich nucleotide composition, and also exhibits secondary structure, with regions that have greater or lesser propensity to exhibit single-stranded and double-stranded character [54]. HIV-1 RNA also associates intimately with Gag during virion genesis. Overall, our data suggests that each of the above factors impacts how A3 proteins bind to viral RNA prior to, during and after virion encapsidation.
Analysis of A3 binding to a complex RNA source (cellular mRNA) revealed that A3F, A3G and A3H each preferred to bind to G-rich or A-rich sites, and motif analysis confirmed this evident G/A-preference. This finding is consistent with a previous study which found that A3-NC complex formation could be driven in vitro by single-stranded RNA molecules that contain G residues, but not single-stranded RNA molecules that lacked G-bases (note that NC is also selective for G-bases) [47]. This modest level of selectivity likely underlies A3 binding to diverse RNA molecules, and would create a profound barrier to the selection of A3-resistant viral strains through changes in A3 binding sites on viral RNA. However, this level of selectivity would also reduce the number of ‘distracting’ RNA binding sites in cellular RNAs that are sampled by A3 proteins, and thereby potentially increase the efficiency with which A3 proteins are incorporated into virions, as compared to a completely nonspecific RNA binding protein.
Previous work has indicated that A3G binds specifically in vitro to single-stranded but not double-stranded RNA [47]. Thus, the accessibility of single stranded RNA would likely affect the pattern of HIV-1 genome binding. Indeed, our CLIP data show that the Rev Response Element (RRE) is disfavored for binding by each human A3 protein examined in this study and binding to the 5' leader was disfavored by A3F and A3G. However, A3H exhibited significant binding to the 5' leader, suggesting that RNA binding specificity is not determined by the presence or absence of RNA structure alone in all cases.
We previously reported that the viral Gag and NC proteins, like A3 proteins, bind to G- and A-rich rich sequences at different stages of virion genesis [42]. Nevertheless, an unexpected finding was that A3 proteins and NC targeted similar sites in the HIV-1 genome in mature virions. These findings suggest that A3 has evolved to partially mimic the RNA binding specificity of the HIV-1 NC domain of Gag, with NC targeting G-rich and A-rich sequences in order to ensure RNA packaging and A3 proteins targeting RNA in a similar manner to drive its own efficient incorporation into nascent virions. The apparently similar RNA binding specificity exhibited by HIV-1 NC and A3 proteins for viral RNA sequences was also evident with 7SL1 RNA, which is abundantly incorporated into virions. However, we also noted that the pattern of A3 binding to both viral and 7SL1 RNAs changed somewhat as virions assembled and matured. One likely explanation for this is the particularly intimate RNA association of the immature Gag protein, transiently, during virion genesis. Perhaps some of the preferred binding sites of the A3 proteins are occluded by Gag during immature virion assembly. Indeed, it is conceivable the RNA occlusion could be a countermeasure against A3 proteins that might be employed by retroviruses that lack a Vif protein.
Overall, this study suggests a model whereby A3 protein encapsidation into virions is facilitated by an RNA binding preference that has been acquired through natural selection so as to strike a balance between selectivity and promiscuity. Thus, only a subset of possible RNA binding sites, that are loosely defined by their nucleotide composition and enriched in lentiviral RNAs, are sampled by A3F, A3G and A3H proteins. Thus, many potentially distracting RNA binding sites are avoided. Conversely, A3F-, A3G- and A3H-RNA binding appears sufficiently non-selective so as to render the evolution of viral escape mutants whose RNA genome does not bind these proteins near impossible, and may thus have driven the acquisition of the vif gene for antagonism of A3 proteins.
The pLHCX-3×HA-A3F and pLHCX-3×HA-A3G plasmids were constructed by PCR-amplification of the open reading frames from pCR3.1 expression vectors encoding A3F and A3G with primers that contain 5' and 3' NotI and XhoI sites respectively. Amplicons were inserted into a pCR3.1 expression vector containing sequence encoding a 3×HA-tag 5' to the NotI site. The 3×HA-A3F and 3×HA-A3G cassettes were then subcloned into pLHCX using SnaBI and HpaI sites. The pLHCX-A3H-3×HA plasmid was generated by PCR-amplification of the open reading frame from a pTR600-A3H-3×HA plasmid [55] with primers that contain terminal HindIII sites for digestion and ligation into the pLHCX vector. HIV-1NL4-3 ΔVif has been previously described [56]. HIV-1NL4-3 ΔVif/PR−was constructed by ligating an AgeI/SpeI fragment encoding an inactive PR protein of a previously described HIVNL4-3/PR−plasmid [57] into HIV-1NL4-3 ΔVif.
Human Embryonic Kidney (HEK) 293T cells and MT4 cells (American Type Culture Collection) were maintained in Dulbecco’s Modified Eagle Medium (DMEM; Corning) or Roswell Park Memorial Institute medium (RPMI; Corning), respectively, supplemented with 10% fetal calf serum (FCS; Corning) and gentamicin. HEK 293T cells and MT4 cells that stably express 3×HA-tagged A3 proteins were produced by transduction of cells with MLV-based retroviral vectors. To generate the MLV-based retroviral vectors, approximately 5 ×105 HEK 293T cells were co-transfected using polyethyleneimine (PolySciences) with 200 ng of a vesicular stomatitis virus G protein (VSV-G) expression plasmid, 1 μg of a MLV GagPol expression plasmid, and 1 μg of a 3×HA-tagged protein-expressing MLV vector. At 48 h post-transfection, cell culture supernatants containing MLV-based retroviral vectors were harvested, filtered and 200 μl was used to transduce approximately 5 ×105 cells in the presence of 5 μg/ml polybrene (Hexadimethrine bromide; Sigma). At 48 h post-transduction, cells were selected in 50 μg/ml hygromycin-B (Corning). Single cell clones were derived by limiting dilution, picked and recultured, and protein expression was confirmed by Western blotting. The HIV-1NL4-3 ΔVif virus that was used to infect cells was prepared by co-transfecting approximately 8 ×106 HEK 293T cells with 22 μg of the HIV-1NL4-3 ΔVif plasmid and 5.5 μg of the VSV-G expression plasmid using polyethylenimine. At 48 h post-transfection cell culture supernatants containing virions were filtered and virus titers were determined on MT4-LTR-GFP indicator cells using FACS analysis.
The mouse monoclonal anti-HA (HA.11 Covance) was used in CLIP assays. The rabbit polyclonal anti-HA (600-401-384 Rockland) and the mouse monoclonal anti-HIV-1 p24CA (183-H12-5C, NIH AIDS Reagent Program) were used in Western blot analyses.
The CLIP method that was used in this study has been described previously [42] and was adapted from previously reported HITS-CLIP and PAR-CLIP protocols [43, 44], and is described briefly here.
The analysis pipeline that was used in this study has been previously described [42]. Briefly, the FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit/) was used to process the raw reads prior to mapping. Reads that were fewer than 15 nt, did not contain the 3' adapter sequence or contained ambiguous nucleotides were excluded from further analyses. Reads were subsequently aligned using Bowtie [58] to the human genome (hg19) concatenated with the HIV-1NL4-3 genome, or to the viral genome alone, allowing for a maximum of 2 mismatches Reads derived from the R region of the HIV-1NL4-3 LTR have been displayed at the 5' end of the genome in our analyses and interpreted with caution. SAMtools [59] was used to generate pileups of the mapped reads which further analysed using in-house scripts. Cluster analysis was performed using PARalyzer [45] using parameters as previously described [42] and the generated clusters were annotated using in-house scripts according to the EMSEMBL v72 database [60]. PARalyzer-generated clusters were then used as input for the cERMIT motif finding tool [46]. Previously described PERL scripts [42] were used for determining nucleotide composition of clusters. Correlation analysis was examined by means of a correlation function (CF), the determination of which has been previously described in detail [42]. Briefly, the existence of a significant level of correlation between read density frequencies in two different data sets was examined by determining a CF. Given two data series, the CF for a certain separation of nucleotides (s) can be defined. High values of CF at s = 0 indicate that the locations and relative peak heights of the binding sites in the two data samples are correlated. Data sets derived from APOBEC3 CLIP-seq experiments in cells (not virions) show significantly positive values of the correlation function for a wide range of separations. This is due to these data sets having a bimodal read density, with all the peaks with high number of reads located in one half of the data array (i.e. the 3' end of the genome). This bimodal data structure introduces a structure in the correlation function, with systematically positive values for s < l and negative values for s > l, where l is the extent of the region that concentrates all high amplitude peaks. We did not attempt to model this bimodal data structure when generating the confidence regions, which results in an overestimation of the significance of positive correlation values when these data sets were analyzed.
|
10.1371/journal.ppat.1007651 | The opportunistic pathogen Stenotrophomonas maltophilia utilizes a type IV secretion system for interbacterial killing | Bacterial type IV secretion systems (T4SS) are a highly diversified but evolutionarily related family of macromolecule transporters that can secrete proteins and DNA into the extracellular medium or into target cells. It was recently shown that a subtype of T4SS harboured by the plant pathogen Xanthomonas citri transfers toxins into target cells. Here, we show that a similar T4SS from the multi-drug-resistant opportunistic pathogen Stenotrophomonas maltophilia is proficient in killing competitor bacterial species. T4SS-dependent duelling between S. maltophilia and X. citri was observed by time-lapse fluorescence microscopy. A bioinformatic search of the S. maltophilia K279a genome for proteins containing a C-terminal domain conserved in X. citri T4SS effectors (XVIPCD) identified twelve putative effectors and their cognate immunity proteins. We selected a putative S. maltophilia effector with unknown function (Smlt3024) for further characterization and confirmed that it is indeed secreted in a T4SS-dependent manner. Expression of Smlt3024 in the periplasm of E. coli or its contact-dependent delivery via T4SS into E. coli by X. citri resulted in reduced growth rates, which could be counteracted by expression of its cognate inhibitor Smlt3025 in the target cell. Furthermore, expression of the VirD4 coupling protein of X. citri can restore the function of S. maltophilia ΔvirD4, demonstrating that effectors from one species can be recognized for transfer by T4SSs from another species. Interestingly, Smlt3024 is homologous to the N-terminal domain of large Ca2+-binding RTX proteins and the crystal structure of Smlt3025 revealed a topology similar to the iron-regulated protein FrpD from Neisseria meningitidis which has been shown to interact with the RTX protein FrpC. This work expands our current knowledge about the function of bacteria-killing T4SSs and increases the panel of effectors known to be involved in T4SS-mediated interbacterial competition, which possibly contribute to the establishment of S. maltophilia in clinical and environmental settings.
| Competition between microorganisms determines which species will dominate or be eradicated from a specific habitat. Bacteria use a series of mechanisms to kill or prevent multiplication of competitors. We show that an opportunistic pathogen, Stenotrophomonas maltophilia, harbours a type IV secretion system (T4SS) that works as a weapon to kill competitor bacterial species. We identified a series of new putative toxic T4SS effectors secreted by S. maltophilia and their cognate immunity proteins. Characterization of one S. maltophilia effector with unknown function (Smlt3024) shows that it reduces growth rate of E. coli cells. Its cognate immunity protein, Smlt3025, presents a structure similar to the FrpD lipoprotein from Neisseria meningitidis. Smlt3024 expressed in the plant pathogen Xanthomonas citri can be translocated into E. coli cells, highlighting the interchangeable characteristic of T4SSs toxins and the conservation of secretion system function. We show that X. citri and S. maltophilia can kill each other in a T4SS-dependent manner, most likely due to differences in their cohorts of effector-immunity protein pairs. This work expands our current knowledge about the function of bacteria-killing T4SSs and the bacterial arsenal fired by these systems during encounters with other species.
| The ecological interactions between bacterial species range from cooperative to competitive and can be mediated by diffusible soluble factors secreted into the extracellular medium or by factors transferred directly into target cells in a contact-dependent manner [1]. Several types of contact-dependent antagonistic interactions between bacteria have been described [1]. Contact-dependent growth inhibition (CDI) is mediated by the CdiA/CdiB family of two-partner secretion proteins in which the outer membrane protein CdiB is required for secretion of the CdiA toxin [2, 3]. The type VI secretion system (T6SS) is a dynamic contractile organelle evolutionarily related to bacteriophage tails, enabling the injection of proteinaceous effectors into target prokaryotic or eukaryotic cells [4, 5]. A specialized secretion system widely distributed among Gram-positive bacteria called Esx pathway or type VII secretion system (T7SS) induces contact-dependent cell death [6, 7]. An atypical bacteriocin system in Caulobacter crescentus called contact-dependent inhibition by glycine zipper proteins (Cdz) was also reported [8]. Another distinct contact-dependent toxin delivery mechanism is that of outer membrane exchange (OME) described in the social bacterium Myxococcus xanthus [9]. Contact-dependent antagonism has also been shown to be mediated via a specialized type IV secretion system (T4SS) that transports toxic effectors into target prokaryotic cells [10, 11].
T4SSs are a highly diverse superfamily of secretion systems found in many species of Gram-negative and Gram-positive bacteria. These systems mediate a wide range of events from transfer of DNA during bacterial conjugation to transfer of effector proteins into eukaryotic host cells [12] and into competitor bacteria [10]. T4SSs have been classified based on their physiological functions as (i) conjugation systems, (ii) effector translocators, or (iii) contact-independent DNA/protein exchange systems [13]. Another common classification scheme divides T4SSs into two phylogenetic families called types A and B [14, 15]; while more finely discriminating phylogenetic analyses based on two highly conserved T4SS ATPases (VirB4 and VirD4) identified eight distinct clades [16, 17].
The model type A VirB/D4 T4SS from Agrobacterium tumefaciens, which is used to transfer tumour-inducing effectors into some plant species [18], is composed of a core set of 12 proteins designated VirB1-VirB11 and VirD4. Electron microscopy studies on homologous systems from the conjugative plasmids R388 and pKM101 [19–21] have revealed an architecture that can be divided into two large subcomplexes: i) a periplasmatic core complex made up of 14 repeats of VirB7, VirB9 and VirB10 subunits that forms a pore in the outer membrane and which is also linked, via VirB10, to the inner membrane and ii) an inner membrane complex composed of VirB3, VirB6 and VirB8 and three ATPases (VirB4, VirB11 and VirD4) that energize the system during pilus formation and substrate transfer. Finally, VirB2 and VirB5 form the extracellular pilus and VirB1 is a periplasmic transglycosidase [22–24]. The X. citri T4SS involved in bacterial killing, and its homologues in other bacterial species (together called X-T4SSs for Xanthomonadales-like T4SSs), share many features with the type A T4SSs from A. tumefaciens and the conjugative T4SSs pKM101 and R388, with one distinctive feature being an uncharacteristically large VirB7 lipoprotein subunit [25] whose C-terminal N0 domain decorates the periphery of the outer membrane layer of the core complex [11, 26].
VirD4 and its orthologs play a key role by recognizing substrates on the cytoplasmic face of the inner membrane and directing them for secretion through the T4SS channel [14, 27–29]. A yeast two-hybrid screen using X. citri VirD4 as bait identified several prey proteins (initially termed XVIPs for Xanthomonas VirD4 interacting proteins) containing a conserved C-terminal domain named XVIPCD (XVIP conserved domain) [30]. These proteins were later shown to be putative antibacterial effectors secreted via the X. citri T4SS into target cells, often carrying N-terminal domains with enzymatic activities predicted to target structures in the cell envelope, including peptidoglycan-targeting glycohydrolases and proteases, phospholipases, as well as nucleases [10]. Furthermore, each T4SS effector is co-expressed with a cognate immunity protein, which is predicted to prevent self-intoxication [10], a feature also observed for effector-immunity pairs associated with T6SSs [31]. Bioinformatic analysis identified potential XVIPCD-containing proteins in many other bacterial species of the Xanthomonadales order, including Stenotrophomonas spp., Lysobacter spp., Luteimonas spp., Luteibacter spp. and Dyella spp. Therefore, these effectors and their cognate immunity proteins were generally designated X-Tfes and X-Tfis (Xanthomonadales T4SS effectors and immunity proteins, respectively) [10, 11].
Stenotrophomonas maltophilia is an emerging multi-drug-resistant global opportunistic pathogen. S. maltophilia strains are frequently isolated from water, soil and in association with plants [32], but in the last decades an increased number of hospital-acquired infections, particularly of immunocompromised patients, has called attention to this opportunistic pathogen [33, 34]. Infections associated with virulent strains of S. maltophilia are very diverse, ranging from respiratory and urinary tract infections to bacteremia and infections associated with intravenous cannulas and prosthetic devices [33]. The ability of Stenotrophomonas spp. to form biofilms on different biotic and abiotic surfaces [35, 36] and its capacity to secrete several hydrolytic enzymes (proteases, lipases, esterases) that promote cytotoxicity both contribute to pathogenesis [37, 38]. In addition, S. maltophilia is naturally competent to acquire foreign DNA, which probably contributes to the multi-drug-resistant phenotype of several strains [32, 39].
S. maltophilia strain K279a contains a cluster of genes (virB1-virB11 and virD4) on its chromosome coding for a T4SS homologous to the X-T4SS of the plant pathogen Xanthomonas citri involved in interbacterial antagonism [10], and their cytoplasmic ATPases VirD4 share 79% amino acid identity (Fig 1A). In this study, we show that S. maltophilia K279a is proficient in inducing the death of several other Gram-negative bacterial species in a T4SS-dependent manner. Interestingly, S. maltophilia and X. citri can duel using their T4SSs and kill each other. A bioinformatic search of the S. maltophilia K279a genome for proteins containing a C-terminal domain conserved in X. citri T4SS effectors (XVIPCD) identified twelve putative effectors. We selected a putative S. maltophilia effector with unknown function (Smlt3024) for further characterization and confirmed that it is indeed secreted in a contact- and T4SS-dependent manner. Heterologous expression of Smlt3024 in the periplasm of E. coli reduced growth rate, which could be counteracted by co-expression with its cognate immunity protein, Smlt3025. Using an X. citri strain that is deficient in target cell lysis due to the lack of nine X-Tfes but proficient in substrate delivery, we show that Smlt3024 can be translocated via the T4SS into target E. coli cells. Furthermore, heterologous expression of the X. citri VirD4 coupling protein in the S. maltophilia ΔvirD4 strain can restore T4SS function. These results highlight the conservation of X-T4SS function and the interchangeable usage of T4SSs effectors by different species. Interestingly, the crystal structure of Smlt3025 revealed a topology similar to the iron-regulated protein FrpD, the cognate binding partner of FrpC, an RTX protein of unknown function secreted by the type I secretion system (T1SS) of Neisseria meningitidis. This work expands our current knowledge about the mechanism of bacteria-killing T4SSs and the bacterial arsenal fired by these systems in encounters with other species.
The genome of S. maltophilia K279a [40] harbours two clusters of genes encoding distinct T4SSs: smlt2997-smlt3008 (annotated as virB) and smlt1283-smlt1293 (annotated as trb) [41]. Comparative sequence analysis showed that the S. maltophilia virB1-11 and virD4 genes are most closely related with their counterparts in the X. citri T4SS involved in bacteria killing (X-T4SS) (37% − 82% identity at the amino acid level), with the three ATPases that energize the system presenting the greatest levels of identity: VirB4 (81%), VirB11 (82%) and VirD4 (79%) (Fig 1A). Phylogenetic analysis based on the amino acid sequences of S. maltophilia VirD4/Smlt3008 grouped the S. maltophilia VirB/T4SS together with the X. citri X-T4SS involved in bacterial killing, while Stenotrophomonas Trb/T4SS, for which no functional information is available, belongs to another group of T4SSs (S1 Fig). The second T4SS from X. citri (encoded by plasmid pXAC64), which was proposed to be involved in conjugation due to neighbouring relaxosome genes and oriT site [30], is located in another branch of the phylogenetic tree, distinct from the two systems described above (S1 Fig).
To investigate the involvement of the S. maltophilia X-T4SS in bacterial antagonism, we created a mutant strain lacking the ATPase coupling protein VirD4 (ΔvirD4) and analysed its ability to restrict growth of other species such as E. coli. Different dilutions of an E. coli culture were mixed with a fixed number of S. maltophilia cells and the co-cultures were spotted onto LB-agar plates containing the chromogenic substrate X-gal and incubated for 24 h at 30°C (Fig 1B). As only E. coli cells naturally express β-galactosidase, they turn blue while S. maltophilia cells are yellow. Growth of E. coli was inhibited by S. maltophilia wild-type, but not by the ΔvirD4 strain (Fig 1B). The phenotype of S. maltophilia ΔvirD4 could be restored by complementing the strain with a plasmid encoding VirD4 (smlt3008) under the control of the PBAD promoter (ΔvirD4 virD4smlt) (Fig 1B). This plasmid promotes low expression levels sufficient for complementation under non-inducing conditions (no L-arabinose) in Stenotrophomonas. Interestingly, transformation of S. maltophilia ΔvirD4 strain with a plasmid encoding VirD4 from X. citri (xac2623) (ΔvirD4 virD4xac) also restored the phenotype (Fig 1B), indicating that the X. citri protein is able to couple substrates to the S. maltophilia translocation apparatus. The S. maltophilia T4SS-dependent antibacterial effect is only detected in co-cultures incubated on solid LB-agar surfaces where cell-cell contact is frequent and long-lasting; no effect on target cell growth is observed when S. maltophilia and E. coli are co-cultured in liquid media (Fig 1C).
To analyse whether the antagonism mediated by the S. maltophilia T4SS is due to target cell lysis, E. coli cells were mixed with different S. maltophilia strains (wild-type, ΔvirD4, ΔvirD4 virD4smlt and ΔvirD4 virD4xac) and spotted onto 96 well plates containing LB-agar with CPRG. CPRG is a cell-impermeable chromogenic substrate hydrolysed by β-galactosidase released from lysed E. coli, thus producing chlorophenol red with an absorbance maximum at 572 nm [26, 42]. Fig 1D shows that S. maltophilia wild-type and complemented strains (ΔvirD4 virD4smlt and ΔvirD4 virD4xac) induce lysis of E. coli with very similar efficiencies (based on the slopes of the curves) while the ΔvirD4 strain does not induce target cell lysis.
Single cell analysis by fluorescence microscopy of S. maltophilia co-incubated with E. coli expressing red fluorescent protein (E. coli-RFP) further confirms that Stenotrophomonas induces target cell lysis in a contact-dependent manner (Fig 1E and S1 Movie). No cell lysis was detected when E. coli was co-incubated with S. maltophilia ΔvirD4 (Fig 1E and S2 Movie). Quantification of E. coli cell lysis over a timeframe of 100 min shows that approximately 50% of E. coli cells in contact with wild-type Stenotrophomonas lysed during this period, while no E. coli cell lysis was detected when mixed with S. maltophilia ΔvirD4 (Fig 1F). It is important to note that during the time frame of these experiments some E. coli cells may be intoxicated without cellular lysis since the time of target-cell lysis may vary after the initial physical contact. Therefore, the quantification presented in Fig 1F most likely sub-estimates the efficiency of the T4SS mediated antagonistic effect. In addition to E. coli, we observed that S. maltophilia is able to kill other Gram-negative bacterial species such as Klebsiella pneumoniae, Salmonella Typhi and Pseudomonas aeruginosa in a T4SS-dependent manner (Fig 2A, S3, S4 and S5 Movies) while no killing was observed using the S. maltophilia ΔvirD4 strain (S15, S16 and S17 Movies). These results are consistent with our previous work in which we demonstrated that X. citri displays an antagonistic effect towards not only E. coli, but also Chromobacterium violaceum (Betaproteobacteria) in a T4SS-dependent manner [10].
As X. citri is, to date, the only other bacterial species experimentally shown to use a T4SS for interbacterial killing, we decided to analyse whether S. maltophilia and X. citri could use their T4SSs to compete with and kill each other. First, we co-incubated S. maltophilia (either wild-type or ΔvirD4) with an X. citri T4SS mutant in which all of the chromosomal virB genes were substituted with the gene for green fluorescent protein (GFP) under the control of the endogenous virB7 promoter (ΔvirB-GFP) [43] and confirmed that S. maltophilia induces lysis of X. citri ΔvirB-GFP in a T4SS-dependent manner (Fig 2B and 2D; S6 and S7 Movies). Next, we co-incubated X. citri-GFP (carrying a functional T4SS) with S. maltophilia wild-type or ΔvirD4 strains. Besides showing that X. citri can induce lysis of S. maltophilia ΔvirD4 (S8 Movie), we observed that when both wild-type species are mixed, they duel and kill each other in a T4SS-dependent manner (Fig 2C and 2E; S9 Movie). S. maltophilia seems to be slightly more effective in killing X. citri via its T4SS, which could be due to differences in the efficiencies of the systems, differences in their repertoires of effectors (see below) and/or the shorter doubling time of S. maltophilia compared to X. citri under the conditions tested.
After confirming that the S. maltophilia X-T4SS is functional and induces target cell death, we decided to search for the effector proteins translocated by this system that were mediating the phenotype. As the VirD4 coupling protein of X. citri complements the ΔvirD4 strain of S. maltophilia (Fig 1B and 1D), we hypothesized that potential substrates secreted via the T4SS of S. maltophilia could be identified by applying a bioinformatic approach using the conserved C-terminal domains of X. citri X-Tfes (XVIPCD) that interact with VirD4 to search the genome of S. maltophilia K279a. Using this approach, we identified twelve S. maltophilia proteins as potential T4SS substrates (X-Tfes) (Fig 3A, S1 Table). Amino acid sequence alignment of C-terminal XVIPCDs from Stenotrophomonas X-Tfes revealed a series of conserved amino acid motifs that are also present in X. citri X-Tfes (Fig 3B) [30], highlighting putative key residues required for VirD4 recognition and secretion by these X-T4SSs.
All identified S. maltophilia effectors are organized in small operons together with an upstream gene encoding a conserved hypothetical protein, reminiscent of the organization of effectors with their immunity proteins [10, 44]. Six of the identified S. maltophilia T4SS substrates harbour domains already described in other bacterial toxins such as lipases, nucleases, lysozyme-like hydrolases and proteins with peptidoglycan binding domains (Fig 3A). Three of these effectors (smlt2990, smlt2992 and smlt3024) are encoded by genes close to the S. maltophilia virB structural locus (genes smlt2997 to smlt3008), further illustrating the link of these effectors with the T4SS. It is interesting to note that six of the identified putative Stenotrophomonas T4SS effectors do not display any known protein domain that could indicate the mechanism mediating antibacterial activity (smlt0113, smlt0332, smlt0500, smlt0502, smlt0505, smlt3024) (Fig 3A). To validate our bioinformatic results and obtain further insight regarding the function of the effectors with unknown function, we selected the products of the smlt3024 gene and its upstream putatively co-transcribed cognate immunity protein (smlt3025) for further characterization.
In its genomic context, smlt3024 seems to be organized in an operon downstream of two genes encoding for its putative cognate immunity protein (smlt3025) and another small protein containing a helix-turn-helix (HTH) domain annotated as a putative transcriptional regulator (smlt3026) (Fig 4A). This operon, along with the putative operons coding for the effector/immunity pairs smlt2990/smlt2989 and smlt2992/smlt2993, is in close proximity to the locus coding the X-T4SS structural genes (smlt2997-smlt3008, Fig 1A).
To determine whether Smlt3024 is indeed an effector secreted via the S. maltophilia T4SS, we cloned an N-terminal FLAG-tagged version of smlt3024 (FLAG-Smlt3024) into the pBRA plasmid under the control of the PBAD promoter and used it to transform both S. maltophilia wild-type and ΔvirD4 strains. These strains were co-incubated with E. coli and spotted onto nitrocellulose membranes placed over LB-agar plates containing 0.1% L-arabinose and incubated for 6 h at 30°C. The membranes were later processed for immunodetection with an anti-FLAG antibody. Results show an increase in signal intensity for FLAG-Smlt3024 when S. maltophilia was co-incubated with E. coli (Fig 4B and 4C), while no increase was detected when S. maltophilia ΔvirD4 was co-incubated with E. coli (Fig 4B and 4C). In addition, no increase in signal intensity could be detected when S. maltophilia FLAG-Smlt3024 was incubated without target E. coli cells (Fig 4B). SDS-PAGE of total protein extracts followed by western blot with anti-FLAG antibody showed that both S. maltophilia wild-type and ΔvirD4 strains were expressing similar levels of FLAG-Smlt3024 (S2 Fig). These results indicate that translocation of Smlt3024 is dependent on a functional T4SS and on contact with a target cell from another species. We interpret the anti-FLAG signal detected by western blot as due to E. coli cell lysis caused by the delivery of FLAG-Smlt3024 along with the full cocktail of S. maltophilia X-Tfes via the T4SS into the target E. coli cells. After target cell lysis, the released FLAG-Smlt3024 binds to the nitrocellulose membrane; hence the assay is an indirect measurement of protein translocation. Although we do not have direct experimental visualization of X-Tfe delivery into target cells, we note that all except for a few T4SSs described to date transfer macromolecules across the bacterial cell envelope directly into the target cell [45–47], so we hypothesize that X-T4SS toxic effectors are translocated directly into the target cell.
If Smlt3024 is indeed a toxic effector translocated by the S. maltophilia T4SS, then we would expect that its expression in the appropriate compartment within E. coli would cause an impairment of bacterial growth. To evaluate the toxicity of Smlt3024 upon expression in E. coli and to establish in which cellular compartment Smlt3024 exerts its effect, we cloned the full-length smlt3024 gene into the pBRA vector placing it under control of the PBAD promoter (inducible by L-arabinose and repressed by D-glucose) both with and without an N-terminal PelB periplasmic localization signal sequence. We also cloned the sequence of the putative Smlt3025 immunity protein into the pEXT22 vector placing it under the control of the PTAC promoter, which can be induced by IPTG. We noted that the published annotated sequence for Smlt3025 [40] has a non-canonical GTG start codon with 4 possible in frame ATG start codons at positions 13, 45, 47 and 50 and that initiation at positions 45, 47 or 50 is predicted to produce proteins with an N-terminal signal sequence lipobox for periplasmic localization as a lipoprotein (Fig 4D) [48]. Therefore, three versions of Smlt3025 were cloned into pEXT22, leading to the production of Smlt30251-333, Smlt302513-333 and Smlt302545-333. E. coli strains carrying the different combinations of pBRA-Smlt3024 and each one of the pEXT22-Smlt3025 plasmids were serial diluted and incubated on LB-agar plates containing either D-glucose, L-arabinose or L-arabinose plus IPTG (D-glucose inhibits and L-arabinose induces expression of Smlt3024; IPTG induces expression of Smlt3025). Results showed that Smlt3024 is toxic when directed to the periplasm of E. coli cells (pBRA-pelB-smlt3024) but not in the cytoplasm (pBRA-smlt3024), and that only Smlt302545-333, which is predicted by the SignalP 5.0 algorithm to be directed to the periplasm [49], could neutralize Smlt3024 toxicity (Fig 4E). These results support the hypothesis that Smlt3025 was mistakenly annotated and that the correct start codon is Met45, Met47 or Met50. Bioinformatic analysis of the closest 100 homologues of Smlt3025 in the non-redundant protein database, shows that most proteins are annotated with initiation codons that align with Met47 of Smlt3025 (S3A Fig). Similar results are obtained when more distantly related Smlt3025 homologues from the KEGG database [50] are aligned (S3B Fig).
To gain some information about the inhibitory mechanism of Smlt3025, we decided to analyse whether this protein could interact directly with Smlt3024 by expressing and purifying full-length Smlt3024 and a soluble version of Smlt3025 (amino acid residues 86–333) lacking its predicted N-terminal signal peptide. Complex formation was analysed using size exclusion chromatography coupled to multiple-angle light scattering (SEC-MALS) (Fig 4F). The MALS analysis calculated average masses for Smlt3024 and Smlt302586-333 of 52.3 kDa and 27.5 kDa, respectively, which are very close to the theoretical values of their monomer molecular masses of 49 kDa and 28 kDa, respectively (Fig 4F). When a mixture of these proteins was analysed by SEC-MALS followed by SDS-PAGE, a new peak was observed containing both Smlt3024 and Smlt302586-333 with an estimated molecular mass calculated by MALS of 74.2 kDa, showing that a stable 1:1 complex (theoretical mass of 77 kDa) was formed between Smlt3024 and Smlt302586-333 (Fig 4F).
To gather further insight on the mechanism by which Smlt3024 could induce toxicity, we decided to perform time-lapse microscopy to evaluate growth and morphology of individual E. coli cells carrying the empty pBRA or pBRA-pelB-smlt3024 plasmids. E. coli carrying the empty plasmid incubated on LB-agar with 0.2% L-arabinose (Fig 4G and S10 Movie) as well as the repressed pBRA-pelB-smlt3024 (0.2% D-glucose) grew normally (Fig 4G and S11 Movie). Upon induction with L-arabinose, cells carrying pBRA-pelB-smlt3024 quickly experienced a strong reduction in growth rate and single cells were smaller (average length of 2.1 ± 0.7 μm after 300 min) compared to the controls incubated in glucose (average length of 3.6 ± 1.2 μm after 300 min) (Fig 4G and S12 Movie). Despite the severe delay in growth rate, E. coli cells expressing PelB-Smlt3024 remained viable and continued growing and dividing for up to 8 h (S12 Movie).
In order to confirm that Smlt3024 produces the same phenotype when delivered by a bona fide X-T4SS into a target cell, we employed an X. citri strain (Δ8Δ2609-GFP) that has an intact and functional X-T4SS but is deficient in inducing target cell lysis due to the sequential deletion of nine X-Tfes genes (see Materials and methods). This strain allows phenotypic analysis of individual effectors without the interference of other lytic toxins. As the structural genes of X. citri and S. maltophilia T4SSs are very similar (Fig 1A) and expression of the VirD4 coupling protein of X. citri can restore the function of S. maltophilia ΔvirD4 (Fig 1B and 1D), we reasoned that X. citri Δ8Δ2609-GFP could be used to deliver S. maltophilia effectors. X. citri Δ8Δ2609-GFP was transformed with pBRA plasmid carrying the operon coding for Smlt3025 (starting from Met45) and Smlt3024. Time-lapse microscopy analysis of X. citri Δ8Δ2609-GFP and E. coli co-cultures grown on agar pads allowed us to measure the doubling times of E. coli cells (Fig 5, S13 and S14 Movies). The average doubling time of E. coli cells that were not in contact with X. citri Δ8Δ2609-GFP or were in contact with X. citri Δ8Δ2609-GFP carrying empty plasmid was 77 ± 23 and 92 ± 66 min, respectively (Fig 5C). However, the E. coli doubling time increased to 173 ± 71 min when in contact with X. citri Δ8Δ2609-GFP expressing Smlt3024 (Fig 5A and 5C). This growth inhibition effect could be reverted by expressing the immunity protein Smlt302545-333 in target E. coli cells in which doubling times were restored to 78 ± 65 min (Fig 5B and 5C). These results confirm the inhibitory effect of Smlt3024 on cell growth upon translocation via a bona fide X-T4SS into target cells. Furthermore, these results demonstrate that X-Tfes from one species can be recognized for transfer by T4SSs from another species, thus highlighting the conservation of X-Tfe secretion signal recognition and X-T4SS function in Stenotrophomonas and Xanthomonas species.
To obtain some insight into the possible contribution of Smlt3024 to T4SS-dependent antagonism, we searched for homologues similar to its amino acid sequence (residues 1–308, excluding the C-terminal XVIPCD) using the PSI-BLAST algorithm [51] against the non-redundant protein sequence database. Three iterations of PSI-BLAST retrieved 815 hits (cutoff e-values < e-6). The first 402 hits with the highest scores are from shorter proteins of unknown function (less than 600 amino acids), which are about the same size of Smlt3024 (440 residues). The PSI-BLAST search also returned 221 hits with lower scores (e-values between e-43 and e-7) from larger proteins (greater than 750 amino acids in length) derived from a wide variety of bacterial genera including Yersinia, Ralstonia, Pseudomonas, Cupriavidus, Snodgrassella, Xanthomonas, Pseudoxanthomonas, Leisingera, Thalassospira, Nitrosomonas, Halocynthiibacter, Vibrio, Neisseria, Thioalkalivibrio, Stenotrophomonas, Rhizobium, Robbsia, Devosia, Sphingomonas, Paraburkholderia, Sphingomonas and Acinetobacter. This group of 221 proteins (S4 Fig) share the following characteristics: i) all except for one align with Smlt3024 via their N-terminal regions (within the first 300 amino acids) and ii) all but six have multiple Repeat in ToXin (RTX) calcium-binding nonapeptide motifs (Pfam: PF00353) [52] or carry hemolysin-type calcium binding protein related domains (Pfam: PF06594). Some also have additional C-terminal domains such as peptidase S8, subtilisin-like, pro-protein convertase P, cadherin-like and IgG-like domains. An analogous search using the JACKHMMER algorithm [53] against the rp75 database produced similar results (S2 Table). Thus, both PSI-BLAST and JACKHMMER searches indicate that Smlt3024 is similar to the N-terminal domain of unknown function often found in larger proteins with downstream Ca2+-binding RTX motifs. One notable exception to the above pattern is the alignment of Smlt3024 with the C-terminal domain of a type VI secretion system tip protein VgrG from Sphingomonas jatrophae strain S5-249 (accession number WP_093316205.1), whose possible significance will be considered in the Discussion.
In order to obtain more information regarding the mechanism of the effector/immunity pair Smlt3024/Smlt3025 we tried to crystalize these proteins to solve their structures by X-ray crystallography. We successfully crystallized a soluble fragment of Smlt3025, corresponding to residues 86–333. Crystals belonged to space group R3, some of which diffracted to around 2 Å resolution. Initial phases were estimated by single wavelength anomalous dispersion using a crystal soaked in sodium iodide and the final model was obtained using data collected from a native crystal (Table 1). The Smlt3025 structure (PDB 6PDK) is organized around a central 8-stranded anti-parallel β sheet (β5-β6-β14-β13-β10-β9-β8-β7). The intervening loops between these β-strands contain α-helices (α1 and α2), 310 helices (η2, η3, η4, η5 and η6) and a small beta-hairpin (β11-β12). The central β-sheet is preceded by two β-hairpins (β1-β2, β3-β4) and a 310 helix (η1) and is followed by a C-terminal helix (η7 and α3; Fig 6A and 6B). An analysis of Smlt3025 homologues using the Consurf algorithm (S5 Fig) identified conserved positions which, once mapped onto the Smlt3025 structure, cluster into the hydrophobic core of the central β-sheet and to the N-terminal β-hairpins (Fig 6C).
Smlt3025 has no significant amino acid sequence similarity with proteins with known 3D structure. Structure-based similarity searches using the DALI algorithm [54] identified a single protein with a Z-score of 9.3, named iron-regulated protein D (FrpD) from Neisseria meningitidis (PDB 5EDF). FrpD is a lipoprotein associated with the N. meningitidis outer membrane that strongly interacts with the N-terminal domain of iron-regulated protein C (FrpC), a 1829 residue protein secreted into the extracellular milieu via a T1SS [55, 56]. FrpC belongs to the RTX protein family, with 43 C-terminal RTX motifs [57], an architecture very similar to most of the 221 proteins identified as Smlt3024 homologues in S4 Fig.
Fig 6D presents a structural alignment between Smlt302586-333 and FrpD. The topologies of the central β-sheets of the two proteins are identical. However, the loops connecting the β-strands have significant differences, for example α2 and the β11-β12 hairpin in Smlt3025 are absent in FrpD. Previous NMR chemical shift perturbation studies identified the surface-exposed portions of the N-terminal β strands (preceding the central β sheet), the C-terminal portion of the last α helix and the unstructured C-terminal tail of FrpD as the probable binding site for FrpC [55, 56]. Although the corresponding surface of Smlt3025 has significantly different structural features at the N-terminus, due to a different relative orientation of its β1-β2 hairpin, the C-terminal α helices and the β3-β4 hairpins of the two proteins superpose well (Fig 6D) and the N-terminal β hairpins are amongst the most well conserved sequences in Smlt3025 homologues (Fig 6C and S5 Fig). These observations raise the hypothesis that Smlt3025 could interact with Smlt3024 in a manner analogous to the FrpD-FrpC interaction.
Competition between microorganisms for nutrients and space often determines which species will thrive and dominate or be eradicated from a specific habitat. S. maltophilia is often found as a member of microbial communities in water, soil and in association with plants. Some Stenotrophomonas species like S. rhizophila can participate in beneficial interactions with plants, but no species were reported to be phytopathogenic, which distinguishes Stenotrophomonas from the phylogenetically related genera Xanthomonas and Xylella [32]. More importantly, an increasing number of hospital-acquired S. maltophilia infections over the last decades has led to the classification of this bacterium as an emerging opportunistic pathogen [33, 34]. Key to the opportunistic behaviour of S. maltophilia strains are their ability to form biofilms and their resistance to multiple antibiotics. In this manuscript, we show that the X-T4SS of S. maltophilia is involved in interbacterial competition, allowing S. maltophilia to induce lysis of several Gram-negative species. The antibacterial property conferred by the X-T4SS probably provides a competitive advantage to S. maltophilia in polymicrobial communities, contributing to increased fitness. S. maltophilia is frequently associated with cystic fibrosis patients [58, 59] and may need to compete with oral and nasal microbiota during infection of susceptible organisms [60, 61]. Our competition experiments showed that S. maltophilia can kill two pathogens that colonize the respiratory tract of susceptible hosts, K. pneumoniae and P. aeruginosa; hence the contribution of S. maltophilia T4SS to colonization and maintenance during polymicrobial infections within mammalian hosts merits further investigation.
The most worrying aspect of pathogenic S. maltophilia strains is their multi-drug resistance phenotype [62]. As S. maltophilia is naturally competent to acquire foreign DNA [32, 39], the T4SS described here could, by inducing target cell lysis and increasing the availability of foreign DNA, be a positive factor in promoting Stenotrophomonas transformation, thus leading to the acquisition of antibiotic resistance genes by horizontal gene transfer. A similar mechanism has already been reported in Vibrio cholerae, which uses a bacterial killing T6SS as a predatory device to induce target cell lysis concomitantly with the uptake of target-cell DNA [63].
The S. maltophilia X-T4SS is homologous to the X. citri X-T4SS and complementation of S. maltophilia ΔvirD4 with the X. citri VirD4 coupling protein restored its full capacity to lyse E. coli target cells. The VirD4 coupling protein interacts with the conserved C-terminal domain (XVIPCD) of X-Tfes described in X. citri [10, 30]; thus it was reasonable to use these conserved regions to search the genome of S. maltophilia for new T4SS effectors. The search rationale proved to be efficient and we identified 12 new putative S. maltophilia T4SS effectors and provided experimental evidence that at least one of them (Smlt3024) is secreted in a T4SS-dependent manner. Due to the conservation of the amino acid sequence of XVIPCD of S. maltophilia T4SS effectors, it is likely that the other 11 putative effectors are also secreted via the T4SS. Translocation of Smlt3024 from the killing-deficient strain X. citri Δ8Δ2609-GFP into target E. coli cells also illustrates the conserved function of both T4SSs systems and confirms the ability of X. citri VirD4 coupling protein to recognize and translocate S. maltophilia effectors by means of their conserved XVIPCDs. Furthermore, recognition and translocation of S. maltophilia effectors by the X. citri T4SS machinery suggests that toxic effectors containing an XVIPCD could be easily exchanged between species in the environment by horizontal gene transfer of effector/immunity protein pairs.
Among the twelve S. maltophilia effector/immunity protein (X-Tfe/X-Tfi) pairs, we believe that special attention should be given to effectors with no recognizable domain annotated in Pfam database—six effectors including Smlt3024. Detailed biochemical and structural characterization of these new effectors could identify new toxic domains and might reveal interesting mechanisms impairing bacteria proliferation, contributing to the design of novel and effective antibacterial drugs. Most of the characterized T4SS and T6SS antibacterial toxins are enzymes that degrade structural cellular components such as peptidoglycan and phospholipids, thus promoting target cell lysis [64]. Recent studies have identified effectors that change cell metabolism, promoting altered cell growth rather than lysis, but these effectors act in the target cell cytoplasm [65, 66]. In this context, the mechanism underlying the apparent periplasmic toxicity induced by Smlt3024, which reduces target E. coli cell growth rate either by ectopic expression or after translocation by X. citri T4SS, is likely to be a mechanism not yet described.
According to our bioinformatic analyses, Smlt3024 presents homology with the N-terminal region of proteins that contain multiple RTX motifs (annotated as RTX toxins or hemolysin-type calcium binding proteins). However, no functional information is available for these N-terminal regions. The crystal structure of Smlt3025 revealed a topology similar to FrpD from N. meningitidis, which is a lipoprotein [55] that is known to bind the N-terminal region of FrpC, an 1829 residue protein that contains 43 RTX repeats between residues 879 and 1705 [57]. Upon secretion by the T1SS, FrpC undergoes Ca2+-dependent trans-splicing via autocatalytic cleavage between Asp414 and Pro415 to form an Asp414-Lys isopeptide bond, which results in covalent linkage of the FrpC1-414 fragment to plasma membrane proteins of epithelial cells in vitro [56]. FrpC was originally proposed to play a role during infection of mammalian hosts; however, subsequent studies analyzing FrpC cytotoxicity towards macrophages in vitro and infection of mammalian hosts with mutant strains failed to detect any cytotoxic effect or virulence attenuation [67]. Considering these findings, we hypothesize that FrpC may in fact be an N. meningitidis T1SS antibacterial effector and FrpD its cognate immunity protein.
The mechanism by which Smlt3024 causes reduction of growth speed after heterologous expression or T4SS-mediated translocation into the periplasm of the target cell is still unknown. Based on the similarity with the N-terminus of RTX proteins, we speculate that Smlt3024 could bind to and inhibit the function of one or more key metabolic or signal transduction components in the periplasm, thus promoting target cell stasis. Inducing target cell stasis could be sufficient in natural scenarios to provide the attacker with a competitive advantage, allowing it to outnumber the target species and establish itself in the environment. It is worth mentioning that Smlt3024 is homologous to Smlt0500, another S. maltophilia X-Tfe (48% identity over the first 308 residues), as are their cognate X-Tfis, Smlt3025 and Smlt0501 (41% identity; both predicted to be lipoproteins). Therefore, Smlt3024 and Smlt0500 could exert their functions via similar mechanisms and it is possible that their combined action could be more detrimental.
In natural settings, many species are likely to have acquired resistance mechanisms against some effectors by means of immunity proteins. Thus, by employing a cocktail of diversified effectors, species deploying an X-T4SS can gain an advantage over competitors. The importance of employing diversified effector-immunity pairs is illustrated by the duelling observed between S. maltophilia and X. citri: these species can kill one another in a T4SS-dependent manner, indicating that each lack immunity proteins against at least a subset of the rival´s set of T4SS effectors. Both S. maltophilia K279a and X. citri 306 carry twelve putative X-Tfe/X-Tfi pairs, but only six of the X-Tfis have homologues with 26–58% identity over segments that vary in size from 99 residues to 265 residues (S3 Table). Hence, these two bacteria could potentially be protected against some homologous cognate X-Tfes from the rival species. However, S. maltophilia is probably susceptible to the action of X. citri X-Tfes XAC4264 (unknown function), XAC2885 (putative fosfolipase), XAC2609 (peptidoglycan hydrolase), XAC1918 (putative peptidoglycan hydrolase), XAC0096 (putative HExxH metallopeptidase) and XAC0151 (unknown function) [10]. Likewise, X. citri can be expected to be susceptible to the action of the S. maltophilia X-Tfes for which it apparently has no immunity proteins: Smlt3024, Smlt0505, Smlt0502, Smlt0500, Smlt0332 and Smlt0273, all of unknown function (Fig 3A).
The above considerations stress the importance of our observations showing that X-Tfes from one organism can be employed by the X-T4SS from another. Therefore, the acquisition by horizontal gene transfer of genes encoding X-Tfe/X-Tfi pairs could be relevant in determining the outcome of encounters between environmental bacteria from the Xanthomonadales order. In addition to Smlt3024 similarity to the N-terminus of a large number of RTX proteins that are often secreted via a type I secretion system [52], one interesting exception is its similarity with the C-terminal region of VgrG from Sphingomonas jatrophae (S2 Table and S4 Fig). VgrG is a secreted component of T6SSs that either interacts with toxic effectors to promote their secretion or itself carries a toxic domain at its C-terminal region [68]. An analogous observation has been made for the S. maltophilia X-Tfe Smlt0332, which is homologous to the C-terminal domains of several VgrG proteins [11]. These observations illustrate the dynamic exchange of effector/toxin domains, not just between bacteria employing similar secretion systems but also their recombination with diverse recognition motifs employed by evolutionarily distinct secretion systems.
This work expands our current knowledge about the function of bacteria-killing T4SSs by increasing the panel of effectors known to be involved in X-T4SS-mediated interbacterial competition and by highlighting the possibility of interspecies exchangeability of X-Tfes, which is dependent on XVIPCD recognition by the VirD4 coupling protein. In addition, the study adds information about the mechanisms S. maltophilia has at its disposal to compete with other species, possibly contributing to its establishment in both clinical and environmental settings.
S. maltophilia K279a [40] and X. citri pv. citri 306 [69] were grown in 2x YT media (16 g/L tryptone, 10 g/L yeast extract, 5 g/L NaCl). E. coli strain K-12 subsp. MG1655 [70] was used in competition assays because of its endogenous expression of β-galactosidase. K. pneumoniae, S. Typhi (ATCC 19430) and P. aeruginosa (PA14) were used for competition experiments. E. coli DH5α and E. coli HST08 were used for cloning purposes and E. coli S17 was used for conjugation with S. maltophilia. The X. citri ΔvirB-GFP strain lacks all chromosomal virB genes and has the msfGFP gene under the control of the endogenous virB7 promoter, while the X. citri-GFP strain has a functional T4SS and expresses GFP as a transcriptional fusion under the control of the virB7 promoter [43]. For time-lapse imaging of S. maltophilia and X. citri strains, AB defined media was used (0.2% (NH4)2SO4, 0.6% Na2HPO4, 0.3% KH2PO4, 0.3% NaCl, 0.1 mM CaCl2, 1 mM MgCl2, 3 μM FeCl3) supplemented with 0.2% sucrose, 0.2% casamino acids, 10 μg/mL thiamine and 25 μg/mL uracil. Cultures of E. coli and S. maltophilia were grown at 37°C with agitation (200 rpm) and X. citri cultures were grown at 28°C with agitation (200 rpm). Antibiotics were used at the following concentrations to select S. maltophilia strains: tetracycline 40 μg/mL and streptomycin 150 μg/mL. For selection of E. coli strains, kanamycin 50 μg/ml and spectinomycin 100 μg/ml were used when appropriate. For induction from the PBAD promoter, 0.2% L-arabinose was added. For PTAC induction, 200 μM IPTG was used. Expression from both promoters was repressed using 0.2% D-glucose.
All primers and plasmids used for cloning are listed in S4 Table. To produce in-frame deletions of virD4 (smlt3008) in S. maltophilia, we used a two-step integration/excision exchange process and the pEX18Tc vector [71]. Fragments of ~1000-bp homologous to the upstream and downstream regions of smlt3008 were amplified by PCR and cloned into pEX18Tc using standard restriction digestion and ligation. The pEX18Tc-ΔvirD4 was transformed in E. coli S17 donor cells by electroporation and transferred to S. maltophilia recipients via conjugation following the protocol described by Welker et al. [72]. Tetracycline-resistant colonies were first selected. Colonies were then grown in 2x YT without antibiotic and plated on 2x YT agar containing 10% sucrose without antibiotic. Mutant clones were confirmed by PCR. To complement the ΔvirD4 strain, the gene encoding full-length smlt3008 was PCR amplified from genomic DNA and cloned into the pBRA vector, which is a pBAD24-derived vector that promotes low constitutive expression in Stenotrophomonas and Xanthomonas under non-inducing conditions. The pBRA construct encoding full-length X. citri virD4/XAC2623 was reported previously [10]. For indirect secretion/translocation assays, the full-length sequence of smlt3024 was cloned into pBRA vector, including a FLAG tag at its N-terminus and transformed into S. maltophilia wild-type and ΔvirD4. Plasmids were transformed into S. maltophilia by electroporation (2.5 kV, 200 Ω, 25 μF, 0.2 cm cuvettes), followed by streptomycin selection. For cloning smlt3024 and smlt3025 into pSUMO–a modified version of pET28a (Novagen), with a SUMO tag between the hexahistidine and the cloning site–we used the soluble portion of Smlt3025 (residues between 86–333) that lacks the N-terminal signal peptide and the full-length Smlt3024. Smlt302586-333 was also cloned into pET28a in order to express the protein with an N-terminal 6xHis tag that was subsequently crystallized (see below). To produce smlt3024 with the pelB periplasmic localization sequence, PCR products were first cloned in pET22b (Novagen; containing the N-terminal pelB sequence). The pelB-smlt3024 construct was subsequently transferred to pBRA using Gibson assembly. For the immunity protein smlt3025, three different constructs were cloned in pEXT22 [73]: one starting at the annotated GTG start-codon and two starting at two downstream ATG codons (Met13 and Met45). The sequences of all constructs containing effectors in pBRA and immunity proteins in pEXT22 were confirmed by DNA sequencing to assure absence of point mutations in the cloned genes and upstream promoter sites using the Macrogen standard sequencing service (https://dna.macrogen.com/). The X. citri Δ8Δ2609-GFP strain was constructed by sequentially deleting the genes coding for X-Tfe/X-Tfi pairs (except for the XAC2610 X-Tfi) from the X. citri genome [10, 11, 30] using the two-step allelic exchange procedure described above (Oka et al., in preparation). This strain has a total of nine deletions which were introduced in the following order: 1) ΔXAC2885/XAC2884; 2) ΔXAC0574/XAC0573; 3) ΔXAC0097/XAC0096; 4) ΔXAC3364/XAC3363; 5) ΔXAC1918/XAC1917; 6) ΔXAC0467/XAC0466; 7) ΔXAC4264/XAC4263/XAC0462; 8) ΔXAC2609::msfGFP; 9) ΔXAC3266/XAC3267. For the 8th deletion, the xac2609 gene was replaced with the msfGFP gene, which allows the strain to be distinguished from target cells during time-lapse fluorescence microscopy.
Bacterial competition was assessed either by analysing target cell growth or target cell lysis. To analyse E. coli growth during co-incubation with S. maltophilia we used a protocol adapted from Hachani et al. [74]. Briefly, strains were subcultured (1:100 dilution) and grown to exponential phase for 2 h at 37°C (200 rpm). Cells were washed with 2x YT, the optical density measured at 600 nm (OD600nm) and adjusted to 1. Serial dilutions (1:4) of E. coli culture was performed in 96 well plates. Equal volumes of E. coli and S. maltophilia cultures at OD600nm 1.0 were mixed into each well. After mixing, 5 μl were spotted onto LB-agar plates containing 100 μM IPTG (isopropyl β-D-1-thiogalactopyranoside) and 40 μg/mL X-gal (5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside) using multichannel pipettes. Plates were incubated for 24 h at 30°C. Competitions in solid and liquid media were performed as described previously [75]. Analysis of target cell death was performed using CPRG (chlorophenol red-β-D-galactopyranoside) as described previously with minor modifications [26, 42]. Briefly, S. maltophilia and E. coli overnight cultures were subcultured by 1:100 dilution and grown at 37°C (200 rpm) to reach OD600nm of approximately 1 (E. coli cultures contained 200 μM IPTG). Cells were washed with LB media, OD600nm adjusted to 1.0 for S. maltophilia strains and OD600nm adjusted to 8.0 for E. coli. The adjusted cultures were mixed 1:1 and 10 μL spotted in triplicate onto 96 well plates containing 100 μL of semi-solid 1.5% 2x YT agar and 40 μg/mL CPRG. Plates were let dry completely, covered with adhesive seals and analysed on a SpectraMax Microplate Reader (Molecular Devices) at 572 nm every 10 min for 3.5 h. E. coli cultures were also spotted onto the same plate as a control for spontaneous cell death. The obtained A572 data was processed using RStudio (www.rstudio.com) and plotted using the ggplot2 package [76]. Background intensities obtained from the mean A572 values containing only E. coli cells were subtracted from all data series. The initial A572 value at time-point 0 min was subtracted from all subsequent time-points to correct for small differences in initial measurements. Finally, the E. coli lysis curves of S. maltophilia ΔvirD4 and complementation strains were normalized with respect to those obtained for the S. maltophilia wild-type strain.
For time-lapse imaging of bacterial killing at the single-cell level, agar slabs containing either 2x YT or supplemented AB media were created by cutting a rectangular frame out of a double-sided adhesive tape (3M VHB transparent, 24 mm wide, 1 mm thick), which was taped onto a first microscopy slide. Into the resulting tray, agar was poured and covered by a second microscopy glass slide to create a smooth surface. After solidification, the second microscopy slide was removed, exposing the agar’s surface onto which 2 μl of cell suspensions were spotted. After cell suspensions were left to dry completely, a #1.5 cover glass (Corning) was laid on top of the agar slab and closed at the sides by the second adhesive layer of the tape, leaving the cell mixtures closely and stably pressed between cover glass and the agar slab. Soon after, phase contrast images together with GFP or RFP excitation images were obtained with a Leica DMI-8+ epi-fluorescent microscope equipped with a Leica DFC365 FX camera, a HC PL APO 100x/1.4 Oil ph3 objective (Leica), a GFP excitation-emission band-pass filter cube (Ex: 470/40, DC: 495, EM: 525/50; Leica) and a Cy3/Rhodamine excitation-emission band-pass filter cube (Ex: 541/51, DC: 560, EM: 565/605; Leica). An incubation cage around the microscope kept temperatures constant at 37°C for E. coli and S. maltophilia experiments and at 28°C for experiments with X. citri. Several separate positions of each cell mixture were imaged every 10–15 min after auto-focusing using the LASX software package (Leica). Images were further processed with the FIJI software using the Bio-Formats plugin [77]. Time-lapse images were visually scored for cell lysis events. Small groups of cells (approximately 2 to 8 cells per colony) containing a mixture of bacterial species in close contact with each other were tagged at time-point zero and followed during 100 min (E. coli vs S. maltophilia competitions) or 300 min (X. citri vs S. maltophilia competitions) and cell lysis events were manually registered. Approximately 100 cells were scored for each assay. Quantification of K. pneumoniae, S. Typhi and P. aeruginosa killing by S. maltophilia was performed as described for E. coli.
For time-lapse imaging of the effect of Smlt3024 delivery into E. coli cells, the X. citri Δ8Δ2609-GFP strain expressing Smlt3025/3024 and E. coli containing the pEXT22-derived constructs expressing Smlt3025 were grown overnight in AB media supplemented with antibiotics. E. coli cells were diluted 100-fold in the same media with 200μM IPTG and grown for an additional 6 h to induce Smlt3025 production from the PTAC promoter. No induction of Smlt3025/3024 expression in X. citri is required due to leaky expression from the PBAD promoter. Before imaging, cells were pelleted and resuspended in AB medium with 0.2% sucrose and 0.2% casamino acids to remove antibiotics, diluted and mixed. To quantify E. coli doubling times, single cells in close contact with X. citri cells at time 0 were marked and followed through time. When mother and daughter cells showed clear separation of the division septa, the time of division was recorded. If either mother or daughter cell were still in contact with X. citri cells after division, subsequent division events of these cells were also counted. Cells that did not divide during the recorded time-lapse movie were not included in the calculations (ND in Fig 5). Doubling times of E. coli cells in the vicinity of but not in contact with X. citri expressing Smlt3024 were also recorded. For each condition, on average 100 cells were tracked. Since measurements started at time 0, independently of the cell-cycle of each marked cell at this time, and each frame of the time-lapse was taken every 20 min, the recorded values provide only a rough estimate of the true doubling times.
To identify putative effectors secreted by the S. maltophilia T4SS, we used the XVIPCDs of known and putative X. citri T4SS substrates (residues in parenthesis): XAC4264(140–279), XAC3634(189–306), XAC3266(735–861), XAC2885(271–395), XAC2609(315–431), XAC1918(477–606), XAC1165(1–112), XAC0574(317–440), XAC0466(488–584), XAC0323(16–136), XAC0151(120–254), XAC0096(506–646) [10, 30] to BLAST search the genome of S. maltophilia K279a (https://www.genome.jp/tools/blast/). A list of S. maltophilia proteins identified by each X. citri XVIPCD with their respective E-values is shown in S1 Table.
Smlt302586-333 and full-length Smlt3024 cloned into pSUMO or pET28a, were transformed into E. coli BL21(DE3) and SHuffle T7 competent E. coli cells (New England BioLabs), respectively, and subcultured into 2x YT medium supplemented with 50 μg/mL kanamycin at 37°C until OD600nm of 0.6 and then shifted to 18°C. After 30 min, protein production was induced with 0.1 mM IPTG. After overnight expression, cells were harvested by centrifugation and resuspended in 20 mM Tris-HCl (pH 8.0), 200 mM NaCl, 5 mM imidazole and lysed by 10 passages in a French Press system. The lysate soluble fraction was loaded onto a 5 mL HiTrap chelating HP column (GE Healthcare) immobilized with 100 mM cobalt chloride and equilibrated with the lysis buffer. After the removal of unbound proteins, the protein was eluted with lysis buffer supplemented with 100 mM imidazole. For the proteins expressed with the SUMO tag, there was an intermediate purification step that began with the removal of the 6xHisSUMO-tag, with the addition of Ulp1 protease to the eluted protein, followed by dialysis at 4°C for 12 h for removal of imidazole. The cleaved target proteins were purified after a second passage through the HiTrap chelating HP column immobilized with cobalt, being eluted in the unbound fraction. Molecular masses of the isolated proteins and the effector-immunity complex were determined by SEC-MALS (size-exclusion chromatography coupled to multi-angle light scattering), using a Superdex 200 10/300 GL (GE Healthcare) coupled to a Wyatt MALS detector. Graphs and the average molecular masses were generated using the ASTRA software (Wyatt), assuming a refractive index increment dn/dc = 0.185 mL/g.
Translocation assays were performed essentially as previously described [10]. Briefly, S. maltophilia wild-type and ΔvirD4 strains carrying pBRA-FLAG-smlt3024 were grown overnight with antibiotics (150 μg/mL streptomycin), subcultured on the next day (1:25 dilution) and grown for an additional 2 h at 37°C (200 rpm). E. coli cells were subcultured (1:100 dilution) in a similar manner. S. maltophilia and E. coli cells were washed with 2x YT, OD600nm adjusted to 1.0, mixed 1:1 volume and 5–10 μL were spotted onto dry nitrocellulose membranes, which were quickly placed onto LB-agar plates containing 0.1% L-arabinose to induce the expression of FLAG-Smlt3024. Plates were incubated at 30°C for 6 h, sufficient to allow detection of secreted proteins and before spontaneous cell death, which would produce background in the dot blot. After 6 h, membranes were washed with 5% low-fat milk diluted in PBS containing 0.02% sodium azide and processed for quantitative dot blot analysis with anti-FLAG rabbit polyclonal antibody, followed by IRDye 800CW anti-rabbit IgG (LI-COR Biosciences) and scanned using an Odyssey CLx infrared imaging system (LI-COR Biosciences). To obtain good signal to noise ratios, the membranes were washed in PBS/Tween (0.05%) at least four times for 1 h each. Quantification of signal intensity was performed using FIJI software [77].
6xHis-Smlt302586-333 at a concentration of 8 mg/ml, was submitted to initial crystallization assays using the sitting drop vapour diffusion method with several commercial crystallization screening kits. 6xHis-Smlt302586-333 successfully crystallized at 18°C, in the Morpheus conditions B4 and D8 (Molecular Dimensions). X-ray diffraction data of the crystals were collected in the MX-2 beamline of the National Laboratory of Synchrotron Light (Campinas, Brazil). Two datasets were acquired, a native dataset at 1.9 Å resolution and an iodine derivative dataset at 2 Å obtained after soaking the crystals for 40 s in the crystallization condition supplemented with 1M NaI. Space group determination and reflection intensity integration was calculated by the XDS program package [78]. Heavy atoms positions were found by SHELX [79], and the automated phasing and model building was performed with CRANK2 [80] within the CCP4i2 package [81]. The preliminary model was used for molecular replacement conducted with Phenix AutoSol [82] and applied to the native dataset to extend the structure resolution to a 1.9 Å resolution. Structural refinement of the model was performed using Phenix [82] and Coot [83]. Secondary structure was assigned by STRIDE [84].
|
10.1371/journal.pgen.1007351 | Cytokinin stabilizes WUSCHEL by acting on the protein domains required for nuclear enrichment and transcription | Concentration-dependent transcriptional regulation and the spatial regulation of transcription factor levels are poorly studied in plant development. WUSCHEL, a stem cell-promoting homeodomain transcription factor, accumulates at a higher level in the rib meristem than in the overlying central zone, which harbors stem cells in the shoot apical meristems of Arabidopsis thaliana. The differential accumulation of WUSCHEL in adjacent cells is critical for the spatial regulation and levels of CLAVATA3, a negative regulator of WUSCHEL transcription. Earlier studies have revealed that DNA-dependent dimerization, subcellular partitioning and protein destabilization control WUSCHEL protein levels and spatial accumulation. Moreover, the destabilization of WUSCHEL may also depend on the protein concentration. However, the roles of extrinsic spatial cues in maintaining differential accumulation of WUS are not understood. Through transient manipulation of hormone levels, hormone response patterns and analysis of the receptor mutants, we show that cytokinin signaling in the rib meristem acts through the transcriptional regulatory domains, the acidic domain and the WUSCHEL-box, to stabilize the WUS protein. Furthermore, we show that the same WUSCHEL-box functions as a degron sequence in cytokinin deficient regions in the central zone, leading to the destabilization of WUSCHEL. The coupled functions of the WUSCHEL-box in nuclear retention as described earlier, together with cytokinin sensing, reinforce higher nuclear accumulation of WUSCHEL in the rib meristem. In contrast a sub-threshold level may expose the WUSCHEL-box to destabilizing signals in the central zone. Thus, the cytokinin signaling acts as an asymmetric spatial cue in stabilizing the WUSCHEL protein to lead to its differential accumulation in neighboring cells, which is critical for concentration-dependent spatial regulation of CLAVATA3 transcription and meristem maintenance. Furthermore, our work shows that cytokinin response is regulated independently of the WUSCHEL function which may provide robustness to the regulation of WUSCHEL concentration.
| Stem cell regulation is critical for the development of all organisms, and plants have particularly unique stem cell populations that are maintained throughout their lifespan at the tips of both the shoots and roots. Proper spatial and temporal regulation of gene expression by mobile proteins is essential for maintaining these stem cell populations. Here we show that in the shoot, the mobile stem cell promoting factor WUSCHEL is stabilized at the protein level by the plant hormone cytokinin. This stabilization occurs in a tightly restricted spatial context, and movement of WUSCHEL outside of this region results in WUSCHEL instability that leads to its degradation. The specific regions on the WUSCHEL protein that respond to the cytokinin signaling are the same regions that are essential for both proper WUSCHEL localization in the nucleus and regulation of its target genes. This spatially specific response to cytokinin results in differential accumulation of WUSCHEL in space, and reveals an intrinsic link between protein stability and the regulation of target genes to maintain a stable population of stem cells.
| Plant meristem development depends largely on positional information. Determining how cells interpret positional cues to regulate gene expression is central to cell fate specification during pattern formation. The concentration of transcription factors (TFs) in developmental fields have been shown to provide positional information in gene expression and cell fate specification in animal systems [1, 2, 3]. The concentration-dependent transcriptional regulation is not well-studied in plants. Recently, the homeodomain TF-WUSCHEL (WUS) has been shown to regulate stem cell gene expression in a concentration-dependent manner. Stem cells located in the central zone (CZ) of the shoot apical meristems (SAMs) of plants sustain the growth and development of all aboveground plant parts [4]. A subset of stem cell daughters that divide relatively infrequently remains in the CZ [5, 6, 7]. The other stem cell daughters that are displaced into the adjacent peripheral zone (PZ) divide rapidly and differentiate as lateral organs, while those that are displaced basally into the rib meristem (RM) differentiate as pith cells and become part of the stem [4]. WUS expressed in the RM is essential for the maintenance of stem cells in the overlying CZ [8, 9, 10]. WUS protein has been shown to migrate into adjacent cells [11], likely through plasmodesmata [12] where it accumulates at a lower level than in the RM. WUS promotes stem cell fate in the CZ, regulates cell division patterns, and also activates CLV3 [9, 13, 14]. CLV3 is a secreted peptide that activates a receptor kinase pathway to restrict WUS transcription [15–18]. A recent study has shown that WUS binds the same cis-elements to activate and repress CLV3 at lower and higher levels, respectively [13]. The concentration-dependent transcriptional regulation has been shown to control CLV3 levels and spatial accumulation. Thus the differential accumulation of WUS in the RM and in the adjacent CZ is critical for regulating CLV3, which in turn impacts WUS transcript levels and the overall size of the SAM [19].
Rodriguez et al. [20] have shown that several interdependent processes involving DNA dependent homodimerization, nuclear-cytoplasmic partitioning, and protein destabilization contribute to higher WUS accumulation in the RM than in the adjacent cells. A specific amino acid residue in the DNA binding homeodomain (referred to as HOD1) is required for both DNA binding and homodimerization. However, HOD1 is not sufficient for full homodimerization, as a second part of the protein (aa134-aa208) also mediates homodimerization (referred to as HOD2). The mHOD1 (single amino acid missense mutation G77E) and ΔHOD2 (deletion of amino acids 134–208) double mutants of WUS have been shown to have higher diffusivity, which implicates DNA binding and homodimerization in the spatial accumulation of WUS. Also, the last 63 amino acid stretch of WUS that contains the acidic region, the WUS-box and the EAR-like domain has been shown to be sufficient for differential accumulation of WUS in the RM and the CZ [20]. A combined analysis of the point mutations in single and double mutants of the WUS-box and the EAR-like domain suggested that the EAR-like domain functions as a nuclear export signal, while the WUS-box functions as a nuclear retention signal, which implicates the nuclear-cytoplasmic partitioning dynamics in the regulation of nuclear concentration [20]. Moreover, ectopic overexpression of the wild type WUS in the CZ resulted in lower WUS protein accumulation [13]. On the contrary, the ectopic overexpression of WUS carrying a potent nuclear localization signal accumulated stably in cells of the CZ, showing that nuclear enrichment leads to WUS protein stability [13]. These studies suggest that higher WUS levels leads to its destabilization, until it reaches a saturating concentration where it becomes stable. Furthermore, a conditional nuclear translocation of WUS using the Dexamethasone (Dex) inducible form of WUS led to low or undetectable amount of the protein within 24 hours of Dex treatment in the CZ cells [20]. In contrast, the cells in deeper cell layers of the RM and the PZ accumulated WUS protein, suggesting that these cells may contain signals that can protect WUS from destabilizing signal/s. Therefore, positional signals and mechanisms that regulate differential accumulation of WUS in SAMs requires further investigation.
The higher WUS accumulation in the RM could simply be due to the synthesis of the protein in these cells [9, 11]. Earlier studies have shown that the exogenous application of cytokinin can induce WUS promoter activity in the RM [21, 22]. Recent studies have shown that the type-B ARRs that function downstream of the cytokinin receptor system bind the WUS promoter and activate transcription [23]. The synthetic cytokinin sensing promoter-pTCSn is active in cells of the RM while its activity is excluded from the L1 and the L2 layers [24, 25], which is explained based on the localized expression of cytokinin receptors in the RM [26]. The higher cytokinin signaling could be established by the WUS-mediated repression of type A ARABIDOPSIS RESPONSE REGULATORS (ARRs), negative regulators of cytokinin signaling [27]. Therefore, synthesis of WUS in the RM, promoted by the cytokinin signaling, may lead to higher WUS protein accumulation. However, as shown by the ectopic overexpression experiments explained in the earlier section, the higher synthesis alone would not lead to a uniformly higher protein accumulation. This suggests that additional levels of post-transcriptional regulation mediated by the positional signals could be important.
Here we present evidence that the RM-localized cytokinin signaling is necessary for promoting stability of the WUS protein, and acts on the acidic domain and the WUS-box transcriptional regulatory domains to stabilize the protein. We show that the Dex inducible form of the transcriptionally inactive, WUS-box mutant form of WUS fails to destabilize itself despite nuclear translocation, providing evidence that the WUS-box functions as a degron sequence. We also demonstrate that cytokinin can negate destabilization of WUS. The work presented here enriches the prevailing paradigm by providing a framework which involves a sensitive role for cytokinin signaling in promoting WUS protein stability in addition to its expression. The multifaceted WUS-box required for nuclear retention, cytokinin sensing, degradation and transcription acts as a common link that differentially interprets spatial cues to regulate the differential WUS accumulation required for concentration-dependent regulation of CLV3 transcription.
We tested the effect of exogenous application of 10 μM 6-benzylaminopurine (6-BAP), which is sufficient to activate transcription of a known cytokinin responsive gene, the type A ARABIDOPSIS RESPONSE REGULATOR 5 (ARR5) [22], on the accumulation of WUS protein. Mock-treated pWUS::eGFP-WUS plants revealed pWUS::eGFP-WUS fluorescence primarily in the nuclei of the apical L3 and basal L3 cells of the SAM, and low levels of accumulation in the L1 and the L2 layers and also in the pith (Fig 1A). Upon 6 hrs of 6-BAP treatment, the pWUS::eGFP-WUS accumulation remained mostly nuclear and extended into the most basally located L3 cells and into the underlying pith (Fig 1B) (n = 8). By 12 hrs of 6-BAP treatment, pWUS::eGFP-WUS nuclear accumulation continued in the basal L3 layers and the pith, along with an increased accumulation in the L2 and the apical L3 layers (Fig 1C) (n = 5). By 24 hrs of 6-BAP treatment, the cells in the basal L3 layer and the pith also showed increased accumulation of the protein, with some accumulation appearing outside of the nuclei (Fig 1D and 1L) (n = 7). The cells located in L2 layer revealed a minimal increase in pWUS::eGFP-WUS accumulation even starting at 6 hrs, yet cells in the L1 layer did not show an increase in pWUS::eGFP-WUS accumulation until 24 hrs of 6-BAP treatment when a small increase was seen. The exposure of the pWUS::dsRed-N7 transcriptional reporter to 10 μM 6-BAP did not reveal a dramatic change in the number of WUS expressing cells (Fig 1I–1K) (n = 7), which is consistent with an earlier study showing that cytokinin at concentrations within the physiological range did not induce WUS transcription [22]. Taken together, these observations reveal that exogenous application of cytokinin resulted in higher WUS protein accumulation without a detectable increase in WUS transcript levels.
An increase in the number of cells that accumulated higher WUS protein in the apical L3, the basal L3, and the pith cells when compared to the cells in the L1 and the L2 layers could be due to the increased cytokinin responsiveness of the deeper cell layers. To determine the relationship between cytokinin responsiveness and the WUS protein accumulation, we analyzed the spatial activation patterns of the synthetic cytokinin responsive promoter-pTCSn::mGFP-ER at various times after 10 μM 6-BAP treatment. The pTCSn::mGFP-ER expression in Mock-treated plants was restricted to the apical L3 and the basal L3 cells (Fig 1E) as also shown in an earlier study [25]. At 6 hrs of 6-BAP treatment a dramatic increase in pTCSn::mGFP-ER reporter activity in the basal L3, along with a slight increase in reporter activity was observed in the apical L3 layers, the underlying pith cells, and the vasculature tissue (Fig 1F) (n = 5). Continued 6-BAP treatment for 12 hrs (Fig 1G) (n = 4) and 24 hrs (Fig 1H) (n = 14) resulted in a dramatic increase in pTCSn::mGFP-ER reporter activity in all cells except in the centrally located L1 layer cells, while a few centrally located L2 layer cells activated the reporter weakly. In summary, a higher WUS protein accumulation was observed in cells that revealed a maximal response to cytokinin treatment.
To test the requirement of the cytokinin signaling in WUS protein accumulation, we examined the spatial patterns of WUS transcription, WUS transcript levels, and WUS protein distribution in the cytokinin triple receptor mutant line, cre1-12;ahk2-2;ahk3-3, that has been shown to eliminate cytokinin signaling [28]. RNA in situ hybridization by using the antisense WUS probe revealed normal expression of WUS in deeper cell layers of both Ler and the smaller sized cre1-12;ahk2-2;ahk3-3 SAMs (Fig 2A and 2B) (n = 5). The number of WUS expressing cells were not significantly different in much smaller sized receptor mutant SAMs (Fig 2C). Moreover, the semi-quantitative RT-PCR analysis did not reveal a striking change in WUS transcript levels despite a dramatic downregulation of CRE1, AHK2, and AHK3 transcript levels in cre1-12;ahk2-2;ahk3-3 mutant line (Fig 2I). To study the effect of cytokinin signaling on WUS protein levels, we introduced the pWUS::eGFP-WUS described in an earlier study [20] into the cre1-12;ahk2-2;ahk3-3 mutant line. In most plants, the pWUS::eGFP-WUS accumulation was undetectable (Fig 2E) (n = 18), except in images acquired at higher detector gain which revealed protein accumulation in 2–3 centrally located cells in the L2 and the apical L3 cell layers (Fig 2F) (n = 7). Taken together, this analysis reveals that cytokinin signaling is required for maintaining WUS protein levels.
In order to understand the significance of the decreased WUS accumulation in cre1-12;ahk2-2;ahk3-3 mutants, we analyzed the expression pattern of CLV3, a direct transcriptional target of WUS [13]. The wild type CLV3 expression is detected in a higher number of cells in the L1 layer when compared to the L2 and L3 layers (Fig 2G). The cytokinin receptor mutants revealed a much smaller CLV3 expression domain with strongest expression detected in the two centrally located cells in the L2 layer (Fig 2H) (n = 6). A relatively higher CLV3 expression in sub-epidermal cells could be due to the lower WUS detected in these cells showing the requirement of higher WUS to repress CLV3 in inner layers. A relatively weaker CLV3 expression in the L1 layer could be due to the limitation of WUS. These results are consistent with the WUS concentration-dependent regulation of CLV3 and cytokinin as a spatial WUS stabilizing signal contributing to the spatial regulation of CLV3.
The results presented in previous sections suggest that higher cytokinin response in the apical L3 and the basal L3 cells may promote WUS protein stability. The WUS protein that migrates into the overlying L1 and L2 cells, and into the underlying pith cells that all exhibit limited cytokinin response may become unstable. The exogenous application of cytokinin was unable to induce higher levels of cytokinin response in the L1 and the L2 layers, which correlates with lower levels of WUS protein accumulation (Fig 1A–1D). To test whether activation of the cytokinin signaling was sufficient to stabilize the protein in the L1 and the L2 layers, we ectopically expressed a type B ARABIDOPSIS RESPONSE REGULATOR 1 (ARR1), a TF which activates downstream targets upon phosphorylation by the cytokinin signaling pathway. The deletion of the phospho-transceiver domain (ΔDDK) has been shown to activate cytokinin signaling constitutively even in the absence of cytokinin, and the fusion of ARR1ΔDDK to the hormone binding domain of the rat glucocorticoid receptor (GR) has been shown to induce cytokinin response upon Dex treatment [29, 30]. We expressed ARR1ΔDDK-GR in the CZ by using the two-component system consisting of the pCLV3::LhG4 driver together with p6xOP::ARR1ΔDDK-GR. To test the ability of this system in inducing ectopic cytokinin signaling, we monitored pTCSn::mGFP5-ER expression at 6 hrs, 12 hrs, 24 hrs, and 48 hrs after Dex treatment. The pTCSn::mGFP5-ER expression after 6 hrs of Dex treatment was elevated in the RM and a few cells in the PZ of the L1 layer (Fig 3B) (n = 6) compared to Mock-treated (Fig 3A) plants. By 12 hrs of Dex treatment, the L1 layer expression expanded inwards towards the CZ, and also into the basal L3 cells and the pith but at slightly reduced levels (Fig 3C) (n = 9). The L2 layer showed weakest response appeared to be the least sensitive to ARR1ΔDDK-GR induction along with the L1 cells in the CZ. At 24 hrs after Dex treatment, the pTCSn::mGFP5-ER expression was observed in all cell layers of SAMs, which was accompanied by an overall increase in SAM size (Fig 3D) (n = 5). The pTCSn::mGFP5-ER expression at 48 hrs after Dex treatment intensified further along with the increase in SAM size (Fig 3E) (n = 4). These observations show that expression of ARR1ΔDDK-GR in the CZ can induce cytokinin signaling. The spatial expression of CLV3 promoter was restricted to the CZ across all time points after ARR1ΔDDK-GR induction (Fig 3K–3O) suggesting that the staggered centripetal pattern of induction of cytokinin response in the CZ and in the basal L3 layers is not due to the misexpression of the CLV3 promoter but it is a non-cell autonomous effect.
Next, we monitored the fate of the pWUS::eGFP-WUS protein, and after 6 hrs of Dex treatment pWUS::eGFP-WUS levels increased in all layers of the SAM, including the L1 and the L2 layers (Fig 3G and S1A Fig) (n = 3). At 12 and 24 hrs after Dex treatment, pWUS::eGFP-WUS continued to accumulate at higher levels in the L2 and L3 layers, the basal L3 layers, and the pith, with a slight decrease in the L1 layer (Fig 3H and S1A Fig) (n = 6). Finally, at 48 hrs after Dex treatment, WUS protein was maintained at higher level in the L2 and the apical L3 layers while a decrease in protein level was observed in the L1 layer and in few cells in the pith (Fig 3J; S1A Fig) (n = 5). The analysis of WUS transcript pattern at 48 hrs of Dex treatment revealed an expanded WUS expression domain in deeper cell layers but the WUS transcripts were largely absent from the L1 and the L2 cell layers (S1B and S1C Fig) (n = 2). These results show that increase in WUS protein accumulation is not due to the de novo WUS transcription in the L1 and the L2 layers and also suggest that the ARR1-mediated activation of WUS transcription requires rib meristem context. The higher WUS accumulation in the L1 and the L2 layers could be due to the higher mobility of the protein from inner layers. However, higher WUS protein accumulation in the inner layers upon 6-BAP treatment was not sufficient for higher WUS accumulation in the L1 and the L2 layers. Taken together, these results show that ectopic induction of cytokinin response in the L1 and the L2 layers was sufficient to partially stabilize WUS.
Since the spatial accumulation of WUS involves several interconnected processes: nuclear-cytoplasmic partitioning, its intercellular movement, and stability, a steady state analysis alone is not sufficient to implicate cytokinin in the regulation of protein stability. Moreover, since cytokinin has been shown to increase WUS transcription [22], an unambiguous argument about cytokinin involvement in regulating WUS protein stability requires transient analysis by using WUS expressed from a heterologous promoter not regulated by cytokinin. Our earlier study has shown that the Dex inducible form of ubiquitously expressed-p35S::eGFP-WUS-GR, resulted in sequential destabilization, starting from the CZ within 6 hrs of Dex treatment and extending into the lateral edge of the PZ and the inner cell layers of the RM within 24 hrs of Dex treatment [20]. To test whether cytokinin can counter the instability, we co-treated wild type p35S::eGFP-WUS-GR seedlings with both 10 μM Dex and 10 μM 6-BAP. The 24 hrs treatment of p35S::eGFP-WUS-GR seedlings with 6-BAP alone resulted in an increase in fluorescence levels in the cytoplasm, especially in deeper cell layers (Fig 4C) (n = 5) when compared to the Mock-treated controls (Fig 4A) which is consistent with the higher activation of cytokinin response in these cells (Fig 1E–1H). The simultaneous treatment with both 10 μM Dex and 10 μM 6-BAP resulted in protein accumulation in the nuclei of all cells, including those that are located in the central part of the SAM (Fig 4D) (n = 11) when compared to the Dex treatment alone (Fig 4B) (n = 5). Though the protein accumulated at a relatively lower level in central part of the SAM when compared to the neighboring cells (Fig 4D). Taken together, these experiments show that cytokinin was able to prevent destabilization of WUS observed upon ectopic overexpression, revealing that cytokinin signaling likely acts directly on the WUS protein. Earlier studies have shown that higher levels of cytokinin can induce WUS transcription [21–24] gradually over several days [31]. It is possible that a higher threshold of cytokinin signaling and acquisition of cellular competence together induce WUS transcription. A relatively insensitive mechanism of cytokinin that induces WUS transcription could provide a long-term control of WUS expression, while the highly sensitive effect of stabilizing the WUS protein could enable rapid response over a shorter timescale.
To further test whether cytokinin directly acts on the WUS protein, we utilized the series of mutant WUS proteins developed in an earlier study [20] (Fig 5M). The N-terminal DNA binding homeodomain also required for homodimerization (HOD1) together with the centrally located 74-aa stretch (amino acids 134–208) homodimerization domain (HOD2) restrict WUS protein diffusivity (Fig 5M) [20]. Additionally, the 63-aa stretch (amino acids 229–292) located at the C-terminus of the protein is sufficient for differential nuclear accumulation of WUS, as it influences nuclear-cytoplasmic partitioning and protein stability [20]. Therefore, we tested the responsiveness of the last 63 amino acid stretch of pWUS::eGFP-WUS (amino acids 229–292) (Fig 5B) to exogenous 6-BAP application. The 24 hrs of 6-BAP treatment led to a much broader accumulation of the truncated eGFP-WUS (amino acids 229–292) (Fig 5H, 5O and 5U) (n = 7), than the full length eGFP-WUS (Fig 5G, 5N and 5T) showing that this part of WUS contains signals for sensing cytokinin. The spread of the truncated eGFP-WUS (amino acids 229–292) further into the PZ and developing leaves is likely due to the increased mobility of this smaller and non-dimerizable form of the WUS protein [20].
To fine map the cytokinin sensing region, we tested the response of the mutant versions of the acidic domain, the WUS-box, and the EAR-like domain that were expressed from the WUS promoter [20]. A previous study has shown that deletion of the acidic domain destabilizes the protein as the mutant protein accumulated only in very few cells in the L3 layer, which could be due to the large-scale effects on the protein structure [20]. Therefore, we generated the pWUS::eGFP-WUS (ADM) construct by introducing point mutations in the acidic domain that previously have been shown to affect transcriptional activity [32]. The acidic domain mutant form of WUS accumulated at very low levels in very few cells (5–6 cells) located in the L3 layer (Fig 5C, 5P and 5V) (n = 5). Exogenous application of 6-BAP to the pWUS::eGFP-WUS (ADM) plants failed to induce the protein accumulation in the apical L3, the basal L3 and the pith cells, showing that the acidic domain is required for cytokinin sensing (Fig 5I, 5P and 5V) (n = 9). The WUS-box (pWUS::eGFP-WUS (WBM)) mutant version which has been shown to result in dramatic non-nuclear accumulation [20], also responded poorly to 6-BAP application and therefore revealed its essential function in cytokinin sensing (Fig 5D, 5J, 5Q and 5W) (n = 8). In contrast, the 6-BAP application was able to induce the accumulation of the EAR-like domain mutant version of WUS (pWUS::eGFP-WUS (EARLM)) in the basal L3 and the pith cells (Fig 5E, 5K, 5R and 5X) (n = 4). Variable accumulation of the pWUS::eGFP-WUS (EARLM) mutant version in different lines also suggests that these lines may have accumulated sub-threshold levels of WUS, leading to various degrees of instability (S2A–S2F Fig). The double mutant of the WUS-box and the EAR-like domain (pWUS::eGFP-WUS (WBM+EARLM)) accumulated stably in the nuclei of cells in the RM and the CZ (Fig 5F), but failed to respond robustly to 6-BAP application, again confirming the requirement of the WUS-box for sensing cytokinin (Fig 5L, 5S and 5Y) (n = 8). Collectively, these results show that the acidic domain and the WUS-box are essential for sensing cytokinin signaling.
Our previous analysis showed that the last 63-aa stretch of WUS contains destabilizing signals [20]. The acidic domain functions as one of the cytokinin sensors along with the WUS-box. The EAR-like domain functions in nuclear export [20]. However, the EAR-like domain mutant lines showed various degrees of instability (S2A–S2F Fig) [20] when compared to the EAR-like domain and WUS-box double mutant which accumulated stably. The WUS-box mutant expressed from the WUS promoter accumulated uniformly in cells located both within and outside the RM, suggesting that the WUS-box may be required for destabilizing WUS in cells outside the RM [20]. However, it is also possible that non-nuclear accumulation could have improved stability. In addition, since the WUS-box is also required for transcriptional activity, the loss of transcriptional function might have led to improved stability. To address these issues, we carried out a transient analysis of the wild type and mutant versions of the WUS protein by using a Dex-inducible system.
We expressed the wild type, the WUS-box (WBM), the EAR-like domain (EARLM), and the double mutant (WBM+EARLM) forms of WUS as glucocorticoid receptor (GR) fusions from the ubiquitous (p35S) promoter. As shown in earlier study [20], the wild type-p35S::eGFP-WUS-GR, failed to accumulate in the CZ after 24 hrs of Dex treatment, however it did accumulate in the nuclei of the basal L3 cells, the pith, the PZ, and the leaves (Fig 4B) (n = 5). The EAR-like domain mutant p35::eGFP-WUS (EARLM)-GR (Fig 4I and 4J) (n = 16) accumulated in nuclei throughout the SAMs upon 24 hrs of Dex treatment in a manner consistent with its role in nuclear export. The WUS-box mutant p35::eGFP-WUS (WBM)-GR (Fig 4E and 4F) (n = 11), and the WBM and EARLM double mutant p35::eGFP-WUS (WBM+EARLM)-GR (Fig 4M and 4N) (n = 18) also accumulated uniformly in SAMs upon 24 hrs of Dex treatment. A relatively higher nuclear accumulation of Dex-treated WBM when expressed from the ubiquitous promoter (Fig 4F) as opposed to the non-nuclear accumulation observed when expressed from the WUS promoter [20] suggests that the nuclear export machinery may be limiting. Though these results show that both the EAR-like domain and the WUS-box are required for destabilization, the instability associated with the EAR-like mutants expressed from the WUS promoter suggests that it may not function directly in destabilization [20]. The stable accumulation of the WBM suggests either that it could function as a degron, or that the transcriptional activity provided by the WUS-box is necessary for destabilization.
To test whether transcriptional activity is required for destabilization, we generated an alternative transcriptionally dead version of WUS that leaves the wild type WUS-box intact. An earlier study has shown that the homodimerization deficient mHOD1 (single amino acid missense mutation G77E) and ΔHOD2 (deletion of amino acids 134–208) double mutant version fails to rescue wus null mutant phenotype (Fig 6D) [20]. The 24 hrs of Dex treatment of p35::eGFP-WUS (mHOD1+ΔHOD2)-GR (Fig 4Q and 4R) (n = 10) resulted in lower protein accumulation as seen with the wild type protein (Fig 4B), though the monomeric WUS accumulated at much lower levels especially in differentiating leaves (Fig 4R). Taken together, these results show that transcriptional activity/function of WUS is not required for its destabilization, therefore the stability observed with the WUS-box mutant version suggests that the wild type WUS-box could function as a degron in addition to its role in transcriptional activity. An earlier study has shown that the WUS-box mutant version of WUS accumulated stably when misexpressed in the L1 layer [13]. Our misexpression of the wild type WUS in the L1 layer (pATML1::eGFP-WUS) (n = 44) resulted in lack of visible protein accumulation, or in very few cases lower levels of non-nuclear accumulation (n = 6) (S4A–S4C Fig). This observation is consistent with the stable nuclear accumulation of the GR fusion form of WUS-box mutant version in outer cell layers supporting a role for the WUS-box in nuclear degradation in cells outside the RM that lack cytokinin signaling.
Cytokinin sensing through the WUS-box, which is also required for nuclear retention, might occur in the nucleus or through stabilization of WUS in the cytoplasm leading to higher nuclear import. To distinguish between these possibilities, we tested the responsiveness of the cytoplasmically-sequestered GR-fused forms of WUS to cytokinin. The 6-BAP application stabilized the mHOD1 and ΔHOD2 domain double mutant p35::eGFP-WUS (mHOD1+ΔHOD2)-GR (Fig 4S) (n = 11) and the EARLM p35::eGFP-WUS (EARLM)-GR (Fig 4K) in the cytoplasm similar to the wild type WUS p35::eGFP-WUS-GR (Fig 4C). The WBM p35::eGFP-WUS (WBM)-GR (Fig 4G) failed to respond as robustly as the EARLM p35::eGFP-WUS (EARLM)-GR (Fig 4K) which is consistent with the WUS-box being one of the cytokinin sensors (Fig 5J). The cytokinin induced increase of WUS in the cytoplasm was also observed with the last 63 amino acid stretch of the WUS (Fig 5H). Similar response patterns of the GR fused and the GR independent versions rules out the possibility the observed effects are artifacts of GR-mediated retention of the WUS in the cytoplasm. In addition, the use of heterologous p35 promoter and the already translated GR-fused forms reveals that the cytokinin-induced WUS protein stability is a post-translational effect. These results show that the cytokinin can stabilize WUS protein in the cytoplasm irrespective of its oligomeric status or DNA binding, which could lead to a higher WUS pool available for nuclear import or diffusion into adjacent cells.
Next, we tested the combined effects of Dex and cytokinin treatments on the monomeric and other WUS mutant forms. The 24hr Dex treatment of the monomeric WUS variant p35::eGFP-WUS (mHOD1+ΔHOD2)-GR (Fig 4R) as shown in the previous section resulted in low protein accumulation similar to that of the wild type in the SAM, except in the pith and leaves where it accumulated at lower levels than the wild type protein (Fig 4B). The 24hr Dex plus 6-BAP treatments (Fig 4T) (n = 6) led to improved protein accumulation of the monomeric WUS, which was readily noticeable in inner cell layers, though the overall protein accumulation was lower and less nuclear than the wild type protein (Fig 4D). These results show that cytokinin can offset instability associated with the monomeric WUS, however, the DNA binding and homodimerization are required for higher nuclear accumulation. The combined 24hr treatments of 10 μM Dex and 10 μM 6-BAP did not result in a dramatic change in WUS forms that accumulated stably in the nucleus after Dex treatment alone, including p35::eGFP-WUS (EARLM)-GR (Fig 4L) (n = 18), p35::eGFP-WUS (WBM)-GR (Fig 4H) (n = 13), and p35::eGFP-WUS (WBM+EARLM)-GR (Fig 4P) (n = 3). These results suggest that the cytokinin signaling is largely responsible for stabilizing WUS until it reaches a level where it can independently remain stable.
Analyses presented in previous sections suggest that WUS can self stabilize at higher levels promoted either by cytokinin signaling or through higher nuclear import as shown in an earlier study [13]. To directly test whether transcriptional activity/function of WUS is required for self-stabilization, we introduced the transcriptionally dead-HOD1 and HOD2 double mutant form of WUS pWUS::eGFP-WUS (mHOD1+ΔHOD2) into wus-1, a null mutant of WUS. The mutant form of WUS was barely detectable in the wus mutant background (Fig 6D) when compared to the normal accumulation observed in the wild type background (Fig 6A). Earlier studies have shown that WUS promoter is active in wus-1 mutant background [33] and the WUS promoter used in this study has been shown to rescue wus null mutant phenotype [11], which rules out lack of WUS transcription as the cause for lower WUS protein accumulation. The 6 hrs (Fig 6E) (n = 7) and 24 hrs (Fig 6F) (n = 7) treatments of 6-BAP were only able to minimally improve accumulation of the HOD1 and HOD2 double mutant form of WUS in the wus-1 mutant background when compared to the 6 hrs (Fig 6B) (n = 3) and 24 hrs (Fig 6C) (n = 3) treatments of cytokinin in the wild type background, showing that functional WUS is required for higher protein accumulation and also for the cytokinin-mediated stabilizing effect. However, the inability of the HOD1 and HOD2 double mutant to respond maximally to cytokinin in wus-1 mutant background could also be due to a decrease in cytokinin responsiveness of the wus mutants. To test this possibility, we analyzed the cytokinin response patterns in wus-1 by using the cytokinin sensor pTCSn::mGFP5-ER. The pTCSn expression was observed in deeper cell layers of wus-1 mutants (Fig 6G), a pattern that was comparable to that of the wild type (Fig 1E). The application of 6-BAP to wus-1 mutants was able to fully induce the expression of pTCSn in the deeper cell layers, in the peripheral edges of the SAM, and in the developing leaves, yet it was excluded or expressed at extremely low levels in the L1 layer within 6 hrs (Fig 6H) (n = 5) of treatment. The overall expression intensified at 24 hrs (Fig 6I) (n = 7), similar to the response observed in the wild type SAM (Fig 1H) revealing that cytokinin response is maintained independently from WUS function. Together these results suggest that functional WUS is required for maintaining its own stability and that cytokinin may amplify this effect by enriching the protein in the nucleus.
WUS, a critical regulator of SAM development, is synthesized in the RM where it accumulates at higher levels and migrates into neighboring cells where it accumulates at lower levels [11]. WUS promotes stem cell maintenance by repressing differentiation factors to maintain differentiation program at a distance from the CZ [34]. Similarly WUS levels must also decrease in the basal L3 layers for their timely differentiation into the pith. Therefore, a precise control of the amount and spatial distribution of WUS protein accumulation is critical which could be regulated at the levels of synthesis, mobility and stability. The WUS has been shown to restrict its own transcription by activating the transcription of CLV3, which encodes a secreted peptide to activate receptor kinase pathway/s [10]. WUS has been shown to activate and repress CLV3 transcription at lower and higher levels respectively to regulate CLV3 levels and restrict its spatial expression to the CZ [13]. The RM-localized cytokinin signaling has been shown to activate WUS transcription [22, 31]. Our work reveals a highly sensitive role for localized cytokinin signaling, mediated by the well-characterized membrane-bound sensor histidine kinases, in stabilizing the WUS protein in the apical L3 cells to maintain spatial concentration which is required for spatial regulation of CLV3 transcription and the meristem size. Moreover, the WUS-independent control of cytokinin response pattern, the stabilizer, may provide robustness to the WUS stability control and spatial regulation.
Our work shows that the higher instability of WUS outside the RM can be offset by cytokinin, which suggests that cytokinin may counter a ubiquitously present degradation signal. Our work also shows that WUS function/transcriptional activity is required for self-stabilization. Our previous work has shown that the WUS protein can be stabilized outside the RM by increasing the nuclear import [13] or by decreasing the nuclear export [20] suggesting that higher levels of nuclear WUS may negate the destabilizing signal. Perhaps at higher levels WUS could act as a better transcriptional repressor, as shown in the case of CLV3 regulation [13], to downregulate genes in protein destabilization pathways, act at the post-translational level to block the destabilizing signal, or simply saturate the destabilization machinery. In an attempt to explore the regulation of WUS degradation, we treated pWUS::eGFP-WUS with proteasome inhibitor-MG132 treatment [35]. Within 4 hours of MG132 treatment, we observed a dramatic decrease (S5B Fig) (n = 8) or complete absence of eGFP-WUS signal (S5C Fig) (n = 12). This suggests that the nuclear degradation of WUS has more complex relationship with MG132 sensitive proteasome degradation pathway. It is possible that the degradation signal/factors necessary for WUS degradation are themselves sensitive to proteasomal degradation, and therefore the inhibition of proteasomal function would elevate their levels leading to a rapid WUS degradation. Future analysis aimed at identifying the degradation signal/s and the nature of the nuclear degradation pathway that destabilizes WUS may provide new insights.
In addition to cytokinin, the WUS-mediated activation of CLV3 in the outer layers (the L1 and the L2) could offset WUS destabilization. This is because WUS fails to accumulate at higher level in the outer layers of clv3 null mutants despite higher synthesis in the inner layers [13], even upon exogenous 6-BAP treatment (S3A and S3B Fig) (n = 4) which also fails to activate cytokinin response in outer layers (S3C and S3D Fig) (n = 14). Therefore, CLV3 could act as an additional signal to fine tune the WUS levels in the outer cell layers where cytokinin signaling is absent, allowing it to function as a concentration-dependent switch in regulating CLV3 levels (Fig 7A). Future work will reveal whether CLV3-mediated signaling interferes with nuclear export leading to a higher nuclear WUS and stability or offsets nuclear destabilization directly.
Cytokinin signaling in the RM/L3 layers acts on the acidic domain and the WUS-box to stabilize WUS. The WUS-box is also required for nuclear retention [20]. The nuclear retention machinery acting in concert with the cytokinin signaling may lead to a higher nuclear accumulation, which could mask the degron like activity of the WUS-box leading to the stability of WUS. Outside the RM, the limitation of cytokinin signaling would lead to a reduced nuclear retention and an increased nuclear export through the activity of the EAR-like domain [20]. A net effect of these processes could lead to sub-threshold levels of WUS, which perhaps changes the protein conformation to expose the WUS-box to the destabilization signal. However, the causal relationship between nuclear export and degradation is not clear. The nuclear export could lead to WUS degradation in the cytoplasm as shown in the case of Aryl hydrocarbon receptor, a ligand (2,3,7,8- tetrachlorodibenzo-p-dioxin (TCDD)) activated nuclear TF [35], or nuclear export could simply decrease the nuclear levels of WUS leading to the nuclear degradation.
The WUS-box is also shown to be critical for transcriptional activity of WUS [32]. Earlier studies on TFs-Aryl hydrocarbon receptor [36], Transforming Growth Factor-ß activated SMAD2 [37], and Interferon-gamma activated STAT1 [38] have shown that the protein instability/higher turnover is coupled to transcriptional activation [39, 40]. Moreover, the unstable TFs in eukaryotes, and eubacteria [39] have been shown to use their transcriptional activation domains (TADs) as degrons. The differential utilization of the multifaceted WUS-box for nuclear enrichment in cytokinin rich cells and as a degron in neighboring cytokinin deficient cells provides a tighter spatial control of the regulation of local concentration of WUS and the transcriptional activation-repression switch. A recent study has suggested that WUS-box could function as a MAPkinase docking site required for generating phosphoisoforms of WUS [41]. Future work aimed at analyzing the in vivo relevance of phosphorylation may provide clues to multiple roles of the WUS-box in regulating WUS concentration and transcription.
Arabidopsis plant growth conditions were maintained as described in earlier studies [11]. For seedling imaging experiments, all plants were grown on ½ MS for 7–8 days and then transferred to plates or liquid ½ MS cultures containing either 10 μM 6-benzylaminopurine (6-BAP) (Acros Organics), 10 μM Dexamethasone (Dex) (Sigma), or 10 μM MG132 (Sigma) for the specified period. clv3-2 [15], wus-1 [8] and cre1-12;ahk2-2;ahk3-3 [28] mutants have been described earlier. pWUS::eGFP-WUS [20], pCLV3::LhG4 [14], pTCS::mGFP5-ER [25] and pWUS:dsRed-N7: [24], pCLV3::mGFP5-ER [19] have been described in earlier studies. pTCS::mGFP5-ER generated in Columbia ecotype was backcrossed twice to clv3-2 and wus-1, and all observations were recorded in wild type erecta background. To examine the WUS protein localization in cytokinin receptor mutants, the pWUS::eGFP-WUS line was crossed with cre1-12;ahk2-2;ahk3-3/+ plants. The F1 plants were backcrossed twice with cre1-12;ahk2-2;ahk3-3/+ to generate cre1-12;ahk2-2;ahk3-3/+ line that was homozygous for pWUS::eGFP-WUS in wild type erecta background.
To create the p6xOP::ARR1ΔDDK-GR vector, the DNA fragment of ARR1ΔDDK-GR was amplified using the primer pair MX318 and MX312rGR (see S1 Table) from the p35S::ARR1ΔDDK-GR plasmid (a kind gift from Dr. Takashi Aoyama, Institute for Chemical Research, Kyoto University), and cloned into the pENTR/D-TOPO vector. After confirming the sequence fidelity, the ARR1ΔDDK-GR fragment was inserted into pMX6xOP::GW (or 6xOP pzp222) vector by the LR reaction and introduced into Landsberg erecta (Ler) background. They were crossed to the pCLV3::LhG4 driver line to generate homozygous plants. The PCR primers used for genotyping have been listed in S1 Table. The transgenic lines carrying the WUS mutant versions-the WUS-box (WBM), the EAR-like domain (EARLM) have been described in an earlier study [20]. The acidic domain mutant version is created by PCR mutagenesis by using the primers listed in S1 Table.
The seedlings of appropriate genotypes were grown for 7–8 days on ½ MS plates and transferred to Mock or 10 μM Dex plates for a specified period as necessary. Tissue collection, fixing, sectioning, and probe detection were performed was performed mostly as described earlier [19, 42, 43], with the following modifications: no salt was included in the ethanol dehydration series, and the RNase digestion step was not performed. Full length WUS cDNA was PCR amplified and cloned into pGEM T-easy (Promega) for probe synthesis using purified plasmids. RNA probe was generated with T7 RNA polymerase (Promega), labeled with dioxigenin-rUTP (Roche). Following hybridization, the probe was immuno-blotted with anti-DIG AB (Roche) and developed with Western Blue alkaline phosphatase (Promega).
For semi-quantitative RT-PCR experiments, three replicates each of seedling tissue from 9 plants were ground in liquid nitrogen. RNA was extracted according to the GeneJet Plant RNA Purification kit (Thermo Scientific). cDNA synthesis was performed using the ThermoScript RT-PCR System (Thermo Scientific). PCR fragments were amplified by employing 35 cycles (WUS), 35 cycles (CRE1), 35 cycles (AHK2), 30 and 35 cycles (AHK3) and 22 cycles (UBQ10). The primers used for amplification are listed in S1 Table.
Seedlings were embedded in 3% agarose melted to 60°C to create a block for tissue sectioning. Excess agarose was trimmed off using a razor, tweezers, and a Zeiss Stemi 2000-C dissecting microscope. The embedded seedling was then oriented in a vertical position, and sliced into two halves using a feather razor blade. Each half was then immediately transferred to a 3% FM4-64x cell membrane staining solution for 10–15 minutes, and then placed on a slide with a coverslip in ample ddH2O. Plants were screened for optimal cut sections using a 20x objective on a Zeiss Axio Imager.A1 fluorescence microscope before final micrographs were captured with a 40x objective on a Leica SP5 Inverted Confocal microscope. eGFP fluorescence was detected with 488 nm excitation and an emission collection between 525 nm-550 nm. FM4-64x membrane stain and dsRed-N7 fluorescence were detected using 543 nm excitation with emission collected between 600 nm-650 nm and 575 nm-625 nm respectively.
Images from confocal microscopy were loaded into the Icy Bioimage software [44] (http://icy.bioimageanalysis.org) and isolated for the eGFP or dsRed-N7 channel. The HK Means and Active Contour plugins were used to detect cell counts. For fluorescence quantification analysis, ROI boxes were drawn in ImageJ [45] (https://imagej.nih.gov/ij/index.html) through the central column of SAM cells (30 μm wide and 100 μm tall) and plot profile was used to quantify fluorescence. This was performed on 4–8 confocal images per treatment, and values for each cell layer were averaged across all samples within a treatment. For cell count measurements, the number of fluorescing cells in each layer was counted for 4–8 confocal images per treatment and subsequently averaged across all samples.
|
10.1371/journal.pntd.0000805 | Neurocysticercosis, a Persisting Health Problem in Mexico | The ongoing epidemiological transition in Mexico minimizes the relative impact of neurocysticercosis (NC) on public health. However, hard data on the disease frequency are not available.
All clinical records from patients admitted in the Instituto Nacional de Neurologia y Neurocirugia (INNN) at Mexico City in 1994 and 2004 were revised. The frequencies of hospitalized NC patients in neurology, neurosurgery and psychiatry services, as well as NC mortality from 1995 through 2009, were retrieved. Statistical analyses were made to evaluate possible significant differences in frequencies of NC patients' admission between 1994 and 2004, and in yearly frequencies of NC patients' hospitalization and death between 1995 and 2009.
NC frequency in INNN is not significantly different in 1994 and 2004. Between these two years, clinical severity of the cases diminished and the proportion of patients living in Mexico City increased. Yearly frequencies of hospitalization in neurology and psychiatry services were stable, while frequencies of hospitalization in neurosurgery service and mortality significantly decreased between 1995 and 2009.
Our findings show a stable tendency of hospital cases during the last decade that should encourage to redouble efforts to control this ancient disease.
| Human neurocysticercosis is a severe parasitic disease caused by the installation of Taenia solium larvae in the central nervous system. Neurocysticercosis is still deeply rooted in Latin-America, Africa and Asia, where it develops its complete life cycle promoted by poor sanitary conditions. It is also emerging in developed countries due to human migration. Although hard data on the evolution of the disease incidence in endemic countries are lacking, its presence is being obscured by the growth of degenerative and metabolic diseases, creating the illusion of having disappeared.
In this article, we show that neurocysticercosis frequency has not significantly changed between 1994 and 2009 among patients attending the Instituto Nacional de Neurología y Neurocirugía, Mexico City, the principal Mexican neurological center. We also show that clinical severity of the cases diminished during this period, associated with the higher proportion of neurocysticercotic patients from Mexico City rather than from the states, where local neurological facilities have improved. These results show that neurocysticercosis is still relevant in México, and that more effective efforts should be put toward its eradication.
| Neurocysticercosis (NC) is a life-threatening and costly parasitic disease, endemic in most non-developed countries and increasing in developed world [1]–[5]. NC real prevalence and incidence are difficult to assess, as symptoms are highly heterogeneous and its diagnosis requires neuroradiological studies, not available to all population at risk. Anyway, some data show the persistence of active transmission in Mexico. Particularly, a partial report including only patients hospitalized at INNN, Mexico, showed no statistically significant decrease of NC frequency between 1995 and 2001 (from 2.4 to 1.8%) [6], transversal surveys in rural communities indicate the persistence of human NC (prevalence>9%) [7], [8] and porcine cysticercosis in rural pigs (up to 30%) [9], [10]. In spite of these data, the epidemiological transition occurring in Mexico, with increased diagnosis of metabolic, neoplastic and degenerative diseases [11], could lead us to disregard the importance of NC in Mexico [12]. Herein, the frequency of NC in all patients who attended at INNN in 1994 was compared with that of 2004, and frequencies of hospitalization and mortality of NC patients in neurology, neurosurgery and psychiatry services between 1995 and 2009 are presented.
This study was performed at the Instituto Nacional de Neurología y Neurocirugía (INNN). INNN is a public, third-level referral center located in Mexico City, where neurological, neurosurgical and psychiatric patients above 15 years old are attended. Admission is reserved to patients lacking social security.
To evaluate the evolution of NC burden on INNN, two approaches were taken. First, all clinical records of patients admitted at INNN in 1994 (n = 4098) and 2004 (n = 4706) were manually, anonymously reviewed. Patients fulfilling criteria of NC definitive diagnosis based on Del Brutto's et al. criteria [13] were selected and their demographical, clinical and radiological characteristics were obtained. Second, the yearly numbers of hospitalized NC patients and deaths due to NC at neurology, neurosurgery and psychiatry services from 1995 to 2009 were obtained from INNN's epidemiological service. Frequencies of NC hospitalized patients and NC mortality with respect to the total number of hospitalized patients in each service were calculated.
Statistical analysis using SPSS software was completed. Chi-square test was made to evaluate statistical differences between proportions and Student t-test between means. 95% confidence intervals of proportions and means were provided. Linear regression was used to test changes during calendar periods, considering calendar year as the independent variable.
This study was approved by INNN Institutional Review Board.
Table 1 summarizes the differences in the frequency of NC cases attended in the INNN in 1994 and 2004. Between the years 1994 and 2004, no significant statistical differences were found neither in the frequency of NC at INNN (100/4098, 2.4% vs. 120/4706, 2.5%, respectively) nor in patients general profile. Two significant differences were observed, though: in 2004, the cases severity (presence of intracranial hypertension) significantly decreased (P = 0.007), and the proportion of patients living in Mexico City significantly increased (P = 0.005).
Figure 1 shows the evolution of NC frequency in hospitalized patients between 1995 and 2009. In neurosurgery service, frequencies of hospitalization of NC patients varied between 0.67 and 7.9% with a statistically significant decrease between these two dates (r = –0.79, P<0.001). In neurology service, frequencies of hospitalization varied between 3.4 and 10.9% without any significant temporal trend (r = 0.12, P = 0.67), while in psychiatry, only five NC patients were hospitalized during the 15 years under study, with no significant tendency (r = –0.38, P = 0.16). Concerning mortality during this period, 28 NC patients died at INNN, with a significant decrease in mortality frequency (r = –0.7, P = 0.003).
The evident epidemiological transition occurring in Mexico, with growing incidence of metabolic, neoplastic, and degenerative diseases [11], sometimes leads us to forget the weight of other “archaic”, infectious disorders, linked mainly with poverty. NC is one of those. Evolution of NC prevalence is not known, as population-based studies are not available. In the present report, during the 10-year period studied the frequency of NC showed no significant decrease in INNN, the most active neurological center in Mexico. However, some significant changes did occur. The proportion of NC patients from Mexico City increased, possibly due to rural population's migration. The acquisition of computerized tomography, between 1994 and 2004, by most of the public hospitals in States neighboring Mexico City may also have resulted in more frequent local NC diagnosis than before and in lessening their attendance to INNN. The only optimistic change detected in this study was the significant decrease in the severity of NC, a change that could be related with its earlier diagnosis. These results were confirmed when only hospitalized patients were considered between 1995 and 2009: the frequency of hospitalized patients in neurology was stable, while NC frequency in neurosurgery service, as well as NC mortality, decreased.
These data strengthen the notion of persistent active transmission, in agreement with the high prevalence of pig cysticercosis (13.3%) reported in recent epidemiological studies performed in rural areas of Mexico [14]. The latter is the most appropriate indicator to demonstrate active transmission, since 80% of pigs in rural communities are consumed before they are one-year-old [10]. In rural areas, the high prevalence of pig cysticercosis is accompanied by a high frequency of human calcified neurocysticercosis cases and an extremely low frequency of vesicular cysticerci [8], [15]. These disproportions between calcified and vesicular cysticerci in human NC indicate that in most cases the parasite is destroyed. On the other hand, the stability of NC frequency in INNN also indicate that, despite this fact, the parasite (vesicular or calcified) still causes symptoms in an important number of infected subjects, resulting in around 2.4% of consultations in INNN are due to NC.
This study demonstrates that NC is still a health problem of Mexico. It is important to note that INNN and Mexico City does not properly represent the whole country. However, the stability of NC frequency in patients attending INNN and in the hospitalization rate in the INNN neurology service must alert medical practitioners and health authorities on the persistence of unresolved health problems related with poverty. Efforts should be encouraged to apply effective measures for their eradication.
|
10.1371/journal.pcbi.1004803 | Revealing the True Incidence of Pandemic A(H1N1)pdm09 Influenza in Finland during the First Two Seasons — An Analysis Based on a Dynamic Transmission Model | The threat of the new pandemic influenza A(H1N1)pdm09 imposed a heavy burden on the public health system in Finland in 2009-2010. An extensive vaccination campaign was set up in the middle of the first pandemic season. However, the true number of infected individuals remains uncertain as the surveillance missed a large portion of mild infections. We constructed a transmission model to simulate the spread of influenza in the Finnish population. We used the model to analyse the two first years (2009-2011) of A(H1N1)pdm09 in Finland. Using data from the national surveillance of influenza and data on close person-to-person (social) contacts in the population, we estimated that 6% (90% credible interval 5.1 – 6.7%) of the population was infected with A(H1N1)pdm09 in the first pandemic season (2009/2010) and an additional 3% (2.5 – 3.5%) in the second season (2010/2011). Vaccination had a substantial impact in mitigating the second season. The dynamic approach allowed us to discover how the proportion of detected cases changed over the course of the epidemic. The role of time-varying reproduction number, capturing the effects of weather and changes in behaviour, was important in shaping the epidemic.
| In 2009, the threat of the new pandemic influenza A(H1N1)pdm09 (referenced in media as ‘swine flu’) created a heavy burden to the public health systems wordwide. In Finland, an extensive vaccination campaign was set up in the middle of the first pandemic season 2009/2010. However, the true number of infected individuals remains uncertain as the surveillance missed a large portion of mild infections. We built a probabilistic model of influenza transmission that accounts for observation bias and the possible impact of the changing weather and population behaviour. We used the model to simulate the spread of influenza in Finland during the two first years (2009-2011) of A(H1N1)pdm09 in Finland. Using data from the national surveillance of influenza and data on social contacts in the population, we estimated that 9% of the population was infected with A(H1N1)pdm09 during the studied period. Vaccination had a substantial impact in mitigating the second season.
| The threat of the pandemic influenza A strain, A(H1N1)pdm09 (‘swine flu’), imposed a huge burden on the public health system in Finland in 2009 [1]. The first A(H1N1)pdm09 season was part of the global pandemic and occurred from September 2009 through January 2010 with a major outbreak in November 2009. To mitigate the epidemic, a national vaccination campaign was started in October 2009, and by February 2010 approximately half of the Finnish population had been vaccinated against A(H1N1)pdm09. The second epidemic season occurred a year later from November 2010 through April 2011. Only sporadic cases were observed before the first epidemic season and between the two seasons.
It is well known that that laboratory-based surveillance of influenza misses the vast majority of infections that occur in the population. Underreporting follows from asymptomatic or non-diagnosed infection or incomplete reporting of influenza cases in primary and secondary health care. More severe cases are diagnosed and reported with a higher probability. This was true also for A(H1N1)pdm09, although special efforts were taken to record cases especially during the early phases of the first season.
Bayesian methodology (evidence synthesis) has been used to analyse influenza outbreaks in the presence of underreporting and mixed data sources. In general, the underlying epidemiological models can be classified as static or dynamic. In static models, cases are typically aggregated by season and the unknown true incidence is estimated as an attack rate (probability of becoming infected during the season) [2–5]. In dynamic models, the process of spread of the infection via transmission is modelled explicitly [6]. The static approach is simpler and requires less computational resources while the dynamic model enables one to answer more complex questions.
Based on a static model, we previously estimated that only 4% of the Finnish population were infected with A(H1N1)pdm09 over the season 2009/2010 and an additional 1% during the 2010/2011 season [2]. The most affected age groups were children and teenagers with attack rates up to 10-12%. The attack rates were much lower in the second season, which was likely due to the relatively high immunity due to natural infection or vaccination in the most influential age groups. In particular, 74-81% of children aged less than 15 years had been vaccinated against A(H1N1)pdm09 before the second season.
However, a static model cannot address the impact of herd immunity induced by vaccination. To properly address the role of vaccination in mitigating the first-season epidemic and lowering the transmission potential before the second season, a more dynamic (i.e. transmission) model is needed. A dynamic model can also address questions about which age groups played the most important role in transmission or why there was a second season despite the fact the influenza strain did not evolve considerably between the seasons to escape population immunity [7]. The effect of time-varying conditions due to weather or public response to the outbreak can also be inferred using a dynamic model [8].
In this study, we built a dynamic probabilistic model of influenza transmission and disease. The model accounts for transmission of influenza in the population, the impact of vaccination, outcomes with varying severity and imperfect detection of infection. We calibrate the model to data on A(H1N1)pdm09 cases and estimate the true incidence of A(H1N1)pdm09 of the first two A(H1N1)pdm09 seasons in Finland.
In all datasets used in this study, information about individuals was aggregated into 16 age groups: 0-4, 5-9, …, 70-74, 75+ years of age. Fig 1 presents the data on registered A(H1N1)pdm09 cases and the coverage of vaccination in Finland 2009-2011. The population sizes were obtained from Statistic Finland (www.stat.fi).
We built a discrete-time dynamical model of influenza transmission and disease in the Finnish population. The time step was one week, corresponding to the resolution in the data. A period of 113 weeks was modelled from week 15/2009 (one month before the first A(H1N1)pdm09 cases of were registered in Finland) through week 22/2011 (after the end of the second season). Within the modelled period, two subperiods are referred to as the first epidemic season (weeks 37/2009 through 1/2010) and the second epidemic season (weeks 46/2010 through 17/2011).
The prior distributions of the model parameters are presented in Table 1. All parameters except the age-dependent susceptibility p and transmission random effect wt were given informative priors. The severity parameters s(sev/inf) and s(IC/sev) were centred around 1% and 10%, respectively. The detection probabilities d t ( mild ) and d(hosp) were centred around 1% and 75%, respectively. These priors were consistent with the ones used in our earlier analysis of the same data [2]. The inflow of infection q was assumed to be extremely small.
We estimated the joint posterior distribution of the model parameters and latent variables using Markov chain Monte Carlo computation (MCMC) with particle Gibbs sampler step [17]. In addition, we applied exact approximate MCMC [18] targeting a smoothed marginal posterior of the model parameters, p(parameters|data)1/25 and p(parameters|data)1/5, to ensure that the peak area of the target posterior is unimodal and well-behaving. Details are provided in S3 Appendix.
Posterior predictive checks were used to to explore how well the model describes the observed data. Sensitivity analysis was performed by comparing the posterior modes (i.e. maximum a posteriori estimates) under different prior settings. Details are provided in S5 Appendix.
If not otherwise stated, the results will be presented in terms of 90% posterior intervals (i.e. the 5th and 95th percentiles) of the estimated quantities. Additional results are presented in S4 Appendix. Exact numerical estimates are presented in S2 Dataset.
The estimated true numbers of A(H1N1)pdm09 infection are shown in Fig 5A and 5B. Fig 6A presents the attack rates, i.e. the numbers of infected per population size. We estimated that 440 000 – 550 000 individuals in total (8.2 – 10.4% of the population, posterior mean 500 000, 9.3%) were infected in Finland during the modelled period. Specifically, the numbers infected were 270 000 – 360 000 (5.1 – 6.7%, posterior mean 320 000, 5.9%) and 140 000 – 190 000 (2.5 – 3.5%, posterior mean 160 000, 3.0%) during the first and the second A(H1N1)pdm09 epidemic seasons, respectively. Only a minor portion of infections (0.3 – 0.4%) occurred outside the two epidemic seasons. In both seasons, the attack rate decreased with age. It was largest in the youngest age group (14 – 19% during the first and 5.5 – 7.6% during the second epidemic season) and smallest in the oldest (5.0 – 6.6% and 4.6 – 6.4%).
Fig 5C presents the cumulative age composition of the infected population per week. The mean age of infection increased with time. Before the peak of the first season, approximately half of all infections occurred among less then 15 years olds. During the second epidemic season only 25% of infections belonged to this age group. The oldest (65+ years) never accounted for a significant portion of the infected population.
Fig 7A shows the posterior distribution of susceptibility p (probability of acquiring infection per contact with an infectious individual) and inflow q (probability of acquiring infection from outside the population). Susceptibility decreased with age: children aged less than 5 years had a 5-fold greater chance to acquire infection per contact than the oldest individuals. Individuals aged 20-29 years were most likely to acquire infection from outside the population.
Fig 7B shows the posterior distributions of the severity parameters s(sev/inf) and s(IC/sev). The hospitalization/infection ratio had a V shape, the infection being more severe among the youngest (s(sev/inf) = 0.7 – 0.9%) and the oldest (1.3 – 1.7%). Children aged 5-14 years had the smallest probability of severe disease per infection (0.3 – 0.4%). The IC/hospitalization ratio did not vary much across age groups, almost repeating the prior information. It was smallest among the youngest (s(IC/sev) = 7 – 8%) and largest (8 – 11%) for those over 30 years.
In our model, influenza transmission, including the outbreaks and periods between epidemic seasons, is modulated by a time-varying reproduction number R0,t = R0 wt (Fig 5E). Before June 2009, R0,t rose above 1 allowing for the minor pre-seasonal outbreak. A significant increase in R0,t in the autumn of 2009 marked the onset of the first epidemic season. After the peak of the first season (November 2009) R0,t dropped below 1 leading to the end of the first outbreak.
By the end of the first epidemic season about 22% of the population were vaccine-protected (Fig 5D), especially in the youngest age groups (53% in <20 year olds, 14% in 20-64 olds, and 11% in >65 olds). This induced herd immunity in the population, so R0,t could raise above 1 without causing an outbreak. By the second epidemic season, 41% of the population acquired immunity from vaccination (posterior mean 52%, 34%, and 47% of individuals aged 0-19, 20-64 and older than 64 years, respectively) and 5 – 7% acquired immunity from infection.
Around October 2010 R0,t started gradually increasing, reaching its maximum in November 2010 and then slowly decreased. For the period November 2010—January 2011, the reproduction number was above 3. This marked the second epidemic season.
The estimates of R0,t outside the epidemic seasons are uncertain, as scarce data are available for these periods. Overall, the product R0,t = R0wt was estimated with smaller uncertainty than wt and R0 individually (see S4 Appendix).
The largest number of potential infections was produced by individuals from age groups 5-14 years old (3.5 – 5.6 infections) (Fig 6B). The smallest number was produced by the oldest age group (0.4 – 0.5 infections). On average, only few infections per week were introduced from outside the population (Fig 6C). The random effect wt and the detection probability d t ( mild ) increased simultaneously during the early phases of the epidemic seasons. However, for any time (t ∈ 0, …, T − 1), the variables wt and d t ( mild ) did not have strong posterior correlation (see S4 Appendix).
Fig 6D shows the number of detected cases per the number of infected (detection ratio; Table 2). We estimated that 2.1 – 2.7% of all A(H1N1)pdm09 infections were detected (specifically 2.5 – 3.3% during the first epidemic season, 1.2 – 1.6% during the second and 1.5 – 2.0% outside seasons). The detection ratio varied by age with posterior means ranging from 3.7% to 1.9%.
We estimated that the detection probability of the mild cases d t ( mild ) reached its maximum before the peak of the first season and decreased subsequently during the outbreak (Fig 5F). During November 2009, the observed numbers of mild infections decreased much faster than the observed numbers of hospitalized cases. According to the model, however, the true numbers of mild and severe infections decreased at the same speed and the observed difference was thus explained by the decline in d t ( mild ). The posterior of the detection probability of hospitalized cases d(hosp) followed the prior closely.
We measured the impact of the vaccination campaign as the number of cases prevented. To estimate this number, we simulated the incidence of infection, using parameter values sampled from the posterior and assuming that no one was vaccinated (va,t = 0). According to this analysis (Fig 8), the second season could have started earlier and caused a larger outbreak, leading to 4-8 times more infections overall (total attack rate would have been 38 – 78%). By contrast, vaccination did not affect the first epidemic season.
We also estimated the impact of the vaccination under a scenario where vaccines were distributed in the same amount but independent of age (va,t = vb,t for all age groups a, b). In this situation our model predicts about twice as many infections overall (total attack rate would have been 15 – 26%).
Using a dynamic transmission model, we estimated that 5.9% (90% credible interval 5.1 – 6.7%) of the Finnish population was infected during the first year of the pandemic A(H1N1)pdm09 strain of influenza in 2009/2010. There was a second season a year later with an attack rate of 3.0% (2.5 – 3.5%) of the population. The vaccination campaign launched in the middle of the first epidemic epidemic season was essential in mitigating the size of the second season, but occurred too late to have an impact on the first season. In both seasons, the proportion of the infected population decreased with age, with the youngest being at least an order of magnitude more likely to be infected than the oldest.
The age distribution of the infected population evolved over time. Before the peak of the first season most infected individuals were children aged less than 15 years. According to the social mixing matrix, estimated from the available data, this age group forms the core group of transmission for infections that spread through droplets in close contact. After the end of the first season, as many as 72% of children aged less than 15 years either had had natural infection (18%) or had been vaccinated (55%) so that the importance of this age group in the chain of transmission decreased. During the second epidemic season, the mean age of infected individuals was higher.
The posterior mean severity of influenza infection, as measured by the hospitalization/infection ratio (parameter s(sev/inf)), was 0.7% when averaged over all age groups and had a clear V shape with the youngest and oldest requiring hospitalization more often. The IC/hospitalization ratio (parameter s(IC/sev)) was driven almost entirely by prior information (around 8% across all age groups).
We estimated that only 2.4% (90% credible interval 2.1 – 2.7%) of infections were recorded by surveillance, i.e. there were 40 – 50 unobserved A(H1N1)pdm09 infections for each detected case. The detection probability peaked early during the first epidemic season, with a clear decline towards the end of the season (Fig 5F). This could reflect the public and governmental concerns increasing initially and then declining as the awareness of the relatively mild impact of the novel A(H1N1)pdm09 virus was revealed. The detection ratio in the second epidemic season was smaller than in the first one. Similar patterns in the detection rates occurred in the UK during the first two years of the pandemic [19].
In our model, the spread of infection is modulated by four quantities: susceptibility to infection (parameter p), the pattern of contacts (contact matrix C), the time varying reproduction number (R0,t) and the rate of inflow of infection (q). Susceptibility to infection was estimated to decrease with age, which is likely to reflect higher levels of pre-existing immunity among older individuals [7]. The contact matrix was based on a survey of daily social contacts in Finland [10].
The standard deterministic SIR model assumes that outbreaks only stop by depletion of the pool of susceptibles. In particular, a second season would be impossible unless the virus evolves to escape the prevailing immunity in the population. Although this is known to happen for seasonal influenza [20], the virus did not change much during the first two years of the pandemic [7]. Vaccination alone cannot explain the observations, as the first season ended in the population with 20% vaccine-induced protection while the second season started with 40%. Therefore, stochasticity in transmission and other mechanisms may be called for.
We applied a time-varying reproduction number (R0,t) of influenza transmission, capturing the impact of changing population behaviour or weather conditions as a stochastic process. In particular, cold and dry weather has been suggested as one of the drivers of influenza transmission [21] and the public behaviour may have changed as the epidemic appeared to be relatively mild. We estimated that R0,t changed markedly with time (Fig 5E).
The model explains the appearance of the second epidemic season when almost half of the population was immune with extraordinary transmission circumstances: the reproduction number was very large (R0,t > 3) for a period October 2010 through January 2011, possibly reflecting a seasonal (weather) effect. Dorigatti et al. [19] reported a similar finding regarding the 3rd wave of A(H1N1)pdm09 in the UK, one year after the 1st and 2nd waves. They inferred that the basic reproduction number increased to 1.5 before the 3rd wave and concluded that this was likely due to the combination of favourable weather conditions and possible evolution of the virus. Of note, we did not factor the possibility of waning immunity in the analysis and, should it have occurred, our estimate of R0,t would be too high.
The rate of introduction of infection to the population (parameter q) mainly plays the role of a primer that initiates the outbreaks. Its influence during the outbreaks (epidemic seasons) was insignificant. Its role was to add stochasticity to the onsets of influenza seasons, thus removing the need to introduce index cases at any fixed time. The estimates of q were notable only for age group 15-29 years, reflecting the fact that the first detected cases in the country belonged to these age groups.
According to our analysis, vaccination played an important role in mitigating A(H1N1)pdm09 transmission in the second season. By the start of the second season, 41% of the population were vaccine-protected while less than 5 – 7% had acquired immunity from infection. We estimated that in the absence of vaccination the affected population would have been about 4 – 8 times larger. It should be noted, however, that these predictions rely heavily on the posterior estimate of the transmission random effect (wt), which in turn may be strongly dependent on the conditions and data in the 2010/2011 season.
In a previous analysis of the same data set [2], we assumed that vaccination did not affect the first season at all, which agrees with the current estimates. Given the two weeks period needed to develop protective immunity after vaccination, it is likely that an effective proportion of immune individuals was only obtained after the end the first epidemic outbreak (Fig 5D). In this study we modelled the vaccine as having a 80% chance to induce complete immunity against the infection. This efficacy was based on a cohort study conducted in a sub-sample of the same population during the same time (i.e. the first season of A(N1H1)pdm09) [11].
We used a discrete-time SIR model with a one-week time step to correspond to the available data. However, a shorter time step would have been more realistic for capturing the dynamics of influenza for which the infectious period is known to last less than a week [6]. In this case, each infection generation in our model likely reflects several actual generations, therefore the basic reproduction number R0,t is an overestimate of what would have been obtained with a smaller time step or a continuous model.
The estimability of model parameters was constrained by the amount of available data. We used informative priors on all of the model parameters except the susceptibility p and the transmission random effect wt and set a strong smoothness constraint on the time-varying processes of transmission (wt) and detection (d t ( mild )). We conducted several sensitivity analyses to study the impact of the choice of the prior distributions (S5 Appendix). We found that increasing the variance of the prior distributions leads to smaller attack rates and vice versa. The prior of the detection probability for mild cases d t ( mild ) was the most influential one. Some estimated quantities were more robust to prior specification. The detection probability d t m i l d was always estimated to increase before the outbreak of the first season. The estimated trends, e.g. the decreasing susceptibility with age and the V shape in the age-specific severity, were also immune to the choice of the prior. The reproduction number R0,t always increased before the seasonal outbreaks.
In our previous study [2] we analysed the same period of the two first years of pandemic influenza in Finland using a static model. A dynamic approach allowed us to take into account the available time-series data to learn about trends in transmissibility and detection. The presented model estimated the total incidence to be 1.5 times higher (see Table 3). A dynamic model also produced larger estimates of the impact of vaccination as it is able to take into account herd immunity effects. Using a static model, we estimated previously that without vaccination the overall attack rate would increase only by 0.8 percentage points.
The attack rates and severity of A(H1N1)pdm09 varied considerably by geographical region (see Table 3). Such variation may be partly due to lack of precision, based on the differences in data availability and in the methods of analysis. Nevertheless, the estimated attack rate in Finland was still smaller than found in other studies. Because of the high per-population risk of hospitalization in Finland (0.06%), the severity of infection (hospitalization/infection ratio) was higher in Finland than elsewhere, probably reflecting differences in the health care system and surveillance. Such differences emphasise the need to calibrate transmission models in each particular setting to best address questions about the performance of surveillance and the impact of influenza seasons.
|
10.1371/journal.pgen.1003045 | Notch-Mediated Suppression of TSC2 Expression Regulates Cell Differentiation in the Drosophila Intestinal Stem Cell Lineage | Epithelial homeostasis in the posterior midgut of Drosophila is maintained by multipotent intestinal stem cells (ISCs). ISCs self-renew and produce enteroblasts (EBs) that differentiate into either enterocytes (ECs) or enteroendocrine cells (EEs) in response to differential Notch (N) activation. Various environmental and growth signals dynamically regulate ISC activity, but their integration with differentiation cues in the ISC lineage remains unclear. Here we identify Notch-mediated repression of Tuberous Sclerosis Complex 2 (TSC2) in EBs as a required step in the commitment of EBs into the EC fate. The TSC1/2 complex inhibits TOR signaling, acting as a tumor suppressor in vertebrates and regulating cell growth. We find that TSC2 is expressed highly in ISCs, where it maintains stem cell identity, and that N-mediated repression of TSC2 in EBs is required and sufficient to promote EC differentiation. Regulation of TSC/TOR activity by N signaling thus emerges as critical for maintenance and differentiation in somatic stem cell lineages.
| Stem cells maintain tissue homeostasis in metazoans. A productive model to study the regulation of stem cell function is the Drosophila posterior midgut. Notch (N) signaling controls intestinal stem cell (ISC) differentiation in this tissue, while ISC proliferation is regulated by growth factor signaling pathways, including Insulin/IGF signaling (IIS). In this study, we explore the interaction between growth signals and N signaling in the control of ISC proliferation and differentiation. We show that TOR signaling, which promotes growth and can be activated by the IIS pathway, is maintained in ISCs in an inactive state by high expression of the TOR inhibitor TSC2. TSC2 expression shelters ISCs from nutritional cues, ensuring their long-term maintenance. In response to N pathway activation in enteroblasts (EB), the ISC daughter cells, TSC2 is transcriptionally repressed and TOR is activated. We demonstrate that this negative interaction between N and TSC2 is required and sufficient for differentiation of EBs into enterocytes (ECs), the absorptive cells of the epithelium. Our findings establish a critical role for TSC in ISC maintenance and provide a mechanism by which N promotes differentiation into the EC fate. The human homologue of TSC2 is an important tumor suppressor, and our study provides new insight into how its regulation controls regenerative processes.
| Regenerative processes in somatic tissues require coordinated regulation of stem cell proliferation and daughter cell differentiation to ensure long-term tissue homeostasis [1]–[3]. The Drosophila posterior midgut epithelium has emerged as an excellent model system to study this regulation [4]–[7]. It is maintained by Intestinal stem cells (ISCs) that divide to self-renew and produce enteroblasts (EB), which undergo differentiation to become either enterocytes (ECs) or enteroendocrine cells (EEs) [8]–[10].
Differentiation in the ISC lineage is controlled by Delta/Notch (Dl/N) signaling. ISCs express Dl and activate N in EBs, thus promoting differentiation into either EEs or ECs. The cell fate decision between ECs and EEs seems to be regulated by the intensity of the Dl signal, i.e. high levels of N activity in EBs result in EC differentiation, while moderate activation of N promotes EE differentiation [10], [11]. Dl-mediated N activation in EBs increases the activity of the Suppressor of Hairless (Su(H)) transcription factor, presumably by replacing the Hairless transcriptional repressor from Enhancer of Split (E(spl)) complex promoters with the Notch intracellular domain (NICD) [12]. How this pathway coordinates cell specification with cell growth and proliferation in the ISC lineage remains unclear.
ISC proliferation is regulated by growth factor and stress signaling pathways [13]–[32]. These pro-mitotic signals include the Insulin/IGF signaling pathway (IIS), which is sufficient and required for ISC proliferation [18], [23], [33], [34]. Activation of the Insulin Receptor (InR) in flies initiates an evolutionarily conserved signaling cascade composed of insulin receptor substrate (IRS, Chico), PI3Kinase (DP110) and Akt, inducing cell proliferation and/or growth and endoreplication [35]–[39]. Interestingly, IIS induces ISC proliferation through both cell-autonomous mechanisms involving the Akt-regulated transcription factor Foxo, as well as through a non-autonomous process in which IIS – induced EB differentiation is critical to allow further ISC divisions [7], [34]. In EBs, InR is sufficient and required for differentiation into ECs [34].
In most Drosophila tissues, cell growth is regulated downstream of Akt by the evolutionarily conserved TSC/Rheb/TOR pathway [37], [40], [41]. As supported by genetic and biochemical studies, this pathway can be activated in response to Akt-mediated phosphorylation of Tuberous Sclerosis Complex 2 (TSC2; encoded by the gene gigas in Drosophila) and subsequent inhibition of the TSC1/2 complex [37], [40]. TSC1 promotes the stability of TSC2, which is a GTPase activating protein for the small GTPase Rheb, inhibiting Rheb-mediated TOR Kinase activation. TOR, in turn, phosphorylates translational regulators, including ribosomal protein S6 Kinase (S6K) and eIF4E Binding Protein (4EBP), resulting in a net increase of protein production in cells.
In addition to Akt-mediated phosphorylation, other signals are likely to play an important role in regulating TSC1/2 activity in vivo [42], [43]. Accordingly, various other regulatory events, including phosphorylation in response to several growth factors and morphogens, ubiquitination, and degradation, have been reported to influence TSC1/2 activity [37], [40], [44]. The TSC1/2 complex thus represents a critical node in signaling networks that arbitrate between cell proliferation and growth in response to increased insulin signaling. Supporting this view, mutations in TSC1/2 result in Tuberous Sclerosis Complex, a rare autosomal dominant disease that is characterized by widespread benign tumor formation [45].
Recent studies suggest that TSC/TOR signaling has an important regulatory role in both vertebrate and invertebrate stem cell lineages. In human embryonic stem cells, activation of S6K by mTOR has been reported to induce differentiation [46], while a recent study in the mouse has identified an interesting non-autonomous function for mTOR activity in ISC support cells, the Paneth cells. Under conditions of dietary restriction, TOR signaling activity is reduced in Paneth cells, resulting in secretion of factors that promote stem cell maintenance and proliferation [47]. In the Drosophila germline, TSC/TOR signaling regulates proliferation and maintenance of germline stem cells (GSCs) [48]–[50]. GSCs mutant for TSC1/2 undergo differentiation, through a so far unknown mechanism [49], [50], while GSCs mutant for the TOR kinase exhibit proliferation defects [49]. TSC/TOR signaling is thus likely to mediate, at least partially, the effects of the dietary status of the organism on GSC proliferation and maintenance [48]–[50][4], [51].
In the Drosophila intestine, TSC/TOR signaling may have a similar function, as ISCs are also regulated according to nutrient availability [33], [52]. Indeed, a recent report shows that loss of TSC in ISCs causes excessive ISC growth and impairs ISC proliferation [53]. Using the ISC and EB driver esgGal4, it was shown that TSC2-RNAi expressing ISCs become large, express less cell cycle markers, have reduced DNA replication, and that these phenotypes are Rapamycin-sensitive. These cells further fail to respond to tissue damage by initiating cell divisions, and exhibit increased DNA content, indicating that they are becoming polyploidy [53]. While these characteristics indicated differentiation of TSC deficient cells, it was shown that TSC2-RNAi expressing cells do not express the EC marker Pdm1, and do not form ECs with brushed borders, suggesting that they may have initiated, but not completed, differentiation. TSC-mediated inhibition of TOR signaling thus seems to be critical to maintain ISC activity and function. It remained unclear, however, what physiological role, if any, TOR activation may have in ISCs or their daughter cells, and how Tor signaling may interact with others pathways regulating ISC commitment and differentiation.
Here, to address these questions, we characterize the function and regulation of TSC/TOR signaling in the ISC lineage in more detail. We find that TSC2 is highly expressed in ISCs, but specifically down regulated in EBs. While, consistent with the previous report [53], high TSC2 expression is required for ISC function, we also find that the down-regulation of TSC2 in EBs, and the resulting TOR activation, are critical for EC differentiation. Our results further suggest that TSC activity promotes lineage commitment of EBs into the EE fate.
To characterize the regulation of TSC/TOR signaling in EBs further, we assessed its interaction with the Notch (N) signaling pathway. We find that N-induced Su(H) activity represses TSC2 expression in EBs. Strikingly, repression of TSC1/2 function is sufficient to commit cells into the EC fate independently of N activity, indicating that TSC2 repression is a central step in N-induced EC differentiation.
We also find that food conditions significantly impact the proliferative capacity of TSC-deficient ISCs. We show that TSC mutant ISCs are capable of generating normal clones of daughter cells on a low calorie (low yeast) diet, but that these lineages decline over time. Rearing flies on high-yeast food, however, causes growth and proliferation phenotypes similar to the ones observed in [53], accelerating the decline of TSC mutant clones. TSC activity in ISCs is thus specifically required to maintain ISC function under high nutrient conditions.
While InR/Akt signaling can activate TOR signaling in many Drosophila tissues [35]–[37], [40], [41], previous reports have suggested an opposing role of InR and TOR signaling in the control of ISC proliferation: While Insulin/IGF signaling (IIS) is required for ISC proliferation, and activation of IIS (by InR over-expression) induces increased proliferation [18], [23], [33], [34], activation of TOR signaling (by loss of TSC function) was found to impair proliferative capacity [53]. Using RNAi-based knockdown of TSC2 in ISCs and EBs, Amcheslavsky et al found that loss of TSC2 increased the size of ISCs. Based on cell cycle markers and EdU incorporation experiments, it was concluded that these cells are not mitotically active. Furthermore, the proliferative capacity of TSC2 homozygous mutant ISCs was assessed using lineage tracing by somatic recombination. However, a mitotic recombination with a repressible cell marker (MARCM, [54]) approach was used in which GFP was expressed under the control of esg::Gal4, which labels only ISCs and EBs (see Materials and Methods in [53]). Clones with more than two cells (including ECs and EEs) that may be formed by TSC mutant ISCs (see below) can not be observed with this approach (see for example Fig. 2A in [53]), and a full lineage analysis of TSC deficient ISCs was thus not possible. While the results reported in [53] thus clearly identified a critical role for TSC2 in maintaining small, diploid ISCs, it remained unclear whether activation of TOR signaling in ISCs would fully impair their proliferative activity and prevent generation of ISC daughter cells. Interestingly, the reported results indicated that TSC2 function is required in ISCs for IIS-mediated induction of proliferation, suggesting that IIS activation does not result in inactivation of TSC in the ISC lineage. It further remained unclear whether TOR signaling has to be continuously repressed by TSC2 in the ISC lineage, or whether TOR activation occurs naturally in the lineage to regulate proliferation, growth or differentiation of ISCs or their daughter cells.
To characterize the relationship between InR and the TOR signaling pathway in ISC lineages in more detail, we generated ISC clones with gain- and loss-of-function conditions for multiple IIS and TOR pathway components (Figure 1). We used MARCM to generate ISC clones that over-express wild-type or dominant-negative insulin receptor (InR; [55]) molecules, that were mutant for the IRS homologue Chico (carrying the loss of function allele chico1 [56]), or that were homozygous for the InR loss-of-function alleles InRE19 or InR353 [57]. Similarly, we generated clones with TOR pathway gain- and loss-of-function conditions by over-expressing Rheb or TSC1 and 2, introducing the TSC1 loss of function allele Tsc1Q87X [58], the TSC2 loss of function allele gigas192 [59], the TOR loss of function alleles Tor2L1, Tor2L19 and TorW1251R [60], or the Rheb loss of function allele Rheb2D1 [61], or expressing dsRNA against TSC2 (TSC2RNAi).
Importantly, we used a MARCM approach in which all daughter cells of mutant ISCs are labeled by GFP (since GFP was expressed under the control of tub::Gal4). The vast majority of GFP+ lineages (>95%) in the midgut were induced by mitotic recombination in response to heat shock, as very few marked cell clones could be observed in control animals (Figure S1C). The number of cells in each clone at a given time point after the heat shock thus accurately reflects ISC proliferation.
Consistent with previous reports [18], [23], [33], [34], gain of InR function increased the number of cells produced in a clone, while loss of InR or chico activity significantly reduced the number of cells produced by an ISC in 7 days (Figure 1A, 1B). Loss of TOR pathway activity also reduced clone sizes at 7 days, and increasing TOR pathway activity (in TSC mutants or Rheb over-expressing clones) resulted in clones that showed no significant difference in average cell numbers at 7 day after induction. In contrast to the observations reported in [53], ISCs with increased TOR pathway (i.e. reduced TSC) activity were thus capable of generating normal ISC lineages in our studies. However, the variability in clone sizes increased in TOR gain of function conditions compared to wild-type clones (compare standard deviations in Figure 1B), indicating that, consistent with [53], individual ISCs may lose the ability to generate normal numbers of daughter cells (see below). As expected, we also observed significantly larger Enterocytes (ECs, defined as the largest polyploid, Dl - negative cell in a clone) in both IIS and TOR gain-of-function conditions at 7 days after clone induction, and significantly smaller cells in IIS/TOR loss-of-function conditions (Figure 1C). This is consistent with previous findings in developmental contexts and in GSCs, showing that IIS and TOR signaling act in concert to promote endoreplication and growth [39], [49].
Our results thus support a positive interaction between InR and TOR signaling in the ISC lineage. We tested whether TOR signaling acts downstream of InR in the regulation of proliferation and growth in this lineage by assessing the frequency of mitotic figures and the size of EC nuclei. We co-overexpressed InR with TSC1 and TSC2, or with dsRNA against S6K (S6KRNAi) using the ISC/EB driver esg::Gal4 in combination with the heat-sensitive Gal4 inhibitor Gal80ts (Figure S1A, S1B; TARGET system [8], [9], [62]). InR over-expression using this driver dramatically increases ISC proliferation rates (as represented by the number of phospho-histone H3 (pH3) positive cells [18]) and increases cell sizes in the gut (as represented by the size of EC nuclei; Figure S1A, S1B). Loss of TOR pathway activity (over-expression of TSC1 and 2 or knockdown of S6K) did not affect InR-mediated proliferation, but significantly prevented the increase in EC nuclear size. In these InR gain-of-function conditions, the TSC/TOR/S6K pathway is thus specifically required to promote growth and endoreplication rather than proliferation in the ISC lineage.
Since these results contrasted with the observations reported in [53], we assessed the phenotypes of TSC deficient ISCs in more detail. A timecourse analysis revealed that loss of TSC1 resulted in clones that initially grew faster than wild-type clones, but declined and became heterogeneous in size at later timepoints (Figure 2A and Figure S2A; see large standard deviations in TSC1 mutant clones at 5, 7, and 15 days, and compare with TSC2 mutant clones in Figure 1B). This indicated an initial increase of proliferative activity in TSC mutant ISCs, followed by a sporadic loss of proliferation in individual ISCs at a later timepoint. Indeed, while many Tsc1Q87X or gigas192 mutant clones, or clones expressing TSC2RNAi, were recovered that contained a single diploid Dl+ ISC even at 15 days after clone induction, at all ages rare clones could also be observed in which the Dl+ cell became large and polyploid, consistent with the phenotype reported by Amcheslavsky et al (Figure S2B, S2C). In our experiments, TSC mutant ISCs did thus not immediately increase in size, but grew and lost function sporadically. This interpretation is supported by the fact that the number of Tsc1Q87X mutant clones observed in the gut declined over time (Figure 2B).
To explore why TSC mutant ISCs exhibited a much less penetrant growth phenotype in our experiments as compared to [53], we tested whether the rate of the spontaneous growth of TSC mutant ISCs might be influenced by dietary conditions, which can modulate TOR activity independently of TSC [35]–[39]. Indeed, the average number of cells generated by Tsc1Q87X mutant ISCs within 7 days was significantly reduced when flies were reared on high yeast food (HY, 15% yeast) compared to our regular food (RF, 2% yeast)(Figure 2C, 2D). Dl+ cells in these Tsc1Q87X mutant clones became large and polyploid, similar to the phenotype described in [53](Figure 2E). Clone sizes were also reduced in wild-type flies reared on high yeast food, but this reduction was less significant than the size reduction of TSC deficient clones (Figure 2D).
Two recent studies have reported strong effects of yeast, the only protein source in fly food, on ISC activity. Both studies reported increased ISC activity in yeast-fed flies compared to flies completely starved of yeast [33], [34]. ISCs thus require a protein source to become fully active, yet our results indicate that they can also lose function when protein levels are too high. This effect is significantly enhanced when TSC is lost, indicating that TSC activity isolates the TOR pathway from dietary stimuli in ISCs, maintaining their function. The role of the TSC1/2 complex in ensuring the long-term maintenance of ISCs is thus reminiscent of its function in GSCs [48]–[50].
Interestingly, Tsc1Q87X mutant, or TSC2RNAi or Rheb over-expressing clones were significantly less likely to contain prospero-labeled EE cells than wild-type clones, suggesting that TOR activation also impaired the commitment of EBs into the EE cell fate, or the terminal differentiation of EEs (Figure 2F, 2G). It remains unclear, however, whether this is a consequence of direct TOR pathway-mediated regulation of prospero expression, or of other events required for EE differentiation.
Our observations thus suggest that the TSC complex promotes ISC maintenance in varying nutritional conditions, influences commitment into the EE fate, and regulates EC growth in the intestinal epithelium.
We hypothesized that these multiple functions of TSC are coordinated by intricate, cell-type specific regulation of TSC activity in the ISC lineage. To start analyzing this regulation, we examined the expression of TSC2 using an anti-Gigas antibody described in [44]. High expression of TSC2 was detected in ISCs (Dl+ cells that do not express GFP under the control of the RU486-inducible EB/EC driver 5966::GS [63], Figure 3A) and in EEs (pros+ cells, Figure 3C. These cells show even higher TSC2 expression than ISCs), and its expression was significantly weaker in EBs (cells expressing bGalactosidase from a Su(H)-GBE::lacZ construct, Figure 3B). Consistent with this expression pattern of TSC2, we found that in wild-type homeostatic conditions, the TOR pathway is highly active in EBs (compared to ISCs or ECs), as determined using an antibody against phosphorylated 4EBP (Figure 3D, this antibody reliably detects changes in TOR signaling activity, see S3A and [49]). Preventing this activation of TOR signaling in EBs was sufficient to impair the formation of normal EBs: over-expression of TSC1/2 or knockdown of S6K (S6KRNAi) specifically in EBs and ECs (using 5966::GS), resulted in the accumulation of small Dl+ cells that also express GFP (Figure 3E). Most of these cells had DNA content that was similar to ISCs, indicating that they are diploid or have not completed endoreplication (Figure S3B). The disruption of the normal asymmetric distribution of Dl in ISC/EB pairs indicates that TOR inactivation in ISC daughter cells inhibits differentiation. A similar disruption of normal EB determination was observed when TSC1/2 were over-expressed in EBs only using Su(H)-GBE::Gal4 [64] (Figure 3F, 3G). Interestingly, these guts also exhibited a significant increase in the number of pros+ EE cells, indicating that inhibiting TOR activity in EBs is sufficient to alter their commitment from the EC fate into the EE fate (Figure S3C).
Combined, our findings suggested that reduced TSC1/2 function in EBs is critical for differentiation of EBs and for lineage commitment into the EC fate. Importantly, these findings also suggested a potential mechanism for TSC2 regulation in the ISC lineage, as down-regulation of TSC2 expression coincides with the activation of N signaling in EBs. We hypothesized that N activation promotes TSC2 down-regulation and tested this idea by over-expressing the N Intracellular Domain (NICD) in ISCs and EBs (using esg::Gal4, Gal80ts). Expression of NICD is sufficient to force differentiation of ISCs into ECs [8], [9]. Consistently, we found that TSC2 expression was undetectable in most esg::GFP+ cells expressing NICD, while in wild-type intestines, more than 50% of all esg::GFP+ cells express high levels of TSC2 (Figure 4A, 4B). We further tested whether N signaling is required for TSC2 repression in the ISC lineage by over-expressing a dsRNA against N (NRNAi) in ISCs and EBs. Expression of NRNAi under the control of esg::Gal4 prevents EB differentiation and results in the formation of ISC tumors characterized by clusters of small, diploid, Dl+ cells [8], [9]. Cells in these tumors were also TSC2 positive, confirming the correlation between ISC identity and TSC2 expression, and suggesting that N signaling is required for TSC2 repression (Figure 4C; TSC2 immunoreactivity was suppressed by TSC2RNAi and enhanced by over-expressing both TSC1 and 2, confirming the specificity of the antibody. Co-expression of TSC1/2 also moderately increased the size of the stem cell tumors, indicating additional enhancement of the NRNAi-caused phenotype).
The N-responsive transcriptional regulator Su(H) has been reported to bind to a cluster of four sites within 1.5 kb in the upstream promoter region of the gigas/Tsc2 gene in Drosophila [65]. Su(H)-mediated transcriptional repression of gigas/Tsc2 was thus a plausible mechanism for N-induced repression of TSC2 expression in EBs. To test this idea, we assessed the regulation of gigas/Tsc2 in co-cultures of S2 cells that constitutively express N or Dl [66](Figure 4D–4E; N activation in N-expressing cells occurs within minutes of exposure to Dl-expressing cells, Figure S4). We first confirmed that Su(H) binds to the upstream promoter region of gigas/Tsc2 using chromatin IP (ChIP, Figure 4D), and found significant enrichment of a region proximal to the transcriptional start site in precipitates from cells with activated N signaling. We further measured transcript levels of gigas/Tsc2 and found reduced expression of this gene within 30 min of N activation (Figure 4E). This repression of gigas/Tsc2 was sustained for at least 24 hours. Protein levels of TSC2 (measured by Western Blot) did not decrease significantly in S2 cells in these experiments (not shown), indicating that in addition to transcriptional repression, posttranslational mechanisms have to be involved in reducing TSC2 protein levels in vivo as observed in ISCs expressing NICD (Figure 4A, 4B). Importantly, these results suggested that Su(H) is a general transcriptional repressor of gigas/Tsc2 expression in Drosophila cells. Accordingly, TSC2 repression in EBs was mediated by Su(H), as inducing ‘Flp-out’ clones [67] expressing Su(H)RNAi was sufficient for the formation of tumors containing small Dl+ and TSC2+ cells (Figure 4F). Consistently, gigas/Tsc2 repression in the S2 co-culture system was prevented when Su(H) was knocked down by RNAi (Figure 4G).
These findings are consistent with a model in which N activation suppresses TSC2 expression in EBs, inducing growth and endoreplication in response to insulin signals. We asked whether TSC2 repression was sufficient and required for ISC differentiation downstream of N, and found that loss of TSC1/2 indeed rescued the tumor phenotype of NRNAi expressing ISCs (in both MARCM clones, and when driven by esg::Gal4; Figure 5, Figure S5). N-deficient ISCs generate tumors because they undergo symmetric divisions and thus generate exponentially growing cell clones. Loss of TSC1/2 prevented this accumulation of Dl+ ISCs in N loss of function conditions and converted NRNAi expressing cells into Dl−, polyploid, EC-like cells. Similar to wild-type ECs, these cells also contained brush borders and expressed the EC marker Pdm1 (Figure 5A–5E, Figure S5A–S5C; brush borders can be observed by staining for phalloidin; Polyploidy measured by intensity of DAPI fluorescence). TSC1 suppression is thus sufficient to fully differentiate N-deficient ISCs into ECs. Consistent with a conversion of these cells into a postmitotic state, the number of cells observed in each cluster of NRNAi expressing cells was significantly reduced when TSC1 or 2 were lost (Figure 5E, 5F).
These results confirm that loss of TSC1/2 in N loss-of-function conditions is sufficient to promote differentiation of ISCs towards the EC fate. For most analyzed phenotypes, inhibition of TSC1 elicited stronger effects than inhibition of TSC2, suggesting that the knockdown of TSC2 is less efficient, or reflecting the fact that loss of TSC1 also results in degradation of TSC2 protein, as TSC1 stabilizes TSC2. While many TSC1/Notch double mutant cells thus are morphologically indistinguishable from wild-type ECs, it is important to note, however, that Notch activation elicits complex gene expression changes in cells, and it remains unclear whether all functional aspects of ECs can be reconstituted in N/TSC1/2 deficient cells. Furthermore, some of these cells retain Dl expression (see example in Figure 5E), indicating that not all of these cells fully differentiate into normal ECs.
Loss of TSC1/2 also rescued the accumulation of pros+ EE cells in N loss-of-function conditions (Figure S5D, Figure S5E), confirming a shift towards the EC fate in TSC-deficient EBs. Furthermore, co-expression of TSC1 and 2 resulted in the maintenance of small, diploid, Dl+ cells even in the presence of NICD, showing that TSC2 repression is required for N-induced ISC differentiation (in both Flp-out clones and when driven by esgGal4, Figure 6). Repression of TSC1/2 function is thus a critical step in the regulation of EB differentiation.
Our results establish a new mechanism by which lineage commitment, differentiation and growth are coordinated in an epithelial stem cell lineage (Figure 7). This mechanism allows for the integration of nutritional signals through the IIS and TOR pathways with Notch-mediated differentiation signals: High expression of TSC2 in ISCs prevents differentiation and is thus critical for stem cell maintenance, while reducing TSC activity in EBs is required and sufficient to promote differentiation into ECs. This dynamic regulation of TSC levels in the ISC lineage intersects with the control of TSC1/2 activity by growth factor signals. Based on current models, we propose that control of TSC2 expression is required to set a threshold for the Akt-mediated inactivation of the TSC1/2 complex downstream of growth factor receptors. The TSC2 expression level would thus determine the cellular response to growth signals in the ISC lineage. Supporting this view, ISCs, which express high levels of TSC2 constitutively, do not differentiate in response to InR over-expression, but rather increase their proliferation rate. EBs, on the other hand, express less TSC2 and respond to InR activation by endoreplicating and growing into ECs.
Robust expression of TSC2 in ISCs thus prevents premature differentiation and growth of ISCs. When IIS is chronically activated (as in high nutrient conditions), however, Akt-mediated TSC1/2 complex inactivation may cause sporadic differentiation and loss of ISCs. Conversely, ISC maintenance might be improved in conditions of chronically low IIS and TOR activity. The TSC1/2 complex thus acts as a ‘buffer’ that improves ISC maintenance by isolating these cells from changing nutritional conditions. Accordingly, TSC1 deficiency leads to differentiation and loss of ISC function when flies are reared under high yeast conditions. Interestingly, these results also indicate that TOR pathway activation is not constitutive in TSC deficient ISCs, but is still inducible by nutritional changes. It is likely that amino acid-sensing signaling pathways, involving Rag GTPase complexes and the MAP4K3 and Vps34 kinases, regulate TOR in these situations [38].
An effect of TOR signaling on stem cell maintenance has previously been described in GSCs, and is consistent with recent findings that suppression of TOR activity, through rapamycin or genetic means, increases lifespan [48]–[50], [68]. While reducing IIS activity in the ISC lineage is sufficient to extend lifespan [23], additional studies will have to be performed to assess the role of TOR signaling in this context.
Three pieces of evidence indicate that transcriptional repression of TSC2 occurs downstream of N activation in the ISC lineage: (i) TSC2 expression is significantly reduced in EBs with high levels of N signaling activity, (ii) forced expression of NICD suppresses TSC2 expression in ISCs, and (iii) loss of TSC2 is sufficient to rescue N loss of function phenotypes in the ISC lineage. This regulatory interaction between TSC2 and N signaling is reminiscent of recent findings in Drosophila sensory organ precursors, mouse embryonic fibroblasts, and mammalian cancer cells [46], [69]–[73]. However, the results reported in these studies indicated that TOR pathway activation could result in increased N cleavage and N pathway activation. In the fly SOP, activation of the TOR pathway phenocopied N gain of function phenotypes, but it was not tested whether these phenotypes were rescued in N loss of function backgrounds. It thus remained unclear whether N activation is a consequence of TOR pathway activation, or whether TOR activation is a required component of the N-induced differentiation pathway in this lineage. Our results demonstrate that in the ISC lineage, TOR activation is sufficient to drive EC differentiation even in the absence of N signaling, supporting a model in which TOR activation occurs downstream of N. In TSC mutant mouse embryonic fibroblasts, on the other hand, TOR -dependent activation of N can be observed, highlighting the close, evolutionarily conserved relationship between these two signaling pathways in the control of cell differentiation, but also suggesting that multiple, context dependent, interaction mechanisms may exist [69].
Our model implies a novel role for Su(H) as a transcriptional repressor of the gigas gene. This interpretation is based on the requirement of Su(H) for the N-mediated repression of the gigas gene both in S2 cells and in the ISC linage, as well as on the binding of Su(H) to the gigas promoter. A function of Su(H) as a transcriptional repressor has not previously been described, and additional studies are needed to explore its mechanism. While binding of Su(H) to the gigas promoter indicates a direct role in transcriptional repression of gigas, it is possible that Su(H) also acts indirectly to repress gigas expression by inducing or cooperating with transcriptional repressors. A candidate group of such repressors are encoded by Su(H) target genes, the classical Hairy and E(Spl) complex. These transcription factors are induced by Su(H) in response to Notch signaling and have been described as transcriptional repressors in other contexts [74]. Putative E(Spl) binding sites are present in the gigas promoter (not shown), and additional studies will therefore be of interest to dissect the requirement for individual E(Spl) complex genes in the regulation of gigas.
Our data further indicate that transcriptional repression of gigas may not be the only mechanism by which TSC2 repression is achieved in EBs. While we find that activation of N is sufficient and required for repression of TSC2 protein in EBs in vivo, our S2 studies indicate that the turnover rate of the TSC2 protein also has to be increased to achieve rapid reduction of TSC2 levels. It can be anticipated that the control of TSC2 ubiquitination by the cul4/ddb1/fbw5 complex may be an important regulatory mechanism here, and it will be of interest to further dissect the interaction of this complex with the N signaling pathway in the ISC lineage [44].
Characterizing these signaling interactions in ISCs in more detail is of significant interest for our understanding of somatic stem cell maintenance, proliferative homeostasis and lineage commitment. The evolutionary conservation of N and TOR signaling, as well as the similarities in the biology of Drosophila and vertebrate stem cell populations [7], indicate that such understanding will provide important insight into human regenerative and proliferative diseases.
The following fly stocks were obtained from the Bloomington Drosophila Stock Center: w1118, UAS-InR, UAS-InRDN, UAS-Rheb, UAS-S6KKQ, tub-Gal80ts, UAS-SuHRNAi (TRiP.HM05110). UAS-TSC1RNAi (Transformant ID 22252), UAS-TSC2RNAi (TID 103417) and UAS-S6KRNAi (TID 104369) were obtained from the Vienna Drosophila RNAi Center. The following lines were gifts from: Esg-Gal4, S. Hayashi; UAS-NICD, UAS-NotchRNAi, and hsFlp; tub-Gal4, UAS-GFP; FRT82B tubGal80, N. Perrimon; Su(H)-GBE-LacZ, S. Bray; Su(H)-GBE-Gal4, S.X. Hou; UAS-TSC1, TSC2, M. Tatar; FRT40A, chico1, FRT82B, InRE19, and FRT82B, InR353, and FRT40, Tor2L1, FRT40, Tor2L19 and FRT40, TorW1251R by D. Drummond-Barbosa; hsFlp; FRT40A, tub-Gal80; tub-Gal4, UAS-GFP and 5966-GS, B.Ohlstein; w, hsFLP; actin, FRT, y+, FRT, Gal4, UAS::RFP, M. Uhlirova; FRT82, Tsc1Q87X and FRT82, Rheb2D1, K. Harvey.
Flies were cultured on yeast-molasses based food at 25°C with a 12 hours light/dark cycle. For TARGET experiments flies were raised at 18°C and shifted to the restrictive temperature (29°C) 3–5 days after eclosion. For clone induction (MARCM and Flp-out), 3–5 day old flies were heat shocked at 37°C for 45 minutes. For 5966GS, flies were maintained for 7 days on RU486 food (100 µl of a 5 mg/ml solution of RU486 was deposited on top of a 10 ml food vial and dried for 16 hours).
Guts were dissected in phosphate-buffered saline (PBS) and fixed for 45 min at room temperature in 100 mM glutamic acid, 25 mM KCl, 20 mM MgSO4, 4 mM sodium phosphate, 1 mM MgCl2, and 4% formaldehyde. All subsequent washes (1 hour) and antibody incubations (4°C overnight) were performed in PBS, 0.5% bovine serum albumin and 0.1% Triton X-100.
Staining with Delta antibody was performed following the methanol-heptane fixation method described in (Lin et al., 2008).
Fluorescent in situ hybridization protocol was adapted from [75] using Tyramide signal amplification (TSA) and Digoxigenin (DIG) labeled RNA probes. The following primers were used to generate RNA probes for pdm1: F 5′-AGT TTG CCA AGA CCT TCA AGC AGC and R 5′-AGG GAT TGA TGC GCT TCT CCT TCT.
Primary antibodies with respective dilutions were: From Developmental Studies Hybridoma Bank: mouse anti- Armadillo and anti-Delta, 1∶100; mouse anti-Prospero, 1∶250; Cell Signaling: rabbit anti-phospho-4EBP, 1∶500; ICN: mouse anti-b-galactosidase, 1∶100; gift from Yue Xiong: rabbit anti-Gigas, 1∶500; gift from Yang Xiao-Hang: rabbit anti-Pdm1; Upstate Biotech: rabbit anti phospho histone H3, 1∶1000; Invitrogen: Alexa Fluor 568 1∶500
Confocal microscopy was performed on a Leica SP5 system. Image processing was done on NIH Image J and Adobe Photoshop.
S2 cell lines stably transfected to express wild type Notch receptor (S2-Mt-N) or Delta ligand (S2-Mt-Dl) from a Cu-inducible metallothionein promoter were obtained from the Drosophila Genomic Resource Center. Both lines were cultured in M3+BPYE medium with 10% heat inactivated Fetal Calf Serum and grown under permanent selection with 0.2 µM Methotrexate (Sigma). N and Dl expression was induced separately with 600 mM CuSO4 for 24 hours and the two cell lines were then co-cultured in 1∶1 ratio for the indicated times.
For RNAi experiments, double-stranded RNAs were synthesized against GFP and Suppressor of Hairless using T7 promoters (Ambion MEGAscript RNAi kit). S2-Mt-N and S2-Mt-Dl cells were cultured separately and only S-Mt-N cells were treated with GFP dsRNA (control) or with SuH dsRNA for 3 days. After two days of dsRNA treatment both cell lines were induced with 600 mM CuSO4 for 24 hours. After three days of dsRNA treatment, S2-Mt-N cells and S2-Mt-Dl cells were co-cultured at a 1∶1 ratio for the indicated times. Knockdown of Su(H) was confirmed by RT-PCR (not shown).
∼1×107 cells were collected from triplicate cell cultures of S2-Mt-N cells (control) or of 2 h co-cultures of S2-Mt-N and S2-Mt-Dl. Cells were cross-linked using ∼1.1% formaldehyde and ChIP was performed using the abcam ChIP kit (ab500). Cells were sonicated on ice using a Branson Sonicator (power 4, 50×10 second pulses with 30 second intervals; average size of genomic DNA fragments was ∼500 bp). Sheared chromatin was incubated with 5 µg of rabbit anti-GFP (invitrogen; negative control) and goat anti-SuH (Santa Cruz Biotechnology) for 24 hours and then precipitated using Protein G Sepharose (Fast Flow; Sigma). De-crosslinking and DNA purification was performed according to kit instructions (ab500). DNA from different ChIP samples was analyzed for enrichment using real time PCR using the following primer sets: gig 1: 5′-ACAAACGCAAAGTTGGCGAC-3′ and 5′-GTGTGCAACCAGTAATTCCTAGCC-3′; gig 2: 5′-AAGTTGTTCCTCAAATCGCTGCCG-3′ and 5′-ATTGAAGTTGTGCAGCTGCGTGTC-3′; actin5C: 5′-ATTCAACACACCAGCGCTCTCCTT-3′ and 5′-ACCGCACGGTTTGAAAGGAATGAC-3′.
Total RNA was extracted using Trizol. cDNAs were synthesized using oligo-dT primers and real-time RTPCR was performed on a BioRad iQ5 detection system (using SYBR Green and ΔΔCt quantification method). Gigas and Suppressor of Hairless expression levels were quantified relative to Actin5c expression.
Cell samples were resolved using 5% (for NICD) or 10% (for tubulin) SDS-polyacrylamide gel electrophoresis, transferred to nitrocellulose membranes using semi-dry transfer, and probed with the following primary antibodies: mouse anti-NICD (DSHB, 1∶10,000), mouse anti-alpha-tubulin (Sigma, 1∶5000). Antibodies were detected using horseradish peroxidase-conjugated secondary antibodies and the ECL detection system (Amersham).
|
10.1371/journal.pbio.1000622 | Clusters of Nucleotide Substitutions and Insertion/Deletion Mutations Are Associated with Repeat Sequences | The genome-sequencing gold rush has facilitated the use of comparative genomics to uncover patterns of genome evolution, although their causal mechanisms remain elusive. One such trend, ubiquitous to prokarya and eukarya, is the association of insertion/deletion mutations (indels) with increases in the nucleotide substitution rate extending over hundreds of base pairs. The prevailing hypothesis is that indels are themselves mutagenic agents. Here, we employ population genomics data from Escherichia coli, Saccharomyces paradoxus, and Drosophila to provide evidence suggesting that it is not the indels per se but the sequence in which indels occur that causes the accumulation of nucleotide substitutions. We found that about two-thirds of indels are closely associated with repeat sequences and that repeat sequence abundance could be used to identify regions of elevated sequence diversity, independently of indels. Moreover, the mutational signature of indel-proximal nucleotide substitutions matches that of error-prone DNA polymerases. We propose that repeat sequences promote an increased probability of replication fork arrest, causing the persistent recruitment of error-prone DNA polymerases to specific sequence regions over evolutionary time scales. Experimental measures of the mutation rates of engineered DNA sequences and analyses of experimentally obtained collections of spontaneous mutations provide molecular evidence supporting our hypothesis. This study uncovers a new role for repeat sequences in genome evolution and provides an explanation of how fine-scale sequence contextual effects influence mutation rates and thereby evolution.
| An intriguing observation made during the comparison of genomes is that insertion and deletion mutations (indels) cluster together with nucleotide substitutions. Two (not mutually exclusive) hypotheses have been proposed to explain this phenomenon. The first postulates that an indel mutation causes an increase in the likelihood of the surrounding sequence incurring nucleotide substitutions, while the second claims that the region of DNA in which such a cluster is located is more likely to sustain both indels and substitutions. Here, we present evidence suggesting that the region of DNA, and not the indel, is associated with the accumulation of clusters of mutations over evolutionary time scales. We find that repeat sequences are closely associated with a large proportion of indels and that the abundance of repeat sequences is linked with regions of increased nucleotide diversity. By analysing molecular data and measuring the mutation rates of genes engineered to contain repeats, we find that the mutation rate can be manipulated by the insertion of long repeat sequences. On the basis of these results, we propose a model in which repeat sequences are prone to cause stalling of the high-fidelity DNA polymerase, leading to the recruitment of error-prone repair polymerases which then replicate the surrounding sequence with a higher-than-average error rate.
| A major challenge of evolutionary genetics is to determine the mechanisms underlying cryptic patterns of mutation rate variation and how they influence evolutionary outcomes [1]. One of the most striking of these trends is the association between indel mutations and nucleotide substitutions [2]–[7]. Inter-species genome comparisons have revealed this trend to be universal to all prokaryotic and eukaryotic genomes examined thus far [4]–[6]. The prevailing explanation for this association is that indels, as “universal mutators” [4], cause the accumulation of nucleotide substitutions in the hundreds of base pairs of sequence surrounding the indel [4],[6]. Although such studies have been unable to unequivocally determine if the clusters are due to a single multimutational event (multiple mutation hypothesis), the indel per se (the mutagenic indel hypothesis), or the region of sequence in which the indel is found (the regional differences hypothesis), the mutagenic indel hypothesis has been adopted by workers in the field [8]–[12].
The mechanism of indel mutagenicity proposed by Tian and co-workers is that indels, when heterozygous, cause paired chromosomes to form heteroduplex DNA during meiosis [4]. This is posited to cause error-prone DNA repair systems to target indel-containing regions, leading to an increased likelihood of nucleotide substitution in the sequence surrounding the indel. Over time, this increase in mutation rate is predicted to leave as its signature the clustering of nucleotide substitutions in the DNA surrounding indels, while corresponding non-indel-containing orthologous sequences should have a lower number of substitutions, in accordance with the background substitution rate. In addition, because the proposed mutagenic effect of the indel is postulated to be dependent on its heterozygosity, the accumulation of substitutions should cease as soon as the indel becomes homozygous in the population. These predictions contrast with the regional differences hypothesis; regional effects are predicted to cause both indel and non-indel haplotypes to accumulate substitutions whether the indel is heterozygous or not. The multiple mutations hypothesis differs from both the regional and indel hypotheses in that clusters of mutations are due to a one-off mutation event. Determining whether mutations have accumulated over time or are due to a single mutation event is difficult without the ability to examine indel divergence on a temporal scale.
Here we use a population genomics approach to tease apart the dynamics of indel divergence using the genomes of Escherichia coli, Saccharomyces paradoxus (S. paradoxus), Drosophila, and humans. We show that it is not the indel but rather the sequence region in which the indel occurs that is associated with the accumulation of nucleotide substitutions over evolutionary time scales. We propose a mechanism whereby a DNA sequence that is prone to cause replication fork stalling causes the recurrent recruitment of error-prone DNA polymerases to certain DNA sites, resulting in an increased likelihood of nucleotide substitutions in the surrounding DNA sequence.
To initiate our investigation into the mechanisms underlying indel-associated mutation, we used a unique population genomics resource: 20 high-quality genomes of the Escherichia/Shigella complex ranging from 0.1% to 2.5% sequence divergence (Table S1A). Employing this range of evolutionary distances facilitates capture of the incipient stages of indel divergence, minimizing the obscuring effect of time unavoidable during analyses of more diverged species. DNA replication and repair in E. coli are well understood and, due to their central and conserved role in all living cells, have provided a useful model for eukaryotic systems [13].
Alignments were created between orthologous regions of pairs of E. coli genomes totalling 96.3 Mb, uncovering 5,390 indels. We then performed stringent tests to ensure that results were not due to artefacts of the alignment process (see Materials and Methods). Following Tian et al. [4], we generated estimates of overall nucleotide diversity, D, (D = 0.01 is equivalent to 1% divergence) and plotted the magnitude of D against sequence intervals of defined distance (designated as windows 1, 2, 3, etc.) from the nearest indel (Figure S1). Figure 1A shows an increase in nucleotide divergence in the sequence window closest to the indel (window 1) for all of the E. coli strain comparisons.
The detection of indel-associated mutation in bacterial species poses a dilemma for the mutagenic-indel hypothesis. Prokaryotes are haploid; following the indel-causing event, the cell has only a brief heterogenote period during which, according to the mutagenic-when-heterozygous hypothesis, the indel is mutagenic. After a few cell divisions, the daughter cell will produce only indel-containing copies of the genome and will not have a non-indel version to recognize that the indel is present (Figure 2). The mutagenic-when-heterozygous theory then predicts (at least in prokaryotes) that nucleotide diversity does not accumulate over time. To test this prediction, we generated pre-defined, non-overlapping sets of old and new indels in E. coli. Old indels are those determined (using an appropriate outgroup) to have occurred before the divergence of the two strains under comparison; new indels are those that have occurred after their divergence (Materials and Methods, Figure S2). As shown in Figures 3A and S3, D values are significantly higher for old indels (black lines) than those for new indels (grey lines). This result demonstrates that, contrary to the mutagenic-when-heterozygous and multiple mutation hypotheses, mutations are accumulating at a higher rate in regions surrounding indels over time.
Background D (Db) is the average difference in the DNA sequences of two aligned orthologous regions. An increase in the number of differences between the nucleotide sequences of two aligned orthologous regions above this average indicates an increase in the rate of the accumulation of substitutions. The mutagenic indel hypothesis states that the indel per se is the cause of an increase in mutation rate and the accumulated nucleotide diversity in the surrounding sequence. A consequence of this is that, of two aligned fragments of DNA, the indel-containing fragment should have a highly elevated D close to the indel and its corresponding non-indel-containing orthologous fragment should have a D equivalent to the background. These predictions can be tested by choosing an orthologous sequence from a third E. coli genome as an outgroup to infer the ancestral state of the aligned sequence, thus allowing us to pinpoint in which of the two aligned genome fragments the indel event has occurred. This is dependent on the assumption of parsimony—if indels are a convergent character, the indel haplotype could be mistakenly assigned. D can be calculated for the sequence windows surrounding an indel-containing region (the indel haplotype) and the corresponding orthologous region without the indel (the non-indel haplotype) with which it is paired. In order to minimize the bias caused by differences in the selective constraints upon aligned sequences, we employed stringent filters to ensure that the sequences compared are strictly orthologous (see Materials and Methods). Figure 3D shows that the values of D for both the indel- and non-indel-containing haplotypes, Di and Dni, are elevated in window 1 as compared to the background nucleotide diversity Db. Although the values of Di in window 1 are often higher than Dni (an average 14% difference in D), this was not significant (two-sample Kolmogorov-Smirnov test, p>0.05, Table S2) for any of the strains compared. By contrast, when Di and Dni are compared to Db, in five out of six comparisons Di is significantly greater than Db (an average 57% difference in D), while Dni is significantly greater than Db in four cases (two-sample Kolmogorov-Smirnov test, p <0.05, Table S2; average 40% difference in D). Thus, for nearly as many instances as the indel haplotype, the non-indel haplotype has a D significantly higher than the background nucleotide divergence, confirming that the regional effect plays a role in the accumulation of nucleotide substitutions.
These results raise the possibility that the accumulation of mutations surrounding indels (Figure 3C) is mainly due to regional effects and not attributable to indels per se. However, this conflicts with the inferences of previous studies [2],[4],[6], that concluded that indels, not regions, are mutagenic. In order to find the cause of this disagreement, we took a closer look at the results of those studies as well as our own data. We noticed that the strains that are less diverged tended to have the largest difference between the indel and non-indel haplotypes (Table S2, Figure S4). Indels detected in the comparisons of two highly similar strains must have happened since their relatively recent divergence. The fact that the more diverged strains differed less between the indel and non-indel haplotypes suggests that the indel-associated effect diminishes over time. When we studied the results of [4] and [6], we found the same trend. For example, using data from [6], when bacterial divergence was plotted against difference between Di and Dni, it showed that the difference between Di and Dni decreases with increasing divergence (Figure S4). A further example is provided by Tian et al.'s [4] analysis of heterozygote alleles at one-third and two-thirds frequencies in yeast. The mutagenic-when-heterozygous mechanism predicts that indels occurring at a higher frequency in a population have been accumulating mutations for longer periods and should thus have a higher D value and a greater difference between Di and Dni. Conversely, the indels at two-thirds frequency have a smaller Di/Dni (1.40) than the indels at one-third frequency (2.23). The fact that indels that have been segregating for longer time have a smaller difference between the indel and non-indel haplotypes indicates that spending more time as a heterozygote actually diminishes the indel-associated effect, contrary to the prediction of the mutagenic-when-heterozygous hypothesis.
The separation of D into Di and Dni allows us to calculate the proportion of D on the indel haplotype that can be attributed to the indel effect and to the regional effect, respectively (see Materials and Methods). Under the assumption that indel-causing events are uniformly distributed since the time of divergence, it follows that the level of divergence between two strains is correlated with the average age of the indels found during comparison. If an indel constantly influences the accumulation of nucleotide substitutions in the surrounding sequence while polymorphic, we expect to see an increase in the difference between Di and Dni over time. Conversely, if indels have a one-time-only effect on nucleotide diversity, we expect to find a decline in this difference over time. We compared Di and Dni for alignments identifying new and old indels (Materials and Methods, Table S3). Figure 4A shows that the difference between Di and Dni decreases with increasing divergence (Pearson's correlation coefficient, r = −0.769, p = 0.0093). This negative correlation is striking when compared to the positive correlation between time since divergence and nucleotide diversity when the indel and region effects are not separated (Figure 3C, Pearson's correlation coefficient r = +0.711, p = 0.00092). This result suggests that it is the region, but not the indel, that is constantly influencing the accumulation of substitutions over evolutionary time scales.
To test whether the aforementioned phenomenon is specific to prokaryotes, we carried out analogous indel analyses using the budding yeast Saccharomyces paradoxus. This organism is suitable for analysis because genome sequences are now available for a variety of its strains [14] and because S. paradoxus, like many multicellular eukaryotes, spends most of its life as a diploid [15]. The results of the analyses with S. paradoxus (Figures 3B, 3E and 4B, Table S3) were in agreement with those obtained using E. coli sequences. The S. paradoxus strains used here (Table S1B) cover a wider range of divergence than the E. coli strains [16]; this allowed us to view the diminishing proportion of the indel-dependent component of D on a longer time scale (Figure 4B, Pearson's correlation coefficient r = −0.963, p = 0.008). We then extended our analysis to Drosophila species (Figure 4C) (see Materials and Methods). Although few species diverged recently enough to be suitable for analysis, the results corroborate our prior findings that the proportion of D attributable to the indel decreases over time (Pearson's correlation coefficient r = −0.980, p = 0.128). It should be noted that the ratio of (Di − Db)/(Dni − Db) was calculated for several yeast and fly alignments with greater divergence than shown in Figure 4; in all cases, this ratio was approximately one (Table S3). All these results suggest that a difference between the indel and non-indel haplotype exists following the indel-causing event but that this difference decreases over time until stabilising with both haplotypes having the same amount of nucleotide diversity. Because our study is able to track indel divergence within a species, this analysis provides unequivocal evidence that nucleotide diversity associated with indels decreases over time.
Mutations arise from inaccurate processing of DNA damage or errors incurred during DNA replication. E. coli possesses five DNA polymerases of which two, Pol IV and Pol V, are error-prone. These polymerases are recruited to stalled replication forks [17],[18] and double-strand breaks [19] to restart DNA replication. Errors made by DNA Pol IV are biased towards frameshifts [20], and though genomes exhibit a bias towards transitions [16], DNA Pol V most often causes transversion mutations [21]–[23]. We analysed the ratio of transition to transversion changes for all aligned E. coli genomes and found that transversions are enriched close to indel and non-indel haplotypes (two-sample Kolmogorov-Smirnov test, p <0.0001) (Figure 5); this is also true for S. paradoxus and other eukaryotes [4]. The accumulation of mutations at a specific site at a higher rate is uncharacteristic of mutations caused by a mutagenic chemical or another random event and is most likely due to the persistent recruitment of error-prone polymerases to that site over evolutionary time. Impediments imposed by polynucleotide repeats or other repeat sequences are suggested to be common causes of DNA replication fork arrest [24]. We performed a computational analysis on the 20 bp immediately flanking our collection of E. coli, S. paradoxus, and Drosophila indels to determine the distribution of repeats around indels. We defined an indel as contiguous with a repeat if it occurred inside or immediately next to a repeat, and as repeat-proximal if some part of a repeat was positioned within 5 bp on either side of the indel. For E. coli, 43% of indels were contiguous with a homopolymer, while 20% were proximal. The corresponding numbers were 45% and 25% for yeast and 31% and 34% for flies, respectively (Figure 6A).
The association between repeat sequences and indels is well understood: repeat sequences are prone to sustain strand slippage mutations [25],[26], which tend to cause indels [19],[27]. We propose a mechanism distinct from strand slippage for the regional increase in nucleotide substitutions, whereby repeat sequences and other polymerase-stalling motifs persistently cause the recruitment of error-prone DNA polymerases. Each time DNA replication is restarted by an error-prone polymerase, DNA surrounding the region will be synthesized with a higher rate of error [17],[18],[28], leading to an increased likelihood of nucleotide substitution. The stalled fork also suffers a high rate of double-strand breaks, another route to error-prone repair [19],[27],[29]. The 3R hypothesis predicts that regions of a genome with increased sequence diversity should be able to be identified by repeat sequence abundance. We tested this prediction by using the recently sequenced genomes of three E. coli strains that we had previously not analysed. We searched for repeat-rich regions by first generating pairwise alignments as for our indel analysis, dividing these into non-overlapping 100-bp windows, and then binning each window according to its number of 4-nucleotide homopolymer repeats (see Materials and Methods). We found that, even when indel-containing windows were excluded, windows with a higher number of repeat sequences had more nucleotide substitutions than those without (83% increase for SE11/REL606 and 71% increase for SE15/REL606 in windows with six repeats). As for indel-based analyses, the more diverged two-strain comparison had a higher value of D, supporting that repeats cause the accumulation of substitutions over time (Figure 6B). We also found that the number of transversions relative to transitions was increased in repeat-rich regions (88% increase in windows with six repeats) (Wilcoxon Sum Rank, p<0.05, Figure 6C, Table S5).
The “bump” in nucleotide substitutions associated with the indel (the difference between Di and Dni) that we and others [4],[6] often observe requires an explanation. The declining ratio of Di/Dni shows that this bump is smoothened over evolutionary time (Figure 4). One explanation for this is that indel mutagenicity is transient because the indel-containing allele is only mutagenic as a heterozygote and its mutagenic effect will vanish when it becomes homozygous. The period for which bacteria exist as heterogenotes for an indel is orders of magnitude less than that for diploid eukaryotes. However, a consistent decrease in Di/Dni is found across taxonomic kingdoms, an observation at odds with the proposal that heterzygosity/heterogenosity causes the indel “bump.” An alternative explanation is that the indel-associated bump in D may be due to the indel-causing event resulting in multiple nucleotide changes. This possibility is not implausible considering the spectrum of mutations in baker's yeast. Lang and Murray [30] found that in 63% of instances where two mutations occurred at the same time one was an indel and the other a nucleotide substitution; yet indels constituted only 6.67% of all mutations observed in that study. Whichever explanation is correct, it is evident that the indel effect is transient and that it is the surrounding sequence that is associated with the accruement of substitutions over evolutionary time scales.
All the inferences made about indels, nucleotide substitutions, and repeat sequences have so far been drawn only from the comparisons of genomes. In order to test predictions made by the 3R and mutagenic indel hypotheses, we utilized the comprehensive collection of spontaneous ura3 mutants gathered by Lang and Murray [30]. This collection comprises 207 ura3 mutant alleles, each of which resulted from a single mutational event in a haploid (and non-indel-containing) gene. The mutagenic indel hypothesis predicts that the clustering of mutations is caused by indels; thus, this set of independently occurring mutants should not cluster. Conversely, the 3R hypothesis states that repeat sequences cause an increase in the likelihood of the surrounding sequence sustaining both indels and nucleotide substitutions; thus, according to this hypothesis, indels and substitution mutations collected from independent mutants should cluster around repeats. Using a model based on a hyper-geometric distribution (Materials and Methods), we first found that indels and substitutions cluster together (p = 0.019), even though most substitutions occurred without a co-occurring indel (97%). Next, we tested for the association of indel/nucleotide substitution mutations with any of the 264 four-nucleotide combinations of A, T, C, and G (e.g., ATCG, ATCA, ATCT, etc.). It is expected by chance that 2 or 3 four-nucleotide combinations should be found to be significant; however, significant associations were found only with the repeat sequences TGTG (p = 0.00027), AAAA (p = 0.0093), and GTGT (p = 0.0098). These results confirm that indels, substitutions, and repeat sequences are associated independently of any initiating mutator indel.
We directly tested whether insertions of repeat sequences could increase the mutation rate of nearby regions in yeast. We engineered a copy of the URA3 gene to contain either a poly(A) repeat, a poly(G) repeat, a poly (TG) repeat, or a random 12-mer sequence in the promoter, verified that these constructs did not abolish URA3 function, and then performed fluctuation tests using the maximum likelihood method to determine the mutation rate to URA3 inactivation. We observed that (G)11 and (G)12 conferred a significant increase in the phenotypic mutation rate compared to the wild type (paired t test, p<0.001, Figure 7). Insertion of a shorter poly(G) sequence also conferred an increased rate, but the changes were less significant. On the other hand, the insertion of a random 12-mer sequence, poly (A), and poly (TG) showed no effect on the mutation rate. The fact that poly(G) causes an increase in the mutation rate is interesting considering that tetranucleotides composed of G or C bases are absent in the URA3 gene and are 5–10-fold less common across E. coli, S. cerevisiae, and Drosophila genomes than A or T tetranucleotides (unpublished data).
In order to determine if clusters of indels and substitutions influenced coding sequences in humans, we used alignments of recent segmental duplications (<5% diverged) [31] to detect indels in the human genome, restricting our analysis to those sequences that had been confirmed as expressed (see Materials and Methods). We found that indels and nucleotide substitutions occurring in human transcribed sequences follow the same patterns observed in other species, confirming that indel/region/repeat-associated mutation impacts genes expressed in humans (Figure 8).
Here we have provided evidence suggesting that regional effects have a strong influence on the accumulation of nucleotide substitutions over evolutionary time scales. Although an indel effect is also observed, we have shown the proportion of D attributable to an indel effect diminishes over time. In addition, it is not possible to formally exclude whether this effect is due to a mutagenic indel effect or a single multiple mutation causing event.
Although we found that many indels are associated with repeat sequences, many are not. This finding may be explained by the existence of other non-repeat polymerase stalling sequence motifs; another possible explanation is that repeat sequences were destroyed by mutation, while the indel remained.
So what is the impact of the indel/region effect on phenotypic evolution? Most indels in E. coli are within 100 bp of the nearest gene (Figure S5). In S. cerevisiae, 25% of promoters contain repeat sequences [32] and 600 seven-nucleotide homopolymer runs have been identified in essential genes [33], putting cis-regulatory regions and coding sequences well within the range of the effect of indel/repeat-associated mutation.
The genomes and accession numbers used for E. coli/Shigella and S. paradoxus analyses are shown in Table S1. Genome sequences for alignments between Drosophila species were downloaded from the UCSC database (http://www.biostat.wisc.edu/~cdewey/fly_CAF1/), while those for melanogastor/melanogastor alignments were downloaded from http://www.dpgp.org. The alignments of recent human segmental duplications were provided by [31]. For pairwise comparisons, genome sequences were aligned using BLAST with default parameters and divided into orthologous regions of at least 3 kb in length and >80% nucleotide sequence identity. Any region that could be aligned to multiple locations was not considered for analysis, ensuring that only orthologous sequences were used. A program was written in Perl script to find indel mutations within orthologous regions; those regions not containing indels were discarded. For three and four genome alignments, orthologous regions that were not common to all strains were discarded and those regions remaining were realigned using ClustalW.
In order to determine in which of two aligned fragments an indel has occurred, an appropriate outgroup was selected using the phylogenetic tree [34] and confirmed by our own approximations of relatedness (Table S4). In addition to establishing in which of the fragments the indel had occurred, the number of nucleotide substitutions occurring in the indel containing haplotype (Di) and non-indel containing haplotype (Dni) was determined by comparison with an outgroup sequence. For instance, when three genomes were aligned to determine indel and non-indel haplotypes, the number of mutations on the non-indel haplotype was counted by comparison of the non-indel fragment with the outgroup, and the number of substitutions on the indel haplotype was calculated by comparing the indel haplotype and the outgroup. Statistical comparisons between indel- and non-indel-containing haplotypes were carried out using the non-parametric Kolmolgorov-Smirnov paired test. See the statistical analysis plan below for more details.
An indel was designated as contiguous with a repeat for cases where the indel occurred inside the repeat (A-AAA, AA-AA, or AAA-A), or immediately next to it (−AAAA or AAAA−) where − denotes the position of the indel. It was defined as near a repeat if any part of a repeat was within five nucleotides on either side of the indel (AAAANNN−, AAAAN−, etc.). For the search for regions of high D on the basis of repeat sequence density, we used three E. coli strains not previously used in this study (E. coli SE11, E. coli SE15, and E. coli B Str. REL606). We searched for repeat-rich regions by first generating pairwise alignments (as described for the indel analysis above), followed by generating non-overlapping 100-bp windows and binning of windows according to the number of homopolymer repeats of at least 4 nt in length. Repeat sequences interrupted by a substitution mutation so that the homopolymer was less than four continuous nucleotides in length were not included. We then calculated total D for each window as well as the D for these classes of mutation: substitution, indel, transition, and transversion. To test for statistically significant differences between different classes or 100-bp windows, we used the Wilcoxon Sum Rank test.
In order to extract indel-flanking sequences for analysis, the positions of indels were recorded in each orthologous region. Next, the sequences (1 kb) both up- and downstream were extracted and examined for additional indels. If one of the flanking sequences was found to contain additional indels, that flanking region was discarded. The sequence surrounding the indel was named and ordered into windows (Figure S1). For every analysis in this study, the nucleotide divergence (D) was calculated for each window using the Jukes-Cantor method [35].
Pairs of recently diverged strains were chosen based on a phylogenetic tree (Figure 1B). Each of these designations as highly related was supported by our own estimations of divergence provided by pairwise alignments (Table S4). Two pairs of recently diverged strains were aligned by performing a new alignment of all four orthologous fragments in ClustalW, giving a total of four aligned genomes. New indels were those that occurred within pairs of recently diverged strains; for indels to be detectable, they must have occurred since the recent divergence of these two strains (see Figure S2). D for new indels was calculated using the alignment of two similar strains, of which one had been found to contain the indel. Old indels were those sites which concurred within recently diverged pairs but were different between the two pairs (see Figure S2). Such indels must have happened before the divergence of the highly similar strains yet after the divergence of the two sets of strains. For calculating D, one from each of the sets of similar genomes was selected, so that two highly diverged genomes were compared and from this comparison D is calculated for old indels. If there are double mutations (sites where the two similar genomes are different from each other and the other diverged pair), these are scored as one substitution because the difference between the two similar strains must have happened since the divergence of the two diverged sets of strains and have already been scored in the new-indel analysis. The background divergence (Db) used for the regression shown in Figure 3C was calculated as the average D from windows 3 to 10 for each E. coli pairwise alignment (window 1 comprises the 50 bp closest to the indel; windows 3 to 10 were assessed as consistently outside the range of influence of the indel) (see Figure 1A). The indel-associated divergence was calculated by subtracting the values obtained for Db from the value of D at window 1.
For pairwise comparisons between indel and non-indel haplotypes, previous studies have used paired t tests, however we found that our data was not normally distributed (Shapiro-Wilk test for normality, p<0.05). We used the two-sample Kolmogorov-Smirnov test to test for the appropriateness of the non-parametric Wilcoxon Sum Rank test for our samples. If the samples were found to be different by the Kolmogorov-Smirnov test, the Kolmogorov-Smirnov test was named and p value given (as was the case for the indel/non-indel analysis). If the two-sample Kolmogorov-Smirnov test found the samples under comparison to be of the same shaped distribution, we carried out and presented the Wilcoxon Sum Rank test and p values (this was the case for the repeat/window analysis).
A comparison of the amount of nucleotide substitutions attributable to the indel and regional effects for indels of different ages would provide for a test of the hypothesis that indel-associated mutations accumulate over time. In principle, this could be achieved by using the sets of old and new indels used for the analysis presented in Figure 3A and 3B; however, the generation of the set of old indels required a four-genome alignment; a fifth genome needs to be added to determine the indel and non-indel haplotypes. Because of our strict criteria for defining orthologous regions, the partitioning of the old and new indel sets into indel and non-indel haplotypes leaves prohibitively few orthologous regions for analysis. An alternative is to consider pairwise sets of alignments. The background nucleotide diversity for each pairwise comparison (Figure 1) provides a measure of relatedness; the greater the average value of background D, the more diverged the two strains. In order to gauge the range and degree of difference across these pairwise comparisons, the sets of background diversity values (provided by the D values for windows 3 to 10, which were chosen because they are outside the range of indel/region-associated influence) were compared. We found that most strains had distinct levels of sequence divergence from each other (Tukey's HSD, p<0.05, Table S4), with an approximately 20-fold difference in D values between the most and least diverged strains (see Table S4 for details). In order to cover a range of pairwise comparisons of increasing divergence, we chose four strains and systematically compared them to strains from clades of increasing divergence. The least divergent outgroup was always chosen. Each value of D can be partitioned into composite fractions (Figure 3D and 3E). Di is attributable to the effect of the indel and the region together, whereas Dni is attributed to the region alone. (Di − Db)/(Dni − Db) provides a measure of the total proportion of Di that is influenced by the indel. If (Di − Db)/(Dni − Db) = 1, none of the increase in nucleotide diversity can be attributed to the indel. As the value increasingly exceeds one, more of the nucleotide substitutions surrounding indels can be attributed to the indel effect. The indels detected in pairwise comparisons of more diverged strains cannot be strictly called “old” indels; these pairwise alignments will also include indels that have occurred relatively recently. However, increasingly divergent strains will be composed of a greater proportion of relatively old indels. This method of comparing indels between less diverged and more diverged strains will therefore underestimate the negative association between indels and the accumulation of nucleotide substitutions.
In order to explore indel divergence in a metazoan genus, we aligned sequenced genomes of the genus Drosophila. However, all pairwise comparisons (except the alignment of D. sechelia and D. simulans) were diverged so much that the difference between Di and Dni was undetectable ((Di − Db)/(Dni − Db) = 1). To possibly obtain alignments of less diverged genomes, we used alignments of 37 genomes available from the D. melanogastor 50 genome project (http://www.biostat.wisc.edu/~cdewey/fly_CAF1/). However, the alignment of any two of these genomes could not give enough indels suitable for analysis; most indels detected within D. melanogastor tended to cluster, leading to the rejection from our analysis of many indel-containing regions. To overcome this, suitable indels found from the alignment of all 35 strains from the Raleigh collection [36] to two of the Malawi strains (MW63 and MW27) [37] were used; indels found in more than one alignment were discarded, and from this set the 100 most and 100 least diverged indel-containing alignments were taken (background divergence was taken as Db and calculated based on the average D of windows 3 to 10).
Each nucleotide site of URA3 was classified as being mutable or not, based on the 5 bp of sequence on each side of that nucleotide, creating a stringent null model for the expected distribution of nucleotide substitutions and indel mutations. The probability of obtaining the observed distribution under the null model was calculated using the hypergeometric distribution:
where for the test for association between indels and substitutions, m is the total number of windows which are defined as mutable, k is the number of times an indel is in a region defined as mutable, N is the number of sliding windows, n is the total number of indel mutations, and for the test for association between repeat sequences and indels and substitutions, k is the number of times a tetranucleotide sequence x is contiguous with a nucleotide site defined as mutable and n is the total number of times a tetranucleotide sequence x appears in URA3.
A single (TG)6, (G)12, or (A)12 tract (or a random 12-mer (AAGTGTCAAATA) as a control) was inserted between positions −4 and −5 of URA3. Because these sequences are inherently unstable, multiple lengths of a homonucleotide tract were recovered during the cloning process, all of which left URA3 functional—providing evidence that alteration in the length of this sequence could not confer the Ura-, 5-FOA-resistant phenotype. Fluctuation tests were carried out in order to determine the mutation rate of altered URA3 genes. These were carried out by first setting up overnight cultures of each strain to be assayed in CSM-Ura media to ensure maintenance of the functional URA3 gene. The following day each strain's culture was diluted so that low numbers of cells (∼1,000) were inoculated into at least 10 independent 100 µl YPD cultures per strain in 96 well plates. Cultures were incubated at 30°C for 2 d without shaking and then spot plated onto dry 5-FOA plates. Aliquots (5 µl) of each culture were pooled, diluted, and subsequently plated onto three YPD plates to determine the total cell count. Each experiment was repeated three times. Mutation rates were calculated using the equation µ = m/Nt, where m is the mutant frequency and Nt is the total number of cells in the culture. m was determined by counting the number of 5-FOA resistant colonies for each of the 3 sets of 10 independent cultures; then calculations were carried out using FALCOR software [38] (http://www.keshavsingh.org/protocols/FALCOR.html#interface), which employs a maximum likelihood method developed by Sarkar, Ma, and Sandri [39]. The resultant value for m (mean mutant frequency) is divided by the total number of cells in the culture Nt. Nt provides a measure of the total cell divisions that have occurred in the culture; therefore, our final unit is number of Ura− mutants per cell division. Error bars are 95% confidence intervals as calculated by FALCOR using a formula devised by [40]. t tests were used to compare all strains to the wild-type strain, using formula 5 on the FALCOR website.
In order to identify indels occurring within the human lineage that may have influenced phenotypic evolution, we used a collection of recent segmental duplications (<5% diverged) [31] and identified them as expressed by comparing with the human mRNA sequence collection (refseq, NCBI). We used the Chimpanzee genome as an outgroup to identify indel and non-indel haplotypes (http://hgdownload.cse.ucsc.edu/downloads.html#chimp). All human segmental duplications were present as a single copy in the chimpanzee genome. The non-indel haplotype corresponds to the human copy that is the same as the chimp single copy at the indel site, while the indel-containing copy is the one that differs from the chimp version at the indel site.
We searched for an association between indel sites and various sequence elements that could have been associated with an increased nucleotide substitution rate. We generated a list of indels found in the E. coli K12 MG1655 genome, the best studied of all E. coli strains for which such sequence elements are well characterized. For each indel, the sequence region flanking 1 kb of the indel was designated as an indel-containing portion of the genome. The frequency with which sequence elements of interest were found in indel-containing portions of the genome compared to the rest of the genome was scored. The sequence elements that were searched were transposable elements and insertion sequences, tRNA genes, recombination sites (as indicated by the chi site), DNA sites prone to breakage (sites identified by the program Twist Flex), and repeat sequences.
|
10.1371/journal.pgen.1003173 | Identification of a BRCA2-Specific Modifier Locus at 6p24 Related to Breast Cancer Risk | Common genetic variants contribute to the observed variation in breast cancer risk for BRCA2 mutation carriers; those known to date have all been found through population-based genome-wide association studies (GWAS). To comprehensively identify breast cancer risk modifying loci for BRCA2 mutation carriers, we conducted a deep replication of an ongoing GWAS discovery study. Using the ranked P-values of the breast cancer associations with the imputed genotype of 1.4 M SNPs, 19,029 SNPs were selected and designed for inclusion on a custom Illumina array that included a total of 211,155 SNPs as part of a multi-consortial project. DNA samples from 3,881 breast cancer affected and 4,330 unaffected BRCA2 mutation carriers from 47 studies belonging to the Consortium of Investigators of Modifiers of BRCA1/2 were genotyped and available for analysis. We replicated previously reported breast cancer susceptibility alleles in these BRCA2 mutation carriers and for several regions (including FGFR2, MAP3K1, CDKN2A/B, and PTHLH) identified SNPs that have stronger evidence of association than those previously published. We also identified a novel susceptibility allele at 6p24 that was inversely associated with risk in BRCA2 mutation carriers (rs9348512; per allele HR = 0.85, 95% CI 0.80–0.90, P = 3.9×10−8). This SNP was not associated with breast cancer risk either in the general population or in BRCA1 mutation carriers. The locus lies within a region containing TFAP2A, which encodes a transcriptional activation protein that interacts with several tumor suppressor genes. This report identifies the first breast cancer risk locus specific to a BRCA2 mutation background. This comprehensive update of novel and previously reported breast cancer susceptibility loci contributes to the establishment of a panel of SNPs that modify breast cancer risk in BRCA2 mutation carriers. This panel may have clinical utility for women with BRCA2 mutations weighing options for medical prevention of breast cancer.
| Women who carry BRCA2 mutations have an increased risk of breast cancer that varies widely. To identify common genetic variants that modify the breast cancer risk associated with BRCA2 mutations, we have built upon our previous work in which we examined genetic variants across the genome in relation to breast cancer risk among BRCA2 mutation carriers. Using a custom genotyping platform with 211,155 genetic variants known as single nucleotide polymorphisms (SNPs), we genotyped 3,881 women who had breast cancer and 4,330 women without breast cancer, which represents the largest possible, international collection of BRCA2 mutation carriers. We identified that a SNP located at 6p24 in the genome was associated with lower risk of breast cancer. Importantly, this SNP was not associated with breast cancer in BRCA1 mutation carriers or in a general population of women, indicating that the breast cancer association with this SNP might be specific to BRCA2 mutation carriers. Combining this BRCA2-specific SNP with 13 other breast cancer risk SNPs also known to modify risk in BRCA2 mutation carriers, we were able to derive a risk prediction model that could be useful in helping women with BRCA2 mutations weigh their risk-reduction strategy options.
| The lifetime risk of breast cancer associated with carrying a BRCA2 mutation varies from 40 to 84% [1]. To determine whether common genetic variants modify breast cancer risk for BRCA2 mutation carriers, we previously conducted a GWAS of BRCA2 mutation carriers from the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA) [2]. Using the Affymetrix 6.0 platform, the discovery stage results were based on 899 young (<40 years) affected and 804 unaffected carriers of European ancestry. In a rapid replication stage wherein 85 discovery stage SNPs with the smallest P-values were genotyped in 2,486 additional BRCA2 mutation carriers, only published loci associated with breast cancer risk in the general population, including FGFR2 (10q26; rs2981575; P = 1.2×10−8), were associated with breast cancer risk at the genome-wide significance level among BRCA2 mutation carriers. Two other loci, in ZNF365 (rs16917302) on 10q21 and a locus on 20q13 (rs311499), were also associated with breast cancer risk in BRCA2 mutation carriers with P-values<10−4 (P = 3.8×10−5 and 6.6×10−5, respectively). A nearby SNP in ZNF365 was also associated with breast cancer risk in a study of unselected cases [3] and in a study of mammographic density [4]. Additional follow-up replicated the findings for rs16917302, but not rs311499 [5] in a larger set of BRCA2 mutation carriers. To seek additional breast cancer risk modifying loci for BRCA2 mutation carriers, we conducted an extended replication of the GWAS discovery results in a larger set of BRCA2 mutation carriers in CIMBA, which represents the largest, international collection of BRCA2 mutation carriers.
Each of the host institutions (Table S1) recruited under ethically-approved protocols. Written informed consent was obtained from all subjects.
The majority of BRCA2 mutation carriers were recruited through cancer genetics clinics and some came from population or community-based studies. Studies contributing DNA samples to these research efforts were members of the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA) with the exception of one study (NICCC). Eligible subjects were women of European descent who carried a pathogenic BRCA2 mutation, had complete phenotype information, and were at least 18 years of age. Harmonized phenotypic data included year of birth, age at cancer diagnosis, age at bilateral prophylactic mastectomy and oophorectomy, age at interview or last follow-up, BRCA2 mutation description, self-reported ethnicity, and breast cancer estrogen receptor status.
The associations between genotype and breast cancer risk were analyzed within a retrospective cohort framework with time to breast cancer diagnosis as the outcome [15]. Each BRCA2 carrier was followed until the first event: breast or ovarian cancer diagnosis, bilateral prophylactic mastectomy, or age at last observation. Only those with a breast cancer diagnosis were considered as cases in the analysis. The majority of mutation carriers were recruited through genetic counseling centers where genetic testing is targeted at women diagnosed with breast or ovarian cancer and in particular to those diagnosed with breast cancer at a young age. Therefore, these women are more likely to be sampled compared to unaffected mutation carriers or carriers diagnosed with the disease at older ages. As a consequence, sampling was not random with respect to disease phenotype and standard methods of survival analysis (such as Cox regression) may lead to biased estimates of the associations [16]. We therefore conducted the analysis by modelling the retrospective likelihood of the observed genotypes conditional on the disease phenotypes. This has been shown to provide unbiased estimates of the associations [15]. The implementation of the retrospective likelihoods has been described in detail elsewhere [15], [17]. The associations between genotype and breast cancer risk were assessed using the 1degree of freedom score test statistic based on the retrospective likelihood [15]. In order to account for non-independence between relatives, an adjusted version of the score test was used in which the variance of the score was derived taking into account the correlation between the genotypes [18]. P-values were not adjusted using genomic control because there was little evidence of inflation. Inflation was assessed using the genomic inflation factor, λ. Since this estimate is dependent on sample size, we also calculated λ adjusted to 1000 affected and 1000 unaffected samples. Per-allele and genotype-specific hazard-ratios (HR) and 95% confidence intervals (CI) were estimated by maximizing the retrospective likelihood. Calendar-year and cohort-specific breast cancer incidences for BRCA2 were used [1]. All analyses were stratified by country of residence. The USA and Canada strata were further subdivided by self-reported Ashkenazi Jewish ancestry. The assumption of proportional hazards was assessed by fitting a model that included a genotype-by-age interaction term. Between-country heterogeneity was assessed by comparing the results of the main analysis to a model with country-specific log-HRs. A possible survival bias due to inclusion of prevalent cases was evaluated by re-fitting the model after excluding affected carriers that were diagnosed ≥5 years prior to study recruitment. The associations between genotypes and tumor subtypes were evaluated using an extension of the retrospective likelihood approach that models the association with two or more subtypes simultaneously [19]. To investigate whether any of the significant SNPs were associated with ovarian cancer risk for BRCA2 mutation carriers and whether the inclusion of ovarian cancer patients as unaffected subjects biased our results, we also analyzed the data within a competing risks framework and estimated HR simultaneously for breast and ovarian cancer using the methods described elsewhere [15]. Analyses were carried out in R using the GenABEL libraries [20] and custom-written software. The retrospective likelihood was modeled in the pedigree-analysis software MENDEL [21], as described in detail elsewhere [15].
The genomic inflation factor (λ) based on the 18,086 BRCA2 GWAS SNPs in the 6,724 BRCA2 mutation carriers who were not used in the SNP discovery set was 1.034 (λ adjusted to 1000 affected and 1000 unaffected: 1.010, Figure S3). Multiple variants were associated with breast cancer risk in the combined discovery and replication datasets (Figure S4). SNPs in three independent regions had P-values<5×10−8; one was a region not previously associated with breast cancer.
The most significant associations were observed for known breast cancer susceptibility regions, rs2420946 (per allele P = 2×10−14) in FGFR2 and rs3803662 (P = 5.4×10−11) near TOX3 (Table 1). Breast cancer risk associations with other SNPs reported previously for BRCA2 mutation carriers are summarized in Table 1. In this larger set of BRCA2 mutation carriers, we also identified novel SNPs in the 12p11 (PTHLH), 5q11 (MAP3K1), and 9p21 (CDKN2A/B) regions with smaller P-values for association than those of previously reported SNPs. These novel SNPs were not correlated with the previously reported SNPs (r2<0.14). For one of the novel SNPs identified in the discovery GWAS [2], ZNF365 rs16917302, there was weak evidence of association with breast cancer risk (P = 0.01); however, an uncorrelated SNP, rs17221319 (r2<0.01), 54 kb upstream of rs16917302 had stronger evidence of association (P = 6×10−3).
One SNP, rs9348512 at 6p24 not known to be associated with breast cancer, had a combined P-value of association of 3.9×10−8 amongst all BRCA2 samples (Table 2), with strong evidence of replication in the set of BRCA2 samples that were not used in the discovery stage (P = 5.2×10−5). The minor allele of rs9348512 (MAF = 0.35) was associated with a 15% decreased risk of breast cancer among BRCA2 mutation carriers (per allele HR = 0.85, 95% CI 0.80–0.90) with no evidence of between-country heterogeneity (P = 0.78, Figure S5). None of the genotyped (n = 68) or imputed (n = 3,507) SNPs in that region showed a stronger association with risk (Figure 1; Table S3), but there were 40 SNPs with P<10−4 (pairwise r2>0.38 with rs9348512, with the exception of rs11526201 for which r2 = 0.01, Table S3). The association with rs9348512 did not differ by 6174delT mutation status (P for difference = 0.33), age (P = 0.39), or estrogen receptor (ER) status of the breast tumor (P = 0.41). Exclusion of prevalent breast cancer cases (n = 1,752) produced results (HR = 0.83, 95% CI 0.77–0.89, P = 3.40×10−7) consistent with those for all cases.
SNPs in two additional regions had P-values<10−5 for breast cancer risk associations for BRCA2 mutation carriers (Table 2). The magnitude of associations for both SNPs was similar in the discovery and second stage samples. In the combined analysis of all samples, the minor allele of rs619373, located in FGF13 (Xq26.3), was associated with higher breast cancer risk (HR = 1.30, 95% CI 1.17–1.45, P = 3.1×10−6). The minor allele of rs184577, located in CYP1B1-AS1 (2p22–p21), was associated with lower breast cancer risk (HR = 0.85, 95% CI 0.79–0.91, P = 3.6×10−6). These findings were consistent across countries (P for heterogeneity between country strata = 0.39 and P = 0.30, respectively; Figure S6). There was no evidence that the HR estimates for rs619373 and rs184577 change with age of the BRCA2 mutation carriers (P for the genotype-age interaction = 0.80 and P = 0.40, respectively) and no evidence of survival bias for either SNP (rs619373: HR = 1.35, 95% CI 1.20–1.53, P = 1.5×10−6 and rs184577: HR = 0.86, 95% CI 0.79–0.93, P = 2.0×10−4, after excluding prevalent cases). The estimates for risk of ER-negative and ER-positive breast cancer were not significantly different (P for heterogeneity between tumor subtypes = 0.79 and 0.67, respectively). When associations were evaluated under a competing risks model, there was no evidence of association with ovarian cancer risk for SNPs rs9348512 at 6p24, rs619373 in FGF13 or rs184577 at 2p22 and the breast cancer associations were virtually unchanged (Table S4).
Gene set enrichment analysis confirmed that strong associations exist for known breast cancer susceptibility loci and the novel loci identified here (gene-based P<1×10−5). The pathways most strongly associated with breast cancer risk that contained statistically significant SNPs included those related to ATP binding, organ morphogenesis, and several nucleotide bindings (pathway-based P<0.05).
To begin to determine the functional effect of rs9348512, we examined associations of expression levels of any nearby gene in breast tumors with the minor A allele. Using data from The Cancer Genome Atlas, we found that the A allele of rs9348512 was strongly associated with mRNA levels of GCNT2 in breast tumors (p = 7.3×10−5).
The hazard ratios for the percentiles of the combined genotype distribution of loci associated with breast cancer risk in BRCA2 mutation carriers were translated into absolute breast cancer risks under the assumption that SNPs interact multiplicatively. Based on our results for SNPs in FGFR2, TOX3, 12p11, 5q11, CDKN2A/B, LSP1, 8q24, ESR1, ZNF365, 3p24, 12q24, 5p12, 11q13 and also the 6p24 locus, the 5% of the BRCA2 mutation carriers at lowest risk were predicted to have breast cancer risks by age 80 in the range of 21–47% compared to 83–100% for the 5% of mutation carriers at highest risk on the basis of the combined SNP profile distribution (Figure 2). The breast cancer risk by age 50 was predicted to be 4–11% for the 5% of the carriers at lowest risk compared to 29–81% for the 5% at highest risk.
In the largest assemblage of BRCA2 mutation carriers, we identified a novel locus at 6q24 that is associated with breast cancer risk, and noted two potential SNPs of interest at Xq26 and 2p22. We also replicated associations with known breast cancer susceptibility SNPs previously reported in the general population and in BRCA2 mutation carriers. For the 12p11 (PTHLH), 5q11 (MAP3K1), and 9p21 (CDKN2A/B), we found uncorrelated SNPs that had stronger associations than the originally identified SNP in the breast cancer susceptibility region that should be replicated in the general population. In BRCA2 mutation carriers, evidence for a breast cancer association with genetic variants in PTHLH has been restricted previously to ER-negative tumors [25]; however, the novel susceptibility variant we reported here was associated with risk of ER+ and ER- breast cancer.
The novel SNP rs9348512 (6p24) is located in a region with no known genes (Figure 1). C6orf218, a gene encoding a hypothetical protein LOC221718, and a possible tumor suppressor gene, TFAP2A, are within 100 kb of rs9348512. TFAP2A encodes the AP-2α transcription factor that is normally expressed in breast ductal epithelium nuclei, with progressive expression loss from normal, to ductal carcinoma in situ, to invasive cancer [26], [27]. AP-2α also acts as a tumor suppressor via negative regulation of MYC [28] and augmented p53-dependent transcription [29]. However, the minor allele of rs9348512 was not associated with gene expression changes of TFAP2A in breast cancer tissues in The Cancer Genome Atlas (TCGA) data; this analysis might not be informative since expression of TFAP2A in invasive breast tissue is low [26], [27]. Using the TCGA data and a 1 Mb window, expression changes with genotypes of rs9348512 were observed for GCNT2, the gene encoding the enzyme for the blood group I antigen glucosaminyl (N-acetyl) transferase 2. GCNT2, recently found to be overexpressed in highly metastatic breast cancer cell lines [30] and basal-like breast cancer [31], interacts with TGF-β to promote epithelial-to-mesenchymal transition, enhancing the metastatic potential of breast cancer [31]. An assessment of alterations in expression patterns in normal breast tissue from BRCA2 mutation carriers by genotype are needed to further evaluate the functional implications of rs9348512 in the breast tumorigenesis of BRCA2 mutation carriers.
To determine whether the breast cancer association with rs9348512 was limited to BRCA2 mutation carriers, we compared results to those in the general population genotyped by BCAC and to BRCA1 mutation carriers in CIMBA. No evidence of an associations between rs9348512 and breast cancer risk was observed in the general population (OR = 1.00, 95% CI 0.98–1.02, P = 0.74) [14], nor in BRCA1 mutation carriers (HR = 0.99, 95% CI 0.94–1.04, P = 0.75) [13]. Stratifying cases by ER status, there was no association observed with ER-subtypes in either the general population or among BRCA1 mutation carriers (BCAC: ER positive P = 0.89 and ER negative P = 0.60; CIMBA BRCA1: P = 0.49 and P = 0.99, respectively). For the two SNPs associated with breast cancer with P<10−5, neither rs619373, located in FGF13 (Xq26.3), nor rs184577, located in CYP1B1-AS1 (2p22-p21), was associated with breast cancer risk in the general population [14] or among BRCA1 mutation carriers [13]. The narrow CIs for the overall associations in the general population and in BRCA1 mutation carriers rule out associations of magnitude similar to those observed for BRCA2 mutation carriers. The consistency of the association in the discovery and replication stages and by country, the strong quality control measures and filters, and the clear cluster plot for rs9348512 suggest that our results constitute the discovery of a novel breast cancer susceptibility locus specific to BRCA2 mutation carriers rather than a false positive finding. Replicating this SNP in an even larger population of BRCA2 mutation carriers would be ideal, but not currently possible because we know of no investigators with appropriate data and germline DNA from BRCA2 mutation carriers who did not contribute their mutation carriers to iCOGS. However, CIMBA studies continue to recruit individuals into the consortium.
rs9348512 (6p24) is the first example of a common susceptibility variant identified through GWAS that modifies breast cancer risk specifically in BRCA2 mutation carriers. Previously reported BRCA2-modifying alleles for breast cancer, including those in FGFR2, TOX3, MAP3K1, LSP1, 2q35, SLC4A7, 5p12, 1p11.2, ZNF365, and 19p13.1 (ER-negative only) [18], [32], [33], are also associated with breast cancer risk in the general population and/or BRCA1 mutation carriers. Knowledge of the 6p24 locus might provide further insights into the biology of breast cancer development in BRCA2 mutation carriers. Additional variants that are specific modifiers of breast cancer risk in BRCA2 carriers may yet be discovered; their detection would require assembling larger samples of BRCA2 mutation carriers in the future.
While individually each of the SNPs associated with breast cancer in BRCA2 mutation carriers are unlikely to be used to guide breast cancer screening and risk-reducing management strategies, the combined effect of the general and BRCA2-specific breast cancer susceptibility SNPs might be used to tailor manage subsets of BRCA2 mutation carriers. Taking into account all loci associated with breast cancer risk in BRCA2 mutation carriers from the current analysis, including the 6p24 locus, the 5% of the BRCA2 mutation carriers at lowest risk were predicted to have breast cancer risks by age 80 in the range of 21–47% compared to 83–100% for the 5% of mutation carriers at highest risk on the basis of the combined SNP profile distribution. These results might serve as a stimulus for prospective trials of the clinical utility of such modifier panels.
|
10.1371/journal.pntd.0003281 | A Comprehensive Assessment of Lymphatic Filariasis in Sri Lanka Six Years after Cessation of Mass Drug Administration | The Sri Lankan Anti-Filariasis Campaign conducted 5 rounds of mass drug administration (MDA) with diethycarbamazine plus albendazole between 2002 and 2006. We now report results of a comprehensive surveillance program that assessed the lymphatic filariasis (LF) situation in Sri Lanka 6 years after cessation of MDA.
Transmission assessment surveys (TAS) were performed per WHO guidelines in primary school children in 11 evaluation units (EUs) in all 8 formerly endemic districts. All EUs easily satisfied WHO criteria for stopping MDA. Comprehensive surveillance was performed in 19 Public Health Inspector (PHI) areas (subdistrict health administrative units). The surveillance package included cross-sectional community surveys for microfilaremia (Mf) and circulating filarial antigenemia (CFA), school surveys for CFA and anti-filarial antibodies, and collection of Culex mosquitoes with gravid traps for detection of filarial DNA (molecular xenomonitoring, MX). Provisional target rates for interruption of LF transmission were community CFA <2%, antibody in school children <2%, and filarial DNA in mosquitoes <0.25%. Community Mf and CFA prevalence rates ranged from 0–0.9% and 0–3.4%, respectively. Infection rates were significantly higher in males and lower in people who denied prior treatment. Antibody rates in school children exceeded 2% in 10 study sites; the area that had the highest community and school CFA rates also had the highest school antibody rate (6.9%). Filarial DNA rates in mosquitoes exceeded 0.25% in 10 PHI areas.
Comprehensive surveillance is feasible for some national filariasis elimination programs. Low-level persistence of LF was present in all study sites; several sites failed to meet provisional endpoint criteria for LF elimination, and follow-up testing will be needed in these areas. TAS was not sensitive for detecting low-level persistence of filariasis in Sri Lanka. We recommend use of antibody and MX testing as tools to complement TAS for post-MDA surveillance.
| Lymphatic Filariasis (LF, also known as “elephantiasis”) is a disabling and deforming disease that is caused by parasitic worms that are transmitted by mosquitoes. The Sri Lankan Anti-Filariasis Campaign provided five annual rounds of mass drug administration (MDA) with diethylcarbamazine and albendazole between 2002 and 2006 in all endemic areas (districts or implementation units), and this reduced infection rates to very low levels in all sentinel and spot check sites. Transmission Assessment Surveys (TAS, surveys for filarial antigenemia in primary school children) performed in 2012–2013 (about 6 years after the last round of MDA) showed that all 11 evaluation units in formerly endemic areas easily satisfied a key World Health Organization target for LF elimination programs. More comprehensive surveillance was performed with other tests to assess LF parameters in 19 study sites in the same eight districts. We detected evidence of persistent LF in all districts and evidence of ongoing transmission in several areas. Exposure monitoring (screening for anti-filarial antibodies in primary school children) and molecular xenomonitoring (detecting filarial DNA in mosquito vectors) were much more sensitive than TAS for detecting low level persistence of filariasis in Sri Lanka. These methods are complementary to TAS, and they are feasible for use by some national filariasis elimination programs. Results from this study suggest that TAS alone may not be sufficient for assessing the success of filariasis elimination programs.
| Lymphatic filariasis (LF, caused by the mosquito borne filarial nematodes Wuchereria bancrofti, Brugia malayi, and B. timori), is a major public-health problem in many tropical and subtropical countries. The latest summary from the World Health Organization (WHO) reported that 56 of 73 endemic countries have implemented mass drug administration (MDA) with a combination of two drugs (albendazole with either ivermectin or diethycarbamazine), and 33 countries have completed 5 or more rounds of MDA in some implementation units [1]. With more than 4.4 billion doses of treatment distributed between 2000 and 2012, the Global Programme to Eliminate Lymphatic Filariasis (GPELF) is easily the largest public health intervention to date based on MDA.
Bancroftian filariasis was highly endemic in Sri Lanka in the past [2]–[4]. The Sri Lankan Ministry of Health' Anti Filariasis Campaign (AFC) used a variety of methods to reduce filarial infection rates to low levels by 1999 [5], [6]. Sri Lanka was one of the first countries to initiate a LF elimination program based on GPELF guidelines [7]. The AFC provided annual MDA with diethylcarbamazine alone for three years starting in 1999. This was followed by five annual rounds of MDA with albendazole plus diethylcarbamazine in all 8 endemic districts (implementation units, IU) between 2002 and 2006. Various types of surveillance have been conducted by AFC and other groups since the MDA program ended in 2006 [8]–[12]. Post-MDA surveillance results (based on detection of microfilariae or Mf in human blood by microscopy) have consistently shown Mf rates much lower than the target value of 1% in all endemic areas [13]. The AFC also conducted school-based surveys for filarial antigenemia in 2008 according to WHO guidelines active at that time. Approximately 600 children were tested for circulating filarial antigenemia (CFA) in 30 schools in each of the 8 endemic districts, and no positive tests were observed (unpublished data, Sri Lanka Ministry of Health). WHO guidelines emphasize that LF elimination programs should provide care for people with acute and chronic clinical manifestations of filariasis [7], and the AFC has an excellent network of clinics that is devoted to this activity [13].
The present study represents a significant expansion of earlier post-MDA surveillance activities in Sri Lanka. Transmission assessment surveys (TAS) were performed according to current WHO guidelines [14], [15] for sampling primary school children to detect filarial antigenemia in each district. While TAS results may be useful for deciding whether MDA can be stopped, TAS cannot guarantee that LF transmission has been interrupted in evaluation units (EUs), which are typically districts with populations that may exceed 1 million. Therefore we conducted more intensive surveillance activities in smaller areas (Public Health Inspector “PHI” areas) that were considered to be at high risk for persistent filariasis to complement the TAS program.
Provisional targets have been proposed for documenting the interruption of filariasis transmission based on studies of the effects of MDA in Egypt, which also has LF transmitted by Culex mosquitoes [16]. Targets proposed for treated populations after at least five years of effective MDA were <2% for filarial antigenemia in communities (which corresponds to a MF prevalence rate of <0.5%), <2% for antibody to the recombinant filarial antigen Bm14 in first grade primary school children, and <0.25% for parasite DNA rates in mosquitoes as assessed by molecular xenodiagnosis (MX). The present study provided an opportunity to gain further experience with these parameters in the post-MDA setting.
Thus, the first aim of this study was to test the hypothesis that LF has been eliminated in Sri Lanka some 6 years after the completion of its national MDA program. The second aim was to assess the relative value of different methods for detecting low level persistence of filariasis after MDA.
Comprehensive surveillance activities in this project used Public Health Inspector (PHI) areas as sentinel sites. PHIs are sub-district health administration units that are comprised of smaller Public Health Midwife (PHM) areas. PHI's typically have populations in the range of 10,000–30,000 people, but they are larger in the country's capital city of Colombo which does not belong to a district. Post-MDA comprehensive surveillance studies were performed in at least two PHIs in each of the 8 LF-endemic districts in Sri Lanka plus two sites in Colombo town. The mean area of these PHIs was 6.3 km2 (range 0.6 km2–24.5 km2). Most PHIs selected for this study were considered to be at increased risk for persistent filariasis based on high infection rates prior to MDA or based on results of microfilaremia surveys conducted after 2006.
Field teams for collection of demographic information and blood specimens consisted of a medical officer, a Public Health Inspector, a phlebotomist, and one or two assistants. Blood samples were collected during the day. Sterile, single use, contact activated BD-microtainer lancets (Fisher Scientific, Pittsburgh, PA) were used for blood collection in community and school surveys. Approximately 300 to 400 µl of blood was collected by finger prick from each study subject into an EDTA coated blood collection vial (Fisher Scientific). Barcode stickers were used to link specimens to data records. Samples were transported to the AFC headquarters laboratory in Colombo in coolers. Plasma was separated from blood samples from school children and stored at −80 C for later antibody testing.
A pilot study was performed in Peliyagodawatta in Gampaha district in 2008 as a training exercise and to test the feasibility of comprehensive LF surveillance in Sri Lanka using methods pioneered in Egypt. This semi-urban area (with a population of about 10,560 in an area of 1.59 km2) was resurveyed in 2011. All other PHIs were only studied once.
The community surveys used a systematic sampling scheme to sample all areas in each PHM within the PHI being studied. The AFC obtained census lists with the numbers of houses in each PHM and PHI along with maps showing the PHMs within PHIs. The number of houses/households needed for each community survey (125) was divided by the number of PHMs in the PHI to get the number of houses to be sampled in each PHM. That number was divided by 4 to get the number of houses to be sampled per quadrant in each PHM. The central house in the quadrant was sampled, and other houses were selected by moving in the 4 cardinal directions from the central house. The sampling interval for houses was calculated by dividing the total number of houses in the PHM quadrant by the number of houses that were to be sampled in that quadrant. For instance, if there were 60 houses in a quadrant and 10 houses were to be sampled, the sampling interval was 6. If a selected house could not be sampled because of absence or refusal, field teams sampled the next house. Community surveys sampled people who were at least 10 years of age, and a maximum of 4 subjects were enrolled per house.
Finger prick blood was collected from children in grades 1 and 2 in primary schools that served children in the study PHIs; approximately 350 school blood samples were collected per PHI. Blood was tested for filarial antigenemia by card test, and plasma was stored for later antibody testing.
Mosquitoes were collected with gravid traps (Model 1712, John W. Hock Company, Gainesville, FL) using liquid bait. The liquid bait was prepared 5–6 days prior to use containing yeast, milk powder and dry straw in water [17]. In some PHI areas cow dung was added to the liquid bait to attract mosquitoes.
Gravid traps were placed adjacent to houses for one to four days; mosquitoes were collected in the morning and traps were replaced in the evening. Traps were placed in shaded, quiet areas near natural breeding sites. Traps were placed in all 4 quadrants of each PHM to ensure sampling from all areas in each PHI.
In the Peliyagodawatta pilot study in 2008, 4835 mosquitoes were collected from 20 trap sites, and the number of pools collected from each trap ranged from 1–10 pools of mosquitoes (range 5–20 mosquitoes per pool). In all subsequent surveys, 4 pools of twenty mosquitoes were collected from each of 50 trapping sites per PHI. Trapped mosquitoes were collected, sorted, dried at 95°C for 1 hr. and placed in tubes for later testing (20 mosquitoes/pool). The tubes were labeled with barcode stickers and transferred to the AFC headquarters laboratory for DNA isolation and qPCR testing.
Washington University personnel trained staff in the central AFC laboratories on standard operating procedures for Mf detection by microscopy, antibody and antigen testing, DNA isolation from mosquitoes, and detection of filarial DNA by qPCR. All samples were tested in AFC laboratories in Colombo.
Circulating filarial antigenemia (CFA) was detected with a simple card test (BinaxNOW Filariasis, Alere Inc., Scarborough, ME) [16], [18].
IgG4 antibodies to recombinant filarial antigen Bm-14 in human plasma were detected by microplate ELISA (Filariasis CELISA, Cellabs Pty Ltd, Brookvale, NSW, Australia) as previously described [19]. Previous studies have shown that this kit is sensitive and specific for infection and/or heavy exposure to filarial parasites. Plasma ELISAs were performed with a single well per sample, and all positive and borderline tests were retested on a different day. Samples that produced an OD value >0.35 in two assays performed on different days were considered to be positive for antibody to Bm14.
Microfilaria (Mf) testing was performed for people with positive filarial antigen tests (in community household surveys, school surveys, and TAS) with three-line blood smears (60 µl total volume of night blood tested).
Mosquitoes were sorted by experienced technicians. Blood fed, gravid, and semi-gravid Culex quinquefaciatus mosquitoes were identified by morphology and sorted into 4 pools of 20 mosquitoes per collection site. Two hundred and seventy-seven pools of mosquitoes (mean pool size of 17) were collected and tested from Peliyagodawatta in the pilot study that was performed in 2008. Approximately 200 pools were tested from each PHI area in later surveys. W. bancrofti DNA was detected in mosquito pools by qPCR as previously described [16], [20]. DNA isolation and PCR analysis for samples from the 2008 pilot study were performed by AFC personnel together with Washington University technicians in St. Louis. All subsequent PCR work was conducted by AFC personnel in the AFC laboratory in Colombo.
Demographic information including age, gender, documentation of informed consent, and a history of compliance with the previously administered MDA program was collected and entered into personal digital assistants (PDA) (Dell Axim ×51, Dell Inc. Round Rock, TX or HP iPAQ 211, Hewlett Packard, Palo Alto, CA) using a preloaded survey questionnaire. Participant data, specimen ID, and test results were linked using preprinted barcode labels as described by Gass et al [21]. AFC deployed 2 or 3 teams for blood collection and 2 or 3 teams for mosquito collection in each PHI, and teams were comprised of a mixture of personnel from the district and from AFC headquarters. Data collected by multiple teams were synchronized at AFC headquarters, and data were transferred to a laptop computer using LF field office data manager software designed by the Lymphatic Filariasis Support Center, Taskforce for Global Health, Decatur, GA. Transferred files were merged to create a master database, which was backed up using an external hard drive. Specimens and laboratory test results were linked to study subject numbers (or to trap site and pool number for mosquito data) using barcodes. Deidentified, cleaned data were transferred into Excel files (Microsoft Corp., Redmond, WA) for analysis at AFC and at Washington University.
GPS coordinates for human and mosquito sampling sites were plotted using ArcGIS 10.2.1 (ESRI, Redlands, CA) to show the location of households surveyed and mosquito trapping sites for each PHI. Waypoints were color coded to show the infection status of household residents and mosquitoes from these collection sites.
TAS were performed in all 8 endemic districts in late 2012 or early 2013 according to WHO guidelines. The TAS program used districts as evaluation units (EUs) in 5 cases. However, 3 districts or areas with large populations (Colombo district plus Colombo town, Gampaha, and Galle) were each divided into two EUs for TAS. All EUs met criteria for conducting TAS by having completed 5 rounds of MDA in 2006 with high MDA compliance rates (>80%). All sentinel and spot check sites in each district had Mf prevalence rates well below 1% for several years prior to TAS. Since Sri Lanka has high primary school attendance rates (>95%), TAS surveys used the cluster method to sample students in 30–35 randomly selected schools per EU[15]. Systematic selection of school children was performed with Survey Sample Builder software, SSB.V.2.1 (http://www.ntdsupport.org/resources/transmission-assessment-survey-sample-builder).
The TAS sampling strategy required filarial antigen testing of approximately 1500 primary grade children in each EU. Blood samples were collected with One Touch Ultra Soft lancet holders with disposable lancets (LifeScan, Inc., Milpitas, CA). Finger prick blood was collected into capillary tubes provided with the BinaxNow Filariasis cards, and 100 µl of blood was added directly to sample application pads of the cards according to the manufacturer's instructions. Tests were performed in the school auditorium, library, or health screening station immediately after blood collection, and read at 10 minutes. Antigen test results (positive or negative) were recorded manually using preprinted data collection forms. Children with positive filarial antigen tests were tested for microfilaremia with night blood smears as described above.
We used the software program PASW Statistics 18 (SPSS, now IBM Corporation, Armonk, NY) and JMP (SAS, Cary, NC). The Chi-square test was used to assess the significance of differences in categorical variables such as antigenemia rates. The correlation between human and mosquito infection parameters was analyzed by the Spearman rank test. Logistic regression was used to assess the independence of risk factors for filarial antigenemia. Graphs were produced with GraphPad Prism V. software (La Jolla, CA). Filarial DNA rates (maximum likelihood estimates with 95% confidence intervals) were calculated with PoolScreen 2.02 [22], [23]. To sharpen the analysis of risk factors for filarial infection, we limited the analysis to 14 PHI areas where one or more people had positive filarial antigen tests. All analyses were performed assuming simple random sampling for simplicity of exposition. A generalized linear mixed model was used to estimate design effects of household-based cluster sampling used in community surveys. This analysis was performed with data from the two PHIs with the highest surveyed CFA rates.
The study protocol for comprehensive surveillance in PHIs was reviewed and approved by institutional review boards at Washington University School of Medicine and at the University of Kelaniya in Sri Lanka (FWA 00013225). Prior to school surveys (both PHI surveys and TAS), study personnel held preliminary meetings with school principals and officials from the Sri Lankan Ministry of Education about the goals and procedures for the study. They also met with parents or guardians to discuss the study design and the significance of the study.
Printed participant information sheets and written consent forms were provided to participants (or to parents/guardians) in Sinhalese, Tamil and English. Written consent was obtained from adults. Participation of minors required written consent from at least one parent or guardian plus assent by the child/minor. Consent was also documented electronically into PDAs by study personnel prior to collection of health information or blood samples. TAS surveys used preprinted paper forms for parental consent and other forms for data collection (school name, child name, age, sex, and CFA result).
Nineteen PHI surveys were conducted in 8 districts and in Colombo town between March 2011 and July 2013. Demographic information for survey participants is provided in Table 1, and results are summarized in Table 2 and Figure 1. Community CFA rates were <2% in 17 of 19 PHIs, but upper confidence limits for CFA were >2% in 5 of 19 PHIs. Microfilaremia rates were <1% in all PHI areas studied. Sixteen of 65 CFA-positive subjects (age range 23–70 yr) were positive for Mf (mean count 14 per 60 µl range 1–51), and 68% of Mf carriers were males. The Unawatuna PHI area in Galle district had the highest rates for several filariasis parameters (Table 2 and Figure 1).
CFA rates were higher in males than females when data from all community surveys were considered (1.01% vs. 0.42%, P<0.001) and when localities with no positive CFA tests were excluded from the analysis (1.39% vs. 0.57%, P<0.001) (Table 3). CFA rates were also higher in adults than in children, and this was especially true for people older than 30 years (Table 3). CFA rates were lower in people who reported having used a bed net the night before their interview (all localities), but the difference was not statistically significant (0.57% vs. 0.92%, P = 0.06). However, the reduced CFA rate in bed net users was significant when localities with no positive CFA tests were excluded from the analysis (0.76% vs. 1.29%, P = 0.04). Bed net users also had lower rates of microfilaremia in these localities (0.17% vs. 0. 52%, P = 0.012).
Reported compliance rates for ingestion of antifilarial medications during the national MDA program were high in most PHIs surveyed, but very low rates were reported in PHIs in Galle district and in Colombo town (Table 2). These results are consistent with low surveyed compliance rates previously reported for these areas [10]. CFA rates in community surveys were significantly lower in people who reported that they had ingested antifilarial medication during the national MDA program (0.45% vs. 1.15%, P = 0.001).
Logistic regression was used to assess the independence of different risk factors for CFA for all surveyed communities and for the subset of communities with one or more subjects positive for CFA (Table 4). Gender, age, and prior MDA treatment were significant independent indicators of risk, but reported bed net use was not.
Intraclass correlations by household in the two locations with the highest filarial infection rates were 0.16 and 0.08, and these values correspond to design effects of 1.6 and 1.3.
CFA rates were very low in children tested in school surveys, and this was consistent with TAS results presented below. Anti-filarial antibodies were detected in primary school children in 17 of 19 PHIs. Antibody rates exceeded the target rate of 2% in 10 of 19 PHIs; five PHIs had borderline elevated antibody rates, and 5 others had higher rates with upper confidence limits >5%. Only three of 137 children with positive antibody tests (out of 6198 children tested for antibody from all 19 PHI areas) had positive CFA tests, and all three of these children were Mf negative.
Community antibody testing was performed in a subset of PHIs that were surveyed in the comprehensive surveillance study (Table S1). Although CFA and Mf rates in these communities were below provisional target levels, community antibody rates were high in all of these PHIs, and this probably reflects high infection rates that were present in these areas prior to implementation of the national MDA program.
Human filariasis parameters tended to be significantly correlated with each other [e.g., community Mf rate vs. community CFA rate (r = 0.63, P = 0.0018), school CFA rate vs. school antibody rate (r = 0.5, P = 0.0142), and community CFA rate vs. school CFA rate (r = 0.69; P = 0.0006)].
More than 17,000 primary grade school children were tested in TAS in 337 schools located in 11 EUs in 8 districts and in Colombo town (Table 5). The numbers of positive CFA tests were well below the TAS threshold level of 18 (critical cut-off value) in all EUs. Thus all EUs “passed” TAS including the coastal Galle District EU, where high rates for filariasis markers were noted in two PHI study areas. None of the 16 children with positive CFA tests in TAS surveys had microfilaremia. All CFA-positive children were treated with anti-filarial medications and follow-up surveys are in progress or planned to further assess people in areas with positive children.
Almost 3,900 pools (20 mosquitoes per pool) of blood fed, gravid or semi-gravid mosquitoes collected in 19 PHI areas were tested for filarial DNA by qPCR (Table 6). Filarial DNA rates exceeded the target of 0.25% in 10 of 19 PHIs. Mosquitoes from both PHIs surveyed in Galle district and one in Matara district had parasite DNA rates of more than 1%, and these rates were comparable to those seen in some filariasis endemic areas in Egypt with continued filariasis transmission following one or two rounds of MDA [24]. Upper confidence limits for filarial DNA rates were ≥1% in 5 of 19 PHIs surveyed. On the other hand, three of 19 PHIs surveyed had no positive mosquito pools. Most of the other filariasis parameters were also low in these PHIs. Mosquito DNA samples from Wattala were retested by qPCR at Washington University and confirmed to be negative.
The percentages of positive mosquito trap sites were highly variable in different PHIs, and these rates were strongly correlated with percentages of pools positive for filarial DNA (r = 0.99, P<0.0001), community CFA rates (r = 0.72, P = 0.0003), and school CFA rates (r = 0.77; P<0.0001). Percentages of mosquito pools positive for filarial DNA were highly correlated with community CFA rates (r = 0.71, P = 0.0001) and school CFA rates (r = 0.79, P<0.0001). In addition, percentages of houses with at least one CFA positive resident were highly correlated with percentages of mosquito trap sites with filarial DNA in mosquitoes (r = 0.75, P = 0.0001) (Table S2) and with percentages of mosquito pools that contained filarial DNA (r = 0.73; P = 0.0002).
GPS data for PHI areas with high and low rates of persistent LF are shown in Figures 2 and S1. These maps show that sampled households and mosquito collection sites were nicely dispersed to cover the study areas. Infections in human and parasite DNA in mosquitoes tended to be dispersed in most study areas.
A pilot LF surveillance study was performed in 2008 in Peliyagodawatta, which is located in Gampaha district just outside of the city of Colombo. The area was resurveyed in 2011, approximately 2.5 years after the baseline study. This is a low income, peri-urban area with high mosquito densities, and no intervention for LF control was undertaken in this area between 2008 and 2011. Results from the two surveys are summarized in Table 7. Several filariasis parameters were lower in 2011 than in 2008. While only the reduction in community CFA was statistically significant, the trend toward reduction was present for all of these parameters apart from Mf rate, which was already very low in 2008.
The first survey in Peliyagodawatta identified 37 amicrofilaremic subjects with positive filarial antigen tests. These people were not treated for LF at that time. Twenty-five of these people were retested in 2010, approximately 18 months after the first survey; others had moved or were otherwise not available for follow-up. Only 12 of 25 subjects were still CFA-positive (48%), and only 1 of 25 was microfilaremic by 60 µl night blood smear. None of the subjects reported symptoms or signs of clinical filariasis during the 18 month interval. All subjects with filarial antigenemia were treated in 2011.
This study has provided interesting data on the status of LF in Sri Lanka approximately 6 years after completion of the country's MDA program, and it has important implications for post-MDA surveillance activities in other LF-endemic countries around the world. Few countries participating in GPELF have been studied as thoroughly as Sri Lanka.
The term “LF elimination” has been interpreted in different ways, but WHO documents clearly state that one goal of LF elimination programs is interruption of transmission [15]. WHO is also responsible for deciding when countries have eliminated LF. Pending their review, we think it is important to recognize the achievements of Sri Lanka's Anti-Filariasis Campaign, which is one of the finest LF elimination programs in the world. The program has reduced Mf rates to less than 1% in all sentinel and spot check sites, all EUs easily passed TAS criteria for stopping MDA, and the AFC has a network of clinics that provide care to thousands of lymphedema patients in all endemic districts. By these criteria, Sri Lanka has achieved several WHO targets and the country is on track to achieve elimination. If WHO determines that Sri Lanka has not met criteria for LF elimination, we believe that the organization should develop criteria and a recognition program for countries that can document this level of superb control, because this pre-elimination status is a significant achievement in public health and an important step on the road to LF elimination. External recognition of “superb control” or “near elimination” may help national programs obtain political support and resources needed for the difficult last mile required for true elimination.
While protocols for transmission assessment surveys are based on solid sampling principles, the sensitivity of TAS for detecting ongoing transmission of LF has not been adequately tested in field studies [15]. Our results clearly show that TAS performed according to WHO guidelines were not sensitive for detecting ongoing LF transmission in Sri Lanka. There are a number of reasons for this. First, we believe that EUs of 1 to 2 million are too much too large, because risk factors that affect LF transmission often vary widely across such large populations/areas. This problem could be mitigated by reducing the size of EUs (for example, to areas with populations of 100,000 or less), but that would significantly increase the cost of TAS. A second problem with TAS is that filarial antigenemia rates in young children are sometimes very low in areas with ongoing LF transmission. Our study showed that CFA rates in school aged children were much lower than those in adults. Therefore, the sensitivity of TAS might be improved by using a similar cluster sampling method to test adults (for example, those attending primary health clinics) instead of children in schools. A recent report from Togo described the use of other types of passive surveillance for assessing LF following MDA [25].
Since anti-filarial antibody rates are uniformly higher than antigenemia rates in LF-endemic populations, another potential solution for the problem of low TAS sensitivity would be to substitute antibody testing for antigen testing in TAS for samples of school-aged children. Antibody results from the present study using a commercially available ELISA kit provide a proof of principle for this approach. However, ELISA testing may not be feasible for all LF programs, and available rapid-format antibody tests have not yet been validated for this purpose.
Results from this study strongly support the use of molecular xenodiagnosis for post-MDA surveillance in areas where LF is transmitted by Culex mosquitoes. MX does not require collection of blood samples or active participation by large numbers of people in endemic areas. However, MX does require cadres of skilled personnel, specialized laboratory facilities, and funds for consumables. While MX was performed by MOH personnel in this study, this required significant external inputs including equipment, supplies, training of personnel, and funds for mosquito collection. Also, additional work is needed to develop and validate sampling methods for assessment of mosquito DNA rates in areas larger than PHIs.
To summarize this section of the Discussion, while TAS surveys may be useful for decisions regarding stopping MDA, they are not sufficient to show that LF transmission has been interrupted. The sensitivity of TAS might be improved by reducing the size of EUs or by sampling adults instead of school-aged children. We recommend antibody testing of children using TAS sampling methods and/or MX (especially in areas believed to be at high risk) to complement antigen-test based TAS, because these methods appear to be more sensitive than TAS for detecting ongoing LF transmission.
This study has provided new insight regarding provisional targets for MDA programs that were suggested in 2007 based on data from Egypt [16]. Since there is uncertainty surrounding all point estimates, we now recommend using confidence intervals to express targets as illustrated in Figure 1. The new suggested target for the antifilarial antibody rate in first and second grade school children is to have an upper confidence limit of <5%. The new target for MX (Culex mosquitoes) is to have an upper confidence limit of the maximum likelihood estimate of <1%. The new target for the community CFA rate (age >9) is to have an upper confidence limit of <2%. This target provides a very high level of confidence that the Mf rate will be less than 0.5% in the community with a much smaller sample size than what would be required for Mf testing. Additional studies will be needed to test the new proposed targets in different regions. We believe that these targets will be helpful for identifying areas that require continued surveillance.
Existing guidelines do not adequately address this issue. Four options to consider are resumption of MDA, implementation of test and treat programs, vector control, and watchful waiting. It may be difficult to justify resumption of MDA when Mf rates are well below 1% when one considers that many of those with persistent infections may have been noncompliant with MDA in the past. Test and treat campaigns may be more efficient for finding and treating those with persistent infections than MDA, and the Sri Lanka AFC has started to do this in Galle district. Our results suggest that adult males and people who do not recall having taken MDA in the past should be considered to be high priority target groups for test and treat programs.
WHO has recommended vector control as a post MDA strategy [26]. Although vector control can be difficult to implement at the scale needed for LF elimination, surveillance results may identify hot spot areas where focused vector control may be feasible. Our finding that CFA rates were lower in people who reported using bed nets is interesting, although the logistic regression analysis suggested that lack of bed net use was not an independent risk factor for filarial infection. Bed nets are popular in Sri Lanka because of the mosquito nuisance factor and the risk of dengue. Beneficial effects of bed nets for LF have been reported from areas with Anopheles transmission [27], [28]. The Sri Lanka government should consider implementing a health education campaign to reinforce the popularity of bed nets and increase usage rates in areas with persistent LF.
The longitudinal data from Peliyagodawatta are intriguing, because they suggest that some areas with filariasis parameters that do not meet our provisional criteria for interruption of transmission may spontaneously improve over time. Thus the strategy of watching, waiting, and retesting may be the best course of action for some areas with persistent LF. Other data from Peliyagodawatta on the natural history of filarial antigenemia in amicrofilaremic individuals in the post-MDA setting are reassuring. These results suggest that there is no pressing need to actively identify and treat asymptomatic and amicrofilaremic persons with positive filarial antigen tests following MDA. This is because the risk of such people developing microfilaremia is low, and antigenemia often clears over time without treatment.
We believe that this study has contributed significant new information regarding post-MDA surveillance and low level persistence of filariasis following MDA. LF elimination is a dynamic process [29], and point estimates of persistent infection may be less important than trends over time. For this reason, we plan to restudy Peliyagodawatta and several other PHIs with elevated LF parameters three years after the evaluations described in this publication.
|
10.1371/journal.pntd.0005740 | Cardio-haemodynamic assessment and venous lactate in severe dengue: Relationship with recurrent shock and respiratory distress | Dengue can cause plasma leakage that may lead to dengue shock syndrome (DSS). In approximately 30% of DSS cases, recurrent episodes of shock occur. These patients have a higher risk of fluid overload, respiratory distress and poor outcomes. We investigated the association of echocardiographically-derived cardiac function and intravascular volume parameters plus lactate levels, with the outcomes of recurrent shock and respiratory distress in severe dengue.
We performed a prospective observational study in Paediatric and adult ICU, at the Hospital for Tropical Diseases (HTD), Ho Chi Minh City, Vietnam. Patients with dengue were enrolled within 12 hours of admission to paediatric or adult ICU. A haemodynamic assessment and portable echocardiograms were carried out daily for 5 days from enrolment and all interventions recorded.
102 patients were enrolled; 22 patients did not develop DSS, 48 had a single episode of shock and 32 had recurrent shock. Patients with recurrent shock had a higher enrolment pulse than those with 1 episode or no shock (median: 114 vs. 100 vs. 100 b/min, P = 0.002), significantly lower Stroke Volume Index (SVI), (median: 21.6 vs. 22.8 vs. 26.8mls/m2, P<0.001) and higher lactate levels (4.2 vs. 2.9 vs. 2.2 mmol/l, P = 0.001). Higher SVI and worse left ventricular function (higher Left Myocardial Performance Index) on study days 3–5 was associated with the secondary endpoint of respiratory distress. There was an association between the total IV fluid administered during the ICU admission and respiratory distress (OR: 1.03, 95% CI 1.01–1.06, P = 0.001). Admission lactate levels predicted patients who subsequently developed recurrent shock (P = 0.004), and correlated positively with the total IV fluid volume received (rho: 0.323, P = 0.001) and also with admission ALT (rho: 0.764, P<0.001) and AST (rho: 0.773, P<0.001).
Echo-derived intravascular volume assessment and venous lactate levels can help identify dengue patients at high risk of recurrent shock and respiratory distress in ICU. These findings may serve to, not only assist in the management of DSS patients, but also these haemodynamic endpoints could be used in future dengue fluid intervention trials.
| Dengue is a viral illness that can lead to severe and potentially fatal complications. The most common complication is fluid leakage from blood vessels, which can cause low blood pressure or dengue shock syndrome (DSS). The majority of patients recover with simple intravenous fluid replacement, however in approximately 30% of DSS cases, recurrent episodes of shock occur, and these patients have a higher risk of fluid overload, respiratory distress and death. We investigated whether using portable echocardiograms (Echo) in the intensive care unit (ICU) to assess cardiac function and intravascular volume parameters as well as blood lactate levels, can help identify these patients. We found patients who developed recurrent shock had higher heart rates and lower Stroke Volume Index (SVI), and higher lactate levels at enrolment than those with 1 episode or no shock. Higher SVI and worse cardiac function after 3 days in ICU was associated with respiratory distress. Admission lactate levels predicted patients who subsequently developed recurrent shock and correlated positively with the total IV fluid volume received. These results demonstrate that Echo-derived intravascular volume assessment and venous lactate levels can help identify dengue patients at high risk of poor outcomes in the ICU, and could assist in the management of severe dengue.
| Dengue is a flaviviral infection that causes substantial morbidity in endemic areas, with 96 million clinically apparent cases each year [1]. Although the majority of infections result in a self-limiting febrile illness, 1–5% of cases can experience more severe manifestations, in the form of organ impairment, coagulopathy and plasma leakage which may lead to intravascular volume depletion and dengue shock syndrome (DSS). The plasma leakage usually resolves around defervescence, and the extravasated fluid then gets reabsorbed, when fluid overload in the form of massive pleural effusions or pulmonary oedema can occur. The resulting respiratory compromise has been associated with an increased risk of death in adult [2] and paediatric severe dengue [3]. The current treatment of DSS is supportive, with careful intravenous fluid replacement. The majority of patients recover after a single crystalloid bolus and in experienced centres the mortality rate is less than 1% [4]. However, in approximately 30% of DSS cases, recurrent episodes of shock occur, which require more intensive treatment with larger volumes of intravenous fluids including colloid boluses; these patients have a higher risk of fluid overload, respiratory distress and poor outcomes [5]. Identifying such individuals early and investigating other potential contributing factors for recurrent shock is needed. Although the main mechanism of DSS is hypovolaemia, it is becoming increasingly recognized that myocardial impairment may play a role in the haemodynamic instability and potentially could contribute to recurrent shock [6, 7]. Cardiac manifestations of dengue are diverse and include functional myocardial impairment, arrhythmias and myocarditis, however the clinical significance of these in DSS has not been well studied [8].
We have shown previously that systemic microvascular dysfunction occurs in more severe dengue infections, but the effect on end-organ perfusion has not been evaluated [9]. Lactate levels can be representative of tissue perfusion and elevated levels are associated with organ failure and predict mortality in septic shock [10, 11]. Serum lactate levels and their prognostic significance in dengue shock syndrome have only been evaluated in small studies and in adults [12–14].
In this study we investigated the association of echocardiographically-derived cardiac function and intravascular volume parameters as well as lactate levels with the clinical outcomes of recurrent shock and respiratory distress in adults and children admitted to ICU with dengue. We hypothesised that DSS patients with cardiac dysfunction and elevated lactate levels would be more likely to develop recurrent shock and also to experience iatrogenic fluid overload and respiratory distress.
This prospective observational study was performed at the Hospital for Tropical Diseases (HTD), Ho Chi Minh City, Vietnam. Individuals aged above 3 years were screened for enrolment if they were admitted to either paediatric or adult Intensive Care Unit (ICU) with a clinical diagnosis of dengue with warning signs or severe dengue [15], and were within 12 hours from ICU admission. In Vietnam patients aged 15 or more are admitted to adult ICU. All patients were reviewed daily until hospital discharge or for up to 5 days from enrolment; at each assessment standardized clinical information was recorded including clinical symptoms and signs, vital signs and all interventions. The amount and type of all intravenous fluids were documented. Portable echocardiograms were performed as soon as feasible after enrolment, and then daily until discharge from ICU. The patients were followed up 10–14 days later.
Ethical approvals were obtained from the Oxford Tropical Research Ethics Committee and the Ethics Review Committee at HTD, and written informed consent was obtained from all participants or the parent/guardian of children.
A full blood count was performed daily and at follow-up. A biochemistry sample for liver and renal function was performed at enrolment and subsequently depending on clinical need. An un-cuffed venous blood sample was taken at enrolment for venous lactate which was processed within 30 minutes of collection.
Dengue diagnostics: Commercial IgM and IgG serology assays (Capture ELISA, Panbio, Australia) were performed on batched acute and convalescent plasma. In addition RT-PCR was performed on the enrolment sample to identify the DENV serotype and measure plasma viraemia levels [16]. Patients were defined as having dengue if the RT-PCR was positive or if the IgM assays were positive at enrolment, or IgM seroconversion between paired specimens and on the basis of their clinical picture. Patients with negative tests at enrolment, but for whom convalescent plasma was not available, were considered unclassifiable.
Echocardiograms were performed at the bedside by one of the investigators (SY, HT, VN), using an M-turbo system (FUJIFILM SonoSite, Inc, USA) with cardiac settings. The Echocardiograms were performed daily and at follow-up 14 days later. The exam included two-dimensional, M-mode and Doppler studies. More detailed methodology can be found elsewhere [6]. All images were stored digitally and a selection reviewed by a cardiologist (CB) in the United Kingdom. The inter- and intra- user variability was checked at regular intervals and was consistently <10%.
Linear regression models were used for the initial analysis, with each cardio-haemodynamic parameter as the outcome and shock status as covariate. The analysis was adjusted for age, gender, and day of illness at ICU admission. Logistic regression was used for the prognostic models, predefined candidate variables at enrolment were used to predict recurrent shock/respiratory distress. Associations between the parameters were assessed by partial correlations controlling for the following potential confounding variables: age, sex and day of illness at enrolment, study day of measurement. Significance of partial correlations was assessed based on their Fisher transformation and corresponding bootstrap standard errors. The cluster bootstrap which resamples patients rather than samples accounted for multiple measurements per patient. To informally adjust for multiplicity, a significance level of 0.01 was used for all comparisons. All analyses were performed with the statistical software R version 3.2.2 and the companion package geepack version 1.2–0.
One hundred and three patients were enrolled between September 2014 and September 2015, 1 patient was excluded from the analysis due to an inconclusive diagnosis and 88 had serial echocardiograms (S1 Fig). The median age was 11 years (IQR 8–14 years). The median time from ICU admission to enrolment was 0.1 hours (IQR: 0.0–1.0 hour), and from enrolment to first echo was 2.5 hours (IQR 1.0–9.3 hours). Twenty-two patients did not develop DSS, 80 patients had DSS, of which 48 had a single episode and 32 had recurrent episodes of shock (4 of these 32 patients required inotropes for refractory shock). Of the patients with 1 episode of shock 34/48 had shock at ICU admission and 14/48 developed shock; median time of 1.75 hours (IQR: 0.75–4.08 hours) after admission. The patients with recurrent shock, 27/32 had shock at ICU admission and 5/32 developed their first episode of shock; median of 1.92 hours, (0.06–6.88 hrs) after admission. The patients who were not in shock at ICU admission, were admitted for more intensive monitoring and/or fluid therapy due to the presence of several warning signs. Nineteen patients developed the secondary endpoint of respiratory distress, 10 (53%) on day 2, 8 (42%) on day 3 and 1 (5%) on day 4, all had radiological evidence of pleural effusions but not pulmonary oedma (S1A Fig). Seven patients had major bleeding, all following shock, 6/7 with recurrent shock and 1/7 with 1 episode of shock. Of the 97 patients for whom RT-PCR was performed, 65 were positive with the following serotypes: 52 DENV-1 (80%); 10 DENV-2 (15%; and 3 DENV-4 (5%). There was 1 death in the study population in adult ICU. 63/102 (62%) were admitted to ICU straight from clinic/community, 39/102 were admitted from the general ward, after a median of 22 hours, but no intravenous fluids were administered on the general wards.
At enrolment, platelet counts and albumin levels were significantly lower, and AST levels significantly higher, in patients who later experienced recurrent shock than those who did not. Lactate levels were also significantly higher in patients who went on to have recurrent episodes of shock than those who had 1 episode or no shock (Table 1). Hyperlactatemia, using a cut-off of >4 mmol/l, was present in 17/32 (53.1%) of the individuals with recurrent shock, 8/48 (16.7%) of the patients with 1 shock episode and was absent in all patients without shock. Considering the 88 patients that had serial echocardiograms, 24/88 (27%) had evidence of left ventricular dysfunction and 6/8 (7%) right ventricular dysfunction. A higher proportion of adults had left and right ventricular dysfunction, 11/16 (69%) and 4/16 (25%) compared to children, 13/72 (18%) and 2/72 (3%).
Patients who developed recurrent shock had a higher enrolment pulse than those with 1 episode of shock or no shock (median: 114 vs. 100 vs. 100 b/min, P = 0.002), and reduced pulse pressure (PP) (median: 20 vs. 20 vs. 30 mmHg, P = 0.001) (Table 2). There was a significantly lower Stroke Volume Index (SVI) in the patients with recurrent shock versus patients with and without 1 shock (median: 21.6 vs. 22.8 vs. 26.8ml/m2, P = 0.002).
SVI was significantly lower at enrolment (study day 1) for patients with recurrent shock compared with no shock (median: 21.6 vs. 26.8mls/m2, P<0.001) and also between patients with shock compared with no shock (median: 22.8 vs. 26.8mls/m2, P = 0.001) (Table 3, Fig 1).
There was a significantly lower cardiac index (CI) between patients with shock compared to no shock on the first study day. The SVI remained lower for patients with recurrent shock versus no shock on study day 2 (median: 22.8 vs. 27.2 mls/m2, P = 0.004). A non-significant trend for higher CI and LMPI was observed on study day 3 in the recurrent shock patients compared to no shock.
Higher SVI on study day 4 was associated with the secondary endpoint of respiratory distress, as well as a trend for higher CI and respiratory distress (S1 Table). The majority of patients only received IV fluids on days 1–2 days (S2 Table), so the higher SVI on days 4 and 5 likely represents fluid re-absorption rather than iatrogenic causes. On study days 3–5, worse left ventricular function (higher LMPI) was associated with respiratory distress. There was an association between the total IV fluid administered during the ICU admission and respiratory distress (OR: 1.03, 95% CI 1.01–1.06, P = 0.001). Respiratory distress presented early (study day 2) in half of the patients, all had evidence of bilateral pleural effusions, suggesting plasma leakage likely causes early respiratory distress in ICU which is later compounded by fluid re-absorption and myocardial impairment on study days 3–5.
Enrolment lactate levels predicted patients who subsequently developed recurrent shock compared to those who did not (Table 4). In addition, higher enrolment lactate levels were also found to predict patients developing respiratory distress (3.9 vs. 3.0 mmol/l, OR 1.46, 95% CI 1.09–2.12, P = 0.008).
The SVI correlated with other parameters of intravascular volume including inferior vena cava collapsibility index (IVCCI) with a negative correlation (rho -0.491, P<0.001) and left ventricular end-diastolic diameter (LVEDD) with a positive correlation (rho 0.354, P<0.001). The IVCCI correlated with LMPI with a positive correlation (rho 0.230, P<0.001). The LMPI, RMPI, LVEDD and IVCCI did not correlate with the amount of IV fluids in the preceding 24 hours.
Enrolment lactate levels correlated positively with the total IV fluid volume received (rho: 0.323, P = 0.001) and also with enrolment ALT (rho: 0.764, P<0.001) and AST (rho: 0.773, P<0.001), but not with any of the cardio-haemodynamic parameters.
We have shown that myocardial impairment was not associated with recurrent shock but was associated with the secondary endpoint of respiratory distress after 3 days of the ICU admission. Lower stroke volume indices during the first 2 days of ICU admission and tachycardia were associated with both recurrent shock and respiratory distress. Higher lactate levels at ICU admission were also predictive for recurrent shock and respiratory distress. These results suggest patients with evidence of severe volume depletion at ICU admission including lower SVI, higher heart rates and venous lactates were more likely to develop recurrent shock and require more intravenous fluids–resulting in respiratory distress from a combination of plasma leakage and myocardial impairment, exacerbated by volume overload from fluid reabsorption in the recovery phase.
Cardiac functional assessment in patients with hypovolaemia is more challenging and hence our use of Doppler derived parameters, which have been shown to be less preload dependent [17]. Myocardial dysfunction was associated with respiratory distress but not with recurrent shock, suggesting the myocardial impairment was sufficient to play a role in fluid overload, following resuscitation and the associated respiratory compromise but not to contribute to the shock syndrome, which appears to be driven predominantly by intravascular volume depletion. These findings are comparable to other echo studies, including one study of Thai children which demonstrated 36% of patients with DSS had reduced systolic function and patients with cardiac impairment were more likely to have fluid overload [18]. The mechanisms underlying this transient myocardial dysfunction in dengue patients remain to be defined, but potential mechanisms may involve some or a combination of the following; myocardial depressant factors, myocardial interstitial oedema, abnormal coronary microcirculation and endothelial dysfunction and also abnormal calcium homeostasis [8, 19, 20].
Most patients admitted to ICU with severe dengue showed signs of intravascular volume depletion, as evidenced by low SVI, CI, and smaller LVEDD and higher IVCCI compared to follow-up. Ejection fractions however remained normal in all the groups, which may be explained by low end diastolic volume and/or diastolic dysfunction- both which may play a role in dengue. SVI was the most robust parameter associated with the severe outcomes of recurrent shock and respiratory distress. Heart rate was significantly higher at enrolment between patients with recurrent shock versus no shock. This confirms a previous study where higher heart rate was found to be useful in predicting children developing profound shock [21]. The IVCCI, although being higher in patients at enrolment compared to discharge, did not discriminate between shock and no shock and was not associated with clinical outcomes. IVCCI has also been shown to correlate with central venous pressure (CVP) and right atrial pressure (RAP) in children and adults [22, 23], and is useful in predicting fluid responsiveness, in mechanically ventilated patients and in spontaneously breathing patients [24, 25]. Due to the coagulopathy and thrombocytopenia in the majority of severe dengue patients, CVP carries a significant risk of bleeding and other non-invasive methods of assessing intravascular volume and guiding fluid therapy are urgently needed [26]. Portable bedside echocardiographic assessment of haemodynamics, particularly the SVI are useful in identifying patients with recurrent shock and could be considered as an alternative to invasive CVP monitoring.
We have shown venous lactates in dengue patients on the first day of admission to ICU is associated with severe outcomes of recurrent shock and respiratory distress. Lactate levels correlated with the total amount of IV fluids received, but did not correlate with other haemodynamic parameters.
The higher lactates likely represent severe volume depletion from plasma leakage causing tissue hypoperfusion, hypoxia and anaerobic glycolysis. In addition to hypoperfusion and excess production of lactate, another mechanism for hyperlactatemia in severe dengue may involve reduced hepatic clearance as moderate hepatic dysfunction occurs in severe dengue [27]. The liver may play a role in the hyperlactatemia in critical illness with circulatory failure, not only by reduced metabolism but also because the liver itself can produce lactate due to hepatic ischaemia. This is supported by our study which showed lactate levels correlated positively with both ALT and AST levels.
A study investigating patients with shock admitted to ICU, found higher lactate levels in patients with early hepatic dysfunction compared to those with no hepatic dysfunction, independent of haemodynamic severity parameters [28]. Altered microcirculation, which we have shown is worse in dengue patients with more severe plasma leakage [9], may play a role in the increased lactate levels, although further studies are required to link the microcirculatory perfusion abnormalities with higher lactate levels.
The current WHO guidelines for managing DSS recommend initial resuscitation with crystalloid fluids for compensated shock, followed by careful on-going assessment including serial HCT measurements and close monitoring of vital signs. Reassessing patients in shock and achieving predefined physiological targets has been a major focus of research in severe sepsis in the last 2 decades [29–31]. The ‘goals’ of resuscitation are currently being readdressed with emerging evidence that a conservative approach to fluid management has better outcomes in certain settings [32]. The balance of administering just sufficient intravenous fluid therapy to maintain haemodynamic stability while avoiding fluid overload and respiratory compromise is extremely difficult and additional cardiovascular monitoring using portable echocardiography in DSS would be beneficial. Stroke volume monitoring may provide improved targeted volume resuscitation. While serial HCT and vital sign monitoring are useful and widely applicable for resource constrained settings, intensive care facilities and associated technologies are improving in many dengue endemic areas, so additional non-invasive cardiovascular assessment is now possible and should be considered where available.
There were some limitations to our study. In order not to interfere with emergency management, some patients had the first echo study after initial fluid resuscitation had commenced. This may therefore underestimate some of the cardiovascular parameters. Secondly, due to restrictions on research blood sampling in paediatric patients, we were unable to take daily lactates so it was not possible to study lactate clearance times. In addition, due to the coagulopathy in many of severe dengue patients, we were unable to take arterial blood gases for assessment of metabolic acidosis to explore relationship with the high lactate levels. As the majority of patients enrolled in this study were children and young adults, the results may not be generalizable to older adult populations with dengue.
In conclusion, this study has identified several simple non-invasive parameters that could assist risk prediction and help tailor management of dengue patients admitted to ICU. We have shown moderate cardiac dysfunction is common in ICU patients with dengue, particularly among adults. The cardiac dysfunction does not appear to play a major part in the haemodynamic instability of dengue shock but likely contributes to the development of fluid overload and respiratory compromise in some cases. Echo-derived volume assessment using stroke volume index combined with heart rate monitoring and venous lactate levels can help identify patients at high risk of recurrent shock. The clinical and therapeutic implications of these findings are potentially important, first as prognostic markers to guide fluid resuscitation and assist in the management of DSS as currently practiced, and second, these echo-derived haemodynamic endpoints could be used in future dengue fluid intervention trials designed to assess alternative strategies intended to improve DSS management and outcome.
|
10.1371/journal.pgen.1000226 | Effects of Aneuploidy on Genome Structure, Expression, and Interphase
Organization in Arabidopsis thaliana | Aneuploidy refers to losses and/or gains of individual chromosomes from the
normal chromosome set. The resulting gene dosage imbalance has a noticeable
affect on the phenotype, as illustrated by aneuploid syndromes, including Down
syndrome in humans, and by human solid tumor cells, which are highly aneuploid.
Although the phenotypic manifestations of aneuploidy are usually apparent,
information about the underlying alterations in structure, expression, and
interphase organization of unbalanced chromosome sets is still sparse. Plants
generally tolerate aneuploidy better than animals, and, through colchicine
treatment and breeding strategies, it is possible to obtain inbred sibling
plants with different numbers of chromosomes. This possibility, combined with
the genetic and genomics tools available for Arabidopsis
thaliana, provides a powerful means to assess systematically the
molecular and cytological consequences of aberrant numbers of specific
chromosomes. Here, we report on the generation of Arabidopsis
plants in which chromosome 5 is present in triplicate. We compare the global
transcript profiles of normal diploids and chromosome 5 trisomics, and assess
genome integrity using array comparative genome hybridization. We use live cell
imaging to determine the interphase 3D arrangement of transgene-encoded
fluorescent tags on chromosome 5 in trisomic and triploid plants. The results
indicate that trisomy 5 disrupts gene expression throughout the genome and
supports the production and/or retention of truncated copies of chromosome 5.
Although trisomy 5 does not grossly distort the interphase arrangement of
fluorescent-tagged sites on chromosome 5, it may somewhat enhance associations
between transgene alleles. Our analysis reveals the complex genomic changes that
can occur in aneuploids and underscores the importance of using multiple
experimental approaches to investigate how chromosome numerical changes
condition abnormal phenotypes and progressive genome instability.
| Most plants and animals have two copies of each chromosome in the normal
chromosome set. Unbalanced numerical changes resulting from gains or losses of
individual chromosomes (aneuploidy) usually have deleterious consequences. For
example, Down syndrome in humans is caused by an extra (triplicate) copy of
chromosome 21. Human tumor cells usually display numerous alterations in
chromosome number and structure. Little is known about how changes in chromosome
number influence gene activity and chromosome integrity, thereby perturbing
physiology and development. We have used the model plant A.
thaliana to study how triplication of chromosome 5 affects gene
expression, chromosome structure, and chromosome packaging in the nucleus. The
results indicate that the presence of an extra chromosome 5 has multiple
effects: (1) substantial changes in gene expression occur, primarily on the
triplicated chromosome 5 but also on the four non-triplicated chromosomes; (2)
broken derivatives of chromosome 5 can be retained in the presence of two normal
copies; and (3) two copies of the triplicated chromosome 5 may show a slightly
enhanced tendency to associate with each other, perhaps to spatially compensate
for the chromosome imbalance. The detrimental effects of aneuploidy are likely
due to concurrent changes in gene expression, chromosome structure, and
arrangement.
| Changes in the number of chromosomes from the normal diploid set can be grouped into
two types: polyploidy and aneuploidy. Polyploidy refers to whole genome duplications
whereas aneuploidy refers to unbalanced losses and/or gains of individual
chromosomes, or parts of chromosomes, from the basic chromosome set. Early work on
plants and insects revealed that aneuploidy has a greater effect on phenotype than
polyploidy [1],[2]. These observations can be explained in terms of
the gene balance hypothesis, which posits that dosage imbalances of genes encoding
regulatory molecules disturb their stoichiometry within multi-protein complexes and
disrupt cellular processes [2]. Consistent with this hypothesis, work in
Drosophila has indicated that genes encoding transcription
factors and members of signal transduction cascades are primarily responsible for
dosage effects on the phenotype [1].
The gene balance hypothesis provides a conceptual framework for investigating in
greater detail the molecular and cytological consequences of aneuploidy. This
information is important for understanding the coordinated operation and expression
of the genome as well as syndromes and disease states associated with abnormal
chromosome numbers. The latter is exemplified by human solid tumour cells, which are
highly aneuploid. The karyotypes of advanced tumour cells typically feature not only
a plethora of chromosome numerical aberrations but also extensive structural
alterations, including translocations and deletions [3]. The co-existence of
chromosome numerical and structural changes in tumour cell nuclei hints that they
are linked in some way, but the basis of this connection is unclear. The genomes of
tumour cells often display a distinctive DNA methylation profile that is
characterized by global hypomethylation accompanied by aberrant hypermethylation of
CpG islands within promoter regions [4],[5]. That aneuploidy might be at the root of these
diverse genomic and epigenomic changes was suggested by a study on trisomic tobacco
plants, in which the chromosome present in triplicate was prone to breakage, local
increases in DNA methylation, and gene silencing [6],[7].
Another aspect of aneuploidy concerns interphase chromosome arrangement and dynamics,
which are increasingly regarded as factors influencing gene activity [8]. Down
syndrome in humans, which is caused by triplication of chromosome 21 (trisomy 21),
is relevant in this context. Chromosome 21 is the smallest human autosome [9], not
the most gene-poor (a distinction that belongs to chromosome 13 [10]), and
it is the only autosome that is compatible with extended life after birth when
triplicated [11]. These observations might be partially explained
if extra chromosomes interfere with chromosome packaging or mechanics such that
triplication of the smallest is the least harmful. However, the ways in which extra
or missing chromosomes in aneuploids might perturb the three-dimensional (3D)
architecture and dynamics of interphase chromosomes are not understood.
The consequences of aneuploidy for global gene expression patterns are only beginning
to be assessed. With respect to Down syndrome, the naïve expectation is
that genes on the triplicated chromosome 21 will be expressed at 1.5 times the level
found in chromosome 21 disomics according to the increase in gene dosage. However,
only a subset of expressed genes on triplicated chromosome 21 appears to be
up-regulated in the expected manner whereas the expression of many genes is adjusted
to the disomic level, indicating dosage compensation [12]. The extent of
trans or secondary effects, in which genes on non-triplicated chromosomes are
misregulated, is still not fully resolved with respect to trisomy 21 [13]–[15]. Trans effects have been
documented in aneuploids of maize [16],[17] and yeast [18],
demonstrating that changes in expression are not restricted to genes on the
numerically altered chromosome. However, information about how global patterns of
gene expression are adjusted following chromosome-wide alterations in gene dosages
is still limited. This issue is complex because unique expression profiles are
likely to result from numerical changes of specific chromosomes or chromosome
regions.
Plants have traditionally provided excellent systems for studying aneuploidy. The
terms trisome and monosome were coined by Blakeslee, Belling and coworkers from
their classic work in the 1920's on the twelve trisomics of Datura
stramonium (Jimson weed), each of which displays a distinctive
phenotype [2]. With respect to mechanisms of epigenetic regulation
and genome composition, plants are arguably more similar to mammals than are yeasts
or Drosophila. For example, both plants and mammals have DNA
methylation, histone H3 lysine 9 and lysine 27 methylation, and proteins of the RNAi
machinery; moreover, their genomes contain substantial amounts of repetitive DNA,
which can potentially affect gene expression and chromosome structural stability
[19].
Insights gained from plants can thus be informative for understanding the effects of
aneuploidy in mammalian cells. Plants have the advantage of generally tolerating
aneuploidy better than mammals, and their chromosome numbers can be more easily
manipulated to allow systematic analyses of the consequences of chromosome numerical
aberrations.
We are using the model plant Arabidopsis thaliana
(2n = 10) to investigate the impact of aneuploidy
on genome structure, expression and 3D organization of interphase chromosomes. All
five trisomics of Arabidopsis
(2n = 10+1) are viable and have a
distinctive phenotype [20]. The genetics and genomics resources available
for Arabidopsis are unsurpassed in the plant kingdom. In addition,
transgenic Arabidopsis lines are available in which distinct
chromosome sites are tagged with fluorescent markers [21],[22], allowing the
identification of specific trisomics at an early stage and subsequent live cell
imaging of fluorescent-tagged sites in interphase nuclei in intact plants. Here we
report the results of experiments using these tools to analyze the molecular and
cytological consequences of chromosome 5 triplication in
Arabidopsis.
The strategy for obtaining chromosome 5 trisomics and for subsequent analysis of
these plants is shown in Figure
1. We started with a diploid parental line that was homozygous for DsRed
(R) and YFP (Y) fluorescent tags on chromosome 5, which is one of the largest
chromosomes in Arabidopsis (Figure 2A). From a cross between the diploid
parent and a tetraploid derivative produced by colchicine treatment, we obtained
triploid plants (F1 generation). Self-fertilization of F1 triploids produced F2
progeny, 33 of which were selected for more detailed investigation. Screening
root nuclei in F2 seedlings for chromosome 5 fluorescent tags allowed us to
predict whether individual F2 plants might be diploid (2R 2Y), chromosome 5
trisomic/triploid (3R 3Y) or chromosome 5 tetrasomic/tetraploid (4R 4Y). The
actual chromosome numbers were subsequently determined by counting metaphase
chromosomes, and the presence of unbalanced chromosome sets was assessed by
array comparative genome hybridization (CGH) (Table S1).
The F2 progeny comprised a complex population containing chromosomally balanced
diploids, triploids and tetraploids, as well as chromosomally unbalanced
trisomics (the most frequently observed chromosome constitution), double
trisomics (2n = 10+1+1), and
near triploids (3X = 15+/−1
or 15+1+1) (Figure 2B). As expected from the screen of chromosome 5 fluorescent
tags, we obtained a number of plants with a triplicated chromosome 5 (3R 3Y);
however, subsequent array CGH and metaphase chromosome counts revealed that only
three of these were true triploids (plants 8-5, 8-6, 9-1; plant 11-5 had 15
chromosomes, but one copy of chromosome 1 was truncated; see below) and just two
were simple chromosome 5 trisomics (plants 6-5 and 6-7) (Table S1A).
The remaining ‘3R 3Y’ plants had an additional extra
chromosome(s), the most common being either chromosome 2 or 4, which are the
smallest of the Arabidopsis chromosome set (Figure 2C).
Representatives of the next generation (F3) were obtained by self-fertilization
of the two trisomic F2 plants (6-5 and 6-7) and two diploid F2 siblings (6-4 and
7-2). From each of the two trisomic F2 parents, we selected around a dozen F3
progeny that were identified by fluorescence microscopy as potential chromosome
5 trisomics (3R 3Y) (Table S1B). Extra copies of chromosome 5 were
confirmed in these plants by array CGH and, in most cases, the expected
chromosome number (2n = 10+1) was
established by counting metaphase chromosomes. From each of the two diploid
parents, we selected for further analysis four F3 progeny that were chromosome 5
disomics (2R 2Y) and confirmed the expected diploid chromosome number by
counting metaphase chromosomes (Table S1B).
Previous work with a trisomic tobacco line suggested that the chromosome present
in triplicate was vulnerable to breakage [6]. Here we used array
CGH to assess genome integrity in selected progeny of
Arabidopsis triploids, including chromosome 5 trisomics from
the F2 and F3 generations (Table S1). Array CGH can detect not only
imbalances of intact chromosomes but also parts of chromosomes resulting from
breakage, thereby revealing the approximate location of a breakpoint.
The first chromosome break we detected was in a triploid plant from the F2
generation (11-5; Table S1), which contained a truncated copy
of chromosome 1 lacking part of the top arm (Figures 2A and 3). The two trisomic F2 plants, 6-5 and 6-7,
had structurally intact genomes as assessed by array CGH. In the F3 generation,
however, we detected chromosome breaks in two trisomic plants (out of 26 tested
by array CGH; Table S1B), one from each trisomic F2 parent. Both of these breaks
affected the triplicated chromosome 5. In one case essentially the entire top
arm of chromosome 5 was deleted (plant 6-5-22), suggesting a break around the
centromere. In the second case, the break occurred in the vicinity of the
DsRed transgene locus, such that the tip of the bottom arm
of chromosome 5 was lost (plant 6-7-10) (Figure 2A and Figure 3).
Although derived from a relatively small sample size, these findings support the
idea that trisomics show enhanced breakage of the chromosome present in
triplicate and/or retention of a fractured chromosome when two intact copies are
present. Because the truncated versions of chromosome 5 appeared in individual
trisomic F3 progeny, they were likely generated during meiosis in the trisomic
F2 parent. The possibility that breaks of the triplicated chromosome occur more
frequently in somatic cells of trisomics than of diploids [23] can be studied in
the future by performing single cell array CGH [24],[25].
Whether the trisomic plants containing truncated versions of chromosome 5 would
transmit the broken chromosome to the next generation is not yet known. In a
pilot study, a second generation chromosome 5 trisomic plant harbouring a break,
again in the vicinity of the DsRed transgene locus (plant
12-16; Figure 2A),
transmitted the truncated chromosome to trisomic progeny. However, array CGH of
five trisomic progeny plants did not detect further deletions of chromosome 5
(data not shown). A more comprehensive study analyzing additional breakpoints in
progeny plants across several generations might uncover evidence for progressive
structural changes after formation of an initial break and reveal whether any
specific DNA sequence features are associated with breakpoints. The current data
suggest that repetitive regions, for example around the centromere and the
DsRed transgene locus, which contains lac
operator repeats [21],[22], are preferential
sites of breakage in trisomics. The chromosome 1 break in the triploid plant
11-5 occurred in an intergenic, nonrepetitive region that does not contain
conspicuous features.
To assess the impact of chromosome 5 triplication on global gene expression, we
carried out gene expression profiling using Affymetrix ATH1 microarrays, which
report on about 21,000 Arabidopsis transcripts of the current
TAIR genome annotation (v7). We were interested in comparing chromosome 5
trisomics and diploid plants with respect to the expression of genes on
triplicated chromosome 5 (primary or cis effects) and the expression of genes on
the four non-triplicated chromosomes (secondary or trans effects). All plants
used for the transcriptome analysis (F2 trisomics 6-5, 6-7 and eight F3 progeny;
F2 diploids 6-4, 7-2 and three F3 progeny) had intact genomes as assessed by
array CGH (Table
S1A,B).
Microarray hybridization signals not only showed a strong systemic effect for the
trisomic chromosome 5 but also a wide range of clear trans effects for
transcripts on the disomic chromosomes (Figure 4) consistent across the relatively
large number of biological replicates analysed. It is noteworthy that many
popular normalization transforms are not appropriate for data sets with
large-scale expression level shifts as seen here because these violate
underlying assumptions of many methods. The consequential distortions and signal
dampening are illustrated for reference in the Supporting Information (Text S1)
and Online Supplement (http://bioinf.boku.ac.at/pub/trisomy2008), where we also discuss
alternative normalization methods ranging from popular established tools used in
previous studies [17],[18] to specialized
approaches such as exploiting CGH data as reference.
Observed expression levels of most transcripts on chromosome 5 reflected the
dosage effect of its increased copy number in chromosome 5 trisomics, whereas
most transcripts on other chromosomes did not change. Examination of expression
differences as a function of average signal intensities in a traditional
M(A)-plot, however, revealed an unexpected intensity
dependence that has no biological explanation (Figure 5): Each transcript is represented by
a dot and error bar, with the difference in expression (trisomics minus
disomics) shown on the y-axis, and the average expression on
the x-axis. Green marks the transcripts on chromosome 5.
Magenta and orange trend lines respectively show the intensity dependence
plus/minus one standard deviation for chromosome 5 and the other chromosomes.
The deviation of the magenta centre trend line from a line parallel to the
horizontal reflects the non-linear response of the detection system. The figure
shows that differential expression is most accurately surveyed when using the
microarray platform for sufficiently strongly expressed transcripts. We thus
focused on the transcripts to the right of the dashed line (roughly half:
2,452/4,790 on chromosome 5 and 7,355/15,725 others), best reflecting the true
trends for all the genes (cf.
Text S1
and Online Supplement for discussion). Both average response and significant
deviations from the chromosomal trends were studied.
Only a minor degree of dosage compensation was observed, with the percentage of
genes on chromosome 5 classed as having similar expression levels in both
trisomic and diploid plants ranging from 3% (by convex decreasing
density estimate [26]) to 11–15%
(89% differential expression for Benjamini-Yekutieli FDR
q<5%). Interestingly, despite the increased
gene dosage, 1% of transcripts on chromosome 5 had significantly
lower expression levels than in the diploid. Whether the observed
down-regulation is due to epigenetic silencing, altered transcription factor
availability, or other mechanism is not yet known. The down-regulated genes,
which are for the most part rather uniformly distributed along chromosome 5
(Figure 4), do not
appear to have any conspicuous common features.
In contrast to the modest number of dosage-compensated and down-regulated genes,
the highest proportion of chromosome 5 transcripts (86–88%)
showed a significant increase in expression (partial or full dosage effect),
reflecting the extra copy of chromosome 5 in the trisomics (88%
significantly upregulated; 14% of expression changes below the trend;
both with Benjamini-Yekutieli FDR q<5%). The
expression increase of 12–13% of transcripts on chromosome
5 was even significantly above the average trend (hyper-dosage effect) for this
chromosome (13% with Benjamini-Yekutieli FDR
q<5%).
To verify this general trend also for chromosome 5 genes with lower expression
levels, we used more sensitive quantitative RT-PCR (qRT-PCR) to quantify
transcript levels of four moderately expressed genes on this chromosome,
selected for their minimal variation during development (http://www.weigelworld.org/resources/microarray/AtGenExpress/)
and five lowly expressed genes. Consistent with the general chromosome 5 trend,
a higher steady-state transcript level in trisomics was indeed observed for the
majority of these genes, confirming a dosage effect (Figures S1
and S2).
A different picture emerged for the secondary or trans effects on the other
chromosomes: While the 12–13% ratio of transcripts
up-regulated relative to the trend was similar, only 8–9%
of transcripts on other chromosomes were significantly down-regulated, giving a
strong 3∶2 skew favoring up-regulation vs
down-regulation. Trans-effects were equally distributed across all chromosomes
(Figure 4,
Fisher's exact test,
p = 33%),
indicating that trisomy 5 has a genome-wide effect on gene expression.
Stress response genes and transcription factors were significantly
overrepresented among the genes involved in trans-effects (Table 1). Indeed, the ten
most-significant trans-effects included four transcription factors, of which
three were strongly up-regulated (AGL19, ANAC019, AtMYB47) and one
down-regulated (MEE3). The prominence of transcription factors in the strongest
trans effects supports the gene balance hypothesis [2]. For the cis
effects, genes involved in responses to abiotic or biotic stimulus and cell wall
components were significantly affected whereas for dosage-compensated genes on
chromosome 5, genes involved in structural roles and ribosome biogenesis were
significantly over-represented (Table 1).
Changes in the expression of genes encoding transcription factors may alter the
expression of numerous target genes and hence contribute to the genome-wide
changes in expression observed in chromosome 5 trisomics. Similarly, changes in
genes encoding epigenetic modifiers might also be expected to influence the
expression of multiple target genes distributed throughout the genome.
Chromosome 5 genes encoding known epigenetic modifiers showed the higher
expression levels of the expected dosage effect in chromosome 5 trisomics. These
include the DNA methyltransferases DRM2, DRM1, and MET1; the histone modifying
enzymes HDA6 and SUVH4; and the SNF2-like chromatin remodeling protein DDM1
(Figure
S3). In addition, epigenetic modifiers encoded on non-triplicated
chromosomes were also involved in the trisomy 5 response. These include two
genes on chromosome 2: ROS1, which encodes a DNA
glycosylase-lyase protein involved in active demethylation of cytosines in DNA
and hence acts antagonistically to MET1, DRM2 and DRM1 [27]; and
RDR5, which encodes an RNA-dependent RNA polymerase related to
those acting in RNAi-mediated pathways in plants [28] (Figure S4).
Previous work has shown a link between components required for DNA methylation
and those for active demethylation of DNA [29]. For example, in
met1 mutants, which have decreased levels of DNA
methylation, ROS1 expression is significantly reduced [29],[30]. One possibility
is that the increased expression of DNA methyltransferases encoded on chromosome
5 might be counterbalanced by increased ROS1 expression to
maintain global DNA methylation at a level compatible with plant viability.
Further work is needed to test this hypothesis.
In summary, transcript expression profiling by microarrays revealed that while
the increased expression of the majority of transcripts
(86–88%) on chromosome 5 reflected a partial, full, or
hyper-dosage effect due to the triplication of this chromosome, there was a
small set of transcripts (3–15%) for which there was
evidence of dosage compensation. In contrast, there were
12–13% of transcripts across all
chromosomes that were up-regulated with respect to their chromosomal
neighborhoods. While there were at least as many transcripts
(13–14%) on chromosome 5 down-regulated relative to the
chromosome trend, down-regulation on other chromosomes was only observed for
8–9% of transcripts.
Generally elevated expression levels reflecting dosage effects for the
triplicated chromosome, a genome-wide 3∶2 skew favoring up-regulation
vs down-regulation in gene specific response, and
dosage-compensation for some genes on chromosome 5 can together account for all
these observations.
To determine whether the up-regulation of ROS1 and
RDR5 in chromosome 5 trisomics is a generic response to an
increased chromosome number or is specific for chromosome 5 trisomics, we used
qRT-PCR to investigate expression of these genes in other F2 trisomics obtained
from self-fertilization of the triploid F1 parents (Figure 2C; Table S1).
Despite their similar behaviour in individual chromosome 5 trisomics (Figure 6, top and middle, left,
compare diploid lanes 1–6 with trisomic lanes
7–12) , ROS1 and RDR5 showed
independent responses in other trisomics. For example, triplication of
chromosome 2 (three plants available for testing) resulted in higher expression
of RDR5 at a level consistent with the increased gene dosage
(Figure 6, top, right, lanes
chr. 2) while ROS1 expression was slightly below the
diploid level, suggesting dosage compensation of this gene in the triplicated
state (Figure 6, middle, right,
lanes chr. 2). Both genes were sharply down-regulated in chromosome 3
and chromosome 4 trisomics, although only single plants were available for
testing (Figure 6, top and middle,
right, lanes chr. 3 and chr. 4). In three plants harbouring
triplications of both chromosome 4 and chromosome 5 (double trisomics), an
intermediate level of ROS1 expression (around that observed in
diploids) was observed (Figure 6,
middle, right, lanes chrs. 4+5). By contrast,
RDR5 was expressed in the double trisomics at a level
comparable to chromosome 5 single trisomics (Figure 6, top, compare lanes chrs. 4+5,
right, with trisomic lanes 7–12, left). One interpretation
of these results is that positive regulators of ROS1 and
RDR5 are on chromosome 5, and in addition, a negative
regulator of ROS1 is on chromosome 4.
The data on ROS1 and RDR5 expression illustrate
the complex variations in the expression of single genes in aneuploids of
different chromosome constitutions. Genes encoding epigenetic modifiers can
change expression independently, regardless of whether they are present on a
numerically altered chromosome. These findings suggest that different
aneuploidies might variably affect epigenetic mechanisms, creating diverse
patterns of epigenetic modifications depending on the chromosome constitution.
Additional work to determine genome-wide distributions of various epigenetic
modifications in different aneuploids is required to test this conjecture.
We also used qRT-PCR to examine the expression of DsRed-LacI and
TetR-YFP transgenes, which are present on chromosome 5 but
not represented on the ATH1 microarray. Interestingly, even though the
DsRed-LacI and TetR-YFP transgenes are
both transcribed by the cauliflower mosaic virus 35S promoter [21],[22], they respond
differently to triplication of chromosome 5. The TetR-YFP gene
was strongly down-regulated in chromosome 5 trisomics compared to diploids
(Figure 6, bottom, right,
diploid lanes 1–6, trisomic lanes 7–12). By
contrast, the average expression of the DsRed-LacI gene
remained at roughly the same level in both diploid and chromosome 5 trisomic
plants, consistent with dosage compensation of this transgene when triplicated
(Figure 6, bottom, left,
diploid lanes 1–6, trisomic lanes 7–12). The
expression of Ds-Red-LacI appears to display more
plant-to-plant variability in trisomics than in diploids, however, suggesting a
stochastic element to the dosage compensation mechanism (Figure 6, bottom, left, diploid lanes 1–6,
trisomic lanes 7–12).
It is unknown why the two 35S promoter-driven transgenes reacted differently upon
triplication of chromosome 5 nor is it clear why the TetR-YFP
transgene undergoes such a steep reduction in expression when triplicated.
Silencing and methylation of a transgene encoding neomycin phosphotransferase in
tobacco was observed when the transgene locus was present on all three copies of
a triplicated chromosome [6]. Both the TetR-YFP and
DsRed-LacI transgene loci comprise complex inserts of the
respective transgene construct [22]. The TetR-YFP transgene is
integrated near a cluster of silent transposon-related sequences and tRNA genes
(At5g20852 to At5g20858) that give rise to numerous small RNAs (http://mpss.udel.edu). By contrast, the DsRed-LacI
transgene is inserted into two overlapping, moderately expressed protein-coding
genes (At5g58140 and At5g58150) in a gene-rich region [21]. Perhaps the
repetitive and silent genomic environment enhances silencing of the
TetR-YFP transgene in trisomics. The basis of
TetR-YFP silencing and whether repressive epigenetic
modifications and/or small RNAs are involved remain to be determined. Although
most down-regulated endogenous genes on triplicated chromosome 5 are not in
repetitive regions, two of the most robustly down-regulated predicted genes
(At5g35480, At5g35490; http://bioinf.boku.ac.at/pub/trisomy2008/nonorm2/down.cis.minA.ldiff.triVsWT.EBFWER.txt)
are divergently transcribed from a common promoter and associated with
transposon-related sequences and numerous small RNAs (http://mpss.udel.edu).
The fluorescent-tagged sites on chromosome 5 are useful for identifying
chromosome 5 trisomics at an early stage of development before the
characteristic phenotype of trisomy 5 is visible. In addition, high resolution
measurements of distances between DsRed and
YFP transgene alleles can be made in interphase nuclei of
living cells and subsequent 3D reconstructions of optical sections of nuclei can
reveal the relative arrangements of the fluorescent tags. In a previous study of
16 different fluorescent-tagged sites distributed throughout the genome in
diploid plants, random arrangements were observed in interphase nuclei of root
cells. There was no indication of allelic pairing (defined as an inter-allelic
distance of ≤ 0.5 µm) or for preferential associations of
ectopic chromosome sites in diploid plants [21]. In the present
study, we compared chromosome 5 trisomics with triploids, both of which have
three YFP dots and three DsRed dots in the context of a chromosomally unbalanced
or balanced genome, respectively (Figure 1). We examined whether the extra copy of chromosome 5 in
trisomics produced any distinctive arrangements of chromosome 5 fluorescent tags
that differed from those observed in the triploid genome.
Six distances – connecting the three YFP dots and the three DsRed dots
– were measured in selected root nuclei in which fluorescent signals
were visible (Figure S5). In sibling triploid and trisomic seedlings of the F2
generation, the distances between the YFP dots and DsRed dots usually differed
within a given nucleus and considerable inter-nuclear variability in distance
measurements was observed for both fluorescent tags (Table
S2A,B). Thus, in both trisomics and triploids, chromosome 5 fluorescent
tags display similar random arrangements. In trisomics, however, we observed an
increased incidence of inter-allelic distances around 0.5 µm (Table S2B).
Although these results might suggest enhanced allelic pairing in trisomics, they
could also reflect the generally smaller inter-allelic distances in these plants
(Table
S2), which in turn is probably due to smaller nuclei in trisomics than in
triploids [21]. The possibility of enhanced allelic
associations in trisomics was supported, however, by 3D reconstructions of
nuclei, which indicated that two of the three alleles of either
DsRed or YFP were more likely to be close to
each other in trisomics than in triploids (group I, Table S2;
Figure
S6). A similar trend was observed in trisomic F3 progeny; however,
analysis of these plants was compromised by problems with epigenetic silencing
of the LacI-DsRed and TetR-YFP transgenes and
by the lack of F3 triploid siblings for comparison (Table S1B
and data not shown).
Although the analysis has involved a limited number of root cell nuclei, it
appears that the presence of an extra chromosome 5 in unbalanced trisomics does
not substantially alter the interphase arrangement of chromosome 5 fluorescent
tags as compared to those observed in chromosomally balanced triploids. A subtle
difference, however, may be a slightly enhanced tendency for two copies of the
triplicated chromosome to be more closely apposed, at least partially along
their lengths, in trisomics than in triploids. This possibility can be studied
in the future with a larger set of trisomic plants and the use of emerging
strategies that minimize silencing of the reporter transgenes [22].
Our studies on the influence of chromosome 5 triplication on chromosome
structural stability, gene expression, and interphase arrangement of chromosome
5 fluorescence tags in Arabidopsis have demonstrated that
trisomy 5 disrupts the genome in a number of ways:
1. Chromosome structural stability: Truncated derivatives of the triplicated
chromosome 5 were regularly observed in trisomic plants. The triplicated
chromosome may be vulnerable to breakage, particularly in vicinity of repetitive
regions, and a truncated chromosome is more likely to be retained when two
intact copies are present. The possibility of structural as well as numerical
deviations in aneuploids underscores the need to perform array CGH for proper
analysis and intepretation of the transcriptome data [31]. The formation and
inheritance of chromosome structural variants in aneuploids might have
evolutionary implications if restructured chromosomes are transmitted to progeny
and eventually fixed in the population [32]. Enhanced
structural instability of aneuploid genomes in somatic cells could have
relevance for human cancer cells, which display progressive chromosome numerical
and structural changes as the tumour evolves [7],[23].
2. Complex changes in gene expression: The transcriptome analysis revealed that
the expression of many genes is affected in chromosome 5 trisomics, primarily on
the triplicated chromosome (cis effects) but also on non-triplicated chromosomes
(trans effects). Most genes on chromosome 5 genes showed higher expression
reflecting a dosage effect, but cases of apparent dosage compensation and even
down-regulation were also observed. Genes involved in responses to stress and
other stimuli were over-represented among genes differentially regulated
relative to the average chromosome trends, and transcription factors were
over-represented in the trans effects. The use of qRT-PCR to analyze expression
of single genes demonstrated variable expression depending on the chromosome
number and constitution, and on the features of individual genes: As shown with
the epigenetic regulators ROS1 and RDR5, genes
on the same chromosome can vary independently in their expression in different
trisomics. In addition, genes under the control of the same promoter can vary in
their response to triplication, as indicated by the two 35S promoter-driven
transgenes, TetR-YFP and DsRed-LacI, on
chromosome 5. The observed variations in gene expression probably depend on
multiple factors including, but not limited to, changes in the dosages of
regulatory molecules and epigenetic factors, and sensitivity of repetitive
regions to copy number changes and gene silencing mechanisms. Transcriptional
changes resulting from aneuploidy must be described in terms of chromosomes
and/or chromosome regions that are numerically altered and whether changes in
expression are in cis or trans regions. Clearly, the choice of microarray data
analysis methods has a substantial impact on results and, in particular,
normalization methods that are robust to large-scale shifts in gene expression
need to be applied in studies of aneuploidy. Although not studied here, cell and
tissue-type differences in gene expression in a given aneuploid might also be
expected [15].
3. 3D organization of fluorescent-tagged sites: Overall, chromosomally unbalanced
trisomics and balanced triploids display equally random interphase arrangements
of fluorescent tagged sites on chromosome 5; however, there may be a slight
tendency for two transgene alleles on the triplicated chromosome to be more
closely associated in trisomics than in triploids. If such associations occur
regularly in trisomics, they might help to induce dosage compensation mechanisms
[33] or spatially compensate for the extra chromosome
in interphase nuclei.
Aneuploidy is usually studied for its developmentally detrimental or pathological
consequences but it also may be important in normal contexts. Recent work has
identified a significant fraction of aneuploid cells in the normal brain
although their physiological significance is not yet known [34]. Given the
strong effect of aneuploidy on global gene expression patterns, it is
conceivable that the formation of aneuploid neurons increases the phenotypic
variability of these cells and their capacity to perform diverse neural
functions.
The plant material in all experiments was Arabidopsis thaliana
landrace Col-0 (the accession used for the design of the ATH1 array). The
transgenic line with YFP and DsRed fluorescent tags on chromosome 5 was
described previously [21]. Seeds were germinated on sterile, solid
Murashige and Skoog medium in plastic petri dishes. Root nuclei in living
seedlings were monitored for YFP and DsRed fluorescence signals as detailed in
previous reports [21],[22]. Seedlings were
then transferred to pots containing a mixture of Huminsubstrat N3 and Vermiculit
Nr.2 (2∶1 v/v) (purchased from a local supplier), and placed in a
culture room with natural light (3000 lux). The photoperiod was 16 h and
temperature was maintained at 23°C. Single leaves were cut from the
plants at a stage of approximately ten rosette leaves (>1 cm in length),
except for plants with extreme aberrant phenotypes, which late were found to
contain an extra copy of chromosome 1. The first cut leaf was selected for RNA
and the second for DNA isolation in order to minimize wounding effects.
Seedlings were treated with colchicine to produce tetraploid progeny according to
an unpublished protocol (Ramon Angel Torres Ruiz, personal communication).
Metaphase chromosome counts were performed using pistil material as described in
protocols 5.2 and 5.3 in a previous publication [35].
Inter-allelic distances and 3D arrangements of fluorescent tagged sites on
chromosome 5 in root interphase nuclei of living, untreated seedlings were
determined using fluorescence microscopy as described previously [21],[22]. The tagged sites
harbor transgene complexes that encode repressor protein-fluorescent protein
fusions proteins (either Tet-YFP or DsRed-LacI) as well as arrays of either
tet or lac operator repeats, to which the
respective repressor protein-fluorescent protein fusion protein can bind [21],[22].
Isolation of genomic DNA (DNeasy mini kit, Qiagen, Hilden, Germany), biotin
labelling of DNA (BioPrime DNA labelling, Invitrogen, Lofer, Austria), and gDNA
hybridization were performed as described [36]. The DNA
concentration was quantified by spectrophotometry (Nanodrop ND-1000; Peqlab,
Erlangen, Germany) and adjusted for gDNA hyridization to 15 µg. ATH1
microarrays were scanned with an Affymetrix GC3000 system and analysed with GCOS
version 1.4 (Affymetrix, High Wycombe, U.K.). For chromosome copy number
variation the disomic transgenic plant, from which all triploid, tetraploid, and
trisomic plants were derived, served as the reference microarray. The array
signals from the derived plants were scaled in GCOS and compared to the diploid
progenitor. Extra chromosomes or chromosomal deletions were then identified
after sorting for probe sets with a “change p-value” call
“Increase” for supernumerical chromosomes or a
“Decrease” call for deletions. In all cases the default
settings were chosen. After excluding probe sets matching to several gene models
(TAIR7) the remaining probe sets were mapped to the Arabidopsis
chromosomes (chromosome map tool at www.arabidopsis.org).
Typically, extra chromosomes are identified by mapping 95% to
98% of probe sets with an “Increase” call to a
unique chromosome e.g. chromosome 5 in case of chromosome 5 trisomy.
Microarrays were normalized and log transformed by the RMAExpress0.5 tool
(http://rmaexpress.bmbolstad.com/). The log ratios of the signal
values were mapped to their chromosomal position. Data on probe set location was
also extracted from TAIR v7 (see microarray data analysis section). Only probe
sets matching to a unique gene model (TAIR7) were selected.
RNA extraction (RNeasy mini kit, Qiagen, Hilden, Germany) and cDNA synthesis
(RevertAid H Minus First strand cDNA synthesis kit, MBI Fermentas, St. Leon-Rot,
Germany) were performed as described previously [37]. The cDNA was
diluted to 75 µl with DEPC-treated double distilled water, and 2
µl was used in a 20 ul PCR reaction. The mixture was set up with 10
µl of QuantiFast SYBR Green PCR (Qiagen, Hilden, Germany), 2
µl cDNA, and 2 µl of each primer (1 µM final
concentration). PCR was performed after a preincubation as suggested by the
supplier (95° C for 5 min) by 40 two-step cycles of denaturation at
95° C for 10 s, and annealing/extension at 60° C for 30 s. The
comparative threshold cycle (Ct) method was used to determine relative RNA
levels (User Bulletin no. 2, Applied Biosystems). GAPC-2 (At1g13440) was chosen
as the internal reference gene (see also [38] for a
comprehensive analysis of reference genes), and expression levels are relative
to a randomly chosen disomic plant. Sequence of the primer sets are shown in
Table
S3.
Total RNA was extracted from rosette leaves (>1 cm in length) using an
RNeasy mini kit (Qiagen, Hilden, Germany). Transcriptomes were analysed using 1
µg of total RNA as starting material. Targets were prepared with the
one-cycle cDNA synthesis kit followed by biotin-labelling with the IVT labelling
kit (GeneChip One-cycle target labelling and control reagents, Affymetrix, High
Wycombe, U.K.) and hybridized for 16 h as recommended by the supplier (Gene
expression analysis manual, Affymetrix). All transcriptome data (CEL and CHP
files) were submitted to a public repository database (http://www.ebi.ac.uk/microarray/, ArrayExpress accession number:
E-MEXP-1454.
|
10.1371/journal.pntd.0001574 | Implementing Preventive Chemotherapy through an Integrated National Neglected Tropical Disease Control Program in Mali | Mali is endemic for all five targeted major neglected tropical diseases (NTDs). As one of the five ‘fast-track’ countries supported with the United States Agency for International Development (USAID) funds, Mali started to integrate the activities of existing disease-specific national control programs on these diseases in 2007. The ultimate objectives are to eliminate lymphatic filariasis, onchocerciasis and trachoma as public health problems and to reduce morbidity caused by schistosomiasis and soil-transmitted helminthiasis through regular treatment to eligible populations, and the specific objectives were to achieve 80% program coverage and 100% geographical coverage yearly. The paper reports on the implementation of the integrated mass drug administration and the lessons learned.
The integrated control program was led by the Ministry of Health and coordinated by the national NTD Control Program. The drug packages were designed according to the disease endemicity in each district and delivered through various platforms to eligible populations involving the primary health care system. Treatment data were recorded and reported by the community drug distributors. After a pilot implementation of integrated drug delivery in three regions in 2007, the treatment for all five targeted NTDs was steadily scaled up to 100% geographical coverage by 2009, and program coverage has since been maintained at a high level: over 85% for lymphatic filariasis, over 90% for onchocerciasis and soil-transmitted helminthiasis, around 90% in school-age children for schistosomiasis, and 76–97% for trachoma. Around 10 million people have received one or more drug packages each year since 2009. No severe cases of adverse effects were reported.
Mali has scaled up the drug treatment to national coverage through integrated drug delivery involving the primary health care system. The successes and lessons learned in Mali can be valuable assets to other countries starting up their own integrated national NTD control programs.
| Neglected tropical diseases (NTDs) are a group of chronic infections that affect the poorest group of the populations in the world. There are currently five major NTDs targeted through mass drug treatment in the affected communities. The drug delivery can be integrated to deliver different drug packages as these NTDs often overlap in distribution. Mali is endemic with all five major NTDs. The integrated national NTD control program was implemented through the primary health care system using the community health center workers and the community drug distributors aiming at long-term sustainability. After a pilot start in three regions in 2007 without prior examples to follow on integrated mass drug administration, treatment for the five targeted NTDs was gradually scaled up and reached all endemic districts by 2009, and annual drug coverage in the targeted population has since been maintained at a high level for each of the five NTDs. Around 10 million people received one or more drug treatments each year since 2009. The country is on the way to meet the national objectives of elimination or control of these diseases. The successes and lessons learned in Mali are valuable assets to other countries looking to start similar programs.
| Neglected tropical diseases (NTDs) are a group of diseases that affect the most vulnerable and the poorest group of the populations in the world [1], [2]. The World Health Organization (WHO) recommends five public health strategies for the prevention and control of the NTDs: preventive chemotherapy (PCT); intensified case management; vector control; provision of safe water, sanitation and hygiene; and veterinary public health [1]. The major NTDs currently being targeted through PCT include lymphatic filariasis (LF), onchocerciasis, schistosomiasis, soil-transmitted helminthiasis (STH, including ascariasis, hookworm infection and trichuriasis) and trachoma. These five major NTDs cause high disease burden with severe disfigurement, disability and blindness, blighting the lives of more than one billion people worldwide and threatening the health of millions more [1]. The drugs needed for these five NTDs are robust, safe, low-cost and available by donation from the pharmaceutical companies or by purchasing at relatively low costs [3]. They can be delivered to the target populations either alone or in combination to prevent morbidity caused by these NTDs, or in some cases, to eliminate the diseases [4], [5].
Mali is landlocked in West Africa with a population of 15.5 million. It is divided into eight administrative regions (Kayes, Koulikoro, Sikasso, Segou, Mopti, Tombouctou, Gao and Kidal) and Bamako. The northern part of the country extends deep into the Sahara desert and the southern region features the Niger and Senegal rivers, where the majority of the country population inhabits. The country's economy centers on agriculture and fishing. Mali is one of the poorest countries in the world and ranked 160 out of 169 countries according to the Human Development Report 2010 [6]. It is endemic with all five major NTDs [7]–[11]. Control of the NTDs before 2007 was through four independent vertical national programs under the Ministry of Health (MoH): the National Onchocerciasis Control Program (PNLO), the National Lymphatic Filariasis Elimination Program (PNEFL), the National Schistosomiasis and Soil-Transmitted Helminths Control Program (PNLS) and the National Blindness Prevention Program (PNLC).
Onchocerciasis was originally prevalent in five regions in the country, including Kayes, Koulikoro, Sikasso, Segou and Mopti, and the PNLO was established in 1986 to address the public health implications of the disease. The eastern part of the endemic regions (Koulikoro rive droite, Sikasso, Ségou and Mopti) was included in the original program area of the Onchocerciasis Control Program (OCP). In 2002 onchocerciasis was declared eliminated as a public health program in large parts of these areas with only epidemiological and entomological surveillance continuing to monitor the prevalence and microfilarial load in the population and to also monitor the infectivity of the vector Simulium damnosum. The western part of the endemic regions (Kayes and Koulikoro rive gauche) was included in the western extension of OCP in 1987 with ivermectin (IVM, donated by Merck & Co.) administration and later with Community Directed Treatment with Ivermectin (CDTI) with support from the African Program for Onchocerciasis Control (APOC) and using the community drug distributors (CDDs). The disease is currently endemic in 17 districts (Sikasso, one of the original 16 districts, was split into two separate districts in 2010) in three regions in Kayes, Koulikoro and Sikasso.
LF, caused by Wuchereria bancrofti, is endemic throughout Mali [10], [12] with the entire population being at risk of disease. The PNEFL was established in 2004 and subsequently a national mapping survey was carried out using Immunochromatographic Test cards confirming LF endemicity across Mali (Dembélé, unpublished data). The MDA for LF started in 2005 in four of the five onchocerciasis districts in Sikasso using CDTI plus albendazole (ALB, donated by GlaxoSmithKline), with support from the Government of Mali.
Both urogenital (caused by Schistosoma haematobium) and intestinal (caused by S. mansoni) forms of schistosomiasis are present in the country [13]. Two national surveys were conducted with the first in 1984–1989 and the second in 2004–2006 [7], [14], [15]. The results confirmed presence of schistosomiasis throughout the country with geographically varying degrees of prevalence. The later survey in 2004–2006 showed a prevalence of 38.3% (ranging 0.0–99.0%) for S. haematobium and 6.7% (ranging 0.0–94.9%) for S. mansoni [15]. Schistosomiasis control in Mali was initiated in the Bandiagara district, Mopti as a component of a dam-building project in 1978 and became a national program (PNLS) in 1982 [13], [15]. The initial control program with praziquantel (PZQ) distribution was implemented by the MoH in collaboration with WHO and with support from the German Technical Co-operation [13], but the MDA ceased later due to lack of further funding. In 2005, the MDA resumed with support from the Schistosomiasis Control Initiative (SCI) with PZQ procured from certified generic manufacturers, targeting school-age children and at-risk adults with PCT through school-based and community-based drug delivery in all endemic regions and Bamako (school-age children only) [16], [17].
STH is a public health problem throughout Mali. The national survey in 2004–2006 (together with schistosomiasis) in school children from 7–14 years of age showed that the whole country is endemic for STH, with prevalence of up to 34.3% with hookworm infection (in Yorosso, Sikasso) (R Dembélé, unpublished data). STH control consists of several drug delivery platforms in Mali. The National Intensified Nutrition Weeks (SIAN, French acronym) deliver vitamin A and ALB twice a year to children of 12–59 months and to women immediately post-partum. In 2004, the PNLS was expanded to include STH, and ALB was delivered at the same time through school-based and community-based drug delivery to those receiving PZQ treatments for schistosomiasis during 2005–2007 with the support from the Schistosomiasis Control Initiative. The population above 5 years also benefits from annual treatment with ALB and IVM from the LF elimination program.
Trachoma as a blinding disease is found in all districts of the eight regions of the country (except Bamako). A national survey in all regions except Bamako was conducted in 1996–1997 [11]. The prevalence of active trachoma, follicular (TF) or intense (TI), was estimated to be 34.9% among children under 10 years of age, and the prevalence of trichiasis among women over 14 years of age was 2.5%, and 1% had central corneal opacity [11]. The PNLC initiated a trachoma control program in 1998 following the national survey adopting the WHO recommended SAFE (Surgery, Antibiotics, Facial washing and Environmental improvement) strategy [18], [19], benefiting from the Zithromax (ZTM) donation program by Pfizer Inc. Significant progress had been achieved in trachoma control since the start of the national program [19]–[21].
As one of the five ‘fast-track’ countries supported by the United States Agency for International Development (USAID) NTD Control Program managed by RTI International [22], Mali launched the integrated national NTD Control Program (NTDCP) in 2007 with technical assistance initially from International Trachoma Initiative (ITI, 2007) and then from Helen Keller International (HKI) from 2008 onward. The overall objectives of the Mali's NTD control program are to eliminate LF, onchocerciasis and trachoma as public health problems and to reduce morbidity caused by schistosomiasis and STH through regular mass drug administration (MDA), with specific objectives of achieving 80% program coverage and 100% geographical coverage yearly within the five-year program plan. This current paper serves as a report on the progress made by the integrated national NTDCP in Mali, drawing from objectives achieved, documented experiences and pertinent lessons learned of the program from 2007 to 2011, and focusing on only aspects of integrated MDA activities.
The existing disease-specific vertical national programs achieved various degrees of coverage throughout the country and mapping of distribution of each NTD was almost complete before integration. These disease-specific national control programs already achieved significant success before 2007 as described in the introduction. Integration of control activities on certain diseases already occurred, e.g. onchocerciasis and LF, and schistosomiasis and STH, on co-delivery of drugs. Building on these successes, in 2007 Mali began to further integrate the existing disease-specific control programs to increase efficiency and program coverage for each target disease. The USAID funds support all the integrated PCT-related activities and procurement of PZQ. Although the integrated NTD control program include other non-MDA components, this paper focuses on the implementation of MDA component only.
The NTDCP is led by the MoH through the National Directorate of Health. The National Steering Committee of the program was established and is chaired by the National Director of Health and its members include members of the Technical Coordinating Committee (TCC, described below), the Head of Planning, Training and Health Information Unit, the Head of Public Health and Safety Division, the Head of the Nutrition Division, the Dean of the Faculty of Medicine, Pharmacy and Odonto-Stomatology (FMPOS), and the representatives of non-governmental developmental organization (NGDO) partners. The Steering Committee meets twice a year to discuss the progress of the program and issues arising.
A National Strategic Plan for integrated control of NTDs (2007–2011) was developed in 2007 as the blueprint to direct the control activities. A new five-year national strategic plan (2012–2016) is being updated and finalized.
Under the National Directorate of Health, Division of Disease Control and Prevention (DPLM) is responsible for coordinating the activities of control and elimination of priority diseases in Mali. The existing four disease-specific national control programs are under the remit of DPLM, which provides an ideal framework for coordination of integrated NTD control activities. The dedicated NTD coordinator at HKI works closely with the four National Coordinators of the disease-specific control programs to facilitate the integrated activities.
Under the DPLM, the TCC was established and is chaired by the Chief of DPLM, comprising four National Coordinators of the disease-specific control programs, the Head of Nutrition Division, the representative from the National Public Health Research Institute, the representative from the National Center of Information, Education and Communication for Health (CNIECS), and the representative of the grantee NGDO (initially ITI and currently HKI). This committee meets every quarter. The program review and planning workshop was conducted annually to review the progress and to plan for the coming year, attended by the TCC members, the Regional Health Directors, and the regional NTD focal persons. The Regional Health Teams in turn planned the MDA activities for each district with the District Health Teams. In Mali, community health centers play a very important role in providing primary health care at local level. Within each district, there are a number of community health centers, each responsible for a number of villages. For long-term sustainability and local capacity building, the NTD control activities were integrated into the primary health care system at local level. Community health center workers (CHCWs) play an important role in the program as their routine health care activities. These CHCWs provided training and supervision of CDDs, and were responsible for drug allocation, treatment data collation in their catchment area, and data reporting to the district health officers. Figure 1 shows the structure of the program.
To integrate the PCT activities of each existing control program, a situation analysis was conducted to map out the overlaps of the disease distribution in each district using the existing disease distribution data. Figure 2 shows the overlapping distributions of the five targeted NTDs in each districts of the country. The PCT strategy for each disease in each district was decided according to the known prevalence of the disease in the district and the WHO PCT guidelines [4].
Drug packages for each district were determined as shown in Figure 3 according to the disease distribution shown in Figure 2 above and the WHO PCT guidelines. There was insufficient evidence and hence lack of clear guidance for combinations of available drugs, therefore, to avoid possible side effects due to combination, different drug packages were delivered in sequential fashion with one week between deliveries, where two or more drug packages were required. For example, where all three packages were required, MDA was organized as ZTM for week 1, ALB/IVM for week 3 and PZQ for week 5. This was also to avoid confusion among the CDDs with managing different dose poles at the same time, considering the relatively low education level in Mali villagers.
Several successful strategies for drug delivery were used by existing disease-specific national control programs, e.g. CDTI for onchocerciasis and LF, school-based and community-based drug delivery for schistosomiasis and STH, and community-based drug delivery for trachoma. Each of these was operating in disease-specific program areas. To scale up each program to a national coverage in the integrated control program, the four existing national programs worked together to plan and coordinate the MDA activities. A number of drug delivery strategies were used in combination in districts to deliver the drug packages by the trained CDDs: 1) School-based distribution by trained school teachers, taking place in schools targeting school-going children; 2) Community-based distribution by CDDs, including door-to-door/household distribution, focal distribution in the market, mosque, or other busy places, and mobile distribution through CDDs travelling on motorbike to households in remote areas, particularly in nomadic zones; and 3) Health center distribution by CHCWs, taking place at the health centers. Before MDA, in villages the trained CDDs work together with village chiefs to register the target population including name, age and sex. They receive drug allocations from community health centers according to the estimated population in each village and take the drugs to the village. CDDs, village chiefs and CHCWs discuss to decide the best drug delivery strategies for each village, mainly using community-based door-to-door distribution and if MDA happens during school terms, school-based distribution as well. In cities/towns, all the above mentioned three strategies are normally used. During MDA, CDDs distribute drugs according to the registered list, and they first confirm that the person has not been treated before treating him/her. The drugs administered are recorded in the register. MDA normally takes 2–3 days in each village. Extensive advocacy was conducted before each round of MDA and sufficient information was given to the general public about the national program. The drugs were voluntarily taken by the persons targeted in the endemic districts.
CDDs in each village were selected by the village and the management team of the community health center, and were used in the program to conduct the MDA activities in communities. The criteria for CDDs include: they were respected by the community; they had ability to read and write; and they were available during the MDA campaigns.
Cascade training for integrated drug administration was carried out throughout the implementation areas. The training sessions started at the regional level and cascaded down to the community level. Training of trainers was organized in the regions and these trainers subsequently trained the CHCWs (as supervisors) at the district level. The supervisors then trained the CDDs at the community health centers. Refresher training was also provided for supervisors and CDDs each year before the MDA campaign started. Table 1 shows the number of people trained or retrained from 2007 to 2011. In view of the usually low educational level of Mali villagers, the NTDCP decided to train the CDDs in the drug administration before each treatment round with different drug packages in order to avoid confusion in CDDs to calculate and administer the drugs using different dose poles. As the national program has matured and in efforts to reduce costs and streamline the program, integrated training is now being introduced. In total, 86,248 persons have so far been trained and retrained across the country.
Advocacy activities undertaken aimed to promote country ownership of the control program through increasing government funding and support to the NTDCP activities and to mobilize resources from existing and potential partners. At sub-country level, advocacy activities were focused on mobilizing support from local authorities at the regional, district and community levels.
Before each campaign, an official notice was sent by the National Director of Health to all Regional Directors of Health to inform them the mass treatment campaign and to request them to achieve the objectives of the program. Several meetings between the various stakeholders (Regional Directorates of Health, the Regional Offices of Education, Social Development, local councils) involved in the control of NTDs were conducted to galvanize interest, support and participation in the campaign. Posters were produced and sent to all health districts and radio and television messages were broadcast to announce the mass treatment campaign.
Meetings with local officials were held to mobilize communities during mass treatment campaign. Images of severe cases of each of the five NTDs were shown, including the short-and long-term signs and symptoms and the treatment available. These meetings also served as means of motivating communities to participate in the mass treatment campaign. These meetings also proved to be effective in mobilizing funds to support CDDs in some districts.
Behavior change communication has been a very important part in the Mali integrated NTD control program. A workshop to develop and harmonize health messages was organized each year. It was followed by the development of audiovisual materials. These messages were broadcast on the various channels during the month immediately preceding the campaigns, and throughout the duration of the campaigns. Posters and banners were also posted strategically during the course of the weeks preceding the campaign. Short documentaries on NTDs and mass treatment campaigns were broadcast on television at least three times during the 20 days preceding the campaign as well as during the campaign. The same schedule was used for broadcasting the radio messages.
Counseling cards on the five NTDs were designed and these cards are used by the CDDs during mobilization and drug distribution to educate people about the disease and the importance of treatment. The cards also contained information for communities to understand the behaviors that could cause or complicate these diseases and the behaviors that could help prevent them from getting the diseases, such as hand washing and face washing. To date, 3,000 counseling cards and 500 posters have been produced.
Data on treatment and serious adverse events (SAEs) in this paper were the CDD-reported data from the NTDCP. During the mass treatment campaign, the CDDs recorded data on drug usage, treatment numbers and SAEs using specific reporting forms. The data were reported to national NTDCP through health reporting system. In 2009, the reported coverage data were verified through a post-PCT coverage survey (details not shown here). In the current paper, to standardize the calculation for all targeted NTDs, national census population was used and population at risk for each NTD was estimated according to the annual projected population figures from the National Directorate of Population, Mali. Eligible population was estimated as the total population at risk for trachoma and 80% of the total population at risk for LF, onchocerciasis, schistosomiasis and STH. The coverage rates were calculated according to the WHO guidelines for drug coverage monitoring, including geographical coverage, program coverage and national coverage [23]. The geographical coverage is the percentage proportion of the targeted districts among the total number of endemic districts for each disease. The program coverage is the percentage proportion of the population treated among the eligible population in the targeted program areas. The national coverage is the percentage proportion of the population treated among the total population at risk in the country.
The cost data were from the HKI program accounts specific for direct expenditure in Mali on the NTD program activities. HKI receives expense receipts after completion of each activity from the NTDCP. The original receipts for all expenses are maintained by HKI, and are spot checked during internal financial reviews as well as during HKI's federally-mandated annual A-133 audit. Expenses, such as vehicle fuel, per diems, and supplies etc. directly incurred during the implementation of program activities, are uniquely coded in HKI's financial system based on the type of activity supported (e.g. training of CDDs, drug transport and distribution, etc.). On a monthly basis, all program expenses are categorized by activities based on these unique codes, and a running cost total is maintained for each activity over the life of the project.
The integrated MDA activities started in 2007. To gain experience of the integrated delivery of different drug packages by the CDDs, the integrated drug delivery started in three regions (Kayes, Koulikoro and Sikasso) which included 24 districts. It was then gradually scaled up to include more regions in the following years to achieve national coverage in 2009. The number of districts covered by MDA each year since 2005 and the cumulative coverage are summarized in Table 2. Onchocerciasis MDA achieved 100% geographical coverage before 2007 and this has been maintained since. Trachoma MDA started in two regions (Kayes and Koulikoro) and already met the program target after three rounds of treatment before 2007. It was gradually expanded to include all other endemic regions in 2009. The significant gain of the integrated NTD program was the scale-up for LF MDA which achieved full national geographic coverage in 2009, and this has since been maintained. The national coverage of LF MDA also benefited STH control throughout the country. MDA for schistosomiasis achieved national coverage for school-age children in 2007, and each endemic district had been targeted two to five times by the end of 2010 according to the endemicity level. In the scarcely populated Kidal region, the mapping of schistosomiasis in this region was not conducted due to the insecurity and will be done later. MDA for schistosomiasis targeting school-age children in this region was delivered based on the historical and clinical knowledge.
With gradual scale-up of geographical coverage, the number of people targeted and treated/retreated each year increased noticeably. The annual treatment numbers for each targeted NTD and the percentage coverage (program coverage and national epidemiological coverage) rates are shown in Table 3 (including data from 2005–06 before integration). Overall, satisfactory program coverage rates had been achieved each year in the targeted areas since 2007 and maintained at high level since 2009, with those for LF, onchocerciasis, STH and trachoma ranging from 76% to over 100%. Although overall program coverage rate for schistosomiasis was relatively lower each year, program coverage rates had been high among school-age children, the main targeted group according to the WHO recommendations. Most notably, the national epidemiological coverage for LF steadily increased over the years to reach over 65%, treating around 10 million people each year since 2009, and this also benefited STH control throughout the country with national epidemiological coverage of 66–75%.
Some minor side effects from taking the drugs had been recorded such as diarrhea and headaches and these were usually dealt with at the community health centers. However, no cases with severe adverse effects have been recorded so far.
The total direct cost of the program in Mali is $3.575 million from the start of the program in 2007 to March 2011, which covers four rounds of drug delivery. The cost shown here does not include the significant contribution from the MoH on housing, logistics, staff salaries etc., and the cost of drugs, which were either donated free of charge or directly procured by RTI. It also does not include any opportunity costs and monetary contributions from local governments, for example, in 2010 the Kayes mayor's office donated an amount of five million francs (CFA) to help motivate the CDDs during the campaign. As expected, the major expenditures were for MDA activities which included training of CDDs, drug transport, storage and administration, and M&E, supervision and annual program reviews (Figure 4).
The concept of integration was to increase efficiency and treatment coverage to deliver the drugs to those who are in need. Since the injection of major funds from the USAID for the integrated NTD control activities, Mali has significantly scaled up MDA coverage for all five targeted major NTDs. Apart from the success in onchocerciasis control already achieved, national geographical coverage has been achieved for LF from 25% in 2006 to 100% in 2009, reaching national coverage of over 65%. After completion of four rounds of MDA, a sentinel site study was conducted in Sikasso and Koulikoro regions in 2010, and will be conducted in other regions accordingly when the criteria are met in each region. The results will be published separately once enough data are collected. Such national coverage of treatment with ALB and IVM for LF greatly benefits the STH control throughout the country. It also continues to benefit the onchocerciasis elimination in the original 16 districts at risk for onchocerciasis. A recent publication from multi-center studies including the Bakoye river focus in Mali showed that the prevalence of onchocerciasis had been reduced to below 1% by 14 years of annual treatment, which confirmed the feasibility of elimination of the disease in Africa [24]. Impact studies in other areas are to be conducted, and Mali is on track to achieve the objective of elimination of onchocerciasis as public health problem which has recently been adopted by the African Program for Onchocerciasis Control as the program objective of onchocerciasis control in Africa [25].
Trachoma control is one of the most successful control programs in the country. Since integration, program coverage rates for trachoma have been very high, ranging from 76% to 97%. The PNLC through the NTDCP has made great progress with determining which districts no longer warrant treatment at district level by conducting impact studies after three to four consecutive rounds of treatment with Zithromax and tetracycline ointment. Currently only 10 districts in the five regions of Gao, Kidal, Koulikoro, Kayes, and Segou still warrant for MDA at district level. The PNLC is working with WHO, HKI, and the Carter Center to pilot a post-endemic surveillance protocol in the districts where trachomatous follicular (TF) prevalence has fallen below 5%. The relatively low annual national coverage for trachoma shown here was mainly due to the progressive starting of the MDA among districts and stopping of MDA in districts that no longer required treatment.
Treatment with PZQ for schistosomiasis at a national level has reached all eight regions plus Bamako with overall program coverage rates ranging from 40% to around 70%. The target population for schistosomiasis treatment in Mali has been primarily school-age children and adults at high risk. Program coverage in school-age children has been maintained over 80% except 72.7% in 2007. According to the WHO guidelines [4], not every district and child require annual treatment, therefore the annual national coverage rates shown here looked relatively low, even though the program coverage for school-age children each year was high. In addition to the USAID funding, Mali also receives support from other partners in country, such as the Organization for the Maintenance of the River Senegal (OMVS) and World Vision International (WVI), whose funds helped to procure PZQ and support distribution costs in the regions of Kayes and Koulikoro.
Integration of vertical national control programs was complex and challenging [26]. As one of the first five ‘fast-track’ countries to implement the integrated program without prior examples to follow, Mali designed and adopted the implementation strategies according to the local context. The program was coordinated by the national NTDCP team and at the local level it was integrated into the primary health care system. The NTD control became a routine activity of the CHCWs. There was a concern that involvement of the CHCWs would impact on their time in responding to curative cares hence interrupting the service provision at the local health facilities [27]. This may have been the case at the beginning of the program; however, this approach would provide the best chance for long-term sustainability of NTD controls [3], [28]. Over 3000 CHCWs countrywide have been trained and retrained, who will be able to provide quality health services at health centers for NTDs once the large-scale MDA is scaled down when the short-term objectives are met. In reality, it is anticipated that such a large-scale intervention would have reduced the demand of the CHCWs' time due to the reduction of morbidity reverted by MDA [29], [30].
Despite the progress of the NTD control program and achievements made, there are a number of difficulties/challenges still in the program and lessons learnt:
Built on the existing success of individual national control programs, the Government of Mali has shown commitment in the control of NTDs in the country. The coordination of NTD control has been integrated at the central level and implementation activities are integrated with the primary health care system at the local communities. With the financial support from the USAID and other donors, Mali has scaled up the drug treatment to a national coverage through integrated drug delivery, with around 10 million people receiving one or more drug packages each year since 2009. With the progress of the program, the focus is now on consolidating the achievements to achieve the goals of eliminating LF and blinding trachoma, perhaps also onchocerciasis, and reducing the morbidity caused by schistosomiasis and STH, in the country by the preset timelines, and on mobilizing resources for the next phase of the NTD control according to the new national strategic plan. Mali's successes and lessons learned can be valuable assets to other countries starting up their integrated national NTD control programs.
|
10.1371/journal.pntd.0002751 | Enhanced Protective Efficacy of Nonpathogenic Recombinant Leishmania tarentolae Expressing Cysteine Proteinases Combined with a Sand Fly Salivary Antigen | Novel vaccination approaches are needed to prevent leishmaniasis. Live attenuated vaccines are the gold standard for protection against intracellular pathogens such as Leishmania and there have been new developments in this field. The nonpathogenic to humans lizard protozoan parasite, Leishmania (L) tarentolae, has been used effectively as a vaccine platform against visceral leishmaniasis in experimental animal models. Correspondingly, pre-exposure to sand fly saliva or immunization with a salivary protein has been shown to protect mice against cutaneous leishmaniasis.
Here, we tested the efficacy of a novel combination of established protective parasite antigens expressed by L. tarentolae together with a sand fly salivary antigen as a vaccine strategy against L. major infection. The immunogenicity and protective efficacy of different DNA/Live and Live/Live prime-boost vaccination modalities with live recombinant L. tarentolae stably expressing cysteine proteinases (type I and II, CPA/CPB) and PpSP15, an immunogenic salivary protein from Phlebotomus papatasi, a natural vector of L. major, were tested both in susceptible BALB/c and resistant C57BL/6 mice. Both humoral and cellular immune responses were assessed before challenge and at 3 and 10 weeks after Leishmania infection. In both strains of mice, the strongest protective effect was observed when priming with PpSP15 DNA and boosting with PpSP15 DNA and live recombinant L. tarentolae stably expressing cysteine proteinase genes.
The present study is the first to use a combination of recombinant L. tarentolae with a sand fly salivary antigen (PpSP15) and represents a novel promising vaccination approach against leishmaniasis.
| More than 98 countries are reported as endemic for leishmaniasis, a vector-borne disease transmitted by sand flies. Drug-resistant forms have emerged and there is an increased need to develop advanced preventive strategies. Live attenuated vaccines are the gold standard for protection against intracellular pathogens such as Leishmania and there have been new developments in this field. The lizard protozoan parasite, L. tarentolae, is nonpathogenic to humans and has been used effectively as a vaccine platform against visceral leishmaniasis in experimental animal models. Correspondingly, pre-exposure to sand fly saliva or immunization with salivary proteins has been shown to protect mice against cutaneous leishmaniasis. Herein, we used DNA/Live and Live/Live prime-boost vaccination strategies against cutaneous leishmaniasis based on recombinant L. tarentolae stably expressing CPA/CPB genes with and without the sand fly salivary antigen PpSP15 in both resistant and susceptible mice models. Assessment of the immune response and parasite burden in vaccinated mice at different time intervals post-challenge demonstrated that combination of recombinant L. tarentolae CPA/CPB with PpSP15 DNA elicits an enhanced protective immune response against cutaneous leishmaniasis in mice. This parasite- and insect vector-derived antigen combination represents an important step forward in the development of new vaccine strategies against Leishmania infections.
| Leishmaniasis is one of the greatest health challenges in nearly 98 countries, contributing to 2 million new clinical cases per year in tropical and subtropical regions of the globe [1]. The disease is transmitted by sandflies and is manifested in several clinical forms, mainly cutaneous leishmaniasis (CL), mucocutaneous leishmaniasis (MCL), and visceral leishmaniasis (VL) [2]. The geographical spread of the various clinical forms depends on vector availability. For instance, over 90% of CL cases occur in Afghanistan, Algeria, Brazil, Iran, Peru, Saudi Arabia and Syria; while, 95% of VL cases are found in Bangladesh, India, Nepal, Sudan, Ethiopia and Brazil [3]. High treatment costs, toxicity of drugs, and the constant emergence of parasite resistance highlight the need for a vaccine. Despite the observation that individuals with a healed primary Leishmania infection are protected against reinfection, no effective vaccine has been developed thus far. Lack of success may be due to our incomplete understanding of the control and regulation of immune responses during infection/reinfection and the mechanisms involved in the development of immune memory. In humans, acquired resistance to L. major infection is mediated primarily by cellular immunity, in particular antigen-specific type 1 T helper (Th1) cells. Similarly, Th1 dependent protection is observed in mouse experimental models of L. major infection. Most efforts for antigen identification have been focused on parasite proteins. More recently, it was shown that immunization with defined sand fly salivary proteins confers protection against leishmaniasis [4]. This suggests that salivary molecules can contribute to protection as a component of an anti-Leishmania vaccine.
Live attenuated vaccines are the gold standard for protection against intracellular pathogens. Importantly, there have been some recent attempts using this approach for the development of Leishmania vaccines [5], [6]. Other approaches manipulate the Leishmania genome to engineer genetically modified parasites by introducing or eliminating particular virulence genes [7], [8], [9]. These approaches are powerful alternatives for the development of new generation vaccines against leishmaniasis. Nonpathogenic to humans Leishmania strains are also being assessed as promising vaccine tools [10]. Vaccination with a L. tarentolae recombinant strain expressing select immunogenic components of L. infantum, including the A2 and the cysteine proteinases A and B (CPA/CPB) genes as a tri-fusion conferred protection against L. infantum infection [11].
In the present study, we evaluated the efficacy of a new prime-boost vaccine combination consisting of a live recombinant nonpathogenic parasite and a vector salivary protein in eliciting a more powerful protective immunity against L. major infection. For this, we combined a recombinant L. tarentolae expressing the CPA/CPB cysteine proteinases with the immunogenic sand fly salivary molecule PpSP15 delivered as a DNA vaccine. We used different prime-boost regimens and evaluated the immunogenicity and protective effectiveness of this novel vaccine combination against L. major infection in both BALB/c and C57BL/6 mice.
All mouse experiments including maintenance, animals' handling program and blood sample collection were approved by Institutional Animal Care and Research Advisory Committee of Pasteur Institute of Iran (Research deputy dated October 2010), based on the Specific National Ethical Guidelines for Biochemical Research issued by the Research and Technology Deputy of Ministry of Health and Medicinal Education (MOHM) of Iran that was issued in 2005.
All solutions were prepared using MilliQ ultrapure (Milli-QSystem, Millipore, Molsheim, France) and non-pyrogenic water to avoid surface-active impurities. G418, and Sodium dodecyl sulfate (SDS) were purchased from Sigma-Aldrich (Sigma, Deisenhofen, Germany). The material for PCR, enzymatic digestion and agarose gel electrophoresis were acquired from Roche Applied Sciences (Mannheim, Germany). Cell culture reagents including M199 medium, HEPES, L-glutamine, adenosine, hemin, gentamicin, DMEM and Schneider were purchased from Sigma (Darmstadt, Germany) and Gibco (Gibco, Life Technologies GmbH, Karlsruhe, Germany), respectively. Fetal Calf Sera (FCS) was purchased from Gibco (Gibco, Life Technologies GmbH, Karlsruhe, Germany). All cytokine kits were purchased from DuoSet R & D kits, (Minneapolis, USA).
A 2.3 kb fragment content CPA/CPB/EGFP fused genes (with stop codons at the end of the EGFP ORF) was digested from pCB6-CPA/CPB/EGFP using SacI and BamHI and then cloned into the corresponding sites of pEGFP-N1 vector (Clontech, Palo Alto, CA) to provide the vector referred to as pEGFP-CPA/CPB/EGFP. After confirmation of the tri-fused gene through PCR and sequence determination, the pLEXSY-NEO2 vector (EGE-233, Jena bioscience, Germany) was used as an integrative vector to incorporate the CPA/CPB/EGFP fusion gene into the genome of the parasite. The CPA/CPB/EGFP was digested from pEGFP-CPA/CPB/EGFP using XhoI and XbaI and cloned into NheI and XhoI sites of the pLEXSY vector (XbaI and NheI are isoschizomers and make compatible sticky ends). For integration, the SwaI was used to linearize the vector at the 5′ and 3′ ends. Then the L. tarentolae (Tar II ATCC 30267) was grown in M199 5% inactivated fetal bovine serum (iFBS) to an optimal concentration. Parasite density was measured by counting the cells dissolved in Hyman's solution (HgCl2 0.5 g, NaCl 1 g, Na2SO4, 10H2O 11.5 g) using a hemocytometer. The pellet was resuspended in ice-cold electroporation buffer (21 mM HEPES, 137 Mm NaCl, 5 mM KCl, 0.7 mM Na2HPO4, 6 mM glucose; pH 7.5) to a final density of 108 parasites/ml, as recommended [12]. A total of 4.0×107 parasites/300 µl were mixed with 5–10 µg linearized DNA for stable transfection in a 0.2 cm electroporation cuvette (BioRad, USA) and stored on ice for 10 min. Electroporations were performed twice at 450 V, 500 µF using Bio Rad Gene Pulser Ecell device (Bio-Rad, USA) and the cell suspension was immediately put on ice for 10 min. Electroporated parasites were then transferred to 3 ml complete M199 media supplemented with 10% iFBS free of antibiotic and incubated at 26°C for 24 hours. Then, the live parasites were collected by centrifugation at 3000 rpm for 10 min at 4°C. Cells were subsequently transferred onto semi-solid plates of M199 medium containing 50% Noble agar (Difco, USA) and 50 µg/ml G418 (Gibco, Germany) and incubated at 26°C. The genotype of transfected parasites was confirmed by Southern blotting using the EGFPORF through incorporation of radiolabeled dCTP in a PCR reaction. In addition, genomic DNA obtained from transfected and wild type (WT) cells was amplified by PCR with specific primers to the upstream and downstream of the flanking region of 5′SSU. Forward primer (F3001) anneals upstream of the 5′SSU on WT genome and reverse primer (A1715) anneals to the backbone of the vector, downstream of 5′SSU and upstream of CPA/CPB/EGFP gene. The sequences for primer F3001 are: 5′ GAT CTG GTT GAT TCT GCC AGT AG 3′ and for primer A1715: 5′ TAT TCG TTG TCA GAT GGC GCA C 3′. The expression of CPA/CPB in the recombinant parasites was confirmed by RT-PCR, Western blot as well as flow cytometry.
The gene coding for PpSP15 (NCBI accession number: AF335487) from the NH2 terminus to the stop codon was amplified from P. papatasi SP15-specific cDNA by PCR as reported previously [13] and cloned into the TOPO TA cloning vector PCRII (Invitrogen). The plasmid VR1020-SP15 was purified using the Endo Free Plasmid Mega kit according to the manufacturer's instructions (QIAGEN, Germany).
Frozen and thawed (F/T) L. major and L. tarentolae CPA/CPB/EGFP antigens were prepared from stationary phase promastigotes. Parasites were washed with PBS (8 mM Na2HPO4, 1.75 mM KH2PO4, 0.25 mM KCl, 137 mM NaCl) prior to 10 times exposition to liquid nitrogen and 37°C water bath alternately. The rCPA and rCPB were also prepared as previously reported [14]. Protein concentrations were measured with a BCA kit (PIERCE, Chemical Co., Rochford III). For preparation of salivary gland homogenate (SGH), P. papatasi females, Israeli strain, were used for dissection of salivary glands 3–7 d after emergence as previously described [15]. Briefly, salivary glands were disrupted by ultra-sonication and centrifuged at 10,000 g for 3 min and the resultant supernatant was dried in a Speed Vac (Thermo Scientific) and reconstituted before use in the listed experiments.
Female BALB/c (H-2d) and C57BL6 (H-2b) mice (6–8 weeks old, weighting 20±5 g) were purchased from the Pasteur Institute of Iran animal breeding facilities. All animals were housed in plastic cages with free access to tap water and standard rodent pellets in an air-conditioned room under a constant 12∶12 h light-dark cycle at room temperature and 50–60% relative humidity. Six groups of BALB/c or C57BL/6 mice (n = 20 per group) were vaccinated in different prime/boost modalities given three weeks apart in the right hind footpad (Table 1). These included, G1: vaccination with L. tarentolae CPA/CPB/EGFP+ and boosting with L. tarentolae CPA/CPB/EGFP+; G2: vaccination with VR1020-SP15 and boosting with L. tarentolae CPA/CPB/EGFP+ followed by VR1020-SP15 the day after; G3: vaccination with L. tarentolae CPA/CPB/EGFP+ followed by VR1020-SP15 the next day and boosting with L. tarentolae CPA/CPB/EGFP+ followed by VR1020-SP15 the next day; G4: Control group, vaccination with PBS; G5: vaccination and boosting with VR1020-SP15; G6: vaccination and boosting with L. tarentolae EGFP+. L. major EGFP+ (MRHO/IR/75/ER) parasites were used for the infectious challenge and were kept in a virulent state by continuous passage in BALB/c mice. The promastigotes were cultured in M199 medium supplemented with 5% iFBS and 40 mM HEPES, 0.1 mM adenosine, 0.5 µg/ml hemin, 2 mM L-glutamine and 50 µg/ml gentamicin at 26°C. For mice challenge, a total of 2×105 stationary phase promastigotes were injected subcutaneously in the left hind footpad 3 weeks after the booster immunization. For G2, G3, G4 and G5, 0.5 pair of sand fly SGH was mixed with parasites and used for challenge.
The profile of cytokine production in the groups vaccinated with L. tarentolae CPA/CPB/EGFP+ (G1) and a combination of L. tarentolae CPA/CPB/EGFP+ and VR1020-SP15 (G2, and G3) and the PBS-immunized control group G4 in both BALB/c and C57BL/6 mice was measured before challenge and at 3 and 10 weeks post challenge in two independent repeats. Briefly, at each time point, 4 mice from each group were sacrificed. Their spleen was treated with a tissue grinder and red blood cells were lysed for 5 minutes using the ACK lyses buffer (NH4Cl 0.15M, KHCO3 1 mM, Na2EDTA 0.1 mM). Splenic cells were then washed, put in culture at 3.5×106 cells/ml and exposed to recombinant antigens rCPA (10 µg/ml) and rCPB (10 µg/ml), F/T lysate of L. major (15 µg/ml), L. tarentolae harboring cysteine proteinase genes of interest (25 µg/ml), and SGH (2pairs/ml). Cell culture supernatants were collected after 24 hours for IL-2 and TNF-α assays and 72 hours later for IFN-γ and IL-4 assays. Cytokine measurements were performed by Sandwich ELISA using the DuoSet R & D kits as per the manufacturer's instructions. The minimum detection limit is 2 pg/ml for mouse IFN-γ and IL-4, 3 pg/ml for IL-2 and 5 pg/ml for TNF-α. All measurements were run in duplicates for two independent experiments. Concanavalin A (Con A; 5 µg/ml) was used in all experiments as a positive control.
For the groups vaccinated with L. tarentolae CPA/CPB/EGFP+ (G1) and a combination of L. tarentolae CPA/CPB/EGFP+ and VR1020-SP15 (G2, and G3) and the PBS-immunized control group G4, mice were bled to obtain serum for determination of antibody responses. The serum sample obtained from each mouse was analyzed by ELISA for specific IgG1 and IgG2a isotype responses three weeks after booster immunization (against F/T lysate of L. tarentolae CPA/CPB/EGFP+ (10 µg/ml) and SGH (2pair/ml) and at 5 weeks after challenge against F/T lysate of L. major (10 µg/ml) and SGH (2pair/ml). Briefly, 96-well plates (Greiner) were coated with each antigen overnight at 4°C. Plates were blocked with 100 µl of 1% BSA in PBS at 37°C for 2 h to prevent nonspecific binding. Sera (1∶100) were added and incubated for 2 h at 37°C. After three washes, goat anti-mouse IgG1-HPR (1∶10,000, Southern Biotech, Canada) or goat anti-mouse IgG2a-HPR (1∶10,000, Southern Biotech, Canada) were added and incubated for 2 h at 37°C. After four washes, plates were incubated for 30 min at 37°C with Peroxidase Substrate System (KPL, ABTS) as substrate. Reactions were stopped with 1% SDS and the absorbance was measured at 405 nm.
The parasite load in different groups of BALB/c and C57BL/6 mice (G1, G2, G3, G4, G5 and G6) were determined by the limiting dilution assay at 3 and 10 weeks post challenge [16]. Briefly, at each time point 4 mice from each group were taken randomly, sacrificed and the lymph nodes (LN) were excised and weighed. After homogenizing, 20 different serial dilutions (10−1 to 10−20) were prepared in Schneider's Drosophila medium supplemented with 10% iFBS and gentamicin (0.01%). Diluted cells were cultured in 96 well plates in duplicate and investigated 7 and 14 days later for positive wells. The parasite load was calculated using the following formula: −Log10 (last dilution with live parasites/weight of homogenized LN).
To demonstrate the in vivo level of infection, the infected footpad (FP) was imaged 10 weeks after challenge with the KODAK Image Station 4000 Digital Imaging System. Briefly, six BALB/c mice from each group (G1, G2, G3, G4, G5 and G6) were treated with a depilatory substance (Nair) to remove hair from their FPs to reduce background auto fluorescence. Afterward mice were temporarily anesthetized intraperitoneally with a mixture of xylazine 2% (7.5 µl), Ketamine 10% (30 µl) and saline solution (260 µl) per mouse and then imaged. Pixel counting and measurement of the lesions were performed using the KODAK molecular image software version 5.3. Measurements were reported as “Net intensity”, a quantitative measurement defined as the number of green pixels in a given area multiplied by the average intensity of each pixel minus the background intensity.
Statistical analysis was performed using Graph-Pad Prism 5.0 for Windows (San Diego, California). Depending on data passing normality tests, ANOVA or Mann-Whitney U tests were computed. P values less than 0.05 were considered significant. The specific test employed is indicated in each figure.
Expression of the 2.3 kb CPA/CPB/EGFP tri-fusion genes in L. tarentolae is under the regulatory control of the rRNA Pol I promoter. We integrated the CPA/CPB/EGFP fragment flanked by 5′(∼860 bp) and 3′SSU (∼1080 bp) sequences into the rRNA locus of L. tarentolae (Figure 1A). The recombinant L. tarentolae strain expressing CPA/CPB/EGFP genes displayed a normal morphology (a drop-like shape) with a normal length of the flagellum comparable to that of the wild type strain. EGFP expression and intensity were verified by visualization using an epifluorescence microscope. The EGFP was attached to the C-terminal end of CPB and fluorescence is distributed through the whole cytoplasm (Figure 1B). Confirmation of CPA/CPB/EGFP expression at the level of RNA and protein was verified using RT-PCR and western blot, respectively (data not shown).
A major requirement of vaccines is to protect the majority of a population that normally displays a high diversity in MHC haplotypes. It is known that L. major causes a non-healing cutaneous infection in susceptible BALB/c mice characterized by progressive skin lesions and visceralization of the parasites to the spleen [17], [18]. In contrast, C57BL/6 mice are naturally resistant against L. major and the infection normally causes transient symptoms and is self-healing [18]. Therefore, we evaluated the immune response in both BALB/c (H-2d) and C57BL/6 (H-2b) mice in the groups vaccinated with L. tarentolae CPA/CPB/EGFP+ (G1), combination of L. tarentolae CPA/CPB/EGFP+ and VR1020-SP15 (G2, and G3) and the PBS-immunized control group G4 (Table 1). It has been shown that IFN-γ and TNF-α are important parameters for vaccine evaluation since they synergize their capacity to mediate killing of pathogens. Furthermore, IL-2 also enhances the expansion of T cells, leading to a more efficient effector responses [19]. Since these effector cytokines mediate protection, we evaluated antigen specific immune responses three weeks after booster immunization by measuring the production of IFN-γ, IL-4, IL-2 and TNF-α in the supernatant of splenocytes in response to rCPA/rCPB or F/T lysate of L. tarentolae CPA/CPB/EGFP.
In susceptible BALB/C mice, the levels of IFN-γ production by splenocytes after rCPA/CPB stimulation were significantly higher (p<0.05) in the G2 and G3 vaccinated groups compared to the control-immunized group G4 (Figure 2A). No significant difference in the levels of IFN-γ production was observed in any of the vaccinated groups when stimulation was done with L. tarentolae CPA/CPB/EGFP (Figure 2A). We further investigated whether splenocytes from the three different vaccinated regimens secreted the Th2-associated cytokine IL-4. Upon stimulation with rCPA/CPB, G3 exhibited a small but significantly higher level of IL-4 as compared to control group G4 (Figure 2A). Stimulation of splenocytes with L. tarentolae CPA/CPB/EGFP resulted in significantly higher levels of IL-4 in G1, G2 andG3 groups as compared to control group G4 (Figure 2A). Furthermore, G1, G2 and G3 produced significantly higher levels of IL-2 compared to G4 (p<0.05) when stimulated with rCPA/CPB or L. tarentolae CPA/CPB/EGFP (Figure 2A). For TNF-α, G1, G2 and G3 showed significantly higher levels in comparison to control group G4 upon stimulation with rCPA/CPB (Figure 2A) but no difference was observed among these groups after L. tarentolae CPA/CPB/EGFP stimulation (Figure 2A).
The antibody response of vaccinated BALB/c mice against L. tarentolae CPA/CPB/EGFP for groups G1, G2, G3 and G4, and against sand fly salivary gland homogenate (SGH) for groups G2, G3 and G4 was determined before challenge (Figure 2B). Higher levels of both IgG1 and IG2a antibodies against L. tarentolae CPA/CPB/EGFP was observed in vaccinated groups G1, G2 and G3 compared to control group G4 (Figure 2B, p<0.05). In both G2 and G3, the level of IgG2a was higher than IgG1 but the opposite was obtained for G1 (Figure 2B, p<0.05). When SGH was used as antigen, both the G2 and G3 groups produced significantly higher levels of IgG2a as compared to control group G4 (p<0.05) while no significant differences were observed for IgG1 levels in all three vaccinated groups (p>0.05).
For resistant C57BL/6 mice, splenocytes produced significantly higher levels of IFN-γ and IL-4 in the three vaccinated groups (G1, G2, G3) as compared to control group G4 when stimulated with either rCPA/CPB or L. tarentolae CPA/CPB/EGFP (Figure 3A, p<0.05). Nevertheless, the level of IL-4 was lower than that of IFN-γ in all vaccinated groups (Figure 3A). Groups G1, G2 and G3 produced significantly higher levels of IL-2 compared to G4 (p<0.05) when stimulated with rCPA/CPB or L. tarentolae CPA/CPB/EGFP (Figure 3A). As for TNF-α, it was only produced upon stimulation with rCPA/CPB where G1, G2 and G3 showed significantly higher levels in comparison to G4 (Figure 3A).
The three vaccinated groups produced significantly higher levels of IgG2a against L. tarentolae CPA/CPB/EGFP compared to control G4 (Figure 3B). The level of IgG1 was only significantly higher in G3 as compared to control G4 (p<0.05). Furthermore, there were no significant differences between IgG1 and IgG2a levels against SGH in the three tested groups G2, G3 and G4 (Figure 3B).
All six groups of vaccinated and control BALB/c mice (Table 1) were challenged with 2×105 late-stationary phase L. major GFP+ promastigotes in their left footpads in the presence (G2, G3, G4, G5) or absence (G1, G6) of SGH. Weekly measurements showed a sharp increase in footpad swelling in the control groups G4 and G6 at weeks 8, 9 and 10 that was significantly larger than that observed in groups G1, G2, G3 and G5 (Figure 4A p<0.05). As a main parameter, the parasite burden was measured in the lymph nodes of all six groups at 3 and 10 weeks post challenge using a limiting dilution assay (Figures 4B). Three weeks after challenge (3WAC), groups G1, G2, G3 and G5 showed a significantly lower parasite load than groups G4 and G6 (Figure 4B) with G2 and G5 showing the lowest lymph node parasite burden (Figure 4B). At the end of week 10, the parasite burden of groups G1, G2, G3 and G5 remained significantly lower (p<0.05) compared to groups G4 and G6 (Figure 4B). In addition, both G2 and G3 has significantly lower parasite load in respect to G1 and G5 (p<0.05). In vivo imaging of fluorescent parasites in the footpad 10 weeks after challenge (10WAC) shows a significant reduction in the level of fluorescence intensity in the footpad of the vaccinated groups G1, G2 and G3 as compared to the control group G4 (Figures 4C–D). Group G3 had the lowest fluorescence intensity with two mice showing no GFP fluorescence (Figure 4C–D). Moreover, the fluorescence intensity of group G3 was significantly lower in comparison to groups G1, G4, G5 and G6 (p<0.05) but was not statistically significant from that of group G2 (Figure 4D).
For assessment of the immune response after challenge, we focused on the groups vaccinated with L. tarentolae CPA/CPB/EGFP+ (G1) and a combination of L. tarentolae CPA/CPB/EGFP+ and VR1020-SP15 (G2, and G3) compared to the PBS-immunized control group G4. Splenocytes stimulated with L. major F/T antigen at 3WAC show that groups G1, G2 and G3 produced significantly higher levels of IFN-γ compared to control group G4 (Figure 5A, p<0.05). Though group G3 produced higher levels of IFN-γ compared to group G2, it also produced significantly higher levels of IL-4 (p<0.05) as compared to groups G1, G2 and G4 (Figure 5B). The difference in the levels of these two cytokines became less pronounced at 10WAC (Figure 5B). All vaccinated groups showed a positive IFN-γ/IL-4 ratio and group G2 had the highest IFN-γ/IL-4 ratio at 3WAC indicative of a Th1 response (Figure 5C). At 10WAC, group G3 had the highest ratio of IFN-γ/IL-4 (p<0.05) in comparison to G1, G2, G4. With regard to IL-2 production, only G2 produced significantly higher levels of this cytokine as compared to G1, G3 and G4 at 3WAC and 10WAC (Figure 5D). TNF-α production was similar at in all vaccinated and control groups at 3WAC but it was significantly higher in the three vaccinated groups (G1, G2 and G3) compared to control group G4 at 10WAC (Figure 5E, p<0.05).
The specific antibody response against L. major in BALB/c mice was measured in the above-mentioned groups at 5 weeks after challenge. Groups G2 and G3 displayed the highest level of IgG2a and IgG1 antibodies to Leishmania compared to group G1, and control group G4 (p<0.05, Figure 5F). The low levels of IgG1 and IgG2a antibodies to Leishmania were similar in groups G1 and G4 (Figure 5F). Regarding anti-sand fly saliva antibodies, the levels of IgG2a antibodies were significantly higher in groups G2 and G3 compared to control group G4 (Figure 5G). Furthermore, the ratio of saliva-specific IgG2a/IgG1 was greater in groups G2 and G3 (Figure 5G).
In C57BL/6 mice the increase in footpad swelling was similar between groups G1, G4 and G6 (Figure 6A). There was a significant decrease (p<0.05) in footpad swelling in groups G2 and G3 in comparison to groups G1, G4, G5 and G6 (Figure 6A). Measurements of parasite burden from lymph nodes at 3 and 10 weeks post-challenge showed that, group G2 had significantly the lowest parasite burden (p<0.05) as compared to groups G1, G3, G4, G5 and G6 (Figure 6B). We also observed at 10WAC a significant decrease in parasite burden in G1, G3 and G5 as compared to control group G4 as well as G6. In addition, the level of parasite burden in G6 is also significantly lower than G4 (p<0.05) (Figure 6B). Overall, these data demonstrate that in C57BL/6 mice priming with VR1020-SP15 and boosting with a combination of the live vaccine and VR1020-PpSP15 elicited a higher protective efficacy than the two other regimens in controlling footpad swelling and parasite propagation up to 10 weeks post-challenge (Figure 6A, B). Although we were able to detect the swelling in the footpad of all groups, we were unable however, to determine the fluorescent intensity by imaging in the footpad of C57BL/6 mice as it was done for BALB/c. In fact, in C57BL/6 mice resistant strain, the level of parasite propagation in the footpads of all groups was limited (below the instrument detection limit).
Similar to BALB/c mice, we focused on the groups vaccinated with L. tarentolae CPA/CPB/EGFP+ (G1) and a combination of L. tarentolae CPA/CPB/EGFP+ and VR1020-SP15 (G2, and G3) compared to the PBS-immunized control group G4 for assessment of the immune response after challenge. Splenocytes stimulated with L. major F/T antigen at 3WAC show that groups G2 and G3 had higher levels of IFN-γ compared to control group G4 (Figure 7A). At 10WAC, IFN-γ production was similar among the vaccinated groups G1, G2 and G3 and was significantly higher than control group G4 (Figure 7A). The levels of IL-4 were lower in group G2 as compared to group G3 at 3WAC (Figure 7B, p<0.05), however this cytokine decreased significantly 10WAC in group G3 compared to groups G1, G2 and G4. Similar to BALB/c mice, the ratio of IFN-γ/IL-4 was higher in groups G1, G2 and G3 compared to the control group G4, particularly at 10WAC (Figure 7C). It is worth to mention that G3 has significantly the highest ratio in compare to G1 and G2 (p<0.05). As for IL-2, its production was significantly higher (p<0.05) in groups G2 and G3 compared to groups G1 and G4 at 3WAC, but no statistical significance was observed among the groups at 10WAC (Figure 7D). Importantly, the induction of TNF-α was significantly higher in Group G2 as compared to control group G4 at 10WAC and was 2-folds higher compared to G1 and G3 (Figure 7E).
Antibodies against L. major in vaccinated groups G1, G2 and G3 showed significantly higher levels of IgG2a in comparison to group G4 (p<0.05, Figure 7F). Furthermore, the levels of IgG1 were significantly lower than IgG2a in these three vaccinated groups (G1, G2 and G3) in comparison to G4 (p<0.05, Figure 7F). Overall, the antibody response to SGH was low (Figure 7G). With the exception of group G2 that produced significantly higher levels of IgG2a compared to groups G3 and G4, the antibody response to SGH was mixed (Figure 7G).
Despite substantial progress in fundamental Leishmania research, there are many unanswered questions concerning pathogenesis of the disease and the acquisition of protective immunity against reinfection. In this respect, immunization with live attenuated strains as a vaccine tool to induce a protective immune response in the host has a long tradition [20]. The major drawback of this approach is that under certain circumstances, the strains may gain virulence and become pathogenic again. To overcome this problem, subunit vaccines, instead of the whole organism, emerged as a vaccination strategy [21]. A number of parasite antigens have been tested for their potential to induce anti-Leishmania responses. The most extensively studied antigens using a wide range of adjuvants and delivery systems are GP63, LACK, CPs, and the poly-antigen Leish111f [22], [23], [24]. In an attempt to engage Leishmania infection at an early stage, salivary proteins of the sand fly have also been evaluated for vaccination. Studies in mice, hamsters and dogs showed promising results with the induction of Th1-like responses and long-term protection against both cutaneous and visceral infections using these salivary proteins [4], [8].
Here, we describe the outcome of a new vaccination strategy with different modalities using a live recombinant nonpathogenic L. tarentolae vaccine expressing CPA/CPB/EGFP combined to a DNA vaccine containing the cDNA for PpSP15, the predominant 15 kDa salivary protein from the sand fly P. papatasi. Our target parasite antigens are cysteine proteinases, which are conserved among different Leishmania species and are highly immunogenic. L. tarentolae, the lizard protozoan parasite, has been previously introduced by Breton et al. [25] as a candidate vaccine against visceral leishmaniasis. Furthermore, we have demonstrated that a recombinant L. tarentolae strain expressing the L. donovani A2 gene elicited a strong protective immunity against virulent L. infantum challenge [26]. Recently, we have shown that vaccination with L. tarentolae expressing A2/CPA/CPB induced a strong parasite-specific Th1 response and conferred protection against L. infantum challenge in BALB/c mice [11]. As for PpSP15, it was shown previously to protect vaccinated C57BL/6 mice challenged with parasites plus SGH [15], [27].
A major requirement of vaccines in general, is that they are able to protect the majority of a population, which normally displays a high diversity in MHC haplotypes. For this reason, we tested the efficacy of the recombinant live L. tarentolae expressing CPA/CPB/EGFP candidate vaccine combined to PpSP15 DNA in eliciting protective immune responses in two different strains of mice. While BALB/c mice develop progressive lesions upon infection with L. major, C57BL/6 mice are naturally resistant and the infection normally causes transient symptoms (contained lesion development and visceralization) and is self-healing. In this study, L. major IR75 was used for an infectious challenge because it is more virulent in comparison to L. major 39 and the Friedlin strain (Modabber F, personal communication). Both strains of mice showed the strongest protective effect following immunization with a prime/boost regimen based on PpSP15 DNA and recombinant L. tarentolae (groups G2 and G3) demonstrating an enhanced vaccine efficacy compared to the sole use of L. tarentolae CPA/CPB/EGFP (G1) or PpSP15 DNA (G5). While group G3 showed a more potent immune response in susceptible BALB/c mice, group G2 showed the strongest immunogenicity in C57BL/6 mice and it was the best group in controlling parasite growth in the lymph nodes of both mice strains. In both strains of mice, immunization with PpSP15 as a DNA vaccine combined to L. tarentolae CPA/CPB/EGFP showed considerable level of protection as demonstrated by footpad thickness measurements and parasite burden. This demonstrated for the first time the effectiveness of co-immunization of a sand fly salivary protein, PpSP15, with live L. tarentolae CPA/CPB/EGFP in controlling the disease. In the case of BALB/c mice, the effect of Live L. tarentolae CPA/CPB/EGFP is less pronounced although we observed a significantly lower parasite burden in G2 and G3 compared to G1, G4, G5 and G6. Inclusion of PpSP15 DNA as a vaccine may be relevant at two levels: i) as an inducer of adaptive immunity, thus reducing lesion pathology and parasite propagation and ii) as an potential enhancer of innate immunity due to the intrinsic properties of this molecule that may contribute to the control of intracellular growth of L. major. Furthermore, there are extensive data showing that live L. major plus CpG DNA prevents lesion development and causes the specific induction of Th17 cells, which enhances the development of a protective cellular immunity against the parasite [28], [29]. Data presented by Mendez et al. [30] suggest that a vaccine combining live pathogens with immunomodulatory molecules may strikingly modify the natural immune response to infection in an alternative manner to that induced by killed or subunit vaccines. Therefore, it may be possible that PpSP15 working as an immunomodulatory molecule and enhancing the development of a protective cellular immunity against the parasite.
Comparing the data obtained in C57BL/6 with BALB/c, the highest level of TNF-α production, indicative of a Th1 response, was seen with group G2 at 10WAC although there were no significant differences in IFN-γ production. Of note, we only checked four key cytokines to demonstrate the immunogenicity of each vaccine modality using the live recombinant L. tarentolae. We acknowledge the need to further investigate the role or contribution of other cytokines when studying live parasite vaccines. Our future efforts should be also focused on the analysis of the immunological memory and the factors that could correlate with the size of the memory pool using these vaccine strategies. One of these aspects is the induction of CD8 T+ cell responses, which remains to be elucidated.
The concept of using live vaccination against leishmaniasis is not new. Actually, the inoculation of live parasites to produce a lesion that heals, named leishmanization, has been the only vaccination strategy implemented at a large scale because it provides lifelong protection against the development of additional lesions [31]. However, this approach was discontinued because of raised non-healing or slow healing lesions in several human cases [31], [32]. During the last few years, several attenuated strains of Leishmania have been developed. As an alternative, various defined genetically modified parasites have been achieved using a gene targeted disruption strategy through homologous recombination. One of the first examples was the in vivo evaluation showed that the dhfr-ts−/−parasites survived but were unable to establish a persistent infection or to cause disease even in the most susceptible mouse strains [33]. Other examples such as LPG2−/− parasites protected highly susceptible BALB/c mice against a L. major virulent challenge even in the absence of a strong Th1 response [34], [35]. In contrast to L. major mutants, L. mexicana LPG2−/− mutants retained their virulence for macrophages and mice [35], which suggested that different Leishmania species possess alternative virulence repertoires to interact with their host. Therefore, major safety constrains, such as a possible reversion to virulence or reactivation in immunosuppressed individuals, are still among the limiting factors against the use of such vaccines. In contrast to all of the above-mentioned approaches, L. tarentolae is non-pathogenic to humans and can be used in immunocompromised individuals. As such, recombinant L. tarentolae could offer more advantages for vaccine development not only against Leishmania, but also against other pathogens. A recombinant L. tarentolae expressing HIV-1 Gag protein induced strong cell-mediated immunity in mice and decreased HIV-1 replication in an ex vivo system, suggesting that this species can be applied as a promising live vaccine against intracellular pathogens [10]. Recently, a recombinant L. tarentolae strain expressing HPV-E7 antigen-green fluorescent protein (GFP) was developed and showed a potential as a live vaccine against HPV infection [36]. Additionally, modification of, and insertion into, the genome of L. tarentolae can be done easily and there is no insert size limitation making it a versatile tool for vaccine development.
Our data clearly demonstrate that group G2 (prime with PpSP15 DNA and boost with L. tarentolae CPA/CPB/EGFP+PpSp15 DNA) has the lowest level of parasite propagation at 3WAC in both mice strains and at 10WAC in C57BL/6 mice. Therefore, apart from the specific immunogenicity of PpSP15, this salivary protein may have an immunomodulatory role that in combination with a live vaccine potentially enhances its efficacy against Leishmania.
In summary, the present study suggests that this new approach that combines a prime-boosting strategy using recombinant L. tarentolae with a sand fly salivary protein offers a promising platform for developing a more effective vaccine against leishmaniasis.
|
10.1371/journal.pgen.1005940 | The Chromatin Remodelling Enzymes SNF2H and SNF2L Position Nucleosomes adjacent to CTCF and Other Transcription Factors | Within the genomes of metazoans, nucleosomes are highly organised adjacent to the binding sites for a subset of transcription factors. Here we have sought to investigate which chromatin remodelling enzymes are responsible for this. We find that the ATP-dependent chromatin remodelling enzyme SNF2H plays a major role organising arrays of nucleosomes adjacent to the binding sites for the architectural transcription factor CTCF sites and acts to promote CTCF binding. At many other factor binding sites SNF2H and the related enzyme SNF2L contribute to nucleosome organisation. The action of SNF2H at CTCF sites is functionally important as depletion of CTCF or SNF2H affects transcription of a common group of genes. This suggests that chromatin remodelling ATPase’s most closely related to the Drosophila ISWI protein contribute to the function of many human gene regulatory elements.
| CTCF is a transcriptional regulator acting as an insulator element interfering with enhancer function and as a boundary between chromatin domains. CTCF has been shown to organise an exquisite array of phased nucleosomes flanking its binding sites. Here we identified SNF2H as the enzyme primarily responsible for organising the extended arrays of nucleosomes adjacent to CTCF sites. We find that SNF2H acts to maintain the occupancy of CTCF at its binding sites, but does not act as a general loading factor for CTCF’s binding partner cohesin. SNF2H’s action at CTCF sites is functionally important as overlapping cohorts of genes are affected by depletion of CTCF or SNF2H. Other transcription factors also organise nucleosomes and we find that the SNF2H and the related enzyme SNF2L contribute to organising nucleosomes at many of these sites.
| The genomes of eukaryotes exist predominantly as chromatin. The fundamental subunit of chromatin is the nucleosome which consists of 147 bp of DNA wrapped around an octamer of histone proteins [1]. Typically nucleosomes are distributed along DNA with defined spacing at distinct loci in a given cell type [2]. In addition, nucleosomes exhibit distinct translational positioning with respect to certain genomic features such as promoters [3–5], origins of DNA replication [6, 7] and the binding sites for transcription factors such as CTCF [8, 9].
CTCF binding has also been found to play a key contribution to the function of insulator elements [10]. Insulators are genetic elements that act to limit the range over which enhancers can act to regulate a gene [11]. Sites occupied by CTCF are frequently observed to also be enriched for subunits of the cohesin complex [12]. Cohesin is a multi- protein complex consisting of two SMC proteins (SMC1 and SMC3) and Rad21 (Scc1) and STAG (Scc3). It forms a ring like complex capable of encircling two DNA strands [13]. This function for cohesin was originally characterised as playing a key role in the association of newly replicated sister chromatids until they segregate in anaphase. However, subsequently it has become clear that cohesin can also mediate interactions between chromosomal loci during interphase. For example, interactions between cohesin and mediator have been found to mediate looping interactions between promoters and enhancers [14]. The combined action of both CTCF and cohesin mediates long range interactions and effects on gene expression [15–18]. In addition, recruitment of cohesin to CTCF binding sites also contributes to insulator activity [19–21]. However, in some cases CTCF sites remain functional following depletion of cohesin [18, 22].
ATP-dependent chromatin remodelling enzymes have been found to play an important role in establishing the positioning of many nucleosomes within the genomes of model organisms [23]. More recently several studies have addressed the roles of members of this family of ATPases in the human genome. For example the human ISWI related remodelling enzymes SNF2H (also known as SMARCA5) has been found to contribute to DNA repair [24], and in a partially redundant fashion to the organisation of a subset of DNase hypersensitive sites [25]. This study also found that SNF2H and CHD4 associate with a significant number of CTCF binding sites and a previous study demonstrated a role for the enzyme CHD8 at CTCF sites [26]. Both CHD8 and SNF2H have been shown to affect enhancer blocking mediated by CTCF at individual loci [26][27]. More recently, the bromodomain PHD finger-containing transcription factor (BPTF) subunit of the NURF complex has been observed to contribute to localised chromatin accessibility at CTCF sites and the regulation of CTCF target genes [28].
SNF2H is known to function as the catalytic ATPase in at least five distinct complexes in mammalian cells, namely ACF, CHRAC, WICH, RSF and NoRC [29]. The accessory subunits with which the SNF2H ATPase subunit is associated with varies in the different complexes. For example, SNF2H is found in association with WSTF in the WICH complex, with Tip5 in NoRC, Acf1 in ACF, and with both Acf1 and CHRAC 15/17 in CHRAC [29]. The related ATPase SNF2L is the ATPase subunit in the Cerf and NURF complexes [29].
To our knowledge no studies to date have investigated the contribution of different remodelling enzymes to the establishment of organised nucleosomal arrays adjacent to CTCF and other transcription factor binding sites. Here we find that SNF2H plays a major role in the establishment of ordered arrays of phased nucleosomes flanking CTCF binding sites. The related enzyme SNF2L plays a minor role at CTCF sites, and contributes to nucleosome positioning adjacent to other transcription factors. Depletion of SNF2H results in alterations to the expression of many CTCF dependent genes indicating a role for this enzyme in CTCF function and raising the possibility that nucleosome phasing contributes to function at gene regulatory elements.
To investigate the contributions of ATP-dependent chromatin remodelling enzymes in nucleosome organisation, we adopted an siRNA based approach to deplete selected enzymes in cultured HeLa cells. CHD1, CHD2, CHD4 (mi-2), SNF2L and SHF2H could be depleted to between 80% and 96% as judged by western blotting (Fig 1A and S1 Fig). Chromatin isolated from these cells was digested with micrococcal nuclease and the nucleosomal ladder was assessed by gel electrophoresis. Subtle changes in the digestion pattern were observed, but in all cases a distinct species of approximately 150 bp was detected (S1 Fig). In order to characterise the distribution of these nucleosomal DNA fragments, they were subject to high throughput sequencing to a depth of 40–350 million paired reads per repeat.
We investigated how depletion of these enzymes affected the organisation of nucleosomes at the promoters of ubiquitously expressed genes. We noticed variation in distribution of nucleosomes across promoters between experimental repeats and realised that this pattern varied with the extent of MNase digestion (Fig 1B and 1C). The extent of MNase digestion could be assessed from the mean length of the mono nucleosome fragments. With fragments digested to a mean length of 147 bp the nucleosome free region was distinct (Fig 1B). With digestion to 169 bp the nucleosome depleted region is partially filled and the -1 nucleosome more prominent (Fig 1C). Using controls with comparable MNase digestion it was not possible to detect significant changes in nucleosome distribution following depletion of SNF2H, SNF2L (Fig 1B and 1C), CHD1, CHD2 or CHD4 (S1 Fig).
We next investigated the organisation of nucleosomes adjacent to CTCF binding sites where strikingly well organised arrays of around 20 positioned nucleosomes have been reported previously [8](Fig 2A). As expected, the organisation of nucleosomes is dependent on CTCF as siRNA depletion of CTCF reduces the nucleosomal pattern (Fig 2A). While depletion of CHD1, CHD2 and CHD4 had little effect on this pattern (S2 Fig), depletion of SNF2H resulted in a significant loss of nucleosome organisation at these sites (Fig 2B). Depletion of SNF2L had a small effect on the nucleosomes adjacent to the CTCF binding site, with progressively weaker effects at nucleosomes located further away (Fig 2C). As SNF2H is present within multiple distinct remodelling complexes in human cells, we next attempted to distinguish which complexes were involved. siRNA depletion of the ACF1, RSF1, TIP5 and WSTF subunits of these complexes did not disrupt nucleosome organisation to the same extent as observed for SNF2H (S3 Fig). We conclude that different SNF2H containing complexes may function with partial redundancy. SNF2L is known to form a complex with subunits of the human NURF complex including BPTF [30, 31]. Depletion of BPTF resulted in a change to the organisation of nucleosomes immediately adjacent to CTCF sites related to that observed with SNF2L suggesting that SNF2L functions at CTCF sites as a component of the NURF complex (S3 Fig).
As SNF2H affects nucleosome organisation at CTCF sites, we examined whether SNF2H is physically associated with CTCF sites by ChIP. SNF2H is enriched at CTCF sites and enrichment at these sites is reduced following depletion of CTCF (Fig 3A). We also noticed that nucleosome occupancy increased at CTCF sites following depletion of SNF2H (Fig 2B). This led us to investigate whether CTCF occupancy was affected following depletion of SNF2H. The ChIP signal for CTCF was indeed found to be reduced following depletion of SNF2H (Fig 3B). This indicates that in addition to organising nucleosomes adjacent to CTCF, SNF2H acts to maintain high CTCF occupancy.
Given that sites bound by CTCF are often also found to be enriched for cohesin, we investigated the effect of depleting CTCF or SNF2H on ChIP enrichment for the cohesin subunit Rad21. Fig 4A shows that enrichment for Rad21 is reduced by approximately 64% following depletion of CTCF. This is consistent with previous studies showing that recruitment of Rad21 to CTCF sites is dependent on CTCF [12, 19–21]. Enrichment of Rad21 is also reduced following depletion of SNF2H (Fig 4B). The reduction in occupancy (36%) is likely to be attributable to 68% reduction of CTCF occupancy following depletion of SNF2H rather than a direct role for SNF2H in Rad21 loading. We also observe that depletion of Rad21 had no effect on SNF2H recruitment to CTCF binding sites (Fig 4C) and consistent with this, depletion of Rad21 had little effect on nucleosome organisation at CTCF binding sites (Fig 4D).
Previous studies have collated ChIP data characterising the interaction sites for some 119 different transcription factors [32] and this information can be used to align nucleosome distribution adjacent to these factors [9]. Here we select 50 factors for which there are over 1000 genomic binding sites characterised in HeLa cells. Consistent with previous studies we find that binding sites for some factors are located in regions of nucleosome depletion or enrichment without precise positioning of adjacent nucleosomes, whereas other factors such as JUN and RFX5 are flanked by arrays of positioned nucleosomes (Fig 5C and 5D and S4 Fig).
While performing this analysis we observed that by ChIP, we could detect enrichment for CTCF at the binding sites for many transcription factors (Fig 5A and 5B and S4 Fig). We reasoned that in some cases CTCF binding sites are located adjacent to the binding sites for other factors. To test this we filtered out any factor binding sites that had a CTCF binding sites within 500 bp. When only binding sites that did not have CTCF sites within 500 bp were considered, CTCF enrichment at the remaining sites was greatly reduced (Fig 5A and 5B).
We noticed that the effect of filtering out adjacent CTCF sites had differing effects on the organisation of nucleosomes. RFX5 sites that do include adjacent CTCF sites have well organised arrays of nucleosomes (Fig 5C, red). In contrast the RFX5 sites that are not adjacent to CTCF sites have less well organised adjacent nucleosomes (Fig 5C, blue). For RFX5 this effect is very significant as 38% of RFX5 sites are within 500 bp of a CTCF site. Depletion of CTCF significantly perturbs the organisation of nucleosomes at RFX5 sites with adjacent CTCF sites (Fig 5E). This shows that the correlation between the presence of adjacent CTCF sites is functionally significant for nucleosome organisation. CTCF also contributes to the recruitment of cohesin at RFX5 sites as this is reduced following CTCF depletion (Fig 5H). However, the proportion of Rad21 that remains associated following depletion of CTCF indicates that RFX5 is capable of recruiting some cohesin independently of CTCF.
In contrast to the observations at RFX5 sites, the nucleosomes distal to JUN sites are affected in a more complex way. The two nucleosomes immediately adjacent to JUN sites are better organised when there are nearby CTCF sites whilst the extended array of nucleosomes extending beyond the third nucleosome is less ordered as assessed by the depth and periodicity of the normalised read depth (Fig 5D). CTCF depletion results in a modest improvement to nucleosome organisation at JUN sites with adjacent CTCF (Fig 5F). JUN sites lacking adjacent CTCF sites show less change of the distal nucleosomal array following CTCF depletion (Fig 5G). Depletion of CTCF has only a minor effect on Rad21 ChIP at JUN sites indicating that JUN can organise cohesin independently (Fig 5I).
The effects of adjacent CTCF sites observed at RFX5 sites are also observed at the binding sites for other transcription factors. For example nucleosomes are better organised adjacent to the binding sites of factors such as BRCA1 and GTF2F1 that have adjacent CTCF sites (S4 Fig). Enrichment of cohesin is often affected in a similar way (S5 Fig). This illustrates a pitfall in the use of averaging to study correlations in the distributions of chromatin associated factors at complex regulatory elements which are likely to include binding sites for many different factors. For this reason we consider only factor binding sites that do not have adjacent CTCF sites for the subsequent analysis.
To investigate the involvement of SNF2H and SNF2L in nucleosome organisation at different factor binding sites, we plotted the organisation of nucleosomes following depletion of each enzyme flanking the binding sites for 50 different transcription factors (S6 Fig). As at promoters significant differences in organisation were observed with different levels of MNase digestion. However differences in chromatin organisation are apparent when compared to control digestions with similar nucleosome fragment lengths. Depletion of SNF2H has effects on nucleosome organisation surrounding binding sites of factors such as JUN (Fig 6B). The effects are most pronounced for nucleosomes distal to the factor binding site. For example the nucleosomes distal to the +3 nucleosome are less well organised at JUN sites following SNF2H depletion (Fig 6B). Similar effects are observed surrounding 24 additional transcription factors (S6 Fig). Depletion of SNF2L was observed to result in a small reduction in occupancy of nucleosomes proximal to a subset of factor binding sites (Fig 6A and 6C and S6 Fig).
For all factors where SNF2H or SNF2L depletion was observed to affect the generation of extended nucleosomal arrays, SNF2H or BPTF were also observed to be present by ChIP (S5 Fig). For example at JUN sites there is enrichment for SNF2H and BPTF by ChIP. However, enrichment for SNF2H and BPTF was also observed at some factor binding sites where nucleosomes are not well organised, for example GTF3C2 (S4 and S5 Figs). Binding of SNF2H was not enriched at all sites, for example as observed at E2H2 and FAM48A sites (S5 Fig) and there is minimal nucleosome organisation at these sites.
To investigate the functional significance of SNF2H dependent phasing of nucleosome arrays we compared the effects of depleting CTCF and SNF2H. Approximately 1000 genes were significantly affected by the transient depletion of either protein. Many of the up and down regulated genes are affected similarly by depletion of CTCF or SNF2H (Fig 7A). This overlap is highly statistically significant with P values lower than 10−50. The most probable explanation for this is that SNF2H is required for the function of a significant subset of CTCF sites. As SNF2H affects both CTCF occupancy and nucleosome positioning it is difficult to distinguish which is dominant. However, it is possible to identify cohorts of genes where CTCF occupancy was either unchanged following SNF2H depletion or reduced. At the genes where CTCF is retained nucleosome positioning is reduced following SNF2H depletion (Fig 7B). In contrast, where CTCF occupancy is lost, nucleosome organisation is completely lost (Fig 7C). To investigate whether changes of CTCF occupancy were correlated with genes that showed changes in expression following depletion of SNF2H, occupancy of CTCF was assessed at all sites within 10kb of genes that changed expression. The changes in CTCF occupancy at genes that changed expression were indistinguishable from the changes observed at all genes (Fig 7D). This suggests that the overlap between genes affected by CTCF and SNF2H depletion cannot accounted for by a simple change in CTCF occupancy.
Previously it has been observed that nucleosomes are organised as phased arrays adjacent to the binding sites of a subset of metazoan transcription factors [9] and at promoters [5, 33]. However, the factors responsible for the organisation of these arrays have not been established. Here we have systematically investigated the contributions of candidate remodelling ATPases in chromatin organisation. At promoters we did not identify a role for any individual enzymes in organising nuclesomes. Consistent with this changes to nucleosome organisation surrounding the binding sites for general transcription factors such as TBP and TAF1 were only modestly affected by depletion of SNF2H or SNF2L (S6 Fig). Our attempts to pursue this further through depleting combinations of remodellers simultaneously did not provide evidence for this. It is possible that the siRNA strategy was less effective at depleting multiple enzymes or that additional as yet unidentified factors contribute to nucleosome organisation at promoters.
During the course of the study we became aware that the distribution of reads surrounding promoters and factor binding sites is sensitive to the extent of MNase digestion. A similar effect has been observed at promoters in yeast [34]. It could arise from differences in the accessibility of nucleosomes in different locations and or differences in the stability of nucleosomes to higher levels of digestion. The end result is that nucleosomes flanking many factor binding sites are enriched at low in comparison to high levels of digestion. Following sequencing a good way to assess the extent to digestion is to measure the mean nucleosomal read length as this decreases with increasing digestion (S1 Table). We sought to ensure that the difference between these mean read lengths was within 5 bp between control and experimental depletions.
At CTCF sites we observed that depletion of SNF2H resulted in a substantial reduction to nucleosomal pattern flanking these sites. This establishes that SNF2H plays the major role in the establishment of the remarkably well organised arrays of nucleosomes observed flanking CTCF sites. Several additional remodelling ATPases including CHD4, CHD8 and SNF2L have also been reported to be recruited to CTCF sites [25, 26, 28]. These enzymes are unlikely to have major roles in the establishment of extended nucleosomal arrays adjacent to CTCF sites as this is so strongly dependent on SNF2H. Depletion of SNF2L or BPTF had minor effects on the nucleosomes proximal to CTCF sites. This localised effect in the SNF2L depletion is consistent with a local alteration to digestion detected using microarrays [28].
The SNF2H ATPase is present within multiple distinct complexes. The effects of depleting distinguishing subunits of these complexes were inconclusive suggesting that there may be some redundancy between different complexes. However, depletion of the ACF1 subunit of the SNF2H containing human ACF complex and the WSTF subunit of the human WICH complex resulted in subtle reductions to nucleosome organisation especially at locations distal to CTCF (S3 Fig). Both the ACF and WSTF complexes have the biochemical capability to organise chromatin [35, 36].
SNF2L has been purified as a component of a distinct remodelling complexes, NURF [31] and CERF [37]. Expression of SNF2L was originally thought to be restricted to brain and gonadal tissue [38] however, more recent studies indicated that it is ubiquitously expressed [39] and has functions in Wnt signalling and at CTCF sites [28, 39, 40]. The CERF complex is found in neural tissues [37]. The biochemical activities of NURF are distinct from those of ACF in that NURF was originally purified based upon its ability to disrupt nucleosomal arrays [30]. The role that human NURF plays in nucleosome positioning adjacent to human transcription factors is consistent with the original assays for DNaseI hypersensitivity. Although NURF repositions nucleosomes, it also interacts with transcription factors [41] and this can result in directional repositioning of nucleosomes adjacent to factor binding sites [42]. Thus, the NURF complex has the biochemical properties to direct the positioning of nucleosomes immediately adjacent to factors such as CTCF.
We also investigated the effects of chromatin remodelling enzymes on nucleosome organisation at the binding sites for 49 additional transcription factors and at promoters. 29 of these organise extended arrays of nucleosomes and SNF2H contributes to nucleosome organisation at most of these (24/29)(S6 Fig). Typically the nucleosomes immediately flanking the factor binding site are best organised. The distance between these +1 and -1 nucleosomes flanking factor binding sites ranges from 258 bp (REST) to 364 bp (RCOR1). This is substantially larger than would be anticipated based on steric occlusion at the factor binding site and the linker observed between adjacent nucleosomes. It may be that many of these factors are bound by additional cofactors. For example, CTCF is known to associate with TAF3 [43]. Following depletion of SNF2H the distance between the +1 nucleosomes flanking factor binding sites increasing by on average 25bp, in addition the average separation between flanking nucleosomes increases from 176 bp to 183 bp. This is consistent with a role for SNF2H in driving nucleosomes together and towards the factor binding site. The effects following SNF2L depletion are relatively minor, but also distinct in that the distance between adjacent +1 nucleosomes reduces by 10bp and the separation between adjacent nucleosomes is reduced from 176 to 173 bp. This suggests that SNF2L complexes may act to move nucleosomes away from bound factors. The finding that different remodelling enzymes act to alter nucleosome positioning with different directionality is reminiscent of the way remodelling enzymes act with different directionalities at yeast promoters [44] and suggests that a similar interplay operates at the binding sites for a range of transcriptional regulators. With the possible exception of SNF2H at CTCF sites, the effects of depleting enzymes result in alterations to the distributions of nucleosomes rather than complete loss. This suggests that as yet unidentified factors are likely to function in a partially redundant fashion with SNF2H and SNF2L.
In vitro, it has been observed that bound transcription factors act as a barrier restricting the positioning of nucleosomes remodelling enzymes [45]. The observations made here provide evidence that the biophysical interplay between bound factors and nucleosome repositioning characterised in vitro is likely to contribute to nucleosome organisation at functional regulatory elements. Nucleosomes positioned adjacent to such barriers could act as a reference point from which progressively distal nucleosomes are organised [46], potentially providing a means of organising chromatin adjacent to any bound factor. This raises the question why are nucleosomes much better organised adjacent to some bound factors than others?
It is possible that targeted recruitment of remodelling enzymes is required in addition to the presence of a barrier. For example both SNF2L and SNF2H interact with CTCF [27, 28]. However, we also observe dependency upon SNF2L and SNF2H at the binding sites for an additional 24 transcription factors. It is difficult to imagine that SNF2H and SNF2L containing complexes possess the capability of recognising such a structurally diverse range of factors. For this reason we consider it likely that in addition to direct association with transcription factors other interactions contribute to the recruitment of these enzymes.
A prime candidate would be modification to histones such as H3 K4 trimethylation which is enriched at the binding sites for many transcription factors [9]. The SNF2L containing NURF complex has specificity for histone H3 methylated at lysine 4 [47] and so this modification is likely to contribute to recruitment. In budding yeast, ISWI chromatin remodelling enzymes have been shown to be recruited by a looping mechanism [48]. As CTCF sites are also sites of gene looping, this mode potentially provides an additional means via which human ISWI containing enzymes could be recruited.
Eukaryotic regulatory elements seldom consist of the binding sites for a single sequence specific regulatory factor in isolation. Instead binding sites for disparate factors are often found within close proximity. This complicates the interpretation of which chromatin features are directly recruited by a transcription factor and which are influenced by neighbouring factors. Our own data show that both nucleosome organisation and cohesin enrichment at many factor binding sites is influenced by the presence of adjacent CTCF sites (S4 and S5 Figs). CTCF is an unusual transcription factor in that its interaction with DNA is very stable [49] and it plays a major role in determining the distribution of cohesin during interphase [15–18]. This means that interference from adjacent CTCF sites may have especially strong effects. Nonetheless, in principle, similar forms of interference could occur between any two transcription factors and affect the distribution of any chromatin feature such as a histone modification or a cofactor. The scope for misinterpretation as a result of this type of interference is especially high when averaged enrichment is considered for many sites. The use of averaging in metazoan genomic datasets is often essential as the read depth with which data has been collected is in many cases not sufficient for analysis at single genes. This is especially relevant for high resolution studies of nucleosome positioning as this requires a high read depth at each base pair. Obtaining data with the required depth is in most cases impractical and averaging at many related sites provides a way round this. To our knowledge, there is one dataset, with a depth of 3.6 billion reads, that potentially does have the depth required to call nucleosome locations at single loci in human cell lines [50]. However, we could not use this to study alignment to CTCF or other transcription factors as ChIP has not been performed to determine the occupancy of these factors in the cell lines used.
The enhanced nucleosome phasing adjacent to sites such as RFX5 and ZNF143 that have adjacent CTCF sites is best explained if these different factors constructively interfere with each other to generate stronger nucleosome phasing (Fig 5 and S4 Fig). This would require that the binding sites are in phase with the nucleosomal repeat. A large proportion of the CTCF sites are immediately adjacent to RFX5 and ZNF143, so this is feasible. At JUN sites nucleosome organisation increases following removal of CTCF sites in silico or following depletion of CTCF (Fig 5 and S4 Fig). This suggests that in this case CTCF destructively interferes with the phasing of nucleosomes established by JUN. Constructive and destructive interference in the phasing of nucleosomes by different factors has also been observed on adjacent coding genes in yeast [51]. These findings indicate the potential for complexity in the way that chromatin is arranged over regulatory elements that contain binding sites for many different factors.
We observed that depletion of SNF2H results in a reduction in the occupancy of CTCF at many sites (Fig 3). Consistent with this depletion of SNF2H has previously been observed to result in decreased binding of CTCF at the H19/Igf2 locus [27]. This suggests that the action of SNF2H promotes CTCF binding. There is a literature supporting a role for ATP-dependent remodelling enzymes in facilitating the binding of transcription factors to chromatin [52] however, this has typically involved enzymes such as SWI/SNF that disrupt chromatin organisation rather than ISWI containing enzymes such as ACF that space nucleosomes evenly. How then could a nucleosome spacing enzyme act to promote factor occupancy? Currently favoured mechanisms for nucleosome spacing involve the enzyme sensing DNA adjacent to nucleosomes such that repositioning occurs towards the side of a nucleosome with a long accessible linker [23]. This results in the repositioning of nucleosomes with a mean location equidistant between neighbouring nucleosomes. Strongly bound transcription factors such as CTCF also potentially reduce access to linker DNA. In this situation a spacing enzyme would be anticipated to move a nucleosome away from the factor bound linker. Indeed the positioning of nucleosomes by ISWI-related complexes has been observed to be affected by transcription factor binding in vitro [53, 54]. The repositioning of nucleosomes away from factor bound sites effectively partitions DNA sequences occupied by transcription factors and nucleosomes. As a result of reduced competition with nucleosomes the factor bound state would be favoured. This contrasts with the action of complexes such as SWI/SNF which move nucleosomes across factor binding sites resulting in dissociation [54].
Reducing competition with nucleosomes may be especially important at CTCF sites as the binding consensus sequence has high GC content and high inherent affinity for nucleosomes [9]. Therefore, unbound CTCF sites are likely to be occupied by nucleosomes. Supporting this increased nucleosome occupancy is observed at CTCF sites that are only occupied in specific cell lines [55] and in our own data following depletion of CTCF or SNF2H (Figs 2A, 2B and 7C). On the other hand when bound by CTCF the action of SNF2H acts to reduce competition with nucleosomes and further stabilise the bound state. The positive feedback favouring both bound and non-bound states may help to explain how the subset of CTCF consensus sequences that are actually bound varies between different cell lines [55] and during differentiation [56].
Following depletion of SNF2H a quite striking increase in the occupancy of nucleosomes well positioned over CTCF sites is observed at locations where CTCF occupancy is reduced (Fig 7C). These well positioned nucleosomes do not by themselves result in the establishment of well-ordered arrays of flanking nucleosomes. This suggests that the level of non-targeted nucleosome spacing activity in human cells is insufficient on its own to establish ordered arrays of nucleosomes. At the sites where well-ordered arrays are observed, there is likely to be a requirement for both a barrier from which the array is established and targeted recruitment of enzymes such as SNF2H and SNF2L to propagate and maintain spaced chromatin.
As CTCF acts to recruit cohesin and SNF2H promotes CTCF occupancy, SNF2H would be expected to influence cohesin occupancy at CTCF sites. This is indeed the case as we observe a reduction in cohesin by ChIP at CTCF sites following SNF2H depletion (Fig 4). A previous study also observed that loading of cohesin was reduced following inactivation of SNF2H [57]. To investigate whether SNF2H contributes to cohesin loading independently of its effect on CTCF binding, the enrichment of Rad21 was plotted at CTCF sites that remain occupied and adjacent to the binding sites for other transcription factors. Enrichment for Rad21 is not affected at these locations following depletion of SNF2H (S7 Fig). This suggests that SNF2H is not a general loading factor for cohesin, but affects its loading at a subset of CTCF sites. We do not believe that a remodelling complex containing cohesin contributes to nucleosome organisation at CTCF sites as depletion of Rad21 has no effect on nucleosome organisation (Fig 4D).
Following SNF2H depletion we observe that nucleosomes become disorganised and CTCF occupancy is reduced. As many CTCF dependent genes show changes in expression following SNF2H depletion, in principle either or both of these effects could contribute to SNF2H function. However, the lack of any difference in CTCF occupancy at the CTCF target genes affected by SNF2H depletion (Fig 7D) suggests that changes in CTCF occupancy do provide a simple means of explaining the effects on transcription. This raises the possibility that nucleosome positioning is functionally significant, but further investigation will be required to establish this rigorously.
The internucleosome spacing adjacent to CTCF sites is 176 bp, 19 bp shorter than the major internucleosome spacing of 198 bp detected in mammalian cells [2]. In addition, the nucleosomes adjacent to CTCF binding sites are unusually well translationally positioned. The presence of similarly well organised nucleosomes over yeast coding genes is correlated with low histone turnover, histone modification and reduced non coding transcription [58–60]. As non-coding transcription also contributes to enhancer function [61] it is possible that the organised nucleosomes adjacent to CTCF sites also affect enhancer function via RNA mediated pathways.
HeLa cells originally obtained from the ATCC Global Bioresource Center were cultured in DMEM (Invitrogen) supplemented with 0.2 mM l-glutamine and 10% FBS.
The siRNA oligonucleotides were purchased from Eurofins MWG used at a final concentration of 7.8 nM. The siRNA transfections were performed using INTERFERin (Polyplus Transfections). All siRNA sequences are listed in Table 1. Cells were transfected three times according to the INTERFERin protocol with 72hours of growing in between transfections.
To check for depletion of proteins after siRNA transfections, whole cell extracts of HeLa cells were prepared by lysing cells in WCE-buffer (20mM Hepes pH7.6, 400mM NaCl, 1 mM EDTA, 25% glycerol, 0.1% NP-40, protease inhibitors) followed by homogenization using a syringe. SNF2L depletion was checked by directly lysing counted cells using urea sample buffer as described by Eckey M. et al, [39].
Primary antibodies for Western blots used were rabbit anti-human SNF2H (Bethyl Laboratories, A301-081A), mouse anti-CTCF (Abcam, ab37477), rabbit anti-RAD21 (Abcam, ab992), rabbit anti-CHD1 (Bethyl Laboratories, A301-218A), rabbit anti-CHD2 (Active Motif, 39364), rabbit anti-CHD4 (Bethyl Laboratories, A301-081A), rabbit anti-ACF1 (Bethyl Laboratories, A301-318A), rabbit anti-WSTF (Cell Signaling, 2152), rabbit anti–TIP5 (Invitrogen, 491037) and mouse anti-beta actin (Sigma, A2228). Primary antibodies for ChIP used were rabbit anti-human SNF2H (Abcam, ab72499), mouse anti-CTCF (Millipore, 17–10044), rabbit anti-RAD21 (Abcam, ab992), rabbit anti-BPTF (Millipore, Abe24), and rat anti-SNF2L (2C4, [39] which has been kindly provided by P. Becker). Secondary antibodies for Western blots used were Alexa Fluor 680 goat anti-mouse (Invitrogen) and Alexa Fluor 790 goat anti-rabbit (Invitrogen) for immunofluorescence staining and analysis using the LI-COR Odyssey CLx.
For verification of the RNAi depletion of TIP5, and SNF2L RNA was isolated using the QIAGEN RNeasy kit according to the manufacturer. 2μg RNA was reverse transcribed into cDNA (QIAGEN, QuantiTect kit). PCR was carried out in a total volume of 15 μl by using 2 μl of cDNA with the Quanta PerfeCTa SYBR Green FastMix and TIP5 transcript primers ((1) for TTCTCCTATGTTGGGATCTAGCA/ rev CAGTGCCATTCTCTGCCACA and (2) for GGCCTACGACTGTCTCTGGAA/ rev TTGGGGATGAAGGTTGCCG) or SNF2L transcript primers ((1) for AAGCGCCTAAATATGAAAAGGA/ rev GCGGTAGTCTCCAGCAGAAAT and (2) for GCTGGAGACTACCGCCCATAG/ rev CAACCAATTCAGTAATCGAATAT) according to Quanta standard protocol using an AB 7500 Real Time PCR Cycler. Beta-actin transcript was used for normalization.
~8 x105 siRNA-transfected cells were crosslinked with 1% formaldehyde for 10 min and quenched for 5 min with 125 mM glycine at room temperature. After washing cells with cold PBS, cells were lysed using cold NP40-lysis buffer (10 mM Tris pH 7.5, 10 mM NaCl, 3 mM MgCl2, 0.5% NP40, 0.15 mM spermine, 0.5 mM spermidine) for 5 min on ice. Cells were pelleted and washed with MNase digestion buffer (10 mM Tris pH 7.5, 15 mM NaCl, 60 mM KCl, 1 mM CaCl2, 0.15 mM spermine, 0.5 mM spermidine) and resuspended in 50 μl MNase digestion buffer. For the digest, 3units MNase S7 (Roche) were added and incubated for 2 min (low digest) or 4 min (high digest) at 37°C. The digest was stopped adding 1/10 vol 10% SDS and 1/10 vol 250 mM EDTA. NaCl was added at a final concentration of 0.2M to reverse the crosslinking at 65°C overnight. The samples were treated with 40 μg proteinase K for 30 min at 45°C and 10 μg RNase A for 30 min at 37°C, followed by phenol-chloroform extraction and purification using a PCR purification kit (QIAGEN). The samples were eluted from the columns with 50 μl 10 mM Tris pH 7.5 and run on a 1.2% agarose gel in 1 x TAE. The gels were stained using SYBR Safe DNA gel stain (Invitrogen) and mono nucleosomes were cut out. Bands were gel extracted with the QIAGEN gel extraction kit. The resulting DNA fragments were used for Illumina library preparations.
~3.2 x107 siRNA-transfected cells were used per ChIP. Cells were cross-linked with 1% formaldehyde for 10 min and quenched for 5 min with 125 mM glycine at room temperature. After washing cells with twice with ice-cold PBS, cell pellets were flash frozen in liquid nitrogen and stored at -80°C. Frozen cell pellets were lysed in 1.8 ml lysis buffer containing 1% SDS, 10 mM EDTA, 50 mM Tris pH 8.1 and protease inhibitors. To shear chromatin to fragments of about 200–500 bp size, samples were sonicated in 300 μl volumes for 15 cycles (7.5 min total sonication time) at high setting using a Bioruptor (Diagenode). Sonicated lysates were then cleared by centrifugation for 10 min at high speed, diluted 1/10 with dilution buffer (1% Triton X-100, 2 mM EDTA, 150 mM NaCl, 20 mM Tris at pH 8.1, 0.1% Brij-35) and incubated with 12 μg of the respective antibody overnight at 4°C. For each ChIP, 200 μl of Protein G Dynabeads (Life Technologies) were pre-incubated with 0.5% (w/v) BSA in PBS overnight. To capture antibody-bound protein-DNA complexes, lysates were incubated with the prepared beads for 3 hrs and subsequently washed twice with 6ml of each wash buffer I (0.1% SDS, 1% Triton X-100, 2 mM EDTA pH 8.0, 20 mM Tris pH 8.1, 150 mM NaCl), wash buffer II (0.1% SDS, 1% Triton X-100, 2 mM EDTA pH 8.0, 20 mM Tris pH 8.1, 500 mM NaCl) wash buffer III (0.25 mM LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA pH 8.0, 10 mM Tris pH 8.1) and TE buffer (10 mM Tris pH 8.1, 1 mM EDTA pH 8.0) in the cold. To elute, reverse-crosslink and purify ChIP DNA the IPure kit from Diagnode was used according to the manufacturer. The resulting DNA was used for Illumina library preparations.
Libaries from ChIP DNA or mono nucleosomal DNA resulting from MNase digests were prepared using the protocol published (Bowman SK et al, BMC Genomics 2013) with modifications. All enzymes, buffers and nucleotides were purchased from Fermentas unless stated differently. In short, DNA was end repaired in 50 μl reactions containing 1x T4 ligase buffer, 0.4 mM dNTPs, 7.5 U T4 DNA polymerase, 5 U Klenow polymerase, 15 U T4 polynucleotide kinase for 30 minutes at room. To purify DNA, 1.8 volumes Agencourt AMPure XP beads were used according to the manufacturer. A-tailing reactions (50 μl) contained cleaned up DNA, 1x Klenow buffer, 2 mM dATP, 15U Klenow 3’-5’ exo—and were incubated for 30 minutes at 37°C. DNA purification was performed using 1.8 volumes Agencourt AMPure XP beads. Adapter ligation reactions in 50 μl volumes contained DNA from A-tailing, 1x T4 ligase buffer, 0.04 μM annealed universal adapter for ChIP samples and 2 μM adapter for mono nucleosomal DNA, 5 Weiss U T4 ligase and were incubated overnight at 16°C. This time DNA was purified using 1.1 volumes of Agencourt AMPure XP beads to avoid co-purifying excess adapter. DNA was eluted using 20 μl H2O of which half was used to amplify DNA in the next step.
For library amplification the PCR reactions contained 5 μl adapter ligated DNA, 1x Phusion HF buffer (Thermo Scientific), 0.3 μM Illumina universal primer, 0.3 μM Illumina barcoded primer, 0.4 mM dNTP, 200mM Trehalose and 3 U Phusion Hot Start II High Fidelity DNA Polymerase (Thermo Scientific). Thermocycling was performed by denaturing for 3 minutes at 98°C; followed by 20 cycles for ChIP DNA and 10 cycles for mono nucleosomal DNA of: 15 seconds at 98°C, 25 seconds at 60°C, and 1 minute at 68°C, and a final extension of 5 minutes at 68°C.
PCR products were resolved on a 1.2% agarose gel in 1x TAE. ~250 bp to 700 bp of ChIP DNA and ~300 bp of mono nucleosomal DNA was extracted using QIAGEN MinElute Kit and sent for sequencing.
Paired end libraries of MNase digested chromatin ChIP DNA were sequenced using illumine HiSeq technology. Fastq files containing raw reads were aligned to human reference genome (ftp://ftp.ccb.jhu.edu/pub/data/bowtie2_indexes/hg19.zip) hg19 by Bowtie2 with option of maximum fragment length of 1500 for chip data and 500 for nucleosome fragments. Fragment length distributions for each sample used are shown in S1 Table. The midpoints of uniquely mapped nucleosomal or ChIP reads were used for further analysis.
Transcription factor tracks [62] in HeLa cells were downloaded using the UCSC table browser [63] of the encode database (http://genome.ucsc.edu/ENCODE/dataMatrix/encodeChipMatrixHuman.html) as narrow Peak file formats. A 2kb region flanking the TFB site was selected for nucleosome or ChIP enrichment analysis. The nucleosome dyads/chip fragment reads coverage was calculated for each base in the 2kb region. This enrichment value at each base was then divided by number of TFB sites and total number of reads in the experiment to obtain normalised reads. The plotted data was normalised to have same mean read counts in the plotted window. The data was smoothed using a 50 bp sliding window for graphical representation. Plots were generated with python’s plotting modules matplotlib and pylab.
All of the data shown in the manuscript was established as being reproducible between repeats of genome wide experiments. In most cases the data plotted is the average of appropriately digested biological repeats. A full description of the data included in each figure is provided in (S1 Table). In some figures the same control enrichments are re-plotted in different panels.
Sequence data is accessible at the European nucleotide archive (ENA) http://www.ebi.ac.uk/ena/data/view/PRJEB8713 under accession number PRJEB8713.
RNA seq analysis pipeline described in [64] was followed for mapping and measuring differential expression of genes. In brief, the paired end reads of each biological replicate was mapped to hg19 human reference genome independently with TopHat [tophat2 -p 8 -r 200 -g 2 -o output folder hg19 reads_R1_001.fastq reads_R2_001.fastq]. The assembled reads were the provided as input in Cufflinks which generates assembled transcripts for each replicate [cufflinks -p 8 -g hg19_genes.gtf -o output folder mapped reads.bam]. Mapped reads were the used as input in Cuffdiff to obtain differential expression results in tabular format [cuffdiff -p 8 -o output cuffdiff hg19_genes.gtf siScr_1.bam, siScr_2.bam siSNF2H_1.bam, siSNF2H_2 bam]. Transcripts having >1.5 fold changes in their expression were selected as differentially regulated and have uncorrected p-value of the test statistic <0.05 and FDR-adjusted p-value of the test statistic <0.05 were used for further analysis.
|
10.1371/journal.pntd.0001732 | Dengue Infection in Children in Ratchaburi, Thailand: A Cohort Study. I. Epidemiology of Symptomatic Acute Dengue Infection in Children, 2006–2009 | There is an urgent need to field test dengue vaccines to determine their role in the control of the disease. Our aims were to study dengue epidemiology and prepare the site for a dengue vaccine efficacy trial.
We performed a prospective cohort study of children in primary schools in central Thailand from 2006 through 2009. We assessed the epidemiology of dengue by active fever surveillance for acute febrile illness as detected by school absenteeism and telephone contact of parents, and dengue diagnostic testing. Dengue accounted for 394 (6.74%) of the 5,842 febrile cases identified in 2882, 3104, 2717 and 2312 student person-years over the four years, respectively. Dengue incidence was 1.77% in 2006, 3.58% in 2007, 5.74% in 2008 and 3.29% in 2009. Mean dengue incidence over the 4 years was 3.6%. Dengue virus (DENV) types were determined in 333 (84.5%) of positive specimens; DENV serotype 1 (DENV-1) was the most common (43%), followed by DENV-2 (29%), DENV-3 (20%) and DENV-4 (8%). Disease severity ranged from dengue hemorrhagic fever (DHF) in 42 (10.5%) cases, dengue fever (DF) in 142 (35.5%) cases and undifferentiated fever (UF) in 210 (52.5%) cases. All four DENV serotypes were involved in all disease severity. A majority of cases had secondary DENV infection, 95% in DHF, 88.7% in DF and 81.9% in UF. Two DHF (0.5%) cases had primary DENV-3 infection.
The results illustrate the high incidence of dengue with all four DENV serotypes in primary school children, with approximately 50% of disease manifesting as mild clinical symptoms of UF, not meeting the 1997 WHO criteria for dengue. Severe disease (DHF) occurred in one tenth of cases. Data of this type are required for clinical trials to evaluate the efficacy of dengue vaccines in large scale clinical trials.
| There is an urgent need to field test dengue vaccine. Efficacy trials need to be conducted in study sites with
sufficiently high dengue incidence to make a robust estimate of vaccine efficacy and where all dengue virustypes
circulate frequently. In this paper, we report on dengue disease surveillance on approximately 3000 primary-school children in seven schools in Muang district of Ratchaburi province, central Thailand, from 2006 through 2009. We report on the characteristics of children in this cohort who fell ill with laboratory confirmed dengue disease. The study showed that approximately four percent of the children had laboratory confirmed dengue per year. All four dengue virus types were found to be the causes of illness in children in all seven schools. This study has shown Muang district of Ratchaburi province to be suitable for dengue vaccine testing and the site has been selected for the world’s first dengue vaccine safety and efficacy study, being conducted from 2009–2014 in children aged 4–11 years.
| Dengue virus (DENV) infection with any one of the four virus serotypes (DENV-1 to -4), and 4) can produce a spectrum of outcomes, ranging from asymptomatic infection to mild undifferentiated fever (UF), classic dengue fever (DF) and the most severe form of illness, dengue hemorrhagic fever (DHF) [1]. Dengue is an important cause of morbidity and mortality in tropical and subtropical regions of the world [2]. In Thailand, dengue was first recognized in Bangkok in 1958, and in 1987 the largest epidemic ever recorded occurred with 174,285 cases [3]–[5]. Data from 1974 to 1993 showed that dengue was common in children aged less than 15 years of age and the incidence rates among children hospitalized with dengue have been consistently highest in the 5–9 year age group [6]. Disease has been caused by all four DENV serotypes and has become an intractable public health problem in the country [6], [7]. There is no specific antiviral therapeutic licensed for treatment of dengue and prevention relies on mosquito control. As several promising live-attenuated vaccines candidates are in the later stages of clinical development, there is an urgent need to field test dengue vaccines, which may ultimately control the accelerating spread of dengue worldwide [8], [9]. Population-based, laboratory confirmed background data on the epidemiology of dengue in high risk age-specific populations along with field site operational suitability are critical for clinical dengue vaccine trials [8], [10].
Our aims were to collect accurate dengue incidence data for four transmission years in primary school children in a dengue hyper-endemic area, and to establish infrastructure for potential large scale trials of candidate tetravalent dengue vaccine. In 2005, a pilot epidemiologic study of symptomatic dengue infection in 481 school-children aged 3–10 years was conducted, which led to this study conducted during 2006–2009.
The study protocol was approved by the Ethical Review Committee for Research in Human Subjects, Ministry of Public Health, Thailand, and the Institutional Review Board, International Vaccine Institute, Seoul Korea.
The study was carried out in the sub-district Namuang (downtown) of Muang district of Ratchaburi province, which is located approximately 100 km west of Bangkok, and lies between the Maeklong River on the east and the Thai-Myanmar border on the west. The sub-district has a population of 38,835 (census 2006) and a total area of 8.7 km2. The principal medical care facility for the province is Ratchaburi Provincial Hospital (RH), a 855-bed tertiary care facility with 90 pediatric beds and 12 pediatricians on staff. In 2005 the hospital served approximately 1,520 outpatients per day. There were 207, 197 and 214 clinically diagnosed dengue patients, admitted to the pediatric dengue ward in 2003, 2004, and 2005, respectively.
This was a prospective cohort study of children attending 7 primary schools. Schools were selected based on their desire to participate in the study, and location within 6 km from RH and the Provincial Health Office (figure 1). Following school-based informational meetings with parents, informed parental consent and signed assent for children >7 years of age were obtained from potential participants. Enrollment criteria were healthy children, no history of chronic illness, ages 3–11 years (grades 1–5) at the time of enrollment, attendance at one of the study schools, and living in a village of Muang district. Exclusion criteria included intent to move outside of the study area within the study period. Children were eligible to remain in the study until graduation from sixth grade. During each January of the study, new children aged 4–5 years were offered the opportunity to enroll to replace children who graduated from the sixth grade.
During the entire four-year study period, active surveillance for school absence and/or children who had a documented fever was conducted by contacting teacher-coordinators daily during school-term and telephoning parents or conducting home visits twice a week during school vacations. School absenteeism was identified each morning by teacher-coordinators at participating schools by comparing names of study participants with reported absences. Participant absenteeism was recorded on a web-based child tracking application at the study field office. Absenteeism was reviewed by the research staff and parents of absent students were contacted by the research staff each afternoon to determine if the child was absent due to a febrile episode.
All parents and teacher-coordinators were provided digital thermometers and were instructed in their use. Parents of a child with a temperature ≥37.5°C were asked to take them to the RH outpatient department (OPD) where there was a special fast track unit with research pediatricians to examine study participants. Children with a fever of ≥38°C or who were considered severely ill were admitted to the inpatient dengue ward (IPD). All illness data were reported at RH on a web- based reporting application.
At the OPD or IPD, an acute-phase venous blood sample (S1) was obtained from each febrile study participant and a convalescent-phase venous blood sample (S2) was obtained 7 to 14 days later. S1 and S2 samples were drawn into serum separator tubes, allowed to clot at room temperature for 1–2 h, then stored at 4°C. Serum was separated into aliquots within 24 hours, and stored at –70°C until laboratory testing. Serum samples were transported in dry ice from RH to Bangkok monthly for dengue and Japanese encephalitis (JE) laboratory testing.
Over the course of the study, diagnostic testing was performed in two laboratories at Mahidol University- Center for Vaccine Development (2006), and at the Faculty of Tropical Medicine, Mahidol (2007–2009) using the same diagnostic algorithm. S1 and S2 were tested for dengue virus specific IgM/IgG by capture enzyme-linked immunosorbent assay (EIA), as described previously [11]. An IgM anti-DENV level ≥40 units was considered indicative of an acute DENV infection. To exclude Japanese encephalitis virus infection and antibody cross-reactivity, specimens were tested concurrently for JE-specific IgM by EIA [11].
S1 samples of the cases with IgM anti-DENV level ≥40 units were further tested for DENV serotype. In 2006, this was performed by mosquito inoculation in Toxorhynchites splendens [12] with detection and serotyping by immunofluorrescence. In 2007–9, a modified nested serotype-specific reverse-transcriptase polymerase chain reaction (RT-PCR) [13] was used to serotype DENV.
While study participants were tracked by name, school, and home address throughout the study, they were given a unique identification number upon enrollment that was used in the electronic data base for transmittal of epidemiologic, clinical and laboratory data, and for data analyses. Individual identifier information was kept in a secured location separate from data forms. Data were entered within 24 hours after the staff identified a participant as being absent, febrile or at the hospital. Data quality was assured by the study field office and hospital staff managers on a daily basis during weekdays. Data from all sources were automatically transferred through a web-based application to a database located at the Data Management Unit (DMU) of the Faculty of Tropical Medicine, Mahidol University, in Bangkok which uses the Mahidol University Information Technology Department Data Procedure SOP. The DMU monitored inconsistency of data entry daily.
On a monthly basis, the entire dataset for the study children was exported into SAS format and archived with a CD backup. Statistical analyses were performed by using SPSS software for Windows (version 17.0; SPSS Inc., Chicago, Illinois). All incidences of the confirmed dengue were calculated as per 100 person-years (percent). Incidence rates in all the children and children aged ≤4, 5–9 and 10–14 years were determined by using the age-specific study population at the time of surveillance as denominator. Chi-square tests were used for determining the differences among the proportions of clinical spectra, dengue serotypes and annual incidences.
In February 2006, 3,015 students aged 3–13 years were enrolled in the study for the start of surveillance. In the subsequent 3 years, loss of students was a result of the graduation of sixth graders and families' relocation. Losses were 51 (1.7%) in 2006, 254 (8%) in 2007, and 384 (14%) in 2008. The higher dropout rate in 2008–2009 was due to 150 subjects' terminations. They were enrolled into Sanofi Pasteur's dengue vaccine efficacy trial which began in February 2009 [ClinicalTrials.gov Identifier: NCT00842530, http://www.clinicaltrials.gov]. Following replacement of dropouts, there were 3,220 (3–.14 years old) subjects in 2007 which declined to 2,833 (4–14 years) in 2008. Since there was no subject replacement in 2009, only 2,316 (5–15 years) subjects remained at the end of the study. No differences in gender distribution were noted from year to year or between schools (data not shown). Mean number of subjects enrolled in the four years was 2,846, with male to female ratio of 1.04∶1. Median age of participants shifted from 9 years in 2006 and 2007 to 10 years in 2008 and 11 years in 2009.
Over the 4-year study period there were 36,934 student-absence episodes- 8,429, 9,438, 10,007 and 9,060, for each respective year of the study. During this same period there were 5,842 febrile illness episodes −1,892, 1,401, 1,527 and 1,022 for each respective year of the study. The mean student-absence episodes and mean febrile episodes per child per each study year are shown in table 1. Per year, the mean number of absences per student was 3.39 and each child had a mean febrile episodes of 0.53 (table 1).
Of the study participants with a febrile illness over the 4-year study period, 73% were brought to RH by their parents. The majority of febrile children (53%) visited RH on days 1–2 after onset of fever, while 30%, 14% and 3% visited RH on days 3–4, days 5–6 and days 7–9 after onset of fever, respectively. In 2006, of 1892 children who had febrile illness, 734 (39%) came to RH of which 154 (8%) were admitted to the IPD and 580 (31%) required only OPD care. Twenty nine percent visited a private hospital and other clinics (IPD 1% and OPD 28%), and the remaining 32% of them bought medicine independently from pharmacy without seeking medical care (self treatment). The rate of hospital visits increased significantly from 39% in 2006 to 94% (1316/1401) in 2007, 88% (1347/1527) in 2008 and 87% (887/1022) in 2009. Numbers of febrile children who were admitted to RH IPD were 257 (18%) in 2007, 161 (10%) in 2008 and 90 (8%) in 2009.
During the study period 58.2% (3401of 5842) of febrile children had acute samples collected (S1: 625 from IPD and 2776 from OPD) and 57.7% (3368 febrile children) had paired sera collected (S2: 625 from IPD and 2743 from OPD) and tested for DENV and JEV infection. There were 394 serological confirmed dengue cases (215 boys and 179 girls). The case number was lowest (51 cases) in 2006, then rose to 111 cases in 2007, reached a peak at 156 cases in 2008 and declined to 76 cases in 2009. Dengue occurred all year round but the highest number was from June to August in 2006, 2008 and 2009. In 2007 the disease was highest from June to December (figure 2). Dengue accounted for 2.73%, 7.78%, 10.18% and 7.50% of the febrile illness episodes in each of the study years, respectively and averaged 6.74% of the children with a febrile illness.
Over the 4 years of the study, there were 2882.32, 3103.66, 2717.27 and 2312.45 person-years of follow-up for the respective years (table 1). There was statistically significant difference in the incidence of dengue year by year (p<0.001).The incidence in 2006 was the lowest (1.77%), increased to 3.58% in 2007, peaked in 2008 (5.74%) and then declined to 3.29% in 2009 (table 1). Mean dengue incidence over the 4 years was 3.6%. Dengue incidence varied by years and by schools. Over the 4 years, school 4 had the highest mean incidence (4%) and school 7 had the lowest mean incidence (2.31%) (figure 3). Among the 3,401 and 3368 samples of S1 and S2 tested, there were 9 serologically-determined acute JEV infection cases (6 and 3 cases in 2007 and 2008, respectively).
Among the serologically confirmed 394 cases, 42 (10.7%) were DHF, 142 (36%) were DF and 210 (53.3%) were UF. Overall the proportion of males and females was 54.6% and 45.4%, respectively, and this male predominance persisted in all disease categories although it was not statistically significant (p = 0.89 chi square test). The age of cases ranged from 4–14 years (mean, 9.4; median, 10; mode 11) with slightly more cases in children aged 10–14 years old than in the 5–9 years old group: 200 (50.8%) vs. 192 (48.7%), respectively. The older age group also exhibited higher numbers of both DHF and DF (25 and 72 cases) than those of the younger age group (17 and 69 cases), respectively. However, there were no significant differences in disease severity, DHF vs DF between the two age groups (p>0.05, chi-square test). Only two children aged 3 and 4 years had dengue (one DF and one UF), so comparative analysis of the less than 4 year-age group with the older age groups was not performed. Mean ages for children with differing DHF grades was: I-11 years (n = 29), II- 10 years (n = 6) and III – 6 years (n = 7). Mean ages of DF and UF were 9.3 and 9.4 years, respectively. There were no deaths. Further details of clinical manifestations have been described [14].
Diagnosis of serotype was attempted on acute serum samples of 394 children who had serological confirmed dengue virus infection. These tests yielded DENV from 333 (84.5%) children, and the detection rate varied with only 67% in 2006 using mosquito isolation compared to 83% in 2007, 86% in 2008, and 96% in 2009 using RT-PCR. Viral detection also varied by day of sample collection, the highest rate (95%) was from the samples collected on days 1 and 2 after onset of fever, followed by days 3 and 4 (89%) and days 5 and 6 (69%).
All four DENV serotypes circulated in every study year (table 2). DENV-1 was the most common serotype detected with 144 isolates (43%) and predominated in every study year, ranging from 34 to 50 percent. DENV-2 (98 isolates, 29%) was the next most common, followed by DENV-3 (66 isolates, 20%) and DENV-4 (25 isolates, 8%). DENV-2 and DENV-3 had the lowest proportions of isolates in 2006 (3% and 9%, respectively). There was a 30 fold increase of DENV-2 to 33% of isolates in 2007 and a further increase to 37% of isolates in 2008 and declining to 25% of isolates in 2009. There was an 8 fold increase of DENV-3 to 26% of isolates in 2007, declining to 10% of isolates in 2008 and again rising to 34% of isolates in 2009. In 2006 DENV-4 was at its highest number: 13 isolates (38%), declining to 7, 3, and 2 isolates in 2007–9, respectively (table 2).
DENV serotype specific incidence differed statistically by year (p<0.002) and varied between schools and between years (number of serotype-specific cases divided by number of children with acute dengue virus infection in the corresponding school). Over the study years, all four DENV serotypes were found in all 7 schools (figure 4). The highest incidence of DENV-1, DENV-2, DENV-3 and DENV-4 was found in schools 5, 4, 1 and 2, respectively.
All 4 DENV serotypes were associated with all illness categories (DHF, DF and UF) (table 3). DENV-1, DENV-2 and DENV-3 were associated with DHF grades 1, 2, and 3. DENV-4 was only observed among children with DHF grade 1.The proportions of DEN1, DEN2, DEN3 and DEN4 among children with DHF were 9.7%(14/144), 10.2% (10/98), 15.2% (10/66) and 8%(2/25), respectively. The proportions of DENV-1, DENV-2, DENV-3 and DENV-4 among children with DF were 39.6% (57/144), 31.6% (31/98), 39.4% (26/66), 44% (11/25), and in UF were 50.7% (73/144), 58.2% (57/98), 45.4% (30/66) and 48% (12/25), respectively. It was found that DENV serotypes had no significant correlation with disease severity (p>0.05 between DHF vs. non-DHF and p>0.05 between UF vs. non-UF) or for rate of hospitalization (p>0.05).
Of all 394 dengue cases, secondary DENV infection was found in 86.3%, the remainder (13.7%) having a primary DENV infection (table 3). The rates of secondary DENV infection found in DHF, DF and UF cases were 95%, 88.7% and 81.91%, respectively. Primary DENV infection in 2 DHF Grade 1 cases was due to DEN3. There was no statistically significant difference in rate of secondary DENV infection in DHF (40/42) vs. those in DF and UF (298/352) (p>0.05).
Altogether 193 cases (47%) were hospitalized. The hospitalization rate by disease severity category was: DHF - 42(100%), DF - 119(84%), and UF - 32(15%). The proportion of hospitalized children with UF (32/172) was significantly lower than that of children with DF (119/142) (p<0.0001), which was significantly lower than those of DHF (42/42) (p<0.001) (table 3).
Our study is a population based epidemiological study driven by a clear objective of site preparation for a future dengue vaccine field trial. The Namuang sub-district of Muang district, Ratchaburi province, was chosen for the study site because it had high reported rates of dengue virus transmission. The incidence of dengue disease was determined from a prospective, long- term active fever surveillance of study participants in a well-defined cohort of school children throughout a 48-month period. Daily tracking of school absenteeism during school days and telephone contact with parents twice a week during school holidays made a broader capture of dengue possible. Most febrile illness cases detected in the cohort surveillance presented at Ratchaburi hospital were examined by pediatricians and had samples collected for dengue diagnostic testing.
An important aspect of this study was that our surveillance system captured children with febrile illness who sought medical care in both the public and private sectors. Prior to the beginning of the study, we met with a private hospital director and pediatricians who worked on the private hospital and clinics in Muang district to inform them of the study project. They were asked to inform us whenever participants attended their hospital/clinics. As a result, our staff visited all participants admitted to the private hospital and obtained acute serum samples from febrile patients. Convalescent samples were subsequently collected at RH. In addition, by end of the first year, we had strengthened parent-staff communications and education by use of telephone explanation and written documents that explained provision of study care with no charge fast track service and telephone consultation service at RH. Our efforts to educate parents were evidenced by increased in parent initiated hospital visits for febrile illness to 87% in the years 2007–2009.
Our rates of dengue incidences over the four-year period, 1.77, 3.58, 5.74 and 3.29 per 100 person-years in 2006–2009 reflect a parallel pattern of dengue incidence reported from the national surveillance data base in 2006–9 for Muang district, Ratchaburi Province: 159, 200, 361 and 155 per 100,000, respectively. Our prospectively determined dengue incidence in Na-muang sub-district is 11 to 21 (average 16.5) fold higher than those derived from the national surveillance database in Muang district. One difference between our incidence rates and the national rates is that we prospectively determined DHF, DF and UF cases among febrile illnesses of primary school children, whereas, the national data presents mainly DHF reported from hospitalized patients in all age groups.
Although there were some methodological differences in how active surveillance was conducted in our study and previously reported studies in childhood cohorts from Kampang Phet, Thailand [15], Kampong Cham, Cambodia [16] and Managua, Nicaragua [17], the incidences of laboratory confirmed dengue found by active surveillance for acute febrile illness were much higher than that of the corresponding national reporting data. Using a calculated expansion factor to estimate differences between national reporting and laboratory based surveillance for dengue incidence, Wichmann et al. 2011 found the average under-recognition of dengue across three cohort studies (Kampang Phet, Ratchaburi and Kampong Cham) of dengue incidence to be more than 8-fold [18]. The Nicaraguan study found under-reporting of dengue cases in relation to national surveillance systems to be 21.3 fold [17]. The national under-reporting of dengue incidence of cases hinders accurate knowledge of disease burden.
All four DENV serotypes were involved in all levels of disease severity observed over the study period. Although there was strong seasonal variation, laboratory confirmed dengue virus infections were observed in every month of the four-year observation period in all 7 schools indicating that there is continuous DENV transmission in Ratchaburi. DENV-1 was the most commonly virus isolated (43% of total), followed by DENV-2 (29%) and DENV-3 (20%). New introduction in 2007 of DENV-2 and DENV-3 after a period of relatively low frequency in occurrence was followed by a severe outbreak in 2008 wherein DEN2 frequency was 30-fold increased and DENV-3 8-fold increased. This observation is consistent with a previous report that in Thailand DENV-2 was a dominant isolate during moderately severe dengue outbreak years and DENV-3 was associated with subsequent severe outbreaks after a time of relatively low frequency [6].
Dengue virus isolation methods differed between 2006 and the subsequent years in that PCR was used from 2007–2009. A lower DENV isolation rate was observed when culture in C6/36 cells and mosquito inoculation (67%) were used compared to detection of virus genome using RT-PCR (>83%) in this study is likely due to lower sensitivity of mosquito methods compared to PCR molecular methods [19], [20]. The overall viral identification rate using combined mosquito inoculation and PCR tests of 84.5% reported in this study is higher than the 65% in the previous report which also used both methods [7].
Secondary-type DENV infection described in this study was based on the ratio of anti-DENV IgM: IgG<1∶8 [11]. It has been observed from serology test results that there is cross reaction between JEV IgG and DENV IgG in S1 as well as in S2 with resulting inability to distinguish JEV IgG from DENV IgG. This observation had also been reported previously [21]. As the Thai National Immunization Program has included JE vaccination since 2000, JEV IgG from vaccination might have had some influence on the observed DENV IgG titers and may have confounded the interpretation of secondary infection data reported here. However, the majority (84.5%) of the patients with secondary DENV infection was accompanied with positive dengue virus detection and had low JEV IgG titer in their acute (S1) specimens suggesting a minimal impact on the validity of anti-DENV IgM/IgG ratio used to estimate secondary DENV infection.
There appears to have been a shift in modal age of dengue incidence over the past four decades in Thailand. A retrospective hospital based study of laboratory confirmed dengue in 15,376 patients from Bangkok, Thailand reported a modal age of 5 years during 1973–79 which increased to 8 years during 1990–99 [6]. We found a modal age of 11 years in the present study. This finding might reflect a somewhat older group presenting with less severe disease (UF) in outpatient settings, but it is consistent with the increasing modal age previously reported [6]. Hospitalization rates for acute symptomatic dengue infection could be considered a measurement of dengue disease severity. The hospitalization rate for DHF, DF, and UF were 100- 84- and 15- percent, respectively. Over the four years of our study, the provision of fast track service without payment, led to progressively better compliance of febrile cases from the cohort reporting directly to hospital OPD for evaluation, minimizing the possibility that cases may have been missed, especially in years 2007–9. However it is possible that the study missed some mild febrile cases that did not attend RH in 2006.
Because of the sensitivity of our surveillance system only 184 (46%) of laboratory confirmed dengue cases (DHF 42 cases and DF 142 cases) met the 1997 WHO case definition [1]. Clinical manifestations of 210 (52.5%) of laboratory confirmed UF cases did not meet the WHO case definition for dengue [1]. UF was the most common clinical manifestation of children infected with dengue virus, whether primary or secondary DENV infections, and can be difficult to distinguish from other childhood febrile illnesses [14]. Our findings that a high proportion of dengue cases manifested as UF has an important impact on present vector control practices. In Ratchaburi province, mosquito control campaigns against dengue occur broadly over the province, four times per year. In addition, implementation of mosquito control measures including source reduction, application of larvicides, and spraying residual insecticide, are implemented in the area of 100 meters in diameter around all the reported houses of patients with dengue infection at all times of year. Due to the current national surveillance system only DHF/DF cases observed in clinical facilities are reported. Mosquito control measures are not implemented around houses of UF patients. This may contribute to inability to adequately control dengue and result in the long persistence and wide spread of dengue in the province as well as in Thailand in general.
We performed an extensive dengue epidemiology study and report accurate background data on dengue incidence in a cohort of individuals at high risk of dengue, children ages 3–15 years living in Namuang subdistrict of Muang district of Ratchaburi province, Thailand. We found sufficiently high incidence of four DENV serotypes for four consecutive years to make vaccine efficacy studies possible. Our study methods and findings fulfilled the epidemiological criteria recommended by WHO for dengue vaccine trial site selection [8]. Following the reports of safety and immunogenicity of phase I studies of live-attenuated tetravalent dengue vaccine in dengue–naïve and dengue-endemic populations [22], [23], the site was selected for the first dengue vaccine efficacy and safety trial, phase 2b. This trial uses a live-attenuated yellow fever vaccine virus based chimeric tetravalent dengue vaccine in a 3-dose vaccine administration of doses spaced at six-month intervals. This trial was launched in early 2009 and dose administration in approximately 4000 participants was completed in March 2011 [ClinicalTrials.gov Identifier: NCT00842530). Safety of the vaccine within the trial enrollment and vaccine administration period has been established. An analysis for vaccine efficacy is planned in 2012. If the vaccine is sufficiently immunogenic and efficacious after long term follow up, it is anticipated that the vaccine might be available for use in 2015.
|
10.1371/journal.pcbi.1004145 | A Dynamical Phyllotaxis Model to Determine Floral Organ Number | How organisms determine particular organ numbers is a fundamental key to the development of precise body structures; however, the developmental mechanisms underlying organ-number determination are unclear. In many eudicot plants, the primordia of sepals and petals (the floral organs) first arise sequentially at the edge of a circular, undifferentiated region called the floral meristem, and later transition into a concentric arrangement called a whorl, which includes four or five organs. The properties controlling the transition to whorls comprising particular numbers of organs is little explored. We propose a development-based model of floral organ-number determination, improving upon earlier models of plant phyllotaxis that assumed two developmental processes: the sequential initiation of primordia in the least crowded space around the meristem and the constant growth of the tip of the stem. By introducing mutual repulsion among primordia into the growth process, we numerically and analytically show that the whorled arrangement emerges spontaneously from the sequential initiation of primordia. Moreover, by allowing the strength of the inhibition exerted by each primordium to decrease as the primordium ages, we show that pentamerous whorls, in which the angular and radial positions of the primordia are consistent with those observed in sepal and petal primordia in Silene coeli-rosa, Caryophyllaceae, become the dominant arrangement. The organ number within the outmost whorl, corresponding to the sepals, takes a value of four or five in a much wider parameter space than that in which it takes a value of six or seven. These results suggest that mutual repulsion among primordia during growth and a temporal decrease in the strength of the inhibition during initiation are required for the development of the tetramerous and pentamerous whorls common in eudicots.
| Why do most eudicot flowers have either four or five petals? This fundamental and attractive problem in botany has been little investigated. Here, we identify the properties responsible for organ-number determination in floral development using mathematical modeling. Earlier experimental and theoretical studies showed that the arrangements of preexisting organs determine where a new organ will arise. Expanding upon those studies, we integrated two interactions between floral organs: (1) spatially and temporally decreased inhibition of new organ initiation by preexisting organs, and (2) mutual repulsion among organs such that they are “pushed around” during floral development. In computer simulations incorporating such initiation inhibition and mutual repulsion, the floral organs spontaneously formed several circles, consistent with the concentric circular arrangement of sepals and petals in eudicot flowers. Each circle tended to contain four or five organs arranged in positions that agreed quantitatively with the organ positions in the pentamerous flower, Silene coeli-rosa, Caryophyllaceae. These results suggest that the temporal decay of initiation inhibition and the mutual repulsion among growing organs determine the particular organ number during eudicot floral development.
| How to determine the numbers of body parts is a fundamental problem for the development of complete body structures in multicellular organisms. Digit numbers in vertebrates are evolutionarily optimized for the specific demands of the organism [1]; the body-segment number in insects is constant despite the evolutionarily diversified gene regulation in each segment [2–4]; and five petals are indispensable to forming the butterfly-like shape that is unique to legume flowers [5]. Studies of animal structures, such as vertebrate limbs and insect segments, strongly suggest that crosstalk between pre-patterns (e.g., morphogen gradients) and self-organizing patterns underlies the developmental process of organ-number determination [6–13]. In plant development, a self-organization based on the polar transport of the phytohormone auxin [14–16] is conserved among seed plants [17] and seems to be the main regulator of the development of a hierarchal body plan, called a shoot, consisting of a stem and lateral organs such as leaves. The number of concentration peaks in most self-organizing patterns, such as Turing pattern and the mechanisms proposed for plant-pattern formation, is proportional to the field size [15, 18, 19]. Despite having a diversified field size for floral-organ patterning, the eudicots, the most diverged clade among plants, commonly have pentamerous or tetramerous flowers containing five or four sepals and petals (the outer floral organs), respectively, and rarely have other numbers of organs [20, 21]. Here, we focus on the developmental properties that so precisely and universally determine the floral organ numbers through self-organizing processes.
Phyllotaxis, the arrangement of leaves around the stem, provides insight into floral development, because studies of floral organ-identity determination [22] have verified Goethe’s foliar theory, which insists that a flower is a short shoot with specialized leaves [23]. Phyllotaxis is mainly classified into two types: spiral phyllotaxis, which has a constant divergence angle and internode length, and whorled phyllotaxis, which has several leaves at the same level of a stem [24]. For spiral phyllotaxis, Hofmeister described a hypothesis of pattern formation in 1868 [24], which we summarize in three basic rules: the time periodicity of primordia initiation, the initiation of a primordium at the largest available space at the edge of the meristem (the undifferentiated stem-cell region), and the relative movement of primordia in a centrifugal direction from the apex due to the growth of the stem tip. Following that hypothesis, numerous mathematical models incorporating contact pressure [25, 26], inhibitor diffusion [27], reaction-diffusion [18, 28], and mechanical buckling of the epidermis [29, 30] were proposed to explain the observed phyllotactic patterns. Over the past ten years, these mathematical models were tested and interpreted in light of modern molecular biology. Several studies have suggested that the competitive polar transport of the auxin accounts for two of Hofmeister’s rules, the periodicity of initiation and the initiation at the largest space, which together are capable of reproducing both spiral phyllotaxis and whorled phyllotaxis [15, 16, 31].
Despite their simple rules and uncertain molecular basis, the phyllotaxis models can account for several of the quantitative properties observed in organ patterning. For example, one model showed that the divergence angle between successive leaves is 180 degrees for the first and second leaves, 90 degrees for the second and third leaves, and oscillating thereafter, converging to the golden angle, 137.5 degrees, which agrees with the phyllotaxis of true leaves in Arabidopsis thaliana after the two cotyledons [32, 33]. Similar oscillatory convergence to a particular divergence angle occurs in the sepal primordia of the pentamerous flower of Silene coeli-rosa, Caryophyllaceae. In S. coeli-rosa, the divergence angle is 156 degrees at first, and then it oscillates, converging on 144 degrees [34]. The golden angle also appears in the floral organs of several Ranunculaceae species [35, 36]. The agreements between the phyllotaxis models and actual floral development suggest that mathematical models can give useful clues to the underlying mechanisms of not only phyllotaxis but also floral organ patterning.
There are at least three fundamental differences, however, between real floral development and the phyllotaxis models. The first difference is the assumption of constant primordium displacement during tip growth, which comes from Hofmeister’s hypothesis and has been incorporated into most phyllotaxis models. Although the helical initiation has been thought to always result in spiral phyllotaxis, many eudicots form the whorled-type sepal arrangements in their blooming flowers subsequent to helical initiation [37] (Fig 1; e.g., Caryophyllaceae [34], Solanaceae [38], Nitrariaceae [39], and Rosaceae [40]). The remnants of helical initiation are more obvious in the pseudo-whorls (e.g., Ranunculaceae [41]), where the distance between each organ primordium and the floral center varies slightly even in the whorls of mature flowers, which usually have more varied floral organ numbers [20, 35], suggesting that post-meristematic modifications of primordia positions [42] play an essential role in generating the whorled arrangement and determining the floral organ number during floral development. In contrast, most phyllotaxis models have assumed constant growth of the primordia, so that the whorls appear only after the simultaneous initiation of several primordia [19]. The second difference comes from the fact that floral development is a transient process, whereas most phyllotaxis models have focused on the steady state of the divergence angle. Although the golden angle (137.5 degrees) is quite close to the inner angle of regular pentagon (144 degrees), the developmental convergence from 180 degrees (cotyledon) to 137–144 degrees in phyllotaxis requires the initiation of more than five primordia, both in A. thaliana leaves and in the mathematical models [16, 33]. In contrast, the divergence angle between the second and third sepal primordia in pentamerous eudicot flower development is already close to 144 degrees [34]. The third difference comes from the accuracy of the floral organ number in many eudicots. Although the polar auxin-transport model reproduced both wild-type and mutant A. thaliana floral organ positioning [43], the organ number in the model was more variable, even with an identical parameter set (Fig 3 in [43]), than that in experimental observations (Table 1 in [44]). Moreover, among eudicot species, the appearance of pentamerous flowers is robust, despite the diversity of the meristem size and the outer structures, including the number and position of outside organs such as bracts [20]. Together, the differences between real floral development and previous phyllotaxis models indicate that floral development requires additional mechanisms to determine the particular organ number.
To resolve the inconsistencies between the earlier models and actual floral development, we set out a simple modeling framework, integrating Hofmeister’s rules with two additional assumptions, namely, the repulsion between primordia that can repress primordium growth and the temporal decrease in initiation inhibition of new primordium, which were proposed independently in the contact pressure model [25, 45, 46] and the inhibitory field model [33, 47, 48], respectively, for phyllotaxis. First, when we incorporated mutual repulsion among primordia into the growth process, a whorled-type pattern emerged spontaneously following the sequential initiation of primordia. The mutual repulsion obstructed the radial movement of a new primordium after a specific number of primordia arose, causing a new whorl to emerge. The number of primordia in the first whorl tended to be four or eight. Second, when we assumed that older primordia have less influence on the initiation of a new primordium, the pentamerous whorl arrangement, which is the most common arrangement in eudicot flowers, became dominant. We analytically show the conditions for the development of tetramerous and pentamerous whorls, and we predict possible molecular and physiological underpinnings.
Following Hofmeister’s rules as mathematically interpreted by Douady and Couder [49], we focused on initiation and growth, the two processes of floral development. In the initiation process, each primordium emerges successively at the least crowded position, depending on a potential function [49]. We assumed periodic initiation to examine how the sequential initiation results in the whorled-type pattern. We allowed the primordia to move during the growth process in response to the repulsion among the primordia, unlike earlier studies that assumed constant growth depending only on the distance from the apex [28, 49].
Following the earlier models [49], we represented the meristem as a circular disc with radius R0 and the primordia as points (Fig 2A). A new primordium arises at the point along the edge of the meristem (R0, θ), in polar coordinate with the origin at the meristem center, where θ gives the minimum value of the inhibition potential Uini. As one of the simplest setups for sequential initiation [37], we followed the assumption of earlier models for spiral phyllotaxis [49], which state that new primordia arise sequentially with time intervals τ, as opposed to the simultaneous initiation studied previously for whorled phyllotaxis [19] (Fig 1). Although the structures outside of the flower, such as bracts and other flowers, as well as the position of the inflorescence axis, may affect the position of organ primordia, the pentamerous whorls appear despite their various arrangement [20]. Therefore, as the first step of modelling of floral organ arrangement, we assumed that whorl formation is independent of any positional information from structures outside of the flower. Thus, we calculated the inhibition potential only from floral organ primordia which are derived from a single floral meristem.
The potential functions for the initiation inhibition by preexisting primordia have been extensively analyzed in phyllotaxis models [16, 47, 49]. The potential decreases with increasing distance between an initiating primordium and the preexisting primordia account for the diffusion of inhibitors secreted by the preexisting primordia [27, 50], and the polar auxin transport in the epidermal layer, as proposed in previous models of phyllotaxis [15, 16, 31] and the flowers [43]. We employed an exponential function exp(−dij/λini) as a function of θ, where dij denotes the distance between a new primordium i and a preexisting primordium j at (rj, θj) as
d i j = R 0 2 + r j 2 - 2 R 0 r j cos ( θ - θ j ) ⋅ (1)
The function decreases spatially through the decay length λini exponentially, induced by a mechanism proposed for the polar auxin transport, i.e., the up-the-gradient model [15, 16]. Up-the-gradient positive feedback amplifies local auxin concentration maxima and depletes auxin from the surrounding epidermis, causing spatially periodic concentration peaks to self-organize [15, 16] and thus determine the initiation position of the primordia [51]. The amplification and depletion work as short-range activation and long-range inhibition, respectively [52], which are common to Turing patterns of reaction-diffusion systems [18]. Since the interaction of local maxima in the reaction-diffusion systems follows the exponential potential [53, 54], the up-the-gradient model likely explains the exponential potential between the auxin maxima, while the rigorous derivation requires further research. The decay length λini depends not only on the ratio of the auxin diffusion constant and the polar auxin-transport rate [15] but also on other biochemical parameters for polar transport and the underlying intracellular PIN1 cycling [55]. Another mechanism, referred to as the with-the-flux model [56, 57], has been proposed for the polar auxin transport. Although with-the-flux positive feedback can also produce spatial periodicity, the primordia position corresponds to auxin minima [57], which is inconsistent with observations [51].
On the other hand, the with-the-flux mechanism can explain auxin drain from the epidermal layer of the primordia to internal tissue [58]. Since the drain gets stronger as the primordia mature [58, 59], the auxin drain could cause decay of the potential depending on the primordia age. The auxin decrease in maturing organs can also be caused by controlling auxin biosynthesis [60, 61]. Therefore, we integrated another assumption, namely that the inhibition potential decreases exponentially with the primordia age at the decay rate α (Fig 2B). Temporally decaying inhibition was proposed previously to represent the degradation of some inhibitors [47, 48] and account for various types of phyllotaxis by simple extension of the inhibitory field model [33]. Taken together, the potential at the initiation of the i-th primordium is given by
U i n i ( θ ) = ∑ j = 1 i - 1 exp ( - α ( i - j - 1 ) ) exp ( - d i j λ i n i ) ⋅ (2)
Most phyllotaxis models have assumed, based on Hofmeister’s hypothesis, that the primordia move outward at a constant radial drift depending only on the distance from the floral center without angular displacement, which makes helical initiation result in spiral phyllotaxis [49]. Here, we assumed instead that all primordia repel each other, even after the initiation, except for movement into the meristematic zone (Fig 2C) following observation of the absence of auxin (DR5 expression) maxima at the center of the floral bud (e.g., [62]). Even at the peripheral zone away from the meristem, the growth is not limited. Hence there is no upper limit for the distance between primordia and the center. The repulsion exerted on the k-th primordium is represented by another exponentially decaying potential when there are i primordia (1 ≦ k ≦ i):
U g , k ( r , θ ) = ∑ j = 1 , j ≠ k i exp ( - d k j λ g ) , (3)
where the decay length, introduced as λg, can differ from λini. The primordia descend along the gradient of potential Ug to find a location with weaker repulsion. The continuous repulsion can account for post-meristematic events such as the mechanical stress on epidermal cells caused by the enlargement of primordia [63, 64] or the gene expression that regulates the primordial boundary [42]. The present formulation (Eq 3) is similar to the contact pressure model, which has been proposed for re-correcting the divergence angle after initiation [25, 45, 46]. Another type of post-initiation angular rearrangement has been modeled as a function of the primordia age employed as i −j −1 in the present model (Eq 2) and the distance between primordia with some stochasticity [65]. Eq 3 accounts for not only the angular rearrangement but also the radial rearrangement with stochasticity in both directions as will be described in the next subsection.
We modeled the initiation process numerically by calculating the potential Uini (Eq 2) for angular position θ incremented by 0.1 degree on the edge of the circular meristem. We introduced a new primordium at the position where the value of Uini took the minimum, provided that the first primordium is initiated at θ = 0. We modeled the growth process by using a Monte Carlo method [66] to calculate the movement of primordia in the outside of the meristem depending on the potential Ug, k (Eq 3, Fig 2C). After the introduction of a new primordium, we randomly chose one primordium indexed by k from among the existing primordia and virtually moved its position (rk, θk) to a new position ( r k ′ , θ k ′ ) in the outer meristem ( r k , r k ′ ≥ R 0 ). The new radius r k ′ and the angle θ k ′ were chosen randomly following a two-dimensional Gaussian distribution whose mean and standard deviation were given by the previous position (rk, θk) and by two independent parameters, (σr, σθ/rk), respectively. Whether or not the k-th primordium moved to the new position was determined by the Metropolis algorithm [66]; the primordium moved if the growth potential (Eq 3) of the new position was lower than that of the previous position (i.e., U g , k ( r k ′ , θ k ′ ) < U g , k ( r k , θ k )). Otherwise, it moved with the probability given by
P M P = exp ( - β Δ U g ) , (4)
where Δ U g = U g , k ( r k ′ , θ k ′ ) − U g , k ( r k , θ k ) and β is a parameter for stochasticity. This stochasticity represents a random walk biased by the repulsion potential. A case PMP = 0 represents that primordia movement always follows the potential (ΔUg < 0). The first primordium stays at the meristem edge r = R0 until the second one arises when PMP = 0 because the growth potential is absent, while it can move randomly outside of the meristem when PMP ≠ 0. To maintain the physical time interval of the initiation process at τ steps for each primordium, the number of iteration steps in the Monte Carlo simulation during each initiation interval was set to iτ, where i denotes the number of the primordia. We also studied the movement following Ug by numerical integration (fourth-order Runge-Kutta method) of ordinary differential equations to confirm the independence of the numerical methods (S1 Fig). All our programs were written in the C programming language and used the Mersenne Twister pseudo-random number generator (http://www.math.sci.hiroshima-u.ac.jp/m-mat/MT/emt.html) [67].
Because the initiation time interval is constant, one possible scenario for forming a whorled pattern should involve decreasing or arresting the radial displacement of primordia (Fig 1, forth row). Therefore, we focused on the change in radial position and velocity to find the whorled arrangement, while angular positions were not taken into account in the present manuscript.
Numerical simulations showed that several whorls self-organized following the sequential initiation of primordia. Although several previous phyllotaxis models showed the transition between a spiral arrangement following sequential initiation and a whorled arrangement following simultaneous initiation [15, 16, 19], they were not able to reproduce the emergence of a whorled arrangement following sequential initiation, which is the situation observed in many eudicot flowers (Fig 1) [34, 37, 38, 40, 41]. In the present model, a tetramerous whorl appeared spontaneously that exhibited four primordia almost equidistant from the meristem center (Fig 2D, left and middle), by arresting radial movement of the fifth primordium at the meristem edge until the seventh primordium arose (arrowhead in Fig 2D, right). Likewise, subsequent primordia produced the same gap in radial distance for every four primordia (Fig 2D, middle and right), leading to several whorls comprising an identical number of primordia (Fig 2D). The radial positions of all primordia were highly reproducible despite stochasticity in the growth process (error bars in Fig 2D–2F, middle and right). Therefore, we identified the whorled arrangement by radial displacement arrest (arrowhead in Fig 2D, right).
The initiation order and angle of the first tetramerous whorl in the model reproduced those observed in A. thaliana sepals [68] (S2A Fig). The first primordium scarcely moved from the initiation point until the second primordium arose because growth repulsion was absent. The second primordium arose opposite the first, whereas the third and fourth primordia arose perpendicular to the preceding two. The angular position of the primordia did not change once the whorl was established because the primordia within a whorl blocked the angular displacement by the growth potential Ug (S3 Fig).
Introducing mutual repulsion among the primordia throughout the growth process caused the whorled arrangement to spontaneously emerge (Fig 2D). This was in contrast to the model of constant growth in which all primordia move away depending only on the distance from the floral apex [49]. A study of post-meristematic regulation by the organ-boundary gene CUP-SHAPED COTYLEDON2 (CUC2) showed that A. thaliana plants up-regulating CUC2 gene have an enlarged primordial margin and have whorled-like phyllotaxis following the normal helical initiation of primordia [42], suggesting that repulsive interactions among primordia after initiation are responsible for the formation of the floral whorls.
In the present model, the meristem size R0 controls the transition from non-whorled (Fig 2E) to whorled arrangement (Fig 2D). Radial spacing of the primordia was regular when R0 was small (Fig 2E, middle) because the older primordia pushed any new primordium across the meristem (Fig 2E, left), causing continuous movement at the same rate (Fig 2E, right). Above a threshold meristem size R0, a tetramerous whorl appeared spontaneously. The primordium number within each whorl increased up to eight with increasing R0, but the number tended to be more variable (S2B Fig). In the A. thaliana mutant wuschel, which has a decreased meristem size, the pattern of four sepals does not have square positions at the stage when the wild-type plant forms a tetramerous sepal whorl [69]. Conversely, the clavata mutant, which has an increased meristem size, has excessive floral organs with larger variation [69]. Our model consistently reproduced not only the transition from the non-whorled arrangement (Fig 2E) to the tetramerous whorled arrangement (Fig 2D) but also the variable increase in the primordia number within a whorl as the meristem size R0 increased.
The pentamerous whorl stably appeared in the presence of temporal decay of initiation inhibition (α > 0 in Eq 2). The whorls comprising five primordia appeared in the same manner as the tetramerous whorls, namely, via the locking of the sixth primordium at the initiation site (Fig 2F, right; S2C Fig).
In order to study the organ number within each whorl extensively, known as the merosity [70], we counted the number of primordia existing prior to the arrest of primordium displacement, which corresponds to the merosity of the first whorl (arrowheads in Fig 2D and 2F, right). We defined arrest of primordium displacement as occurring when the ratio of the initial radial velocity of a new primordium immediately after initiation to that of the previous primordium was lower than 0.2. The definition does not affect the following results as long as the ratio is between 0.1 and 0.6. We found that the key parameter for merosity is the relative value of R0 normalized by the average radial velocity V = σ r / 2 π (see S1 Text) and the initiation time interval τ (Fig 3). The arrest of radial displacement did not occur below a threshold of R0/Vτ (the left region colored red in Fig 3A), whereas the whorled arrangement appeared above the threshold value of R0/Vτ. As R0/Vτ increased further, tetramery, pentamery, hexamery, heptamery, and octamery appeared, successively (Fig 3A).
The present model showed dominance of special merosity, i.e., tetramery and octamery in the absence of temporal decay of inhibition (α = 0 in Eq 2; Fig 3A); pentamery in the presence of temporal decay (α > 0; Fig 3B and 3C), in contrast to previous phyllotaxis models for whorled arrangement in which the parameter region leading to each level of merosity decreased monotonically with increasing merosity [19]. The major difference between α = 0 and α > 0 was that θ3, the angular position of the third primordium, took an average value of 90 degrees when α = 0 (arrowhead in Fig 3A bottom magenta panel) and decreased significantly as α increased (arrowhead in Fig 3A bottom cyan panel). In a pentamerous flower Silene coeli-rosa, the third primordium is located closer to the first primordium than the second one [34]. This is consistent with the third primordium position at α > 0, indicating the necessity of α, as we will discuss in the next section. The parameter region R0/Vτ for pentamery expanded with increasing α, whereas the border between the whorled and non-whorled arrangements was weakly dependent on α (Fig 3C). The tetramery, pentamery, and octamery arrangements were more robust to R0/Vτ and α than the hexamery and heptamery arrangements. Dominance of the particular number also appears in the ray-florets within a head inflorescence of Asteraceae [71], in which radial positions show the whorled-type arrangement [72]. Meanwhile, the leaf number in a single vegetative pseudo-whorl transits between two to six by hormonal control without any preference [73].
Moreover, the transition between the different merosities occurred directly, without the transient appearance of the non-whorled arrangement. This is in contrast to an earlier model [19] in which the transition between different merosity always involved transient spiral phyllotaxis. The fact that the merosity can change while keeping its whorled nature in flowers (e.g., the flowers of Trientalis europaea[74]) supports our results. To our knowledge, ours is the first model showing direct transitions between whorled patterns with different merosities as well as preferences for tetramery and pentamery, the most common merosities in eudicot flowers.
To further validate our model of the pentamerous whorl arrangement, we quantitatively compared its results with the radial distances and divergence angles in eudicot flowers. Here we focus on a Scanning Electron Microscope (SEM) image of the floral meristem of S. coeli-rosa, Caryophyllaceae (Fig 4A–4C) [34], because S. coeli-rosa exhibits not only five sepals and five petals in alternate positions, which is the most common arrangement in eudicots, but also the helical initiation of these primordia, which we targeted in the present model. In addition, to our knowledge, this report by Lyndon is the only publication showing a developmental sequence for both the divergence angle Δθk, k+1 = θk+1 −θk (0 ≤ Δθk, k+1 < 360) and the ratio of the radial position, rk/rk+1, referred to as the plastochron ratio [75], in eudicot floral organs. Reconstructing such developmental sequences of both radial and angular positions is an unprecedented theoretical challenge, while those which describe the angular position alone for the ontogeny of spiral phyllotaxis (180 degree, 90 degree and finally convergence to 137 degree [16, 33]; the ‘M-shaped’ motif, i.e., 137, 275, 225, 275 and 137 degrees [76, 77]) have been reproduced numerically.
By substituting the initial divergence angle between the first and second sepals of S. coeli-rosa into Δθ1,2 = 156 but not any plastochron data into the simulation (θ1 = 0 and θ2 = 156 degree), we numerically calculated the positions of the subsequent organs (Fig 4D). The observed divergence angle Δθ2,3 = 132 degree indicates α > 0, because Δθ2,3 = Δθ1,3 = (360−156)/2 = 102 degree at α = 0, in the present model setting r1 ≅ r2. Even when r1 > r2, the divergence angle was calculated as Δθ2,3 = 113 degree (r1 = R0+2Vτ, r2 = R0+Vτ, R0 = 1, Vτ = 0.14, and λini = 0.05 estimated from the S. coeli-rosa SEM image [34]; see S4 Fig for detail), which is still less than the observed value. As α became larger, the inhibition from the second primordium became stronger than that from the first one, making Δθ2,3 consistent with the observed value in S. coeli-rosa (Fig 4E, top).
For the subsequent sepals and petals, the model faithfully reproduced the period-five oscillation of the divergence angle and the plastochron ratio until the ninth primordium (Fig 4E), notably in the deviation of the divergence angle from regular pentagon (144 degree) and the increase of plastochron ratio at the boundary between the sepal and petal whorls. Moreover, a similar increase in the plastochron ratio occurred weakly between the second and third primordia in the first whorl (closed arrowhead in Fig 4E), indicating a hierarchically whorled arrangement (i.e., whorls within a whorl). Such weak separation of the two outer primordia from the three inner ones within a whorl is consistent with the quincuncial pattern of sepal aestivation that reflects spiral initiation in many of eudicots with pentamerous flowers (e.g., Fig 2D–E in [21]). Even with an identical set of parameters, the order of initiation in the first pentamerous whorl can vary depending on the stochasticity in the growth process. The variations of the initiation order in simulations may be caused by the absence of the outer structure, because the axillary bud seems to act as a positional information for the first primordia in S. coeli-rosa floral development (Fig 4B). The positioning of the five primordia in the first whorl was reproducible in 70% of the numerical replicates, within less than 20 degrees of that in S. coeli-rosa or that of the angles in a regular pentagon. Mismatches in the inner structure (from the tenth primordium, i.e., the last primordium in petal whorl) might be due to an increase in the rate of successive primordia initiation later in development [35], which we did not assume in our model. The agreements between our model and actual S. coeli-rosa development of sepals and petals in both the angular and the radial positions suggests that the S. coeli-rosa pentamerous whorls are caused by decreasing inhibition from older primordia.
A possible mechanism to arrest the radial displacement of a new primordium, a key process for whorl formation (arrowheads in Fig 2D and 2F), involves an inward-directed gradient of the growth potential Ug, k (Eq 3) of a new primordium so that its radial movement is prevented. To confirm this for tetramerous whorl formation (Fig 3A), we analytically derived the parameter region such that the radial gradient of the growth potential at the angle of the fifth primordium Ug,5 (Eq 3), which is determined by the positions of the preceding four primordia, is inward-directed. For ease in the analytical calculation, we set α = 0 and PMP = 0. The first four primordia positions were intuitively estimated (see S2 Text) as
r 1 = R 0 + 3 τ V , θ 1 = 0 r 2 = R 0 + 3 τ V , θ 2 = 180 r 3 = R 0 + 2 τ V , θ 3 = 90 r 4 = R 0 + τ V , θ 4 = 270 , (5)
which agreed with the numerical results with an error of less than several percent regardless of the parameter spaces. Hereafter we demonstrate a case Vτ = 6.0. The position of the fifth primordium derived from the positions of four existing primordia (Eq 5) becomes θ5 = 90 when R0 ≤ 2, whereas θ5 ∼ 135 when R0 > 2 (S5 Fig). Next, we calculated the potential for the fifth primordium in radial direction by substituting Eq 5 and the position of the fifth primordium θ5 into Eq 3. The function becomes
U g , 5 ( r , θ 5 ) = ∑ j = 1 4 exp ( - d 5 j λ g ) = ∑ j = 1 4 exp ( - r j 2 + r 2 - 2 r j r cos ( θ j - θ 5 ) λ g ) ⋅ (6)
The potential exhibits a unimodal (2 < R0 < 10; Fig 5A) or bi-modal (R0 < 2, R0 > 10; Fig 5B and 5C) shape. At R0 < 10, the potential gradient at the initiation position of the fifth primordium ∂Ug,5(r, θ5)/∂r∣r = R0 is outward-directed (Fig 5A), providing almost constant growth resulting a non-whorled arrangement in the simulations (Fig 3A, red region). At R0 > 10, we defined the radial position of the local maximum closest to the fifth primordium as rmax (open arrowhead in Fig 5B and 5C; red squares in the upper half of Fig 5D) and the local minimum as rmin (blue circles in Fig 5D; 0 < rmin < rmax). The potential gradient ∂Ug,5(r, θ5)/∂r∣r = R0 has a negative value when R0 < rmin or rmax < R0 (Fig 5C), causing the fifth primordium to constantly move outward. On the other hand, the potential gradient is positive, i.e., directed inward (Fig 5B), when rmin < R0 < rmax (between the two solid arrowheads in Fig 5D), causing the arrest of radial movement of the fifth primordium. The values of rmin and rmax, analytically calculated as function of R0 and τ (solid black line in Fig 5E), were faithfully consistent with the parameter boundaries between the non-whorled pattern and the tetramerous-whorled pattern and between the tetramerous-whorled and pentamerous-whorled patterns, respectively, in the numerical simulations (Fig 5E). The assumption r1 = r2 (Eq 5) according to our numerical results (Fig 2D), which is a similar setup to co-initiation of two primordia, is not a necessary condition for consistency (S6 Fig). Thus the inward-directed gradient of the growth potential (Eq 3), which works as a barrier to arrest the outward displacement of the fifth primordium, causes the formation of tetramerous whorl.
The inward radial gradient of the potential Ug, k (Eq 3) also accounted for the emergence of pentamerous whorls at α > 0. Unlike the case of α = 0, the angular position of the third primordium θ3 at the global minimum of Uini decreases from 90 degrees as α increases (Fig 6A). For example, the recursive calculations for the minimum of Uini gave the angular positions of the two subsequent primordia, θ3 ≅ 62 and θ4 ≅ 267, respectively, at α = 2.0 (Vτ = 6.0, R0 = 20.0, and PMP = 0). Those angular positions were consistent with the numerical results (e.g., Fig 2F and S2B Fig). The gradient of the growth potential ∂Ug,5(r, θ5)/∂r at the edge of the meristem for the fifth primordium that arises at θ5 ≅ 129 is negative (Fig 6B). Therefore, the fifth primordium moves outward at constant velocity so that the tetramerous whorl is unlikely to emerge. The inward-directed potential at the position of the new primordium first appears when the sixth primordium arises around 343 degrees, which was derived by the recursive calculation (Fig 6C). The first primordium (the rightmost potential peak in Fig 6C) prevents the outward movement of the sixth primordium (red circle in Fig 6C). Arrest of radial displacement of the sixth primordium is maintained until the seventh primordium arises to allow the radial gap between these primordia to appear (i.e., a pentamerous whorl emerged). After the appearance, the growth potential gradients of the sixth and the seventh primordia become outward-directed, providing their constant growth with keeping the radial gap to the first whorl. Likewise, the other merosities can be explained by similar recursive calculations of the angular position from the initiation potential (Eq 2) and the radial gradient of the growth potential (Eq 3).
Based on these analytical results (Figs 5 and 6) and the dimensionless parameter G = τV/R0, which represents the natural logarithm of the average plastochron ratio [49, 75], we quantitatively compared the present model against previous phyllotaxis models assuming simultaneous initiation based on the initiation potential [19]. The tetramerous and pentamerous whorls appeared in, at most, 1.3-fold and 1.2-fold ranges of G, respectively, in the earlier study (Fig 4D in [19]); however, they appeared in much wider ranges in our model (i.e., 3-fold to 5-fold and 1.2-fold to 5-fold ranges of G, respectively; Fig 3C). Here, another key parameter is the temporal decay rate of the initiation inhibition α that shorten the transient process approaching to the golden angle (Fig 6A) than those of spiral phyllotaxis [32, 33]. λini, representing the gradient of the initiation potential (Eq 2), little affected the border between the whorled and non-whorled arrangements at α = 0 (Fig 7A and 7B); λini affected the border only when α ≠ 0 (Fig 7C and 7D). The independency of λini at α = 0 is consistent with the result shown by the previous model, which did not incorporate temporal decay of the potential and indicated that the phyllotactic pattern depends little on the functional type of initiation potential [49]. On the other hand, the gradient of the growth potential (Eq 3) regulated by λg caused a drastic transition between the whorled and non-whorled arrangements (Fig 7E and 7F). Unlike G, λini, and α (Fig 6A), λg hardly affects the angular position, as demonstrated in the previous sections, but it controls how far the growth potential works as a barrier to determine the merosities of the whorls (Fig 7E and 7F). Thus, λg, α, and G differentially regulate phyllotaxis of the floral organs, suggesting the involvement of distinct molecular or physiological underpinnings.
We have seen that both the mutual repulsion of growth regulated by λg and the temporal decay of initiation inhibition controlled by α are responsible for the formation of tetramerous and pentamerous whorls following sequential initiation. These mechanisms can be experimentally verified by tuning λg and α. Here, we discuss several candidates for the molecular and physiological underpinnings.
Future studies should also clarify the limits and applicability of the common developmental principle elucidated here by exploring more complex development in a wide variety of flowers. Because our model assumes sequential initiation of the primordia, it does not cover the floral development of all eudicots; sepal primordia arise simultaneously in some eudicot clades (Fig 1; e.g., mimosoid legume [94]). Likewise, in later development, several primordia arise at once in the stamen and carpel whorls (e.g., Ranunculaceae [35]). The transitions between simultaneous and sequential development have two additional intriguing implications for evolutionary developmental biology. First, the initiation types may affect the stochastic variation of floral organ numbers, possibly caused by the absence or presence of pseudo-whorls (Fig 1) and the noisy expression domain of homeotic genes [95]. Second, such transitions occur even in animal body segmentation [3, 4], possibly caused by evolution of both gene regulatory network topologies and embryonic growth [7, 9–11]. The limitations of the model can be reduced by introducing initiation whenever and wherever the potential (Eq 2) is below a threshold, allowing simultaneous as well as sequential initiation [19]. The threshold model exhibiting both types of initiation does not by itself result in the dominance of particular merosities [19]. Incorporating two mechanisms, mutual growth repulsion and temporally decreasing inhibition at the point of initiation, into the threshold model could explain the dominance of particular merosities following both the sequential and the simultaneous initiation of floral organ primordia (Fig 1). Another problem is the absence of trimerous whorls in the present model (Fig 3). The transition between the trimery and tetramery or pentamery, and vice versa, occurred multiple times during the evolution of angiosperms. Therefore, trimerous flowers are scattered across the basal angiosperms, monocots, and a few families of eudicots [96, 97]. Elucidating the developmental mechanisms underlying the transitions between the different merosities, as well as those between sequential and simultaneous initiation, will be an important avenue for future studies.
One problem in determining floral organ number is how to generate whorls comprised of a specific number of organs. By introducing a growth assumption (i.e., continuous repulsion among primordia throughout development, which was originally proposed as the contact pressure model [25, 45, 46] and is supported by experimental observations [42]) into a dynamical model of phyllotaxis [49], we showed that the whorled arrangement arises spontaneously from sequential initiation. Moreover, when we allowed the inhibition to decay over time [33, 47, 48], pentamerous whorls became the dominant pattern. The merosity tended to be four or five in much larger parameter spaces than those in which it tended to be six or seven. The emergence of tetramerous and pentamerous whorls could be verified experimentally by tuning the two parameters α and λg.
|
10.1371/journal.ppat.1003289 | Low-Volume Toolbox for the Discovery of Immunosuppressive Fungal Secondary Metabolites | The secondary metabolome provides pathogenic fungi with a plethoric and versatile panel of molecules that can be deployed during host ingress. While powerful genetic and analytical chemistry methods have been developed to identify fungal secondary metabolites (SMs), discovering the biological activity of SMs remains an elusive yet critical task. Here, we describe a process for identifying the immunosuppressive properties of Aspergillus SMs developed by coupling a cost-effective microfluidic neutrophil chemotaxis assay with an in vivo zebrafish assay. The microfluidic platform allows the identification of metabolites inhibiting neutrophil recruitment with as little as several nano-grams of compound in microliters of fluid. The zebrafish assay demonstrates a simple and accessible approach for performing in vivo studies without requiring any manipulation of the fish. Using this methodology we identify the immunosuppressive properties of a fungal SM, endocrocin. We find that endocrocin is localized in Aspergillus fumigatus spores and its biosynthesis is temperature-dependent. Finally, using the Drosophila toll deficient model, we find that deletion of encA, encoding the polyketide synthase required for endocrocin production, yields a less pathogenic strain of A. fumigatus when spores are harvested from endocrocin permissive but not when harvested from endocrocin restrictive conditions. The tools developed here will open new “function-omic” avenues downstream of the metabolomics, identification, and purification phases.
| Several fungal pathogens produce bioactive small molecules, commonly known as secondary metabolites (SMs) that contribute towards disease development in susceptible hosts. Genome assessment of human pathogenic Aspergillus species indicates these fungi have the capabilities of producing hundreds of SMs, most of which are currently not characterized for their effect on human health and the immune system. This lack of knowledge is directly correlated to the difficulties of obtaining assayable quantities of pure metabolites. To overcome this roadblock in assessing the potential impact of SMs on the immune system, our laboratories have developed a two-tiered cost-effective, high-throughput program utilizing microfluidic platforms and a novel zebrafish model to identify SMs inhibiting neutrophil chemotaxis. Using minimal and physiologically relevant amounts of SMs, this systematic approach has identified the A. fumigatus spore SM, endocrocin, as a potent chemotaxis inhibitor. Interestingly, the production of endocrocin is temperature dependent and virulence studies with the endocrocin null mutant implicates the temperature at which the fungus forms spores as a factor in disease development.
| The secondary metabolome provides filamentous fungi with a biologically active panel of molecules, deployed in the presence of competing/host organisms or specific microenvironmental factors, and increasingly found to afford both physical and competitive fitness to the producing fungus [1]. Although the study of fungal secondary metabolism has reached the ‘omics era, with the development of tools for efficient genetic exploration [2] and the improvement of HPLC and LC-MS methods [3], a significant challenge remains in identifying the biological activity of the purified compounds. The minute quantity of metabolites (nano- to micrograms) collected from the latter methods and the large number of isolated compounds (hundreds to thousands) are important limiting factors in this endeavor. Thus, as the methods for identifying SM gene clusters and the compounds they produce are becoming well established [4], there is an increasing need for improved assays, compatible with the fungal metabolomics process, that can reveal the biological activity of metabolites produced and break a bottleneck in scientific advancement.
Aspergillus spp. SMs are of particular interest in medical research as the genus is genetically accessible, produces a plethora of bioactive compounds [5], and contains several opportunistic pathogenic species including A. fumigatus and A. nidulans whose SMs are assessed in this study [6], [7]. Though it is likely that a number of factors together contribute in making these species effective pathogens, SMs play an important role in the virulence of Aspergillus-related diseases as direct toxins and modulators of the immune response [8], [9]. As the innate immune response is the primary line of defense against fungal spores in the lung, inhibition of essential functions of these cells may confer to the fungi an ability to evade immune clearance, and increase its pathogenicity. These findings highlight the necessity of mapping the interactome between fungi and host organisms to establish the pathomechanism of fungal diseases as well as to bioprospect.
Leading to this current study are the series of works showing LaeA, a global regulator of secondary metabolism to be a virulence factor not only in pathogenic Aspergilli [10]–[13] but in all filamentous pathogenic fungi assessed to date [12], [14], [15]. Examination of the laeA mutant in A. fumigatus implicated unidentified SMs in development of invasive aspergillosis [10], [11]. As several studies have shown A. fumigatus culture filtrates to inhibit neutrophil chemotaxis [16]–[18], we considered it possible that LaeA-regulated SMs could be chemotaxis inhibitors. Complicating this hypothesis, however, is the fact that LaeA regulates dozens of SM clusters, all of which can produce multiple derivatives from the same biosynthetic pathway, whose purification results in small available quantities [19].
Traditional in vitro neutrophil migration models, often performed in well-plates, do not allow a good level of control over the migration microenvironment, do not allow imaging of the cells during the migration process, and require large amounts of purified compound (micrograms to grams in hundreds of microliters to milliliters). Advances in microscaled assays have demonstrated enabling characteristics, with the ability to use microliters or less of reagents, develop high-throughput applications, and design assays with more control over the micro-environment [20]. The development of open systems, that interface with existing fluid handling equipment, contributed in making microscaled assays more accessible and better suited for screening libraries of individual compounds [21]. Further, these approaches enable the development of functional cell-based assays, such as arrayed leukocyte recruitment assays [22]. However, these have not been applied for identifying SMs modulating the fungal-immune interaction, nor for screening of fungal SMs.
Appropriate in vivo models are also required to support in vitro progress. Current models such as Galleria [23] or Drosophila [24], are accessible and inexpensive but do not fully recapitulate the vertebrate immune system. More relevant models such as the murine model are logistically challenging, require excessive amounts of purified SMs, and are still fraught with deficiencies in visualization of innate cell response [25]. Recently, the development of the zebrafish embryo model has proven to be a vertebrate model well suited for leukocyte studies as it is readily accessible, small, and transparent and has been established as a model system for infectious diseases including fungi [26].
Here, we demonstrate a two-tiered screening approach for identifying the immunosuppressive and neutrophil recruitment inhibitory activity of LaeA-regulated SMs, capitalizing on advances of microscale in vitro systems and an original zebrafish model. An arrayed microfluidic in vitro neutrophil recruitment platform was developed, compatible with manual and automated pipettes, allowing for rapid assessment of the neutrophil recruitment inhibition properties of purified Aspergillus SMs. Passive open microfluidic methods were employed for creating arrayed gradient-generation devices operable in typical biological laboratory settings and minimizing reagent use. Bioactive metabolites identified from this platform were then assessed in an in vivo zebrafish recruitment assay. The zebrafish assay is enabling as it significantly reduces the quantity of compounds required and provides a quickly assessed window into innate immune response. Using this approach, we report the identification of the neutrophil recruitment inhibition activity of endocrocin, an A. fumigatus SM, and further characterized its localization in spores of the growing fungus.
In order to systematically test the large number of compounds provided by liquid chromatography methods, a microscale platform for assessing neutrophil chemotaxis properties was designed. The use of tubeless microfluidic-interfacing methods minimizes dead volumes and static gradient-generation methods further minimizes volume requirements [22], [27]. Thus, neutrophil recruitment can be assessed with as little as 3 µL of purified compound, using only a simple micropipette. In this approach, the gradient is generated in a reproducible way by leveraging a flow bypassing method based on ensuring that no undesired flow passes through the gradient channel, rather a second flow path of significantly lesser fluidic resistance diverts the bulk of the flow (Figure 1A,B). The reliability of the fluidic handling mechanism used, and the simplicity of the design, enables rapid loading protocols and large arrays of microdevices. In order to achieve the highest throughput possible, batch-processing algorithms were developed, enabling the quantification of neutrophil migration properties from an endpoint phase-contrast image of the migrating neutrophils. Together, these advances allow the processing by hand of up to 300 migration data-points per experimental run in an embodiment that can be readily interfaced using handheld pipettes and automated liquid handlers.
The static gradient-generation device was verified using fluorescent dyes and food colorant (Figure 1B), demonstrating that the gradient establishes in minutes as assessed by fluorescent imaging of AlexaFluor488 (Figure 1C). The microfluidic gradient, once established, was able to maintain steady for several hours, allowing the reliable measure of neutrophil (or other leukocyte) migration over the course of a typical neutrophil experiment (Figure 1C). The microfluidic channel can be entirely prepared in 3 simple pipetting steps, which require seconds, and can be interface-able with electronic and robotic liquid handlers (Figure 1D). Using the flow generation method described, as well as the static gradient-generation methods, we demonstrate the ability to create large arrays of gradient devices which can be prepared in large batches allowing the operation of several hundred per day (Figure 1E). As these contain very little dead volumes (being devoid of tubes and actuation equipment) a single blood draw of 10–20 mL is sufficient to operate thousands of these channels, provided the appropriate liquid handling and data acquisition equipment is used.
In order to systematically test the large number of compounds provided by liquid chromatography methods, the microscale platform was used to screen for compounds inhibiting neutrophil recruitment. The purified compounds were introduced in conjuncture with a known chemoattractant (fMLP) in the gradient-source reservoir, causing the recruitment of neutrophils into the gradient microchannels (Figure 2A,B). A decrease in the number of neutrophils recruited compared to the positive control suggests an inhibitory effect and is considered as a positive hit for the screen. Using this method, a set of 20 purified compounds (each inserted in a volume of 3 µL at 10 µM, n = 9) from A. nidulans and A. fumigatus was screened (Figure 2C). Results show that four compounds – 8-hydroxyl emodin, austinolide, F-9775B, and endocrocin – display a significant decrease in neutrophil migration.
As in vitro assays do not account for the complexity of whole organisms, we developed a low-volume in vivo neutrophil recruitment assay based on advances in zebrafish models. Zebrafish are a very attractive model for developing neutrophil research assays as they are transparent, allow the observation of immune cells in vivo, low-cost, and importantly have a strong immunological resemblance to the human system [26]. However, most neutrophil recruitment assays in zebrafish to date were performed through a wounding assay in which the fish is wounded by a needle stab or a cut in the fin or the body [28]. This approach has been successful but requires the individual manipulation of each fish, which is not amenable to screening applications. We developed a neutrophil recruitment inhibition assay based on the ability of a known chemoattractant, leukotriene B4 (LTB4), to diffuse through the skin of the zebrafish and induce neutrophil recruitment out of the Caudal Hematopoietic Tissue (CHT). In conjuncture with a neutrophil migration inhibitor added to the well in which the zebrafish are bathing, the level of recruitment can be reduced and easily assessed (Figure 3A). We validated this assay using LY294002, a known neutrophil chemotaxis inhibitor targeting PI3K, and found that neutrophil recruitment to the CHT was entirely suppressed (Movie S1 and S2).
This assay is rapid and accessible as many zebrafish are treated simultaneously in a multi-well plate and it does not require the use of wounding or micro-injection methods. Furthermore, the fish can be readily fixed and neutrophils quantified using simple optical microscopy (Figure 3B). Based on availability, three of the four compounds identified in the in vitro screen were tested for their inhibitory properties to neutrophil chemotaxis. Results show that only endocrocin displayed a significant reduction in the number of neutrophils recruited to the tail fin (Figure 3B,C). We further assessed the general cytotoxicity of the compound, and did not observe a reduction in zebrafish survival after 4 days at endocrocin concentrations up to 10 µM (data not shown). Similarly, endocrocin did not induce an increase in neutrophil death in the timescale and concentrations used in the microfluidic assays (Figure S1).
We investigated properties of endocrocin production in order to gain insight on its mode of interaction with a host. Endocrocin is a LaeA regulated anthraquinone whose biosynthesis pathway was recently identified in A. fumigatus [29]. Given the potential role of endocrocin for modulating the immune response, we sought to characterize the mode of production and tissue specificity of this SM. We analyzed the production of endocrocin in both the WT strain and a ΔencA strain of A. fumigatus grown on GMM-agar at temperatures varying from 25°C to 42°C (Figure 4A). At temperatures below 35°C, the WT fungus produces endocrocin, while at higher temperatures it does not. As expected, the ΔencA strain did not produce endocrocin. The location of endocrocin production was characterized by analyzing crude extracts from different fractions of the fungal culture grown on solid medium: the conidia, the mycelia (top agar), and secreted metabolites (bottom agar). Endocrocin was only observed in the conidial fractions and not in the fractions containing mycelia (top agar) or soluble secreted factors (bottom agar, Figure 4B).
We identified the functional range of endocrocin inhibition by performing a dose response assay of endocrocin using the microfluidic neutrophil migration platform. Results show that a significant reduction of neutrophil inhibition can be observed at concentrations as low as 100 nM, with an inhibition of up to 40% for concentrations of 10 µM (Figure 5).
Finally we assessed the pathogenicity attribute(s) of products from the endocrocin gene cluster in vivo by using the established Toll-deficient Drosophila invasive aspergillosis model [30], [31]. This model, while not possessing the immune system of mammalians, is utilized successfully for studying virulence of A. fumigatus and, importantly, meets the affordable and rapid methodology of this study. Drosophila were inoculated with A. fumigatus wild-type and ΔencA spores harvested from 25°C (endocrocin stimulating) or 37°C (endocrocin restrictive) environments. Attenuated virulence was observed when flies were inoculated with spores of ΔencA grown at 25 but not 37°C, consistent with the observed temperature-dependent production of endocrocin (Figure 6A,B).
Modern metabolomic techniques provide the ability to identify and isolate hundreds of microbial compounds in academic research settings. Downstream of these techniques, there is an increasing need for low-volume platforms that can help map the multi-kingdom interactome in a low-cost laboratory setting. We present here a novel and economical approach for identifying immunosuppressive properties of microbial compounds. The two-tiered approach offers an accessible solution for performing a screen both in vitro, on human primary neutrophils, and in vivo, on whole organism zebrafish models. Both methods described are compatible with natural product extraction methods with small yields and are scalable as they allow batch processing. The combination of assessing the response of human primary cells - a more precise model for medical studies - and using zebrafish for a more holistic in vivo assessment of specific innate immune responses, covers a highly relevant scope of models for predicting the effect of metabolites in humans.
While a multitude of microfluidic gradient and neutrophil migration platforms have been demonstrated to date, this is the first systematic assessment of the neutrophil migration properties of Aspergilli SMs in a microfluidic embodiment. It is worthy to note, however, that due to material interaction and a higher surface-to-volume ratio in microscale devices, it is possible for the compound tested to be sequestered by the material and not be able to diffuse to the neutrophils, leading to false negatives. Using this microscaled approach, we find that four of the compounds tested displayed migration inhibitory properties. The in vivo zebrafish assay complements the in vitro assay as it uses an equally scalable and potentially automatable approach, as specific manual treatment of the fish used for each compound is not necessary. One potential caveat is the necessity for the compound to traverse the skin barrier in order to have an effect on the neutrophils. In this aspect, the assay can be made more sensitive by pre-incubating the fish in the purified compounds prior to adding the chemoattractant. Together, these assays represent a new avenue for the discovery of immunosuppressive properties of microbial SMs. As an example, we found that one of the compounds selected from the in vitro screen, endocrocin, displayed significant leukocyte recruitment inhibitory properties.
Endocrocin was only recently identified as a LaeA regulated SM in A. fumigatus [28]. Coupling the observations that laeA loss yields reduced virulent strains and that several studies have shown uncharacterized components of A. fumigatus filtrates inhibit neutrophil chemotaxis, we suggest that endocrocin is one of these components. The fact that endocrocin did not result in zebrafish death nor increase neutrophil death under the assay conditions used (Figure S1) further supports a more specific function for this metabolite than mere cytotoxicity.
The study of endocrocin showed that it is located in the fungal spores, and may be released upon contact with a humid environment such as the host lung tissues or upon early germination of spores that managed to evade macrophage clearance. Previous studies have found that certain fungal metabolites, associated with the cell wall of the inhaled spores, have the ability to interact detrimentally with the lung epithelia [32], [33]. This study provides insight into the role of spore-borne metabolites as “protective” constituents during early lung colonization, in which the spores are pre-armed with metabolites that could provide the germinating spore with an advantage over the immune system. Spores containing these “protective” attributes that could modulate or negate early host innate confrontations – the first line of defense towards Aspergillus infections – may provide the fungus with leverage during this initial host-pathogen arms race. Thus, spore-borne metabolites, highly dependent on the micro-environmental conditions in which the spores originate (temperature, nutrient sources, microbial interactions, etc) may play an important and understudied role in the pathomechanism of fungal opportunistic diseases.
Purified fungal metabolites were obtained by culturing Aspergillus species (A. fumigatus and A. nidulans wild-type and mutant stains) in liquid shake or solid agar conditions. For the liquid shake culture, the supernatant was collected and small molecules extracted by freeze drying and methanol extraction. For solid agar cultures, the agar was homogenized and soaked in 800 mL of 1∶1 CH2Cl2/Methanol for 24 h. After filtration, the combined extract was evaporated in vacuo to yield a residue, which was suspended in water (500 mL) and then partitioned with ethyl acetate (500 mL) three times. The combined ethyl acetate layer was evaporated in vacuo to afford a crude extract (the weight for each deletant is list below). The crude extract was applied to a Si gel column (Merck, 230 to 400 mesh, ASTM, 20×80 mm) and eluted with 250 mL CH2Cl2/Methanol mixtures of increasing polarity (fraction A, 1∶0; fraction B, 19∶1; fraction C, 9∶1; fraction D, 7∶3). Each fraction was examined by high performance liquid chromatography–photodiode array detection–mass spectrometry (HPLC–DAD–MS) and the fraction contained target natural products was applied to a gradient HPLC on a C18 reverse phase column (Phenomenex Luna 5 µm C18 [2], 250×10 mm) with a flow rate of 5.0 mL/min and measured by a UV detector at 254 nm. The gradient system was acetonitrile (solvent B) and 5% Acetonitrile/H2O (solvent A) both containing 0.05% Tetra-Fluoroboric Acid.
The neutrophils were purified from whole blood of consenting self-reported healthy donors. A volume of 7 mL of whole blood was placed in a 15 mL conical tube and 7 mL of Polymorphoprep liquid (Axis Shield, UK) was layered above. The conical tube was centrifuged for 20 min at 1200 rpm followed by 10 min at 1700 rpm. The neutrophil layer was removed, placed in a 50 mL conical tube, and 1× PBS was added to fill the tube up to 50 mL. The conical tube was subsequently centrifuged for 10 min at 1500 rpm. The pellet was re-suspended in 9 mL of de-ionized H2O for 30 s after which 1 mL of 10× PBS and 40 mL of 1× PBS were added. The conical tube was centrifuged again for 10 min at 1500 rpm, and the pellet was re-suspended in 1 mL of PBS. The neutrophils were counted and the cell suspension was diluted to a final concentration of 4 million per mL.
A silicon-based mold for the microfluidic device-arrays was fabricated by creating the design on Illustrator (Adobe, USA), and printing on a high-resolution film (Imagesetter, Inc, Madison, USA). The main channels of the microfluidic device were designed to be 1 mm wide and 400 µm tall, while the gradient channel is 35 µm tall, 150 µm wide, and 1 mm long. A photo sensitive epoxy, SU-10 (Microchem, USA), was spun on a 150 mm diameter wafer (WRS, USA) to a thickness of 35 µm and backed for 10 min at 95°C. The first film defining the migration channels was placed on the wafer and exposed to 200 mJ of UV light from an Omnicure light source (EXFO, Canada), after which the wafer was post-backed at 95°C for 5 min. A second layer of epoxy, SU-100, was spun on the wafer to a thickness of 400 µm, backed for 90 min at 95°C, exposed with 1200 mJ of UV light with the second film defining the microfluidic channels, and backed for 30 min. A third layer of SU-100 was spun and exposed with the same parameters as the second layer, albeit with the third film defining the ports. The wafer was developed in SU-developer (Microchem, USA) for 3 hours on a shaker, washed with acetone, rinsed with iso-propanol, and dried using compressed air.
Microfluidic arrays were designed and fabricated using silicone polymer-based soft-lithography methods by replicating a silicon-SU8 master mold [34]. In brief, a silicone polymer, PDMS (Sylgaard 184, Dow Corning, USA), was mixed with a ratio of 1∶10 of curing agent to monomer base and placed in a vacuum for 30 min for degassing. The molding process was performed by placing, in order, on a hot plate a transparency sheet (Cheap Joe's, USA), the silicon-based mold, degassed PDMS, a second transparency sheet, a 1 mm thick layer of silicon foam (Mc Master, USA), a rectangle of glass, and a 5 kg weight. The hot plate was heated to 80°C for 3 hours and cooled down completely prior to PDMS removal. The PDMS layer was peeled off the silicon-based mold, bonded non-covalently in a polystyrene Omnitray dish (NUNC, USA). The Omnitray was filled with 20 mL of PBS and placed in a vacuum chamber for 15 min in order to fill the microdevices. The superfluous PBS was removed, 4 µL of mHBSS (1× PBS containing 0.1% HSA and 0.2 mM of HEPES buffer) were placed on the output well and 4 µL were placed on each input well in order to replace the fluid contained in each device.
Purified dried compounds were resuspended in DMSO to 1 mM and subsequently diluted 1∶100 in mHBSS containing 100 nM of fMLP – a known chemoattractant. 4 µL of neutrophil suspension at 4.106 cells per mL was inserted into the sink channel of the device, followed by 3 µL of compound in the source channel, and the Omnitrays were placed in a CO2 incubator. After 45 min, the Omnitrays were removed, placed on a Nikon Eclipse microscope, and phase-contrast images of the migration channels were taken at 10× magnification. Image analysis was performed using the software package Je'Xperiment developed in-house (source code available on sourceforge.net or upon request) allowing the identification and quantification of the neutrophils invading the migration channel on a batch processing scale. A Wilcoxon rank-sum test was performed on sets of 5 data points to evaluate the statistical significance of the effect of each compound. Those with a p-value of less than 0.05 were considered statistically significant.
Zebrafish were maintained according to the protocols approved by the University of Wisconsin-Madison Research Animal Resources Center. Zebrafish larvae at 3 days post fertilization were placed in the wells of a 24 well-plate (20 per well). The larvae were pre-incubated in 500 µL of E3 (egg water) containing 10 µM of the purified fungal compound in DMSO. After 1 hour, LTB4 was added to each well at a final concentration of 30 nM. The larvae were fixed after 30 min and stained using Sudan Black [26]. Numbers of neutrophils recruited to the ventral fin of each fish were counted manually and representative images were taken with a phase contrast upright microscope. Statistical significance was estimated using a Kruskal-Wallis test followed by Dunn's Multiple Comparison test in the software GraphPad Prism (USA). The animal handling protocols were performed according to the Guide of the Care and Use of Animals of the National Institutes of Health.
Aspergillus fumigatus strains used in this study are listed in Table S1. Strains were maintained as glycerol stocks and activated on glucose minimal media (GMM). Conidia were harvested in 0.01% Tween 80 and enumerated using a hemocytometer. Strain TFYL7.1 was constructed by targeted gene deletion of encA using a deletion cassette made via double-joint fusion PCR described in [29]) into strain AF293.1. Internal primers to encA was used to confirm the absence of its open reading frame and single integration of the deletion cassette was confirmed via Southern analysis using two different restriction digest profiles as described in [29]) (data not shown).
For temperature-dependent characterization of endocrocin production, A. fumigatus WT and ΔencA strains were point-inoculated at 1×104 conidia/inoculum onto solid GMM and incubated at temperatures ranging from 25–42°C without light selection. A 1.2 cm diameter core was removed from the middle of the fungal culture and homogenized in 2 mL of 0.01% Tween 80. The homogenized mixture was extracted with equal volumes of ethyl acetate and routine vortexing at room temperature over the course of 30 min. The mixture was centrifuged for 5 min at 3,500 rpm and 1 mL of the ethyl acetate layer was removed and allowed to evaporate at room temperature to yield a dried crude extract. For TLC analysis, the crude extract was reconstituted in 50–100 µL of ethyl acetate and 5–10 µL was spotted onto a 250 µM analytical silica plate (Whatman, Cat 4410-222) and subjected to a toluene: ethyl acetate: formic acid (5∶4∶0.8) resolving phase. Plates were visualized at 366 nm using a FOTO/Analyst® Investigator gel imaging system (Fotodyne Inc).
For tissue specific extraction of endocrocin, a suspension of 1×106 conidia in molten GMM top agar (0.75% w/v agar) was overlaid over solid GMM bottom agar (1.5% w/v agar). SM from different developmental parts of the fungus were obtained as follows: conidia was obtained by gently tapping the fungal culture with the petri dish lid side down as described previously [35], The remaining conidia and conidiophores were removed by gently scraping the surface of the fungal culture with 0.01% Tween 80 followed by multiple washes to remove residual conidia/conidiophores. The mycelia (free of aerial conidiophores/conidia) were obtained by peeling off the top agar from the culture plate. The bottom agar that is now free of mycelia (as inspected under the microscope) was extracted. A diagram that depicts the various sampling parts of the culture plate can be found in Figure 3B. Extraction process and TLC analyses were performed as described above.
We developed an image-processing algorithm that is able to identify un-labeled neutrophils in a phase-contrast image and quantify their number of migration distance (Figure1F). Used in conjuncture with a software platform developed in-house, called Je'Xperiment (Source code available on sourceforge.net or on request), which allows batch processing and data mining, we show a first step towards creating a high-throughput solution for quantifying the effect of fungal SMs on the inhibition of leukocyte migration.
Flies were generated by crossing flies carrying a thermosensitive allele of Toll (Tl r632) with flies carrying a null allele of Toll (Tl I-RXA) [36]. Two- to four day old adult female Toll-deficient flies were used in all of the experiments. Twenty flies were infected with each A. fumigatus strain used in this study. A. fumigatus isolates (ΔencA and wild-type) were grown on yeast extract agar glucose (YAG) either at 37°C or 25°C. Conidia were collected in sterile 0.9% saline from 2 days old cultures. The conidial concentration suspension was determined by using a hemacytometer and adjusted to 1×108 per mL. The dorsal side of the thorax of 20 CO2 anesthetized flies was punctured with a thin (10 µm) sterile needle that had been dipped in a concentrated solution of A. fumigatus conidia (107 spores/mL). As a negative control group, Toll-deficient flies were punctured with a 10 µm sterile needle and monitored daily for survival. Flies that died within 3 h of the injection were considered to have died as a result of the procedure and were not included in the survival rate analysis. The flies were housed in a 29°C incubator to maximize expression of the Tl r632 phenotype [31]. The Toll-deficient flies were transferred into fresh vials every 3 days. Fly survival was assessed daily over 10 days. Each experiment was repeated 3 times on different days and at the same time of the day to eliminate variability due to circadian rhythm.
Neutrophils were obtained from whole blood of self-reportedly healthy donors, from which we obtained informed and written consent at the time of the blood draw with approval of the University of Wisconsin-Madison Center for Health Sciences Human Subjects committee.
|
10.1371/journal.pntd.0002835 | Enhanced Passive Bat Rabies Surveillance in Indigenous Bat Species from Germany - A Retrospective Study | In Germany, rabies in bats is a notifiable zoonotic disease, which is caused by European bat lyssaviruses type 1 and 2 (EBLV-1 and 2), and the recently discovered new lyssavirus species Bokeloh bat lyssavirus (BBLV). As the understanding of bat rabies in insectivorous bat species is limited, in addition to routine bat rabies diagnosis, an enhanced passive surveillance study, i.e. the retrospective investigation of dead bats that had not been tested for rabies, was initiated in 1998 to study the distribution, abundance and epidemiology of lyssavirus infections in bats from Germany. A total number of 5478 individuals representing 21 bat species within two families were included in this study. The Noctule bat (Nyctalus noctula) and the Common pipistrelle (Pipistrellus pipistrellus) represented the most specimens submitted. Of all investigated bats, 1.17% tested positive for lyssaviruses using the fluorescent antibody test (FAT). The vast majority of positive cases was identified as EBLV-1, predominately associated with the Serotine bat (Eptesicus serotinus). However, rabies cases in other species, i.e. Nathusius' pipistrelle bat (Pipistrellus nathusii), P. pipistrellus and Brown long-eared bat (Plecotus auritus) were also characterized as EBLV-1. In contrast, EBLV-2 was isolated from three Daubenton's bats (Myotis daubentonii). These three cases contribute significantly to the understanding of EBLV-2 infections in Germany as only one case had been reported prior to this study. This enhanced passive surveillance indicated that besides known reservoir species, further bat species are affected by lyssavirus infections. Given the increasing diversity of lyssaviruses and bats as reservoir host species worldwide, lyssavirus positive specimens, i.e. both bat and virus need to be confirmed by molecular techniques.
| According to the World Health Organization rabies is considered both a neglected zoonotic and a tropical disease. The causative agents are lyssaviruses which have their primary reservoir in bats. Although bat rabies is notifiable in Germany, the number of submitted bats during routine surveillance is rarely representative of the natural bat population. Therefore, the aim of this study was to include dead bats from various sources for enhanced bat rabies surveillance. The results show that a considerable number of additional bat rabies cases can be detected, thus improving the knowledge on the frequency, geographical distribution and reservoir-association of bat lyssavirus infections in Germany. The overall proportion of positives was lower than during routine surveillance in Germany. While the majority of cases were found in the Serotine bat and characterized as European bat lyssavirus type 1 (EBLV-1), three of the four EBLV-2 infections detected in Germany were found in Myotis daubentonii during this study.
| Lyssaviruses are non-segmented negative-strand RNA viruses of the order Mononegavirales, family Rhabdoviridae and causative agents of rabies in bats and other mammals as well as in humans [1]. While rabies in dogs and other carnivores has been known since antiquity, the first evidence of rabies in haematophagous and insectivorous bats was reported from the Americas in the first half of the 20th century [2]. Since 1954, bat rabies cases have also been reported from other continents. Antigenic and genetic analyses revealed the diversity of different lyssavirus species, and to date, besides classical rabies virus (RABV), thirteen additional lyssaviruses have been discovered, mostly in bats [3]. Beyond Europe, Lagos bat virus (LBV), Mokola virus (MOKV), Duvenhage virus (DUVV), Shimoni bat virus (SHBV), and Ikoma lyssavirus (IKOV) were found in Africa. In Asian bat species, Aravan virus (ARAV), Khujand virus (KHUV), and Irkut virus (IRKV) were isolated. With the exception of MOKV and IKOV, all of those viruses were detected in bats [3]. In Australia, which has a long history of freedom from classical rabies, Australian bat lyssavirus (ABLV) is found in insectivorous and pteropid bats [4].
In Europe, bat rabies is also caused by several lyssavirus species. Between 1977 and 2012, a total of 1039 bat rabies cases were reported from European countries (http://www.who-rabies-bulletin.org). The majority was characterized as European bat lyssavirus type 1 (EBLV-1) isolated from Eptesicus bat species (E. serotinus, E. isabellinus) [5]. Genetically, EBLV-1 can be divided in two subtypes, EBLV-1a and 1b [6], [7]. While the EBLV-1a subtype is predominantly found in Central and Eastern Europe (France, The Netherlands, Denmark, Germany and Poland), EBLV-1b has been reported from Spain, France, Southern Germany, and central Poland [8]–[11].
European bat lyssavirus type 2 (EBLV-2) has been isolated from Daubenton's bats in the UK, Switzerland, Finland and Germany, and from Pond bats (M. dasycneme) in The Netherlands [12]. As of today, three Natterer's bats (Myotis nattereri) infected with the novel Bokeloh bat lyssavirus (BBLV) have been found in Germany and France [13]–[15]. A single detection of the West Caucasian bat virus (WCBV) in a Schreiber's bent-winged bat (Miniopterus schreibersii) has been reported from Western Caucasus Mountains [16]. Interestingly, specific RNA from a putative new lyssavirus named Lleida bat lyssavirus (LLEBV) was detected in brain material from the same bat species collected in Spain [17].
The public health relevance of bat rabies in general is highlighted by the fact that most of the bat associated lyssaviruses have caused human rabies [18]. In Europe, both EBLV-1 and EBLV-2 were responsible for four confirmed human casualties [19]. Also, sporadic spill-over infections of EBLV-1 to terrestrial mammals have been reported, i.e. in sheep in Denmark, two cats in France and a stone marten (Martes foina) in Germany [20]–[22].
Because of the zoonotic character of bat lyssaviruses knowledge about distribution, abundance and epidemiology is important to estimate and subsequently reduce the public health risk posed by bat rabies. Guidelines for the surveillance of bat lyssaviruses in Europe were established by the European research consortium Med-Vet-Net [23] and supported by EUROBATS [24]. The investigation of sick or dead bats for lyssavirus antigen in brain samples (passive surveillance) and testing of oro-pharyngeal swab samples and serum samples from free-living indigenous bats (active surveillance) for the presence of viral RNA or virus neutralizing antibodies, respectively, were recommended. However, the levels of active and passive bat rabies surveillance in Europe are still very heterogeneous despite previous recommendations [5]. Based on published data, active surveillance provides only limited information and cannot replace passive bat rabies surveillance [25].
Comprehensive passive bat rabies surveillance was conducted in The Netherlands [26], the United Kingdom [27], France [28] and Germany [10]. With the exception of Germany, passive surveillance in these countries is realized by only one or two cooperating departments investigating all bats submitted from the whole country. In contrast, rabies diagnosis in Germany is the responsibility of the sixteen federal states [10]. Dead or diseased bats with symptoms suggestive of rabies, particularly after contact with humans (bites and scratches) have to be submitted and tested for lyssavirus infection in the regional veterinary laboratories.
While cases of this notifiable disease in carnivores and bats were reported to the National Reference Laboratory for Rabies at the WHO Collaborating Centre for Rabies Surveillance and Research (FLI Riems, Germany), the number of bats tested negative was only sporadically submitted. Furthermore, the identification of bats to species level is generally missing as in some other European countries [5]. Therefore, routine bat rabies surveillance in Germany has relied on limited and opportunistic sampling which may not be representative of the true epidemiological situation [10].
To overcome these limitations and to obtain further information on the epidemiology of bat rabies in Germany an enhanced passive retrospective surveillance study was started at FLI in 1998. In this study, the focus was on dead bats excluded from routine diagnostic testing. This included bats obtained from (private) collections from different parts of Germany. Each sample was identified to species level, partly by molecular tools and tested for lyssavirus infection. Here, we present the data from this study and compare it with published data from routine diagnostic screening.
Dead bats were submitted under the prevailing laws of the respective federal states and following EUROBATS guidelines [24]. Because this study was in the frame of a surveillance programme conducted by the national reference laboratory for rabies no further permits were necessary.
Starting in 1998, on a federal state level bat conservationists, as well as various institutions and authorities (e.g. Nature and Biodiversity Conservation Union (NABU, Germany), Museum of Natural Science, wildlife care centers) were requested and encouraged to submit dead bats for rabies diagnosis irrespective of the circumstances of acquisition. Archived or newly acquired bats were submitted from all 16 German federal states by local bat biologists during long-time monitoring or routine inspection of maternity roosts, wintering grounds or were killed by cats, wind turbines or unintentional removal of roosts. All bats were stored frozen prior to submission in a chilled state.
Usually, bat carcasses were submitted with additional information, e.g. geographical origin, date found, sex, age and species identification. Bats without species information were determined to genus or even to species level using external morphological features [29]. Bat carcasses which were degraded or damaged and those suspected to represent cryptic bat species (e.g. Myotis mystacinus, M. brandtii and M. alcathoe) were identified by a mitochondrial cytochrome b (cyt b) gene specific PCR [30] if not restricted by museum specific preservation requirements. For this purpose, wing membrane samples were collected and stored separately in Eppendorf tubes at −80°C until analysis. For DNA preparation a small piece (1.0×1.0 mm) of each sample was lysed overnight (56°C, 400 rpm) using 3 µl proteinase K (10 mg/ml) and 300 µl lysis buffer (50 mM KCL, 10 mM TRIS-HCL (pH 9.0), 0.45% nonidet NP 40 and 0.45% Tween 20). After centrifugation (1 min, 13000 rpm) the supernatant was stored at −20°C. For PCR amplification two primer pairs (CytB Uni fw: 5′-CATCMTGATGAAAYTTYGG-3′ and CytB Uni rev: 5′-ACTGGYTGDCCBCCRATTCA-3′ [30]; HG for: 5′-CACTACACATCAGAYAC-3′ and HG rev: 5′-AAGGCGAAGAATCGRGT-3′) were used to obtain fragments of about 950 bp and 400 bp, respectively. The latter primer mix was developed based on reference material submitted to the AHVLA to identify all bat species indigenous to the UK (data not shown).
The PCR reaction mix (total volume of 25 µl) consisted of RNase-free water (17.65 µl), 25 mM of each dNTP (0.5 µl), 50 mM MgCl2 (0.75 µl), 10 pmol/µl of each primer (0.5 µl), 10x PCR RxN Buffer (2.5 µl), (0.1 µl) Platinum-Taq DNA Polymerase (Invitrogen, Darmstadt, Germany) and 2.5 µl template DNA. The amplification was performed with the following temperature profile: 3 min at 94°C (initial denaturation), followed by 50 cycles of 30 s at 94°C (denaturation), 30 s at 47°C (annealing), 1 min at 72°C (elongation) and a final extension at 72°C for 10 min. Amplification of the expected products was confirmed in a 1% agarose gel stained with ethidium bromide or SYBR safe DNA gel stain. PCR products were then purified (NucleoSpin Gel and PCR Clean-up kit, Macherey-Nagel, Düren, Germany) and sequenced using BigDye Terminator v1.1 Cycle Sequencing Kit (Applied Biosystems/Life Technologies, Carlsbad, CA, USA). The cytochrome b sequences were compared with published sequences of European bat species (GenBank) using the Basic Local Alignment Search Tool (BLAST, http://blast.ncbi.nlm.nih.gov/Blast.cgi) and the species determination was finalized by identification of the species of the highest nucleotide sequence similarity (≥90%).
Rabies diagnosis was performed on bat brain samples which were removed either by opening of cranium or, in case of natural scientific collections, by puncture of foramen occipitale magnum using a 26-gauge needle. Lyssavirus antigen was detected by standard fluorescent antibody test (FAT) using commercially available polyclonal fluorescein isothiocyanate (FITC)-labelled anti-rabies conjugates (Behring, Marburg; SIFIN, Berlin, Germany) following standard protocols [31]. Additional tests included virus isolation in cell culture, reverse-transcription quantitative real-time polymerase chain reaction (RT-qPCR) and sequencing following RT-PCR was performed to confirm positive FAT results.
For virus isolation, FAT positive or inconclusive bat brain samples were homogenized in a volume of 1000 µl sterile minimum essential medium (MEM-10, with 2% Streptomycin). The resulting brain suspensions (500 µl) were subjected to the RTCIT [32], using a mouse neuroblastoma cell line (MNA 42/13, No. 411, cell culture collection for veterinary medicine, FLI). Infected cells were incubated for three days at 37°C and 5% CO2 and then tested using FAT. A result was confirmed negative after the third consecutive cell passage.
RNA was extracted from 200 µl brain suspension or RTCIT supernatant using TRIzol Reagent (Invitrogen, Darmstadt, Germany)/peqGOLD TriFast (peqlab Biotechnologie GmbH, Erlangen, Germany) method. The RNA pellet was re-suspended in a volume of 20 µl bidistilled water. Samples were analysed for the presence of viral RNA using quantitative real-time PCR (RT-qPCR) specific for EBLV-1/-2 as described [25]. In cases of inconclusive FAT results a conventional panlyssavirus RT-PCR was additionally performed [33].
All EBLV-isolates were further characterized by sequence analysis [34]. RNA was subjected to one-step RT-PCR using primers JW12 and JW6 E [33] followed by sequencing. Briefly, after amplification, PCR-products were run in a 1% agarose gel stained with ethidium bromide, excised and purified essentially as for the molecular bat species identification. Sequences were manually checked for quality, trimmed to the first 400 bp using SeqMan (Lasergene, DNASTAR, Madison, WI, USA)) and submitted to NCBI GenBank (Table S1). Sequence alignment and subsequent phylogenetic analysis was performed using MEGA 5. Further representatives of EBLV-1 and 2 were derived from GenBank for comparison (Table S2).
From 1998 to June 2013 a total of 5478 bats from all German federal states (N = 16, Figure 1b) were investigated. The annual number of submissions to FLI of obtained specimens varied between 30 and 1200 individuals. The bats encompassed specimens from the entire study period and before, with the oldest sample originating from 1981.
Among all samples, 21 out of the 23 indigenous bat species in Germany were included (Table 1), although the proportion of bat species differed per federal state. The majority of bat samples originated from Lower Saxony (N = 1252), followed by Baden-Wuerttemberg (N = 736) and Saxony-Anhalt (N = 692). In contrast, in three and two of the remaining federal states the sample size was less than 90 and 15, respectively.
With the exception of a single carcass of the Lesser horseshoe bat (Rhinolophus hipposideros), all other species investigated belonged to the family Vespertilionidae. Among those, the most frequently tested bat species were the Common pipistrelle and Noctule bat followed by Serotine bat and Brown long-eared bat (Table 1). A total of 330 bats could not be identified to species level using external morphological criteria. Cyt b sequences were obtained from 119 bats, representing 15 different species (Table 1). The sequence similarity ranged between 92% and 100% when compared to publicly available sequences. Wing membrane samples from the remaining 211 individuals from natural scientific collections were not available.
Most positive specimens were found in bats from Lower Saxony (N = 27), Saxony-Anhalt (N = 10) and Berlin (N = 5) (Figure 1d). Bat rabies was detected in animals from additional 10 German federal states although only sporadically (1–3 cases). No lyssavirus infection was found in bats originating from Rhineland-Palatinate (N = 108), Baden-Wuerttemberg (N = 736) and Bavaria (N = 252) (Figure 1b–d).
Except for a single Serotine bat for which sufficient brain material was not available, lyssaviruses were successfully isolated and sequenced from 54 and 55 bats, respectively, which had been tested FAT-positive (Table 1). The presence of EBLVs was confirmed in five different bat species (E. serotinus, P. pipistrellus, P. nathusii, Pl.auritus and M. daubentonii). The majority of viruses were identified as EBLV-1, predominately isolated from E. serotinus (N = 48). Single lyssavirus infections in other species were also characterized as EBLV-1 (Table 1). The phylogenetic analysis of the N gene derived sequences identified the two lineages of EBLV-1, i.e. five out of the 52 available sequences were characterized as EBLV-1b found in Serotine bats originating from Saarland (N = 1), Saxony-Anhalt (N = 3) and Saxony (N = 1) (Figure 2a). Some clustering was observed for EBLV-1a isolates from the same or from neighbouring federal states, with occasional exceptions. The nucleotide sequence divergence within the EBLV-1a group was <1%.
Of 160 Daubenton's bats tested, three (1.88%) individuals were rabies positive, and EBLV-2 was isolated in each case (Table 1, Figure 2b). Those infected bats were submitted from Saxony-Anhalt [35], Thuringia and Hesse. Overall, in 47 smears from different bat species investigated using the FAT small fluorescing structures indicative for lyssavirus antigen was found, but the infection could not be confirmed by other methods, e.g. RTCIT, EBLV-1/-2 specific RT-qPCR and conventional RT-PCR. Furthermore, 13.1% (N = 718) of all submitted bats could not be investigated because the carcasses were mummified or organs had autolysed (Table 1).
Of all animals tested by FAT and with a reference to a date (month, N = 3714) the peak of bat finds were in July, August and September, with a second peak in February and March (Figure 3). Of those, the percentage of bats tested EBLV-positive was highest in July (N = 11, 1.95%) and August (N = 12, 1.96%). Altogether, 50 Serotine bats tested rabies positive by FAT, of which 18 were males and 11 females. Four of the positive cases were juvenile animals whereas the remaining animals were sub-adults or adults.
In Germany, routine bat rabies surveillance has been reliant on limited and opportunistic sampling as a result of international, national and federal state specific legislation that often restricts the handling, submission and even testing of bats [10]. Because the current knowledge about distribution, abundance and epidemiology of bat lyssaviruses is rather fragmentary, we initiated long-term enhanced passive bat rabies surveillance in Germany. When this study was started in 1998 it was a prerequisite to collect samples through extensive lobbying, education and awareness training of local bat biologists at a federal state level to encourage submissions of bat carcasses. Only with the 2006 Agreement on the Conservation of Populations of Bats in Europe by EUROBATS [24], which for the first time established a basis for legitimized bat rabies surveillance in Europe, the collection, submission and testing of indigenous dead bats was enabled. Within 15 years of this retrospective study (1998-June 2013) with more than 5000 bats, a six fold higher number of indigenous bats was tested for lyssavirus infections, compared to the number of bats examined during 50 years of routine surveillance in Germany [10]. We could therefore demonstrate that enhanced passive surveillance can generate an increase in submissions. Comparable studies were also conducted in The Netherlands (1984–2003, N = 3873), the UK (1987–2004, N = 4883), France (1989–2004, N = 934), and Switzerland (1976–2009, N = 837) [5], although we were able to sample more bats in a shorter period of time. Together with routine surveillance (data not shown) the passive bat rabies surveillance in Germany appears to be more intense than in most other European countries [5].
In this enhanced passive surveillance bats were submitted which had rarely been included in routine surveillance because they have a low likelihood of human contact (e.g. bats from caves, forests etc.) or bats from the countryside. Samples from e.g. private bat collections were included that had been stored in freezers for up to 25 years. Despite this long period of storage it was still possible to isolate lyssavirus from those brain tissues.
In this retrospective study, 56 additional bat rabies cases were detected that otherwise would have been missed. Together with the number of positive cases from routine surveillance (1977–2012, N = 243) (www.who-rabies-bulletin.org) Germany is one of the countries in Europe with the highest number of reported bat rabies cases. Evidently, the high level of surveillance contributes to this fact. However, the influence of other factors such as abundance of reservoir species and virus prevalence needs to be studied further.
To date three different bat lyssaviruses have been reported from Germany. While the presence of EBLV-1 has been known for a considerable period of time [10], EBLV-2 [33] and BBLV [13] were first isolated in 2007 and 2010, respectively, during routine bat rabies surveillance. In this retrospective study we confirm both the circulation of EBLV-1 and EBLV-2 in Germany.
However, samples included in this study comprise almost all bat species indigenous in Germany. Depending on the geographical distribution, population density and their habitat use the number of submissions varied considerably per bat species. Similar to the UK, the Netherlands and Switzerland the Common pipistrelle was the most frequently submitted bat species [26], [27], [36]. In fact, this synanthropic species is also one of the most abundant bat species in Europe. Although rabies in this species had been reported before [37], [38], in this study we confirmed an EBLV-1 infection for the first time by RTCIT, RT-qPCR and sequencing. EBLV-1 infections in species other than the Serotine bat (E. serotinus) was also found in a single Nathusius' pipistrelle bat and a Brown long-eared bat. All those viruses were identified as EBLV-1 and showed geographical clustering thus indicating that they resemble spill-over infections from infected Serotine bats [10].
In contrast to North America, where RABV is found in most bat species with distinct lineages [39], this pattern was not found for EBLV-1 in European bats.
A total of 49 Serotine bats tested EBLV-1 positive confirming that this bat species is the main reservoir for EBLV-1. Surprisingly, despite testing of individuals of this bat species originating from various regions in Germany, the majority of EBLV-1 cases was found in Serotine bats from the northwest of Germany, supporting previous studies [10]. Although the Serotine bat is abundant all over Germany, the density is higher in the northern lowlands of Germany [40], suggesting that the intraspecies transmission rate is higher so that more cases are detected, both in routine as well as in enhanced surveillance. While the situation appears similar in the Netherlands [26], the distribution of positive EBLV-1 cases in France differs insofar as those infections were detected in many parts of the country irrespective of the altitude [28].
The majority of lyssaviruses were characterized as EBLV-1a. Similar to previous studies from the Netherlands and Germany [10], [26], [41], EBLV-1a sequences showed a very high level of identity. However, genetic clusters appear to be linked to defined geographic regions (Figure 2a). In the past, EBLV-1b had been sporadically detected in the German federal state of Saarland close to the French border [10], [41]. Surprisingly, we could confirm the presence of the EBLV-1b subtype also in central and eastern parts of Germany. Genetically, those isolates are more closely related to an EBLV-1b isolate from Poland than with the other Saarland isolate (Figure 2a). Since Serotine bats generally do not migrate, the sporadic occurrence of EBLV-1b variants in Germany and Poland remains puzzling. However, our results could reflect a recent eastward spread of EBLV-1b although this needs further investigation.
Whilst during routine surveillance in Germany only a single Daubenton's bat was found to be infected with EBLV-2 [33], we report three additional cases. Because Daubenton's bats are associated with forest habitat, detection of grounded bats by the public is limited. This is reflected by the low number of Daubenton's bats tested for lyssavirus infections in other European studies [5]. In our study a total of 160 Daubenton's bats were tested for bat lyssaviruses resulting in an estimated prevalence of 1.88% for EBLV-2. This is comparable to estimates for Switzerland (4.6%) and the UK (3.6%). The pond bat was associated with EBLV-2 infection in the Netherlands [26]. We were only able to test seven individuals of this species, and thus cannot properly assess whether this species serves as a true reservoir host or represent a spill-over infection from Daubenton's bat.
Generally, irrespective of the low number of tested bats, the prevalence of EBLV-2 in Daubenton's bats seems to be lower than EBLV-1 in Serotine bats. In our study we found EBLV-1 in 13.28% of all tested Serotine bats, whilst in Spain (E. isabellinus) and in the Netherlands this proportion of positives was 20% and 21%, respectively [26], [42].
Overall, 47 brain smears of 11 bat species, including the reservoir bat species E. serotinus, M. daubentonii and M. nattereri, showed a particulate staining pattern morphologically similar to anti-rabies staining in FAT. This could be regarded as typical but lyssavirus infection could not be confirmed with further tests e.g. RT-qPCR, RT-PCR or RTCIT. False positive results have been shown to occur in diagnosis of classical rabies, but at a very low percentage [43]. Degradation of samples and microbial contamination may lead to certain cross-reactions with the anti-rabies conjugates. Furthermore, cross-reactions with other viral encephalites, such as West Nile and Powassan flavivirus infection cannot be excluded [44]. Besides in reservoir species, a large proportion of these unspecific results were found in the Noctule bat. Previously, bat rabies cases had been reported sporadically in this species [45], although cases were not confirmed by virus isolation and/or sequence analysis. Further investigations are needed to establish the cause of this observation.
While in this retrospective study none of the 159 Natterer's bats tested positive for BBLV, during an enhanced passive bat rabies study performed in the German federal state Bavaria BBLV was isolated from a single Natterer's bat [15]. Although with more than 20000 captures per year this is the most handled bat of all reservoir bats species (M. nattereri, E. serotinus, M. daubentonii, M. dasycneme) known to exist in Germany [D. Brockmann, Bat Marking Centre Dresden, Germany, pers. communication], this bat species is clearly underrepresented in passive surveillance due to its sylvatic mode of life making it difficult to find large numbers of dead bats of this species.
As stated before, EBLV-1 infections were not only detected in E. serotinus but also in three other indigenous bat species. In contrast to this study, during routine surveillance only about half of FAT positive bats were determined to species level where the majority of cases in Serotine bats were identified [10]. Thus, it cannot be excluded that more bat species are affected by lyssavirus infections. This demonstrates the importance of species identification for epidemiological evaluation as previously shown for the UK [46]. In cases where morphological species identification was not possible due to either the quality of the specimens (e.g. damaged, degraded) or the absence of morphological criteria, (i.e. cryptic bat species), samples were genetically characterized. By this we were able to characterize more than 100 bat specimens which helped to complete the dataset. Given the increasing diversity of lyssaviruses and reservoir bat species, lyssavirus positive specimens, i.e. both bat and virus need to be confirmed by molecular techniques. For example, the bat species originally described as associated with KHUV and ARAV may be incorrect [4]. Similarly, SHBV isolated from Hipposideros vittatus in Kenya, was initially described as Hipposideros commersonii [47]. Furthermore, genetic information on hosts may allow for a comparison with the viral evolution as shown for North America [39] and Eastern Europe [48].
Based on the experience gained in this project, we propose that enhanced passive surveillance for bat rabies should be continued to complement routine diagnosis. Thus, it is a prerequisite to collect dead bats as “fresh” as possible and freeze them as soon as possible. To this end, a close cooperation with all stakeholders involved in bat handling, monitoring and research is essential. Those dead bats should eventually be transferred to a central point where rabies diagnosis can be performed. In parallel, bats involved in human contact have to be tested by the responsible regional veterinary laboratories, to allow for prompt veterinary and human public health response. All bats need to be identified to species level by morphological and/or molecular techniques. Finally, it is of eminent importance that all data are combined into a comprehensive evaluation.
Research activities, particularly surveillance efforts to gain insights into the epidemiology of bat lyssaviruses, can be regarded as a true bat conservation effort, since a greater understanding of this zoonosis can help to reduce unjustified fear and misconceptions.
With enhanced passive surveillance 56 additional bat rabies cases were detected also in federal states where rabies in bats had not been found previously. Considering the large number of animals tested the prevalence was lower than in routine surveillance and likely represents the true level of lyssavirus infections in indigenous bats in Germany. Although the vast majority of cases were found in the known reservoir species Eptesicus serotinus, spill-over cases were also observed. In conclusion, all bat species need to be sampled and identified, and, since some bat species are still underrepresented, the enhanced surveillance should be maintained.
|
10.1371/journal.pcbi.1002363 | OptCom: A Multi-Level Optimization Framework for the Metabolic Modeling and Analysis of Microbial Communities | Microorganisms rarely live isolated in their natural environments but rather function in consolidated and socializing communities. Despite the growing availability of high-throughput sequencing and metagenomic data, we still know very little about the metabolic contributions of individual microbial players within an ecological niche and the extent and directionality of interactions among them. This calls for development of efficient modeling frameworks to shed light on less understood aspects of metabolism in microbial communities. Here, we introduce OptCom, a comprehensive flux balance analysis framework for microbial communities, which relies on a multi-level and multi-objective optimization formulation to properly describe trade-offs between individual vs. community level fitness criteria. In contrast to earlier approaches that rely on a single objective function, here, we consider species-level fitness criteria for the inner problems while relying on community-level objective maximization for the outer problem. OptCom is general enough to capture any type of interactions (positive, negative or combinations thereof) and is capable of accommodating any number of microbial species (or guilds) involved. We applied OptCom to quantify the syntrophic association in a well-characterized two-species microbial system, assess the level of sub-optimal growth in phototrophic microbial mats, and elucidate the extent and direction of inter-species metabolite and electron transfer in a model microbial community. We also used OptCom to examine addition of a new member to an existing community. Our study demonstrates the importance of trade-offs between species- and community-level fitness driving forces and lays the foundation for metabolic-driven analysis of various types of interactions in multi-species microbial systems using genome-scale metabolic models.
| Microorganisms rarely live isolated in their natural environments but rather function in consolidated and socializing communities. Despite the growing availability of experimental data, we still know very little about the metabolic contributions of individual species within an ecological niche and the extent and directionality of interactions among them. This calls for development of efficient modeling frameworks to shed light on less understood aspects of metabolism in microbial communities. Here, we introduce OptCom, a comprehensive mathematical framework for metabolic modeling and analysis of microbial communities, which relies on a multi-level/objective optimization formulation to properly describe trade-offs between individual vs. community level fitness criteria. OptCom is general enough to capture any type of interactions (positive, negative or combinations thereof) and is capable of accommodating any number of microbial species involved. We first demonstrate the capability of OptCom to quantify known metabolic interactions in a well-characterized microbial community. We next apply it to more complex communities to assess the optimality levels of growth for each microorganism, elucidate the extent and direction of inter-species metabolite transfers and examine addition of a new member to an existing community. Our study lays the foundation for metabolic-driven analysis of various types of interactions in multi-species microbial systems.
| Solitary species are rarely found in natural environments as most microorganisms tend to function in concert in integrative and interactive units, (i.e., communities). Natural microbial ecosystems drive global biogeochemical cycling of energy and carbon [1] and are involved in applications ranging from production of biofuels [2], [3], biodegradation and natural attenuation of pollutants [4], [5], [6], bacterially mediated wastewater treatment [7], [8] and many other biotechnology-related processes [9], [10]. The species within these ecosystems communicate through unidirectional or bidirectional exchange of biochemical cues. The interactions among the participants in a microbial community can be such that one or more population(s) benefit from the association (e.g., through cooperation), some are negatively affected, (e.g., by competing for limiting resources), or more often than not a combination of both. These inter-species interactions and their temporal changes in response to environmental stimuli are known to significantly affect the structure and function of microbial communities and play a pivotal role in species evolution [11], [12], [13], [14], [15], [16].
Recent advances in the use of high-throughput sequencing and whole-community analysis techniques such as meta-genomics and meta-transcriptomics promise to revolutionize the availability of genomic information [16], [17], [18]. Despite the growing availability of this high-throughput data, we still know very little about the metabolic contributions of individual microbial players within an ecological niche and the extent and directionality of metabolic interactions among them. This calls for development of efficient modeling frameworks to elucidate less understood aspects of metabolism in microbial communities. Spurred by recent advances in reconstruction and analysis of metabolic networks of individual microorganisms, a number of metabolic models of simple microbial consortia have been developed. Efforts in this direction started with the development of metabolic model for a mutualistic two-species microbial community [19]. The metabolic network of each microorganism was treated as a separate compartment in analogy to eukaryotic metabolic models [20], [21]. A third compartment was also added through which the two organisms can interact by exchanging metabolites. The same approach was employed for the metabolic modeling of another syntrophic association between Clostridium butyricum and Methanosarcina mazei [22]. Lewis et al [23] have also described a workflow for large-scale metabolic modeling of interactions between various cell lines in the human brain using compartments to represent different cells. Similarly, Bordbar et al [24] developed a multi-tissue type metabolic model for analysis of whole-body systems physiology. Alternatively, others proceeded to identify and model synthetic interactions among different mutants of the same species using genome-scale metabolic models. For example, Tzamali et al [25] computationally identified potential communities of non-lethal E. coli mutants using a graph-theoretic approach and analyzed them by extending dynamic flux balance analysis model of Varma and Palsson [26]. The same researchers have recently extended their study to describe the co-growth of different E. coli mutants on various carbon sources in a batch culture [27]. Wintermute and Silver [28] identified mutualistic relationships in pairs of auxotroph E. coli mutants. Each pair was modeled using an extended form of the minimization of metabolic adjustment (MOMA) hypothesis [29]. More recently, the concept of inducing synthetic microbial ecosystems not by genetic modifications but rather with environmental perturbations such as changing the growth medium was introduced [30].
All these studies aimed primarily at modeling communities where one or both species benefit from the association while none is negatively affected. The first study to characterize a negative interaction between two microorganisms using genome-scale metabolic models was published by Zhuang et al [31] where similar to [25], [27] an extension of the dynamic flux balance analysis [32] was employed to model the competition between Rhodoferax ferrireducens and Geobacter sulfurreducens in an anoxic subsurface environment. The same procedure was also employed in a study that characterized the metabolic interactions in a co-culture of Clostridium acetobutylicum and Clostridium cellulolyticum [33]. A wide range of methods beyond flux balance analysis have been used to model microbial communities [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45]. For example, Taffs et al [46] proposed three different approaches based on elementary mode analysis to model a microbial community containing three interacting guilds. Other studies have drawn from evolutionary game theory, nonlinear dynamics and the theory of stochastic processes to model ecological systems [39], [40], [43].
Despite these efforts, all existing methods for the flux balance analysis of microbial communities are based on optimization problems with a single objective function (related to individual species), which cannot always capture the multi-level nature of decision-making in microbial communities. For example, the flux balance analysis model described in [19] is applicable only to syntrophic associations, where the growth of both species is coupled through the transfer of a key metabolite. The dynamic flux balance analysis models introduced by Zhuang et al [31] and Tzamali et al [25], [27] rely on solving separate FBA problems for each individual species within each time interval. In all cases these methods cannot trade off the optimization of fitness of individual species versus the fitness function of the entire community. Therefore, there is still a need to develop an efficient modeling procedure to address this issue and to analyze and characterize microbial communities of increasing size with any combination of positive and/or negative interactions.
Here, we introduce OptCom, a comprehensive flux balance analysis framework for microbial communities, which relies on a multi-level optimization description. In contrast to earlier approaches that rely on a single objective function, OptCom's multi-level/objective structure enables properly assessing trade-offs between individual vs. community level fitness criteria. This modeling framework is general enough to capture any type of interactions (positive, negative or combination of both) for any number of species (or guilds) involved. In addition, OptCom is able to explain in vivo observations in terms of the levels of optimality of growth for each participant of the community. We first analyze a simple and well-determined microbial community involving a syntrophic association between D. vulgaris and M. maripaludis [19] to demonstrate the ability of OptCom in recapitulating known interactions. Next, OptCom is employed to model the more complex ecological system of the phototrophic microbial mats of Octopus and Mushroom Springs of Yellowstone National Park and compare our results with those obtained using elementary mode analysis [46]. OptCom identifies the level of sub-optimal growth of one of the guilds (SYN) in this community to benefit other community members and/or the entire population. Finally, we use OptCom to elucidate the extent and direction of inter-species metabolite transfers for a model microbial community [47], identifying the proportion of metabolic resources apportioned to different community members and predicting the relative contribution of hydrogen and ethanol as electron donors in the community. Addition of a new member to this community is also examined in this study.
OptCom postulates a separate biomass maximization problem for each species as inner problems. The inner problems capture species-level fitness driving forces exemplified through the maximization of individual species' biomass production. If preferable, alternate objective function (e.g., MOMA [29]) could be utilized in the inner stage to represent the cellular fitness criteria. Inter-species interactions are modeled with appropriate constraints in the outer problem representing the exchange of metabolites among different species. The inner problems are subsequently linked with the outer stage through inter-organism flow constraints and optimality criteria so as a community-level (e.g., overall community biomass) objective function is optimized. Figure 1A schematically illustrates the proposed concept. OptCom is solved using the solution methods previously developed for bilevel programs [48], [49], [50], [51] (see Text S1 for details of the optimization formulation and solution). Note that since OptCom yields a (non-covex) bilinear optimization problem, all case studies presented in this paper were solved using the BARON solver [52], accessed through GAMS, to global optimality.
It is important to note that OptCom can be readily modified to account for the case when one or more organisms show a form of cooperative behavior that benefits the whole population, but comes at the expense of growing at rates slower than the maximum possible [15], [53]. To quantify the deviation of community members from their optimal behavior, we introduce a metric called optimality level for each species k (i.e., ck). The optimality level for each one of the microorganisms is quantified using a variation of OptCom which we refer to as descriptive. Descriptive OptCom incorporates all available experimental data for the entire community (e.g., community biomass composition) as constraints in the outer problem and all data related to individual species as constraints in the respective inner problems while allowing the biomass flux of individual species to fall below (or rise above) the maxima () of the inner problems (see Figure 1B). We note that here the optimum biomass flux for each species () is community-specific as it is computed in the context of all microorganisms striving to grow at their maximum rate (using the formulation given in Figure 1A). An optimality level of less than one for a microorganism k implies that it grows sub-optimally at a rate equal to 100ck % of the maximum () to optimize a community-level fitness criterion while matching experimental observations. Alternatively, an optimality level of one implies that microorganism k grows exactly optimally at a rate equal to whereas a value greater than one indicates that it achieves a higher biomass production level than the community-specific maximum (i.e., super-optimality) by depleting resources from one or more other community members. It is worth noting that super-optimality is achievable for a species only at the expense of sub-optimal behavior of at least one other member in the community. The identified combination of sub- and/or super-optimal behaviors of individual species is driven by the maximization of a community-level criterion (e.g., maximize the total community biomass).
OptCom can capture various types of interactions among members of a microbial community. Symbiotic interactions between two (or more) populations can be such that one or more species benefit from the association (i.e., positive interaction), are negatively affected (i.e., negative interactions), or combination of both. Mutualism, synergism and commensalism are examples of positive interactions, whereas parasitism and competition are examples of negative interactions. A pictorial representation of how these interactions can be captured within OptCom by appropriately restricting inter-organism metabolic flows is provided in Figure 2 (see Text S1 for implementation details).
We first explore the capability of OptCom to model and analyze a relatively simple and well-characterized syntrophic association between two microorganisms, namely Desulfovibrio vulgaris Hildenborough and Methanococcus maripaludis. Syntrophy is a mutualistic relationship between two microorganisms, which together degrade an otherwise indigestible organic substrate. A prominent example of syntrophic interactions is interspecies hydrogen transfer, where the hydrogen produced by one of the species has to be consumed by the other to stimulate the growth of both microorganisms [54], [55], [56], [57]. In these communities degradation of a substrate by fermenting bacteria is energetically unfavorable as it carries out a reaction, which is endergonic under standard conditions. However, if this fermenting bacteria is coupled with a hydrogen scavenging partner such as methanogenic bacteria, the organic compound degrading reaction can proceed [58]. Methanogens use hydrogen and energy gained from the first reaction and reduce CO2 to methane [56], [58].
Here we focus on such a syntrophic association between Desulfovibrio vulgaris Hildenborough and Methano- coccus maripaludis S2, for which genomes-scale metabolic models as well as experimental growth data for the co-culture are available [19]. With lactate as the sole carbon source and in the absence of a suitable electron acceptor for the sulfate reducer, M. maripaludis provides favorable thermodynamic conditions for the growth of D. vulgaris by consuming hydrogen and maintaining its partial pressure low. Stoylar et al [19] modeled this microbial community as a multi-compartment metabolic network and employed FBA to identify community-level fluxes by maximizing the weighted sum of the biomass fluxes of two microorganisms.
Here we examine the applicability of OptCom for modeling a more complex microbial community containing three interacting guilds, the phototrophic microbial mats of Octopus and Mushroom Springs of Yellowstone National Park (Wyoming, USA) [60]. The inhabitants of this community include unicellular cyanobacteria related to Synechococcus spp (SYN), filamentous anoxygenic phototrophs (FAP) related to Chloroflexus and Roseiflexus spp and sulfate-reducing bacteria (SRB) as well as other prokaryotes supported by the products of the photosynthetic bacteria [46], [60]. Diel (day-night) variations in metabolic activities of members of this community were observed before [61], [62], [63]. During the day when the mat is oxygenated cyanobacteria appear to be the main carbon fixer, consuming CO2 and producing storage products such as polyglucose as well as O2 as a by-product of photosynthesis. High levels of O2 relative to CO2 stimulate the production of glycolate. Glycolate is then used as a carbon and energy source by other community members such as photoheterotrophic FAP. At night, the mat becomes anoxic and cyanobacteria start to ferment the stored polyglucose into small organic acids such as acetate, propionate and H2. FAP can incorporate fermentation products photoheterotrophically while SRB oxidizes the fermentation products under anaerobic condition and produces sulfide [60], [64], [65], [66]. A schematic diagram representing the interactions in this community is given in [46].
This microbial community has been previously modeled and analyzed by Taffs et al [46] using a representative microorganism for each guild: Oxygenic photoautotrophs related to Synechococcus spp were chosen to represent the mat's primary carbon and nitrogen fixers. FAP from the family Chloroflexaceae, were selected to represent metabolically versatile photoheterotrophs that capture light energy as phosphodiester bonds but require external reducing equivalents and carbon sources other than CO2. A SRB guild representative whose metabolic behavior was based on several well-studied sulfate-reducing bacteria was also included in the community model description [46]. The metabolic networks representing central carbon and energy metabolism for each guild were then constructed and three different modeling approaches based on the elementary mode analysis were employed to elucidate sustainable physiological properties of this community [46]. Here, we focus only on daylight metabolism (for which more experimental data is available) to assess the efficacy of OptCom in describing carbon and energy flows and the biomass ratio between guilds.
In a recent study, Miller et al [47] established a model microbial community to better understand the trophic interactions in sub-surface anaerobic environments. This community was composed of three species including Clostridium cellulolyticum, Desulfovibrio vulgaris Hildenborough, and Geobacter sulfurreducens. Cellobiose was provided as the sole carbon and energy source for C. cellulolyticum whereas the growth of D. vulgaris and G. sulfurreducens were dependent on the fermentation by-products produced by C. cellulolyticum. D. vulgaris and G. sulfurreducens were supplemented with sulfate and fumarate, respectively, as electron-acceptors to avoid electron acceptor competition [47]. The experimental measurements for the biomass composition of the community showed that, as expected, C. cellulolyticum was the dominant member in the co-culture and confirmed the presence of D. vulgaris and G. sulfurreducens. It was, however, not possible to quantify experimentally the flow of shared metabolites among the community members as their concentrations were below the detection limits. Therefore, the authors proposed an approximate model of the carbon and electron flow based on some measurements of the three species community at steady-state, pure culture chemostat experiments and data from the literature [47].
Here, we model this microbial community by making use of the corresponding bacterial metabolic models and employ OptCom to elucidate the inter-species interactions. The metabolic models of C. cellulolyticum (i.e., iFS431) and G. sulfurreducens were reconstructed by Salimi et al [33] and Mahadevan et al [69], respectively. A basic metabolic model of D. vulgaris containing 86 reactions was introduced by Stolyar et al [19], however, this model had only a compact representation of the central metabolism. For example, the model was not able to support growth in the presence of acetate or ethanol as the sole carbon source. Therefore, we expanded this model by adding new reactions from a first draft reconstructed model in the Model Seed [70] and the KEGG database [71] using the GrowMatch procedure [50] (see Text S1 for details). The updated model of D. vulgaris consists of 145 reactions and is capable of supporting growth on acetate as well as ethanol. This model is available in the supplementary material (Table S1).
Here, we introduced OptCom, a comprehensive computational framework for the flux balance analysis of microbial communities using genome-scale metabolic models. We demonstrated that OptCom can be used for assessing the optimality level of growth for different members in a microbial community (i.e., Descriptive mode) and subsequently making predictions regarding metabolic trafficking (i.e., Predictive mode) given the identified optimality levels. Unlike earlier FBA-based modeling approaches that rely on a single objective function to describe the entire community [19], [30] or separate FBA problems for each microorganism [25], [27], [31], [33], OptCom integrates both species- and community-level fitness criteria into a multi-level/objective framework. This multi-level description allows for properly quantifying the trade-offs between selfish and altruistic driving forces in a microbial ecosystem. Species and community level fitness functions are quantified by maximizing the biomass formation for the respective entity. We note, however, that the physiology of microbial communities is highly context and environment dependent and a universal community-specific fitness criterion does not exist. Studies similar to those conducted for mono-cultures that examine and compare various presumed hypotheses on cellular objective function [82], [83], [84], [85], [86], [87] or algorithms that identify/test a relevant objective function using experimental flux data [88], [89] are needed in the context of multi-species systems.
An important goal of studying natural and synthetic microbial communities is their targeted manipulation towards important biotechnological goals (e.g., cellulose degradation, ethanol production, etc.). This has motivated researchers to construct simple synthetic microbial ecosystems, which are amenable to genetic and engineering interventions, for biotechnology- and bioenergy-related applications. As an example, Bizukojc et al [22], have proposed a co-culture composed of Clostridium butyricum and Methanosarcina mazei to relieve the inhibition of fermentation products and increase production of 1,3-propanediol (PDO) by Clostridium butyricum. Mixed cultures have been also established for overproduction of polyhydroxyalkanoates (PHA) [90], [91] and ethanol [92], [93], [94], [95], [96]. For example, Clostridium thermocellum, which is used for ethanol production, has been found to be capable of utilizing hexoses, but not pentose sugars generated from breakdown of cellulose and hemicellulose [96]. Therefore, cultivation of C. thermocellum with other thermophilic anaerobic bacteria capable of utilizing hexoses as well as pentose to produce ethanol (e.g., Clostridium thermosaccharolyticum and Thermoanaerobacter ethanolicus) has been previously examined in vivo [92], [93], [94], [95], [96]. The multi-objective and multi-level structure of the OptCom procedure, introduced here, can help assess the metabolic capabilities of such synthetic ecosystems. Taking a step further, OptCom can be readily modified to identify the minimal number of direct interventions (i.e., knock-up/down/outs) to the community leading to the elevated production of a desired compound (e.g., by considering the overproduction of desired compound as the outer problem objective function), thus extending the applicability of strain design tools such as OptKnock [48], OptStrain [49], OptReg [97] and OptForce [98]. It is worth noting that a key bottleneck to the modeling and analysis of microbial communities is the paucity of genome-scale models for all participants in a complex microbial community. Overcoming this barrier would require the development of high-throughput metabolic reconstruction tools such as the Model Seed [70] resource. Given that microbial communities change with time (e.g., day/night cycle) and also location (e.g., nutrient gradients), approaches that would be able to capture temporal and spatial varying inter-species metabolic interactions are needed. For example, the separate FBA problems for each individual species in the dynamic flux balance analysis methods of Zhuang et al [31] and Tzamali et al [25], [27] can be integrated with OptCom to account for inter-species interactions and community-level fitness driving forces within each time interval.
|
10.1371/journal.pcbi.1002257 | Metabolic Regulation in Progression to Autoimmune Diabetes | Recent evidence from serum metabolomics indicates that specific metabolic disturbances precede β-cell autoimmunity in humans and can be used to identify those children who subsequently progress to type 1 diabetes. The mechanisms behind these disturbances are unknown. Here we show the specificity of the pre-autoimmune metabolic changes, as indicated by their conservation in a murine model of type 1 diabetes. We performed a study in non-obese prediabetic (NOD) mice which recapitulated the design of the human study and derived the metabolic states from longitudinal lipidomics data. We show that female NOD mice who later progress to autoimmune diabetes exhibit the same lipidomic pattern as prediabetic children. These metabolic changes are accompanied by enhanced glucose-stimulated insulin secretion, normoglycemia, upregulation of insulinotropic amino acids in islets, elevated plasma leptin and adiponectin, and diminished gut microbial diversity of the Clostridium leptum group. Together, the findings indicate that autoimmune diabetes is preceded by a state of increased metabolic demands on the islets resulting in elevated insulin secretion and suggest alternative metabolic related pathways as therapeutic targets to prevent diabetes.
| We have recently found that distinct metabolic disturbances precede β-cell autoimmunity in children who later progress to type 1 diabetes (T1D). Here we performed a murine study using non-obese diabetic (NOD) mice that recapitulated the protocol used in human, followed up by independent studies where NOD mice were studied in relation to risk of diabetes progression. We found that young female NOD mice who later progress to autoimmune diabetes exhibit the same lipidomic pattern as prediabetic children. These metabolic changes are accompanied by enhanced glucose-stimulated insulin secretion, upregulation of insulinotropic amino acids in islets, elevated plasma leptin and adiponectin, and diminished gut microbial diversity of the Clostridium leptum subgroup. The metabolic phenotypes observed in our study could be relevant as end points for studies investigating T1D pathogenesis and/or responses to interventions. By proceeding from a clinical study via metabolomics and modeling to an experimental model using a similar study design, then evolving further to tissue-specific studies, we hereby also present a conceptually novel approach to reversed translation that may be useful in future therapeutic studies in the context of prevention and treatment of T1D as well as of other diseases characterized by long prodromal periods.
| Type 1 diabetes (T1D) is an autoimmune disease that results from the selective destruction of insulin-producing β-cells in pancreatic islets. The diagnosis of T1D is commonly preceded by a long prodromal period which includes seroconversion to islet autoantibody positivity [1] and subtle metabolic disturbances [2]. The incidence of T1D among children and adolescents has increased markedly in the Western countries during the recent decades [3] and is presently increasing at a faster rate than ever before [4], [5]. This suggests an important role of environment and gene-environment interactions in T1D pathogenesis.
Metabolome is sensitive to both genetic and early environmental factors influencing later susceptibility to chronic diseases [6]. Recent evidence from serum metabolomics suggests that metabolic disturbances precede early β-cell autoimmunity markers in children who subsequently progress to T1D [2]. However, the environmental causes and tissue-specific mechanisms leading to these disturbances are unknown. Given its relatively low disease incidence in the general population and even among subjects at genetic risk [1], studies on early phenomena of T1D pathogenesis in humans are a huge undertaking as they require long and frequent follow-up of large numbers of subjects [2], [7], [8] to be able to go “back to the origins” of the disease once a sufficient number of subjects in the follow-up have progressed to overt disease. In order to effectively prevent this disease it is thus fundamental to identify suitable experimental models that recapitulate findings from such large-scale clinical studies while being amenable to mechanistic studies at the systems level.
The non-obese diabetic (NOD) mouse is a well characterized model of autoimmune disease [9] which has been widely used in studies of T1D. It is clear that the NOD experimental model does not completely mimic the immune system and T1D pathogenesis in man [10]. Only a fraction of NOD mice progress to disease, with the incidence of spontaneous diabetes being 60%–80% in females and 20%–30% in males [9]. There is thus a stochastic component to diabetes pathogenesis in NOD mice, believed to be due to random generation of islet-specific T cells [11]. The disease incidence does seem to depend on the environment and there is evidence indicating that it is the highest in a relatively germ-free environment [12] and that gut microbiota may affect disease incidence via the modulation of the host innate immune system [13].
Herein we performed a murine study in NOD mice that recapitulated the protocol used in human studies [2] and applied a reverse-translational approach (Figure 1) to (1) map the lipidomic profiles of T1D progressors in human studies to lipidomic profiles in NOD mice and derive a surrogate marker to stratify mice according to risk of developing autoimmune diabetes, then (2) perform multiple follow-up studies in NOD mice where metabolic phenotypes, tissue-specific metabolome and transcriptome as well as gut microbiota are characterized in the context of early disease pathogenesis.
Our first objective was to validate whether the NOD mouse was a good model of T1D able to recapitulate the lipidomic-based metabolic phenotypes observed in the longitudinal study of children who later progressed to T1D (Type 1 Diabetes Prediction and Prevention project; DIPP) [2], [8]. Hence we performed a murine study using NOD mice and using a similar protocol as applied in human studies (Study 1). A total of 70 NOD/Bom mice (26 female) were monitored weekly with serum collection from age 3 weeks until either (a) the development of diabetes (progressor group), or (b) followed until 36 weeks of age in females and 40 weeks in males in the absence of a diabetic phenotype (non-progressor group) (Figure 2A). Similarly as in the DIPP study [2], we were primarily interested in early pre-diabetic differences of lipidomic profiles, in mice of the same colony, between diabetes progressors and non-progressors.
Lipidomic analysis using established methodology based on Ultra Performance Liquid Chromatography (UPLC) coupled to mass spectrometry (MS) [14] was performed on a complete sample series from 26 female mice (12 progressors, 14 non-progressors) and 13 male (7 progressors, 6 non-progressors) mice, comprising a total of 1172 samples or 30 samples/mouse on an average (733 samples from female and 439 from male mice), with 154 lipids measured in each sample. When comparing the lipid concentrations of diabetes progressors and non-progressors, the first weeks of life (3–10 weeks) were characterized by an overall lipid-lowering trend among the female progressors, while the period close to the disease onset (15 week and older) was characterized by elevated triglycerides and phospholipids (Figure 2B). No such changes were observed in male mice (Figure S1). The NOD female progressors had similar levels of glycemia (Figure 2C) than the non-progressors but to our surprise the progressors exhibited higher fasting as well as glucose-stimulated plasma insulin levels (2-way ANOVA p = 0.025 for diabetes progression) (Figure 2D) despite the fact that no body weight differences were evident between progressors and non-progressors at 10 weeks of age (Figure 2E). To account for multiple comparisons, false discovery rates among significantly differing lipids were estimated using q-values [15], [16].
Together, these results imply that the mice who later progress to diabetes are characterized by enhanced glucose-stimulated insulin secretion (GSIS) at an early age or that they are inappropriately insulin resistant for their degree of body weight. In fact this increased GSIS associated to early evolutive stages towards T1D is consistent with our earlier findings indicating that the children who later progress to diabetes are characterized by low serum ketoleucine and elevated levels of the more insulinotropic aminoacid leucine prior to seroconversion to insulin autoantibody (IAA) positivity [2], [17].
In order to systematically investigate similarities between early metabolic phenotypes of autoimmune diabetes progressors in mice and men, we proceeded with comparative analysis of longitudinal lipidomic profiles from female NOD mice and DIPP study children [2]. One inherent challenge in the studies of early disease pathophysiology is variable disease penetration. The metabolic profiles may individually change at different paces, and it is not obvious how they should be compared between individuals or species in the context of the disease process. We recently introduced a concept that the maturation of metabolic profiles with age, such as during normal development or early disease pathogenesis, can be described in terms of metabolic states derived using the Hidden Markov Model (HMM) methodology [18]. Instead of observing progression of average lipidomic profiles (Figure 2B), our approach allows for each individual's lipidomic profiles to mature at their own pace. Such individual profiles are captured into a set of progressive HMM states (described by mean lipid profiles) using an underlying statistical model.
Firstly, we proceeded with the analysis of previously reported longitudinal lipidomic profile data from the DIPP study children [2]. The nested case-control study included 56 T1D progressors and 73 matched non-progressors, comprising a total of 1196 samples or 9.3 samples per child on average between birth and the diagnosis of T1D (in progressors). We applied the HMM methodology to study the longitudinal lipidomic profiles in DIPP children and identified a three-state HMM, developed separately for T1D progressors and non-progressors, to describe the progression of metabolic states at early ages (up to 3 years) (Figure 3A,B). As expected based on the earlier report [2], the first state corresponding to the first year of life was characterized by low triglycerides and specific phospholipids in T1D progressors (Figure 3C). In both progressors and non-progressors the states followed a similar time course (Figure 3B), but the first and third states, corresponding to the first and third years of life, respectively, were qualitatively different between the two groups. On the other hand, there were no such clear differences in the second state, corresponding to the second year of life in average. Such multi-stage progression of lipidomic profiles during the first 3 years of life was not detected when examining them cross-sectionally in different age cohorts.
We then applied the HMM methodology to study the longitudinal lipidomic profiles of female NOD mice and identified a three-state HMM to describe the progression of metabolic states at early ages (3–10 weeks) (Figure 4A,B). The first two states, corresponding to mean ages of approximately 4 weeks and 6 weeks, respectively, were similar to the first state in DIPP children (Figure 3C) and characterized by decreased phospholipids and triglycerides among the progressors (Figure 4B). In the third state, corresponding to approximately 7 weeks of age when a large fraction of the NOD mice already seroconvert to islet autoantibodies [9], the differences observed in the first two states have disappeared. Instead, the levels of proinflammatory lysophosphatidylcholines (lysoPCs) were increased in diabetes progressors (1%–10% confidence interval for progressors having higher concentrations, see Figure 4B).
The similarity of state progression in children (Figure 3) and female NOD mice (Figure 4B) presenting with diabetes suggests that the early disease stages as reflected in the lipidomes share similar metabolic perturbations. However, it is always a challenge to compare species exhibiting differences in systemic lipid metabolism as well as diet-related effects on the lipidomic profiles. Consequently the mapping of molecular lipids between mouse and man may not be trivial and our results should be considered qualitative.
In order to compare progression of mouse and human lipidomic profiles we applied a mapping algorithm [19] that captures their dependencies across the two species. By using this strategy it is possible to compare lipidomic profiles across species, and we therefore sought for the disease effect by a two-way analysis on progressors/non-progressors vs. men/mice. By this approach, we uncovered associations of functionally and structurally related lipids between the species (Figure 4C) and confirmed strong association of diminished phospholipids with the development of the disease at an early age (HMM state 1). We can thus conclude that the lipid changes seen in children prior to the first seroconversion to autoantibodies are also characteristic of the early changes in female NOD mice progressors.
Seroconversion to islet autoantibody positivity is associated with transiently elevated lysoPC serum levels in children who subsequently progress to T1D [2]. Here we measured the IAA levels in NOD mice at 8 weeks of age and similarly confirmed that the IAA-negative (IAA−) progressor female NOD mice had elevated lysoPC as compared to IAA- non-progressors at a marginal significance level (p = 0.091, see Figure 2F). Intriguingly, IAA positivity had the opposite association with diabetes progression since the IAA-positive (IAA+) mice with high lysoPC were protected from diabetes (Figure 2F). It can be speculated that due to their opposite association with disease progression IAA measurement in combination with lysoPC may help stratify the NOD mice according to their risk of developing diabetes. We derived a surrogate marker by combining autoantibody positivity and lysoPC concentration, which reasonably well discriminated between progressors and non-progressors (χ2 = 5.75, Pχ2 = 0.0044; Figure S2), with the NOD mice in the assigned “High-risk” group being at 4.3-fold higher risk (95% lower tolerance bound = 2.6, as calculated from 1000-fold resampling) of developing autoimmune diabetes as compared to the mice in the “Low-risk” group.
In an independent experiment normoglycemic female NOD mice from the same colony as in the first experiment were sacrificed at 8 (n = 57) or 19 (n = 14) weeks of age and blood, liver and pancreas samples were collected (Study 2). We selected sixteen 8-week-old mice (seven were IAA+) and thirteen 19-week-old mice (six were IAA+) for UPLC-MS based serum lipidomics analysis for subsequent risk stratification using the algorithm described above. Mice at high risk of developing diabetes showed a tendency towards more severe insulitis (Figure 5A), therefore providing an independent validation of the surrogate marker. In parallel liver and islet transcriptomics was performed in 19-week-old mice.
When comparing high- and low-risk mice, independent of IAA level, the pathway analysis of islet gene expression data using Gene Set Enrichment Analysis (GSEA) [20] expectedly revealed upregulation of several apoptotic and immunoregulatory pathways in the high-risk group (Table 1 and Table S1). These pathways were associated with the autoimmune status, as they were also upregulated when comparing IAA+ and IAA− mice independent of diabetes risk. In support of our findings from pre-diabetic mice, some of the upregulated gene products of these pathways are in fact known to be implicated in progression to autoimmune diabetes, including CD3 from the CTLA4 pathway [21], pro-inflammatory chemokine CCL5 (or RANTES) from the toll like receptor signalling pathway [22], [23], [24] and the IL-7 pathway [25] (Table S2).
Several upregulated pathways in high-risk mice were not associated with the IAA titer. These pathways associated with high risk of developing diabetes were mainly metabolic pathways and included upregulated genes from TCA cycle and glycolysis/gluconeogenesis (Table 1). In order to directly measure the metabolic products of these pathways, we performed metabolomic analysis of islets using two-dimensional gas chromatography coupled to time-of-flight mass spectrometry (GC×GC-TOFMS) [26]. Metabolomics confirmed dysregulation of energy and amino acid metabolism in the islets of high-risk mice (Figure 5B), as several key metabolites of these pathways were found upregulated, including glutamic and aspartic acids, as well as at a marginal significance level all three branched chain amino acids (BCAAs). These elevated amino acids are known insulin secretagogues in β cells [27]. In agreement with this, the insulin signaling pathway was upregulated in the livers of high-risk mice (Table S3). The top ranking gene in this pathway, Glucose-6-phosphatase, catalytic, 2 (G6PC2; fold change high- vs. low-risk group +11%, P = 0.0034), controls the release of glucose from liver into the bloodstream. However, the animals included in this study, as in the earlier longitudinal study, were normoglycemic and there were no differences in body weight between the two groups. The metabolic changes in β cells and liver can thus explain the observed elevated GSIS in mice at high risk for developing autoimmune diabetes (Figures 2C–E).
We recently observed that the serum metabolome of germ-free mice is similar to pre-autoimmune metabolomes of children who later progress to T1D [28], thus implying that gut microbiota of T1D progressors may be devoid of important constituents or has an impaired function that predisposes the children to T1D. Given the observed similarities of metabolomes of diabetes progressors in mice and men (Figure 4), we hypothesized that the observed metabolic differences between the high- and low-risk mice may be reflected in differences in their gut microbial composition.
We characterized the microbial composition of caecum samples from high- and low-risk mice from Study 2 using predominant bacterial as well as five different bacterial group-specific (namely Eubacterium rectale – Blautia coccoides group, Clostridium leptum group, Bacteroides spp., bifidobacteria, and lactobacilli) denaturing gradient gel electrophoresis (DGGE) methods as previously described [29], [30]. With such an approach to profile microbiota it is possible to detect the phylotypes that constitute over 1% of the specific group in question [29], [31]. Analysis of the composite dataset, which included all the different group-specific DGGE results, showed that the total bacterial composition did not markedly differ between the groups but was slightly more coherent in low-risk mice than in high-risk mice (see the deviation bars in Figure S3). In addition, the high-risk mice had significantly diminished diversity of the Clostridium leptum group of the Firmicutes phylum (Figure 5C).
There is evidence from clinical studies that insulin resistance is a risk factor for progression to T1D [32], [33]. It is also known that the NOD genetic background may predispose the mice to insulin resistance [34]. To test for insulin resistance as a potential explanation for the observed metabolic phenotype of high-risk mice, we performed two independent studies in another NOD colony where (Study 3) 36 female NOD/MrkTac mice were tested for GSIS, glucose and insulin tolerance, and plasma leptin between 8 and 11 weeks of age; and (Study 4) 42 female NOD/MrkTac were sacrificed at 10 weeks of age and tested for insulitis, plasma leptin and adiponectin. As before, serum lipidomics and IAA assays were performed to stratify the mice into high- and low-diabetes-risk groups.
We confirmed the elevated GSIS in high risk mice (Figure 6A) but found no significant difference in glucose responses to intraperitoneal glucose or insulin between the groups (Figures 6B,C), in the Homeostatic model assessment of insulin resistance (HOMA-IR) index or GLUT4 expression in white adipose tissue and muscle (Figure 6D,F). In agreement with the results from older mice (Figure 5A) and in further support of the surrogate marker applied to stratify the mice according to disease risk, the 10-week old female NOD mice at higher risk of developing diabetes have already signs of more insulitis than their low-risk counterparts, although the average degree of insulitis is mild in both groups (p<0.05, see Figure 6G). Surprisingly, the adipose tissue derived hormones leptin (p<0.05, see Figure 6H) and adiponectin (p<0.05, see Figure 6I) were both elevated in plasma of high-risk mice despite no significant differences in weight or adiposity (p>0.05, see Figure 7A–C). However, both adiponectin and leptin correlated with the gonadal adipose tissue mass (Figure 7D,E).
Given that the metabolic profile is normalized in children following seroconversion to autoantibody positivity [2], we proposed earlier that the generation of autoantibodies may be a physiological response to early metabolic disturbances. In the present study (mice from Study 2), we investigated the pathways in the IAA+ low-risk female mice and compared them to all other groups. The IAA+ low-risk mice were characterized by several elevated signaling pathways in the islets, including the IL-4 and IL-6 pathways (Table 1). IL-4 is known to be protective from diabetes in NOD mouse [35]. Conversely IAA+ low-risk mice had reduced expression of pathways mainly related to mitochondrial function and TCA cycle, BCAA catabolism, beta oxidation and oxidative phosphorylation. It is unclear how downregulation of these pathways may protect against T1D. However downregulation of these pathways will lead to a state of reduced production of reactive oxygen species (ROS) [36] which may explain at least in part the conserved β cell functionality. This would offer a potential protective mechanism linking decreased ROS production to the prevention of β cell apoptosis in IAA+ mice which do not progress to diabetes. Our results stress the need for similar studies in terms of protection from diabetes in individuals who seroconverted but did not progress to overt disease.
This study emphasizes the translatability of our previous findings from the large-scale clinical study [2] into the tissue-specific context. Also, our study highlights that specific metabolic disturbances are identifiable early on during the evolutive stages and could potentially be linked to pathogenic mechanisms implicated in the progression to autoimmune diabetes. Although the specific causes, likely to be diverse amongst humans and between the NOD mouse and humans, of these early metabolic disturbances remain to be established, our findings pave the way for studies focused on how the metabolism and the immune system interact in early stages of the disease.
The lipidomic profiles associated with progression to T1D in children [2] were similar to early lipidomic profiles in female but not male NOD mice that later progressed to autoimmune diabetes. It is known that female NOD mice are more likely to progress to autoimmune diabetes [9] although the reasons for this are poorly understood. Notably, in humans the T1D incidence is nearly 2-fold higher in men than women [37]. In the present study, we have not pursued the reasons for the gender-specific metabolic differences in NOD mice and have instead focused on studies in female mice since they displayed the similar metabolic patters as observed in human studies.
Both in man and mouse, the metabolic states as determined by HMM followed the similar progression in disease progressors and non-progressors (Figures 3B and 4A), but were qualitatively different between these two groups (Figures 3C and 4B). However, notably no major qualitative differences were observed in the second state in the human study and in the third state of the mouse study. These two states correspond to the ages when the first diabetes-associated autoantibodies have appeared in many of the human T1D progressors [2] and NOD mice [9], respectively. In the DIPP study we have previously shown that the seroconversion to autoantibody positivity appears to normalize the metabolic profiles, suggesting that the immune system may be involved in the metabolic regulation and vice versa. In fact, the metabolic demands of T cells are extraordinary, rivaling that of cancer cells [38], [39]. For example, differentiating T cells consume 10-fold more glutamine than other cells in the body [39], and we in fact found that glutamine is diminished in children within a period of months prior to seroconversion [2]. Concentration changes of circulating metabolites as detected in our studies may thus have a direct effect on T-cell function. We therefore hypothesize that the second metabolic state in human progressors, and similarly the third state in NOD mice, reflect the period following the seroconversion when the metabolic profiles have been restored to normal levels via interaction with the immune system.
In the NOD mice, this apparent interaction between the immune system and metabolic status is underlined by the opposite association of the IAA and lysoPC (Figure 2F) at 8 weeks of age, i.e., in the age group corresponding to the third state in the HMM model (Figure 4B). Based on this observation, a surrogate marker was derived combining information on IAA positivity and lysoPC concentration (Figure S2), which was utilized to stratify mice according to risk of developing autoimmune diabetes in subsequent studies. The so-classified high-risk mice had higher degree of insulitis, a histopathological hallmark of progression to diabetes in NOD mice [9], as assessed in two independent studies (Figures 5A and 6G). While the association of the surrogate marker with established characteristics of progression to T1D validated our approach in the present study, it also suggests that prediction of autoimmune diabetes in NOD mice using combined metabolic and immune markers may be feasible. However, further prospective studies are needed in different NOD strains, similar in design as our Study 1, to determine and validate such markers. As already demonstrated in our study, the use of such markers sensitive to disease risk may facilitate investigations of early pathophysiological phenomena at a tissue-specific level prior to any symptoms of the disease.
Our results indicate that early stages of progression to T1D are characterized by acute increased response to high glucose-stimulated insulin secretion. Furthermore, this response is accompanied by increased concentration of insulinotropic amino acids and other markers of energy metabolism in the islets and more specifically of insulin signaling pathways in the liver. Together with human data [2], our study provides compelling evidence that increased GSIS is an event that heralds diabetes progression already in pre-autoimmune stages of the disease pathogenesis. In NOD mice, elevated GSIS at young age may be an initial response associated with early insulitis. Our data suggest that this response might reflect a state of insulin resistance; however our insulin tolerance tests do not support this insulin-resistant component in diabetes progressors. One potential link may be increased circulating insulin concentrations-suppressing leptin [40] and insulin-sensitizing adiponectin [41]. Adiponectin is known to be elevated in patients with T1D, but very limited data exist on its levels during the pre-diabetic period [42]. Leptin, however, is known as an important immune regulator [43]. Leptin is a negative regulator of CD4+CD25+ regulatory T cells [44] and promoter of Th1 immune responses [45], [46]. In fact administration of leptin accelerates autoimmune diabetes in female NOD mice [47]. Of interest, endoplasmic reticulum stress is known to induce leptin resistance [48]. Together, our findings from the studies of female NOD mice at high- and low-risk of T1D within the same colony suggest that elevated leptin in high-risk mice is a consequence of early metabolic stress, and that leptin may play a role in mobilization of deleterious Th1 immune responses characteristic of T1D [49]. This would offer an explanation for the epidemiological findings that obesity [50] and decreased insulin sensitivity [51] are risk factors of T1D. Given the global rise of obesity and related metabolic complications among children [52], our study thus suggests that improving insulin sensitivity while avoiding harmful immune responses in genetically susceptible individuals may be an important new strategy for early T1D prevention.
Our study also implicates that early metabolic disturbances in progression to autoimmune diabetes associate with diminished diversity of specific bacterial groups such as C. leptum group. This is in agreement with a recent pilot study in the DIPP cohort where phylum Firmicutes was found decreased in the four children who later progressed to diabetes [53]. Microbial communities are sensitive to disturbances and may subsequently not return to the their original state [54]. Interestingly, diminished diversity of the anti-inflammatory commensal bacterium Faecalibacterium prausnitzii from the C. leptum group characterizes also Crohn's disease [55]. Our study thus revealed a candidate microbial group which may be further considered in the context of diabetes prevention.
The fact that the diabetes-associated differences in microbial composition were observed among the mice of the same colony suggests that the observed diminished microbial diversity is rather a consequence than a primary cause of immunological or metabolic responses. The mechanisms by which the gut microbial community is modulated by specific metabolic and immune factors associated with progression to T1D are at present unclear and demand further investigation. However, these findings may still be important by having a role in early disease pathogenesis. In fact, recent study revealed that microbes from C. leptum group induce regulatory T cells in colonic mucosa [56]. Diminishment of C. leptum diversity along with elevated leptin may therefore be two mechanisms which promote negative regulation of CD4+CD25+ regulatory T cells, and therefore also promote the autoimmune response [57].
Our study uncovered multiple factors which may contribute in parallel to progression towards autoimmune diabetes. It is unlikely that any of them is a primary cause to initiate the disease process. Instead, as an early mathematical model of T1D describing changes in numbers of β-cells, macrophages, and Th-lymphocytes concluded, the “onset of type 1 diabetes is due to a collective, dynamical instability, rather than being caused by a single etiological factor” [58]. In this context, understanding the spatial and temporal balance of different disease-contributing factors is important [59]. The study design such as ours may help identify the early factors contributing to the disease as well as their mutual dependencies.
Finally, the metabolic phenotypes described here could be relevant as end points for studies investigating T1D pathogenesis and/or responses to interventions. By proceeding from a clinical study via metabolomics and modeling to an experimental model using a similar study design, then evolving further to tissue-specific studies, we hereby present a conceptually novel approach to reversed translation (Figure 1) that may be useful in future therapeutic studies in the context of prevention and treatment of T1D as well as of other diseases.
All experimental procedures were approved by the Committee for Laboratory Animal Welfare, University of Turku.
The mice were kept in an animal room maintained at 21±1°C with a fixed 12∶12 hr light-dark cycle. Standard rodent chow (Special Diet Services, Witham, UK) and water were available ad libitum. The colonies of NOD/Bom mice used were bred and maintained in the animal facilities of University of Turku and originated from mice purchased from Taconic Europe (Ry, Denmark). 26 female and 44 male NOD mice (Study 1) underwent weekly blood sampling by venopuncture from the tail vein starting at 3 weeks of age until the mice developed diabetes (blood glucose ≥14.0 mmol/in two consecutive weeks) or until female mice reached 36 weeks and male mice 40 weeks of age. Serum was separated and quickly frozen in −70°C for metabolomic analysis. Blood samples for detection of insulin autoantibodies (IAA) were collected from tail vein at the age of 8 weeks. Plasma samples for insulin were collected between noon and 2 PM after 4 hr fast and two days later 5 minutes after intraperitoneal glucose (1 g/kg) administration at the age of 10 weeks. Another set of euglycemic NOD/Bom female mice (Study 2) was sacrificed with decapitation under CO2 anesthesia at the age of 8 weeks (n = 57) or 19 weeks (n = 14), and blood, liver and pancreas samples were collected.
Two separate batches (n = 36 and 42, Studies 3 and 4) of female NOD/MrcTac were delivered from Taconic USA (Hudson, NY, USA) at 5 weeks of age. In Study 3, intraperitoneal glucose tolerance test was performed after 4 hr fast at 8 weeks of age by administering glucose (10% [wt/vol], 1 g/kg body weight) and measuring tail vein blood glucose and serum insulin. Serum samples for lipidomics and IAA were collected from tail vein at 10 weeks of age. Intraperitoneal insulin tolerance test was performed after 1 hr fast at 11 weeks of age by administering human insulin (1.0 IU/kg body weight, Protaphane, Novo Nordisk, Bagsvaerd, Denmark). In Study 4, mice were sacrificed at 10 weeks of age after 4 hr fast by cardiac puncture under anesthesia. Gonadal white adipose tissue (WAT) depot was carefully dissected and weighted, and was used as a marker of adiposity. Serum samples for IAA, lipidomics and adipokine panel assays, gonadal WAT, gastrocnemius muscle and pancreas samples were collected, and stored at −70°C until analyses. HOMA-IR, an estimate of insulin resistance, was calculated as fasting insulin (µIU/ml)×fasting glucose (mmol/l)/22.5. Statistical significances were analyzed with Student's t-test or two-way ANOVA using GraphPad Prism 4.
Blood glucose was measured with Precision Xtra™ Glucose Monitoring Device (Abbott Diabetes Care, IL). Plasma insulin was analyzed with Mouse Ultrasensitive ELISA kit (Mercodia, Uppsala, Sweden) or together with leptin with Milliplex Mouse Adipokine Panel (Millipore, Billerica, MA, USA). Plasma adiponectin was measured with Mouse Adiponectin ELISA kit from Millipore.
Pancreatic islets were isolated using Ficoll 400 (Sigma-Aldrich, St Louis, MI, USA) gradient method [60]. In brief, the pancreata were incubated with Collagenase P (0.5 mg/ml, Roche Diagnostics, Mannheim, Germany) in HBSS containing 10 mM HEPES, 1 mM MgCl2, 5 mM Glucose, pH 7.4 for 17 min. After two rounds of washing, the pellet was resuspended in Ficoll 25%, and the densities 23%, 20% and 11% were layered on top. After centrifugation, the islet layer between densities 23% and 20% was collected and washed twice before snap freezing the pellet for metabolomic analysis or homogenization in lysis buffer for RNA extraction. Samples were stored in −70°C until analyses.
Pancreata from euglycemic NOD mice were cryosectioned. 5 µm sections with >20 µm intervals were stained with hematoxylin & eosin and graded for insulitis as follows: 0, no visible infiltration, I peri-insulitis, II insulitis with <50% and III insulitis with >50% islet infiltration. Total 678 islets from eight female 10-week-old low-risk mice (60–123 islets/each) and 633 islets from eight high-risk mice (59–102 islets/each), and 52 islets from four female 19-week-old low-risk mice (11–17 islets/each) and 28 islets from three high-risk mice (7–10 islets/each) were graded. Statistical significance was analyzed with Student's t-test or Chi Square test using GraphPad Prism 4.
Murine IAA were measured by a radiobinding microassay (RIA) with minor modifications to that previously described for human IAA [61]. Mouse sera (2.5 µl) and serial dilutions of standard samples (5 µl) of a serum pool obtained from persons with a high IAA titer were incubated for 72 h with 15,000 cpm mono125I-(TyrA14)-insulin (Amersham, GE Healthcare, Buckinghamshire, UK) in the presence or absence of an excess of unlabeled human recombinant insulin (Roche Diagnostics, Mannheim, Germany). Antibody complexes were precipitated by adding 50 µl TBT buffer (50 mM Tris, pH 8,0, 0,1% Tween 20) containing 8 µl Protein A and 4 µl Protein G Sepharose (Amersham). After repeated washings the bound radioactivity was measured with a liquid scintillation detector (1450 Microbeta Trilux, Perkin Elmer Life Sciences Wallac, Turku, Finland). The specific binding was calculated by subtracting the non-specific binding (excess unlabeled insulin) from total binding and expressed in relative units (RU) based on standard curves run on each plate. The cut-off value for mouse IAA positivity was set at the mean+3SDS in 29 BALB-mice, i.e. 0.90 relative units (RU).
Serum samples (10 µl) in Eppendorf tubes were spiked with a standard mixture containing 10 lipid compounds at a concentration level of 0.2 µg/sample, and mixed with 10 µl of 0.9% sodium chloride and 100 µl of chloroform∶methanol (2∶1). After 2 min vortexing and 1 hr standing the samples were centrifuged at 10000 rpm for 3 min and 60 µl of the lower organic phase was taken to a vial insert and spiked with 20 µl of three labelled lipid standards at a concentration level of 0.2 µg/sample.
The lipidomics runs were performed on a Waters Q-Tof Premier mass spectrometer combined with an Acquity Ultra Performance LC™ (UPLC; Milford MA). The solvent system consisted of 1) water with 1% 1 M NH4Ac and 0.1% HCOOH and 2) LC/MS grade acetonitrile/isopropanol (5∶2) with 1% 1 M NH4Ac, 0.1% HCOOH. The gradient run from 65% A/35% B to 100% B took 6 min and the total run time including a 5 min re-equilibration step was 18 min. The column (at 50°C) was an Acquity UPLC™ BEH C18 (1×50 mm, 1.7 µm particles) and the flow rate was 0.200 ml/min. The lipids were profiled using ESI+ mode and the data collected at a mass range of m/z 300–1200. The data was processed by using MZmine software (version 0.60) [62], [63] and the lipid identification was based on an internal spectral library [64]. Data was normalized using the appropriate internal standards as previously described [14], [65].
Depending on the protein concentrations of PBS buffered cell solutions, 20–40 µl samples were taken for islet metabolomic analysis. 10 µL of an internal standard labeled palmitic acid-16,16,16-d3 (250 mg/l) and 400 µl of methanol solvent were added to the sample. After vortexing for 2 min and incubating for 30 min at room temperature, the supernatant was separated by centrifugation at 10,000 rpm for 5 min. The sample was dried under constant flow of nitrogen gas and derivatized with 25 µl of MOX (1 h, 45°C) and MSTFA (1 hr, 45°C). 5 µl of retention index standard mixture with five alkanes (125 ppm) was added to the metabolite mixture.
Islet samples were analyzed by two-dimensional gas chromatography coupled to time of flight mass spectrometry (GC×GC-TOFMS). The instrument used was a Leco Pegasus 4D (Leco Inc., St. Joseph, MI), equipped with an Agilent GC 6890N from Agilent Technologies (Santa Clara, CA) and a CombiPAL autosampler from CTC Analytics AG (Zwingen, Switzerland). The modulator, secondary oven and time-of-flight mass spectrometer were from Leco Inc. The GC was operated in split mode with a 1∶20 ratio. Helium with a constant pressure of 39.6 psig was used as carrier gas. The first dimension GC column was a non-polar RTX-5 column, 10 m×0.18 mm×0.20 µm (Restek Corp., Bellefonte, PA), coupled to a polar BPX-50 column, 1.50 m×0.10 mm×0.10 µm (SGE Analytical Science, Ringwood, Australia). The temperature program was as follows: initial temperature 50°C, 1 min→295°C, 7°C/min, 3 min. The secondary oven was set to 20°C above the oven temperature. Inlet and transfer line temperatures were set to 260°C. The second dimension separation time was set to 5 s. The mass range used was 45–700 amu and the data collection speed was 100 spectra/second. Raw data were processed using Leco ChromaTOF software, followed by alignment using Guineu software (version 0.7) [66]. The metabolites were identified by using an in-house reference compound library together with The Palisade Complete Mass Spectral Library, 600K Edition (Palisade Mass Spectrometry, Ithaca, NY).
RNA extraction from islets was carried out with Rneasy minikit (QIAGEN GmbH, Hilden, Germany) and from liver, skeletal muscle (m. gastrocnemius) and gonadal white adipose tissue with Trizol reagent (Invitrogen, Carlsbad, CA) followed by RNase-free DNase I treatment (QIAGEN GmbH) and purification with Rneasy minikit. Pancreatic islets and liver for microarray analysis were collected from 19-week-old euglycemic female NOD/Bom mice. Skeletal muscle and adipose tissue for GLUT4 mRNA expression were collected from 10-week-old female NOD/MrkTac mice.
GLUT4 mRNA expression in skeletal muscle and gonadal white adipose tissue was measured by quantitative real-time PCR. CDNA synthesis was performed with High Capacity RNA-to-cDNA Kit according to manufacturer's protocol. Real-time PCR was performed with 7300 Real Time PCR system, pre-designed TaqMan® Gene Expression Assay for GLUT4 and TaqMan® Endogenous Control Assay for β-actin. The 20 µl PCR reactions contained 8 µl cDNA, 8 µl TaqMan® Gene Expression Master Mix, 1 µl GLUT4 TaqMan Gene Expression Assay, 1 µl b-actin TaqMan Endogenous control Assay and 2 µl depc water. Cycling parameters for real-time RT-PCR were as follows: 50°C for 2 min, 95°C for 10 min followed by 40 cycles of 95°C for 15 seconds and 60°C for one minute. GLUT4 mRNA levels were expressed relative to β-actin, which was used as a housekeeping gene. Relative gene expression was calculated using the comparative CT method and RQ = 2−ΔΔCT formula. All reagents were from Applied Biosystems (Foster City, CA, USA).
RNA amplification was performed from 300 ng total RNA with Ambion's (Austin, TX) Illumina RNA TotalPrep Amplification kit (cat no AMIL1791). IVT reaction overnight (14 hr), during it cRNA was biotinylated. Both before and after the amplifications the RNA/cRNA concentrations where checked with Nanodrop ND-1000 (Wilmington, DE) and RNA/cRNA quality was controlled by BioRad's Experion electrophoresis station (Hercules, CA).
The samples were hybridized in the Finnish DNA Microarray Centre, at the Turku Centre for Biotechnology. 1.50 µg each sample was hybridized to Illumina's MouseWG-6 Expression BeadChips, version 2 (BD-201-0602) at 58°C overnight (18 hr) according to Illumina® Whole-Genome Gene Expression Direct Hybridization protocol, revision A. Hybridization was detected with 1 µg/ml Cyanine3-streptavidine, GE Healthcare Limited (Chalfont, UK) (cat no PA43001). Chips were scanned with Illumina BeadArray Reader, BeadScan software version 3.5. The numerical results were extracted with Illumina's GenomeStudio software v. 1.0 without any normalization.
Bead Summary data, exported from Illumina's GenomeStudio software, was preprocessed using beadarray package [67] of R/Bioconductor [68] as follows. Data was transformed to logarithm (base 2), and normalized using quantile method [69], which equalizes the distribution of probe intensities across a set of microarrays.
Gene Set Enrichment Analysis (GSEA) [20], a commonly used pathway analysis technique for microarray gene expression data analysis, uses a Kolmogorov-Smirnov like statistic to test whether selected gene sets are enriched among the most up or down regulated genes. Multiple hypothesis testing was addressed by computing the false discovery rate q-values based on random permutation of membership of genes across gene sets as implemented in the GSEA software [20]. Linear Models for Microarray Data (LIMMA) approach [70] identifies differentially expressed genes by fitting a linear model to the expression data of each gene, and computing moderated t-statistic using posterior residual standard deviations to account for the gene-specific variability of expression values. Here, we used the R/Bioconductor package [68] and LIMMA [70] for testing differential expression of genes. We then performed pre-ranked GSEA analysis using the moderated t-statistic for ranking the gene list, to test for enrichment of gene sets from a variety of pathway databases such as Gene Ontology (GO) [71], GenMAPP [72], BioCarta (http://www.biocarta.com), Signal Transduction Knowledge Environment (STKE) (http://stke.sciencemag.org/), and KEGG [73] curated in Molecular Signatures Database (MSigDB) [20].
Leading edge genes of an enriched pathway are the genes that account for the enrichment signal [20]. For selected pathways that are found statistically significant by GSEA, the pathway profiles are calculated as average expression of all leading edge genes. This matrix of pathway profiles of selected pathways was then augmented with selected metabolite profiles. Then the numerical values in this matrix were normalized with the autoantibody-negative low-risk group (IAA− & LR), i.e., each numerical value of a variable is divided by the average values from IAA− & LR samples, and transformed to logarithmic (base 2) scale. Then the variables were scaled for unit variance. Finally, hierarchical clustering was applied using Euclidean metric and complete linkage method [74] for computing inter-cluster distances. An R package called gplots (http://www.r-project.org/) was used for the clustering and displaying the numerical values as a heat map.
DNA was extracted from 200 mg of fecal sample from caecum using FastDNA Spin Kit for Soil (QBIOgene, Carlsbad, CA,) with modifications to the manufacturer's instructions [29]. PCR-DGGEs of predominant bacterial PCR-DGGE and five different group specific PCR-DGGEs (bifidobacteria, Lactobacillus-group, Eubacterium rectale – Blautia coccoides clostridial group (Erec-group), Clostridium leptum clostridial group (Clept group), and genus Bacteroides) were performed as described previously [30].
The comparison of the profiles and the quantification of the amplicons were performed using BioNumerics software version 5.1 (Applied Maths NV, Sint-Martens-Latem, Belgium). The statistical analysis of amplicon numbers was performed with the Student's t-test with unequal variances. Clustering was performed with Pearson correlation from each bacterial group separately besides using composite datasets (included predominant bacterial DGGE and five group specific DGGEs) in which amplicons with the total surface area of at least 1% were included in the similarity analysis. Principal component analysis was performed with the BioNumerics software.
R statistical software (http://www.r-project.org/) was used for data analyses and visualization. The concentrations were compared using the Wilcoxon rank-sum test, with p-values <0.05 considered statistically significant. Due to the large number of tests, one test for each of the 154 lipids, for the differences in mean concentrations between the progressor and non-progressor groups some p-values may be small due to chance. In order to quantify the number of such false significant findings we estimated the false discovery rates using q-values [15], [16]. A q-value is associated for each lipid with the interpretation that among those lipids that have p-value less than or equal to the p-value of the lipid a fraction q are falsely stated significant. To account for multiple comparisons, false discovery rates among significantly differing lipids were estimated using q-values [15], [16]. False discovery rates were computed using the R package q-value. The fold difference was calculated by dividing the median concentration in high-risk group by the median concentration in low-diabetes-risk group and taking the base-2-log of the resulting value. This makes interpretation easy as values greater/smaller than zero correspond to up/down-regulated lipids in the high-risk group. In clustering we applied a customized correlation based distance metricWhere and denote the concentrations of lipids an in the sample set. Ward's method was then applied in hierarchical clustering using this distance measure [75].
Metabolic state development in diabetes progressors and non-progressors was modeled by separate Hidden Markov Models [18], making it possible to align individuals based on metabolic states instead of age, and to compare the metabolic states in progressors and non-progressors. The modeling assumptions under which the models are fitted to data are that individuals share a similar developmental progression but the timing of the states may vary, and that metabolite profiles in each state may be different for progressors and non-progressors. Model fitting was done by the standard Baum-Welch algorithm using the MATLAB toolbox by Kevin Murphy (http://www.cs.ubc.ca/~murphyk/Software/HMM/hmm.html). The model structure was validated by the bootstrap in the same way as in our earlier studies [18], and confidence intervals were estimated with non-parametric bootstrap (5000 samples).
Let and be two data matrices with and samples, , and dimensions and , respectively. The task is to find a permutation of samples in such that each sample in is matched with in . We assume a one-to-one matching of samples between the two data matrices. Since the data matrices do not lie in the same data space, it is not possible to use distance as the matching criterion. We have recently introduced a methodology based on statistical dependencies between the data sets to solve this problem [19]. The idea is to compute from the data features or statistical descriptors that maximize statistical dependencies, and do the matching based on the descriptors. In practice, we project the data onto a lower-dimensional subspace such that the statistical dependencies between the datasets are maximized, and find a matching of samples in this comparable subspace.
In order to find disease effects shared by NOD mice and humans in the DIPP study, we first paired metabolites of the two organisms, then estimated the metabolic states of progressor and non-progressor men and mice by HMMs, and finally did a bootstrap-based two-way analysis on progressors/non-progressors vs. men/mice to identify disease and organism effects and their interactions. The data-driven pairing or the metabolites and the four HMMs were computed as described above. The two-way analysis of disease effect was done by first removing the organism effect, represented with a single mean parameter estimated by least squares, and then computing bootstrap confidence intervals for the disease effect of pooled men and mice. Organism and cross effects were estimated analogously.
|
10.1371/journal.pcbi.1007338 | MHC binding affects the dynamics of different T-cell receptors in different ways | T cells use their T-cell receptors (TCRs) to scan other cells for antigenic peptides presented by MHC molecules (pMHC). If a TCR encounters a pMHC, it can trigger a signalling pathway that could lead to the activation of the T cell and the initiation of an immune response. It is currently not clear how the binding of pMHC to the TCR initiates signalling within the T cell. One hypothesis is that conformational changes in the TCR lead to further downstream signalling. Here we investigate four different TCRs in their free state as well as in their pMHC bound state using large scale molecular simulations totalling 26 000 ns. We find that the dynamical features within TCRs differ significantly between unbound TCR and TCR/pMHC simulations. However, apart from expected results such as reduced solvent accessibility and flexibility of the interface residues, these features are not conserved among different TCR types. The presence of a pMHC alone is not sufficient to cause cross-TCR-conserved dynamical features within a TCR. Our results argue against models of TCR triggering involving conserved allosteric conformational changes.
| The interaction between T-cells and other cells is one of the most important interactions in the human immune system. If T-cells are not triggered major parts of the immune system cannot be activated or are not working effectively. Despite many years of research the exact mechanism of how a T-cell is initially triggered is not clear. One hypothesis is that conformational changes within the T-cell receptor (TCR) can cause further downstream signalling within the T-cell. In this study we computationally investigate the dynamics of four different TCRs in their free and bound configuration. Our large scale simulations show that all four TCRs react to binding in different ways. In some TCRs mainly the areas close to the binding region are affected while in other TCRs areas further apart from the binding region are also affected. Our results argue against a conserved structural activation mechanism across different types of TCRs.
| The interaction between T-cell receptors (TCRs) on the surface of T-cells and peptides bound by Major Histocompatibility Complexes (MHCs) on the surface of antigen presenting cells is one of the most important processes of the adaptive immune system [1]. In the case of MHC class I molecules intracellular proteins are degraded by proteasomes into peptides, the peptides are loaded onto MHCs, and subsequently the peptide/MHC (pMHC) structures are presented on the cell surface. The TCRs of T-cells bind to pMHCs with their six hypervariable Complementarity Determining Regions (CDRs) and thereby scan the pMHC (Fig 1A and 1B). Based on this interaction further downstream signalling cascades can be activated and an immune response can be elicited against a particular antigenic peptide. The TCR/pMHC interaction is of relatively low affinity (KD ~0.1–500 μM) and degenerate: One TCR can recognise multiple pMHC and one pMHC can be recognised by multiple TCRs but not every TCR can recognise every pMHC.
A long standing question has been how TCR binding to pMHC results in changes in the cytoplasmic domains of the TCR/CD3 signalling subunits (e.g. phosphorylation), a process termed TCR triggering. Several mechanisms of TCR triggering have been proposed [2], which can be grouped roughly into segregation/redistribution, aggregation, and conformational change models. Conformational change models are of two types [2]: one group, based on the observation that pMHC binding will impose a mechanical pulling force on the TCR [3], proposes that this mechanical force somehow alters the conformation of TCR relative to the membrane and/or CD3 subunits [4–6]. The second group postulates that pMHC binding is accompanied by an allosteric conformation change transduced through the TCRαβ heterodimer [7]. While there is growing support for mechanical models [8–11], evidence for ‘allosteric’ models is more equivocal (reviewed in [2]). Rossjohn and colleagues have reported a change in the AB loop of TCR Cα domain accompanying binding to agonist pMHC [7,12], but structural studies have failed to identify a conformation change conserved in all TCRs upon pMHC binding (reviewed in [2]).
The experimental studies described above relied on X-ray crystallography or NMR spectroscopy, which cannot measure dynamic changes in the TCR structure at atomistic resolution, leaving open the possibly that pMHC binding results in conserved changes in TCR dynamics. Fortunately, computational methods such as Molecular Dynamics (MD) or Monte Carlo simulations can be used to explore this possibility (reviewed in [13]). Recent approaches include TCR/pMHC interface H-bond network analysis [14,15], binding free energy [16] and detachment [17,18] simulations, effects on the MHC [19], peptide [20], and on the TCR [21] or CDR loop characterisation [22]. In a previous large scale study we analysed the same TCR/MHC in combination with 172 different peptides of known experimental immunogenicity [23]. The aim was to use simulations to seek distinct dynamical TCR behaviours for encounters with immunogenic versus non-immunogenic peptides. While no such differences were identified to be involved in discrimination, this does not rule out a role for conformational dynamics in triggering itself. In the current study we explore this by simulating four different TCRs in their agonist pMHC bound and unbound state and then look for conserved conformational features that distinguish bound from unbound TCRs. Using a total of 26 000 ns of simulation time we show that no such conserved and non-obvious features distinguish bound and unbound TCR.
We extracted the structures of the LC13-TCR/FLRGRAYGL/HLA-B*08:01 (accession code 1MI5), A6-TCR/LLFGYPVYV/HLA-A*02:01 (accession code 1AO7), JM22-TCR/GILGFVFTL/HLA-A*02:01 (accession code 1OGA), and 1G4-TCR/SLLMWITQC/HLA-A*02:01 (accession code 2BNR) from the Protein Data Bank (PDB) [24]. Constant TCR regions were included in all simulations as we have previously shown that the constant regions are important for reliable conclusions from molecular TCR and antibody simulations [25].
The LC13 was chosen as a model system as this system has been the target of extensive MD simulations (e.g. [23,26]) due to the availability of a large number of experimentally tested mutations. Additionally we chose the A6, JM22, and 1G4 systems as they have been investigated computationally before (e.g. [14,15,18,27]) and all three TCRs bind to HLA-A*02:01 while the LC13 TCR binds HLA-B*08:01. This allows us to investigate if there are conserved TCR reaction features across MHC types as well as within HLA-A*02:01.
Eight different structures (LC13-TCR, A6-TCR, JM22-TCR, and 1G4-TCR with and without pMHC) were simulated. Each structure was immerged into a dodecahedronic simulation box filled with explicit SPC water allowing for a minimum distance of 1.2 nm between protein and box boundary. Na+ and Cl- ions were added to achieve a neutral charge and a salt concentration of 0.15 mol/litre. Protonation states were determined automatically by Gromacs [28]. Energy minimisation using the steepest descent method was applied. The systems were warmed up to 310 K using position restraints. Hydrogen atoms were replaced by virtual sites to allow for an integration step of 5 fs for the production runs [29]. Final production runs were carried out using Gromacs 4 [28] and the GROMOS 53a6 force field [30]. Parts of the LC13 simulations were taken from our previous work in [21] and [23] while parts of the A6, JM22, and 1G4 simulations were taken from [14] and [15].
Multiple replicas (identical parameter but different seeds) per simulation are important for reproducible conclusions as the results of several studies [15–17,25,31] and a systematic evaluation have shown [32]. Therefore we use 100 LC13 TCR simulation replicas at 100ns each. A boot strapping analysis of these 100 replicas (S1 Fig) shows that 10 replicas reduce the variance between replicas by about 70% while more replicas reduce the variance only slowly further (25 replicas: 80% and 50 replicas 87%). Therefore we simulated the A6, JM22, and 1G4 TCR with and without pMHC for 100 ns using 10 replicas totalling 26 μs (Table 1).
Trajectories were manually inspected using VMD [33] and the vmdICE-plugin [34]. Solvent accessible surface area (SASA), root mean square fluctuation (RMSF), radius of gyration (RG), hydrogen bonds (H-bonds), and distances were calculated by the GROMACS [28] modules gmx sasa, gmx rmsf, gmx gyrate, gmx hbond, and gmx distance respectively and imported into pymol/Matlab using gro2mat [35]. The H-bond networks were visualized using pyHVis3D [14].
We used three different types of measurements to quantify the magnitude of difference between descriptors of TCRpMHC and TCR simulations (modified from our previous study [19]): Firstly, the simple difference in the mean values that is referred to as:
d=X¯TCRpMHC−X¯TCR
Where X-TCRpMHC and X-TCR are the mean values over all frames and replicas of descriptor X (e.g. SASA or H-bonds of a region). The value d helps to quantify the actual magnitude of difference e.g. TCRpMHC simulations have on average 0.5 H-bonds less between their TCR chains than TCR simulations. Secondly, we normalise d by the range of the combined distributions excluding the highest and lowest 2.5% of the values:
d/r=X¯TCRpMHC−X¯TCRrange(X¯TCRpMHC;2.5−97.5%∪X¯TCR;2.5−97.5%)
The value d/r helps to quantify the magnitude of difference related to the width of the combined distributions. Thirdly, we calculate the total variation difference (tvd) to quantify the difference in the probability distributions:
TVD(f1,f2)=12∫|f1(XTCRpMHC)−f2(XTCR)|dx
Where f1XTCRpMHC is the normalized distribution of all TCRpMHC simulation frames and replicas and f2(XTCR) the normalized distribution of all TCR simulation frames and replicas. The tvd is normed between 0 and 1 where a tvd value of 0 represents perfect overlap of the distributions and a tvd value of 1 represents no overlap. In contrast to d and d/r the tvd does not have a sign i.e. is always positive.
We investigated the six CDR loops according to the IMGT [36] definition as extracted from the STCRDAB database [37] as well as the AB loop of the TCR α-chain which was previously hypothesised as influenced by antigen recognition of the LC13 TCR [7], the loops positioned between the variable and constant TCR domains as the linker of the β-chain (“CβFG”) was hypothesised to be important for TCR mechanosensing [10,11], αA and αΒ helices of Cβ as they were reported to be important for CD3 interaction [38], the F and C-strand of Cα as they were reported to be involved in a possible allosteric TCR signalling mechanism [39], and Cα DE and Cβ CC’ loops [40]. These regions and their sequence positon in our four different TCRs are summarised in S1 Table.
In order to evaluate how likely an observed descriptor difference (in d, d/r and tvd) would be seen by chance we performed permutation tests. We merged the n replicas of TCRpMHC simulations with the n replicas of the TCR simulations into one group of size 2n. From this 2n group we picked randomly and with repetition n members for group 1 and n members for group 2. Then we calculated d, d/r and tvd between group 1 and group 2. We repeated this procedure 1000 times and obtained a distribution of d, d/r and tvd values. Finally we determine the quartile (q) of the observed d, d/r and tvd between TCRpMHC and TCR within these distributions of random boot strapping samples as a indicator of significance. E.g. q = 0.98 for d means that 98% of all randomly created group pairs have a smaller d value between them than the d value between TCRpMHC and TCR simulations. An example for a difference in CDR1 loop distance between TCRpMHC and TCR simulations that is likely to be true for the LC13 TCR (d = 0.9 Å and q = 1.0 (i.e. none of the 1000 permutations had a larger difference)) and unlikely to be true the JM22 TCR (d = 0. 3 Å and q = 0.7 (i.e. about 300 of the 1000 permutations had a larger difference) is given in S2 Fig.
We have analysed 61 properties of our 4 TCRs in the pMHC bound and unbound state based on a total of 26 000 ns of simulation time. An overview of these results is given in Table 2 and the results are discussed in detail in the subsequent sections.
The distances (DIST) between the CDR loops of a TCR can be a descriptor for an opening or closing of the TCR binding interface. Especially the CDR3α and CDR3β loops that are positioned centrally over the scanned peptide (Fig 1B) and could be further apart as the pMHC presses between them or could be closer together as a result of binding interface rigidification. Our results show that the first is true for the LC13 and JM22 TCR while the A6 TCR shows a wider distribution and higher distance for TCR simulations (Fig 2). The 1G4 TCR does not show significant changes. The CDR1 and CDR2 distances also do not show patterns conserved across TCRs (Table 2).
The radius of gyration (RG) is a proxy for the compactness of a structure. A high RG indicates a more extended structure while a low RG indicates a more compact structure. We measure the RG of all atoms of the six CDR regions as well as the ABloop, the variable/constant domain linkers and 6 further regions within the TCR constant domains to investigate if pMHC presence has “cramping” effect on any of these regions that could be involved in signalling. Several regions and TCRs show significant differences between their TCRpMHC and TCR states (Table 2); however, these are not conserved. The LC13 and JM22 TCR tend to have lower CDR RGs in their TCRpMHC simulations than in their TCR simulations while the opposite tends to be the case for A6 and 1G4. As an example we show the RG of CDR3α in Fig 3: All differences found are highly significant but they have opposing signs. In contrast the RG of the ABloop and variable/constant linkers are almost unaffected (Table 2).
The solvent accessible surface area (SASA) quantifies the extent that a region is exposed to solvent (here measured using the gmx sasa method of Gromacs). When two proteins bind the solvent accessible area of the binding interface will be reduced. This is the case for all six CDRs of all four TCRs upon pMHC binding (Table 2). The more interesting question is if there is also a change in the SASA if the SASA is measured as if no pMHC would be present for TCRpMHC simulations. i.e. is the protruding of the CDRs altered by pMHC presence? Here we obtain a picture that is partly similar to the RG-analysis. The A6 and 1G4 TCR which tend to have higher RGs in TCRpMHC simulations tend also to have increased SASAs. However, for the LC13 and JM22 TCR the reduced RG in TCRpMHC simulations seems not to lead to a decreased SASA. Fig 4 shows this effect for CDR3α.
The root mean square fluctuation (RMSF) is an indication of how stable areas of a structure are. A potential signal transduction could be the increased or decreased flexibility of an area. We have investigated the RMSF of all TCR residues in Fig 5. Areas of special interest are marked with dashed lines. For these areas permutation tests are given in Table 2. As expected the RMSFs of the CDRs are lower in TCRpMHC simulations than in TCR simulations due to the restricted degrees of freedom in the binding interface. The changes in CDR RMSF between TCR and TCRpMHC simulations for the LC13, JM22 and 1G4 TCRs are high and mostly significant while the changes for the A6 TCR are lower and mostly not significant (compare Fig 5 and Table 2).
In contrast by far the largest observed difference in the RMSF is found in the ABloop of the A6 TCR. In TCRpMHC simulations this loop is about 50% more flexible than in TCR simulations. This is not the case for the other three TCRs. Note that the difference in ABloop arrangement between TCR and TCRpMHC was originally described for the LC13 TCR [7] and not for the A6 TCR as observed here.
We also investigated the number of H-bonds between the TCR chains as a proxy for spatial re-arrangement between the TCR chains (Fig 6). For the JM22 TCR there are on average 1.37 H-bonds less between the TCR chains for TCRpMHC simulations than for TCR simulations. For the LC13 TCR this number is also slightly reduced by 0.51 H-bonds. For the A6 TCR there is a change of 0.61 H-bonds in the opposite direction but due to wider variability this number is not significant based on permutation tests (quartile 0.76). Also for the 1G4 TCR the change is insignificant.
To investigate the overall H-bond network in TCRpMHC and TCR simulations we used pyHVis3D [14] which creates a three dimensional graphical representation of the H-bond distributions (Fig 7). This analysis shows different pictures for the four different TCRs. LC13 TCRpMHC simulations show an increased H-bond presence in the area around CDR3α and CDR3β as compared to LC13 TCR simulations which might indicate an interface rigidification.
This effect is also present in the JM22 TCR but it is much less pronounced than in the LC13 TCR. In the JM22 TCR the area around the linker between the variable and constant region of the TCRβ chain is mainly affected by pMHC presence causing an increased presence of H-bonds which is not observeable in the LC13 TCR.
A6 TCR simulations without pMHC show a higher H-bond frequency in the CDRs of the TCR α-chain while intra TCR chain H-bonds in this area are increased for TCRpMHC simulations. In the constant area of the TCR α-chain H-bonds dominate for TCRpMHC simulations.
Similar to JM22 also 1G4 simulations show a higher H-bonds frequency around the linker between the variable and constant region of the TCRβ chain for TCRpMHC simulations. In contrast to all other TCRs the H-bond patterns of the 1G4 TCR around the CDR loops show a very mixed picture.
Taken together this shows that there are no conserved differences in the H-bond patterns between TCR and TCRpMHC simulations.
We have presented an MD study of four different TCRs (LC13, JM22, A6, and 1G4) in their pMHC bound and unbound form using a total of 26 000 ns. To our knowledge this is the first study that investigates four different TCRs on such a large scale.
The most similar study was published by Hawse et al. [41] who investigated the A6 and DMF5 TCR and found a global rigidification and dampened coupling in the linker between variable and constant TCR domains upon pMHC binding using computational mutagenesis with gradient-based minimization [27] and experimental hydrogen/deuterium exchange. Also Cuendet et al. [18] investigated the detachment of the A6 and B7 TCR from HLA-A2 using about 4 000 ns of steered MD code. This study found interesting binding interface characteristics but did not address the different dynamics within the TCR in the bound and unbound form. In two studies we have previously investigated the LC13 TCR [21,23]. In one of these studies [21] we found multiple significant differences between the pMHC bound and unbound LC13 TCR. These included CDR distance distributions, CDR compactness as well as differences in the TCR hydrogen bond network. In the current study we investigated if these results hold true for other TCR/MHC combinations. Surprisingly, we obtained very different results for JM22 TCR/HLA-A*02:01 compared with LC13 TCR/ HLA-B*08:01. Thus the linker between the C and V regions of the TCR β-chain had a lower RG and RMSF for JM22 TCRpMHC simulations. Furthermore a reduced RG was seen in CDR3α and CDR3β while it was increased in CDR2β. Because LC13 and JM22 bind different MHCs, we investigated two more TCRs (A6 and 1G4) which bind the same MHC as JM22. Even when binding the same MHC type the dynamics within the TCRs vary significantly in, for example, RGs, SASAs, RMSFs (Table 2).
These differences between TCRs upon pMHC binding are consistent with our experimentally measured finding that the JM22, A6 and 1G4 TCRs have very different energetic footprints on HLA-A*02:01 [15]. Other experimental studies support the conclusions drawn in our study. With regard to changes at the binding interface, several structural studies have demonstrated local conformational changes upon TCR binding to pMHC [1], while thermodynamic and kinetic studies of several TCRs, including JM22 [42], are consistent with a reduction in conformational flexibility upon binding. With respect to conformational changes distal from the interface, while the AB-linker of the LC13 TCR has been described to differ between the X-ray structure of the HLA-B*08:01 bound and unbound structure [7], for other TCRs this finding could not be replicated. Similarly the B4.2.3 TCR was reported to be effected in its H3 loop by the binding of H2-Dd presenting a 10-mer HIV-env peptide [39]. But no further support for this being a conserved mechanism could be found in any of the other 10 TCRpMHC complex structures for which a separate TCR structure exists [39].
While our finding argue against pMHC binding inducing conformation changes through allosteric mechanism in the TCR, they remain consistent with models proposing that conformational changes are introduced by mechanical mechanisms [4,5,8–11]. For example, pulling [8,9] and shearing [10] forces have been shown to enhance TCR triggering; with the CβFG loop seemingly be affected by side-wards pulling on the TCR [11].
We conclude that TCR structural dynamics do not differ between TCR/pMHC and TCR simulations in conserved and non-obvious ways. Taken together with previous studies our findings argue against a role for allosteric conformation change models in TCR triggering. Our results are consistent with mechanical models of conformational change, as well as aggregation and kinetic-segregation models.
|
10.1371/journal.pcbi.1002523 | Conformational Spread in the Flagellar Motor Switch: A Model Study | The reliable response to weak biological signals requires that they be amplified with fidelity. In E. coli, the flagellar motors that control swimming can switch direction in response to very small changes in the concentration of the signaling protein CheY-P, but how this works is not well understood. A recently proposed allosteric model based on cooperative conformational spread in a ring of identical protomers seems promising as it is able to qualitatively reproduce switching, locked state behavior and Hill coefficient values measured for the rotary motor. In this paper we undertook a comprehensive simulation study to analyze the behavior of this model in detail and made predictions on three experimentally observable quantities: switch time distribution, locked state interval distribution, Hill coefficient of the switch response. We parameterized the model using experimental measurements, finding excellent agreement with published data on motor behavior. Analysis of the simulated switching dynamics revealed a mechanism for chemotactic ultrasensitivity, in which cooperativity is indispensable for realizing both coherent switching and effective amplification. These results showed how cells can combine elements of analog and digital control to produce switches that are simultaneously sensitive and reliable.
| Bacteria swim to find nutrients or to avoid toxins. Their swimming is powered by the rotation of flagella (hair-like structures) that act as propellers. Each flagellum is driven by a rotary molecular engine (the bacterial flagellar motor) that can rotate in either a counterclockwise or clockwise direction and switches between the two directions are frequent and rapid. Although the motor has been studied in detail, we do not understand how it is able to reliably switch direction – a critical function that gives bacteria the ability to steer. In this paper we examined a mathematical model describing how a potential gearbox in the motor might work inside a ring of identical proteins. We compared the output of this model with experimental data on switching speed and other measures of motor function, finding excellent agreement. This is an exciting finding not only because the operation of the motor itself is important, but also because protein complexes play an important and ubiquitous role in cellular signal transduction and therefore, “conformational spread” may be a widespread mechanism for signal propagation in biology.
| Bacterial chemotaxis enables the cell to move towards favorable environments. This sensing ability relies closely on collective coordination of several operation modules in the signal transduction pathway (reviewed in [1][2]). The first component of this system is responsible for detecting environmental signals and converting them into intracellular signals. At the surface of the cell, detection of attractants and repellents is mediated by a series of chemoreceptors in the cytoplasmic membrane, the methyl-accepting chemotaxis proteins (MCPs). The second component is the intracellular chemotactic pathway, which processes extracellular signal and converts it into one that is used to determine the behavior of the bacterial flagellar motors: the concentration of the soluble cytoplasmic protein CheY. Binding of repellents induces phosphorylation of CheY, whereas binding of attractants results in CheY dephosphorylation. At the end of the chemotactic pathway lies the final component of the system – the motor block – which changes its switching bias in response to changes in CheY-P (phosphorylated CheY) concentration. On a typical E. coli cell surface, there are 4–5 functioning bacterial flagellar motors. When most of the motors on the membrane spin counterclockwise (CCW), flagellar filaments form a bundle and propel the cell steadily forward; if a few motors (can be as few as one) spin clockwise (CW), flagellar filaments fly apart and the cell tumbles. Therefore the cell repeats a ‘run’-‘tumble’-‘run’ pattern to perform a biased random walk for chemotaxis in a low Reynolds number world [3].
The essential feature of the motor that allows effective chemotaxis is its ability to switch direction quickly and reliably in response to small changes in environmental conditions. Previous studies have revealed that the motor switching responds ultrasensitively to changes in intracellular CheY-P concentration: a high concentration of CheY-P in the cytoplasm of the cell stimulates more CW rotation, while a low concentration of CheY-P results in more CCW rotation. In WT E. coli, the cytoplasmic concentration of CheY-P is around 3 mM and the flagellar motors show stochastic reversals of rotation every second or so [4]. A small change in CheY-P concentration up or down disrupts this equilibrium and produces a large shift toward either CW or CCW rotation. The sensitivity coefficient for the change in rotational bias (time spent in CCW vs. CW) as a function of CheY-P concentration (the Hill coefficient) is ∼10 at the most sensitive part of the region of operation [5].
How the flagellar motor accomplishes this switching behavior is not fully understood, partly because structural data are difficult to obtain. It is known that CheY-P molecules interact with a ring-shaped assembly of about 34 identical FliM protein subunits and this unit is believed to be responsible for determining the direction of rotation [6][7]. For several decades, a series of models have attempted to explain the dynamic behavior of the motor switch and identify the underlying kinetic mechanisms that control the steady state behavior of motor switches [8][9]. Tu and Grinstein [10] used a theoretical argument to suggest that in a dynamical two-state (CW and CCW) model, temporal changes in CheY-P concentration drive the switching behavior of the motor at long time scales and produces a power-law distribution for the durations of the CCW states. Bialek et al. [11] used the bacterial motor as a model system to evaluate the noise limitation of intracellular signaling, concluding that the motor switch operates close to the theoretical limit imposed by diffusive counting noise.
A key test for any model of motor switching is the ability to explain how small changes in extracellular concentration are converted into large changes in motor output. To explain this ultrasensitivity, the possibility of cooperative binding of CheY-P to the FliM subunits of the motor switch complex has been suggested [12]. However, studies focused on this binding step [13][14] have determined a Hill coefficient of ∼1 for it, which eliminates the possibility that the amplification is driven by cooperative CheY-P binding to the motor and suggests that a separate, post-binding step within the switch complex is responsible. Duke et al. [15] described a stochastic allosteric model that qualitatively reproduces the ultrasensitive switching and locked state behavior of the motors assuming energetic coupling between neighbor units on the FliM ring inspired by the classic Ising phase transition theory. In particular, this model can reproduce the Hill coefficient of the switch, the nonlinear dependence of rotational bias on CheY-P concentration and the equilibrium between the CW and CCW locked states. The model was based on two assumptions: (a) each subunit of the ring can exist in one of two conformations: CCW and CW state and undergoes a conformational change catalyzed by the binding of CheY-P and (b) a coupling between neighboring subunits favors a coherent configuration and this leads to the propagation of conformational changes along the ring.
Although this model is able to qualitatively reproduce the equilibrium behavior of the motor switch, further work is needed to test its ability to reproduce the dynamics of the switching behavior. Here we investigated in detail the behavior of the conformational spread switching model and its ability to reproduce measurements of locked state intervals, the Hill coefficient and other measures of motor dynamics. We then performed a parameter space search to identify the parameters required to best match experimental findings and make further predictions.
Our Monte Carlo model is based on the approach of Duke et al. [15] and our previous work [16], which we briefly describe here. In addition, we make it more general by extending the assumption of symmetry in their original model to include asymmetric cases. The centerpiece of the model is a multi-protein complex or oligomer (to simulate the FliM ring), the individual protomers of which are identical to one another and arranged in a closed ring of size 34. The interface between adjoining ring units represents domains at the boundary between proteins in a biological multi-protein complex. Each protomer can at any time be in either an active (here denoted A and shaded dark in Figure 1a) or inactive (here denoted I and left unshaded in Figure 1a) state, leading to CW and CCW rotation state, respectively. Each protomer can also be bound (here denoted B) or not bound (here denoted N) to a single CheY-P molecule. Then each protomer can make transitions between four possible states, AB↔AN↔IN↔IB↔AB. The model assumes that each protomer can flip reversibly between the two mechanical conformations (CW and CCW) and ligand binding/unbinding changes its chemical conformations, all of which together contribute to a free energy diagram shown in Figure 1a. To reproduce high sensitivity, the model further assumed a coupling energy between the mechanical conformations of adjacent protomers, which favors alike conformations between neighbors, but that the rate constant for CheY-P binding to a protomer is affected only by the conformation of the protomer itself.
In the original model of Duke et al. [15], it is assumed that the free energy of the active state, relative to that of the inactive state, changes from +EA to −EA when a protomer binds ligand, for simplicity (Figure 1a). In our model, to make it more general, we introduce two separate energy differences between the active and inactive states: EA0 when a protomer is unliganded and EA1 when it is liganded (Figure 1b).
Under the assumption of energy symmetry, the free energy change associated with CheY-P binding can be modeled as , where is the CheY-P concentration required for neutral bias. In the asymmetric case, we use the same definition of , however, does not lead to neutral bias since . In later calculations, we use a numerical method to search for c0.5(asymmetric) as a function of . Finally, the model includes a cooperative energy term (EJ, here called cooperativity) so that the free energy of a protomer is lowered by EJ for each neighbor that is in the same state (Figure 1c). This interaction is crucial because it leads to the stochastic creation of semi-stable ‘domains’: regions of the ring whose constituents are either all in the active or all in the inactive state. These domains can then either (a) shrink and disappear, returning the ring to its previous coherent state or (b) grow to encompass the entire ring, a state in which it will remain until another stochastically growing domain of the opposite type will lead to another ring switch.
Given all these energy combinations, a protomer can make transitions between all possible states at rate constants describing mechanical conformational changes by: is the sum of the free energy changes associated with changes in activity and interaction. The fundamental flipping frequency, ωa, was set as 104 s−1, a typical rate of protein conformational change and consistent with previous modeling of the switch complex [15]. Lacking information about λa, the parameter that specifies the degree to which changes in the free energy affect forwards as opposed to backward rate constants, an intermediate value of λa = 0.5 was selected.
The free energy associated with CheY-P binding depends only on the conformation of the protomer bound, not on adjacent protomers. So the rate constants describing chemical conformational changes are:where c is the concentration of CheY-P, c0.5 is the concentration of ligand at which protomers of the ring are 50% occupied on average under the symmetry assumption. ωb is the characteristic binding rate and ΔG(N→B) is the free energy associated with CheY-P binding. A value of ωb = 10 s−1 was selected based on the experimentally determined CheY-P binding rate [17], and consistent with previous modeling of the switch complex [15][16], and λb = 0 such that the binding rate is independent of protomer conformation. In the case of asymmetric EA, the CheY-P concentration corresponding to neutral bias can only be solved numerically.
We use custom written C++ code to generate a Monte-Carlo simulation of the conformational spread model [16] (including cases of both symmetric and asymmetric energy). At the beginning of each simulation, each protomer on the ring is set to active and with CheY-P bound. Later on, each protomer n of the ring is assigned two transition times, An and Bn, at which it will undergo a conformational change associated with (A) change between CCW and CW state and (B) CheY-P molecule binds on or off. The program progresses iteratively by locating the event in the [A1, A2……A34, B1, B2……B34] array with the earliest execution time, and after change its state accordingly (either mechanical state or chemical state), new transition time An and Bn of that protomer is updated by t−t0 = −ln(rand)/k, where k is the rate constant for the next transition, t0 is the simulation time when the calculation is made and rand is a random number generated in the interval 0 to 1 [18]. If the transition was associated with a change in mechanical conformation of that protomer, then transition times An+1 and An−1 for the two adjacent protomers are also recalculated (for a closed ring of protomers, we defined An+1 for n = 34 to be A1 and An−1 for n = 1 to be A34). The activity of all protomers on the ring is recorded at integer number MΔt, where Δt = 0.1 ms as the output sampling time interval and M goes up to 50,000,000. The algorithm continues to update protomer activities on the ring until the simulation time exceeds a specified maximum. Following Duke et al. [15], we assume that switching in the bacterial motor is controlled by the C-ring in the motor complex, which contains 34 copies of the protein FliM and therefore set n = 34 in our model unless otherwise stated.
In our model we have in total 4 free parameters: EA0, EA1, EJ, c. The CheY-P concentration c is expressed in the unit of c0.5 and when we change it we see the ring operate at different bias and therefore the response curve can be plotted. In the following sections, when we make predictions about ring switching time, switching interval etc., we searched the parameter space EA0, EA1 and EJ across the ranges 0.5 kBT≤EA0≤1.5 kBT , 0.5 kBT≤EA1≤1.5 kBT , and 3.5 kBT≤EJ≤4.5 kBT (shown in Table 1), but for each parameter set, we only present results at neutral bias for simplicity.
A typical screenshot of the ring with multiple domains, labeled with a symbol legend, is shown in Figure 2a. We simulated the qualitative behavior of the ring for a few carefully chosen special cases under the symmetry assumption. Typical screenshots of the ring representing different regimes in the parameter space are shown in Figure 2b.
If the activation energy is zero (EA = 0, Figure 2b, top row), growing domains can only form at random through cooperativity between neighbors, but are unstable and unable to grow sufficiently quickly to encompass the ring because any one of their constituent protomers has a high probability of flipping. When the cooperativity energy becomes high, a coherent ring conformation starts to emerge as the coupling between neighboring protomers is sufficiently strong to lock the whole ring in one conformation. However, as the activation energy is zero, the switch complex loses its ability to respond to ligand concentration changes and switching between coherent inactive and coherent active states can be very slow.
In the absence of cooperativity (EJ = 0, Figure 2b, middle row), the ring displays random salt-and-pepper patterns reflecting the underlying stochasticity of the ligand binding and unbinding process. When the activation energy becomes high, absolute coupling between chemical conformation and mechanical conformation starts to emerge, and the protomers exist in inactive form only when there is no ligand bound and change to active form once ligand binds. In this case, a coherent active conformation of the ring only exists when there are 34 ligands bound to the ring and for a coherent inactive conformation of the ring we find 0 ligand bound.
When cooperativity and activation energy are both present at appropriate magnitudes (EA = 1 kBT, EJ = 4 kBT as discussed in reference [15], Figure 2b, bottom row), the ring spends most of its time locked in either a coherent inactive or active conformation, with transitions between the two (switches) accomplished rapidly by means of a spreading domain. In order to achieve both ring stability and coherent switching, cooperativity is needed to ensure the presence and growth of domains, and activation energy is needed to stabilize these domains by ligand binding. In this energy regime of the model parameter space, the conformational spread model best simulate the performance of the flagellar motor switching responding to external signals.
Ring activity, locked states and switching.
We simulated the behavior of the 34-protomer ring with the method introduced earlier and EA0 = EA1 = 1 kBT, EJ = 4 kBT. A typical result is shown in Figure 3. The top panel is a time series of the number of active ring protomers (34 active protomers correspond to the CW state and 0 to CCW in our convention). The middle panel shows the number of protomers with bound ligand. This graph matches that of the locked states (i.e. the two fit on top of each other if superimposed) along both axes: ligand binding makes the active state more favorable and the active state binds ligand more strongly and the two effects cooperate to produce locked state and switching behavior.
The bottom panel shows the number of independent domains (see Figure 2a for an illustration of typical domain formation present on the ring at any time). Domains appear within a locked CCW or CW state because of stochastic flipping events in protomers and their growth is driven by ligand binding and unbinding and protomer-protomer cooperativity. The domains are transient features of the ring's behavior. They can either (a) disappear or (b) grow (alone or by fusing with nearby domains) to encompass the whole ring, with these latter events corresponding to switches and occurring very rarely: <1% of domains lead to a switch. We find that at the parameter values identified by Duke et al. [15], the number of independent domains almost never exceed 6 (and that such a ring state only exists ∼0.0279% of the time). The vast majority of the time, the ring either contains one domain (coherent state, 83.77% of the time) or contains two domains (15.34% of the time). Four domains are present 0.8685% of the time.
In addition to being able to reproduce the locked coherent state on the ring and fast switching behavior, a separate key test of the conformational spread model is its ability to reproduce the relationship between changes in CheY-P concentration and motor bias. The Hill coefficient (the maximum sensitivity of the switch) is defined in this case using the relation:where Y is the CW bias, h is the Hill coefficient, c is the concentration of the CheY-P and c0.5 is the concentration required for neutral bias (in the case of asymmetric model, replace c0.5 to c0.5(asymmetric)). Here we estimated the Hill coefficient by fitting a linear equation to a plot of log[Y/(1−Y)] against log(c/c0.5), with the slope of this line corresponding to h. For each parameter set, its characteristic Hill curve can be generated by long time simulation with varying c, and plot CW bias of the simulated trace as a function of c.
The sensitivity of ring activity to changes in ligand concentration depends more strongly on the activation energy and considerably less on the cooperativity. A lower sensitivity can be brought about by a lower activation energy or by a lower cooperativity, with the activation energy having the dominating influence. However, with a 34 protomer ring, the cooperativity can be no less than the critical cooperativity required for coherent switching to occur, i.e. EJ>3.5 kBT [15]. In table 1, we present the Hill coefficient calculated for parameters EA0, EA1 and EJ across the ranges 0.5 kBT≤EA0≤1.5 kBT , 0.5 kBT≤EA1≤1.5 kBT , and 3.5 kBT≤EJ≤4.5 kBT. We see that the experimentally determined Hill coefficient ∼10 can be reproduced by a large parameter sets.
The behavioral features of the ring can be further characterized by the distributions of (a) the times spent in the two locked states and (b) the times required for both CCW→CW and CW→CCW switches. Here we used simulated ring state data to obtain the theoretical length of the locked state intervals predicted by the model. Because we have direct access to the fundamental protomer states, filtering and threshold algorithms are not needed to identify the intervals (and switches, respectively). Because transitions between the two locked states are not instantaneous, we needed an unambiguous way to define CCW and CW intervals, respectively. We defined such an interval as the time (in simulation steps) between when the ring enters a fully locked state (0 or 34 active protomers, respectively) and when it next enters the other fully locked state (i.e. 34 or 0 active protomers, respectively).
Distributions of locked state intervals obtained from simulation traces (EA0 = EA1 = 1 kBT and EJ = 4 kBT at neutral bias) equivalent to 30000 seconds of real time are shown in Figure 4 (a). To make a comparison, the log-linear plot of the distributions at low (0.2) and high (0.8) CW biases are also shown in Figure 4(b). Least-squares fitting of exponential curves to the simulation data are shown overlaid. We see that the locked state distribution follows an exponential distribution. In table 1, we presented the mean locked state interval values calculated for parameter EA0, EA1 and EJ across the ranges 0.5 kBT≤EA0≤1.5 kBT , 0.5 kBT≤EA1≤1.5 kBT , and 3.5 kBT≤EJ≤4.5 kBT. Within the parameter range of our simulation, the minimum value of mean locked state time is 0.13 s and the maximum value is 22.17 s (shown in Table 1). The mean locked state time increases when the energy of activation or cooperativity is increased, with the activation energy EA having the dominant influence.
The essential feature of interest of the model proposed by Duke et al. [15] is that the ring can simultaneously achieve very rapid switches and very stable locked states. This qualitatively matches what is observed in the rotary motors of flagellar bacteria such as E. coli, which can rotate at hundreds of RPM stably for a long period but switch direction quickly (on the order of ms) and stochastically. Distinct from the classic MWC model, which requires coherent switches to happen instantaneously, in our model switches occur by a mechanism of conformational spread. We defined a switch time as the time (in simulation steps) between when the ring leaves a fully locked state (0 or 34 active protomers, respectively) and when it next enters the other fully locked state (i.e. 34 or 0 active protomers, respectively).
We simulated the behavior of the ring at the optimal activation energy and cooperativity values identified earlier (EA0 = EA1 = 1 kBT, EJ = 4 kBT) for different values of bias. The empirical distributions thus determined are shown in Figure 5.
In contrast to the distributions of locked state intervals, the switch times follow a peaked gamma distribution. At the parameter value chosen, the mean lies between 58–61 ms for low (0.2), middle (0.5) and high (0.8) biases and these are statistically independent of bias and of direction of switch. In table 1, we presented the mean switch time values calculated for parameter EA0, EA1 and EJ across the ranges 0.5 kBT≤EA0≤1.5 kBT , 0.5 kBT≤EA1≤1.5 kBT , and 3.5 kBT≤EJ≤4.5 kBT. The result of our simulation shows that the mean switch time values changes across the ranges from 12.03 ms to 322.65 ms. The mean switch time increases when activation or cooperativity energy is increased, with the activation energy having the dominant influence.
To confirm that typically one domain of opposite conformation (rather than several) grows to encompass the entire ring, we also computed power spectra for the ring activity traces in order to characterize the spectral properties of the ring switch complex. If switching events are associated with a single nucleation event (a Possion step), we expect the power spectra of the trace to be monotonically decreasing with a ‘knee’, i.e. display a Lorenzian profile. In contrast, if switching events are associated with multiple hidden steps, as for example in a closed biochemical system with hidden reactions, then we expect a non-Lorentzian profile with a peak (a local maximum) at a characteristic frequency related to the number of steps involved [19].
Our simulation results (Figure 6) show the spectra thus obtained are Lorentzian without a local maximum at long times. This behavior is observed at different values of bias. These results offer an internal confirmation of the model results shown in Figure 4, which indicate that the distributions of times spent in the locked states are exponential. Such a system would be expected to display power spectra with Lorentzian profiles. However, because the power spectra and locked state time distributions are computed independently and by different methods, the result that they predict the same behavior is an important internal test of the model. In particular, the power spectra results confirm that the locked state time distributions are not an artifact of our algorithm for detecting the start and end of a locked state.
In our recent experimental paper [16], we used a high-resolution optical system to measure the switching time and locked state interval of bacterial flagellar motors. The experimental observations confirmed that the switching time distribution follows a broad gamma distribution with mean switch time 18.72 ms and the locked state interval follows an exponential distribution with mean interval value 0.75 s at neutral bias. Hence we use mean switch time (∼18 ms), mean locked state time (∼0.75 s) and Hill coefficient (∼10) to parameterize our model.
Table 1 shows a coarse parameter search of our model with predictions of the switching time, locked state interval, and Hill coefficient. We identify the 0.55 kBT≤EA0≤0.95 kBT , 0.55 kBT≤EA1≤0.95 kBT , and 4.05 kBT≤EJ≤4.25 kBT region for a fine parameter search (Table 2). We see with only a few parameter sets, the conformational spread model is able to reproduce the three experimentally determined quantities. With the results shown in Table 2, we find 3 groups of values that fit the experimental values determined by Bai et al. [16]. They are EA0 = 0.55kBT EA1 = 0.75kBT EJ = 4.15 kBT, EA0 = 0.75kBT EA1 = 0.55kBT EJ = 4.15 kBT, EA0 = EA1 = 0.65kBT EJ = 4.15 kBT. We therefore expect a conformational spread model with activation energy ∼0.65 kBT and coupling energy ∼4.15 kBT can well reproduce experimental observations. Please see Figure 7 for a visual summary of our computational results. For simplicity, we only showed those values with EA0 = EA1 = EA and the best-fit parameter set has been labeled by a square.
Although we have identified a best-fit parameter set that can well reproduce experimental findings, we have to note that these fit values are sensitive to the parameters fixed earlier, especially to the fundamental flipping frequency ωa. Here we investigate how mean locked state time and mean switch time respond to changes of ωa while other parameters remain fixed. We see in Figure 8 that the fundamental flipping frequency is a scaling factor of the system. Both mean locked state time and mean switch time are inversely scaled by the flipping frequency. When flipping frequency is higher, each protomer on the ring makes more attempts to flip to the opposite conformation and therefore the locked state becomes less stable (hence mean locked state time decreases) and transition becomes much faster (hence mean switch time decreases); when flipping frequency is lower, each protomer on the ring makes fewer attempts to flip to the opposite conformation and therefore locked state becomes more stable (hence mean locked state time increases) and transition becomes much shorter (hence mean switch time increases).
In the above sections, we have determined the best-fit model parameters using experimental results. It will be interesting to test the ring behavior at different sizes using those values. When Duke et al. [15] first proposed the conformational spread model, they identified that EJ>kBT ln N (N is the size of the ring) is the condition under which a large ring has the characteristic of a coherent switch. In the case of 34 protomers, this condition requires that EJ>3.5kBT. When this condition is met, in time series of ring activity, we see for the majority of time that, the ring stays in complete active (active protomer = 34) or complete inactive (active protomer = 0) state. This invokes an empirical mathematical definition of ‘coherent switch’: the active number of protomers on the ring has to be in 0 or N for greater than 65% of the total simulation time.
We then simulated the ring activity at sizes of 10, 60, 100 protomers with activation energy 0.65 kBT and coupling energy 4.15 kBT at neutral bias and the result is shown in Figure 9. From the requirement of EJ>kBT ln N we expect to see coherent switch behavior for ring sizes at 10, 34, and 60, but not at 100.
Indeed, in Figure 9, we see with the same parameter set, the smaller the ring is, the easier a switch happens. At ring size of 10, 34, 60 protomers, we see clear locked states in the trace and the switching events are fast. However, at the ring size of 100, a switch across the ring becomes very difficult and the time spent during a switch is comparable to the time the ring stays in a locked state. In Table 3, we made predictions about the mean locked state intervals, mean switch times and Hill coefficient of the switch response with activation energy 0.65 kBT and coupling energy 4.15 kBT at different ring sizes.
In this study, we undertook a comprehensive numerical simulation analysis of a general model of stochastic allostery in a protein ring and evaluated the ability of such a model to explain the switching, sensitivity and locked state behavior of the rotary bacterial motor. We modeled the gearbox of the motor as a ring of 34 identical protomers, a geometry inspired by the FliM structure in the motor complex, believed to be responsible for motor switching. The model is able to qualitatively reproduce the motor behavior, such as locked rotation in CCW or CW state and fast switching between the two. Furthermore, based on a comprehensive parameter space search, the model can also quantitatively account for the experimentally determined switch time, locked state interval and Hill coefficient of the motor. Specifically, we found a unique set of values that fit the experimental value best, activation energy must be around 0.65kBT and the cooperativity around 4.15kBT. The bounds around these values are tight. Smaller or larger energies result in rings that either (a) spend too long or too little time in the locked states, (b) do not have the required sensitivity or (c) far away from this parameter regime, fail to switch coherently.
With the ring operating in this parameter set, time traces of ring state (measured as the number of active protomers) indicate that the ring spends most of its time in one of the two locked states, with rare (every 0.5–2 seconds) switches between the two being accomplished very rapidly (on the order of milliseconds). The trace of ligand activity (measured as the number of ring protomers having bound ligand molecules) mirrors the ring state, with the two driving each other: binding of more ligand drives active domain formation, which in turn leads to a preference for having ligand bound and conversely. Rather than being completely locked in one stable state with all protomers being either active or inactive, the ring displays constant activity in the form of nascent domains of the opposite state to the locked state, seen as ‘noise’. For the vast majority of the time, only one such domain exists, and the presence of two (but no more) growing domains is frequently, but not always, associated with a switching event.
The model predicts that the time spent by the ring in the locked states corresponding to CW (all protomers active) and CCW (all protomers inactive) is exponentially distributed. The model can also predict the Hill coefficient (∼10) measured for the sigmoidal curve that relates CheY-P concentration to motor bias. Near the optimal parameter point identified, the distributions of the switching times are gamma-like with a peak around 5–8 ms.
To be effective, a switch must achieve two globally conflicting properties. It must accomplish sensitivity by amplifying small changes in the effector, but only over a narrow critical range (the switching point). Outside this range, it must accomplish reliability by being unresponsive to changes in the effector. The allosteric switching model explains how the motor simultaneously meets these competing design requirements. Near the critical CheY-P concentration, a highly cooperative mechanism (EJ≫1kBT) is used to amplify small stochastically occurring nascent domains that can rapidly grow to encompass the entire switching complex. The resulting digital switch displays the desired selective ultrasensitivity but switches chaotically. In order to ensure switch reliability, CheY-P binding must moderately stabilize the protomer active state, providing a mechanism for biasing the whole switch complex by continuously varying the CheY concentration over a large range: a strategy typical of analog control. The values and ratio of the strengths of the two mechanisms must be tightly controlled in order for the switch complex to be functional. We hypothesize that this control is accomplished through the biochemical structure of the protomers and ring, which are genetically determined and so robust to intracellular noise during the cell's life. In light of recent studies of digital cellular signaling [20], we wish to further suggest that the combination of analog and digital control here proposed to explain the behavior of the bacterial switch complex may be a motif typical of biological switch design.
In this study we have focused on a ring consisting of 34 protomers because it is believed that the C-ring in the E. coli flagellar motor, which consists of 34 copies of the protein FliM, acts as the motor direction switch. However, numerous examples of protein rings and other interconnected protein complex geometries are known, including DNA polymerase sliding clamps, voltage-gated ion channels, ATP synthase etc. Each of these rings may hypothetically accomplish its function using a conformational spread mechanism, but would consist of different numbers of protomers. In our model this can be simulated by changing N, the number of elements in the ring. In this paper, we have narrowed our study to a closed protein ring. However, we have to point out that the conformational spread model as well as the numerical method we presented here can be easily modified to describe one dimensional allostery regulation in a protein chain or a strand of DNA molecules. The model can also be modified to describe signal transduction and amplification on a two dimensional plane, which will be of great use in studying functions of cellular receptors.
Allostery is a widespread mechanism in biology and conformational change is the basis for a large subset of all protein function. Since protein complexes are the workhorses of the cell, we expect models similar to this and the idea of conformational spread in general to be increasingly important in systems biology and biophysics. Investigating the applicability of conformational spread models to other biological systems will be the subject of future work.
Bacterial chemotactic exploration depends on the ability of the flagellar motors at the base of the flagella to perform two tasks: (1) remain stable in their current direction of rotation for long periods (seconds) as required and (2) switch quickly between the two directions in response to the environmental changes detected by the chemotaxis pathway. These properties make the bacterial switch an exquisite computational element that combines ultrasensitivity and reliability. In this paper we presented an analysis of a model featuring conformational spread that aims to explain the mechanism of the motor switch. Simulations confirm that this model is able to reproduce the characteristics of the motor observed in experiments. We speculated that stochastic models of conformational spread will be a common theme in protein allostery and signal transduction.
|
10.1371/journal.pntd.0002422 | Patterns of Migration and Risks Associated with Leprosy among Migrants in Maranhão, Brazil | Leprosy remains a public health problem in Brazil with new case incidence exceeding World Health Organization (WHO) goals in endemic clusters throughout the country. Migration can facilitate movement of disease between endemic and non-endemic areas, and has been considered a possible factor in continued leprosy incidence in Brazil. A study was conducted to investigate migration as a risk factor for leprosy. The study had three aims: (1) examine past five year migration as a risk factor for leprosy, (2) describe and compare geographic and temporal patterns of migration among past 5-year migrants with leprosy and a control group, and (3) examine social determinants of health associated with leprosy among past 5-year migrants. The study implemented a matched case-control design and analysis comparing individuals newly diagnosed with leprosy (n = 340) and a clinically unapparent control group (n = 340) without clinical signs of leprosy, matched for age, sex and location in four endemic municipalities in the state of Maranhão, northeastern Brazil. Fishers exact test was used to conduct bivariate analyses. A multivariate logistic regression analysis was employed to control for possible confounding variables. Eighty cases (23.5%) migrated 5-years prior to diagnosis, and 55 controls (16.2%) migrated 5-years prior to the corresponding case diagnosis. Past 5 year migration was found to be associated with leprosy (OR: 1.59; 95% CI 1.07–2.38; p = 0.02), and remained significantly associated with leprosy after controlling for leprosy contact in the family, household, and family/household contact. Poverty, as well as leprosy contact in the family, household and other leprosy contact, was associated with leprosy among past 5-year migrants in the bivariate analysis. Alcohol consumption was also associated with leprosy, a relevant risk factor in susceptibility to infection that should be explored in future research. Our findings provide insight into patterns of migration to localize focused control efforts in endemic areas with high population mobility.
| In Brazil, leprosy remains a significant public health problem in endemic clusters of high transmission risk throughout the country. Migration is thought to be a factor associated with continued leprosy transmission, as migration has also been found to be associated with other Neglected Tropical Diseases (NTDs). We analyzed the association between past five year migration and leprosy as part of a larger epidemiological study evaluating risk factors for infection among recently diagnosed leprosy cases (n = 340) and a matched clinically unapparent control group (n = 340) in the northeastern state of Maranhão. Among migrants with leprosy, 23.5% (n = 80) migrated in the past five years, with 16.2% (n = 55) of the control group. Past five year migration was significantly associated with leprosy, and remained significant after controlling for household and familial contact as potential confounders. Factors found to be associated with leprosy among past 5-year migrants included alcohol consumption, poverty, and household, family and other leprosy contact. Key patterns of movement emerged from the study that may aid future regional leprosy control efforts.
| Leprosy continues to be an endemic disease in many parts of the world. Brazil has globally the second highest new case incidence [1]. National leprosy prevalence of 1.54/10,000 in 2010 [2] remains above the WHO goal of <1 per 10,000. Highly endemic areas of the disease continue to persist despite large-scale national efforts to control the disease. A challenge in disease control efforts is compounded as leprosy can be diagnosed many years after infection took place due to the long incubation period, and mild early symptoms of the disease may be overlooked. Migration has been found to be a social determinant of disease [3], and has been hypothesized as a risk factor in continued leprosy incidence [4], [5], [6]. In fact, earlier research in Brazil highlighted the increased distribution of leprosy along new corridors coinciding with frontier expansion connecting southern agricultural areas to the north of Brazil [7], as well as periurban migrant settlements on the outskirts of urban centers [4]. Migrants move between endemic and non-endemic areas in Brazil and often live in substandard conditions. As an infectious disease caused by Mycobacterium leprae, leprosy primarily affects the skin and peripheral nerves and causing sensory loss. While nasal mucosa is considered the main transmission site, new research indicates that oral presence of M.leprae bacilli may be an additional mode of transmission [8]. Maranhão, the study area of this research, has the third highest prevalence of leprosy (5.34/10,000) in the country [2] and is among the states with the highest out- and return- migration rates [9].
The proliferation of leprosy in Brazil continues largely in conditions of poverty that include poor housing and sanitation, high household density, illiteracy and low socioeconomic levels both at the micro and macro levels [4], [10]–[13]. Rapid population growth and uncontrolled urbanization, often as a consequence of migration for employment and differential access to services between rural and urban areas, has facilitated the expansion of these poor social and environmental conditions on the peripheries of cities associated with leprosy infection [4]–[5], [7], [13]. Additionally, new road construction and railways have enabled movement between rural communities and urban areas. These developments in transportation have been argued to explain the expanded distribution of leprosy in Brazil [4]–[6]. Nevertheless, household leprosy contact continues to be the primary risk factor associated with leprosy infection [14]. Proximity to the household contact has been seen as relevant in terms of increased risk [15]. Consanguineous contact has also been found to be associated with leprosy. Findings from Moet et al. (2006) suggest evidence of a genetic relationship independent of physical contact for leprosy infection.
Migration has been found to be an impediment to both leprosy elimination and control efforts. Prior research has suggested that migration may influence transmission and distribution of the disease [5], [16] as well as other neglected tropical diseases (NTDs) [3], [17]–[23]. This study explores the spatial and temporal patterns of migration in individuals with leprosy in Maranhão. The study also examines risk factors associated with leprosy among individuals who have migrated in the past five years (past 5-year migrants). Comparison of risks associated with leprosy and migration is challenging in a homogeneous population. However evaluation of specific risk factors that differentiate leprosy among past 5-year migrants from a clinically unapparent control group without clinical signs of leprosy who migrated in the past five years in this investigation, sheds light on those factors that are of importance when considering leprosy infection and expression of disease. The study has three specific aims: 1) to examine if migration in the past five years is a risk factor for leprosy; 2) to describe and compare geographic and temporal patterns of migration among past 5-year migrants with leprosy and a control group without clinical signs of leprosy; 3) to examine the social determinants of health associated with leprosy among past 5-year migrants.
Written approval was obtained from the Ethical Review Board of the Federal University of Ceará (Fortaleza, Brazil). Permission to perform the study was also obtained by the Maranhão State Health Secretariat, the State Leprosy Control Program and municipalities involved. Informed written consent was obtained from study participants, or their parent/guardian in the case of minors, after explaining the objectives of the study. Interviews were conducted in private.
The research was conducted in four leprosy endemic municipalities in the state of Maranhão, Brazil: Santa Inês, São José de Ribamar, Codó, and Bacabal. These municipalities are located in a major endemic cluster identified by the Brazilian Ministry of Health as a high-risk area for leprosy transmission [16]. Santa Inês, (population 77,282) [24], Codó (population 118,038) [24], and Bacabal (population 100,014 same) [24] are small townships in the interior of Maranhão that are largely surrounded by rural agricultural production, while São José de Ribamar (population 163,045) [24] is on the outskirts of the capital city, São Luis. Most households are small brick or mud and palm residences with rudimentary plumbing and hammocks to accommodate the multigenerational inhabitants.
A case-control study was designed as part of an extended epidemiological investigation on risk factors associated with leprosy infection in four highly endemic municipalities in Maranhão, as part of the MAPATOPI study. The MAPATOPI study is an interdisciplinary project to support and improve the Brazilian leprosy program in Maranhão, Pará, Tocantins, and Piaui. Variables associated with past five year migration among those diagnosed with leprosy between 2009–2010 were compared with a matched clinically unapparent control group without clinical signs of leprosy. Migration was defined as those who resided outside of the municipality of their current residence, and is limited to five years as this is the average incubation period from leprosy infection to symptom onset. Past five year migration data is also collected in the Brazilian National Household Survey [9]. A detailed analysis of socio-cultural, health service related and economic variables that were collected as part of the larger epidemiological study will be explored elsewhere.
The case group was identified through the database of the National Information System for Notifiable Diseases (Sistema de Informação de Agravos de Notificação – SINAN) and included adults 15 and older in each of the four sites diagnosed with leprosy in 2009–2010 (n = 394). Individuals under 15 years of age, those previously diagnosed with leprosy and relapsed, living outside of the highly endemic cluster and who could not be located through multiple contact attempts were excluded from the study. The control group (n = 391) was selected from the Programa Saúde da Família (Program for Family Health). This program registers all families in the catchment areas of the clinic by community health workers. At the clinics, we randomly selected intake forms from the Program for Family Health for age and sex at each clinic and contacted those individuals for inclusion in the control group. Each of the matched controls were clinically evaluated for signs of leprosy. Any individual with a clinical suspicion of leprosy was excluded from the study and referred to municipal health centers for further diagnostic testing.
Data collection was conducted between April and August 2010. Data collection was coordinated through the Municipality Health Secretariats with the support of the Maranhão State Health Secretariat. Study participants were recruited by community health agents for the study. They were interviewed by trained health professionals at the local health care centers, or in patient homes when disability or age prevented health center attendance. Information on demographics, socioeconomic status, healthcare access, migration, behavior and stress was collected through structured questionnaires. Clinical data were also collected through patient medical records.
Data were entered twice using EpiInfo software version 3.5.1 (Centers for Disease Control and Prevention, Atlanta, USA) and cross-checked for entry-related errors. Statistical tests were used to assess normality. Included are data sets with information related to migration. Any cases that did not have complete migration data were excluded from the analysis. Of the 340 leprosy cases and 340 matched controls, we first identified 135 (19.9%) past 5-year migrants in the case (n = 80) and control groups (n = 55). The distribution of key demographic, spatial and temporal migration pattern variables among past 5-year migrants in the case and control groups was examined and tested by the use of Fishers exact test for significant differences in the stratified sample of past 5-year migrants.
We then conducted bivariate analyses comparing cases (n = 340) and controls (n = 340) using Fishers exact test to examine if past five year migration was associated with leprosy diagnosis. As household contact remains the most significant known transmission risk to date for leprosy infection [14], [15], we additionally undertook multivariate logistic regression analysis controlling for family (parent, child and/or sibling) and household (consanguineous and/or non-consanguineous) contact with leprosy.
Next, stratified bivariate analyses using Fishers exact tests were used to determine differences in the association among social determinants of health (socioeconomic status), psychosocial (alcohol use and life stressors) and biosocial factors (leprosy contact exposure) for case and control groups of past 5-year migrants (n = 135).
A total of 394 leprosy cases and 391 controls were interviewed. There were 23 relapsed leprosy cases and 12 controls suspected of leprosy who were excluded from the study. Eight respondents refused to participate. Complete migration data was available for 680 respondents. Of the 340 leprosy cases and 340 matched clinically unapparent controls, 23.5% of those with leprosy (n = 80 cases) and 16.2% (n = 55) of the control group without clinical signs of leprosy migrated in the past 5 years before diagnosis. Only 4.4% (n = 15) of cases migrated after diagnosis. Table 1 reflects migration into and out of major endemic clusters identified by the Brazilian Ministry of Health as high-risk areas for leprosy transmission [6] (Figure 1), and other demographics and migration variables. These variables were not significantly associated with leprosy among past 5-year migrants prior to diagnosis (test results not shown). Leprosy cases were largely among the youngest age group (15–29) migrating, with an equal distribution between males and females. More than one-third of those with leprosy who migrated in the past five years were illiterate. The majority of leprosy cases migrated within cluster 1, which includes the northern states of Pará, Piauí, Tocantins and Maranhão. More than half (56.3%) of cases moved between municipalities in Maranhão, followed with fewer cases to neighboring Pará (11.8%), Piauí (3.9%) and Tocantins (2.0%), and one-fifth of migrants were drawn to non-contiguous states. All those with leprosy migrated into a highly endemic cluster on at least one occasion, not including their current residence.
Nearly one in six migrants with leprosy migrated for employment in the last five years and this was slightly less than expected for internal population movement. Typical of internal population flow, most migration in Maranhão was to urban areas (60.3%) compared to rural areas (33.3%), and both rural and urban areas (7.7%). Social networks in migration destination sites for those with leprosy had a higher tendency to be family contacts with whom they lived (81.0%) than work contacts (17.7%). This may be an explanation for the significant number of respondents who always had a contact prior to migrating (79.8%). Migrants with leprosy lived on average with 8.61 people per household while migrating.
Past five year migration prior to diagnosis was found to be significantly associated with leprosy as shown in Table 2 which represents the results of the multivariate logistic regression analysis. Past five year migration remained significantly associated with leprosy after controlling in separate models for 1) household contact (consanguineous and/or non-consanguineous); 2) family contact (parent, child and/or sibling; 3) and household and family contact in multiple logistic regression models.
Key social, biosocial, and behavioral factors were found to be associated with leprosy (Table 3). Household, familial and other contact with someone infected with leprosy was significantly different for leprosy infected past 5-year migrants compared to control group migrants. Genetic association of closely related kinship shows a significant difference for contact with parent/child/sibling (OR: 7.82; CI 95%: 2.32–33.38; P-value = 0.0001). Contact regardless of consanguinity (OR: 4.99; CI 95%: 1.7–16.51; P-value = 0.001) and actual household contact (OR: 5.54; CI 95%: 1.49–30.46; P-value = 0.004) was also significant. An important behavioral factor distinguishing migrants with leprosy compared to the clinically unapparent control group was past five year alcohol consumption (OR: 4.46; CI 95%: 1.43–14.15; P-value = 0.005).
Income and other socioeconomic variables showed significant differences between migrants with leprosy and the control group. Income less than the minimum wage (OR: 2.12; CI 95%: 0.97–4.71; p-value = 0.049) as well as poor access to public waste services (OR: 3.1; CI 95%: 1.1–10.02; p-value = 0.03) and family illiteracy (OR: 2.67; CI 95%: 1.13–6.51; p-value = 0.02) were found to be associated with leprosy among past 5-year migrants.
Education, presence of BCG scar, zone of residence and lifestyle stressors - separation from family and friends, loss of employment or income, marital separation or death of close friend or relative- were not significantly associated with leprosy among past five year migrants.
Leprosy was introduced to Brazil through European colonization and later through slave movement so that by the 1600's, leprosy was well established in the country [25]. More recently, migration has been hypothesized to be an impediment to leprosy control, and spatial analysis indicates the introduction of leprosy through inter and intra-state population movement in Brazil [5], as well as expanded distribution of leprosy through migration [26]. Population movement can put both migrants and non-migrants at risk when diseases move between endemic and non-endemic areas. Latent symptomology, characteristic of leprosy, could facilitate the distribution of disease when no symptoms are present, or when mild symptoms are overlooked. The migrant lifestyle poses similar marginalized socioeconomic, behavioral and environmental risks that have been well established as factors associated with leprosy transmission [4], [10]–[13], [27]–[28].
Leprosy in this study was found to be significantly associated with past five year migration. Susceptibility among migrants may, in part, be due to spatial and temporal patterns of movement in and between areas identified by the Brazilian Ministry of Health as highly endemic clusters for leprosy transmission [16]. While we found no significant difference between key spatial and temporal variables and past five year migration among those with leprosy compared to the clinically unapparent control group, more than half of movement for internal migration among those with leprosy was within the leprosy endemic cluster in the state of Maranhão. Few migrated to the other nine endemic clusters in Brazil, a third migrated to other non-endemic areas, and less than half migrated outside of Maranhão. From an operational perspective for leprosy control in Brazil, this provides sufficient evidence to suggest future surveillance of population flow between municipalities in Maranhão, which should involve comparison of the distribution of leprosy incidence over the five year latency period. Should these areas be identified as emerging endemic areas, service delivery strategies should target these as focal points for state control efforts.
Maranhão continues to be a state with higher net out- and return- migration [9]. Interstate population movement, such as to neighboring Pará, draws many poor migrants from Maranhão's interior leprosy endemic areas to the employment found in large-scale mining and agriculture industries [29]–[30]. Interstate movement necessitates cross-border cooperation for leprosy control and may aid in identifying impending high-risk areas for disease distribution. In fact, other research showed that 5.2% of leprosy patients in Cluster 1 (including Maranhão and neighboring states of Tocantins, Piaui and Pará) were diagnosed outside of their municipality of residence between 2001–2009 [31]. Municipalities in Maranhão and neighboring Pará, which have the third and fifth highest new case incidence in the country respectively, would be good targets for future collaborative surveillance projects.
Our findings indicate that the majority of migration in Maranhão continues to be between rural and urban areas, consistent with other research on population flow in Brazil [32]. However population movement documented in our study appears to be of longer duration than is typical for temporary circular migration. Rural to urban migration is a common solution to reduce poverty, as more and regular job opportunities tend to exist in urban areas [33]–[35]. This often places migrants at higher risk for disease morbidity and mortality due to poor living conditions in urban slums [36]. Kerr-Pontes et al.'s (2004) [4] ecological study in Brazil's northeast demonstrated that urban population growth due to uncontrolled urbanization and migrant influx from Brazil's rural interior, was a predictor of leprosy incidence.
We found that population movement is clearly facilitated through strong destination based social networks as a precursor to migration. These social networks tend to be family-based, as indicated by migrant co-habitation arrangements. On an individual level, social networks enable population movement by reducing the cost of migration through benefits such as established shared housing and employment networks, thus making migration a more attractive option to pursue. On a community level, social networks that facilitate migration can have a cumulative effect in sending municipalities to perpetuate and build upon migrant flow between origin and destination sites [37]. Because of the social nature of these community relationships to kinship, friendships and working relationships, migration can be highly localized to movement between specific neighborhoods in sending and receiving communities.
Short-term movement, as Skeldon (2003) [38] points out, is less likely to be measured through census surveys, thus monitoring population movement should be undertaken at the municipality level and integrated into larger databases to establish early warning systems.
Exposure to an index patient has been identified as the primary determinant of leprosy infection among their contacts. The magnitude of the effect of contact in our study was highest among close family contact – parent, child, and/or siblings - followed by consanguineous and/or non-consanguineous household contact and lastly other contact, which could include social and distant family exposure. The possibility of genetic susceptibility to leprosy infection, through close family kinship has been significantly associated with leprosy among contacts [14]–[15], [39] which supports our findings of leprosy association with close kinship among past 5-year migrants. At the household level, other research has shown that proximity to and intensity of exposure to leprosy increases the risk of transmission, as much as five to nine times that of non-household contacts [14]–[15], [39]–[42], although leprosy clustering among neighboring residences in areas of high population density and poverty has social contact risk similar to household contacts [43]. Contact with multibacillary diagnosis in the household has also been associated with increased risk [14]–[15], [41]–[42] and indicates late diagnosis and long-term exposure to contacts. As the majority of migrants in our sample were diagnosed with multibacillary leprosy, this has significant implications for transmission and also for leprosy associated complications and disability.
Migration was significantly associated with leprosy in our logistic regression models controlling for household and close family contact independently. The independent association with household and close consanguineous exposure could indicate some relationship to familial social networks in migrant destination sites. This, in addition to intensity of exposure due to high household density during migration, suggests both the genetic relationships and social environment surrounding migration may figure prominently in explaining leprosy diagnosis.
The majority of individuals in contact with an index patient are not susceptible to the disease. As such, Sales et al. (2011) [14] suggest that leprosy surveillance should explore multiple factors that may contribute to the risk for infection. While many behavioral, demographic, and socio-environmental variables were included in the analysis, we found socioeconomic status and past five year alcohol consumption among migrants with leprosy were significantly associated with leprosy in comparison to clinically unapparent migrants in the control group. Brazil has one of the highest alcohol-attributable disability-adjusted life years (DALYs) in the world. According to the World Health Organization, there is evidence for an association between alcohol consumption and infectious disease [44]. Current alcohol use however was not significant. This may be the result of recently diagnosed migrants abstaining from alcohol use due to multi-drug therapy treatment. A substantial concern, however was that nearly one in five migrants with leprosy were currently drinking alcohol, which has been associated with leprosy relapse in Brazil [12]. Alcohol consumption can interact with medication absorption [45] and could render leprosy treatment less effective. This can contribute to the elevation of risk for transmission to exposed contacts.
Low socioeconomic status was additionally associated with leprosy among past 5-year migrants. Other research in Cluster 1 also found poverty associated with migration prior to diagnosis among those with leprosy (unpublished data). While poverty is ubiquitously associated with leprosy throughout the literature, it should be noted that these results were taken after the migration period and thus may not be an adequate measure of socioeconomic level during migration. Low socioeconomic status among migrants with leprosy may be linked to restricted employment as the result of disability due to leprosy, or difficulty in sustaining employment during treatment. Despite this, family illiteracy and inaccessibility to public waste collection, proxies for low socioeconomic status in Brazil, were significantly higher for migrants with leprosy compared to the control group. Socioeconomic status, the primary social determinant of health, should be the topic of further investigation both during and after migration.
Leprosy was found to be associated with past five year migration, even after controlling for confounders. In the comparison of past 5-year migrants, leprosy was associated with both household consanguineous and/or non-consanguineous contact, close family and other social leprosy contact, consistent with research identifying contact exposure as the major determinant of leprosy transmission [14]–[15]. However, the magnitude of effect for leprosy among migrants in our study was most significant among close family and household contacts. As migration in Maranhão was largely facilitated through family networks, contact surveillance should include migration site residence contacts as well as current residence contacts.
While patterns of migration, including movement in and between highly endemic clusters, were not different among migrants with leprosy and clinically unapparent migrants in the control group, important facets of migration emerged that could benefit leprosy control at the state and national level. State control programs should consider monitoring past five year residence among those newly diagnosed with leprosy to identify intra- and inter-state migration flow. This may provide early warning systems for localized disease control in areas yet to be identified as high-risk areas.
Alcohol consumption in the years prior to diagnosis may be associated with susceptibility to leprosy. Alcohol consumption and consumption frequency should be included in future investigations. This research will help to determine the extent that alcohol consumption plays a role in the dynamics of both transmission and expression of leprosy. As alcohol consumption has also been associated with leprosy relapse, further attention should be given to alcohol consumption during treatment, patient relapse and contact exposure to leprosy. Other substances should also be given attention in future research.
Other research in Brazil has found a spatial relationship to migration and distribution of leprosy and an association of leprosy with poor socio-economic conditions [4]–[6]. Our research shows that in endemic areas leprosy is not only associated with population movement itself, but, most importantly with the social conditions of the migrant in the endemic areas, their behavior, and contact with leprosy in the family and household.
|
10.1371/journal.ppat.1008022 | Exposure to opposing temperature extremes causes comparable effects on Cardinium density but contrasting effects on Cardinium-induced cytoplasmic incompatibility | Terrestrial arthropods, including insects, commonly harbor maternally inherited intracellular symbionts that confer benefits to the host or manipulate host reproduction to favor infected female progeny. These symbionts may be especially vulnerable to thermal stress, potentially leading to destabilization of the symbiosis and imposing costs to the host. For example, increased temperatures can reduce the density of a common reproductive manipulator, Wolbachia, and the strength of its crossing incompatibility (cytoplasmic incompatibility, or CI) phenotype. Another manipulative symbiont, Cardinium hertigii, infects ~ 6–10% of Arthropods, and also can induce CI, but there is little homology between the molecular mechanisms of CI induced by Cardinium and Wolbachia. Here we investigated whether temperature disrupts the CI phenotype of Cardinium in a parasitic wasp host, Encarsia suzannae. We examined the effects of both warm (32°C day/ 29°C night) and cool (20°C day/ 17°C night) temperatures on Cardinium CI and found that both types of temperature stress modified aspects of this symbiosis. Warm temperatures reduced symbiont density, pupal developmental time, vertical transmission rate, and the strength of both CI modification and rescue. Cool temperatures also reduced symbiont density, however this resulted in stronger CI, likely due to cool temperatures prolonging the host pupal stage. The opposing effects of cool and warm-mediated reductions in symbiont density on the resulting CI phenotype indicates that CI strength may be independent of density in this system. Temperature stress also modified the CI phenotype only if it occurred during the pupal stage, highlighting the likely importance of this stage for CI induction in this symbiosis.
| Insects often harbor heritable symbiotic bacteria that infect their cells and/or bodily fluids. These heritable bacteria are passed from mother to offspring and can have substantial effects on host insect biology, and include bacteria like Cardinium that cause mating incompatibilities between symbiont-infected and uninfected individuals. Often, the extent of these symbiont-conferred modifications correlates with the bacterial density in the host. The appearance of these phenotypes is also affected by temperature stress, which often reduces bacterial density. However, here we find that temperature-altered strength of Cardinium-induced mating incompatibility in a whitefly parasitoid wasp can be independent of Cardinium density. While heat treatment reduced the symbiont density and the phenotype, as expected, cold treatment also reduced symbiont density but increased the degree of mating incompatibility. Here, the prolonged duration of the host pupal development in the cold treatments appeared to be more important for phenotype strength. These results suggest that the connection between bacterial density and phenotype strength may not be as general as previously thought. Furthermore, the modification of this manipulative phenotype has implications for the effectiveness of the host, Encarsia suzannae, as a biological control agent.
| Temperature has profound effects on the biology of ectothermic organisms as well as on the microbiota inhabiting these organisms [1–4]. The heritable symbionts of arthropods, which are primarily vertically transmitted from mother to offspring, are closely integrated with host biology [5], and may be particularly vulnerable to temperature stress [4, 6]. These symbionts rely on strategies, ranging from mutualistic to parasitic, to modify host phenotype in ways that increase the likelihood of their maternal transmission [5]. While transmission of these symbionts and their phenotypic effects are usually stable under benign conditions, both aspects may be threatened as the environment becomes less hospitable [4]. Extreme temperatures can reduce symbiont density [7, 8, 9], induce phenotypic failure [7–10], and/or reduce rates of vertical transmission, resulting in uninfected offspring [8, 9]. From the host’s perspective, the limited thermotolerance of the symbiont compared to the host is a vulnerability, since long-term associations are likely to make hosts metabolically or reproductively dependent on these symbionts, even those that originally acted as reproductive parasites [11–14]. In these cases, temperature-induced disruption of these symbioses can be disastrous for the host, reducing host fitness by rendering hosts vulnerable to natural enemies [10, 15] or causing reproductive failure [13, 16, 17].
Temperature extremes thus pose a clear challenge to heritable symbioses, including those involving one of the best-studied symbionts, Wolbachia, a diverse genus of widespread bacteria (Phylum Alphaproteobacteria). Wolbachia typically acts as a reproductive parasite of hosts using one of four major strategies, the most common of which is cytoplasmic incompatibility (CI)[18]. CI is a mating incompatibility between symbiont-infected males and uninfected females, resulting in inviable offspring. Offspring viability is restored (or “rescued”) in crosses with infected females [18]. CI increases infected female fitness at the cost of uninfected female fitness, thus driving the symbiont infection through a population [19–21]. Some Wolbachia strains also improve host resistance to viruses and macroparasites, including several important human pathogens, instigating the use of this symbiont as a means of controlling important human diseases, including dengue fever, zika, and chikungunya [22–24].
Multiple studies using different host species and Wolbachia strains have found that Wolbachia’s CI phenotype can be modified by temperature [7, 8, 25]. These changes to CI appear to be generally mediated by changes in Wolbachia density, which in turn influences CI strength [26]. Temperature stress reduces Wolbachia density, thereby weakening CI [8, 25]. Temperature-induced Wolbachia reduction also leads to partial failure of the rescue phenotype and reduced efficacy of vertical transmission [8, 27]. From these studies, it seems that temperature stress in some habitats could limit the ability of some Wolbachia strains to maintain high levels of infection or spread through novel populations, limiting both the natural spread of this symbiont as well as efforts to use this symbiont in disease vector control [8].
While there is well over three decades of research studying the effects of temperature on Wolbachia symbioses [28–30], we know very little about how temperature effects a widespread and unrelated bacterium, Cardinium hertigii (Phylum Bacteroidetes; [31, 32]), that also induces CI in its hosts [33–37]. A CI-inducing Cardinium strain infects the parasitoid wasp, Encarsia suzannae (Hymenoptera: Aphelinidae, = Encarsia pergandiella; [33, 38]). Genome and transcriptome comparisons between this CI-Cardinium strain from E. suzannae, cEper1, and CI-Wolbachia strains found little homology between the two bacterial genomes, indicating that CI arose independently in these distantly related symbionts [39, 40]. Despite the two bacteria using different molecular mechanisms to induce CI, the resulting phenotype is strikingly similar; both CI mechanisms ultimately kill embryos by inhibiting proper chromosome pairing in early mitotic cycles, leading to chromatin bridging between daughter cells, eventually triggering widespread aneuploidy and embryonic mortality [41, 42].
Although Cardinium symbioses present an alternative study system for the study of CI, research concerning the impact of environmental factors upon this manipulative symbiont is limited to a handful of studies investigating Cardinium presence or absence at various temperatures. One such study found that Cardinium prevalence was reduced in natural populations of Culicoides biting midges (Diptera: Ceratopogidae) occupying arid regions, offering evidence for a destabilizing effect on the symbiosis attributed to extreme temperature fluctuations in those regions [43]. More recently, an analysis of global Cardinium infection frequencies found increased Cardinium prevalence in insect host populations occupying warmer regions (mean temperature of ~25°C) compared to cooler ones [44]. Other studies using laboratory cultures of the spider mite, Metaseiulus occidentalis, found that mite cultures held at 34°C for >1 year were either uninfected with Cardinium or had a reduced amount of the symbiont compared to cultures maintained at room temperature [45, 46]. Together, these studies provide some evidence for a general destabilizing effect of extreme temperatures on Cardinium-host symbioses; yet we still do not know if temperature modifies other aspects of Cardinium symbioses, particularly the CI phenotype.
Here we examined how temperature stress modifies this manipulative symbiosis. We tested the effects of prolonged exposure, including one or more host stages, to either high or low temperatures on both Cardinium CI modification (in males) and rescue (in females; Fig 1). We next tested whether observed temperature-induced changes to Cardinium’s CI modification machinery and rescue capabilities were due to changes in symbiont density. We also tested the effect of temperature stress on the duration of host pupal development, a host factor that appeared to correlate with Cardinium’s CI strength in a previous study [47]. We next used temperature shock treatments during specific life stages to test whether temperature’s influence on CI was life-stage dependent. Finally, we tested the effect of temperature stress on vertical transmission rates of this heritable symbiont.
Similar to previous studies of this system, we found that at the constant temperature of 27°C, the CI cross resulted in ~70% offspring mortality (Fig 2; [33, 47–49]). We also found that prolonged exposure of male hosts to either warm (32°C day/29°C night; 16:8 h) or cool (20°C day/17°C night; 16:8 h) temperatures caused dramatically different levels of CI-induced offspring mortality compared to males reared at 27°C (Fig 2). Prolonged exposure to warm temperatures, beginning either when the host was a larva or pupa, resulted in a significant decrease in CI-induced offspring mortality (Fig 2; S1 Table; Logistic regression F1,32larva = 7.15, p = 0.01; F1,31pupa = 7.49, p = 0.01). Interestingly, exposure to cool temperatures beginning in the pupal stage had an opposite effect on CI strength and resulted in an increase in CI-induced mortality (F1,38 = 9.98, p = 0.003). In this treatment, the CI phenotype became not only stronger but also less variable than CI typical of this Cardinium strain (Fig 2). However, we found that neither warm nor cool temperatures affected CI strength when the exposure period only included the adult stage (Fig 2; F1,37cool-adult = 0.04, p = 0.85; F1,30warm-adult = 0.85, p = 0.36). We found no effect of temperature on offspring viability in control (“N”) crosses, in which no CI-induced mortality is expected, showing that changes in offspring mortality in warm and cool treatments are not due to a general effect on male viability (Fig 2; Logistic regression F1,40cold-larva = 0.26, p = 0.62; F1,39cold-pupa = 2.31, p = 0.14; F1,38cold-adult = 0.08, p = 0.77; F1,31warm-larva = 0.43, p = 0.52; F1,30warm-pupa = 0.0002, p = 0.99; F1,29warm-adult = 2.56, p = 0.12).
When we tested the effectiveness of the ability of Cardinium in females to rescue Cardinium-modified sperm, the 27°C control treatment showed comparable offspring survival between the rescue cross (“R”; infected mother x infected father) and the control cross (“N”; infected mother x uninfected father), indicating that cEper1 reliably rescues CI when mating individuals bear the same Cardinium strain (Fig 3; [33, 47]). We found that exposure of females to warm temperatures (32°C day/29°C night; 16:8 h) beginning during the pupal and adult stages decreased offspring survival in both R and N crosses compared to those same crosses at 27°C (Fig 3, S1 Table; Logistic regression F1,54pupa-N = 11.78, p = 0.001; F1,53adult-N = 18.26, p = <0.0001; F1,53pupa-R = 19.02, p = <0.0001; F1,52adult-R = 15.74, p = 0.0002). The heat-exposed control (N) crosses showed lower offspring survival rates than the 27°C controls, which suggests that heat reduced the viability of the experimental females independent of the CI phenotype. To account for this general reduction in viability, we subsequently compared offspring survival between the rescue (R) cross and its corresponding control (N) cross for each heat treatment. We then found an additional significant reduction in offspring survival in the R cross compared with its corresponding N cross when females were exposed to heat starting in the larval stage (Fig 3; F1,28 = 20.13, p = 0.0001). These results suggest a partial failure of the rescue phenotype with heat, although pupal and adult exposure to heat did not significantly influence rescue beyond the general reduction in viability (F1,28pupa = 2.85, p = 0.1; F1,25adult = 0.11, p = 0.74). Cool temperatures (20°C day/17°C night; 16:8 h) did not decrease offspring survival in N crosses, suggesting no effect of cold on the viability of females (F1,52larva = 2.53, p = 0.12; F1,51pupa = 0.89, p = 0.35; F1,50adult = 0.54, p = 0.47). For this reason, we compared cold R rescue crosses to the R cross of females kept at the 27°C control temperature (Fig 3). Like the heat treatment, the longest cold exposure resulted in reduced offspring survival in R crosses, indicative of partial rescue failure (Fig 3; F1,56 = 10.65, p = 0.002), but neither exposure at pupal or adult stages reduced rescue significantly (F1,55pupa = 3.21, p = 0.08; F1,54adult = 0.25, p = 0.62).
We tested the effect of temperature stress on Cardinium density in male and female hosts. We found that Cardinium reached variable densities in adult males at the control rearing temperature of 27°C, with densities ranging from 0.25 Cardinium: host cells to about 2:1 Cardinium:host cell (Fig 4a). Both warm and cool temperatures reduced Cardinium density in the male larval and pupal treatments (Fig 4a, S2 Table; Mann-Whitney U-test pwarm-larva = 0.009; pwarm-pupa = 0.04; pcool-larva = 0.004; pcool-pupa = 0.04), although we found no significant changes to Cardinium density in either of the male adult treatments, which also did not show evidence of temperature-modified CI strength (Figs 2 and 4a). While warm and cool treatments both suppressed Cardinium density in male hosts, these temperature treatments showed contrasting effects on CI strength. Furthermore, the Cardinium density of individuals across treatments did not correlate with CI strength (Spearmen’s rho = 0.05, p = 0.72).
In female wasps reared at 27°C, Cardinium density was less variable than in males, with females harboring about 2.5 Cardinium: 1 host cell (Fig 4b). Temperature treatments all significantly reduced Cardinium density in females relative to those kept at 27°C, regardless of temperature range or exposure period (Fig 4b, S3 Table; Mann-Whitney U-test p = 0.012 for all comparisons to control).
Because cool temperatures simultaneously reduced Cardinium density yet resulted in significantly stronger CI, we looked for another factor that could be responsible for the cold-modified CI phenotype. We found that cool temperatures significantly prolonged the duration of the pupal stage from ~6 days at 27°C to ~14 days (Fig 5, S4 Table; Mann-Whitney U-test p = <0.0001). Additionally, we found that warm temperatures significantly reduced pupal developmental time (from ~6 days to ~ 4 days) when those exposures began in the larval or pupal stage (Fig 5, S4 Table; p = <0.01). The adult treatments underwent their temperature exposure after the pupal stage, and we found no difference in pupal duration between adult-exposed wasps and control wasps kept at 27°C (Fig 5). As in [47], we also found that the duration of the pupal stage positively correlated with CI strength (Fig 6; Spearman’s rho = 0.57, p = <0.0001).
We designed a second experiment using temperature-shocks of equal duration to test whether temperature effects on Cardinium’s CI were dependent on host life stage. We found that heat shock (40°C for 2 hours) during the pupal, but not the adult stage, reduced offspring mortality in CI crosses (Fig 7, S1 Table; F1,24pupa = 5.48 p = 0.03; F1,23adult = 0.04, p = 0.84). Unlike our previous experiment (Fig 2), we found no difference in CI-induced mortality between the control and either the pupal or adult cold-shock (4°C for 2 hours) treatments (Fig 7, S1 Table; F1,25pupa = 1.76, p = 0.2; F1,24adult = 0.23, p = 0.64). Perhaps notably, cold-shock during the pupal stage did not prolong pupal development (S1 Fig). While Cardinium density was slightly reduced in males that experienced shocks during the pupal stage, this difference was not significantly different compared to control densities (S2 Fig).
Prolonged exposure to high and low temperatures resulted in a reduction of the ability of Cardinium to rescue its CI (Fig 3) and a reduction of the density in female hosts (Fig 4b). This rescue failure could have resulted from vertical transmission failure, which would generate uninfected eggs susceptible to CI. To test this hypothesis, we exposed female hosts to one of three temperature treatments that started during their larval stage: a warm (32°C day/29°C night; 16:8 h), cool (20°C day/17°C night; 16:8 h), and intermediate control (27°C) treatment, and then examined their offspring for Cardinium infection. We found that vertical transmission was perfect (100% of offspring were infected) at the 27°C control temperature and the cool temperature treatment (Table 1). However, heat treated wasps showed a significant reduction in the number of infected offspring, which translated to a 90% transmission rate (Table 1; Exact binomial test p = <0.0001).
We tested the effect of temperature stress on a symbiosis between a minute parasitoid wasp, E. suzannae, and its CI-inducing heritable symbiont, Cardinium. Male hosts exposed to warm temperatures during the larval or pupal life stages showed faster pupal development, as well as reduced Cardinium density and reduced expression of the CI phenotype relative to hosts kept at a benign temperature. On the other hand, exposure to cool temperatures beginning during the pupal stage strengthened CI expression, despite also reducing Cardinium density. This stronger CI likely resulted from the prolonged pupal stage of cold-treated males, as cold shock treatments that did not prolong the pupal stage or modify Cardinium density also did not result in stronger CI. We also found that lifetime exposure to both warm and cool temperatures reduced Cardinium density in female hosts. Although both temperature ranges reduced Cardinium density in female hosts, heat exposure weakened the CI rescue phenotype more than cold. Why heat exposure caused more rescue failure is unclear, although the decreased vertical transmission rates we observed from heat-treated females to their offspring could account for additional mortality in these rescue crosses.
The symbiosis between Cardinium and E. suzannae, like other heritable symbioses, is vulnerable to environmental stress [4]. Specifically, warm temperatures can weaken the CI phenotype of this manipulative symbiont, with lifetime exposure to higher temperatures almost completely nullifying the phenotype. Like most symbionts, Cardinium infection is expected to carry a fitness cost to infected females under neutral conditions, and, like other symbionts, Cardinium is expected to rely on its CI phenotype to compensate for maintenance costs, allowing it to successfully invade new host populations when above the unstable equilibrium threshold [19–21, 48, 49]. With its CI phenotype weakened, the cost of harboring Cardinium could prevent its spread into regions experiencing prolonged bouts of warmer temperatures.
Warm temperatures also caused Cardinium transmission rates to drop from essentially perfect (100% of offspring infected) to a 90% infection rate. Such a decrease, coupled with modification and rescue failure, may be enough to destabilize the Cardinium symbiosis and cause a drop in Cardinium frequency in an infected population. A similar scenario occurs with Wolbachia that infect Trichogramma wasps, in which the Wolbachia falls in frequency during warm periods and re-invades populations in cooler months [50]. However, given that symbiont strains can show substantial variation in their susceptibility to temperature [8, 10], we caution against applying our results concerning cEper1 (the Cardinium strain used in this study) to other strains of Cardinium. If a strain of Cardinium is thermally sensitive, like cEper1, it may have dynamic infection frequencies that vary seasonally, although this should be investigated further using population-level studies and seasonal field collections.
Our study highlights the importance of the male pupal stage for this Cardinium strain’s CI modification step, which has mechanistic implications for Cardinium CI modification. Temperature exposure only affected CI strength if it occurred throughout the pupal stage, suggesting that this symbiont is actively modifying male sperm as the testes develop during the pupal stage. Furthermore, E. suzannae adult males emerge with more than enough mature sperm for ~14 successive matings in a laboratory setting (MRD, personal observation). During spermatogenesis the cytological contents, including any heritable symbionts, are dumped, leaving mature sperm bacteria-free [51, 52]. Given that male E. suzannae emerge with most of the sperm they will use in their lifetime, and that these sperm likely don’t contain Cardinium, we suspect that the majority of Cardinium’s active modification of male chromatin occurs in the pupal stage. Indeed, adult males of many parasitoid species emerge with a majority of their lifetime sperm, and the pupal stage is potentially of prime importance for the modification step of CI symbionts in these hosts [7, 53]. Future gene expression studies on Encarsia–Cardinium symbioses should focus on the pupal stage to catch expression of all possible bacterial genes involved in CI modification.
The strength of Wolbachia CI is generally dependent on its within-host density [8, 21, 26, 50, 52]. Reduction of Wolbachia density, whether by temperature or some other means, results in loss or weakening of the CI phenotype [8, 25, 54]. Wolbachia also harbors a bacteriophage, WO, which may further modify Wolbachia density and CI strength in response to extreme temperatures [7]. While we found that Cardinium density is also reduced by warm and cold temperature stress, this density reduction did not result in consistent changes to the CI phenotype. Instead, the duration of the pupal stage of male hosts seems to have a larger impact on this symbiont’s phenotype. This can be seen most clearly in the case of cold exposure, which prolonged the pupal stage and increased CI mortality. A similar scenario occurs in Wolbachia-infected Nasonia vitripennis, a species of parasitic wasp, that also shows prolonged pupal development and stronger CI after long-term exposure to cool temperatures. However, this Wolbachia strain harbors the WO phage and it’s possible that increased phage densities may also be responsible for the stronger CI phenotype [7]. It’s not known whether Cardinium must be present within a spermatid in order to modify it or whether the CI factors employed by Cardinium diffuse across membranes of developing sperm, as has been shown for Wolbachia [55]; either case is compatible with a time-limited mechanism for modifying sperm during the pupal stage. Extension of the pupal stage may grant time for Cardinium to modify a greater proportion of sperm and generate stronger CI. This contrasts with what is observed in Drosophila melanogaster-Wolbachia symbioses, in which slower developing males exhibit weaker CI through an unknown mechanism that is also independent of density [56]. Time could be a limiting factor for modification steps in CI symbioses which occur in host organisms that produce most of their lifetime sperm prior to adult emergence, like many parasitic wasps [7, 53, 57], but not for hosts that continuously mature sperm, like Drosophila flies [56, 58].
One caveat to the apparent lack of dependence of CI strength on density in the current study is that qPCR, our method and the standard method for estimating symbiont density, does not give any indication of symbiont health. It’s possible that while the warm and cool treatments equally depressed Cardinium density, the heat treatment may have damaged Cardinium or denatured proteins and other effector molecules involved in expression of CI, while the cold treatment only repressed Cardinium replication. Nevertheless, while density may be a reliable indicator of phenotypic strength for some symbionts, more examples are emerging in which phenotype strength and symbiont density are decoupled, and this rule may not be as general as originally thought [7, 9, 10, 56, 59]. In the case of low-density symbionts, like the CI-Cardinium strains infecting Encarsia wasps and citrus thrips (Pezothrips kellyanus) [37] or some CI-Wolbachia strains [59], other factors may be of equal or greater importance than symbiont density. These include host factors, like developmental time in this study, or the density of extrachromosomal elements harbored by many heritable bacteria [7]. These phages or plasmids can harbor genes responsible for the symbiont-conferred phenotype [60–63], replicate independently of the host bacterium [7, 10], and may respond differently from their host bacterium to environmental stress [7, 10]. In the case of Wolbachia, temperature stress inversely modifies the density of Wolbachia and its WO phage, which harbors the genes required for CI expression [7, 62]. The cEper1 strain of Cardinium does not have a phage but does harbor a plasmid [39]. The role of the plasmid in Cardinium symbioses and temperature stress is not yet clear.
With a rapidly changing climate, the susceptibility of heritable symbionts to temperature stress confers similar vulnerability to the insect hosts that often depend on them, including hosts of economic or medical importance. Here, we show that a symbiosis involving Cardinium and its wasp host is modified by temperature. Whether this has cascading effects upon the host population remains to be seen. However, Encarsia spp. are used as widespread biological control agents for agricultural pests like whiteflies and armored scales [64–68]. Given the potential for temperature to disrupt these symbioses, further investigations into how environmentally susceptible Cardinium-Encarsia symbioses impact wasp and whitefly populations are warranted.
Encarsia suzannae (= E. pergandiella [38]) is a species of autoparasitic wasp (Hymenoptera: Aphelinidae). Like all Hymenoptera, these wasps are haplodiploid; females develop from fertilized, diploid eggs and males develop from haploid, unfertilized eggs. In autoparasitic wasps, females lay diploid eggs within whitefly nymphs, where they develop as solitary primary parasitoids. Male eggs are laid within developing aphelinid whitefly parasitoids enclosed within the remnant whitefly cuticle, and male E. suzannae develop obligately as hyperparasitoids [69]. Laboratory cultures of female E. suzannae were maintained on Bemisia tabaci (sweet potato whitefly) grown on Vigna unguiculata (cowpea); male wasps were maintained by exposing pupae of a second parasitoid species, Eretmocerus emiratus, to virgin E. suzannae females. Wasp cultures were kept at 27°C and 16 D: 8 N photoperiod at ambient humidity unless otherwise noted.
The E. suzannae culture of this study was collected from the Rio Grande Valley of Texas, USA [33]. This wasp naturally harbors a CI-inducing strain of Cardinium hertigii, cEper1 [33, 39, 40]. Additionally, an uninfected wasp culture was established by feeding adults antibiotic-spiked honey, allowing for mating crosses between uninfected C(-) and cEper1 infected C(+) individuals [33]. Uninfected individuals used in this study were from a culture that had been antibiotically-cured at least 10 generations previously. The Er. emiratus used for generating male E. suzannae were not infected with Cardinium.
Male C(+) wasps were generated by placing several (~5–7) virgin C(+) E. suzannae in a 35mm Petri dish with 25 Er. emiratus pupae on moist filter paper along with a small amount of honey. We observed females parasitize for 3 hours, removing parasitized Er. emiratus with a fine paint brush to limit superparasitism (multiple eggs oviposited in a single host). Parasitized Er. emiratus (i.e. developing male E. suzannae) were placed into individual 1.2 ml vials. Then, using an experimental design modified from [7], we tested the effect of temperature stress on Cardinium’s ability to induce CI in crosses between C(+) males and C(-) females by exposing these developing males to a variety of temperature treatments (Fig 1). One subset of males was kept at 27°C in a Percival incubator throughout development to serve as a control group. Other males were kept in Percival incubators set to either a “warm” or “cool” temperature range of 32°C day/ 29°C night or 20°C day/ 17°C night, respectively. We monitored internal incubator temperatures with HOBO data loggers. These ranges, particularly the “warm” temperatures, reflect the temperature extremes at which E. suzannae will tolerate constant exposure in the laboratory (MRD, personal observation). Temperature treatments were further split into three groups that began at either the larval 1st instar (4-day old wasps), pupal (7-day old), or adult (upon emergence) stage [70]. All temperature treatments continued until two days post-adult emergence. This resulted in a total of seven temperature treatments that differed in their temperature range as well as the host life stages that they included (Fig 1). As cEper1’s CI phenotype was previously found to positively correlate with male pupal developmental time [47], we also recorded the duration of the pupal stage for each individual male.
Two days after male adult emergence, males were mated either with virgin C(+) females for the control cross (“N”), to test for temperature-induced effects on general male viability, or with virgin C(-) females for the inviable “CI” cross to test for the effect of temperature on the strength of Cardinium’s CI phenotype. Mating was visually confirmed by observing male and female wasps confined in a 1.2 ml vial plugged with cotton for ~5 minutes or until mating. Mating typically occurred < 1 min after bringing the pair together, but in cases where mating had not yet occurred after 5 minutes, the female was removed, discarded, and replaced. After mating, males were stored at -80°C and females were transferred to individual arenas to parasitize whiteflies. The arenas consisted of a 35mm Petri dish with a ventilated lid, containing a cowpea leaf disc resting on 1% agar. The cowpea leaf disc bore ~50 2nd– 3rd instar B. tabaci nymphs. Females were allowed to parasitize whiteflies in these arenas for 24hrs and the dish was then incubated at 27°C. Ten days later, we counted the number of whiteflies that either contained a developing E. suzannae pupa (resulting from a successful cross) or were developmentally arrested (i.e. parasitized by E. suzannae but the wasp failed to develop–a reliable proxy for the dead wasp embryos indicative of CI [33]). We then compared the proportion of arrested whiteflies across treatments using logistic regression with a quasibinomial distribution. We compared male pupal developmental time using the non-parametric Kruskal-Wallis test and post hoc multiple pairwise Mann-Whitney U Tests with Benjamini-Hochberg-corrected p-values in R v 3.3.1 [71]. We also compared male pupal developmental time and CI strength using Spearmen’s rho correlation.
To test the effect of temperature stress of Cardinium’s ability to rescue its CI phenotype, we first generated female C(+) E. suzannae by allowing mated C+ female E. suzannae to parasitize leaf dishes with 2nd-3rd instar whiteflies for 24hrs. Leaf discs with developing female wasps were then exposed to one of seven temperature treatments in Percival incubators as previously described. Two-days after adult emergence, experimental females were mated with either C(-) males (N; control cross) to test for temperature-induced effects on female viability, or C(+) males (R; rescue cross) that developed at 27°C to test for effects on Cardinium’s rescue phenotype. Here a decrease in offspring survival relative to the control would indicate a reduction in the rescue ability of Cardinium in the temperature treated female host. After mating observations, females were transferred to individual parasitism arenas and given 24hrs to parasitize whitefly hosts. Females were then removed and stored at -80°C. Ten days later, we counted the number of surviving wasp pupae and arrested whiteflies and analyzed rescue efficiency across temperature treatments using logistic regression with a quasibinomial distribution, as in the CI experiment.
After the crossing experiments, we next tested whether changes in Cardinium’s CI modification and rescue phenotypes were due to changes in the within-host density of this symbiont. Experimental wasps (both male and female) had been stored at -80°C for DNA extraction. Extractions consisted of homogenizing a single wasp in 3 μL of 20 mg ml -1 proteinase k, then adding the homogenate to 50 μL of 5–10% w/v Chelex [72]. Samples were incubated at 37°C for 1 hour with periodic vortexing, followed by incubation at 97°C for 8 min and storage at -20°C. We estimated Cardinium density relative to host cells by performing quantitative PCR (qPCR) using Maxima SYBR Green/ROX qPCR Master Mix (2×) (ThermoFisher Scientific) with primers for the single-copy Cardinium gyrB gene [47] and the host Ef1a gene on a Bio-Rad CFX Connect Real-Time cycler [73]. We also created standards via serial dilutions of PCR products from both primer sets. PCR products were diluted to a concentration of 1.0 ng/μL DNA, which was confirmed with a Qubit 4.0 fluorometer prior to serial dilution. Samples were run in triplicate and each qPCR plate included standards for both primer sets to correct for between-plate differences in reaction efficiency. Raw Cq values were averaged and corrected before conversion to relative density [47].
We then compared Cardinium density across temperature treatments for male and female wasps using the Kruskal-Wallis non-parametric test with post hoc multiple pairwise Mann-Whitney U Tests with Benjamini-Hochberg-corrected p-values. We also compared Cardinium density and CI strength using Spearmen’s rho correlation. All statistical analyses were performed in R v. 3.3.1 [71].
The experiments described in the current study tested the effect of prolonged temperature stress on the Cardinium CI phenotype and thus included one or more host life stages and different durations of temperature exposure. To test for life stage-specific effects of temperature stress on the CI phenotype, we performed temperature “shocks” of uniform duration on male C(+) E. suzannae, during either their pupal stage or upon adult emergence. Temperature shocks consisted of a two-hour ramping step, where temperatures increased from 27°C to 40°C (heat shock) or decreased from 27°C to 4°C (cold shock), followed by a two-hour period at 40°C or 4°C, and then a two-hour ramp back to 27°C in a Percival incubator. After the temperature shock, males were maintained at 27°C until mating two days after emergence. As in the previous CI experiment, males were kept in individual 1.2 ml vials with a small amount of honey throughout development. Mating and parasitism were otherwise performed and analyzed as in the first experiments, and males were kept at -80°C for DNA extractions and Cardinium density estimates using qPCR.
In addition to modifying symbiont-induced phenotype and/or density, temperature stress can destabilize symbiont vertical transmission [8]. Inefficient or “leaky” vertical transmission can slow the spread or cause the loss of symbionts within a population. We tested the effect of temperature stress on Cardinium vertical transmission by exposing C(+) female E. suzannae to one of three temperature treatments: a 27°C control, 32°Cday/29°Cnight (warm), or 20°Cday/17°Cnight (cool). Female wasps were generated as previously described and kept at 27°C until they were 4-days old (1st instar larvae), when they were moved to their respective temperature treatments. Exposure continued through their development until two-days post adult emergence, when the females were mated to C(-) males, a control cross that should generate viable progeny regardless of female infection status. After mating, females were placed on arenas with whitefly nymphs and given 24hrs to parasitize hosts. These offspring were kept at 27°C until adult emergence, at which point they were frozen at -80°C. Offspring DNA was extracted and tested for the presence of Cardinium DNA using Cardinium-specific 16S rRNA primers [49]. PCRs included a positive and negative control and were run on a 1% agarose gel with SYBR Safe to visualize the product. Negative samples were re-run with the Cardinium primers. We next confirmed the presence of wasp DNA in negative samples using general CO1 primers [74]. Negative samples that failed to show DNA bands with the CO1 primers were discarded from the final analysis. We then tested whether the vertical transmission rates were significantly lower than 1.0 (perfect transmission) using the Exact Binomial test with 95% confidence intervals.
|
10.1371/journal.ppat.1002784 | The Rhoptry Proteins ROP18 and ROP5 Mediate Toxoplasma gondii Evasion of the Murine, But Not the Human, Interferon-Gamma Response | The obligate intracellular parasite Toxoplasma gondii secretes effector proteins into the host cell that manipulate the immune response allowing it to establish a chronic infection. Crosses between the types I, II and III strains, which are prevalent in North America and Europe, have identified several secreted effectors that determine strain differences in mouse virulence. The polymorphic rhoptry protein kinase ROP18 was recently shown to determine the difference in virulence between type I and III strains by phosphorylating and inactivating the interferon-γ (IFNγ)-induced immunity-related GTPases (IRGs) that promote killing by disrupting the parasitophorous vacuole membrane (PVM) in murine cells. The polymorphic pseudokinase ROP5 determines strain differences in virulence through an unknown mechanism. Here we report that ROP18 can only inhibit accumulation of the IRGs on the PVM of strains that also express virulent ROP5 alleles. In contrast, specific ROP5 alleles can reduce IRG coating even in the absence of ROP18 expression and can directly interact with one or more IRGs. We further show that the allelic combination of ROP18 and ROP5 also determines IRG evasion and virulence of strains belonging to other lineages besides types I, II and III. However, neither ROP18 nor ROP5 markedly affect survival in IFNγ-activated human cells, which lack the multitude of IRGs present in murine cells. These findings suggest that ROP18 and ROP5 have specifically evolved to block the IRGs and are unlikely to have effects in species that do not have the IRG system, such as humans.
| Toxoplasma gondii can infect any warm-blooded animal and is transmitted orally by consumption of tissue cysts. To facilitate transmission, the parasite must balance induction and evasion of host immune responses to allow parasite growth and persistence, while avoiding excessive parasite burden, which can kill the host before infectious cysts are formed. Different strains of Toxoplasma have likely evolved specific effector molecules to modulate the immune responses of different hosts. In many mammals, including mice but not humans, the cytokine interferon gamma (IFNγ) induces the immunity-related GTPases (IRGs), which are essential to the murine immune response to Toxoplasma. They function by binding to and disrupting the parasite-containing vacuole. However, some Toxoplasma strains prevent the IRGs from disrupting the parasitophorous vacuole. It was previously shown that the secreted Toxoplasma kinase ROP18 promotes virulence in mice by phosphorylating the IRGs, leading to their inactivation. We report that ROP18 requires another virulence factor, the secreted pseudokinase ROP5, to prevent IRG accumulation, and these two proteins determine the majority of strain differences in IRG evasion, even for divergent strains for which virulence determinants have not been studied. Additionally, we show that ROP18 and ROP5 do not affect Toxoplasma survival in IFNγ-stimulated human cells. Thus, ROP18 and ROP5 are strain- and host-specific determinants of Toxoplasma immune evasion.
| Toxoplasma gondii is a widespread intracellular parasite capable of infecting most warm-blooded animals and is an important opportunistic pathogen for immunocompromised individuals and unborn fetuses. Toxoplasma resides within a non-fusogenic parasitophorous vacuole and has three apical secretory organelles, the micronemes, rhoptries and dense granules, which secrete proteins into the host cell during invasion that mediate important host-pathogen interactions [1]. In general, an asymptomatic but chronic infection is established in immunocompetent humans. However, in rare cases Toxoplasma can cause severe disease even in immunocompetent people. Diverse disease outcomes may be due to genetic differences between infecting strains [2].
Toxoplasma has a partially clonal population structure of 12–15 [3], [4] haplogroups with the majority of North American and European isolates belonging to the canonical types I, II and III strains [5], [6], although haplogroup 12 has been recently shown to be prevalent in wild animals in North America [6]. In mice, these strains differ in virulence, with type I strains having an LD100 of just one parasite, compared to the LD50 of ∼103 or ∼105 parasites for types II and III strains, respectively [7], [8]. Type I strains may also be more virulent in humans, as they are more frequently isolated from cases of congenital or severe ocular toxoplasmosis than from animals [5], [9]. Interestingly, in South America, more genetically diverse strains are isolated, while the canonical strains are rarely found [10]. Some of these strains are associated with high mortality rates in mice [11]. Additionally, there are high rates of ocular toxoplasmosis in humans in South America [12], [13], and some strains isolated from French Guiana have been reported to cause severe disseminated toxoplasmosis even in healthy individuals [14]. The determinants of canonical strain-specific differences in murine virulence are well studied, but the same determinants for non-canonical strains or for human infection remain unknown.
Mice and humans use divergent immune mechanisms to resist Toxoplasma. Interferon-γ (IFNγ) is essential to murine Toxoplasma resistance, and IFNγ-deficient mice die after infection even with avirulent strains [15]. Some of the important downstream effectors of this immune activation are the IFNγ-inducible immunity-related GTPases (IRGs), which belong to the dynamin family of GTPases and can cooperatively oligomerize to vesiculate membranes. Mice deficient in individual members of the IRG family die of toxoplasmosis, but at different stages of infection, and expression of the IRGs is required even in non-hematopoietic cells, suggesting IRGs have non-redundant, crucial roles in the innate immune response against Toxoplasma [16]–[18]. Different IRGs are sequentially and cooperatively loaded onto the parasitophorous vacuole membrane (PVM) with Irgb6 and Irgb10 initiating and stabilizing the loading of the other members [19]. The IRGs are able to disrupt the PVM and kill the parasite [20], [21].
While mice have 23 IRG genes, humans have only two IRG genes: IRGC which is expressed only in the testis and IRGM which is expressed independently of IFNγ induction and has a truncation in the nucleotide-binding G-domain [22]. Despite these differences, IRGM plays a role in autophagy-mediated destruction of Mycobacterium tuberculosis and Salmonella typhimurium in human cells, and some variants are associated with increased risk for Crohn's disease [23], [24]. Thus, IRGM may have an immune role, but its lack of GTPase activity suggests a distinct mechanism of action in humans. Humans do have other known IFNγ-mediated mechanisms of resistance to Toxoplasma. For instance, IFNγ-induced indoleamine 2,3-dioxygenase (IDO1) degrades cellular tryptophan for which Toxoplasma is auxotrophic, thereby inhibiting Toxoplasma growth [25], [26]. The NALP1 inflammasome also mediates the innate immune response to Toxoplasma, and NALP1 was recently identified as a susceptibility locus for human congenital toxoplasmosis [27].
Toxoplasma strain differences in evasion of murine immune responses exist. For instance, type I strains are able to prevent the accumulation of IRGs on the PVM, while types II and III strains are susceptible to killing by the IRGs even when co-infecting the same cell as a type I parasite [28]. Because strain-specific evasion of the IRGs is correlated with increased virulence in the mouse, it is likely that the genetic determinants of IRG evasion will also be associated with virulence. Quantitative Trait Locus (QTL) mapping analyses of the virulence of F1 progeny derived from type I×II, I×III and II×III crosses have identified the genetic loci associated with virulence, and subsequent experiments have identified the causative genes within these loci.
ROP18, a highly polymorphic rhoptry protein kinase, was identified as a virulence locus in the II×III QTL study and the only virulence locus in the I×III cross [7], [29]. ROP18 is highly expressed in types I and II strains but an insertion in the promoter prevents expression in type III strains. Addition of a type I or II copy of ROP18 into an avirulent type III strain makes that strain become virulent [7], [11]. Recently, it was shown that type I ROP18 can phosphorylate a conserved threonine in the G-domain of Irga6 and Irgb6, disrupting their accumulation on the PVM [30], [31]. However, type II strains have the highest percentage of vacuoles coated with IRGs [19], [28] despite the fact that a type II copy of ROP18 is also able to make a type III strain virulent, suggesting that other polymorphic proteins are involved in IRG evasion [7]. ROP18 was also shown to promote the degradation of the endoplasmic reticulum-associated transcription factor ATF6-β, compromising CD8 T cell-mediated adaptive immune responses [32]. Importantly, ROP18-mediated ATF6-β degradation occurs in human as well as murine cells.
The ROP5 locus, which consists of a family of 4–10 tandem duplicates of highly polymorphic genes encoding for rhoptry pseudokinases that localize to the PVM, is another important virulence determinant in mice [33], [34]. Deletion of ROP5 in a type I strain significantly attenuates virulence. Furthermore, ROP5 was the only significant virulence locus identified in the recent I×II QTL analysis and was the main virulence locus in the II×III QTL study [7], [34]. Both types I and III strains have a virulent ROP5 locus, but the mechanism by which ROP5 affects virulence and which of the three major ROP5 isoforms, A, B or C, [33] are necessary to complement the virulence of type II are not known.
A third virulence locus, identified in the II×III QTL study, contains the rhoptry protein kinase ROP16, which in types I and III strains leads to sustained phosphorylation and activation of STAT3/6 [35]. It was recently shown that ROP16 and the dense granule protein GRA15, suggested to be the fourth virulence locus in the II×III QTL study [36], affect the accumulation of p65 guanylate binding proteins (GBPs) on the PVM in infected murine cells [37]. Because GBPs are also dynamin family members and were found on the same vacuoles as the IRGs, ROP16 and GRA15 might also affect the accumulation of the IRGs on the PVM. Furthermore, since the GBPs are present in humans, ROP16 and GRA15 could possibly affect survival in IFNγ-stimulated human cells.
Because the murine and human immune responses to Toxoplasma are so different, it cannot be assumed that ROP18, ROP5, ROP16 and GRA15, which determine Toxoplasma virulence in mice, similarly affect survival in human cells. Furthermore, it is currently unknown for most of these proteins what effects they have outside the clonal lineages from which they were identified. Many of the exotic strains are highly virulent in mice, but because they are so divergent from the canonical strains and the exotic strains have not been used in QTL or gene manipulation studies, it is not known what factors drive virulence in these strains. For example, IRG evasion has not been measured for the exotic strains, and it may be that this is strictly a type I phenotype.
In this study, we find that ROP18 can only inhibit accumulation of the IRGs on the PVM of strains that also express virulent ROP5 alleles. Expression of ROP18 in strains that do not express virulent ROP5 alleles does not affect IRG accumulation or in vivo virulence. In contrast, specific ROP5 alleles can reduce IRG coating even in the absence of ROP18 expression and directly interact with Irga6 to inhibit its oligomerization. Non-canonical strains exhibit differences in evasion of IRG-mediated killing as well, and the allelic combination of ROP18 and ROP5 also correlates with strain differences in IRG evasion and virulence for these strains. However, neither ROP18 nor ROP5 markedly affect parasite survival in IFNγ-activated human cells.
Type II strains have the highest percentage of IRG-coated vacuoles compared to types I and III strains [19], [28] even though they possess a ROP18 allele capable of conferring virulence to a type III strain [7]. To determine if, like ROP18I [30], [31], the increased virulence due to ROP18II is correlated with reduced IRG coating in a type III background, we measured the percentage of vacuoles coated with Irgb6 by immunofluorescence in IFNγ-stimulated mouse embryonic fibroblasts (MEFs) infected with type I, II, III, III + ROP18I, or III + ROP18II (Figure 1A). Indeed, transgenic expression in the type III strain CEP of either ROP18I or ROP18II decreased the average number of vacuoles coated with Irgb6 from 45% to 23% (P = 0.001) for ROP18I or 29% (P = 0.003) for ROP18II (Figure 1B). Although it is generally assumed that once the PVM is coated, it will eventually lead to killing of the parasite inside, it has also been shown that Toxoplasma can escape a coated vacuole and invade a new cell [37], [38]. Therefore, to measure killing of Toxoplasma, 100 parasites were seeded on a monolayer of MEFs, either previously stimulated for 24 hours with IFNγ or left untreated, and the number of plaques that form after 4–7 days of growth was determined. Type III had an average of 45% plaque loss when comparing plaques formed on IFNγ-stimulated MEFs to unstimulated MEFs. This percentage plaque loss was similar to the percentage of vacuoles coated with Irgb6, suggesting that coated vacuoles are eventually destroyed. Furthermore, plaque loss is drastically reduced in Atg7 deficient MEFs (Figure S1) in which the IRGs are misregulated as previously reported for Atg5 deficient MEFs [19], [39], suggesting the killing observed is indeed due to the IRGs. Similar to the decrease in Irgb6 coating, the plaque loss of type III + ROP18I or ROP18II was significantly decreased to 18% (P = 0.0002) and 21% (P = 0.0004), respectively (Figure 1B). The 23% PVM coating and 18% killing of type III + ROP18I is similar to the 25% coating and 35% plaque loss of the type I strain GT1. Thus, ROP18 expression can likely explain most of the difference in IRG coating and killing between type I and type III strains. Despite the ability of ROP18II to reduce IRG coating of type III strain vacuoles and subsequent killing of the parasite, type II strains are still very susceptible to the IRGs, with 70% Irgb6 coating and 73% plaque loss for Pru (type II) (Figure 1B). Thus, there must be at least one other gene involved in IRG evasion that is shared between types I and III but different in type II.
It was recently demonstrated that the ROP5 cluster of pseudokinases accounts for most of the variation in virulence between types I and II strains and between types II and III strains, with types I and III strains possessing a virulent ROP5 locus [33], [34]. Therefore, the ROP5 locus is an excellent candidate for explaining strain differences in IRG evasion. We tested a potential role of ROP5 in mediating ROP18-independent strain differences in IRG evasion by using the S22 strain, an avirulent F1 progeny from a II×III cross [40] which possesses the avirulent ROP18III and ROP5II alleles. We compared the percentage plaque loss and percentage of Irgb6 coated vacuoles between S22 and an S22 transgenic strain carrying the cosmid LC37, which contains the ROP5 locus from the RH (type I) genome and was previously shown to have significantly increased virulence [33]. Expression of ROP5I significantly reduced the Irgb6 coating from 48% to 28% (P<0.001), and the plaque loss from 38% to 27% (n.s.) (Figure 2A). Thus, ROP5I can function independently of ROP18I/II to prevent IRG accumulation on the PVM and subsequent killing of the parasite.
While ROP5 can function independently of ROP18 in reducing IRG accumulation on the PVM of S22 + LC37 vacuoles, type II strains, which have a virulent allele of ROP18 and an avirulent ROP5 locus, have a high percentage of IRG-coated vacuoles. This suggests that either ROP18 cannot function independently of ROP5, or that ROP18 is inhibited in the type II background. We expressed ROP18II in S22 and in S22 + LC37 to determine if ROP18II can function in the absence of virulent ROP5 alleles. ROP18II only slightly reduced Irgb6 coating in S22 from 47% to 41% (n.s.) and plaque loss from 39% to 24% (n.s.). However, ROP18II significantly reduced Irgb6 coating from 31% to 7% (P<0.001) and plaque loss from 27% to 9% (P<0.01) when expressed in S22 + LC37 (Figure 2A). Together, this suggests that ROP18 needs the virulent ROP5 locus for its function. That the Irgb6 coating and plaque loss in S22 + LC37 + ROP18II are similar to those in RH (type I) signifies that these two genes are sufficient to complement IRG evasion and plaque loss in the S22 background.
To determine if the interactive effect of ROP18 and ROP5 on parasite survival also occurs in vivo, we infected outbred CD-1 mice by intraperitoneal injection with S22, S22 + ROP18II, S22 + LC37 or S22 + LC37 + ROP18II tachyzoites expressing firefly luciferase and followed parasite growth and dissemination using in vivo imaging. On the third day after infection, the parasite burden in S22 + LC37 and S22 + LC37 + ROP18II-infected mice was 10-fold higher than in S22 or S22 + ROP18II-infected mice. By day six, both strains containing the LC37 cosmid had disseminated throughout the peritoneal cavity, but S22 + LC37 + ROP18II-infected mice had 35-fold higher luciferase activity than S22 + LC37-infected mice (P = 0.03), which in turn had 10-fold higher activity than S22 + ROP18II-infected mice (P = 0.1) and 30-fold higher activity than S22-infected mice (P = 0.06). While S22 + ROP18II had a greater parasite burden than S22, this was not significant (P = 0.27). Indeed, S22 + LC37 + ROP18II killed 100% of the mice in the acute stage of infection at both a low and high dose (Figure 2B and C). Likewise, in keeping with the increased IRG evasion of S22 + LC37 but not S22 + ROP18II, S22 + LC37 showed increased virulence compared to S22, but S22 + ROP18II-infected mice survived the infection and did not show significant differences compared to S22 infected mice (Figure 2D). Thus, overall these results suggest that ROP18 only affects virulence in the context of a virulent ROP5 locus.
Although mouse virulence has been determined for many non-canonical strains [11], it is unknown what factors determine virulence in these strains. We wondered if virulent non-canonical strains could also evade IRG-mediated killing, or if IRG evasion is specific to type I strains. We measured the percentage plaque loss in IFNγ-stimulated MEFs as well as percentage Irgb6-coated vacuoles for strains from haplogroups 1–11 [6], [41]. In general, IRG evasion correlates with virulence as strains that have a mortality rate of greater than 90% in CD-1 mice also have 25% or less Irgb6-coated vacuoles and plaque loss (Figure 3A). However, some exceptions are CASTELLS and COUGAR, which exhibit greater than 50% Irgb6 coating and plaque loss in IFNγ-stimulated MEFs, despite a high mortality rate in mice [11]. These strains may have a different mechanism underlying their virulence in mice besides IRG evasion.
For most strains, the Irgb6 coating and plaque loss correlates with their ROP18 allele (Figures 3A and S2) [11]. For example, CASTELLS and P89, as well as the type III strains CEP and VEG, have between 40% and 50% Irgb6 coating, and all of these strains do not express ROP18 because they have a ROP18III-like allele that contains an insertion in the promoter [11]. The strains that express a type I-like allele of ROP18, with the exception of BOF, display 25% or less Irgb6 coating. Type II strains and COUGAR are highly susceptible to the IRGs with 70% and 53% Irgb6 coating respectively, despite having the virulent ROP18II allele. For type II strains, the avirulent ROP5 locus likely explains the high degree of Irgb6 coating, but it is unknown what versions of ROP5 are present in the non-canonical strains.
For most of the strains mentioned above, Irgb6 coating correlates with their ROP18 allele, suggesting that they also have a virulent ROP5 locus, as this is necessary for ROP18 to function (Figure 2). It is currently unknown what determines the virulence and IRG evasion properties of the ROP5I/III locus because both copy number and amino acid sequence of the individual copies differ between the canonical strains [33]. To identify differences that may be associated with virulence or IRG evasion, we sequenced the different ROP5 isoforms of strains from haplogroups 1–11 (GenBank JQ743705-JQ743783). Based on the Toxoplasma genome sequence (www.ToxoDb.org) and our own genome sequencing of seven non-canonical Toxoplasma strains (Minot et al., submitted), we identified four distinct ROP5 open reading frames that we amplified and sequenced separately using isoform specific primers. Sequence chromatograms indicated that two or more alleles were present for the second ROP5 reading frame. We therefore cloned the PCR product from this ROP5 gene and sequenced multiple clones to obtain sequences from the different alleles, but some alleles may still be missing. Sequences from this expanded paralog matched what has previously been called both ROP5-B (minor) and C (major) genes (Figure 3B) [33], [34]. We could not differentiate B and C alleles for all strains if they were not similar to the canonical strains, so we refer to them here as B copies. We determined that besides the three major ROP5 copies that were previously described, 2 other highly divergent ROP5 isoforms exist that we call ROP5L-A and ROP5L-B (Figures S3 and S4). Interestingly, ROP5L-A and ROP5L-B are highly conserved between strains, but we find that these isoforms are not expressed in tachyzoites (Figure S3) so they will not be discussed further. The previously described ROP5 genes (A, B and C) [33] are highly divergent with strong evidence for diversifying selection (Figure 3C)., especially in surface exposed residues in the kinase domain [42]
In general, for ROP5-A and for ROP5-B and C, which cluster together, alleles can be divided into distinct groups with the BOF, P89, CAST and GPHT strains grouping with the virulent types I and III alleles (Figure 3B). A second allelic group consists of the strains VAND, RUB, GUY-KOE, GUY-DOS and GUY-MAT. The ability to confer virulence of this allelic group is unknown but because these strains are all highly virulent [11] and able to evade the IRGs, these alleles are likely virulent. A third very divergent group of alleles contains the strains MAS, CASTELLS and TgCatBr5, but there is less diversity in the ROP5-A, B and C isoforms present in these strains. The COUGAR allele is most similar to but divergent from the second group, but interestingly, COUGAR has only one B/C allele. The avirulent ROP5 locus from type II is also divergent, and a phylogenetic analysis of all ROP5 alleles indicates that the type II ROP5-B and C genes are more closely related to ROP5-A than to ROP5-B or C of the other strains. These results suggest that ROP5-B and/or C could be important for IRG evasion and virulence since type II strains and COUGAR have high levels of IRG coating (Figure 3A) and seem to have either ROP5 alleles that are all ROP5-A-like (type II) or are missing ROP5-C (COUGAR) (Figure 3B).
Next, we tested whether differences in ROP5 expression or copy number could account for strain differences in IRG evasion. For example, BOF has virulent ROP18 and ROP5 alleles but is highly coated by Irgb6 (Figure 3A–B). To estimate copy number differences between the strains we have sequenced, we plotted the sequencing coverage of the ROP5 locus versus the average genome coverage, as this was previously shown to be a good estimate for copy number [43]. Most of the strains had about twice as many reads at ROP5-A and B as the rest of the genome, while MAS and TgCatBr5 have 3–5 copies of each gene (Figure 3D). However, coincident with our inability to amplify ROP5-A, we found that BOF is missing ROP5-A and has only one copy of ROP5-B.
We also looked at ROP5 expression levels determined using RNA-Seq data from 24 hour infections of murine bone marrow derived macrophages with each strain (Figure 3D). BOF has barely detectable expression of ROP5-B and no expression of ROP5-A, likely explaining its high Irgb6 coating despite having a similar ROP5-B/C amino acid sequence to types I and III. Indeed BOF + LC37 has virtually no Irgb6 coating (0.33%) compared to BOF (40% Irgb6 coating, P = 0.001) (Figure 4A). ROP5 expression levels can also likely explain many intra-haplogroup strain differences where ROP18 and ROP5 coding sequence are the same; for example, VEG has higher ROP5 expression levels compared to the other type III strain CEP, and VEG has slightly reduced IRG coating compared to CEP. Thus, higher ROP5 expression is correlated with reduced IRG coating, suggesting a non-enzymatic, dose-dependent role for ROP5 in IRG evasion.
Because the LC37 cosmid that reduced Irgb6 coating and plaque loss in S22 and BOF contains ROP5-A, B and C it is unknown which of these isoforms (or which combination) is important for IRG evasion. However, the fact that type II ROP5 alleles are less divergent and more similar to ROP5-A suggests type II is missing ROP5-B and C. Additionally, ROP5-C was previously described as the major allele with A and B as minor alleles when trace reads were assembled for the ROP5 coding region of types I, II and III [34]. Therefore, we tested if ROP5-AIII, ROP5-CIII or LC37, which contains the entire ROP5 locus, could complement IRG evasion in the type II background. Although some of the effects we see in the type II background will be due to an interaction with ROP18, because ROP18 is present in all backgrounds, we can still compare the effects of individual ROP5 genes. We find, as expected, that expression of ROP5-AIII in the type II strain Pru led to only a slight but significant reduction in Irgb6 coating (51%, P<0.05), but expression of ROP5-CIII in Pru led to a significant reduction of IRG coating (36%, P<0.001) similar to that of Pru + LC37 (38%, P<0.001) compared to a heterologous control (62%) (Figure 4B). The 36% IRG coated vacuoles in Pru + ROP5-CIII is comparable to the 25% IRG coated vacuoles for GT1, suggesting that the lack of ROP5-C may account for the excessive IRG accumulation on type II vacuoles.
To see if ROP5-CIII can also increase the survival of type II parasites in vivo, we infected CD-1 mice with Pru, Pru + ROP5-AIII, Pru + ROP5-CIII or Pru + LC37. The growth and dissemination of Pru and Pru + LC37 was determined by in vivo imaging of luciferase activity. On the third day post infection, Pru + LC37-infected mice had twice the parasite burden of Pru-infected mice (Figure 4C and D). By day six, there was 50 fold higher luciferase activity in Pru + LC37-infected mice (P<0.0001), and the parasites had disseminated throughout the peritoneal cavity. Indeed, 100% of Pru + LC37-infected mice died within 11 days of infection even at the lowest dose (Figure 4E). Mice infected with Pru parasites expressing only ROP5-AIII or ROP5-CIII survived the infection (Figure 4E) but Pru + ROP5-CIII-infected mice had more ruffled fur and lost significantly more weight (Figure 4F) than Pru-infected mice throughout the course of infection(P = 0.01 at 15 days post infection) while Pru + ROP5-AIII-infected mice continued to gain weight. Together, these results suggest that while expression of ROP5-CIII can reduce Irgb6 coating of type II parasites, ROP5-CIII only partially enhances the survival of type II parasites in vivo, and the whole ROP5 locus is required to significantly increase virulence in mice.
It is not clear how ROP5 inhibits IRG accumulation at the PVM, but other pseudokinases have been shown to serve as protein scaffolds or to regulate the activity of kinases [44]. Since ROP18 requires ROP5 for fully efficient IRG evasion, and there is an interactive effect of adding ROP18 and ROP5 to the S22 strain, it is possible that ROP5 and ROP18 interact directly. To test this hypothesis, we immunoprecipitated ROP5 and ROP18II-HA from MEFs infected with CEP or CEP + ROP18II-HA for one hour with or without previous IFNγ stimulation. We were unable to detect by western blot co-immunoprecipitation of ROP18 and ROP5 (Figure 5A). Furthermore, when recombinant, tagged ROP18 kinase domain (Lim et al., submitted) is added to cell lysates from IFNγ-stimulated or unstimulated MEFs infected for one hour with Pru + ROP5-CIIIHA, and ROP5 is immunoprecipitated with anti-HA, we do not co-immunoprecipitate ROP18 (Figure S5A) indicating that there is no direct interaction between ROP5-CIII and the ROP18 kinase domain. Next we tested the hypothesis that ROP18 is only active in the presence of virulent ROP5 alleles by immunoprecipitating ROP18II-HA from MEFs infected with S22, S22 + ROP18IIHA, and S22 + LC37 + ROP18IIHA for one hour with or without previous IFNγ stimulation for use in an in vitro kinase assay. We found that there was no difference in the activity of ROP18 immunoprecipitated from parasites with or without a virulent ROP5, as measured by the phosphorylation of an optimized substrate (Lim et al., submitted) in vitro, (Figures 5B and S5B). This established that ROP18 was active in all backgrounds and indicated that there are no irreversible effects of ROP5 on ROP18 kinase activity.
Because ROP5 does not directly interact with or irreversibly affect ROP18 kinase activity, we next tested the hypothesis that ROP5 directly interacts with one or more IRGs. We immunoprecipitated HA-tagged proteins from IFNγ-stimulated or untreated MEFs infected for one hour with Pru, Pru + ROP5-AIII-HA, Pru + ROP5-CIII-HA, or RH + GRA15II-HA and lysed in the presence or absence of GTPγS (a non-hydrolyzable form of GTP). Co-immunoprecipitated proteins were separated by SDS-PAGE and identified by mass-spectrometry. We did not recover any ROP18 peptides, again suggesting that ROP5 does not directly interact with ROP18. We did, however, recover 13 peptides (38% sequence coverage) from Irga6 only in the Pru + ROP5-CIII-HA infected samples lysed in the presence of GTPγS (Figure 5C) suggesting a specific interaction between ROP5-C and Irga6 because the other HA-tagged, PVM associated proteins did not co-immunoprecipitate Irga6 under these conditions. Under different buffer conditions and in the absence of GTPγS, we also recovered 4 peptides of Irga6 and 2 peptides (9.8% sequence coverage) of Irgb10 only in the Pru + ROP5-CIII-HA infected samples (data not shown). Because ROP5 lacks kinase activity [42] but reduces IRG localization to the PVM, we wondered if Irga6 binding by ROP5 could inhibit Irga6 oligomerization, which is necessary for its activity. To test this hypothesis, we measured the GTP-mediated oligomerization of recombinant Irga6 by dynamic light scattering in the presence of recombinant maltose binding protein (MBP)-tagged ROP5 or MBP alone. We found the predicted hydrodynamic radius of Irga6 to be reduced in the presence of ROP5 but not MBP (Figure 5D). Thus, we find that ROP5-CIII binds and inhibits the oligomerization of at least one IRG.
It was recently reported that p65 guanylate-binding proteins (GBPs), members of the dynamin superfamily that includes the IRGs, also accumulate on the Toxoplasma PVM alongside the IRGs [37]. Because ROP16 and GRA15 were shown to affect GBP coating, we were interested to see if ROP16 and GRA15 also affect IRG coating. We measured the effect of ROP16 and GRA15 on IRG coating and IRG-mediated killing in types I, II and III genetic backgrounds. In a type I background, the deletion of ROP16, the transgenic expression of GRA15II, or both in combination did not significantly alter IRG coating or killing (Figure 6A and not shown). Likewise, type IIΔgra15, type II transgenically expressing ROP16I, and type III transgenically expressing GRA15II showed no statistical differences in Irgb6 coating or plaque loss compared to their parental strains. Thus, while these genes may affect GBP coating, they do not significantly alter Irgb6 accumulation.
Not all of the F1 progeny in the I×II cross that have the type I ROP5 are as virulent as type I in mice [34] indicating that there are other genes besides ROP5 and ROP18 that affect virulence. While the genetic location of the dense granule protein GRA2 has not been verified as a QTL affecting virulence, an RHΔgra2 strain is one of the few type I knockouts that have reduced mouse virulence [45]. While the reason for this reduced virulence is unknown, it is known that GRA2 functions in the formation of the tubulovesicular network in the Toxoplasma PVM [46], which creates negative curvature in the PVM that might help to attract Toxoplasma proteins secreted into the host cell back to the PVM [47]. Indeed, it has been shown that the RHΔgra2 strain has reduced ROP18 localization to the tubulovesicular network in the Toxoplasma PVM [47]. We therefore hypothesized that this GRA2-dependent ROP18 and ROP5 localization and/or localization of other proteins, would be important for IRG evasion. Indeed, the RHΔgra2 strain has significantly increased IRG coating to 36% (P<0.001) and increased plaque loss on IFNγ-stimulated MEFs to 24% (P = 0.08) (Figure 6B). Therefore, a protein required for the formation of the PVM structure also affects IRG accumulation.
We wondered if there are strain differences in the survival of Toxoplasma in IFNγ-stimulated human cells since strain differences in virulence have been primarily studied in mice, and human cells lack the multitude of IRGs present in murine cells. We measured the percentage plaque loss of different types I, II and III strains as well as of non-canonical strains in human foreskin fibroblasts (HFFs) pre-stimulated for 24 hours with IFNγ (Figure 7A). In general, the percentage plaque loss in IFNγ-stimulated HFFs is higher than in IFNγ-stimulated MEFs. The type I strains RH and GT1 have plaque losses of 54% and 63%, respectively, while the type II strains ME49 and Pru have plaque losses of 73% and 96%, respectively and the type III strains CEP and VEG have plaque losses of 90% and 67%, respectively. The non-canonical strains range in plaque loss from 47% (GUY-DOS) to 67% (CASTELLS). Thus, strain susceptibility to IFNγ-mediated killing in human cells does not correlate with that of murine cells.
As we have shown, ROP18 and ROP5 are responsible for most of the strain differences in IFNγ-susceptibility in murine cells, but Toxoplasma IFNγ-susceptibility in murine cells does not correlate with IFNγ-susceptibility in human cells. To test if ROP18 affects IFNγ-mediated killing in human cells, we first examined type III strains transgenically expressing a virulent copy of ROP18. Neither ROP18I nor ROP18II expression in type III decreases the percentage plaque loss compared to the parental strain (Figure 7B), suggesting that ROP18 is not responsible for strain differences in IFNγ-mediated killing in human cells. To see if ROP5 affects survival in IFNγ-activated human cells, we compared the percent plaque loss in IFNγ-stimulated HFFs between S22 and S22 + LC37 and between Pru and Pru + LC37. The plaque loss decreases from 87% for S22 to 76% for S22 + LC37 (P = 0.03) and from 96% in Pru to 88% for Pru + LC37 (P = 0.01) (Figure 7C). Although the differences in plaque loss due to expression of ROP5 are significant, the differences are minimal (±10%). Thus, virulent alleles of ROP18 and ROP5 do not largely affect parasite survival in IFNγ-stimulated human cells.
We report that the precise allelic combination of the Toxoplasma polymorphic ROP18 and ROP5 genes determines Toxoplasma strain differences in susceptibility to killing by IFNγ-stimulated MEFs, even for non-canonical strains. We also show that ROP18 and ROP5 function by inhibiting the accumulation of and subsequent killing by the IRGs. Toxoplasma strains also differ in their susceptibility to killing by IFNγ-stimulated HFFs, but this is not determined by ROP18 or ROP5.
Previous studies on the role of ROP18 in mediating strain differences in IRG accumulation on the PVM have produced inconsistent results. Initial studies of in vivo primed macrophages infected with the type III strain CTG expressing ROP18I and L929 cells expressing ROP18I infected with the type II strain ME49 showed minimal effects of ROP18 on Irgb6 and Irga6 coating [19], [28]. More recently, ROP18I was shown to phosphorylate a conserved threonine in the switch 1 loop of the GTPase domain of Irga6 and Irgb6 leading to their subsequent inactivation [30], [31]. Here, we report that both ROP18I and ROP18II can prevent the accumulation of IRGs on the PVM but only when expressed in a genetic background that contains the virulent ROP5 locus. The lack of virulent ROP5 in type II strains therefore likely explains why L929 expression of ROP18I did not affect IRG accumulation on type II vacuoles in those cells [19].
Previously it was shown that the avirulent strain S22 transgenic for the cosmid LC37, containing ROP5, had slightly fewer Irgb6 coated vacuoles (∼72%) than wild type S22 (87%) in IFNγ-stimulated MEFs, but growth inhibition as measured by uracil uptake was not affected [19]. In contrast, we see a significant decrease in the percentage of vacuoles coated with Irgb6 and increased parasite survival when comparing S22 + LC37 with S22. This could be due to the concentration of IFNγ, the exact assay used or the genotype of the host cells used, as the IRGs are divergent between mouse strains. We find that LC37 also reduces Irgb6 coating and promotes parasite survival in Pru and BOF, and that ROP5-C can explain most of the reduction in IRG coating in vitro. However, Pru + LC37 was significantly more virulent in mice than Pru + ROP5-CIII suggesting the other ROP5 genes may have additional roles besides IRG evasion, in mouse virulence.
Currently, all Toxoplasma genes that determine strain differences in virulence were identified using pairwise crosses between types I, II and III. Because types I and III are progeny from a cross(es) between type II and a strain named alpha (similar to type VI) and beta (similar to type IX), respectively, these three strains are closely related to each other [48]. In recent years it has become appreciated that in South America, many other highly divergent strains exist, and types I, II and III are rarely isolated. To date, no studies have been done to determine the virulence determinants of these strains. Here we report that for these strains the allelic combination and/or expression level of ROP18 and ROP5 also determine how well these strains evade the accumulation of the IRGs and their virulence in mice. Surprisingly, even though the North American/European and South American strains diverged an estimated one million years ago [41], they all use the same two genes to evade the murine IFNγ response. This suggests that evasion of host IRGs is crucial for Toxoplasma. However, most strains do not completely evade the IRGs as this would be an unsuccessful strategy to ensure transmission in mice as the host would be killed before infectious cysts are formed. This could mean that ROP18/ROP5 allelic combinations of highly virulent strains might have evolved to evade the IRGs of species that are more resistant to Toxoplasma, for example rats [49], and that mice are just an accidental host or it could be an artifact of the mouse lab strains commonly used. Strains such as type II, type III, BOF (VI), P89 (IX) and CASTELLS (IV) that either lack (type II) or do not express (BOF) virulent ROP5 alleles or do not express ROP18 (type III, P89 and CASTELLS) and are therefore less virulent in mice seem better adapted to cause chronic infections in mice. Indeed, the large majority of Toxoplasma isolates in North America and Europe belong to type II [2].
ROP5 reduces IRG coating of the Toxoplasma PVM independently of ROP18 despite a lack of kinase activity [42]. Many pseudokinases have been shown to act as scaffolds or regulators of active kinases [44]. We find that ROP5 is not necessary for ROP18 kinase activity in vitro nor did we find evidence for any direct interactions between ROP5 and ROP18 (Figure 5A). We find instead that ROP5 directly interacts with and inhibits the oligomerization of Irga6 (Figure 5C and D). Expression levels of ROP5 seem to correlate with the intra-haplotype differences in IRG coating between CEP and VEG, supporting a non-enzymatic, dose-dependent inhibition of the IRGs by ROP5. Importantly, both the IRGs and the ROP5 locus have expanded, perhaps due to an evolutionary arms race whereby new host IRG genes required new ROP5 genes so Toxoplasma could continue to evade IFNγ-mediated killing. Although ROP5 can inhibit IRG oligomerization, we see an interactive effect with ROP18 on IRG-coating and virulence. Perhaps the reduced oligomerization of IRGs in the presence of virulent ROP5 alleles is reversible, but this de-oligomerization might provide access for ROP18 to bind and phosphorylate the IRGs on the threonines in their switch I loop, to prevent re-activation. If this model is correct than the interaction of ROP18 with the IRGs [30], [31] should only occur in the presence of virulent ROP5 alleles.
To defend itself against the IRGs Toxoplasma must have evolved a mechanism to ensure appropriate trafficking of ROP18 and ROP5 to the PVM upon their secretion into the host cytoplasm. The N-terminal amphipathic helices (RAH domains) of both proteins are required for efficient localization to the PVM, and it was speculated that their specificity for the PVM versus other membranes might be because of a preference for negative curvature [47]. Indeed, we found that RHΔgra2 parasites that have a disrupted tubulovesicular network [46], which provides much of the negative curvature of the PVM, have increased IRG accumulation. This indicates that the attraction of ROP5, ROP18 and possibly other secreted proteins to the PVM, which is attenuated in RHΔgra2 [47], outweighs the possible attraction the IRGs may have for the negative curvature of the PVM [50]. It is likely that the increased IRG accumulation on the PVM of RHΔgra2 accounts for its decrease in virulence [45].
Because all Toxoplasma strains seem to rely on ROP5 and ROP18 for evasion of the murine IFNγ response, these proteins could be attractive drug targets if they are also involved in evasion of the human IFNγ response. However, we find that although there are significant strain differences in susceptibility to IFNγ-mediated killing by HFFs, ROP5 and ROP18 do not markedly affect survival in those cells. This might not be surprising because humans do not possess the large variety of IRGs of murine cells (23 members) but only a single member (IRGM) that is not regulated by IFNγ [22]. The effector mechanisms induced by IFNγ in human cells that are effective against Toxoplasma include tryptophan degradation [25], iron depletion [51], P2X7-mediated death of the host cell [52] and activation of the NALP1 inflammasome [27]. While the IRGs do not mediate vacuolar destruction in human cells, we wondered if another group of dynamin-related large GTPase, the GBPs, could be involved in IFNγ-mediated killing by HFFs, but we failed to see GBP1 at the PVM in HFFs (data not shown).
The Toxoplasma strains that were most resistant to IFNγ-mediated killing by HFFs have also been shown to be able to cause severe disease even in immunocompetent humans. In future studies, strain differences in survival in IFNγ-activated HFFs may provide insight into that mechanism.
A rat monoclonal antibody against HA (3F10, Roche, 1∶500 dilution), a goat polyclonal antibody against mouse TGTP (A-20, Santa Cruz Biotechnology, 1∶100 dilution), a mouse monoclonal antibody against Toxoplasma surface antigen (SAG)-1 (DG52) [53], and a mouse polyclonal antibody against the N-terminus of ROP5 [54] were used in immunofluorescence assays or immunoprecipitations. Secondary antibodies for immunofluorescence were coupled with Alexa Fluor 488 or Alexa Fluor 594 (Molecular Probes). Secondary antibodies used in Western blotting were conjugated to peroxidase (Kirkegaard & Perry Laboratories). Mouse IFNγ from Peprotech and human IFNγ from AbD serotec were dissolved in DMEM with 10% FBS.
Parasites were maintained in vitro by serial passage on monolayers of human foreskin fibroblasts (HFFs) at 37°C in 5% CO2. The following representatives for each haplotype were used: RH and GT1 for type I, ME49 and Pru for type II, CEP and VEG for type III, MAS and CASTELLS for type IV, GUY-KOE and GUY-MAT for type V, GPHT and BOF for type VI, CAST for type VII, TgCatBr5 for type VIII, P89 for type IX, GUY-DOS and VAND for type X and COUGAR for type XI. A Pru strain engineered to express firefly luciferase and GFP (PruΔHXGPRT A7) [55], a CEP and RH strain engineered to express clickbeetle luciferase and GFP (CEPΔHXGPRT C22 and RH 1-1) [56], CEP + ROP18II, Pru + ROP16I [7], RHΔgra2 [45], RHΔrop16 [57], RH + GRA15II and CEP + GRA15II [36] have been described previously. HFFs were grown as described previously [36]. WT C57BL6/J MEFs were a gift from A. Sinai (University of Kentucky College of Medicine, Lexington, KY), Atg7+/− and Atg7−/− MEFs [58] from Masaaki Komatsu (The Tokyo Metropolitan Institute Medical Science) and all MEFs were grown in HFF media supplemented with 10 mM Hepes. All parasite strains and cell lines were routinely checked for Mycoplasma contamination and it was never detected.
Monolayers of MEF cells grown on coverslips and incubated for 24 hours with or without 1000 U/ml IFNγ. Parasites were allowed to invade for 20 minutes, non-invading parasites were then washed away with PBS 3 times, and the infection proceeded for 1 hour. The cells were then fixed with 3% (v/v) formaldehyde in PBS for 20 minutes at room temperature, permeabilized with 0.2% saponin and blocked in PBS with 3% (w/v) BSA and 5% (v/v) FBS. Percent Irgb6 coating was determined in a blind fashion by finding intracellular parasites and then scoring Irgb6 coating as positive or negative.
The coding sequence for ROP5 from types I (GT1), II (ME49), and III (VEG) was predicted from ToxoDB genomic sequence using ORF Finder (NCBI). ROP5 genomic DNA from additional strains was amplified by PCR with isoform specific primers confirmed by sequence chromatograms. ROP5 was amplified with the following primers forward 5′CGATTCACGCTTTCCATGT′3, reverse 5′TCCTTCAGCGGAAAACAGAT′3 for ROP5-A, forward 5′CATTTCATGCCTTCCCAGTT′3, reverse 5′GCGCTCGAGTACTTGTCCTG′3 for ROP5-B/C, forward 5′GTCCCTGGAAAACTGTTTCG′3, reverse 5′GTGAACAGAGAGCGTCCAA′3 for ROP5-D, forward 5′ATTCTGCAATGCCCAAAAGA′3, reverse 5′TTCATGTTGGATACGGCAAC′3 for ROP5-E and 5′AAAAGGCGCGGCGAGCTAGCGTC′3 as an alternate forward primer for ROP5-A for MAS and CASTELLS.
The ROP5-B/C PCR products produced mixed sequence and therefore the PCR product was cloned and multiple clones were sequenced.
The following primers were used to sequence ROP5-A and ROP5-B 5′ATAGGTAACCGGGACGCTTG′3, 5′CCACTTCGGAAGAGACTTGC′3, 5′GGACAGACGCAGGCTTTTAC′3
The following primers were used to sequence ROP5-D and ROP5-E 5′TGAGCTGAAAACCGACTTCAC′3, 5′GGTGACTGGAACACTCGACA′3, 5′TTTTCCGGACCTTGTCTTTG′3, 5′TTCGGGAGAGACTTGCTCAG′3, 5′GCTGTGACAGTTCCGACTCA′3
Sequences were aligned using ClustalX and Neighbor-Joining phylogenetic trees were made with Molecular Evolutionary Genetic Analysis (MEGA) software version 4.1 with 1000 bootstraps and default settings [59]. The Non-synonymous Analysis Program (SNAP) was used to calculate the proportion of synonymous and non-synonymous changes in coding regions [60].
The coding region and putative promoter (766 bp upstream of the start codon for ROP5-A and 681 bp upstream for ROP5-B) of ROP5-A and ROP5-B was amplified from type III Toxoplasma genomic DNA by PCR (A forward, 5′-CCACGCATTCTTCCACTCAGTACCG-3′; B forward, 5′-CCACAATGGCTACCAGGTCCTGCG-3′; A/B reverse, 5′-CTACGCGTAGTCCGGGACGTCGTACGGGTAAGCGACTGAGGGCGC-3′). The coding region of ROP18, along with putative promoter (742 bp upstream of the ATG start codon), from type I Toxoplasma genomic DNA was amplified by PCR. (Forward 5′-CACCAGATTCGAAACGCGGAAGTA-3′; Reverse 5′-TTACGCGTAGTCCGGGACGTCGTACGGGTATTCTGTGTGGAGATGTTCCTGCTGTTC -3′). These primers amplified these genes specifically as confirmed by sequencing and the sequence matched the previously published data [30]–[34]. Sequence coding for an HA tag was included in the reverse primer (denoted with italics) to C-terminally tag the protein. ROP5-AIIIHA, ROP5-CIIIHA and ROP18I were then cloned into pENTR/D-TOPO (Invitrogen), and then cloned into pTKO-att (described in [58]) by LR recombination (Invitrogen). The pTKO-att-ROP5IIIHA vectors were then linearized by digestion with HindIII (NEB), which does not cut inside either gene. Linearized vector was transfected into PruΔHXGPRT parasites by electroporation as described previously [58]. The pTKO-att-ROP18 vector was linearized by digestion with NdeI (NEB) and transfected into CEPΔHXGPRT C22 parasites by electroporation. Stable integrants were selected in media with 25 mg/ml mycophenolic acid (Axxora) and 25 mg/ml xanthine (Alfa Aesar) and cloned by limiting dilution. To express ROP18II in the S22 and S22 LC37 parasite strains, 35 µg of pTKO-att-ROP18IIHA [1] was linearized by HindIII (NEB) and 1 µg of pTUB-CAT were co-transfected by electroporation. Stable integrants were selected by passage of 106 parasites every 2 days in 2 µM chloramphenicol. Expression of ROP18 and ROP5III was confirmed by IF and Western blot for HA staining (Figure 6A–D). The LC37 cosmid from the pSC/Ble library (gift of M.J. Gubbels, Boston College, Boston, MA) was expressed in PruΔHXGPRT A7 and BOF by transfecting 50 µg cosmid and selecting twice extracellularly for 1.5 hours with 5 µg/ml phleomycin. Integration was confirmed by PCR with the Type I ROP5 specific forward primer (5′-TTTTCCGCAGGCCGTGGC-3′) and ROP5A/B reverse for Pru and amplification of ROP5-A for BOF.
For the plaque assays, 100–300 parasites per well were added to monolayers of MEFs seeded the day before or HFFs seeded two days before and either previously stimulated with 1000 U/ml mouse IFNγ, 100 U/ml human IFNγ or left unstimulated for 24 hours before infection in a 24 well plate in either MEF media or DMEM with 1% FBS for HFFs. Infections were then incubated for 4–7 days at 37°C and the number of plaques was counted using a microscope.
CD-1 (Charles River Laboratories) mice were intraperitoneally (i.p.) infected with 500 or 5000 syringe-lysed tachyzoites in 300 µl PBS using a 28 gauge needle. On days 3, 6 and 12 post infection, parasite burden and dissemination was measured by bioluminescence emission imaging. Mice were injected i.p. with 3 mg firefly D-luciferin (Gold Biotechnology) dissolved in PBS, anesthetized with isofluorane, and imaged with an IVIS Spectrum-bioluminescent and fluorescent imaging system (Xenogen Corporation). Images were processed and analyzed with Living Image software. The MIT Committee on Animal Care approved all protocols. All mice were maintained in specific pathogen-free conditions, in accordance with institutional and federal regulations.
For genomic sequencing, DNA was isolated from freshly lysed Toxoplasma parasites using a Trizol-based extraction (Invitrogen). This DNA was subsequently prepared for high-throughput sequencing according to the Illumina single-end genomic DNA kit protocol (COUGAR, CASTELLS and MAS) and 36 nucleotides of each library was sequenced on an Illumina GAII and processed using the standard Illumina pipeline. Paired-end sequencing Illumina libraries were constructed for the genomic DNA of P89, GUY-KOE, TgCatBr5, BOF using the Nextera Illumina compatible DNA sample prep kit (Epicenter) and amplified with the modified PCR protocols described previously [61]. Sequence reads were aligned to the Toxoplasma and human genomes using the Maq software package [62]. Reference Toxoplasma genomes from a type II (Me49), a type I (GT1) and a type III (VEG) strain were obtained from http://toxodb.org (release 6.3). For RNA sequencing, murine bone-marrow derived macrophages (BMDM) were seeded in 6 well plates at 70% confluency and infected with different strains of Toxoplasma at three multiplicity of infections (MOIs): 15, 10 and 7.5. After 24 hours total RNA was extracted from all infected cells using the Qiagen RNeasy Plus kit. Integrity, size and concentration of RNA was then checked using the Agilent 2100 Bioanalyzer. The RNA was processed for high-throughput sequencing according to standard Illumina protocols. Briefly, after mRNA pull down from total RNA using Dynabeads mRNA Purification Kit (Invitrogen), mRNA was fragmented into 200-400 base pair-long fragments and reverse transcribed to into cDNA, before Illumina sequencing adapters were added to each end. Libraries were barcoded and subject to paired end sequencing on the Illumina HiSeq2000 (40+40 nucleotides) and processed using the standard Illumina pipeline. All libraries were spiked with trace amounts of the phiX bacteriophage for quality control purposes. After sequencing, the samples were de-barcoded to separate reads from the multiplexed samples using a custom Perl script. Reads were assembled into full sequences by mapping to exons and across exon junctions using the organism's genomes as a template. Maq was used to estimate Toxoplasma transcript abundance for ROP5 and ROP18 based on our sequenced alleles. A more detailed analysis of the genome and RNA-seq data will be described elsewhere.
Immunoprecipitations were each performed with 5 µg of rat anti-HA (3F10, Roche) or mouse anti-ROP5 [54] conjugated to 25 µl of protein G dynabead slurry (Life Technologies). The HA antibodies were crosslinked at room temperature with 5 mM Bis(Sulfosuccinimidyl) suberate (BS3) (Pierce) prepared in conjugation buffer (20 mM sodium phosophate, 150 mM sodium chloride, pH 7.5) for 30 minutes and quenched by adding 50 mM Tris-Cl, pH 7.5 for 15 minutes and finally washed with conjugation buffer. For each immunoprecipitation (IP) condition, 4.2×106 MEFs were infected at an MOI of ∼5–10, with the strain and condition indicated. After 1 hour, uninvaded parasites were washed away with PBS and the infected cells were treated with 0.25% trypsin for 5 minutes at 37°C. The cells were quenched and harvested with growth media and subsequently washed with PBS + 1 mM PMSF and lysed for 15 minutes at 4°C with light agitation in 1 ml of IP lysis buffer [50 mM HEPES-KOH (pH 7.5), 300 mM NaCl, 10 mM β-glycerophosphate, 1 mM NaF, 0.1 mM NaVO4, 1 mM PMSF, 1% NP-40, and protease inhibitor cocktail (Roche)]. The lysate was then centrifuged at 16,000 g for 30 minutes at 4°C and the supernatant was collected. For ROP18 binding assays, 1 µg of ROP18 recombinant kinase domain [residues 187–554, fused to a series of N-terminal fusion tags consisting of: (His6)-(glutathione S-transferase)-(maltose binding protein)-(Streptococcus protein B1 domain)-(TEV cleavage site), (Lim, D et al., submitted)] were incubated for 30 minutes before adding conjugated and crosslinked antibody beads described above and agitating them for 3 hours at 4°C. The beads were washed 3 times with IP wash buffer [10 mM HEPES-KOH (pH 7.5), 150 mM NaCl, 20 mM β-glycerophosphate, and 0.5% NP-40], washed 3× with HEPES-buffered saline (HBS) and boiled in sample buffer. The samples were western blotted with anti-GST HRP conjugate (GE Healthcare Life Sciences) and anti-HA (3F10, Roche) antibodies.
Immunoprecipitations for kinase assays were performed as above but with several changes. The cleared lysates were incubated with 10 µg of rat anti-HA (3F10, Roche) per IP reaction and incubated for 90 minutes, washed 5 times with the IP lysis buffer and 3 times with IP wash buffers and HBS (all buffers contained 300 mM NaCl). Half of the beads were boiled in sample buffer for western blotting with anti-HA and the other half used for the kinase assay. Kinase assays using a ROP18 model peptide substrate (NH3-KKKKKWISEHTRYFF-CONH2) (Lim, D. et al., submitted) were conducted at room temperature with a reaction buffer consisting of 50 mM HEPES pH 7.5, 300 mM NaCl, 10 mM DTT, 10 mM MgSO4 and 60 µM cold ATP. Each reaction contained 0.5 mM of peptide substrate and 2 to 10 µCi of 32P-γ-ATP. Reactions were stopped after 30 minutes by spotting on Whatman P81 phospho-cellulose paper, which were then dried and washed with 0.425% phosphoric acid until no significant radioactivity remained in the washes. Radioactivity captured on P81 filters was then quantified by phosphorimage analysis with ImageQuant 5.2 software (Molecular Dynamics). The radioactivity detected was normalized to the amount of protein immunoprecipitated as determined by the above Western blot.
Immunoprecipitations were performed as above with a monolayer of confluent MEFs in a T175 lysed in the presence or absence of 0.5 mM GTPγS (Sigma) and precipitated using 30 µg of HA antibody. The washed beads were boiled in sample buffer and samples were subjected to SDS–PAGE and colloidal coomassie (Invitrogen) staining. For mass spectrometry analysis, proteins were excised from each lane of a coomassie-stained SDS-PAGE gel encompassing the entire molecular weight range. Trypsin digested extracts were analyzed by reversed phase HPLC and a ThermoFisher LTQ linear ion trap mass spectrometer. Peptides were identified from the MS data using SEQUEST algorithms44 that searched a species-specific database generated from NCBI's non-redundant (nr.fasta) database.
Recombinant Irga6 [residues 1–413, fused to a series of N-terminal fusion tags consisting of: (His6)-(glutathione S-transferase)-(maltose binding protein)-(Streptococcus protein B1 domain)-(TEV cleavage site), (Lim et al, submitted)] oligomerization was monitored in 50 mM Tris/5 mM MgCl2/2 mM DTT by dynamic light scattering (DLS). Oligomerization was initiated by the addition of 10 mM GTP (Sigma) to 20 µM Irga6 in the presence or absence of 40 µM recombinant (His6)-MBP-tagged ROP5-CIII or (His6)-MBP. The reaction was mixed by pipetting and immediately transferred to a quartz cuvette and equilibrated to 37°C. DLS was performed using a DynaPro NanoStar Light Scatterer (Wyatt Technologies) with an acquisition time of 10 sec over 35 minutes and analyzed using the DYNAMICS software version 7.1.4. The mean hydrodynamic radius of the population was estimated using the standard curve of molecular weight for globular proteins and is not equal to the actual size of the oligomer.
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 MIT Committee on Animal Care (assurance number A-3125-01) approved all protocols. All mice were maintained in specific pathogen-free conditions, and all efforts were made to minimize suffering.
Sequences can be accessed in GenBank: ROP5-BL sequences JQ743705–JQ743719, ROP5-AL JQ743720–JQ743735, ROP5A–C sequences JQ743736–JQ743783.
|
10.1371/journal.pcbi.0030232 | CATHEDRAL: A Fast and Effective Algorithm to Predict Folds and Domain Boundaries from Multidomain Protein Structures | We present CATHEDRAL, an iterative protocol for determining the location of previously observed protein folds in novel multidomain protein structures. CATHEDRAL builds on the features of a fast secondary-structure–based method (using graph theory) to locate known folds within a multidomain context and a residue-based, double-dynamic programming algorithm, which is used to align members of the target fold groups against the query protein structure to identify the closest relative and assign domain boundaries. To increase the fidelity of the assignments, a support vector machine is used to provide an optimal scoring scheme. Once a domain is verified, it is excised, and the search protocol is repeated in an iterative fashion until all recognisable domains have been identified. We have performed an initial benchmark of CATHEDRAL against other publicly available structure comparison methods using a consensus dataset of domains derived from the CATH and SCOP domain classifications. CATHEDRAL shows superior performance in fold recognition and alignment accuracy when compared with many equivalent methods. If a novel multidomain structure contains a known fold, CATHEDRAL will locate it in 90% of cases, with <1% false positives. For nearly 80% of assigned domains in a manually validated test set, the boundaries were correctly delineated within a tolerance of ten residues. For the remaining cases, previously classified domains were very remotely related to the query chain so that embellishments to the core of the fold caused significant differences in domain sizes and manual refinement of the boundaries was necessary. To put this performance in context, a well-established sequence method based on hidden Markov models was only able to detect 65% of domains, with 33% of the subsequent boundaries assigned within ten residues. Since, on average, 50% of newly determined protein structures contain more than one domain unit, and typically 90% or more of these domains are already classified in CATH, CATHEDRAL will considerably facilitate the automation of protein structure classification.
| Proteins comprise individual folding units known as domains, with a significant proportion containing two or more (multidomain structures). Each domain is thought to represent a unit of evolution and adopts a specific fold. Detecting domains is often the first step in classifying proteins into evolutionary families for studying the relationship between sequence, structure, and function. Automatically identifying domains from structural data is problematic due to the fact that domains vary substantially in their compactness and geometric separation from one another in the whole protein. We present a novel method, CATHEDRAL, which iteratively identifies each domain by comparing a query structure against a library of manually verified domains in the CATH domain database through computational structure comparison. We find that CATHEDRAL is able to outperform the majority of popular structure comparison methods for finding structural relatives. Furthermore, it is able to accurately identify domain boundaries and outperform other methods of structure-based domain prediction for the majority of proteins. CATHEDRAL is available as a Webserver to provide domain annotations for the community and hence aid in structural and functional characterisation of newly solved protein structures.
| Proteins comprise individual folding units known as domains, with a significant proportion containing two or more units (multidomain structures) [1]. Each domain adopts a specific fold, and it is estimated that there are up to several thousand such folds in nature [2–4]. As the domain is thought to be an important evolutionarily conserved unit, several structural classifications, such as SCOP [5] and CATH [6], have sought to cluster them into fold groups and evolutionary families. Although a given pair of structures in these families can diverge below similarities of ≤30% in their sequence, these relatives still maintain a comparable topology or fold in the core of the structure [6,7].
More than 7,000 new proteins structures were deposited in the Protein Data Bank (PDB; http://www.rcsb.org/pdb) [8] in 2005. Furthermore, analysis of version 2.6 (April 2005) of the CATH database shows that nearly 50% of classified structures are multidomain. Although many are close in sequence to previously solved structures, the structural genomics initiatives have concentrated their resources on proteins with low sequence similarity to existing structures. As a consequence, they often require considerable manual analysis to be classified in the CATH domain database. That said, the vast majority of newly solved structures contain previously observed folds, although they are often quite remote homologues. In this situation, structural comparison algorithms can be essential to facilitate the automatic and semiautomatic classification of domains.
The number of larger multidomain structures has been gradually increasing since the formation of the PDB, with improvements in techniques for structure determination. We can expect this trend to continue, as recent analyses of completed genomes have suggested that the proportion of multidomain structures in some organisms, particularly eukaryotes, may be as high as 80% [1]. Figure 1 shows that the majority of multidomain chains comprise two domains, although some structures have been solved with three, four, five, and six domains.
A further complication is that approximately 20% of domains from multidomain proteins in the PDB are discontiguous [9] (Figure 2); that is, the structure of the individual domains is formed from disconnected regions of the polypeptide chain. Both automated and manual approaches to domain boundary recognition have problems in assigning these domains.
Various computational methods have been developed to automatically detect domain boundaries in multidomain structures (see [10]) through a posteriori knowledge of domain folding and interactions. Several approaches assume that a domain makes more internal contacts (intradomain) than external contacts (contact with residues in the remainder of the structure). For example, the DOMAK algorithm of [11] derives a “split value” from the number of contacts measured when a protein is divided into two parts. Optimal values are obtained when the separate parts of the split structure are distinct domains. The protein domain parser (PDP) takes a similar approach and looks to divide multidomain structures so that the number of internal contacts in each putative domain increases. By contrast, the parser for protein unfolding units (PUU) algorithm by [12] uses a harmonic model to describe interdomain dynamics, and is used to define domain units in the FSSP database [13]. A further approach, used by the DETECTIVE algorithm [14], attempts to determine the hydrophobic core at the heart of each domain unit. The original CATH classification protocol [9] used a consensus approach by combining the results from the three independent methods, described above (PUU, DOMAK, and DETECTIVE).
Although each method reports 70%–80% accuracy in benchmarking tests, our experience of updating the CATH database suggest that these methods frequently (∼80%–90% of the time) produce results that are inconsistent with one another. As a consequence, manual validation becomes the only secure way to resolve these conflicting predictions. A recent analysis by Holland et al. [10] showed that all automatic methods run into difficulties when assigning boundaries for certain architectures that do not fit their chosen model of a domain unit—for example, an alpha horseshoe domain, which does not form a compact structure. The authors suggested improvements achieved by a heuristic method that accounts for exceptions to the theoretical domain model. An alternative approach would be to compare a given protein chain against a library of known domain folds.
Although many of the algorithms described above effectively delineate domains for a large percentage of protein chains in the PDB (even those which contain novel folds), they provide no indication as to how similar each predicted domain is to folds already classified within the CATH database. Therefore, it is still necessary to compare the excised domain against the CATH library to classify the fold. Since manual validation of domain boundaries and structure-based database scans are both slow, this has remained one of the major bottlenecks in the CATH classification process.
As discussed above, there are a limited number of folds, and a novel multidomain structure could well comprise those that have already been classified. This concept of recurrence is not new, and has been successfully exploited by other structural classifications. For example, the DALI algorithm is used to detect recurrent folds for classification in the DALI Domain Database [13], while the SCOP database uses manual inspection to locate known folds.
Many methods exist to find recurring domains using pure sequence approaches (e.g., MKDOM [15], SMART [16], and PFam [17]). However, these are designed to identify individual protein families within gene sequences, rather than predict structural domains. Others, such as SnapDragon [18] and Rigden's covariance analysis [19], attempt to infer domain boundaries through prior prediction of tertiary structure. Nagarajan and Yona [20] used a combination of PSI-BLAST multiple alignments, predicted structural features, and neural networks to identify the transition between domains in the sequence (i.e., the boundaries). The authors were able to correctly predict the domain architecture for 35% of multidomain proteins when compared with SCOP.
Recent analyses of structures solved by the structural genomics initiatives—which are frequently targeted because they have no clear sequence similarity to existing structures and may adopt novel folds—show that approximately 90% are similar to those already observed in the PDB through sequence or structure comparison [21,22]. Therefore, exploiting the concept of domain recurrence to detect known folds in newly determined multidomain structures is a sensible strategy to classify the majority of new structures. Moreover, several fast and powerful algorithms for structure comparison now exist that could be used to perform this task. Some of these compare secondary structures between proteins [23–25], while others operate at the residue level (DALI [26], SSAP [27], COMPARER [28], STRUCTAL [29], and CE [30]).
The performance of an automatic structural alignment method should be measured both on its ability to generate biologically meaningful alignments and its capacity to accurately detect fold similarities and homologous protein structures. As Kolodny et al. [29] highlight, not all structural comparison methods are as good at scoring their alignments as they are at producing them. A root-mean–squared deviation (RMSD) value, or any linear transformation of this, often remains dependent on the number of aligned residues. Some algorithms (e.g., CE [30]) are optimised to find highly conserved regions between two protein structures with a low RMSD. This can be useful in detecting similarities within extremely diverse superfamilies and fold groups. However, this approach does not necessarily give a globally optimal alignment, and can assign high significance to matching small structural motifs that may not be in equivalent positions in the two structures being compared. Hence, for the purpose of domain boundary recognition, it is also vital to consider the number of aligned residues as a proportion of those residues in the larger of the two structures as well as the RMSD of superposed residues.
This paper reports the development of the CATHEDRAL algorithm, a novel domain identifier that exploits the fold-recurrence philosophy. CATHEDRAL is an acronym for CATH's existing-domain recognition algorithm. It compares a novel multidomain protein structure against a library of previously classified folds in the CATH database [6] by modifying and combining features from two established structural similarity algorithms. A secondary-structure–matching algorithm, GT (using graph theory) [25], which is very fast and reasonably accurate, is combined with a residue-based method that uses double-dynamic programming (DDP) [27], and is therefore slower but very accurate. By combining these approaches, a 100-fold to 1,000-fold increase in speed is achieved, depending on the size of the query structure, at no cost to the quality of the domain alignments. This enables regular scans of newly determined protein structures and rapid classification of their constituent domains into the CATH database.
To investigate the efficacy of CATHEDRAL in producing quality alignments, it has also been benchmarked against other publicly available structure comparison algorithms at the single-domain level. By aligning domains in a consensus SCOP/CATH dataset, CATHEDRAL was found to give comparable and, in many cases, superior performance for fold recognition. In addition, when assessing the fidelity of the structural alignments in comparison to hand-curated structural alignments with respect to BAliBASE [31], it consistently performed better than other approaches by aligning more residues correctly.
The rationale behind CATHEDRAL was to use a fast secondary structure–based graph theory (GT) algorithm to discover putative fold matches for a given protein domain/chain structure, which could subsequently be more accurately aligned using a residue-based method exploiting DDP. To evaluate the performance of GT and DDP for fold recognition and accurate alignment, we first created a dataset of single CATH–SCOP domains and compared this approach with other publicly available methods before optimising the algorithm for discovering domain folds in multidomain chains. This benchmarking was also performed on a larger dataset of domains in the nonredundant CATH library (version 2.6) and produced almost identical results.
Once CATHEDRAL has identified putative domain matches for a query multidomain structure, all domain hits to the chain are ranked by the SVM score, and domain boundaries are assigned using the protocol described in Methods. CATHEDRAL was able to assign 90% of domains in the query dataset to the correct fold group, with 80% of these within ten residues of the actual boundary (Figure 9). Although our dataset only contained multidomain chains in which all component domains were represented in the CATH library, this is not always the case in classifying novel structures. Indeed, assigning erroneous folds to chains could adversely affect the quality of the domain boundaries. Figure 10 shows a plot of coverage according to the percentage of accurate boundaries (i.e., within 10 residues). It can be seen that once the SVM score cutoff is increased above 2, the coverage drops dramatically. However, the accuracy of the domain boundaries does not increase significantly, suggesting that this is an appropriate threshold for CATHEDRAL. Figure 9 shows the coverage of all chains in the dataset with respect to the accuracy of their predicted domain boundaries.
CATHEDRAL was developed as a method to be applied unilaterally to all protein chains to be classified into the CATH database. As it is not known a priori whether a given chain contains more than one domain, it is important that the algorithm does not split whole-chain domains unnecessarily. To analyse whether this would pose a problem, the iterative version of CATHEDRAL was also applied to the single-domain CATH–SCOP dataset. In less than 4% of cases, CATHEDRAL predicted that these structures contained more than one domain.
The major speed increase in CATHEDRAL is due to the fact that GT preselects representatives for DDP to align to the query chain. By default, it takes all relatives (nonredundant at 35% sequence identity level) in each of the top ten top-scoring fold groups identified by GT, even if this results in thousands of comparisons, as occurs in large fold groups such as the Rossmann and TIM barrel folds. This can produce very long running times for some query chains. Nevertheless, it is important to find the closest structural relatives for each assignment to reduce the number of insertions and deletions and therefore increase the accuracy of the domain boundary.
We explored whether only a limited number of relatives from each fold could be taken without compromising the fidelity of the domains boundaries. However, given that GT does not accurately discriminate between homologues and domains with the same fold, it was decided to take at least one relative from each superfamily in the target fold group and explore the effect of varying this number.
CATHEDRAL was run as described above (by targeting the top ten fold groups at each iteration), but the number of nonredundant representatives (fr) taken from each superfamily to be aligned by DDP was varied. Figure 11 shows a plot of the number of correctly assigned domain boundaries (within ten residues of manually validated boundary) at each of these levels. It appears that taking any more than seven representatives from each superfamily does not increase the number of good assignments, and hence appeared to be an appropriate level to set the fr parameter.
Figure 12 shows the relationship between the accuracy of the domain boundary and the sequence identity between the assigned domain region and best structural match used to assign the boundary. When sequence identity exceeds 10%, there is an increase in the number of correct domain boundaries. It could be expected that the closer the relative from which the assignment is made, the greater chance of it being correct. However, it is encouraging to note that 60% of assignments with sequence identities between 5% and 10% show very little deviation from the manually verified boundaries.
Structural embellishments of the core of a fold are responsible for the majority of examples where there is a disagreement between a manually assigned boundary and those predicted by CATHEDRAL. Figure 13 illustrates this problem by showing a domain assignment for a catalase HPII [32] (PDB code 1iph) domain, through similarity to its closest match in the CATH library [33] (PDB code 1beb). The matched domain is much smaller than the query, and hence CATHEDRAL is only effective at aligning the core of the fold (shown in red). A number of large insertions in the catalase domain cannot be assigned purely by structural comparison, and these sites are therefore not included within the domain, causing a substantial discrepancy from the correct boundary assignment.
Recent analyses of CATH superfamilies has revealed that in 40% of well-populated superfamilies (nine or more diverse relatives at <35% sequence identity), there is 2-fold or more variation in the sizes of the domains (as measured by the numbers of secondary structures in the domain) [7]. Therefore, in these superfamilies, it may be difficult to obtain accurate boundaries until a close structural relative is deposited in the PDB.
To place the performance of CATHEDRAL in context, we compared its ability to assign domains boundaries with two other methods: hidden Markov models (HMMs) and domain predictions from structure (PDP).
Our dataset of protein chains was scanned against HMMs built from each structure in the CATH library using the HMMer suite of programs [34]. Domain boundaries were then assigned to the query chains in the same way as CATHEDRAL, but using the HMM E-value instead of the CATHEDRAL SVM score to rank hits. We found that the HMM method was only able to discover 65% of domain folds within the dataset chains. One of the main reasons for this low coverage was that 11% of the chains were not annotated with any domains using an E-value threshold of 0.001. Of the domain boundaries assigned, only 33% were within ten residues, compared with 80% for CATHEDRAL. It is possible that the number of assigned domains could have been increased by using a less conservative E-value threshold. However, this is unlikely to improve the overall quality of the domain boundaries, given the low quality of those that were assigned by the HMM alignments. The domain recognition performance is on a par with the method of Nagarajan and Yona [20], who predicted the correct domain architecture of 35% of a dataset of multidomain PDB chains. However, by incorporating structural information they were able to increase the percentage of boundaries within ten residues to 63%.
CATHEDRAL finds domain boundaries for a query chain by using structural alignment to known folds in CATH. To compare our approach with other methods that do not exploit the concept of fold recurrence, but instead are based on ab initio analysis of structural properties such as residue contacts, we applied the PDP method to our multidomain chain dataset. PDP was able to predict correct domain boundaries (within ten residues) for 67% of the chains in the dataset. Although this is lower than CATHEDRAL, it is substantially higher than the 33% achieved by HMM methods. Furthermore, the performance of PDP is still impressive given the problem of distinguishing domain units in a chain based purely on structural properties such as internal contacts and hydrophobicity.
More than 50 structural comparison algorithms have been published in the literature in the last 30 years, the vast majority of which are not in regular use by the bioinformatics or structural biology communities. Those which have gained popularity tend to have a Web-based interface for users to submit their own structures or structures from the PDB. CATHEDRAL has been implemented as a crucial part of the CATH classification protocol, and a new Webserver was created to provide users to investigate domain assignments and homologue recognition with their protein structure of choice (http://cathwww.biochem.ucl.ac.uk/cgi-bin/cath/CathedralServer.pl).
We have developed a protocol for domain boundary assignment in multidomain proteins (CATHEDRAL) that exploits the recurrence of folds in different multidomain contexts. This was devised because a high proportion (currently >90% [21]) of domains in newly determined structures contain folds that have been previously classified in CATH.
CATHEDRAL first scans a query structure against a library of folds from the CATH databases. The algorithm first uses GT to perform a secondary structure–based comparison and identify putative domain fold matches in the query structure. A representative sample of nonredundant superfamily relatives from the top ten folds are then recompared to try to obtain a better alignment and refine the domain boundaries. This latter step exploits a DDP algorithm that has been guided by information on equivalent secondary structures identified by the GT match.
CATHEDRAL combines the power of two established structural comparison algorithms in order to develop a fast and accurate protocol for homologue recognition and domain assignment. CATHEDRAL misses ∼10% of the domains in the target dataset. Of these, ∼30% are too small (fewer than three secondary structures) and so are ignored by the CATHEDRAL protocol. Manual inspection revealed that a further ∼20% are distorted or irregular structures giving poorly defined graphs. The remaining ∼50% are missed because they do not pass the score similarity cutoff, as the relatives are too distant and related structural motifs in neighbouring fold groups are better matched.
The CATH classification of protein folds gives a discrete description of fold space. However, there are difficulties in identifying distinct folds in some populated regions of fold space where the structural universe could be more reasonably represented as a continuum [6]. In many cases, as the size of the protein increases, the repertoire of folds appears to consist of extensions to existing motifs. It has been shown by Koppensteiner et al. [35] that it is possible to “walk” from one α/β sandwich fold to another, through the extension of α/β motifs. Furthermore, certain motifs, described as “attractors,” occur as the core of a protein's structure more frequently than others [36]. Recent analyses of the overlaps between fold groups has shown that for some protein architectures (αβ sandwiches and mainly β sandwiches), extensive overlaps between fold groups are observed due to large common structural motifs [37].
For nearly 80% of the test set, all domain boundaries within the multidomain were correctly assigned within ten residues. This is a considerable improvement over a previous consensus protocol (DBS; [9]) described above, for which on average only 10%–20% of domains could be identified as having reliable boundary assignments from agreement between three independent methods. Furthermore, as known folds are recognised by CATHEDRAL, individual domains can be simultaneously classified in the CATH database, without the need for further structure comparison as in previous classification protocols. The method is currently being extended to assign a confidence level or p-value to the boundary and fold assignments predicted by CATHEDRAL. Furthermore, at present, CATHEDRAL assigns domains to a query chain in an iterative fashion. It could be conceived that a better prediction of boundaries and fold assignments could be attained by considering a number of different classifications. The best of these could be identified as the prediction with the highest confidence value.
Since CATH aims to maintain high quality domain boundary assignments [38], results returned by the CATHEDRAL algorithm will continue to be manually assessed. However, the high accuracy of the approach will considerably facilitate this process. Since the proportion of domain folds classified within CATH is likely to continue to increase significantly in the next decade due to the progress of the structural genomics initiatives, the CATHEDRAL algorithm will considerably increase the speed of classification of new multidomain structures and their constituent folds within CATH.
CATHEDRAL and DDP (a modified version of the SSAP algorithm [27]) were benchmarked against other publicly available structural comparison methods, STRUCTAL [29], DALI [26], LSQMAN, and CE [30] (see Text S1 for description of methods). An all-against-all structural comparison was performed on the 6,003 unique CATH domains (<35% sequence identity to each other) from 907 fold groups for each of the different structural comparison methods, giving more than 18 million individual comparisons. To minimise any bias in the CATH dataset, a dataset that was a subset of CATH version 2.6.0 and SCOP verson 1.65 was also constructed. Each of 6,003 CATH (SRep) domains was checked to see if it had an equivalent SCOP domain containing at least 80% of the same residues. All domains satisfying this criterion were mapped to their CATH and SCOP superfamilies. These superfamilies were then compared, and only those sharing 80% of the same members were identified. This restricted the CATH–SCOP dataset to 1,779 SReps encompassing 406 folds.
There are several publicly available methods that have been endorsed by widespread community use and/or validation by comparative benchmarking against established methods. We selected a range of methods, many of which had been previously benchmarked by Kolodny et al. [29] for their performance in fold recognition and alignment accuracy.
Protocol used to compare the performance of CATHEDRAL and full DDP in fold recognition and alignment accuracy with other established methods—fold recognition. Structure alignment methods were compared using ROC curves. These plot true positive rate (sensitivity) against the false positive rate (1 − specificity) for different similarity scores returned by the individual methods. A binary classifier was defined by the CATH hierarchy whereby a positive match is one in which both domains share the same fold classification. The matches for each method were ordered by the structural similarity score of their alignment, and the number of true positives and false negatives were calculated at varying thresholds.
Protocol used to compare the performance of CATHEDRAL and full DDP in fold recognition and alignment accuracy with other established methods—alignment accuracy.
Kolodny and coworkers tested several measures for assessing the accuracy of structural alignments [29]. They identified redundant measures, and alignment accuracy was subsequently compared using the two geometric measures shown in Equations 1 and 2 below: SAS, and SiMin where Nmat represents the number of aligned residues and L1/L2 represents the length of the respective domains. The different measures attempt to balance the different properties that describe a “good” alignment, weighting the RMSD against the length of the alignment as a fraction of the length of the proteins aligned.
As CATHEDRAL is exploiting fold recognition to obtain reliable domain boundary assignment, we developed a further measure that scores the global alignment accuracy. As opposed to SiMin, which gives a good score for a small fold appearing as a conserved motif within a much larger fold, SiMax (Equation 3 below) takes account of the proportion of residues aligned in the larger domain structure to determine whether a significant fold match has been achieved.
All the measurements are in angstroms, and the percentage of alignments within a particular distance in angstroms were calculated for each measure (SAS, SiMax, and SiMin).
In addition to these geometric measures, alignment accuracy was also assessed by comparison against a set of manually curated alignments. BAliBASE is a database of manually refined multiple structure alignments specifically designed for the evaluation and comparison of multiple sequence alignment programs. The alignments in BAliBASE are selected from the FSSP [36] or HOMSTRAD [39] structural databases, or from manually constructed structural alignments taken from the literature. When sufficient structures are not available, additional sequences are included from the HSSP database [40]. The VAST Webserver [23] is used to confirm that the sequences in each alignment are structural neighbours and can be structurally superimposed. Functional sites are identified using the PDBsum database [41], and the alignments are manually adjusted to ensure that conserved residues and secondary structure elements are correctly aligned.
A total of 14 BAliBASE multiple alignments were selected, comprising 108 pairwise structural comparisons. All the alignments represented single-protein domain chains that shared less than 25% sequence identity, making alignment nontrivial. All three major protein classes were represented, and the quality of the alignments generated by the different structure comparison methods are measured by the score, fm, which quantifies the number of amino acids correctly aligned in the structural alignment divided by the total number of aligned residues in the BAliBASE alignment. CE was not included in this analysis, as it only identifies the largest continuous motif.
CATHEDRAL was benchmarked to calculate its ability to delineate domains within multidomain proteins, as well as correctly recognising the fold of the constituent domains. A set of representatives from 1,071 multidomain S35 sequence families (clustered by single linkage at 35% sequence identity) was selected. From this set, those chains containing domains from folds with less than two S35 sequence families were removed. The remaining set contained 964 chains comprising 1,593 domains. These originated from 245 distinct fold groups and 462 superfamilies.
To identify domain boundaries in a novel multidomain structure, CATHEDRAL scans the query structure against a library of folds classified in the CATH database (see Text S1 for description of CATH hierarchy) that are derived from contiguous domain representatives from each sequence family (in which relatives have at least 35% sequence identity) in version 2.6 of the CATH database. This comprised 4,707 domains, covering 907 folds. A secondary structure graph of each domain was generated as described in Harrison et al. [25].
Iterative protocol used by CATHEDRAL. CATHEDRAL uses an iterative protocol illustrated in Figure 14. As described above, novel multidomain proteins are first scanned against a library of domain folds from CATH using the secondary structure GT algorithm. All folds containing hits in the top ten ranked fold hits are then selected for further analysis. To improve the alignment of the matched regions and thereby identify the closest structural neighbour, fr representatives from each superfamily in the selected folds are compared against the matched region using the DDP algorithm.
As matches to small domains (fewer than five secondary structures) can produce insignificant E-values (see [25]) when compared to large chains, these were isolated from the original CATH library and scanned only after all large domains had been assigned by CATHEDRAL.
A variety of different scoring schemes were assessed for their ability to recognise true matches, together with a combination of several measures using an SVM (see below). If the score suggests that the match is valid, the region is accepted as a putative domain and the alignment used to indicate the residues that can be excluded from the multidomain structure (and score matrix) in future searches. A new graph is constructed from the remaining secondary structures, and the GT and subsequent DDP search is repeated to identify another putative domain. CATHEDRAL continues for up to ten iterations or until there are fewer than three secondary structures left to be assigned.
Identification of corresponding secondary structures in the multidomain protein and a single domain structure using GT. GT was first introduced for protein structure comparison by Artymiuk and coworkers [42]. CATHEDRAL uses a new implementation of this approach [25] (see Text S1) that includes further structural features (e.g., chirality) to obtain a better resolution between related and unrelated folds. A robust statistical framework was also derived to calculate expectation values (E-values) that can be used to assess the significance of each comparison (see [25] for a detailed description of the GT algorithm used in CATHEDRAL).
Generating a residue alignment of the fold match using DDP. Once a putative domain within the multidomain structure has been matched to a fold in the CATH database, an accurate alignment between this domain and the target structure can be obtained using a residue-based method that exploits DDP.
CATHEDRAL uses the global alignment version of the DDP algorithm, described in Taylor and Orengo [27]. This choice followed assessment of the performance obtained using the global and local alignment versions [43]. The global alignment version is better able to handle proteins with discontiguous domains, as the alignment produced by the local version in such cases was found to match only one of the fragments of the discontinuous domain. The break in the discontiguous domain appears to the alignment program as a large gap, and the local alignment score within the gapped region rapidly falls to zero, thus terminating the alignment incorrectly.
Using secondary structure matches from the GT filter to guide residue alignment by DDP. The full DDP algorithm is computationally expensive because it makes an exhaustive search of all possible pathways through the residue and summary level matrices, although this search can be constrained by imposing a window on the score matrix [27]. Fortunately, it is not necessary to compare all the equivalent positions between two related proteins to obtain an accurate residue alignment. Therefore, the clique information identifying matching secondary structures can be used to exclude large regions of the score matrix by populating a binary matrix, which dictates which residues to compare. First, residues in equivalent secondary structures must pair with one another. As equivalent strands and helices can vary in length (e.g., a helix with 11 residues could be aligned to one with eight residues), it must be an all-versus-all pairing (represented by a square of “1” values in the matrix). Similarly, residues on the end of aligned secondary structures could potentially be paired with residues in the loop regions, so the boundary is extended by 10 residues on either side.
Second, although the alignment for residues outside the clique is unknown, it is possible to exclude certain pairings. The clique effectively orientates the alignment and dictates that if helix 1 in protein A is equivalent to helix 2 in protein B, it cannot simultaneously be equivalent to helix 3 in protein B. Moreover, it gives the overall direction of the alignment and allows the regions between the clique secondary structures to be linked together.
Finally, the alignment of embellishments of the core clique secondary structures at the start and ends of the domains is unspecified. However, it is known that these cannot be aligned to any of the core residue pairs. Hence, the starts and ends of the domains are paired up for DDP to decide where the equivalences lie. As outlined in the DDP description in Text S1, residue pairs possessing similar torsional angles and accessibility within these matching secondary structure blocks are then selected for comparison.
Cliques indicate blocks within which residues in matching secondary structures should be aligned. Gaps between these blocks are also possible locations for the residue alignment algorithm to search. The rest of the score matrix can be ignored. This typically gives a significant reduction in the number of residue pairs that must be compared in the first pass of the DDP algorithm. As well as speeding up the alignment, it also reduces the amount of noise in the summary score matrix accumulated in the first pass, as fewer nonequivalent residue pairs are compared. Similarly, once a domain has been matched in the multidomain structure, the block associated with that domain need not be subsequently searched. These restrictions on the search space result in much faster comparisons without significantly affecting the ability to recognise equivalents.
Adapting the CATHEDRAL protocol to favour global matches over local motif matches. The accuracy of the secondary structure–matching algorithm improves with clique size because for larger cliques there are more equivalent geometric relationships identified. This is because a clique that has N nodes contains N(N − 1) / 2 edges. Matches identified using GT are therefore more secure when the clique is large, independent of the residue similarity score. Furthermore, because the scoring scheme for graph-matching breaks down for the very small folds (fewer than three secondary structures [25]), to maintain the integrity of CATHEDRAL's predictions, these very small proteins are excluded by the algorithm.
As CATHEDRAL iterates toward a solution, the CATH database is repeatedly scanned. However, some large folds contain structural motifs that match well to small folds. These motif matches sometimes rank higher in the match list because the geometry is very well conserved, and the selection of these matches over equivalent folds can therefore confuse the identification of domain boundaries. This effect can be avoided by attempting to match only large domains first; that is, two passes of CATHEDRAL are performed. The first pass only allows matches to folds in CATH that have graphs of five or more nodes. Once CATHEDRAL has reached its termination, it is applied again to the folds in CATH that have graphs with three or four nodes.
This strategy results in the smallest folds only being compared against regions of the multidomain protein that are not part of a large fold, as well as typically increasing CATHEDRAL's speed by 50% or more since fewer searches are required. Hence, CATHEDRAL essentially assigns all large domains first before attempting to align smaller domains to any remaining unassigned regions. To aid the assignment of discontiguous domains, in the first iteration, the top hit is also required to be contiguous (i.e., the assigned region comprises one continuous sequence segment).
Scoring the structural similarity of the domain region aligned by DDP. To assess whether a given structural hit represents a true fold match within the multidomain protein, several measures of similarity are calculated. The structural similarity score returned by the DDP algorithm is normalised to lie in the range of 0–100 (with 100 for identical structures) irrespective of the protein sizes [44]. This score is based on similarities in the vectors between Cβ atoms of equivalent residues in the aligned proteins and is normalised to take account of the size of the largest domain being compared.
A rigid body superposition of the structures is also generated from the equivalent residues identified by the alignment. RMSD of the aligned Cα atoms is calculated, and a cutoff can be imposed on the local structural similarity (see above) to select only the most similar residue pairs when generating the superposition of the structures. A cutoff of 30 (with 100 representing identical residues) is used to ensure the most equivalent residues pairs are used to calculate the SAS.
Using an SVM to validate structural matches. Determining domain boundaries in protein chains through iterative fold assignment presents several challenges. For example, there is the problem of mis-assigning folds that simply match a large structural motif that does not correspond to a significant “global” match to the domain region. Discontiguous domains can also present problems for structural alignment algorithms. Several similarity measures can be considered when gauging whether a match is valid. Manual experimentation can be used to explore and optimise the combination of these measures, or machine learning methodologies can be used. In CATHEDRAL, we exploited an SVM to perform the optimisation automatically and to determine when a significant domain structure match to a classified fold in CATH was occurring.
In addition to the similarity measures provided by the GT and DDP algorithms, we also considered other features (e.g., the proportion of residues matched between the two structures, and similarity in domain sizes) to help improve recognition of global similarity between domain structures. We used the SVMLight package [45] to combine these features using a linear kernel.
To train the SVM, 5-fold cross-validation was used to assess the performance of the SVM models. That is, the dataset was split into five sets, and each one was successively used as the test set, while the model was trained on the other four sets. This reduces any potential bias caused by random fluctuations in the composition of the training and test sets. The error cost for positive examples was weighted according to their ratio to negative examples.
Features included the raw score, E-value, and clique size (number of matched secondary structures) returned by the GT comparison. In addition, the raw score derived from the DDP algorithm together with the residue overlap (percentage of residues in the CATH domain aligned against the putative domain region), CATH domain size, sequence identity, and SAS. To improve the ability of the classifier to avoid bias toward one feature, each was normalised between 0 and 1.
Identifying domain boundaries and handling discontinuous domains. The individual DDP similarity score of equivalent residue pairs, normalised to lie between 0–100, indicates where residue similarity is good (high), where it is poor (low), and where it is nonexistent (residue score is zero). Since only individual domains from CATH are scanned against the multidomain structure, the alignment can be used to find domain boundaries, because the residue pair score falls to zero at the boundary.
When CATHEDRAL determines which fold to assign to a region of the protein chain, it is also making a judgment of where the domain boundaries lie. The fidelity of this latter process is arguably dependent on the structural similarity between the domain region in the chain and the domain it has matched in the library. Although domain boundaries can be assigned to the chain in the same step as taking the highest scoring hits to each region of the chain, the accuracy can be improved by modifying the boundaries once all assignments have been made.
Subsequently, domain assignments that contain regions of the chain that overlap with one another are processed as a last step in the protocol. Conflicts are resolved by assuming that the highest-scoring domain is most likely to have the correct boundaries. The boundaries of the overlapping domain are cropped to exclude the shared region. Second, some chains may contain small regions at the start and end that are unassigned. This is often fewer than 20 residues and is unlikely to contain another domain, or comprise an additional segment of a discontiguous domain. In these instances, CATHEDRAL assigns the extra residues at the beginning and end of the chain to the first and last domains, respectively. Similarly, some chains contain small regions between assigned segments. In these cases, CATHEDRAL splits the unassigned residues equally between the two neighbouring segments. |
10.1371/journal.pgen.1000385 | Ploidy Reductions in Murine Fusion-Derived Hepatocytes | We previously showed that fusion between hepatocytes lacking a crucial liver enzyme, fumarylacetoacetate hydrolase (FAH), and wild-type blood cells resulted in hepatocyte reprogramming. FAH expression was restored in hybrid hepatocytes and, upon in vivo expansion, ameliorated the effects of FAH deficiency. Here, we show that fusion-derived polyploid hepatocytes can undergo ploidy reductions to generate daughter cells with one-half chromosomal content. Fusion hybrids are, by definition, at least tetraploid. We demonstrate reduction to diploid chromosome content by multiple methods. First, cytogenetic analysis of fusion-derived hepatocytes reveals a population of diploid cells. Secondly, we demonstrate marker segregation using ß-galactosidase and the Y-chromosome. Approximately 2–5% of fusion-derived FAH-positive nodules were negative for one or more markers, as expected during ploidy reduction. Next, using a reporter system in which ß-galactosidase is expressed exclusively in fusion-derived hepatocytes, we identify a subpopulation of diploid cells expressing ß-galactosidase and FAH. Finally, we track marker segregation specifically in fusion-derived hepatocytes with diploid DNA content. Hemizygous markers were lost by ≥50% of Fah-positive cells. Since fusion-derived hepatocytes are minimally tetraploid, the existence of diploid hepatocytes demonstrates that fusion-derived cells can undergo ploidy reduction. Moreover, the high degree of marker loss in diploid daughter cells suggests that chromosomes/markers are lost in a non-random fashion. Thus, we propose that ploidy reductions lead to the generation of genetically diverse daughter cells with about 50% reduction in nuclear content. The generation of such daughter cells increases liver diversity, which may increase the likelihood of oncogenesis.
| The liver comprises many different types of cells, including hepatocytes. Hepatocytes perform numerous physiological functions, such as detoxification, metabolism, and protein synthesis. Hepatocytes have the ability to fuse with blood cells, generating hybrid hepatocytes that contain nuclei from both fusion partners. In cases of genetic liver disease, fusion between diseased hepatocytes and normal blood cells can result in the formation of hybrid hepatocytes that function normally. In this series of experiments, we show that fusion hepatocytes produce daughter cells with one-half the amount of DNA found in the parental fusion hepatocyte. Furthermore, we show that the daughter cells are genetically distinct from each other. The increase in genetic diversity within the liver could give rise to hepatocytes lacking proper growth control, potentially resulting in tumor formation and cancer.
| Cell divisions in mitosis are thought to always produce daughter cells with the same chromosome content as the parental cell. Our recent studies with fusion-derived polyploid hepatocytes challenge that ideology. We propose that polyploid hepatocytes can undergo ploidy reductions, leading to the generation of genetically distinct daughter cells with reduced DNA content.
Our group previously showed that transplantation of wild-type bone marrow into fumarylacetoacetate hydrolase (Fah) knockout mice leads to the generation of fusion-derived hepatocytes [1],[2]. In this murine model for the human disease hereditary tyrosinemia type 1, hepatocytes expressing FAH have a strong selective growth advantage and can repopulate the diseased host liver [3],[4]. Fah−/− mice can be bred and kept healthy by administering the drug 2-(2-nitro-4-trifluoro-methylbenzol)-1,3-cyclohexanedione (NTBC) in their drinking water [5]. This drug blocks tyrosine catabolism upstream of FAH and, therefore, prevents the accumulation of fumarylacetoacetate, the toxic substrate of FAH. NTBC withdrawal induces liver injury and results in death from liver failure 4–8 weeks later. In bone marrow transplanted Fah−/− mice, fusion between the Fah+/+ donor blood cells and Fah−/− host hepatocytes results in polyploid cells that have a selective advantage and can completely repopulate the liver [1],[2],[6]. Furthermore, FAH positive hepatocytes can be serially transplanted into secondary and tertiary Fah−/− recipients, thereby expanding the pool of hybrid hepatocytes and making them amenable to extensive genetic and cell biological analysis [3],[7]. Bone marrow transplantation has been shown to generate fusion-derived hepatocytes by numerous investigators [8]–[12]. However, direct differentiation of hematopoietic precursors into liver epithelial cells cannot be excluded, but it is clear that the majority of fusion-derived hepatocytes arise by fusion of donor blood cells with preexisting hepatocytes [1],[2],[6].
In our previous studies, chromosomal analysis of hepatocytes from bone marrow transplanted mice revealed the presence of diploid fusion-derived hepatocytes [1]. This result was surprising since fusion-derived cells should be at least tetraploid. Thus, we hypothesized that fusion-derived hepatocytes could undergo ploidy reductions during regeneration, leading to genetic diversity among daughter cells.
The present study rigorously examines whether fusion-derived hepatocytes undergo ploidy reductions. First, cell fusion experiments show conclusively that fusion-derived polyploid hepatocytes generate daughter cells with one-half DNA content. Unexpectedly, a high degree of aneuploidy was seen among fusion-derived cells. Secondly, ploidy reduction events were associated with independent marker segregation. Hepatocytes derived by cell fusion were expected to retain markers from each cell participating in the fusion event. Indeed, analysis of liver sections from repopulated mice showed the majority of nodules harboring both donor and host markers. However, we also detected a low (but highly reproducible) percentage of FAH positive nodules lacking additional markers. Third, to exclude the possibility that diploid hepatocytes expressing donor markers arose from transdifferentiation of hematopoietic cells, we employed a Cre-loxP reporter system in which ß-galactosidase (ß-gal) was only expressed in hepatocytes generated though cell fusion. As expected, polyploid hepatocytes expressed ß-gal and FAH. Consistent with the cytogenetic results, hepatocytes with diploid DNA content also expressed ß-gal and FAH. Finally, we tested individual cells for donor and host markers. Single cell PCR analysis of diploid daughter cells revealed a heterogeneous population containing a combination of donor and host markers. Taken together, our results demonstrate that fusion-derived polyploid hepatocytes undergo ploidy reduction events, generating heterogeneous populations of lower ploidy daughter cells.
Cell fusion produces hybrids with increased centrosome and chromosome numbers. Numerous studies have suggested that tetraploidy and aberrant centrosome numbers can result in genetic instability and cancer [13],[14]. To test whether hepatocytes generated by cell fusion in vivo are genetically stable, we karyotyped hepatocytes from serially transplanted mice. Lethally irradiated Fah−/− recipients were transplanted with cKit+ Linneg/lo Sca1+ (KLS) bone marrow cells from wild-type or ROSA26 (lacZTg/0) donors in a sex-mismatched fashion. Following NTBC withdrawal and liver repopulation, hepatocytes were serially transplanted into female Fah−/− recipients, allowing fusion-derived hepatocytes to undergo successive rounds of proliferation (Figure 1A). Importantly, only FAH positive cells (i.e., fusion products) but not Fah−/− hepatocytes can repopulate secondary recipient livers [3]. After completed repopulation, unselected hepatocyte metaphases were analyzed by standard G-banding techniques. Fusion between two diploid cells generates a tetraploid cell. Normal hepatocytes polyploidize in an age-dependent manner and can be diploid, tetraploid, octaploid or higher (reviewed in [13]). Therefore, the chromosome content of hybrid cells arising from hepatocyte-blood fusion must be tetraploid or greater.
The chromosome content of fusion-derived hepatocytes (i.e., positive for Y-chromosome) varied widely (Figure 1B). Approximately 15% of the metaphases had exactly 80 chromosomes with the percentage increasing to 30% of metaphases harboring nearly 80 chromosomes (80±5). Furthermore, 9% of metaphases had exactly 40 chromosomes, and the percentage increased to 14% containing nearly 40 chromosomes (40±4). The presence of diploid and nearly diploid fusion-derived hepatocytes suggests that they were produced by ploidy reduction events of tetraploid fusion products. Fusion hepatocytes with intermediate numbers of chromosomes were also found (>40 and <80, >80 and <160) (Figure 1B). Nearly all of the Y-chromosome containing metaphases had numerical chromosome abnormalities (Table 1). These numerical aberrations may have resulted from the fusion process or by DNA damage sustained by Fah−/− hepatocytes during NTBC withdrawal [15].
The distribution of chromosomes among fusion-derived hepatocytes clearly supported a pattern of cell fusion, ploidy reduction and polyploidization (Figure 1C). For example, fusion between a female diploid blood cell (40,XX) and male diploid hepatocyte (40,XY) results in a tetraploid fusion-derived hepatocyte (80,XXXY). A ploidy reduction event generates two types of daughter cells (40,XY and 40,XX) that can polyploidize, giving rise to tetraploid cells (80,XXYY and 80,XXXX, respectively). Because we could not distinguish between daughter cells that lost the Y-chromosome and host female hepatocytes, we focused exclusively on Y-chromosome positive metaphases. Fusion-derived tetraploid hepatocytes (∼80,XXXY) were detected in 21% of the metaphases analyzed. Approximately 14% of Y-chromosome positive metaphases were nearly diploid (∼40,XY), corresponding to daughter cells arising through ploidy reduction. Furthermore, 8% of cells were ∼80,XXYY, which is the expected karyotype for polyploidized diploid cells.
In addition to finding fusion-derived hepatocytes arising from diploid-diploid cell fusion, we detected cells generated by fusion between male tetraploid hepatocytes (80,XXYY) and female diploid blood cells (40,XX). Hexaploid cells (120,XXXXYY) were detected in 17% of the metaphases. Furthermore, triploid cells (either ∼60,XYY or XXY), which are the predicted daughter cells of hexaploid reduction events, were identified in 17% of metaphases. Together, these data strongly support the emergence of daughter cells containing one-half DNA content from fusion-derived hepatocytes. Similar to normal diploid hepatocytes, diploid daughter cells either remain diploid or polyploidize to generate tetraploid hepatocytes.
After demonstrating that fusion-derived hepatocytes could generate daughter cells with one-half chromosome content, we hypothesized that ploidy reduction events should also lead to marker segregation among daughter cells. If ploidy reduction occurs in a tetraploid fusion-derived hepatocyte, there is a 50% chance of losing a heterozygous or hemizygous marker. Liver sections from mice repopulated by fusion-derived hepatocytes in serial transplantation experiments (Figure 1A) were stained for FAH, Y-chromosome and ß-gal activity (Figures 2A–2D). Typically, FAH is co-expressed with the Y-chromosome (Figure 2A) and ß-gal (Figure 2D). However, while most regenerating nodules expressed all markers of cell fusion, a fraction of FAH positive nodules (2–5%) were Y-chromosome negative (Figures 2B and 2C) or ß-gal negative (Figure 2D). Based on the expected loss-of-heterozygosity frequency of one-half, the observed frequency suggests that 4–10% of the FAH positive nodules were initiated by cells that had undergone ploidy reductions.
Heterogeneous FAH positive nodules (lacking one or more donor markers) were consistently found in repopulated mice (Figure 2E). Similar results were obtained regardless of the transplantation scheme (F>M>F or M>F>F). Approximately 25–50% of primary recipients contained Y-chromosome negative FAH positive nodules, and this number increased to 90–100% of secondary recipients. ß-gal negative FAH positive nodules were seen in all serially transplanted mice analyzed. These data demonstrate that ploidy reduction events leading to the formation of heterogeneous daughter cells occur in a large fraction of livers repopulated by fusion-derived hepatocytes.
To facilitate the genetic analysis of fusion-derived hepatocytes, a Cre-loxP system was used to track fusion products. Fah−/− mice were bred with transgenic animals expressing Cre-recombinase via a hepatocyte-specific albumin promoter (Alb-Cre) [16]. Lethally-irradiated recipient mice were transplanted with bone-marrow from a ROSA26 reporter (R26R) mouse (Figure 3A) [17]. These animals harbor a floxed allele of the lacZ gene at the ROSA26 locus. In this transplantation scheme, ß-gal is only expressed when the R26R and Alb-Cre alleles are combined in the same cell. Hepatocytes were isolated from repopulated mice and found to express ß-gal, conclusively showing that these cells were derived by fusion between donor blood cells and host hepatocytes (Figure 3B). Fusion-derived cells also expressed FAH, indicating the successful activation of wild-type Fah supplied by donor hematopoietic cells (Figure 3C).
Next, we examined reduction of DNA content and marker segregation in fusion hepatocytes. Single cell hepatocyte suspensions (containing a mixture of fusion-derived hepatocytes and host hepatocytes) from repopulated mice were loaded with the DNA dye Hoechst 33342 and analyzed by flow cytometry. In control experiments, hepatocyte ploidy populations were readily identified in non-transplanted mice (Figure S1A and S1B). As expected, livers isolated from aged mice (Figure S1A) contained fewer diploid hepatocytes than livers isolated from young mice (Figure S1B). Furthermore, diploid hepatocytes were isolated with high purity. Sorted 2n hepatocytes were >99% pure (Figure S1C), and they contained a single Y-chromosome (Figure S1D). Analysis of mice repopulated to ∼40% by fusion hepatocytes revealed populations of diploid and polyploid hepatocytes (Figure 3D). The diffuse polyploid population is consistent with the chromosome distribution in Figure 1B. To utilize ß-gal as a marker of cell fusion, hepatocytes were loaded with fluorescein di-ß-D-galactopyranoside (FDG), a substrate that becomes fluorescent when cleaved by the enzyme. As expected from the overall degree of FAH repopulation, ß-gal was expressed by a fraction of polyploid hepatocytes (27±11% of all cells) (Figure 3E). Moreover, interrogation of diploid hepatocytes revealed a subpopulation of ß-gal positive cells (8±3% of all diploids). Hepatocyte populations of different ploidy were also FACS-sorted and subjected to FAH immunocytochemistry (Figure 3F). A portion of unfractionated hepatocytes expressed FAH (36±8%). Consistent with ß-gal expression, a subpopulation of diploid hepatocytes also expressed FAH (10±3%). Together, these results show that fusion-derived hepatocytes undergo ploidy reduction events, generating diploid hepatocytes that express ß-gal and FAH.
Although the mechanism by which fusion-derived hepatocytes undergo ploidy reductions is unknown, the data clearly suggest a model in which individual chromosomes/markers segregate independently of one another. For example, as described in Figure 3A, fusion between a donor diploid blood cell (female, R26RTg/0, Fah+/+) and recipient diploid hepatocyte (male, Alb-Cre, Fah−/−) generates a tetraploid cell containing a single Y-chromosome and a single copy of R26R. The Alb-Cre genotyping assay fails to distinguish between hemi- and homozygous mice. Thus, tetraploid fusion-derived hepatocytes undergoing ploidy reductions should generate a pair of diploid daughter cells, and each daughter cell should have a 50% chance of inheriting either R26R or the Y-chromosome.
We performed single cell genotyping of diploid hepatocytes to determine whether chromosomes/markers were lost in cells that had undergone ploidy reduction events. Hepatocytes from repopulated livers (Figure 3A) were FACS-purified on the basis of DNA content and genotyped for donor (Fah and R26R) and host (Cre and Y-chromosome) markers. One of the major obstacles to single cell genotyping is PCR failure, resulting from DNA degradation or template inaccessibility [18]. In our hands, PCR failure ranged from 0 to 40%, which is consistent with published rates [18]. To minimize the effects of PCR failure, each marker was detected with two independent primer sets, thus reducing the net dropout rate to 0% (Fah), 13.2% (R26R), 0.5% (Cre) and 2.5% (Y-chromosome). As a control, single splenocytes from repopulated mice were genotyped and found to contain only host markers (cells 1 and 2) or donor markers (cells 3–5) (Figure 4A). These results are consistent with a high degree of donor engraftment seen in our KLS and bone marrow transplanted mice (Duncan, Hickey and Grompe, unpublished results).
Single cell genotyping was performed on 157 diploid hepatocytes derived from two independently transplanted mice. Representative PCR data is shown (Figure 4B). Cell 7 contained only host markers, suggesting that it was host-derived. Cell 1, which was positive for all makers, was fusion-derived. All of the remaining cells illustrated loss of one (cells 2 and 3) or more markers (cells 4–6). Overall, Fah was detected in 29% of the cells. Detailed analysis of the diploid Fah positive fusion products revealed that 57% lost the R26R transgene, 33% lost the Cre transgene and 70% lost the Y-chromosome (Figure 4C). Although PCR failure may account for a small percentage of the observed marker loss, the high degree of marker loss represents loss-of-heterozygosity at the indicated locus. Furthermore, loss of one or more markers was identified in subpopulations of diploid Fah positive hepatocytes (Figure 4D). Only 13% of cells, for example, contained all four markers. Loss of a single marker was detected in 13% (Y-chromosome) and 17% (R26R) of the cells. Loss of two markers (R26R/Y-chromosome, Cre/Y-chromosome) or three markers (R26R/Cre/Y-chromosome) was found in 24%, 17% and 15%, respectively, of diploid hepatocytes. Together, these data showed that diploid daughter cells were genetically unique, suggesting that autonomous markers segregate independently during ploidy reduction events.
In this study, we demonstrated that fusion-derived hepatocytes could undergo ploidy reductions. Initially, serial transplantation experiments were performed. Cytogenetic analysis showed that 14% of fusion-derived hepatocytes were nearly diploid. Surprisingly, fusion-derived hepatocytes were highly aneuploid. While most regenerating nodules expressed all markers of cell fusion, a fraction of FAH positive nodules (2–5%) were Y-chromosome negative or ß-gal negative. This frequency suggests that 4–10% of the nodules were initiated by cells that had undergone ploidy reduction events. Next, we utilized a ß-gal reporter system to track fusion products in discrete ploidy populations. Polyploid hepatocytes expressed FAH (indicating that fusion-derived cells were reprogrammed to express donor genes) and ß-gal (demonstrating that these cells were derived by cell fusion). Significantly, diploid hepatocytes also expressed ß-gal and FAH, establishing that these cells originated from polyploid fusion-derived hepatocytes. Finally, we carefully tracked donor/host markers in hepatocytes that had undergone ploidy reductions by single cell genotyping. The majority of diploid daughter hepatocytes (87%) were negative for one or more markers, giving rise to a heterogeneous population of cells. These results suggest that markers/chromosomes segregate independently during ploidy reduction events.
Hepatocyte polyploidization has been documented in many species (reviewed by Gupta [13]), but ploidy reversal has not been rigorously characterized. Our experiments provide proof-of-concept that ploidy reversal does occur in fusion-derived hepatocytes. Several reports also suggest that normal hepatocytes may undergo ploidy reduction. For instance, treatment of rodents with hepatotoxins thioacetamide [19] and carbon tetrachloride [20] led to a dramatic increase in diploid hepatocytes and concomitant decrease in polyploid hepatocytes over 72 hr. Differential proliferation and/or cell death was not seen among diploid or polyploid hepatocytes [20]. Thus, it is possible that normal polyploid hepatocytes undergo ploidy reductions, but this hypothesis remains to be tested.
The high degree of aneuploidy displayed by fusion-derived hepatocytes is surprising. It is unclear whether aneuploidy resulted directly from the fusion and/or ploidy reduction events or indirectly as a consequence of the Fah repopulation model [5]. Furthermore, we cannot exclude the possibility of stochastic chromosome loss during mitosis [21]. Thus, aneuploid hepatocytes could arise from the random loss of chromosomes by fusion-derived hepatocytes undergoing extensive proliferation.
A number of possibilities could explain how diploid hepatocytes are generated from polyploid fusion-derived hepatocytes. First, it is theoretically possible that binucleated fusion-derived hepatocytes could simply complete cytokinesis (Figure 5A). Normal binucleated polyploid hepatocytes are formed through failed cytokinesis [22],[23]. For example, a mononucleated diploid hepatocyte undergoes a regular mitosis, but then separation of the two daughter cells fails, generating a binucleated tetraploid cell with two diploid nuclei [22]. Whether binucleated hepatocytes could resume cytokinesis is unclear, but it remains a possibility. In the context of fusion-derived hepatocytes, the completion of cytokinesis would generate two mononucleated diploid daughter cells, each with the same genotype as the original fusion partners. As seen in Figure 4D, subsets of diploid hepatocytes contained a donor marker (Fah) and a recipient marker (Cre and/or Y-chromosome), proving that these cells were genetically distinct from the original fusion partners. Therefore, a cytokinesis-type mechanism can be excluded.
The second possibility is chromosome loss via multipolar mitosis, which can lead to the random segregation of chromosome content among two or more daughter cells [24]. Fusion-derived hepatocytes have increased numbers of centrosomes, which could result in the formation of multiple spindle poles during prophase. Thus, multipolar mitotic events could enrich for daughter cells with diploid chromosome content (Figure 5B). However, multipolar mitosis cannot adequately explain the clustering of fusion-derived hepatocytes with atypical chromosome counts. For example, triploid hepatocytes with ∼60 chromosomes (XXY or XYY) comprised 17% of the metaphases analyzed (Figure 1C). These daughter cells likely originated from hexaploid fusion-derived hepatocytes. It is difficult to imagine how multipolar mitosis would enrich for cells with such abnormal chromosome counts. Furthermore, if ploidy reduction were achieved by multipolar mitosis, then each chromosome should be lost with the same low frequency (i.e., 1/19 for autosomes). Single cell genotyping of diploid daughter hepatocytes showed loss of R26R (located on chromosome 6 [25]) and the Y-chromosome at 50% or greater (Figure 4C). This high degree of marker segregation strongly suggests that chromosome/marker loss occurs in a non-random fashion.
Another possibility to explain the emergence of daughter hepatocytes with one-half DNA content is cell division without DNA replication (Figure 5C). This type of ploidy reduction was first described in the mosquito Culex pipiens [26],[27] but has never been described in mammalian cells. In this model, fusion-derived hepatocytes could proceed through G1 phase of the cell cycle, skip S-phase and progress to G2/mitosis. Pairing between homologous chromosomes would ensure proper chromosome segregation. This type of mechanism accounts for the generation of diploid daughter cells (Figures 1B, 3E and 3F) as well as enrichment for atypical triploid daughter cells (Figure 1B). Moreover, the high degree of marker loss seen in diploid daughter cells (Figure 4C) is possible through a chromosome pairing interaction. Rigorous testing of all potential mechanisms must be performed to elucidate the cellular processes governing ploidy reductions in fusion-derived hepatocytes.
Finally, direct transmission of DNA via horizontal gene transfer (HGT) into diploid hepatocytes must be considered. HGT among somatic cells involves phagocytosis of apoptotic cells followed by nuclear uptake/integration of whole chromosomes or chromosome fragments by the engulfing cell [28],[29]. HGT was hypothesized to induce hepatocyte reprogramming in xenotransplantation experiments [30]. In our studies, diploid host hepatocytes could acquire genes from apoptotic donor blood cells, resulting in hepatocyte reprogramming while maintaining a nearly diploid chromosome count (Figure 5D). However, the presence of multiple donor markers on different chromosomes by HGT is expected to be rare, and we found that nearly half of diploid fusion-derived hepatocytes harbored at least two donor markers (Figure 4D). Therefore, our data strongly argue against an HGT-type of mechanism.
Regardless of the mechanism, ploidy reduction events have significant implications. In the context of fusion-derived hepatocytes, ploidy reductions can be a confounding factor when tracing markers during stem cell transplantation. Donor markers can be lost during ploidy reductions, thus leading to an underestimate of engraftment. Similarly, host markers can be lost from hybrids, obscuring the existence of fusion and giving the false impression of transdifferentiation. Because cell fusion between transplanted cell types and target organs has been described in many experimental systems (reviewed in [31]), the possibility of ploidy reductions needs to be considered when interpreting cell transplantation experiments. Furthermore, we propose that ploidy reduction events may contribute to tumorigenesis. The independent segregation of chromosomes from polyploid cells results in genetically heterogeneous diploid daughter hepatocytes. Individual daughter cells could lose tumor suppressors, generating a subset of hepatocytes with oncogenic potential.
The Institutional Animal Care and Use Committee of the Oregon Health and Science University approved all mouse experiments.
The following inbred mouse strains were used: wild-type (C57Bl and 129), transgenic ROSA26 (C57Bl and 129) [32], Fah−/− (C57Bl and 129) [33], transgenic R26R-lacZ C57Bl [34], transgenic Albumin-Cre C57Bl [35].
KLS cells were sorted from mouse bone marrow, as described [36]. Antibodies used for cell sorting are described in Protocol S1.
Primary hepatocytes were isolated by two-step collagenase perfusion [4] and cultured hepatocytes were isolated by trypsinization. For detection of hepatocyte ploidy populations, hepatocytes (2×106/ml) were incubated with 15 µg/ml Hoechst 33342 (Sigma) and 5 µM reserpine (Invitrogen) for 30 min at 37°. Cells were analyzed and/or sorted with an InFlux flow cytometer (Cytopeia) using a 150 µm nozzle. Dead cells were excluded on the basis of 5 µg/ml propidium iodide (Invitrogen) incorporation. Cells adhering to each other (i.e., doublets) were eliminated on the basis on pulse width. Ploidy populations were identified by DNA content using an ultraviolet 355 nm laser and 425–40 nm bandpass filter. Sorted hepatocytes were collected in DMEM with 4.5 g/l glucose (HyClone) containing 50% fetal bovine serum (FBS) (HyClone). The purity of sorted populations was determined at the end of each sort, and only highly purified populations (>99% pure) were used for subsequent assays.
Transplantation of hematopoietic cells (either bone marrow or KLS cells) was performed as previously described [1],[12]. Briefly, hematopoietic cells (either 4×106 bone marrow cells or 3–4×103 KLS cells) were injected retro-orbitally into groups of congenic Fah−/− recipient mice. KLS cells were co-transplanted along with 3×105 bone marrow cells derived from an unirradiated recipient mouse. Host mice were lethally irradiated with 12 Gy using a 137Cs irradiator 4–24 hr prior to transplantation, obtained by two doses of 6 Gy each. Mice were maintained on NTBC drinking water (8 µg/ml). NTBC was withdrawn 3–12 weeks post transplantation, providing a selective environment for the proliferation of FAH positive fusion-derived hepatocytes.
For serial transplantation experiments, hepatocytes were isolated from primary transplanted mice and 1–3×105 cells injected intrasplenically into Fah−/− recipient mice. NTBC was stopped immediately, allowing for selection of FAH positive cells [3].
Freshly isolated primary hepatocytes were seeded at 1–2×103 cells/cm2 on Primaria tissue culture plastic (Beckton Dickinson). Cells were incubated in hepatocyte culture medium containing DMEM with 4.5 g/l glucose (HyClone), 10% FBS (HyClone), nonessential amino acids (Cellgro) and antibiotic-antimycotic (Cellgro). Nonadherent cells were removed after 4 hr. Hepatocytes were then incubated in culture medium supplemented with 100 ng/ml human epidermal growth factor (Invitrogen) plus insulin, transferrin, selenium and ethanolamine (ITS-X, Invitrogen). After ∼40 hr, hepatocytes were treated with 150 mg/ml colcemid (Sigma) for 2–4 hr and harvested by trypsinization. After extensive washing, slides were incubated for 10 min in 56 mM KCl with 5% FBS and fixed with methanol:acetic acid (3∶1 ratio). For karyotype analysis, chromosomes were G-banded with a standard trypsin/Wright's stain protocol. Fluorescent in situ hybridization on interphase hepatocytes was performed using a Cy3-labeled whole chromosome paint (mouse Y-chromosome) per manufacturer's instructions (Cambio).
Histological analyses were performed as described [3]. For FAH immunocytochemistry, hepatocytes (either unfractionated or FACS-purified) were allowed to adhere to collagen-coated Lab-Tek II, CC2-treated chamber slides (Nunc) in hepatocyte culture medium for 24 hr. Slides were washed extensively, fixed with methanol and dehydrated with acetone. After blocking in 5% normal donkey serum, cells were incubated with a custom rabbit polyclonal FAH antibody diluted 1∶1000 and detected with 10 ng/ml donkey anti-rabbit secondary antibody conjugated to Alexafluor 555 (Invitrogen). Nuclei were visualized with 200 ng/ml Hoechst 33342. For detection of ß-gal activity, hepatocytes were plated in culture medium on Primaria plastic. After 24 hr, adherent hepatocytes were subjected to X-Gal (5-bromo-4-chloro-3-indolyl-ß-D-galactopyranoside, Invitrogen) staining as described [37]. Alternatively, flow cytometry was used to detect ß-gal activity in Hoechst-stained hepatocytes (for the detection of ploidy populations) using FDG reagent (Invitrogen) per manufacturer's instructions.
Single hepatocytes or splenocytes were FACS-sorted into individual wells of a 96-well PCR plate, lysed and subjected to semi-nested PCR, as described in Protocol S2.
Fisher's exact test (two-sided) was used to determine statistical significance. P values less than 0.05 were considered statistically significant.
|
10.1371/journal.ppat.1004649 | Role of Pentraxin 3 in Shaping Arthritogenic Alphaviral Disease: From Enhanced Viral Replication to Immunomodulation | The rising prevalence of arthritogenic alphavirus infections, including chikungunya virus (CHIKV) and Ross River virus (RRV), and the lack of antiviral treatments highlight the potential threat of a global alphavirus pandemic. The immune responses underlying alphavirus virulence remain enigmatic. We found that pentraxin 3 (PTX3) was highly expressed in CHIKV and RRV patients during acute disease. Overt expression of PTX3 in CHIKV patients was associated with increased viral load and disease severity. PTX3-deficient (PTX3-/-) mice acutely infected with RRV exhibited delayed disease progression and rapid recovery through diminished inflammatory responses and viral replication. Furthermore, binding of the N-terminal domain of PTX3 to RRV facilitated viral entry and replication. Thus, our study demonstrates the pivotal role of PTX3 in shaping alphavirus-triggered immunity and disease and provides new insights into alphavirus pathogenesis.
| Chikungunya virus (CHIKV) and Ross River virus (RRV) are arthropod-borne viruses associated with massive epidemics affecting millions of people worldwide, causing widespread distribution of alphaviral-induced arthritis. The rising prevalence of alphavirus infections and, critically, the lack of therapeutic treatments warrant urgent attention to elucidate the innate immune responses elicited, which serves as the first line of host defense against alphavirus. Ironically, robust innate immune responses have been associated with both protective and pathogenic outcomes. Here, we identified PTX3 as an innate protein involved in acute CHIKV and RRV infection in humans. Using an established acute RRV disease mouse model, we revealed a pathogenic immunoregulatory role of PTX3 which led to enhanced viral infectivity and prolonged disease. Transient overexpression of PTX3 in a human epithelial cell line identified the importance of PTX3 N-terminus in binding RRV and modulating viral entry and replication. Collectively, our study identified a previously undescribed pathogenic role of PTX3 during virus infection and shed insights into the sophisticated innate immune responses launched against virus invasion.
| Arthritogenic alphaviruses including Ross River virus (RRV) and chikungunya virus (CHIKV) are the causative agents of the widespread arthropod-borne illnesses, Ross River virus disease (RRVD) and chikungunya fever (CHIKF) respectively [1]. RRV is endemic to Australia, Papua New Guinea and South Pacific islands. An average of ~6,000 cases of RRVD endemic to Australia are reported annually [2], and ~500,000 individuals were infected during its first outbreak in Fiji [3]. CHIKV, which is closely related to RRV, has caused large sporadic outbreaks globally, with the largest recorded outbreak of up to 6.5 million cases in India [4]. Recently, 470,000 suspected and confirmed cases of CHIKF have been reported in the Americas [5]. In both RRVD and CHIKF, clinical symptoms include fever, myalgia, fatigue and maculopapular rash [1,6]. Debilitating persistent polyarthritis is the clinical hallmark of alphaviral diseases, often affecting joints in the hands, wrists, elbows, knees and feet, which can persists for months to years post infection [7–9]. In addition, we have recently identified severe pathological bone loss as another characteristic of alphaviral disease which may contribute to the chronic persistent arthralgia [10]. Emerging clinical evidence has demonstrated an increased tendency of CHIKF patients to develop RA [11], and RRVD patients with pre-existing arthritis such as RA have prolonged rheumatic symptoms after infection [12]. These studies suggested a potential link between alphaviral-induced arthritis and other bone diseases, highlighting alphavirus infection as a possible predisposing risk factor for development of complicated bone disorders [13]. The persistency of debilitating polyarthralgias has a serious impact on quality of life and the economy, with an estimated cost of 34 million euros per year solely in the La Reunion CHIKV outbreak [14]. Symptomatic relief is the only therapeutic option currently available, due partly to a lack of understanding of the immune responses elicited during alphaviral infection.
The cellular and humoral arms of innate immunity serve as the first line of host defense against alphaviral invasion. Despite the importance of the innate immune system in the defense against alphaviral infection, increasing evidence of a pathogenic role for innate mediators has also surfaced over the past few years. Excessive production of soluble innate mediators such as interleukin-6 (IL-6), granulocyte macrophage-colony stimulating factor (GM-CSF), tumor necrosis factor-α (TNF-α), interferon-γ (IFN- γ), macrophage chemoattractant protein-1 (MCP-1) and macrophage migration inhibitory factor (MIF) [15–17] contributes to alphaviral disease pathogenesis. Recent evidence that alphavirus-induced diseases can be exacerbated by overt expression of complement factor 3 (C3) [18] and mannose binding lectins (MBLs) [19] highlights the significance of the complement cascade in modulating alphaviral disease pathogenesis.
Long pentraxin 3 (PTX3) is a pattern recognition molecule which belongs to the humoral arm of innate immunity. PTX3 has a role in all three complement pathways, enhancing the activation, inflammation and cell lysis processes [20]. PTX3 can be secreted by a broad range of cell types including neutrophils [21], monocytes, macrophages and myeloid DCs [22] in response to inflammatory signals such as TNF and IL-1 [23]. Upon pathogen encounter, the release of PTX3 enables cells of monocyte-macrophage lineage to recognize and opsonize the pathogen, presenting it to activated phagocytic cells of the immune system for elimination [24]. Elevated expression of PTX3 has been implicated in many inflammatory and autoimmune diseases, including pulmonary infection [25], giant cell arteritis [26], atherosclerosis [27] and rheumatoid arthritis [28]. Intriguingly, PTX3 is thought to have both protective [29,30] and pathogenic functional roles [31] in the immune system.
PTX3 has a variety of ligands, including complement components, microbial moieties, extracellular matrix proteins, growth factors and P-selectin [16]. The interaction of PTX3 and P-selectin is involved in the regulation of inflammation and leukocyte recruitment through attenuation of polymorphonuclear leukocyte (PML, also known as neutrophils) rolling at sites of inflammation [32]. Consequently, this affects the physiological functions of PMNs in pathogen defense and modulates inflammatory processes.
The role of PTX3 in alphavirus-induced diseases has yet to be established. In this study, we identified the crucial involvement of PTX3 during acute alphaviral infections using specimens from CHIKF and RRVD patients. Characterization of PTX3-/- mice and PTX3-overexpressing HEK 293T cells revealed pathological roles of PTX3 in enhancing viral infectivity during acute RRV infection, which was dependent on the binding interaction between RRV and PTX3. In summary, our data demonstrated the crucial role of PTX3 in modulating alphavirus-induced immune responses and disease manifestation through its N-terminal interaction with the virus particles leading to enhanced viral entry and replication.
Elevated levels of PTX3 have been associated with both protective and pathogenic functions in several inflammatory diseases. To investigate the involvement of PTX3 during acute alphaviral infection, we analyzed PBMCs and serum from CHIKF and RRVD patients for levels of PTX3 using qRT-PCR and ELISA, respectively. Transcriptional expression of PTX3 in PBMCs collected from CHIKF patients was significantly higher compared to controls (Fig. 1A). Further segregation of the CHIKF patient cohort based on viral load (Fig. 1B) and disease severity (Fig. 1C) [15] revealed significantly higher transcriptional expression of PTX3 in patients with higher viral load and more severe disease. Similarly, ELISA analysis of serum specimens collected from acute RRVD patients revealed significantly higher levels of serum PTX3 compared to healthy controls (Fig. 1D). Taken together, these data indicate that PTX3 is induced as part of the innate immune response during acute alphaviral infection and its expression is associated with viral load and disease severity.
To determine the expression of PTX3 during alphaviral disease progression, we utilized an established mouse model of acute RRVD [33]. RRV-infected and mock-infected mice were sacrificed at 2 (peak viremia phase), 5 (disease onset phase), 10 (peak disease phase) and 15 (recovery phase) days post infection (dpi). The serum, quadricep muscles and ankle joints were harvested for analysis. High levels of serum PTX3 were detected in RRV-infected mice across all time points, particularly at 2 and 10 dpi, in contrast to consistently low levels of PTX3 in serum from mock-infected mice (Fig. 2A).
To further investigate PTX3 expression at the sites of inflammation, total RNA was extracted from tissues and analyzed by qRT-PCR. A high level of PTX3 expression was observed at 2 dpi in the ankle joint, with levels declining as the disease progressed. In contrast, quadricep muscles showed peak PTX3 expression at 10 dpi, a time that correlated with the peak of disease (Fig. 2B). IHC was also performed in quadriceps harvested from RRV- and mock-infected mice at 10 dpi (Fig. 2C). Pronounced tissue damage was observed in the striated muscle fibers, which was associated with the presence of inflammatory infiltrates. Increased PTX3 expression was observed in the inflammatory infiltrates of quadricep muscles at peak disease (Fig. 2C).
PTX3 is secreted by a vast array of cell types. To identify the source(s) of PTX3 production during acute RRV infection, we harvested splenocytes from mock- and RRV-infected mice at 2 dpi for flow cytometry analysis. Total leukocytes (CD45+) demonstrated significant elevation of intracellular PTX3 after RRV infection. Further segregation of the total leukocytes into various cellular subsets revealed PTX3 induction after RRV infection in only 2 subsets of cells—neutrophils (CD11b+ Ly6Cint) and inflammatory monocytes (CD11bhi Ly6Chi). No induction of PTX3 was observed in NK cells (NK1.1+ CD3-), T cells (CD3+ CD19-) and B cells (CD3- CD19+) (Fig. 2D).
High expression of PTX3 during inflammatory diseases has been associated with differential effects [34]. To determine the role of PTX3 in RRV disease, PTX3-/- and wild-type (WT) C57BL/6 mice were infected with 104 PFU RRV and monitored for the development of RRVD clinical signs for up to 18 dpi. Disease onset in RRV-infected WT mice occurred at 3 dpi, with ruffled fur and very mild hind limb weakness (clinical score 2), while in PTX3-/- mice disease onset was significantly delayed commencing at 5 dpi. RRV-infected PTX3-/- mice also demonstrated milder disease signs between 2 to 7 dpi, compared to the RRV-infected WT mice (Fig. 3A). In contrast, there was no significant difference in clinical presentation between PTX3-/- and WT mice during peak disease (from 8 to 10 dpi). From 11 dpi, PTX3-/- mice showed faster disease recovery than WT mice and by 15 dpi regained full function of hindlimbs. In contrast, WT mice continued to display signs of hindlimb weakness until 18 dpi.
To examine the role of PTX3 in modulating RRV replication in vivo, viral titre was determined in serum, ankle joints and quadricep muscles harvested at 2 and 10 dpi. As seen in Fig. 3B, viral titres in the serum and ankle joints of RRV-infected PTX3-/- mice were significantly reduced compared to WT mice at 2 dpi. There were no significant differences between PTX3-/- and WT mice in viral titres recovered from the quadricep muscles. At 10 dpi, viral titres recovered from the ankle joints of RRV-infected PTX3-/- mice were also lower than in WT mice. Titres in serum and quadricep muscles from both PTX3-/- and WT mice were below the level of detection at this time (Fig. 3C). To confirm these observations, viral load quantification in ankle joints and quadricep muscles were performed using qRT-PCR. Consistent with previous results, higher viral load was detected in the ankle joints of WT mice at 2 and 10 dpi (S1A Fig.), whereas no difference in viral load was detected between RRV-infected WT and PTX3-/- mice in the quadricep muscles (S1B Fig.).
Collectively, our data indicate that PTX3 deficiency delays the development of RRV clinical signs in infected mice during early infection and assists in rapid recovery in the latter stages of disease. Additionally, the absence of PTX3 also reduced the level of viremia and viral load in the ankle joints of RRV-infected mice.
We next sought to determine the effects of PTX3 on the expression of inflammatory mediators IFN-Ɣ, TNF-α, IL-6 and iNOS in the early and late phases of RRVD. The quadricep muscles were collected from RRV-infected PTX3-/- and WT mice at early (2 dpi) and peak (10 dpi) RRV disease. At 2 dpi, IFN-Ɣ (Fig. 4A), TNF-α (Fig. 4B), IL-6 (Fig. 4C) and iNOS (Fig. 4D) levels were significantly reduced in RRV-infected PTX3-/- mice. However, at 10 dpi, IFN-Ɣ, TNF-α, IL-6 and iNOS levels were significantly upregulated in RRV-infected PTX3-/- mice compared to WT animals. Collectively, these data demonstrate that the absence of PTX3 results in delayed inflammatory responses in quadricep muscles of RRV-infected mice, as well as enhanced production of these immune mediators in the latter stages of infection.
Having demonstrated the effect of PTX3 on the induction of soluble inflammatory mediators during acute RRV infection, we next investigated the effect of PTX3 on leukocyte recruitment during in vivo infection. As shown in Fig. 2C, localized cellular infiltration in quadricep muscles of RRV-infected mice occurs at peak disease (10 dpi). To examine the effect of PTX3 on cellular recruitment during early RRV infection, mice were inoculated via the peritoneal route with RRV. At 6 hpi, flow cytometry analysis of peritoneal lavages revealed significantly increased numbers of neutrophils and inflammatory monocytes in the peritoneal cavity of RRV-infected PTX3-/- mice compared to WT mice (Fig. 5A). This early influx of neutrophils and inflammatory monocytes coincides with the chemotactic responses observed in the quadricep muscles of PTX3-/- mice. Among the 5 cytokines investigated, CCL2 and MIF were higher in quadriceps of RRV-infected PTX3-/- mice at 2 dpi compared to WT mice, but not during peak disease (S2A, B Fig.). No significant difference in chemotactic responses of CCL3 (S2C Fig.), CXCL1 (S2D Fig.) and CXCL2 (S2E Fig.) was observed between the PTX3-/- and WT mice at 2 and 10 dpi.
To investigate the effects of PTX3 deficiency on cellular infiltrates during peak RRV disease, mice were infected subcutaneously with 104 PFU RRV and the quadricep muscles examined at 10 dpi. Previously we have shown that inflammatory monocytes and NK cells are the major cells recruited into muscles during localized inflammation [35]. As seen in Fig. 5B, the number of inflammatory monocytes was significantly reduced in PTX3-/- mice compared to WT controls. Infiltration of NK cells, however, was not affected by deficiency of PTX3. Together, these results suggest that acute production of PTX3 dampens early recruitment of neutrophils and inflammatory monocytes, but enhances the egress of inflammatory monocytes in the latter stages of infection.
We next determined the direct effect of PTX3 on the RRV infection process using HEK 293T cells overexpressing PTX3. HEK 293T cells were transiently transfected with a plasmid expressing PTX3 for 20 h and approximately 5 μg/ml of PTX3 could be detected in supernatants using ELISA at this time. In vector-transfected HEK 293T cells, PTX3 could not be detected regardless of RRV infection (S3 Fig.). Overexpression of PTX3 in HEK 293T cells resulted in a significant increase in viral titres recovered from supernatants of RRV-infected cells, compared to cells transfected with control vector, when infected with MOI 0.1, 0.5 and 1 (Fig. 6A, S4A Fig.). This data suggests a direct effect of PTX3 in enhancing RRV replication.
To support that the presence of PTX3 enhanced viral titres, supernatants from vector- and PTX3-overexpressing HEK 293T cells were harvested at 20 h post transfection and incubated with untransfected HEK 293T cells. In the presence of RRV, untransfected HEK 293T cells treated with supernatant from PTX3-overexpressing HEK 293T cells supported significantly increased virus production compared to cells treated with supernatants from vector-treated control cells (Fig. 6B). These data confirmed that the presence of PTX3 is crucial for enhancing virus production.
To confirm that the results of enhanced virus production was due to PTX3 enhancing RRV replication, HEK 293T cells transiently transfected with vector or hPTX3 plasmids were harvested at 20 hour post transfection (hpt) (Fig. 7A) and subjected to a second round of transfection with RRV T48 plasmid through electroporation. At 3 h and 6 h post RRV transfection, cells were harvested for flow cytometry analysis, which demonstrated a significant increase in virus antigen detected within PTX3-, RRV-transfected HEK 293T cells compared to vector-, RRV-transfected control (Fig. 7B). No virus was detected in the supernatant of these RRV-transfected cells at 3 and 6 hpi (Fig. 7C).
To further characterize the effect of PTX3 during alphaviral infection, we examined the potential of PTX3 to directly interact with the virus and enhance viral entry. We quantified the viral load in PTX3-overexpressing HEK 293T cells at early time points following a one-hour virus adsorption step. Typically, alphavirus particles attach to and enter cells during the adsorption phase of infection (0 hpi), with the replication of alphavirus genome commencing 5 to 6 hpi [36]. Therefore, following an hour of virus adsorption, the detection of viral antigens present at 0 hpi is indicative of binding and entry, and 6 hpi is indicative of the synthesis of new virus particles. Detection of intracellular viral antigens in RRV-infected PTX3-overexpressing HEK 293T cells revealed a significant increase in the number of RRV antigen positive cells at 0 and 6 hpi compared to vector-transfected cells (Fig. 6C), indicating that PTX3 facilitates viral entry. This result was further confirmed with qRT-PCR viral load analysis, which detected increased viral load within PTX3-expressing cells at 0, 1, 2, 4, 5 and 6 hpi, compared to vector control (S4B Fig.). At 4 hpi, the first round of virus replication was observed when a sudden spike in viral load was detected (S4B Fig.). Interestingly, in conjunction with increased viral entry in the RRV-infected PTX3-overexpressing cells, we also observed a significant increase in intracellular PTX3 expression, compared to the mocked-infected controls (Fig. 6D, 6E). Furthermore, flow cytometry analysis showed up to 90% of RRV+ cells were PTX3+, suggesting the co-localization of RRV with PTX3 during acute infection (S5 Fig.). Similar results were obtained for CHIKV infection of PTX3-expressing HEK 293T cells. Enhanced viral titres were recovered from the supernatant of PTX3-expressing CHIKV-infected cells when compared to vector controls (S6A Fig.). Further evaluation of CHIKV-infected cells at 0 and 6 hpi demonstrated significant increase in viral entry in PTX3-expressing cells in conjunction with increased intracellular PTX3 expression (S6B, C Fig.).
To demonstrate that the effect of PTX3on enhancing RRV entry and replication contributed to the increased level of virus detected in the in vivo studies, we performed RRV infection on primary fibroblasts isolated from tails of PTX3-/- and WT C57BL/6 mice. At 24 hpi, RRV infection of WT fibroblasts resulted in significant up-regulation of PTX3 mRNA expression compared to mock-infected WT fibroblasts (Fig. 8A). Moreover, viral titres in supernatants from WT fibroblasts were significantly enhanced compared to fibroblasts from PTX3-/- mice (Fig. 8B). To further demonstrate the importance of PTX3 in enhancing RRV replication, recombinant mouse PTX3 was pre-incubated with RRV prior to infection of PTX3-/- primary fibroblast cultures. Virus titres recovered from supernatants of PTX3-RRV complex-infected PTX3-/- fibroblasts at 24 hpi were significantly enhanced compared to RRV-infected PTX3-/- fibroblasts (control) (Fig. 8C). Furthermore, the effects of PTX3 deficiency on viral entry into primary fibroblasts during the early stages of infection were examined. Consistent with our earlier findings, significantly lower viral load was detected in PTX3-/- primary fibroblasts compared to WT after RRV infection at both 0 and 6 hpi (Fig. 8D). Similarly, RRV infection of WT fibroblasts led to increased PTX3 expression compared to mock-infected controls at 0 and 6 hpi (Fig. 8E). Immunofluorescence staining of the WT fibroblasts also revealed more intense expression of PTX3, particularly within the cytoplasm, after RRV infection at 0 and 6 hpi (Fig. 8F).
Collectively, these data demonstrate that PTX3 promotes viral entry and replication at the early stages of RRV infection (0 and 6 hpi) within host cells.
Previous studies have demonstrated that PTX3 binds to a range of microbes, including viruses. For cytomegalovirus [29] and influenza virus [30], recognition by PTX3 was shown to neutralize virus infectivity. To test whether PTX3 can bind to RRV, a microtitre plate-binding assay was performed. Microtitre wells coated with RRV were incubated with increasing concentrations of recombinant mouse PTX3 (rmPTX3) and RRV-PTX3 binding was determined. As seen in Fig. 9A, PTX3 bound to RRV in a dose-dependent manner. Similarly, a microtitre plate binding assay performed on CHIKV also demonstrated that PTX3 bound to CHIKV dose-dependently (S6D Fig.). Next, we examined whether PTX3 colocalizes with RRV during infection. During RRV infection of PTX3-overexpressing HEK 293T cells, RRV colocalized with PTX3 in the cytoplasm at 24 hpi (Fig. 9B). Similarly, RRV infection of HeLa cells, which are highly permissive to RRV infection and express endogenous PTX3 (S7A Fig.), demonstrated clear evidence of PTX3 colocalization with RRV in the cytoplasm during infection (S7B Fig.). These data show that during acute RRV infection, PTX3 forms a complex with RRV and colocalizes in the cytoplasm of the host cells, which may facilitate viral entry and replication processes.
To confirm that the enhanced infectivity observed during acute RRV infection is specific to PTX3 and not to other acute phase immune proteins, a separate experiment was performed using another acute phase protein—MBL. As previously reported, serum MBL expression was significantly elevated in patients suffering from acute RRVD when compared to healthy controls (Fig. 10A) [19]. In the acute RRVD mouse model, elevated serum MBL-C was seen at both 2 and 15 dpi (Fig. 10B). Using a microtitre binding assay, a clear dose-dependent binding interaction between RRV and MBL-C was observed (Fig. 10C). Next, we infected C2C12 cells (Fig. 10D) with either complexed PTX3-RRV or MBL-RRV in order to identify the specificity of acute phase immune proteins in enhancing RRV infectivity. Enhanced infectivity was observed in cells infected with PTX3-RRV complex at 6, 12 and 24 hpi; however, no significant difference in infectivity was observed between RRV- or MBL-C-RRV complex-infected cells (Fig. 10E).
PTX3 consists of a conserved pentraxin C-terminal domain and a unique N-terminal domain. To determine the functional domain that is crucial for its functionality, we first examined the binding efficiency of recombinant human PTX3 (rhPTX3) N- and C-terminal fragments (Fig. 11A) to RRV. Full-length rhPTX3 bound to RRV in a dose-dependent manner (Fig. 11B) and the majority of binding activity could be mapped to the N-terminal domain. Removal of the N-terminal domain led to a significant reduction in RRV binding (Fig. 11C).
We next compared N- and C-terminal domains of rhPTX3 for their ability to facilitate RRV entry and replication. Briefly, RRV was pre-incubated with full-length rhPTX3, N-terminal-rhPTX3, or C-terminal-rhPTX3 and these mixtures were then added to HEK 293T cells. Examination of viral entry at 0 and 6 hpi revealed that N-terminal-rhPTX3 was approximately 30% less efficient in its ability to facilitate RRV entry, compared to full-length-rhPTX3. In contrast, removal of the N-terminal led to a complete ablation of PTX3-enhanced infection (Fig. 11D). Despite retaining approximately 70% of its ability to facilitate viral entry, infection of cells with RRV-N-terminal-rhPTX3 complex led to a reduced ability to enhance viral replication, compared to full-length rhPTX3. However, higher viral titre was still recovered from cells infected with RRV-N-terminal-rhPTX3 complex when compared to control infected with only RRV. No difference in viral titre was observed in cells infected with RRV-C-terminal-rhPTX3 complex (Fig. 11E).
Taken together, these data indicate that the N-terminal domain of PTX3 is responsible for the binding interaction with RRV and its functionality in facilitating viral entry.
Robust innate immune responses serve as the first line of host defense against alphavirus invasion. However, dysregulation of innate responses can also promote pathogenicity and disease. Consistent with this, we have previously identified overt expression of pro-inflammatory cytokines [37,38] and complement components [18] as pathogenic events in alphaviral diseases.
In the current study we sought to determine the role of PTX3, an acute phase protein associated with activation of the complement cascade [39], in the pathogenesis of alphaviral disease. During the acute phase of alphaviral infection, PTX3 was highly induced in serum and PBMCs of RRVD and CHIKF patients, respectively. Notably, the magnitude of PTX3 induction in CHIKF patients was dependent on viral load and disease severity. Similar observations have been reported for the short pentraxin C-reactive protein (CRP), which is a common laboratory marker for diagnosis of alphaviral infection [40,41]. Previously, Chow and colleagues reported that elevated expression of CRP was associated in CHIKF patients with high viral load and severe disease [15]. In addition to elevated PTX3 expression in alphavirus-infected patients, we also report abundant expression of PTX3 in serum and spleen of RRV-infected mice at the early stage of infection. During peak disease, PTX3 expression was also observed within the cellular infiltrates and further characterization identified inflammatory monocytes and neutrophils as the cellular sources of PTX3 during acute RRV infection. These findings indicate PTX3 is induced in response to alphaviral infections in humans and in mice. Elevated serum PTX3 expression has been observed in patients suffering from several arthritic conditions, including rheumatoid arthritis (2.08 ± 0.99 ng/ml), psoriatic arthritis (1.79 ± 0.80 ng/ml), polymyalgia rheumatic (2.08 ± 0.95 ng/ml), ankylosing spondylitis (2.48 ± 1.07 ng/ml) as well as other diseases such as giant cell arteritis (1.98 ± 1.05 ng/ml) and systemic lupus erythematosus (1.03 ± 0.84 ng/ml) [42]. Herein, the strong induction of PTX3 in RRVD (serum PTX3: 36.79 ± 8.443 ng/ml) and CHIKF patients suggests that PTX3 may also be included as a laboratory marker of acute alphaviral infection.
Dual roles of PTX3 have been reported in several pathogen-induced inflammatory diseases. Overexpression of PTX3 has protective effector function during bacterial infection with Aspergillus fumigatus [21,43], Pseudomonas aeruginosa [44] and uropathogenic Escherichia coli [45], as well as viral infections such as murine cytomegalovirus [29] and influenza virus [30]. Nevertheless, PTX3 expression has also been associated with exacerbated inflammatory responses and disease outcomes in intestinal ischemia-reperfusion injury [46] and pulmonary infection with Klebsiella pneumonia [47]. As PTX3 expression was associated with disease severity during acute alphaviral infections, we utilized an established RRVD mouse model [33] to examine the role of PTX3 during alphavirus infection. Deficiency of PTX3 was associated with delayed disease onset. While PTX3-/- mice displayed similar clinical manifestations at peak of disease, these mice recovered more rapidly than WT animals. It has previously been reported that pro-inflammatory cytokines, including IFN-Ɣ, TNF-α and IL-6, and massive cellular infiltration contribute to inflammatory disease during alphaviral infections [37]. Indeed, delayed IFN-Ɣ, TNF-α and IL-6 responses were observed in quadricep muscles of PTX3-/- mice during the peak of RRVD. In addition, PTX3-/- mice showed diminished infiltration of inflammatory monocytes to the quadricep muscles during peak disease. Indeed, PTX3 has been shown to regulate leukocyte recruitment through interaction with P-selectin, leading to attenuation of cellular recruitment [32]. Using a peritoneal exudate model, we demonstrated increased recruitment of neutrophils and inflammatory monocytes in PTX3-/- mice during early stages of infection. This observation may be associated with early upregulation of CCL2 and MIF, which are crucial for the recruitment of RRV-induced cellular infiltration [17,48] during early infection. PTX3 has been shown to bind apoptotic cells promoting deposition of complement components C3 and C1q [49]. Previously, it has been reported that C3 deposition during RRV infection contributes to the destruction of skeletal muscle tissues [18]. Hence, it is likely that the absence of PTX3 in our current study ameliorates complement-induced damage of muscle tissues in RRV-infected mice. Furthermore, we observed higher induction of iNOS in quadricep muscles of PTX3-/- mice at peak RRVD. iNOS expression was recently shown to be pivotal in mediating skeletal muscle regeneration after acute damage [50]. These observations suggest PTX3 plays an immunomodulatory role during alphaviral infection. Moreover, the diminished infiltration of inflammatory monocytes and higher expression of iNOS during peak RRVD may contribute to rapid recovery from disease in the PTX3-/- mice. Collectively, these data identify PTX3 as a pathogenic factor that shapes the progression of alphaviral disease through modulation of RRV-induced immune responses.
PTX3 is a pattern recognition molecule that interacts with viruses such as murine cytomegalovirus [29] and influenza virus [30], through which it can act to inhibit infection of target cells. In our study, in vitro and in vivo approaches were used to demonstrate that PTX3 promotes RRV infection and replication in host cells. Alphaviruses gain entry into host cells through receptor-mediated endocytosis, although the exact cell surface receptors involved remain poorly defined [51]. Herein, we demonstrate that both RRV and CHIKV can bind to PTX3. RRV and CHIKV infection of PTX3-expressing HEK 293T cells led to enhanced viral entry and replication. In addition, treatment of PTX3-/- primary fibroblasts with rPTX3 also resulted in enhanced viral replication during early RRV infection, likely due to the formation of PTX3-RRV complex which enhances early viral entry events and replication. These data suggest that the extracellular interaction between PTX3 and RRV was involved in facilitating viral entry into host cells. The aggregates formed between RRV and PTX3 may promote more efficient multivalent binding to cell surface receptor/s for RRV, thereby promoting enhanced receptor-mediated endocytosis and viral entry. Alternatively, PTX3 may opsonize RRV and promote its uptake via putative (at this stage unknown) cell surface receptors for PTX3.
In addition to demonstrating the potential of PTX3 enhancing RRV entry into cells, we also report that the distribution of intracellular PTX3 was altered during RRV infection. Intracellular PTX3 migrates from perinuclear space to cytoplasm during infection and PTX3 co-localized with RRV in the cytoplasmic space suggests the possibility of intracellular associations between PTX3 and RRV. These interactions may further promote productive viral infection, perhaps by enhancing genomic replication. Indeed, we demonstrated that cells co-transfected with PTX3 and RRV, and harvested prior to the release of new virions had elevated levels of intracellular virus antigen. This result further supports the hypothesis that intracellular associations of PTX3 and RRV may promote viral replication processes. Moreover, the presence of PTX3 was crucial for enhanced viral replication during RRV infection of WT mice and PTX3-overexpressing HEK 293T cells. Together, this study shows that PTX3-RRV interaction gives rise to pathogenic effect, enhancing viral entry and replication, in contrast to previous studies using other viruses such as murine cytomegalovirus [29] and influenza virus [30], where PTX3 binding was associated with virus neutralization, thereby contributing to a protective host response.
PTX3 is a structurally complex multimeric protein, comprising a highly conserved C-terminal domain shared across all members of the pentraxin family and a unique N-terminal domain whose structure is poorly characterized. We showed that the N-terminal domain is crucial for PTX3 binding to RRV and PTX3-mediated enhancement of RRV infection. However, removal of the C-terminal domain did affect the ability of the N-terminal domain of PTX3 to modulate viral replication, resulting in only partial enhancement of viral replication compared to full-length PTX3. Previous studies have reported the importance of an intact quaternary structure in order for PTX3 to retain its binding and biological efficacies [52]. Therefore, full-length PTX3 with intact quaternary structure would be necessary to retain its biological role of enhancing RRV replication.
Taken together, the data presented in this study provides the first evidence of a role for PTX3 in enhancing RRV uptake and replication during early alphaviral infection. PTX3 has previously been associated with protective functions against a number of viruses, including influenza virus [30], human/murine cytomegalovirus [7] and coronavirus murine hepatitis virus [53], in contrast to the pathogenic role identified in the current study. Our findings demonstrate a previously undescribed pivotal role of PTX3 in shaping alphaviral disease progression through immunomodulation and facilitating viral infection and replication processes during the acute phase of infection. In conclusion, our findings provide new insight into the role of PTX3 in acute alphaviral infection. The newly identified role of PTX3 in enhancing RRV infection and replication also sheds light on the poorly defined route of alphavirus entry into host cells. Given the diverse functional roles of PTX3 as well as its ability to bind to a variety of immune factors, further study is required to define the exact PTX3-triggered immune pathways induced in alphaviral-induced arthritic diseases. Identification of such pathways will be an important step towards the future development of therapeutic interventions.
Animal experiments were approved by the Animal Ethics Committee of Griffith University (BDD/01/11/AEC). All procedures involving animals conformed to the National Health and Medical Research Council Australian code of practice for the care and use of animals for scientific purposes 8th edition 2013. CHIKV human PBMC samples were collected from 20 patients that were admitted to the Communicable Disease Centre at Tan Tock Seng Hospital during the 2008 Singapore CHIKF outbreak. All patients were diagnosed with CHIKF and blood were collected at the acute phase (median of 4 days after illness onset) of infection [54], with written informed consent obtained from all participants. The study was approved by the National Healthcare Group’s domain-specific ethics review board (DSRB Reference No. B/08/026). All RRV human serum samples had been submitted to the Centre for Infectious Diseases and Microbiology Laboratory Services (CIDMLS), Westmead Hospital for diagnostic testing and laboratory investigation of RRV with written and oral informed patient consent. Serum from healthy individuals was provided by Australian Red Cross with written and oral informed consent, approved by Griffith University Human Research Ethics Committee (BDD/01/12/HREC). No new human samples were collected as part of this study. Serum samples were de-identified before being used in the research project.
PBMC specimens of 20 patients were classified into viral load (high viral load, HVL; n = 10 and low viral load, LVL; n = 10) and disease severity (severe illness; n = 10 and mild illness; n = 10) groups, as described previously [15]. Briefly, the HVL and LVL groups had mean viral loads of 1.31 × 108 PFU/ml and 1.95 × 104 PFU/ml respectively, while severe illness were defined as having a temperature of higher than 38.5°C, pulse rate more than 100 beats/min or platelet count less than 100 × 109 cells/L. Serum specimens were collected from 21 acute cases of RRV-induced polyarthritis patients in Australia. PBMCs and serum specimens isolated from 10 healthy volunteers were used as controls. All specimens were stored at -80°C until use.
Stocks of the WT T48 strain of RRV were generated from the full-length T48 cDNA clone, kindly provided by Richard Kuhn, Purdue University, West Lafayette, IN. The CHIKV variant expressing mCherry (CHIKV-mCherry) was constructed using a full-length infectious cDNA clone of the La Reunion CHIKV isolate LR2006-OPY1 as described previously [55].
HEK 293T, HeLa and C2C12 cells were cultured in DMEM supplemented with 10% FBS. Primary fibroblasts were isolated from tails of WT and PTX3-/- mice using a previously described protocol [56] and cultured in DMEM supplemented with 20% FCS. Transient transfection of PTX3 plasmids [57] was performed using Lipofectamine 2000 (Invitrogen) following manufacturer’s instructions. Electroporation of RRV T48 infectious plasmid clone [33] was performed using Eppendorf Eporator. Recombinant N-terminal and C-terminal PTX3 proteins were purified as described in [58]. Recombinant mouse and human PTX3, and mouse MBL-C were purchased from R&D.
HEK 293T cells and primary tail fibroblasts were plated at a density of 1.0 × 105 per well on 24-well plates overnight, prior to infection with RRV or CHIKV (MOI 1) for 1 h at 37°C in humidified CO2 incubator. Virus overlay was removed and 1 ml of pre-warmed growth medium was added to the monolayer of cells, marking the 0 hour post infection (hpi). Cells were incubated at 37°C in humidified CO2 incubator and were harvested accordingly.
All titrations were performed by plaque assay on Vero cells as described previously [59].
Microtiter plates (Sarstedt) were coated overnight at 4°C with 0.1M carbonate-bicarbonate coating buffer alone or containing either 104 PFU RRV or CHIKV (UV-inactivated for 30 min). Non-specific binding sites were blocked by 5% BSA in PBS for 1 h at room temperature. Recombinant PTX3 or MBL-C binding to virus was performed by incubating recombinant proteins on virus-coated microtitre plate for 2 h at 37°C. Biotin-conjugated anti-PTX3 or anti-MBL-C detection antibody (R&D) was added and incubated at room temperature for 2 h. The optical density at 450 nm was read using the streptavidin conjugated to horseradish-peroxidase (HRP) substrate (R&D).
Total RNA extraction was performed using TRIzol reagent (Life Technologies) following manufacturer’s instructions. Quantification of total RNA was measured by NanoDrop 1000 spectrophotometer (Thermo Scientific). Extracted total RNA (10 ng/μL) was reverse-transcribed using an oligo (dT) primer and M-MLV reverse transcriptase (Sigma Aldrich) according to the manufacturer’s instructions.
qRT-PCR was performed using SsoAdvanced Universal SYBR Green Supermix (BIO-RAD) in 12.5 μl of reaction volume. Reactions were performed using QuantiTect Primer Assay kits (Qiagen) and BIO-RAD CFX96 Touch Real-Time PCR Detection System on 96-well plates. Cycler conditions were as follows: (i) PCR initial activation step: 95°C for 15 min, 1 cycle and (ii) 3-step cycling: 94°C for 15 sec, follow by 55°C for 30 sec and 72°C for 30 sec, 40 cycles. Dissociation curves for each gene were acquired using CFX Manager software to determine specificity of amplified products. The fold change relative to healthy donors/mock samples for each gene was calculated with the ΔΔCt method using Microsoft Excel 2010. Briefly, ΔΔCt = ΔCt(patient/infected)–ΔCt(healthy donor/mock) with ΔCt = Ct(gene-of-interest)—Ct(housekeeping gene-GAPDH/HPRT). The fold change for each gene is calculated as 2-ΔΔCt.
Standard curve was generated using serial dilutions of RRV T48 infectious plasmid DNA as described previously [33]. Quantification of viral load was performed using SsoAdvanced Universal Probes Supermix (BIO-RAD) in 12.5 μl reaction volume to detect nsP3 region RNA, using specific probe (5-ATTAAGAGTGTAGCCATCC-3’) and primers (forward: 5’-CCGTGGCGGGTATTATCAAT-3’; reverse: 5’-AACACTCCCGTCGACAACAGA-3’)[60]. Reactions were performed using BIO-RAD CFX96 Touch Real-Time PCR Detection System on 96-well plates. Cycler conditions were as follows: (i) PCR initial activation step: 95°C for 3 min, 1 cycle and (ii) 2-step cycling: 95°C for 15 sec, followed by 60°C for 45 sec, 45 cycles. Standard curve was plotted and copy numbers of amplified products were interpolated from standard curve using Prism Graphpad software to determine viral load.
Transfected HEK 293T cells were seeded on poly-L-lysine-coated coverslips for staining. Cells were fixed with 2% paraformaldehyde (PFA), permeabilized in PBS containing 0.1% Triton X-100, and blocked with 20% goat serum in PBS. Cells were incubated with rat monoclonal anti-PTX3 (MNB4, Abcam) or mouse monoclonal anti-alphavirus (3581, Santa Cruz) primary antibody in PBS, followed by goat anti-rat AF488 (Invitrogen) or goat anti-mouse AF647 (Invitrogen) secondary antibody. Cells were washed, mounted, and examined with a confocal laser-scanning microscope (Fluoview FV 1000, Olympus) at 60x magnification. Images were collected and processed using FV1000-ASW software.
ELISAs were performed using DuoSet ELISA Development kit (R&D systems) following manufacturer’s instructions.
To analyze PTX3 intracellular expression, transfected HEK 293T cells were fixed with 2% PFA and permeabilized with 0.1% Saponin (Sigma Aldrich) in PBS. Indirect intracellular staining was performed with rat anti-PTX3 (MNB4, Abcam) primary antibody, followed by AF488-conjugated anti-rat (Life Technologies) secondary antibody. To identify the various cell populations present in splenocytes, peritoneal lavage and quadriceps harvested from mice, cells were first incubated with anti-mouse CD16 / CD32 (FC block, BD Pharmingen) and stained with the following antibodies: APC-conjugated anti-mouse GR1, PE-conjugated anti-mouse F4/80, FITC-conjugated anti-mouse CD11b, APC-conjugated anti-mouse Ly6c, APC-conjugated anti-mouse CD3, FITC-conjugated anti-mouse CD19, PE-conjugated anti-mouse CD45, or PE-Cy7-conjugated anti-mouse NK1.1 (BD Pharmingen). For detection of alphavirus antigens, indirect intracellular staining was performed using mouse monoclonal anti-alphavirus (3581, Santa Cruz) primary antibody, followed by AF488-conjugated anti-mouse (Life Technologies) secondary antibody. Data acquisition was performed using CyanADP (Beckman Coulter), and analysis was done by Kaluza Flow Analysis Software (Beckman Coulter).
6–8 week-old C57BL/6 male and female mice, of equal distribution, were inoculated intraperitoneally with 105 PFU RRV in 500 μl of PBS, to study the early effect of PTX3 deficiency on recruitment of neutrophils and inflammatory monocytes. Peritoneal lavage was harvested at 6 hpi with 5 ml of ice-cold PBS.
For the acute RRV mouse model, 21-day-old C57BL/6 male and female mice, of equal distribution, were inoculated subcutaneously in the thorax below the right forelimb with 104 PFU RRV in 50 μl. Mock-infected mice were inoculated with PBS diluent alone. Mice were weighed and scored for disease signs every 24 h. Mice were assessed based on animal strength and hind-leg paralysis, as outlined previously [33], using the following scale: 0, no disease signs; 1, ruffled fur; 2, very mild hindlimb weakness; 3, mild hindlimb weakness; 4, moderate hindlimb weakness; 5, severe hindlimb weakness, 6, complete loss of hindlimb function; and 7, moribund. Mice were euthanized, quadriceps and ankle joints were removed and homogenized using QIAGEN Tissuelyser II then centrifuged at 12, 000 × g, 5 min, 4°C. Blood was collected via cardiac puncture. Serum was isolated by centrifugation at 12, 000 × g, 5 min, 4°C. For analysis of infiltrating inflammatory cells by flow cytometry, mice were sacrificed and perfused with PBS at 7 dpi. Quadricep muscles were harvested, weighed, minced, and digested with DMEM containing 20% FBS, 1 mg/ml of collagenase IV (Roche) and 1 mg/ml of DNase I (Roche), for 1 h at 37°C. Cells were strained through a 40 μm strainer (BD Biosciences) and washed with DMEM containing 20% FBS and viable cells were counted by trypan blue exclusion.
For histology, quadricep muscles harvested were fixed in 4% PFA, followed by paraffin-embedding. Five-micrometer sections were prepared. IHC was performed on dewaxed, rehydrated, 5 μm paraffin-embedded tissue sections. Sections were incubated with 20% goat serum (Gibco) in 5% BSA/PBS for 20 min. Primary antibody staining was performed using rat anti-mouse PTX3 (MNB1, Abcam) in PBS, incubated overnight, at 4°C, in humidified chamber. Tissue sections were washed in PBS for three times at 5 min intervals. Secondary antibody staining was performed using HRP-conjugated anti-rat IgG2b (Serotec) incubated for 30 min, room temperature, in a humidified chamber. Colour was developed with 3,3’-diaminobenzidine (DAB) Peroxidase Substrate Kit (Vector Laboratories), according to manufacturer’s instructions and counter-stained with hematoxylin (Vector Laboratories).
All statistical analyses were performed using Prism 5.01 (Graph-Pad Software). Analysis of PTX3 expression profiles in comparison between healthy and RRVD or CHIKF patients, HVL and LVL CHIKF patients’ groups, and severe and mild illness CHIKF patients’ groups was done using Mann-Whitney U test. Comparisons of PTX3 expression among different time points post infection in WT mice, PTX3 expression in mock- and RRV-infected mouse splenocytes, clinical scoring of between PTX3-/- and WT mice, viral replication and viral entry among RRV-infected HEK 293T cells and fibroblasts, were performed using two-way ANOVA with Bonferroni post-test. Comparisons of viral replication and viral entry among RRV-, FL-PTX3, N-term-PTX3 and C-term-PTX3-RRV infected groups were analyzed using one-way ANOVA with Bonferroni post-test. Analyses of all other experimental groups were performed using student unpaired t-test. P values less than 0.05 were considered statistically significant.
|
10.1371/journal.pgen.1003393 | A Gene Transfer Agent and a Dynamic Repertoire of Secretion Systems Hold the Keys to the Explosive Radiation of the Emerging Pathogen Bartonella | Gene transfer agents (GTAs) randomly transfer short fragments of a bacterial genome. A novel putative GTA was recently discovered in the mouse-infecting bacterium Bartonella grahamii. Although GTAs are widespread in phylogenetically diverse bacteria, their role in evolution is largely unknown. Here, we present a comparative analysis of 16 Bartonella genomes ranging from 1.4 to 2.6 Mb in size, including six novel genomes from Bartonella isolated from a cow, two moose, two dogs, and a kangaroo. A phylogenetic tree inferred from 428 orthologous core genes indicates that the deadly human pathogen B. bacilliformis is related to the ruminant-adapted clade, rather than being the earliest diverging species in the genus as previously thought. A gene flux analysis identified 12 genes for a GTA and a phage-derived origin of replication as the most conserved innovations. These are located in a region of a few hundred kb that also contains 8 insertions of gene clusters for type III, IV, and V secretion systems, and genes for putatively secreted molecules such as cholera-like toxins. The phylogenies indicate a recent transfer of seven genes in the virB gene cluster for a type IV secretion system from a cat-adapted B. henselae to a dog-adapted B. vinsonii strain. We show that the B. henselae GTA is functional and can transfer genes in vitro. We suggest that the maintenance of the GTA is driven by selection to increase the likelihood of horizontal gene transfer and argue that this process is beneficial at the population level, by facilitating adaptive evolution of the host-adaptation systems and thereby expansion of the host range size. The process counters gene loss and forces all cells to contribute to the production of the GTA and the secreted molecules. The results advance our understanding of the role that GTAs play for the evolution of bacterial genomes.
| Viruses are selfish genetic elements that replicate and transfer their own DNA, often killing the host cell in the process. Unlike viruses, gene transfer agents (GTAs) transfer random pieces of the bacterial genome rather than their own DNA. GTAs are widespread in bacterial genomes, but it is not known whether they are beneficial to the bacterium. In this study, we have used the emerging pathogen Bartonella as our model to study the evolution of GTAs. We sequenced the genomes of six isolates of Bartonella, including two new strains isolated from wild moose in Sweden. Using a comparative genomics approach, we searched for innovations in the last common ancestor that could help explain the explosive radiation of the genus. Surprisingly, we found that a gene cluster for a GTA and a phage-derived origin of replication was the most conserved innovation, indicative of strong selective constraints. We argue that the reason for the remarkable stability of the GTA is that it provides a mechanism to duplicate and recombine genes for secretion systems. This leads to adaptability to a broad range of hosts.
| Double-stranded DNA viruses are extremely abundant and evolve rapidly, yielding highly diverse viral populations. The transfer of bacteriophage DNA from one bacterial cell to another is regulated by the viral genome. Bacteriophage sequences may account for up to 20% of the bacterial chromosome, but most insertions are highly unstable and the presence/absence patterns of prophage genes vary even in otherwise nearly identical genomes. In generalized transduction, bacterial DNA is by mistake packaged into the phage capsid and transferred into another cell. Gene transfer agents (GTA) differ from viruses in that they transfer random pieces of the bacterial genome and that the fragments are shorter (<14 kb) than needed to encode the phage particle [1]. Although genes for putative gene transfer agents are widespread in bacterial genomes, the selective forces that drive their evolution and maintenance are still largely unexplored.
The best-studied agents are RcGTA from Rhodobacter capsulatus, which resembles a small, tailed bacteriophage and packages 4.5 kb DNA fragments [2], [3], [4], and VSH-1 in the intestinal spirochaete Brachyspira which transfers 7.5 kb DNA fragments [5], [6]. The 15 genes that encode the RcGTA are clustered, whereas the bacterial genes that control their expression are scattered around the R. capsulatus genome. Regulation is mediated by quorum-sensing systems and responds to changes in nutrition and stress in the environment [7]. Although functional RcGTA particles have so far only been identified in Rhodobacter capsulatus and Ruegeria pomeroyi [8] all members of the Rhodobacterales have complete RcGTA-like gene clusters, and most bacteria of other alphaproteobacterial orders contain partial clusters [4]. Phylogenies inferred from the capsid protein sequences show a similar topology as the 16S rDNA tree, indicating relationship by vertical descent [4]. The conservation of the RcGTA genes is remarkable given that lifestyles are extremely diverse and alphaproteobacterial genome sizes differ by one order of magnitude.
In this contribution, we have searched for factors that can explain the emergence of GTAs. We have used Bartonella as our model since they belong to the Alphaproteobacteria but have evolved their unique GTA [9] that is unrelated to the RcGTA [4]. Moreover, a peculiar amplification of a segment of several hundred kb in size has been observed in Bartonella henselae [10], [11] and Bartonella grahamii [9]. The peak of the amplification is located in a region that contains a few phage genes, suggesting that the origin of replication is derived from a bacteriophage [9], [11]. This resembles run-off replication (ROR) in Salmonella, where a phage origin in a prophage amplifies surrounding chromosomal sequences by accident [12]. Based on studies performed in B. grahamii it has been shown that the combination of the two phage-derived systems results in the production of phage particles that contain genomic DNA in direct proportion to the level of amplification from the ROR-region [9]. However, the evolutionary significance of the newly identified GTA in Bartonella has not yet been demonstrated.
Bartonella differ from most other members of the Rhizobiales and Rhodobacterales in that they are adapted to diverse mammalian hosts where they infect endothelial cells and erythrocytes. The infections are normally asymptomatic, but human pathogens like Bartonella bacilliformis and Bartonella quintana cause Oroya fever and trench fever, respectively. Because of the many hosts involved, and the opportunity for transmission to novel hosts with the aid of blood-sucking arthropods, Bartonella is also a good model organism for studies of the molecular mechanisms involved in adaptive radiation [13].
The acquisition of a type IV secretion system (VirB) and the associated genes for effector proteins have been suggested to represent the key innovation event that triggered adaptive radiation in two Bartonella lineages [13]. The virB operon encodes a pilus structure that injects a combination of effector proteins directly into the primary host cell niche, causing modulations of a variety of host cytoplasmic functions [14], [15], [16]. It was hypothesized that the virB gene cluster was transferred from a conjugative plasmid into the ancestral strains of these two lineages in two separate events [13]. Another gene cluster for a conjugative T4SS, trw, which mediates binding to erythrocytes [17], has also been imported into Bartonella from a plasmid [18]. Both the virB and the trw gene clusters are necessary for successful infections of B. tribocorum in a rat model [17], [19]. Although it is generally agreed that the surface components of these systems evolve at high rates, different mechanisms have been proposed. Positive selection for nucleotide substitutions is one hypothesis [13], higher fixation rates for recombination events due to diversifying selection is another [10], [18].
Despite remarkable progress in our understanding of the function of the different T4SSs in Bartonella, a comprehensive understanding of the evolution and plasticity of their genomes is still lacking. Here, we present the sequences of six new genomes of non-pathogenic isolates from wild and domestic animals and connect the discovery of the unique GTA with the acquisition and evolution of several different host adaptation systems. We propose that the acquisition of secretion systems along with a phage-derived system to modify them underlies adaptive radiation of all lineages in the genus Bartonella.
Whole genome shotgun sequencing was performed on six Bartonella isolates (Table 1). The isolates were selected to provide a broad sampling of the known phylogenetic diversity of the genus Bartonella. Four isolates have been described previously, including strains from a marsupial, B. australis NH1 [20], a cow, B. bovis (Bermond) 91-4 [21] and two dogs, B. vinsonii berkhoffii Winnie isolated from a Pekingese [22] and Tweed isolated from a Labrador [23]. Two isolates were obtained as part of this study; both were cultivated from blood samples taken from moose at two different sites in Sweden. These isolates were most similar to B. bovis and B. schoenbuchensis based on 16S rRNA sequence data, and were classified as B. bovis m02 and B. schoenbuchensis m07a, respectively. Experimental infection of bovine endothelial cells by the moose isolate B. bovis m02 is shown in Figure S1.
The six new genomes are at various stages of completion, including two fully resolved genomes of B. australis NH1 and B. vinsonii Winnie (Table S1). Unresolved gaps in the draft genomes covered phage genes in B. vinsonii Tweed and long stretches of approximatively 20–100 kb of repeated genes with internal repetitions in B. bovis m02 and 91-4, and B. schoenbuchensis m07a (Figure S2). The architectures of the Bartonella genomes, as inferred from comparisons to previously sequenced genomes are largely conserved, with the exception of a major inversion in B. bacilliformis, which was confirmed by PCR and sequencing, and several smaller inversions across the terminus of replication (Figure S2). All single-copy genes and at least one copy of all duplicated genes were resolved in all draft genomes. For consistency, the newly sequenced genomes were annotated along with the previously published genomes through our pipeline using the manually annotated B. grahamii genome as the reference (Figure S3).
Plasmids of 23–28 kb have previously been identified in B. grahamii (pBGR3, NC_012847) [9] and B. tribocorum (plasmidBtr, NC_010160) [24], both of which include a copy of vbh, a type IV secretion system (T4SS). Contigs 8 to 10 in B. schoenbuchensis R1 (FN645513-15) are also likely derived from a plasmid, which we have here designated pSc. The m07a strain of B. schoenbuchensis sequenced as part of this study harbors two plasmids, one of which is 59 kb (pML) and shows homology to pBGR3 and pSC. The other (pMS) is only 2 kb and contains only three genes (two copies of repA, mob) that shows homology to genes on the 2.7 kb cryptic plasmids pBGR1 and pBGR2 identified in B. grahamii isolate IBS 376 and WM10, respectively (NC_006374, NC_004308) [25].
Combined with previously published data, a total of 16 Bartonella genomes (8 complete and 8 draft genomes) were included in our comparative genomics study (Table S2). These genomes range in size from 1.4–2.6 Mb, comprising about 1100–2000 CDSs per genome. To place the genomic data in an evolutionary context, we inferred a reference phylogeny of the 16 Bartonella isolates and 6 representative outgroup species that are also members of the Rhizobiales. For this purpose, we clustered the encoded proteins into families of protein homologs using a Markov chain algorithm [26]. Core genes present in a single copy in all genomes were extracted from the clustered families and complemented with ribosomal protein genes (which are duplicated in B. bacilliformis) and groEL, groES and gltA (which are duplicated in the outgroup species). This resulted in a dataset of 428 genes. The mean substitution frequencies for these genes was estimated to less than 0.12 substitutions per nonsynonymous site (Ka) and up 1.09 substitutions per synonymous site (Ks) in all pairwise comparisons within the genus Bartonella (Table S3). Synonymous substitution frequencies between Bartonella species and the outgroup were above saturation level (Table S3).
The phylogenetic analyses based on the concatenated nucleotide alignment using maximum likelihood and Bayesian methods revealed three highly supported groups (Figure 1A). Group A includes the ruminant-adapted strains, group B contains isolates from cats, rats and squirrels and group C is composed of the human pathogen B. quintana and of strains isolated from cats, dogs and rodents. Moreover, the phylogeny placed B. australis as the earliest diverging species with high support (bootstrap support = 100%; posterior probability = 1.0). This finding was surprising since earlier studies have placed B. australis near to the C-group strains B. grahamii and B. tribocorum [20]. We tested five possible placements of B. australis in single gene trees using both nucleotide and amino acid alignments, and found that in 67% to 85% of all trees the likelihood was the highest for a placement as either the earliest diverging species (Figure 1B, same topology as inferred from the concatenated nucleotide alignment, Figure 1A) or as a sister species to group B (Figure 1C) (Figure S4). Only 5% to 15% of the trees indicated a placement near to group C.
The B. australis genome is slightly more GC-rich than the other genomes and we reasoned that its placement as the deepest diverging species could be due to attraction to the more GC-rich outgroup taxa for rapidly evolving genes. Indeed, sequence divergence level was the factor that most strongly correlated with tree topologies: the most divergent gene sets placed B. australis as the earliest diverging species (pp = 0.98; bootstrap = 77–93%) (Figure 1B), whereas the least divergent gene sets placed B. australis as a sister taxa to the B-group strains (pp = 0.94; bootstrap = 90–95%) (Figure 1C) (Figure S5). Consistently, a t-test revealed a significant (p = 0.039) correlation between the mean Ka-values and the placement of B. australis as the deepest diverging lineage. To examine the support for the rooting, we calculated the likelihood of seven alternative placements of the root. In 70% of the single gene trees inferred from nucleotide sequences but in only 35% of the trees built from amino acid alignments, the highest likelihood was associated with a rooting on the B. australis branch (Figure 1B) (Figure S6).
To further improve the rooting in the analysis, we included Bartonella tamiae Th239, whose draft genome was recently released in Genbank. This was for two reasons: the genomic GC content is 38%, which is similar to the GC content of the other Bartonella genomes, and a previous phylogenetic analysis based on the neighbor-joining method with the gltA gene indicated that it is a very early diverging species, although there was no bootstrap support for this placement [27]. To reduce the influence of putative horizontal gene transfers, we applied a discordance filter to a concatenated amino-acid alignment of 425 proteins [28] in which 0 to 40% of the proteins producing the most deviant tree topologies were removed. The tree topology obtained from an alignment where 10% of the genes were removed showed that B. tamiae is the earliest diverging species with 100% bootstrap support, followed by B. australis, with 96% bootstrap support (Figure S7). Importantly, the topology was unchanged even after the removal of up to 40% of the most discordant genes (Figure S8). This suggests that the tree topology is not affected by the inadvertent inclusion of a subset of horizontally transferred genes with a different evolutionary history. Moreover, the analysis provided additional support to the hypothesis for the placement of B. australis as the earliest diverging species (excluding B. tamiae), although a small set of highly conserved genes indicates that it clusters with the B group strains.
Both of these two topologies (Figure 1B and 1C) are in conflict with the proposal that B. australis is related to the rodent strains B. grahamii and B. tribocorum [20]. Since we have sequenced the exact same strain as deposited by the authors [20], the discrepancy resides in the use of different genes and gene sequences for the phylogeny. We argue that the previous clustering with rodent Bartonella species was a PCR artifact since only 4 of the 6 B. australis genes that were used to construct the published tree are identical in sequence to those of the B. australis genome, and none of these 4 genes, including the gltA gene, supported a grouping with the rodent isolates (data not shown). Consistently, a previously published phylogeny of Bartonella species based on the gltA gene also indicated that B. australis clusters separately from B. grahamii and B. tribocorum [27]. The other two PCR amplified genes (rpoB and ftsY) differ in sequence from those of the B. australis genome, suggesting that they may have been amplified from DNA of a contaminating rodent Bartonella species. This would explain the affiliation of B. australis with B. grahamii and B. tribocorum in the study by Fournier et al. [20].
The clustering of B. bacilliformis with the ruminant A-group strains was consistently observed in all analyses (Figure 2; Figures S4, S5, S6, S7, S8). This finding is notable since all previous attempts to identify a sister clade of B. bacilliformis that could provide indications of its natural host reservoir have failed. The lack of affiliation with other Bartonella species has been taken as an argument to suggest that B. bacilliformis represents the earliest diverging species in the genus and that this species has retained more ancestral features than the other species [13], [24], [29] (Figure 1D). Our phylogeny is inconsistent with this hypothesis, and, to the best of our knowledge, no other circumstantial data suggest that B. bacilliformis is more “ancestral” than the other species. The new placement of B. bacilliformis as a sister clade to Group A is most likely due to the inclusion of B. australis and as many as four different A-group species in our concatenated genome tree.
Rather than representing ancestral features, the small genome size and the high virulence properties of B. bacilliformis are characteristic of reductive genome evolution following a recent host switch, as observed for B. quintana, the agent of trench fever [30] and Rickettsia prowazekii, the agent of epidemic typhus [31]. In analogy, we suggest that the establishment of B. bacilliformis in the human population was a rare event that was started from a small founder population that successfully made a host shift from ruminants, possibly from local camelids, to humans. The global distribution of ruminant Bartonella species and ruminant-infecting vectors represent a large reservoir of ruminant-infecting Bartonella species. But even if incidental infections by ruminant Bartonella species occur frequently, the lack of global vector systems may restrict their establishment in the human population. Indeed, the small geographic area in South America in which infections with B. bacilliformis occurs is thought to correspond to the geographic area of the sandfly transmitting the pathogen [32].
Having inferred a robust species phylogeny, we set out to identify genes acquired in the Bartonella last common ancestor (BLCA), thereby providing clues to the basis of the subsequent radiation of Bartonella species with different host preferences. To this end, we clustered all CDSs in the 16 Bartonella genomes and the outgroup species into families of protein homologs, resulting in 11,315 protein families of which 1828 families contained Bartonella proteins. Some families contain only orthologs encoded by single-copy genes whereas other families are larger and contain several paralogous proteins. We inferred the loss and gain of the protein families along the branches of the tree (Figure 2) with parsimony character mapping [33], [34]. For the interpretation of the gene content analyses we have considered both of the supported topologies (“BAnh1 early”, Figure 1B, and “Radiation”, Figure 1C). The analysis revealed in both cases the loss of about 1,500 protein families in the BLCA, which was balanced by the acquisition of about 100 protein families (Figure 2), resulting in a large efflux of genes. This estimate of acquired functions does not include duplicated and rapidly evolving surface proteins (see below) that are too divergent in sequence and/or size to be included in distinct protein families, given the criteria used for protein clustering.
To establish a complete host-arthropod lifecycle, Bartonella must be able to attach to both nucleated host cells and erythrocytes [38]. Attachment to nucleated cells is mediated by autotransporter adhesins encoded by the badA genes [39], which are tandemly duplicated and located outside the SSC region, while the two gene clusters shown to be important for internalization into endothelial cells and for binding to erythrocytes, virB and trw, are located inside the SSC region in most genomes. Since host-specificity presumably resides in the sequences and structures of proteins that mediate these internalization processes, and since these are likely to have changed in response to an expansion or alteration of the host range, it is of interest to identify and characterize all clade- and species-specific gene acquisitions.
We can think of two hypotheses to explain the linkage of gene clusters for secretion systems to the phage-derived origin of replication and the gene transfer agent: selection for regulation of gene expression or for enhanced recombination rate and enforced collaboration among cells. According to the regulation hypothesis, this may be a mechanism to regulate the level of transcription of the acquired secretion systems and surface proteins. Given the slow doubling time of Bartonella, the amplified genes could serve as templates for transcription before being degraded, packaged or recombined back into the genome. Such a process will result in an increase of the copy number that correlates inversely with the distance from the ROR region, and hence the mere placement of the gene could potentially influence its expression level. A rapid increase in gene copy numbers could be beneficial during invasion of host cells or at other critical life stages of Bartonella.
However, we consider such a regulatory scheme in its simplest form unlikely. First of all, a microarray study of expression changes in B. grahamii showed that although genes for secretion systems and phages located inside the amplified segment were upregulated, many core genes located in the same segment were not [9], arguing against a dosage effect. Moreover, copy number increases are expected to be similar for the virB and trw gene clusters for T4SSs, yet the virB gene cluster is expressed during invasion of endothelial cells [50], whereas the trw genes are expressed during invasion of erythrocytes [17], [40]. This does not preclude that increased copy numbers enhances the expression levels of some genes, but other regulatory signals must also be involved.
Alternatively, selection may act to increase levels of gene transfer and recombination of the secretion systems. Recombination can generate diversity that allows for rapid adaptation (to new hosts in the case of Bartonella), but is associated with the cost of disadvantageous alterations of core genes and disruptions of beneficial allele combinations [51]. With recombination targeting a specific chromosomal region, the positive aspects of recombination would be preserved, while the cost would largely be avoided.
There are several lines of evidence favoring this hypothesis. First, as shown by Touchon et al. [52], the amount of recombination correlates well with GC content at third codon position and we observed a gradual increase in GC content over the amplified region that follows the level of amplification with the peak located in the ROR region (Figure 9A, 9B). Second, it has been shown that DNA segments located in the amplified region are more frequently encapsidated into phage particles, increasing the likelihood for recombination with homologous DNA sequences in other bacterial cells [9]. Third, the higher GC content in the amplified region correlates with a higher abundance of imported genes (Figure 9C), although the increase in GC content is not restricted to the imported genes but also observed for core genes in the same region. Imported genes are here defined as genes present in any Bartonella strain, with homologs in distantly related bacteria but not in the outgroup species. Fourth, the fraction of synteny blocks shared among most or all Bartonella genomes is lowest in this region (Figure 9D). Finally, higher fixation rates for recombination events than in the genome overall has been demonstrated for several genes within the amplified region, including genes in the virB and trw clusters [10], [18]. Thus, the incorporation and maintenance of a gene transfer agent in the Bartonella genome may primarily be a mechanism to facilitate recombination and thereby speed up the evolution of adaptive traits.
Although the latter hypothesis is supported by the data and seductive at species level, it remains to be explained how it functions at the level of the individual bacterial cell since such a collaborative system is prone to cheaters in the population. This may not be a problem in the case of genes coding for surface components that are required for an individual bacterial cell to attach to a receptor on the host cell surface. In these cases, selection will act on the maintenance of these systems in every single bacterial cell rather than on the population as a whole. However, genes coding for proteins that are secreted into the host cytoplasm to modulate host cellular functions and pathways, such as the bep genes and possibly the ctxA gene homologs, should be prone to deletions in the individual cell. Secreted proteins are of importance for the population at large, but often costly to make for the individual cell. The population is therefore vulnerable to the emergence of cheaters who rip off the benefits without contributing to the production of the common good. This is particularly true for the bep genes, which are expressed by a few bacterial cells to block endocytosis, thereby making it possible for the large majority of cells, which may not necessarily need to have the bep genes, to enter the host cell through the formation of the invasome. Experimental studies have identified cheaters in bacterial populations that depend on the secretion of iron-scavenging molecules [53], digestive enzymes and quorum-sensing molecules [54] and antibiotic resistance factors [55], [56]. Despite the risk for collapse of the social trait, a dilemma referred to as “the tragedy of the commons” [57], [58] secretion is widespread in bacteria, not the least in host-adapted bacteria.
To solve this problem, mechanisms must exist that force selfish cells to cooperate. One strategy to enforce a cooperative behavior is to associate genes for secreted compounds with mobile elements, thereby ensuring a constant re-acquisition of the genes [59]. Indeed, genes for secreted proteins are overrepresented on mobile elements in 22 Escherichia and Shigella genomes [60]. Thus, the benefits of associating genes for secreted molecules with a GTA and a separate origin of replication could be several-fold: to generate variability and rapid spread of beneficial genes as well as to enforce collaboration among cells in the population. This could help explain why “social” genes coding for secreted effector proteins tend to be located inside the amplified SSC region, whereas genes for other surface molecules that mediate attachment and generate a benefit only to the bacterial cell in which the protein is produced may be located elsewhere.
This study provides the broadest genomic study so far of vector-borne bacteria that infect mammals. Contributing 6 new Bartonella genome sequences to the previously sequenced 10 genomes, we identified a recently discovered gene transfer agent (BaGTA) and a phage-derived origin of replication as the major innovations in the last common ancestor of Bartonella that have since been maintained in all species. In contrast to a rapid turnover of other phage genes, the BaGTA and the phage-derived origin of replication are highly conserved in sequence and gene order and they are located at the same sites in all Bartonella genomes examined in this study, suggesting a single integration event. Previous studies of B. grahamii have shown that the phage-derived origin of replication drives the amplification of a surrounding chromosomal segment of several hundred kb, which is then digested into 14 kb DNA fragments and packaged into the phage particles [9]. In this study, we have extended these findings by demonstrating that the BaGTA can transfer genes between different strains of B. henselae. Because of the extreme conservation of the BaGTA and the associated origin of replication, it is reasonable to believe that the BaGTA serve as a gene transfer agent in all species examined here. Located within the amplified segment are many species- and clade-specific gene clusters for autotransporters and other secretion systems. In total, at least eight gene clusters for putative adhesins and secretion systems have been integrated into this region. Taken together, this suggests that the main role of the BaGTA is to shuffle genes for secretion systems located in this region among cells in the Bartonella populations.
We hypothesize that the plasticity given to this region by the BaGTA has rescued Bartonella from the reductive evolutionary processes that are operating on host-specialized bacteria adapted to otherwise sterile intracellular environments [61]. Previous studies have associated adaptive radiation in Bartonella with two independent acquisitions of the virB gene clusters for T4SSs along with genes for associated effector proteins [13]. Our tree topology suggests that the two clades containing these systems are not sister groups, which provides additional support to the hypothesis of two independent events. Experimental studies on the virB and trw gene clusters for type IV secretion systems have shown that they are essential for the infection of endothelial cells and erythrocytes, respectively. Amplification and recombination of this region facilitates duplication and functional divergence, contributing to the generation of a diverse set of proteins that can bind to a variety of host cell molecules in a broad range of hosts.
This organization may also facilitate reintroduction of the secretion systems into cells and populations that have lost them. Since secretion systems and adhesins are outer surface proteins that may be recognized by the host immune system, selection may favor downregulation or loss at some stages during the infection. For example, in regions where hosts and vectors are abundant and the propensity for super-infections is high it may be beneficial to shed or modulate some of these genes. We observed here that one of two closely related moose isolates have lost most of the genes for a flagellar type III secretion system. Likewise, we have shown previously the presence/absence patterns of genes for filamentous hemagglutinin (a type V secretion system), differs in mouse-infecting B. grahamii populations sampled from geographically close locations [62]. An additional benefit is that repeated reintroduction of these genes provides a mechanism that forces all cells in the population to contribute to the production of the secreted molecules. More detailed studies of the evolution of these systems during different stages of the infection in individual hosts could help clarify the interplay between secretion systems, infection and immunity.
The results presented in this paper contribute to a greater general understanding of the evolutionary significance of gene transfer agents and their role in driving adaptive radiation processes. The identification of BaGTA is remarkable given the prevalence of RcGTA-like gene clusters in most other Alphaproteobacteria, including genera such as Brucella and Agrobacterium that are closely related to Bartonella. The two GTAs are not related, suggesting that BaGTA has replaced RcGTA. Features associated with BaGTA, such as the phage-derived origin of replication and the amplification has not been identified in genomes that host the RcGTA. Future analyses of conserved phage elements will be needed to determine whether the association of gene transfer agents with adaptive traits is a common theme in bacterial genomes.
The type strains Bartonella australis NH1, isolated from a grey kangaroo [20], and B. bovis (Bermond) 91-4, isolated from a cow [21] were obtained from the Culture Collection at the University of Göteborg, Sweden (CCUG numbers 51999 and 43828, respectively). The strains Bartonella vinsonii berkhoffii Winnie [22] and Tweed [23], isolated from dogs, were obtained from Ricardo Maggi and Edward Breitschwerdt (Department of Clinical Sciences, College of Veterinary Medicine, North Carolina State University, Raleigh, USA). The strain Bartonella bovis m02, isolated from a moose in Sweden, was obtained from Martin Holmberg (Department of Medical Sciences, Uppsala University, Uppsala, Sweden). The strain Bartonella schoenbuchensis m07a was isolated from a blood sample collected from a moose close to Dyltabruk, Örebro county, Sweden, on October 28, 2007. The sample was cooled after collection, and stored at −80°C for 30 days before cultivation on hematin agar plates in a humidified 5% CO2 incubator at 35°C for 5–7 days.
Figure 3, Figure 4, Figure 6A, Figure 8, Figure S2, and Figure S10 were drawn with genoPlotR 0.7 [88]. Figure 9 was obtained with R 2.12.1 [89]. All trees were drawn with FigTree (Andrew Rambaut, available on the author's website: http://tree.bio.ed.ac.uk/software/figtree/).
|
10.1371/journal.pbio.0050203 | Targeted Inhibition of miRNA Maturation with Morpholinos Reveals a Role for miR-375 in Pancreatic Islet Development | Several vertebrate microRNAs (miRNAs) have been implicated in cellular processes such as muscle differentiation, synapse function, and insulin secretion. In addition, analysis of Dicer null mutants has shown that miRNAs play a role in tissue morphogenesis. Nonetheless, only a few loss-of-function phenotypes for individual miRNAs have been described to date. Here, we introduce a quick and versatile method to interfere with miRNA function during zebrafish embryonic development. Morpholino oligonucleotides targeting the mature miRNA or the miRNA precursor specifically and temporally knock down miRNAs. Morpholinos can block processing of the primary miRNA (pri-miRNA) or the pre-miRNA, and they can inhibit the activity of the mature miRNA. We used this strategy to knock down 13 miRNAs conserved between zebrafish and mammals. For most miRNAs, this does not result in visible defects, but knockdown of miR-375 causes defects in the morphology of the pancreatic islet. Although the islet is still intact at 24 hours postfertilization, in later stages the islet cells become scattered. This phenotype can be recapitulated by independent control morpholinos targeting other sequences in the miR-375 precursor, excluding off-target effects as cause of the phenotype. The aberrant formation of the endocrine pancreas, caused by miR-375 knockdown, is one of the first loss-of-function phenotypes for an individual miRNA in vertebrate development. The miRNA knockdown strategy presented here will be widely used to unravel miRNA function in zebrafish.
| The striking tissue-specific expression patterns of microRNAs (miRNAs) suggest that they play a role in tissue development. These small RNA molecules (∼22 bases in length) are processed from long primary transcripts (pri-miRNA) and regulate gene expression at the posttranscriptional level. There are hundreds of different miRNAs, many of which are strongly conserved. Vertebrate embryonic development is most easily studied in zebrafish, but genetically disrupting miRNA genes to see which miRNA does what is technically challenging. In this study, we interfere with miRNA function during the first few days of zebrafish embryonic development by introducing specific antisense morpholino oligonucleotides (morpholinos have been used previously to interfere with the synthesis of the much larger mRNAs). We show that morpholinos targeting the miRNA precursor can block processing of the pri-miRNA or directly inhibit the activity of the mature miRNA. We also used morpholinos to study the developmental effects of miRNA knockdown. Although we did not observe gross phenotypic defects for many miRNAs, we found that zebrafish miR-375 is essential for formation of the insulin-secreting pancreatic islet. Loss of miR-375 results in dispersed islet cells by 36 hours postfertilization, representing one of the first vertebrate miRNA loss-of-function phenotypes.
| MicroRNAs (miRNAs) have a profound impact on the development of multicellular organisms. Animals lacking the Dicer enzyme, which is responsible for the processing of the precursor miRNA into the mature form, cannot live [1–3]. MiRNA mutants have been described only for Caenorhabditis elegans and Drosophila, reviewed in [4]. From these studies, it is clear that invertebrate miRNAs are involved in a variety of cellular processes, such as developmental timing [5,6], apoptosis [7,8], and muscle growth [9]. Analysis of conditional Dicer null alleles in mouse has indicated a general role for miRNAs in morphogenesis of the limb, skin, lung epithelium, and hair follicles [10–13]. Overexpression studies in mouse have implicated specific vertebrate miRNAs in cardiogenesis and limb development [14,15]. In zebrafish, embryos lacking both maternal and zygotic contribution of Dicer have severe brain defects [2]. Strikingly, the brain phenotype of maternal-zygotic Dicer zebrafish can be restored by injection of miR-430, the most abundant miRNA in early zebrafish development. Despite all these studies describing functions for miRNAs in development, no vertebrate miRNA mutant has been described to date. Genetically, it is challenging to obtain mutant miRNA alleles in zebrafish, because their small size makes them less prone to mutations by mutagens, and for many miRNAs, there are multiple alleles in the genome or they reside in families of related sequence.
Temporal inhibition of miRNAs by antisense molecules provides another strategy to study miRNA function. 2′-O-methyl oligonucleotides have been successfully used in vitro and in vivo to knock down miRNAs [16–18]. Morpholinos are widely applied to knock down genes in zebrafish development [19] and have recently been used to target mature miR-214 in zebrafish [20]. However, off-target phenotypes are often associated with the use of antisense inhibitors.
Here, we show that morpholinos targeting the miRNA precursor can knock down miRNAs in the zebrafish embryo. Several independent morpholinos can knock down the same miRNA, and these serve as positive controls to filter out off-target effects. Morpholinos can block miRNA maturation at the step of Drosha or Dicer cleavage, and they can inhibit the activity of the mature miRNA. We show that inhibition of miR-375, which is expressed in the pancreatic islet and pituitary gland of the embryo [21], results in dispersed islet cells in later stages of embryonic development, whereas no effects were observed in the pituitary gland. The morpholino-mediated miRNA knockdown strategy presented here, is an extremely fast and well-controlled method to study miRNA function in development.
Since it is difficult to obtain a genetic mutant for a miRNA in zebrafish, we looked for alternative strategies to deplete the embryo of specific miRNAs. Antisense molecules such as 2′-O-methyl and locked nucleic acid (LNA) oligonucleotides have been used to inhibit miRNAs in cell lines [16,18,22], Drosophila embryos [23], and adult mice [17]. We tried to use these molecules to inhibit the function of endogenous miRNAs in the zebrafish embryo. Although they can be used to suppress the effects of miRNA overexpression [24], injection of higher concentrations required to obtain good knockdown of endogenous miRNAs resulted in toxic effects, when injecting 1 nl solution at a concentration of approximately 10 μM and 50 μM for LNA and 2′-O-methyl oligonucleotides, respectively (unpublished data). Therefore, we switched to morpholinos because these are widely used to inhibit mRNA translation and splicing in zebrafish embryos [19], and have also been shown to target miRNAs in the embryo [2,20,24]. We injected 1 nl of 600 μM morpholino solution with a morpholino complementary to the mature miR-206 in one- or two-cell–stage embryos. Subsequently, embryos were harvested at 24, 48, 72, and 96 hours postfertilization (hpf), and subjected to in situ hybridization and Northern blotting (Figure 1A and 1B). This analysis showed that the mature miRNA signal is suppressed up to 4 d after injection of the morpholino. The knockdown effect was specific for this miRNA; parallel in situ analysis of the same embryos with a probe for miR-124 did not show any effects on expression of this miRNA (Figure 1B). Thus, miRNA detection can be specifically and efficiently suppressed during embryonic and early larval stages of zebrafish development using morpholinos antisense to the mature miRNA.
The zebrafish embryo can be used to monitor the effect of miRNAs on green fluorescent protein (GFP) reporters fused to miRNA target sites [24]. To determine the effect of a morpholino in this assay system, we constructed a GFP reporter for miR-30c and tested it in the presence and absence of a mature miR-30c duplex. Injected miR-30c silences this GFP reporter, which is in line with previous reports using similar strategies in the embryo (Figure 1C) [2,20,24]. Co-injection of the miR-30c duplex and a morpholino targeting mature miR-30c rescues the reporter signal, whereas injection of a control morpholino did not reverse the silencing by miR-30c. These data indicate that a morpholino can block the activity of a mature miRNA duplex in a functional assay .
There are three possible explanations for the observed reduction in the detection signal for a miRNA that is targeted by a morpholino. First, the hybridization of a morpholino could disturb isolation of the miRNA. Second, the morpholino could destabilize the miRNA. Third, the morpholino could inhibit the maturation of the miRNA.
To examine the effect of a morpholino on the isolation of a mature miRNA, we incubated a mature miR-206 duplex and a control duplex (miR-205) with a morpholino against miR-206 in vitro. After isolation, samples were analyzed by Northern blotting for the presence of miR-206 and miR-205. We could still detect miR-206, indicating that there is no effect of the morpholino on the RNA isolation procedure (Figure 1D). However, when morpholino and miRNA duplex were incubated together in vitro and loaded on a denaturing gel without isolation, we observed a decrease in the signal for miR-206, indicating that the morpholino can bind to the miRNA in vitro and still does so in the denaturing gel.
Next, we wanted to know whether a morpholino could affect the stability of a mature miRNA in vivo. Therefore, we injected a mature miR-206 and a control duplex (miR-205) together with a morpholino against miR-206 in the embryo. After incubation for 8 h, RNA was isolated and subjected to Northern blot analysis to probe for injected miR-206 and injected miR-205. In contrast to the data obtained for endogenous miR-206, there was no decrease observed in the amount of injected miR-206 in the morpholino-injected embryos (Figure 1D) (endogenous miR-206 is not yet expressed at this stage).
Since these data show that there is no effect of a morpholino on miRNA isolation or stability, we conclude that morpholinos deplete the embryo of miRNAs by inhibiting miRNA maturation. If this is the case, then we expect morpholinos targeting other regions of the miRNA precursor to act as well as the morpholinos designed against the mature miRNA, and this is indeed what we find (see next section).
Injection of antisense oligos in embryos might result in off-target effects. Thus, phenotypic data retrieved from antisense knockdown experiments should be treated with caution. In Drosophila, 2′-O-methyl oligo–mediated knockdown of embryonically expressed miRNAs caused defects that clearly differed from the phenotype of the corresponding knockout fly [9,23]. In sea urchin experiments, off-target effects of morpholino knockdowns are well documented, though low incubation temperatures favor off-target interactions [25].
To filter out off-target effects, we sought a control strategy that would allow us to compare effects of morpholinos with independent sequences targeted to the same miRNA. Because our data on morpholinos targeting the mature miRNA suggested that miRNA biogenesis might be affected, we designed morpholinos targeting the Drosha and Dicer cleavage sites of the precursor miRNA (Figure 2A). We decided to test this strategy on miR-205, since it is expressed relatively early, and there are only two, but identical, copies in the fish genome. Four different morpholinos were designed to inhibit miR-205 biogenesis: two targeting the Drosha cleavage site complementary to either the 5′ or 3′ arm of the stem, and two morpholinos similarly targeting the Dicer cleavage site (Figure S1). These morpholinos were injected under similar conditions as described for miR-206 and compared to the morpholino targeting mature miR-205. Interestingly, all five morpholinos induced complete or near-complete loss of miR-205 (Figure 2B).
Many miRNAs are highly expressed during later stages of embryonic development [21]. Therefore, we tested how long the effect of the morpholinos would last. Although for this series of morpholinos the knockdown is best at 24 hpf, the effect is still significant up to 72 hpf (Figure 2C).
Next, we tested a similar series of morpholinos against the miR-30c precursor and analyzed miR-30c expression by Northern blotting (Figure S2). However, we only observed knockdown for the morpholino targeting mature miR-30c, but not for the other four morpholinos targeting the miR-30c precursor. This could be because miR-30c resides in a family of closely related species, with more sequence variability in the regions outside of the mature miRNA. The precursors of the family members might not all be targeted by these morpholinos (Figure S2). Thus, not all miRNAs are equally prone to knockdown by morpholinos that target the miRNA precursor.
To investigate the effect of morpholinos on exogenously introduced pri-miR-205, we injected mRNA derived from a GFP construct with pri-miR-205 in the 3′ UTR. Again, we could not detect mature miR-205 derived from this construct after targeting by morpholinos (Figure 2D). Interestingly, the miR-205 precursor also could not be detected in the embryos co-injected with morpholinos, whereas pre-miR-205 could be detected in the absence of morpholinos (Figure 2D). Because pri-miR-205 was cloned in the 3′ UTR of GFP, we monitored GFP fluorescence after injection of this construct. In the presence of a morpholino, GFP fluorescence increased (Figure 2E), suggesting accumulation of the primary miRNA. Therefore, we performed reverse transcriptase PCR (RT-PCR) on 8-h-old embryos injected with GFP-pri-miR-205 and a control mRNA (luciferase) (Figure 2F). In the presence of a morpholino, the GFP-pri-miR-205 mRNA level is higher compared to control embryos that were not injected with morpholinos. This experiment confirms the GFP data and shows that morpholinos targeting the miRNA precursor inhibit Drosha cleavage.
Next, we tested whether processing of the pre-miRNA might also be inhibited by morpholinos. Therefore, we injected a miR-205 precursor in the one-cell–stage embryo. Northern analysis showed that the precursor was processed into mature miRNA in the embryo (Figure 2G). However, co-injection of the overlap loop and non-overlapping loop morpholinos blocked processing completely. There was only a little effect of morpholinos targeting the Drosha cleavage site, probably because they only partially overlap the precursor.
A similar analysis was performed for miR-375, which is expressed in the pancreatic islet and pituitary gland [21], and has two copies in the zebrafish genome, which differ in the regions outside the mature miRNA.
Overlap loop and loop morpholinos were designed for both miR-375–1 and miR-375–2, and a morpholino against the miRNA star sequence could be used to target both copies of miR-375 simultaneously (Figure 3A). The efficacy of all morpholinos was assessed by determining their effect on injected pri-miR-375–1 or pri-miR-375–2 transcripts (Figure 3B). As expected, each morpholino targeted the transcript to which it was directed. However, the star miR-375 morpholino did not knock down miR-375 completely. In addition, morpholino oligonucleotide (MO) miR-375 did not interfere with processing of miR-375 from pri-miR-375–1, possibly because this primary transcript forms a more stable hairpin. In all cases, the lack of a signal for mature miR-375 coincided with the absence of pre-miR-375, which could be detected in the absence of a complementary morpholino.
Next, all morpholinos were injected separately and in combination, and embryos were subjected to Northern blotting to determine endogenous miR-375 expression at 24 and 48 hpf (Figure 3C). In contrast to the results obtained by in situ hybridization (see last section), the morpholino to mature miR-375 only slightly decreased the expression of miR-375. However, MO miR-375 could inhibit the activity of a mature miR-375 duplex in a GFP-miR-375-target reporter assay (Figure 3E). The morpholinos targeting only one copy of miR-375 reduced miR-375 expression, with the strongest effect for the morpholinos targeting pri-miR-375–1. However, simultaneous injection of morpholinos targeting pri-miR-375–1 and pri-miR-375–2 completely knocked down mature miR-375, indicating that both transcripts are expressed.
To further determine the contribution of each transcript to mature miR-375 accumulation, we performed in situ hybridization for pri-miR-375–1 and pri-miR-375–2 (Figure 3D). Both transcripts could not be detected in wild-type embryos. However, pri-miR-375–1 was detected in the pancreatic islet and the pituitary gland in embryos injected with the miR-375–1 loop morpholino and the morpholino to miR-375 star. Similarly, pri-miR-375–2 was only detected in embryos injected with the miR-375–2 loop morpholino, the morpholino to miR-375 star and mature miR-375. Thus, both transcripts are expressed in the pituitary gland and the pancreatic islet, similar to miR-1 in the developing mouse heart [15]. Together, this indicates that these morpholinos inhibit primary miRNA processing and result in primary miRNA accumulation, as we described for miR-205.
In conclusion, our data demonstrate that morpholinos targeting the miRNA precursor can interfere with primary miRNA processing at either the Drosha or Dicer cleavage step and that morpholinos targeting the mature miRNA can inhibit their activity in a functional assay. Taken together, our data show that different morpholinos targeting the same miRNA may serve as positive controls for miRNA knockdown phenotypes in the embryo.
To identify functions for individual miRNAs in zebrafish embryonic development, we knocked down a series of 11 conserved vertebrate miRNAs (Table S1) and analyzed their expression after morpholino knockdown. Injected embryos were monitored phenotypically by microscopic observation until four days postfertilization (dpf). Knockdown of most miRNAs resulted in loss of in situ staining for the respective miRNA. However, we could not observe gross morphological malformations after knockdown of these miRNAs (Figure 4A). Therefore, we analyzed embryos injected with morpholinos against miR-182, miR-183, or miR-140 in more detail, because we could easily stain the tissues that express these miRNAs (Figure 4B). Embryos injected with morpholinos against miR-182 or miR-183, which are expressed in the lateral line neuromasts and hair cells of the inner ear, were treated with DASPEI, which stains hair cells. Embryos injected with a morpholino against miR-140, which is expressed in cartilage, were subjected to Alcian Blue staining, a cartilage marker. However, staining of these specific cell types that express the miRNA did not uncover any defects upon knockdown (Figure 4B).
In conclusion, knockdown of many miRNAs does not appear to significantly affect zebrafish embryonic development, at least not to the extent that can be visualized by the methods used in these examples.
MiR-375 is known to be expressed in the pancreatic islet and the pituitary gland, and was first isolated from pancreatic beta cells [21,26]. This miRNA is conserved in vertebrates and may regulate insulin secretion by inhibiting myotrophin [26].
We injected a morpholino against mature miR-375 into the one-cell–stage embryo. This morpholino effectively knocked down miR-375 in the first 4 d of development (Figure 5A), and it could also block the activity of an injected miR-375 duplex, as monitored by its effect on a GFP reporter silenced by miR-375 (Figure 3E).
During the first 5 dpf, there was no clear developmental defect except for a general delay in development. At around 7 dpf, approximately 80% of the injected embryos died. Next, we analyzed the development of both the pituitary gland and the pancreatic islet, by in situ hybridization with pit1 and insulin markers. This analysis revealed no change in the formation of the pituitary gland (Figure 5B). However, analysis of insulin expression showed a striking malformation of the islet cells in 3-d-old morphant embryos (Figure 5B). Wild-type embryos have a single islet at the right side of the midline, whereas the miR-375 knockdown embryos have dispersed insulin-positive cells. The effect is sequence specific, because a morpholino complementary to the mature miR-375 morpholino inhibited the pancreatic islet phenotype (Figure 5E).
The pancreatic islet consists of four cell types, α, β, δ, and PP, expressing glucagon, insulin, somatostatin, and pancreatic polypeptide, respectively. Insulin is the first hormone expressed, and somatostatin co-localizes partially with insulin, whereas glucagon-expressing cells are distinct [27]. A more detailed analysis using somatostatin and glucagon as marker genes revealed a similar pattern of scattered islet cells in the miR-375 morphant (Figure 5C).
In zebrafish, insulin is first expressed at the 12-somite stage in a few scattered cells located at the midline, dorsal to the yolk [28]. Insulin-positive cells migrate posteriorly and converge medially to form an islet by 24 hpf. To look at the development of the pancreatic islet in time, we collected MO miR-375 and noninjected control embryos at different stages, and investigated the expression of insulin (Figure 5D). At the 16-somite stage, insulin-positive cells are scattered at the midline in both noninjected and MO miR-375–injected embryos, and a presumptive islet is formed by 24 hpf. Subsequently, when the insulin-positive islet is moving to the right side of the embryo in later stages, the islet breaks apart and insulin-positive cells become scattered in morphant embryos (Figure 5D). Also, in later stages, the phenotype persists, although miR-375 is re-expressed at approximately 5 dpf in morpholino-injected embryos (Figure 5A).
Next, we analyzed the effect of all miR-375 control morpholinos described in the previous section, by staining for insulin (Figure 6A). Both the dispersion phenotype and the knockdown were striking for embryos injected with MO miR-375. Injection of the overlap loop and loop morpholinos targeting pri-miR-375–1 also resulted in scattered insulin-positive cells at 72 hpf, although the effect was weaker compared to MO miR-375. The miR-375–2 loop and overlap loop morpholinos hardly induced any scattering of insulin-positive cells, whereas the effect was very strong in embryos injected with morpholinos to pri-miR-375–1 and −2 simultaneously. The effect of the miR-375 star morpholino on insulin-positive cells was moderate compared to MO miR-375.
To further prove the specificity of the pancreatic islet phenotype, we injected two control morpholinos against let-7 and miR-124 and analyzed these for miR-375 and insulin expression. None of these control morpholinos showed loss of miR-375 expression or abnormal development of the islet cells (Figure 6A).
Next, we analyzed miR-375 knockdown embryos with markers staining the endocrine or exocrine pancreas (Figure 6B). Similar to insulin staining, islet1 expression showed dispersed islet cells in embryos of 48 hpf and 72 hpf, but not 24 hpf. Embryos injected with MO miR-375 exhibited delayed development of the exocrine pancreas, liver, and gut as shown by ptf1a and foxa2 staining. At 72 hpf, these markers showed a similar pattern in MO miR-375–injected embryos as in noninjected embryos at 48 hpf. However, co-injection of miR-375–1/2 loop morpholinos did not delay development of the exocrine significantly, but these embryos still displayed the scattered insulin-positive cells (Figure 6A). This shows that loss of miR-375 mainly results in malformation of the endocrine pancreas, whereas surrounding tissues that do not express miR-375 are not affected.
Functional data on miRNAs in vertebrate development have been obtained mainly from overexpression studies and analysis of conditional Dicer knockouts. For example, the role of miR-430 in zebrafish brain morphogenesis has become clear from experiments that rescued Dicer null mutants by injection of an miRNA duplex that mimicked a miR-430 family member [2].
MiRNA expression can be conveniently studied in zebrafish embryos. However, dissecting miRNA function by disrupting miRNA genes is difficult in zebrafish, because the miRNA is too small to efficiently search for mutations by a target-selected mutagenesis approach [29]. In addition, it is unclear what such point mutations would do to processing or function of the miRNA.
It has been shown previously that morpholinos can target miRNAs in the zebrafish embryo [20,24]. In a recent study, mature miR-214 was targeted by a morpholino in zebrafish, and this resulted in a change in somite shape, reminiscent of attenuated hedgehog signaling [20]. Although the phenotype could be rescued by simultaneous inhibition of a negative regulator of hedgehog signaling, no positive control morpholinos were reported that could mimic the phenotype. In addition, data were lacking that showed an effect of the morpholino on endogenous miR-214 levels.
The results in this paper show that morpholinos targeting the miRNA precursor form a reliable and efficient tool to deplete the embryo of miRNAs during the first 4 d of development, when most organ systems are formed and miRNAs are expressed. We have shown that miRNA expression can be inhibited by targeting the mature miRNA, the precursor miRNA or the primary miRNA. Our data show that such morpholinos can inhibit miRNA processing at the Drosha cleavage step or the Dicer cleavage step, probably by steric blocking, although the exact mechanism is unclear. In addition, morpholinos targeting the mature miRNA can inhibit their activity, probably by preventing binding to a target mRNA.
We used morpholinos targeting the mature miRNA for a set of 13 conserved vertebrate miRNAs to identify their developmental functions. By microscopic analysis, we could not observe clear defects associated with loss of 11 of these miRNAs during the first 4 d of embryonic development, although in situ hybridization revealed specific loss of most knocked-down miRNAs. Because all the targeted miRNAs are expressed in very specific tissues and we did not investigate most morphants in much detail by marker analysis, we may have missed subtle defects. In addition, many miRNAs reside in families of related sequence (e.g., let-7 and miR-182), and these should possibly be targeted simultaneously by different morpholinos to obtain a biological effect. Furthermore, in those instances in which miRNAs of unrelated sequence target a similar set of mRNAs when expressed in the same tissue [21], removing only one miRNA might not have a profound impact on transcript levels or expression. Finally, microarray analysis and computational predictions have shown that a single miRNA may regulate hundreds of mRNAs [30,31], but that some miRNAs act as a backup for mRNAs that are already repressed transcriptionally [32]. Thus, knockdown of such miRNAs might not dramatically affect gene expression, but ensure robustness of protein interaction networks as for example miR-7 in Drosophila [33].
In zebrafish, there are two copies of miR-375, and in human and mouse only one copy has been identified [34]. To verify the miR-375 knockdown phenotype, we designed control morpholinos targeting both precursors simultaneously (MO miR-375 star) and separately. Complete knockdown was only observed in those instances in which both miR-375 copies were targeted simultaneously. This also led to scattered islet cells, proving the specificity of the phenotype. However, knockdown with miR-375–1/2 loop morpholinos did not delay development as seen in the knockdown with the mature miR-375 morpholino. This shows the strength of using control morpholinos and excludes the delayed development as a relevant miR-375 loss-of-function phenotype. A moderate version of the phenotype was also observed in embryos injected with a morpholino specifically targeting miR-375–1. Thus, a reduction in the level of miR-375 already disturbs islet integrity. Similar to mouse miR-1 [15], miR-375 copies survived evolution and are expressed similarly in time and space, probably to ensure the high intracellular concentration of miR-375 necessary to repress many weakly binding targets.
In a forward genetic screen, several mutants were identified with improper development of the endocrine pancreas [35]. These mutants fall into three classes: (1) mutants with severely reduced insulin expression; (2) mutants with reduced insulin expression and abnormal islet morphology; and (3) mutants with normal levels of insulin expression and abnormal islet morphology. However, in all of these mutants, islet cells do not merge into an islet from their first appearance at approximately the 14-somite stage. Our miR-375 knockdown phenotype differs from this, because in the first instance, an islet is formed at approximately 24 hpf, but in later stages, the islet falls apart into small groups of cells. This rules out a general role for miR-375 in early endocrine formation as is seen for Wnt5 [36], but rather indicates a role in maintenance of tissue identity, which is assumed to be a general function of miRNAs in development [21]. It is as yet unclear which miR-375 targets are involved in the phenotype. Work in cell lines has implicated miR-375 in insulin secretion by targeting myotrophin [26]. The zebrafish homolog of myotrophin also contains a seven-nucleotide seed match to miR-375 (unpublished data), but future studies should reveal whether this target or many other predicted targets are relevant to the phenotype. The specific expression of miR-375 in the pancreatic islet and its implication in insulin secretion make it a candidate drug target in diabetes, e.g., to influence insulin levels in the blood. However, our data show that if miR-375 is used as a drug target, developmental side effects need to be taken into account.
Morpholinos were obtained from Gene Tools LLC (http://www.gene-tools.com) and dissolved to a concentration of 5 mM in water. Morpholinos were injected into one- or two-cell–stage embryos at concentrations between 200 μM and 1,000 μM, and per embryo, one nl of morpholino solution was injected.
RNA oligos (Table S2) were obtained from Sigma (http://www.sigmaaldrich.com) and dissolved to a concentration of 100 μM in distilled water. Oligos were annealed using a 5x buffer containing 30 mM HEPES-KOH (pH 7.4), 100 mM KCl, 2 mM MgCl2, and 50 mM NH4Ac. Typically, 1 nl of a 10 μM miRNA duplex solution was injected.
All morpholino sequences used in this study are listed in Table S1.
The miR-30c and miR-375 reporter constructs were made by cloning two annealed oligos containing two perfectly complementary miRNA target sites into pCS2 (Clontech, http://www.clontech.com) containing a gfp gene between BamHI and ClaI restriction sites. A construct containing pri-miR-205 was made by amplifying a genomic region (801 base pairs) containing the miR-205 precursor (miR-205-hairpinF ggcattgaattcataaCCTCTTACCTGCATGACCTG; miR-205-hairpinR ggcatttctagaGTGTGTGCGTGTATTCAACC). The resulting PCR fragment was cloned between XbaI and EcoRI restriction sites of PCS2GFP. Pri-miR-375–1 and pri-miR-375–2 constructs were made by amplifying genomic regions containing miR-375–1 and miR-375–2 precursors (WKmiR-375–1F-pCS2 gcccgggatccTGTGTCTTGCAGGAAAAGAG; WKmiR-375–1R-pCS2 attacgaattcTCAAACTCTCCACTGACTGC; and WKmiR-375–2F-pCS2 gcccgggatccGCCCTCCCATTTGACTC; WKmiR-375–2R-pCS2 attacgaattcAATGAGTGCACAAAATGTCC), and cloning of the resulting PCR fragments into the BamHI and EcoRI sites of pCS2. mRNA was synthesized using SP6 RNA polymerase. Luciferase mRNA was derived from pCS2 containing luciferase between BamHI and EcoRI sites.
In situ hybridization was performed as described previously [37]. LNA probes for miRNA detection were obtained from Exiqon (http://www.exiqon.com) and labeled using terminal transferase and DIG-11-ddUTP. cDNA clones for pri-miR-375–1, pri-miR-375–2, pit1, insulin, somatostatin, and glucagon were used for antisense DIG-labeled probe synthesis by T7 or Sp6 RNA polymerase.
For Northern blotting, total RNA was isolated from ten embryos per sample using Trizol reagent (Invitrogen, http://www.invitrogen.com). RNA was separated on a 15% denaturing polyacrylamide gel. Radiolabeled DNA probes complementary to miRNAs or 5S RNA (atcggacgagatcgggcgta) were used for hybridization at 37 °C. Stringency washes were done twice for 15 min at 37 °C using 2 × SSC 0.2% SDS. Alternatively, DIG-labeled LNA probes were used for hybridization at 60 °C and stringency washes were performed at 50 °C with 2x SSC 0.1% SDS for 30 min and 0.5x SSC 0.1% SDS for 30 min.
For RT-PCR, RNA was isolated with Trizol, treated with DNAse (Promega, http://www.promega.com) and subsequently purified again using Trizol. cDNA was made with a poly dT primer. Primers used for amplification were miR-205-hairpinF and miR-205-hairpinR, and lucF (ATGGAAGACGCCAAAAACATAAAG) and lucR (ATTACATCGATTTACACGGCGATCTTTCC).
For Alcian Blue staining, embryos were fixed for 1 h at room temperature in 4% PFA in PBS, rinsed for 5 min in 50% MeOH, and stored overnight in 70% MeOH at 4 °C. Next, embryos were incubated for 5 min in 50% MeOH and for 5 min in 100% EtOH. Embryos were stained at room temperature with Alcian Blue (Sigma) for 90 min with continuous shaking. Subsequently, embryos were rinsed in 80%, 50%, and 25% EtOH for 2 min each and two times in water containing 0.2% Triton and neutralized in 100% Borax solution. Finally, embryos were incubated for 60 min in digest solution (60% Borax solution, 1 mg/ml colleganase-free and elastase-free trypsin, 0.2% trypsin) and stored in 70% glycerol.
Staining of the hair cells was done by incubating live embryos for 5 min in a 200 μM solution of Daspei (Sigma) in + chorion. After rinsing twice in + chorion, embryos were anesthetized using MS222 and mounted in methylcellulose.
|
10.1371/journal.pbio.3000172 | Lateral hypothalamic neurotensin neurons promote arousal and hyperthermia | Sleep and wakefulness are greatly influenced by various physiological and psychological factors, but the neuronal elements responsible for organizing sleep-wake behavior in response to these factors are largely unknown. In this study, we report that a subset of neurons in the lateral hypothalamic area (LH) expressing the neuropeptide neurotensin (Nts) is critical for orchestrating sleep-wake responses to acute psychological and physiological challenges or stressors. We show that selective activation of NtsLH neurons with chemogenetic or optogenetic methods elicits rapid transitions from non-rapid eye movement (NREM) sleep to wakefulness and produces sustained arousal, higher locomotor activity (LMA), and hyperthermia, which are commonly observed after acute stress exposure. On the other hand, selective chemogenetic inhibition of NtsLH neurons attenuates the arousal, LMA, and body temperature (Tb) responses to a psychological stress (a novel environment) and augments the responses to a physiological stress (fasting).
| Adjusting sleep-wake behavior in response to environmental and physiological challenges may not only be of protective value, but can also be vital for the survival of the organism. For example, while it is crucial to increase wake to explore a novel environment to search for potential threats and food sources, it is also necessary to decrease wake and reduce energy expenditure during prolonged absence of food. In this study, we report that a subset of neurons in the lateral hypothalamic area (LH) expressing the neuropeptide neurotensin (Nts) is critical for orchestrating sleep-wake responses to such challenges. We show that brief activation of NtsLH neurons in mice evokes immediate arousals from sleep, while their sustained activation increases wake, locomotor activity, and body temperature for several hours. In contrast, when NtsLH neurons are inhibited, mice are neither able to sustain wake in a novel environment nor able to reduce wake during food deprivation. These data suggest that NtsLH neurons may be necessary for generating appropriate sleep-wake responses to a wide variety of environmental and physiological challenges.
| We first analyzed the distribution of Nts neurons in the LH by generating transgenic mice expressing green fluorescent protein (GFP) exclusively in Nts neurons (Nts-Cre::L10-GFP; henceforth “Nts-GFP” mice; see Methods section). We found that Nts neurons are densely packed in the perifornical LH, intermingled with orexin and MCH neurons. We also observed another dense population of Nts neurons in the subthalamic nucleus of the basal ganglia located dorsolateral to the LH and a less-dense population in the dorsomedial hypothalamus (DMH) lying medial to the LH. However, this study specifically focuses on the Nts neurons in the LH (NtsLH), and all brain injections were aimed at this population.
As Nts neurons were found densely packed in the perifornical LH region, we examined whether these neurons co-express MCH or orexin by immunolabelling brain sections from Nts-GFP mice (n = 6) for MCH and orexin. We found that none of the GFP+ neurons were labeled for MCH, whereas 3.7 ± 0.8% of GFP+ neurons were labeled for orexin (Fig 1A and 1B), indicating that a small fraction of Nts neurons may express orexin. To determine whether this overlap between Nts and orexin expression in LH neurons is an artifact of developmental expression of Nts/Cre (the Nts-GFP mouse might express GFP congenitally), we stereotaxically injected a Cre-dependent adeno-associated viral (AAV) vector containing mCherry (AAV8-hSyn-DIO-hM3Dq-mCherry, henceforth “AAV-hM3Dq”; University of North Carolina Vector core, United States; see below) into the LH (anteroposterior: −1.7 mm, ventral: 5.1 mm, lateral: ±1.1 mm) of adult (8 wk old) Nts-Cre mice (n = 4). Six weeks after the injections, we perfused the mice and immunolabeled the brain sections for mCherry (to label virally transfected, Cre-expressing Nts neurons) and orexin (Fig 1C). In these brain sections, we did not find any double-labeled neurons in the LH, indicating that NtsLH neurons are a distinct population from orexin neurons, but some of them may produce orexin during development.
We next examined the projections of NtsLH neurons using conditional anterograde tracing to understand the neuronal targets through which NtsLH neurons may regulate sleep-wake and Tb. We unilaterally microinjected a Cre-dependent AAV coding for channelrhodospsin-2 (ChR2) and the fluorescent tag mCherry (AAV8-EF1α-DIO-ChR2-mCherry, henceforth “AAV-ChR2”; University of North Carolina Vector core, US) [32, 33] into the LH of Nts-Cre mice (n = 6). Six weeks after the injections, we perfused the mice and processed the brain sections for immunohistochemical labeling of mCherry to identify ChR2-expressing Nts neurons and their axon terminals. AAV-ChR2 injections in three out of six mice were restricted to the LH (Fig 2A) without spread to adjacent regions, including the subthalamic nucleus or DMH, and only these cases were used to identify NtsLH projections. A high density of Nts terminals were found in the VTA, ventrolateral periaqueductal gray (vlPAG), parabrachial nucleus (PB), locus coeruleus (LC), retrorubral region, substantia innominata, diagonal band of Broca, ventral pallidum, nucleus accumbens, raphe pallidus (RPa), parapyramidal region (Ppy), and lateral preoptic area (Fig 2B–2G). We observed a moderate density of Nts terminals in the dorsolateral septum and supramamillary nucleus. While all NtsLH projections were predominantly ipsilateral, we also observed some less-dense contralateral projections in many of the target regions.
We used optogenetic tools to activate NtsLH neurons and investigated their role in sleep-wake control. We stereotaxically injected AAV-ChR2 bilaterally into the LH (anteroposterior: −1.7 mm, ventral: 5.1 mm, lateral: ±1.1 mm; Fig 3A) of Nts-Cre mice (n = 7) and implanted them with bilateral optical fibers (targeting 0.2 mm dorsal to the LH) for illumination with blue laser light [33], electrodes for recording electroencephalography (EEG) and electromyography (EMG) [34] and telemetry transmitters [35] for recording Tb and LMA.
Injections of AAV-ChR2 into the LH of Nts-Cre mice resulted in robust expression of ChR2-mCherry in NtsLH neurons. In contrast, injections of the same AAV into the LH of WT littermates did not result in any expression of mCherry, indicating the Cre dependency of the AAV-ChR2. The AAV injections in Nts-Cre mice were largely restricted to the LH and zona incerta, with little or no spread medially into the DMH (Fig 3B).
We first tested the response of NtsLH neurons to photo-illumination using ex vivo whole cell current clamp recordings. Illumination with blue laser light (473 nm at 1, 5, and 10 Hz) evoked action potentials in ChR2-mCherry–expressing NtsLH neurons in a frequency-dependent manner (Fig 3C). In vivo, 5-Hz stimulation for 2 h prior to perfusion caused robust cFos expression in mCherry-expressing NtsLH neurons, demonstrating that blue light illumination consistently drives activity in these neurons (Fig 3D).
To assess whether optogenetic activation of NtsLH neurons influences sleep-wake states, we applied blue laser light pulses (473 nm, 10 ms) with different frequencies for 10 s specifically during NREM or REM sleep. We applied laser stimulations at frequencies of 1, 5, and 10 Hz (10 mW light power at the tip of the optical fibers) after either 30 s of stable NREM sleep or 10 s of stable REM sleep during the light period (from 10:00 AM to 5:00 PM). We applied 10 photostimulations of each frequency for each state. Photostimulations during NREM sleep in AAV-ChR2–injected Nts-Cre mice resulted in rapid transition to wake (Fig 3E). On average, about 65% of 1-Hz stimulations and 93% of 5-Hz and 10-Hz stimulations during NREM sleep produced a rapid (within 1–5 s of light pulse onset) arousal response (Fig 3F), characterized by EEG desynchronization, EMG activation, and behavioral wakefulness in Nts-Cre mice (Fig 3E). In addition, the amount of wakefulness during the 30-s period immediately after cessation of photostimulations showed a frequency-dependent increase, indicating that higher stimulation frequencies produce more rapid and longer-lasting wake responses (Fig 3G). In contrast, photostimulation during REM sleep had no effect on sleep-wake or EEG/EMG activity in Nts-Cre mice (Fig 3H–3J). Mice remained undisturbed and REM sleep continued during the entire stimulation period (Fig 3H and 3I). These findings demonstrate that activation of NtsLH neurons rapidly trigger NREM-wake but not REM-wake transitions.
Although brief 10-s activations of NtsLH neurons are sufficient to evoke NREM-wake transitions, they are not sufficient to induce detectable changes in Tb or LMA. We therefore applied a continuous 5-Hz stimulation (10-ms pulse) for 30 min during the light period (at 10:00 AM) in Nts-Cre mice injected with AAV-ChR2. We observed a significant increase in Tb (1.35 ± 0.17°C increase) during the 30-min stimulation period (Fig 4A and 4B) when compared with 30 min prior. Tb began to drop almost immediately upon cessation of the stimulation and returned to baseline within the next approximately 30 min (Fig 4A). Because the photostimulations evoked immediate arousal in all mice (except in one mouse that was in REM sleep when the photostimulations began), and the mice were awake during the entire stimulation period (98.78% ± 1.22% versus 18.89% ± 6.44% during the 30 min prior; P < 0.001), we calculated the average Tb specifically during wake episodes (Tbwake). We found that the Tbwake during the photostimulations was significantly higher (1.11 ± 0.3°C increase; P = 0.016) than Tbwake during the 30 min prior to stimulations (Fig 4C), suggesting that the hyperthermia was not a mere coincidence of wake transitions. On the other hand, total LMA counts did not differ across these periods (Fig 4D and 4E), although the EMG activity (measured as integral EMG) during the photostimulations was 151% higher than prestimulation values (P = 0.004; one-way ANOVA; Fig 4F). The temporal correlation between the photostimulations and Tb on the one hand and the increase in Tb and EMG activity, even without substantial increase in LMA, on the other hand suggests that NtsLH neurons may participate in thermogenesis.
While optogenetic activation is suitable for studying acute state transitions (with millisecond-timescale precision), chemogenetic activation is better suited for studying long-term (minutes to hours) changes in sleep-wake behavior, LMA, and Tb. Therefore, we activated NtsLH neurons using chemogenetic tools and assessed changes in sleep-wake, LMA, and Tb. We stereotaxically injected either an AAV vector (AAV-hM3Dq) coding for the excitatory designer receptor exclusively activated by designer drugs (DREADD) [15, 37] or a control vector (AAV8-DIO-hsyn-mCherry; henceforth “AAV-mCherry”) into the LH of Nts-Cre mice (Fig 5A) and implanted them with telemetry transmitters for simultaneous recordings of EEG, EMG, Tb, and LMA [15].
Similar to the optogenetic experiments, the AAV injections induced specific expression of the viral vector product in NtsLH neurons, which was primarily restricted to the LH (Fig 5B). Bath application of 5 μM clozapine-n-oxide ([CNO] the ligand for hM3Dq receptors) depolarized hM3Dq-mCherry–expressing NtsLH neurons and increased their firing rate in ex vivo brain slices (Fig 5C). Moreover, intraperitoneal (i.p.) injections of CNO (0.3 mg/kg) 2.5 h prior to killing induced robust cFos expression in hM3Dq-mCherry–expressing NtsLH neurons (93.76% ± 1.52% of mCherry+ neurons expressed cFos after CNO in AAV-hM3Dq–injected Nts-Cre mice versus 16.78% ± 4.45% in AAV-mCherry–injected controls; Fig 5D). These data demonstrated that CNO effectively activated hM3Dq-expressing NtsLH neurons both ex vivo and in vivo.
To assess sleep-wake, LMA, and Tb changes following chemogenetic activation of NtsLH neurons, we i.p. injected saline (vehicle) or CNO (0.3 mg/kg) at 9:50 AM (light period) or 6:50 PM (dark period) and recorded EEG, EMG, LMA, and Tb for 12 h using the telemetry transmitters. Administration of CNO into Nts-Cre mice injected with the control vector AAV-mCherry (negative controls) did not produce any significant changes in sleep-wake, Tb, or LMA (data can be found from the Open Science Framework, https://osf.io/nmrpq/). Conversely, in the Nts-Cre mice expressing hM3Dq, CNO injections during the light period (9:50 AM) resulted in prolonged and uninterrupted wake (without any NREM or REM sleep) for 4–6 h (Fig 5E and S1 Table). The first NREM sleep and REM sleep were observed 275.00 ± 33.00 min and 312.66 ± 38.31 min, respectively, after CNO injections, significantly later than in the saline condition (40.74 ± 4.11 min and 68.29 ± 6.40 min, respectively; Fig 5F). Towards the end of the 12-h recording period, specifically hours 10–12 after CNO, we observed that REM sleep amounts, bout number, and mean duration were significantly higher than those after saline treatment, indicating a REM sleep rebound (Fig 5E and S1 Table). Although the NREM sleep increase during this same period did not reach statistical significance, a trend was observed, along with a significant increase in the number of NREM sleep bouts, indicating rebound increase in NREM sleep as well (Fig 5E and S1 Table). Finally, CNO injections at the dark onset (6:50 PM), when the circadian drive for wake is high, produced similar sleep-wake changes in Nts-Cre mice (S1 Fig and S2 Table).
The increase in wakefulness after CNO was accompanied by significant increases in Tb and LMA during both light (Tb: 1.56 ± 0.18°C and LMA: 584.81% ± 98.92% increase) and dark periods (Tb: 0.85 ± 0.09°C; LMA: 269.18% ± 22.45% increase) (Fig 5G and 5H and S1 Fig). Hyperthermia and higher LMA after CNO during the dark period were particularly interesting as their levels were much higher than the usual circadian increase in LMA and Tb at that time of day. Moreover, LMA per unit time of wake was significantly elevated after CNO (light period: 3.01 ± 0.50 versus 1.11 ± 0.13 counts per min after saline; P = 0.0084; dark period: 3.29 ± 0.61 versus 1.46 ± 0.19 counts per min after saline; P = 0.039), indicating a hyperactive phenotype. These data suggest that NtsLH neurons directly increase LMA and Tb in addition to regulating sleep-wake behavior.
Thus, chemoactivation of NtsLH neurons produced robust wake, hyperactivity, and hyperthermia, which are commonly observed after acute stress in rodents. While hyperthermia itself is considered a sensitive marker of stress [38], we also counted the cFos+ neurons in the paraventricular nucleus (PVH) whose activation in response to stressors leads to corticosterone secretion) [39, 40] as an additional stress marker. CNO injections almost doubled the cFos+ neurons in the hM3Dq-injected Nts-Cre mice when compared with negative controls (197.13 ± 32.57 versus 103.67 ± 18.01; P < 0.05, Mann–Whitney U test; S2 Fig), suggesting that activation of PVH occurs concurrently with the activation of NtsLH neurons.
Next, we chemogenetically inhibited NtsLH neurons to determine whether these neurons are necessary for spontaneous wake and regulation of Tb under baseline conditions. We stereotaxically injected an AAV encoding inhibitory DREADDs (AAV8-DIO-hsyn-hM4Di-mCherry; hereafter “AAV-hM4Di”) [34] into the LH of Nts-Cre mice (n = 7) (Fig 6A). Similar to experiment 5, we first confirmed the specific expression of hM4Di in NtsLH neurons by immunohistologically staining brain slices 4 wk after viral injections (Fig 6B). Next, we assessed the response of hM4Di-transfected NtsLH neurons to CNO in ex vivo brain slices and found that bath application of CNO (10 μM) caused complete inhibition of hM4Di-mCherry–expressing neurons (Fig 6C). Similarly, in vivo application of CNO (1.5 mg/kg, i.p.) in Nts-Cre mice 2.5 h before perfusions resulted in a complete absence of cFos in hM4Di-mCherry–expressing NtsLH neurons (2.86% ± 1.52% of mCherry+ neurons expressed cFos in AAV-hM4Di–injected Nts-Cre mice versus 16.78% ± 4.45% in AAV-mCherry–injected controls; Fig 6D).
Administration of CNO (1.5 mg/kg, i.p.) during the light or dark period did not cause any major alterations in sleep-wake states in Nts-Cre mice expressing hM4Di in NtsLH neurons (Fig 6E and S3 Fig). Hourly percentages, bout numbers, and bout durations of wake, NREM and REM sleep, as well as latencies to NREM and REM sleep after CNO were not significantly different from those after saline injections (Fig 6E and 6F, S3 Fig, S3 Table, and S4 Table). Similarly, total LMA and mean Tb after CNO were not significantly different from those after saline (Fig 6G and 6H and S3 Fig). These data indicate that NtsLH neurons may not be critical for sleep-wake regulation or maintenance of Tb under baseline conditions.
Although the inhibition of NtsLH neurons did not alter sleep-wake, LMA, and Tb during baseline conditions, activation of NtsLH neurons produced robust wake, hyperactivity, and hyperthermia as well as increased cFos expression in the PVH—all of which are usually observed in response to stress [31]. We therefore hypothesized that NtsLH neurons may specifically engage in stress-induced behavioral and physiological arousal. To test this hypothesis, we chemogenetically inhibited NtsLH neurons and assessed sleep-wake, LMA, and Tb in response to (a) the psychological stress induced by a novel environment and (b) the physiological/metabolic stress induced by fasting.
To study the role of NtsLH neurons in the stress response induced by a novel environment, we injected mice expressing hM4Di in NtsLH neurons with CNO (1.5 mg/kg, i.p.) or saline at 9:50 AM and immediately placed them in a new, clean cage with fresh bedding material (Fig 7A). Mice placed in a new cage after saline injections (saline+new cage) showed a significant increase in wake, LMA, and Tb for approximately 4 h compared with saline injections in the home cage (Fig 7B–7E and S5 Table). The first NREM sleep and REM sleep bouts in the new cage were observed after 218.74 ± 15.71 min and 264.03 ± 11.51 min, respectively (Fig 7C). In contrast, when the mice were injected with CNO before placing them in a new cage (CNO+new cage), sleep onset occurred faster with significantly decreased latencies for both NREM sleep (137.89 ± 13.65 min) and REM sleep (161.60 ± 11.05 min) (Fig 7C). Consistently, the percentage of time spent in NREM and REM sleep in the third hour (NREM: 41.57% ± 9.90% versus 0.00% ± 0.00% after saline; P < 0.0001; REM: 4.13% ± 1.83% versus 0.00% ± 0.00% after saline; P = 0.13) and the fourth hour (NREM: 56.74% ± 5.29% versus 21.14% ± 10.22% after saline, P = 0.0007; REM: 6.01% ± 1.01% versus 1.09% ± 0.77% after saline, P = 0.034) after CNO+new cage were higher than in the saline+new cage condition (Fig 7B). Similarly, the LMA and Tb during the period of 3–5 h after CNO+new cage were also substantially lower than after saline+new cage (Fig 7D and 7E). These findings demonstrate that NtsLH neurons are necessary for modulating arousal, LMA, and hyperthermia in response to psychological stress induced by a novel environment.
Based on the established role of the LH in feeding and energy homeostasis [41, 42], we then hypothesized that NtsLH neurons might contribute to the regulation of sleep-wake, LMA, and Tb in response to acute fasting (or mealtime hunger)—a form of physiologic/metabolic stress. Therefore, we tested whether NtsLH neurons contribute to these responses after metabolic stress by inhibiting hM4Di-expressing NtsLH neurons at dark onset by i.p. injections of CNO in the absence of food (Fig 8, S4 Fig, and S6 Table). Food was removed during the same time as i.p. injections and mice were not habituated to fasting or food restriction prior to these experiments. We recorded sleep-wake behavior, LMA and Tb during the 24-h fasting period following CNO.
When fasted after saline injections (saline+fasting), Nts-Cre mice (expressing hM4Di in NtsLH neurons) displayed an increase in wakefulness, with a corresponding decrease in NREM sleep for 3 h (between 2 and 4 h after injections; S4 Fig). This wake increase was accompanied by an increase in LMA (55% higher than in the saline+fed condition). Following this period (i.e., 4 h after fasting), both Tb and LMA began to drop (as expected in fasting mice), and we observed a significant increase in hypothermia (<34°C) and hypoactivity between 11 and 16 h after saline+fasting (S4 Fig). Similarly, NREM sleep during the same period was higher than in the saline+fed condition (S4 Fig).
Interestingly, chemoinhibition of NtsLH neurons exaggerated the sleep-wake and Tb responses to fasting. When fasted after CNO injections (CNO+fasting), Nts-Cre mice expressing hM4Di in NtsLH neurons displayed increased wakefulness for 6–7 h (Fig 8B). While the amount of wakefulness during the first 3 h after CNO+fasting was comparable to that in the saline+fasting condition, it was significantly higher during the seventh hour after CNO+fasting (94.47 ± 5.41% versus 59.10 ± 14.19% after saline; P = 0.044). Importantly, the latency to NREM sleep was significantly longer after CNO+fasting compared with saline+fasting (Fig 8C). LMA levels remained elevated for 6–7 h after CNO+fasting compared with an increase lasting only 3 h after saline+fasting (Fig 8D and S4 Fig). Similarly, Tb levels started falling 7 h after CNO+fasting (compared with 5 h after saline), but fell more rapidly and with a greater magnitude between 13 and 15 h after CNO+fasting (versus saline+fasting; Fig 8E); this later response may not be a direct consequence of the inhibition of NtsLH neurons, because the plasma half-life of CNO is short (<1 h) and its behavioral effects generally last about 4 to 8 h [43–45]. In contrast, this deeper fall in Tb between 13 and 15 h is presumably a consequence of enhanced wake and hyperactivity during the first 6–7 h after CNO. These results indicate that NtsLH neurons are critical for the fine-tuning of sleep-wake behavior, LMA, and Tb in times of caloric scarcity. Naturally, when food is unavailable, mice must reduce their activity and levels of wakefulness to conserve energy and guard against metabolic stress. Our findings suggest that NtsLH neurons are critical for this response.
Our results establish NtsLH neurons as a distinct neuronal population in the LH without overlap with either the orexin or MCH population. Using optogenetic and chemogenetic approaches, we demonstrate that (1) acute activation of NtsLH neurons results in rapid arousals from NREM sleep but not from REM sleep; (2) sustained activation of NtsLH neurons causes prolonged wake, hyperactivity, and hyperthermia; and (3) inhibition of NtsLH neurons attenuates the arousal and hyperthermia response to psychological stress, but augments sleep-wake, LMA, and Tb responses to metabolic stress.
Our data indicate that NtsLH neurons are distinct from orexin and MCH neurons in the LH, which is consistent with a previous report [46] but contradicts another study that reported a >80% overlap between of NtsLH and orexin neurons [26]. It is likely that the specificity of the antibodies and mRNA probes used by Furutani and colleagues prevented them from distinguishing between signals from NtsLH cell bodies and dense Nts terminals on orexin neurons [26]. While we observed a small fraction of Nts neurons expressing orexin in the Nts-GFP reporter mice, no such colocalization was observed when Nts neurons were visualized in adult mice. Thus, it appears that some NtsLH neurons may be capable of synthesizing orexin during development, but not in the adult stage. Consistently, single-cell gene expression analysis revealed no evidence of Nts mRNA in orexin neurons in adult mice [47]. These observations establish NtsLH as a distinct population of LH neurons in adult mice. While NtsLH neurons do not express MCH or orexin, subpopulations of NtsLH neurons may co-express other neuropeptides, including galanin and corticotrophin-releasing hormone, and classical neurotransmitters such as GABA and glutamate [20, 48–50].
We show that brief optogenetic activation of NtsLH neurons produces immediate transitions to wake from NREM sleep, while sustained chemogenetic activation causes arousals lasting several hours, suggesting that NtsLH neurons play a crucial role in both initiation and maintenance of wakefulness. While rapid arousals from NREM sleep were consistently evoked by photoactivation of NtsLH neurons, arousals from REM sleep were never evoked. Because even high-frequency photostimulations failed to arouse mice from REM sleep, it is not likely that differences in arousal threshold between NREM sleep and REM sleep are responsible. On the other hand, it is possible that NtsLH neurons become unresponsive to external stimuli during REM sleep, which is an inherent property of hypothalamic thermosensitive neurons [51, 52]. Considering the hyperthermia induced by NtsLH neuronal activation, these neurons may belong to a cold-sensitive neuronal population and may engage in cold defense behavior.
Another possibility is that the arousal from REM sleep involves pathways and mechanisms different from those involved in the arousal from NREM sleep and that NtsLH neurons may be a part of the latter. Interestingly, the previously identified wake-promoting cell groups in the forebrain, such as GABAergic neurons in the LH [11], bed nucleus of stria terminalis (BNST) [53], and basal forebrain [54–56], also did not elicit wakefulness from REM sleep. Considering the presence of sleep-wake alternation without REM sleep in the isolated forebrain [57, 58], it is probable that most wake-promoting neurons in the forebrain are wired for arousal from NREM sleep but not from REM sleep.
Besides the rapid but relatively short arousals induced by optogenetic stimulation of NtsLH neurons, we show that arousals induced by chemogenetic activation of NtsLH neurons were long-lasting and accompanied by a hyperactivity and hyperthermia, suggesting that the NtsLH neurons may regulate LMA and Tb in addition to sleep-wake states. Although wake is associated with increased activity, the LMA count per unit time of wake was significantly higher after NtsLH activation than during baseline wake, suggesting mice were hyperactive. Wake and LMA do not always go hand in hand. For example, increased wake after systemic administration of certain pharmacological agents (e.g., modafinil) or after activation of the wake-promoting cell groups in the brain (e.g., PB) were not accompanied by increased LMA [59, 60]. On the contrary, increased LMA after loss of MCH neurons in the LH was not accompanied by an increase in wake [15]. Thus, the increased LMA after NtsLH activation is not necessarily due to higher wake amounts.
Similarly, subchronic (30-min) photoactivation of NtsLH caused an increase in Tb that was neither related to state changes nor accompanied by an increase in LMA, even though chemoactivation increased Tb with a concurrent increase in LMA. While chemoactivation makes neurons more responsive to natural inputs, photoactivation causes rhythmic monotonous firing of neurons. Such differential activation of NtsLH neurons could have led to different downstream responses, contributing to the differential locomotor responses after optogenetic- versus chemoactivation. Nevertheless, these data clearly demonstrate that NtsLH activation may increase Tb independent of wake or hyperactivity and thereby suggest a direct role for NtsLH neurons in thermogenesis. Lack of arousal response after activation of NtsLH neurons during REM sleep indicates the potential cold-sensitive nature of these neurons [51, 52], further supporting this idea. Future studies are, however, necessary to test the cold sensitivity of NtsLH neurons and the thermoregulatory deficits induced by their loss.
In contrast to chemogenetic activation, chemogenetic inhibition of NtsLH neurons had no effect on wake, LMA, or Tb, suggesting that these neurons may not be necessary for the regulation of spontaneous wakefulness or Tb control under baseline conditions. Because the increased wake, hyperthermia, and hyperactivity after chemogenetic activation of NtsLH neurons resembled a stress response, with concurrent increase in cFos in PVH neurons, we hypothesized that these neurons could be important for stress-induced arousals. We found that chemoinhibition of NtsLH neurons indeed attenuates wakefulness, LMA, and Tb responses to stress induced by a novel environment. Interestingly, NtsLH inhibition paradoxically amplified the sleep-wake, LMA, and Tb responses to metabolic stress induced by fasting. Clearly, the responses to stress should differ depending on the type, magnitude, and duration of the stress [61]—while it may be necessary to increase wake to explore a novel environment and search for potential threats and food sources, it is also necessary to decrease wakefulness and reduce energy expenditure during metabolic challenges, such as during prolonged absence of food. Thus, the observed changes in response to novelty stress and metabolic stress are not actually paradoxical but instead indicate that NtsLH neurons might integrate stress stimuli and generate the appropriate responses. It is also likely that NtsLH neurons are heterogenous (in terms of co-expression of other neurotransmitter or molecular signatures) [20], and these different subsets of neurons may orchestrate responses to different stressors through their differential output pathways. Further studies are required to identify these subsets of NtsLH neurons selectively responding to different stressors.
Similar to chemoinhibition of NtsLH neurons, attenuated arousal response to a novelty stress was observed in orexin neuron-ablated mice, and exaggerated arousal response to fasting was observed in MCH-knockout mice [62, 63]. In addition, orexin neuron-ablated mice also did not exhibit increased arousal levels during fasting [63]. Both orexin and MCH neurons express receptors for Nts [22, 64], and both receive inputs from NtsLH neurons. Importantly, NtsLH neurons are the only cell population in the LH that expresses MCH receptors [65]. Nts has been shown to activate orexin neurons in vivo and ex vivo [26]. In contrast, activation of Nts terminals on orexin neurons may inhibit orexin neurons by releasing galanin [66]. Thus, it is possible that Nts neurons may excite or inhibit orexin neurons by releasing either Nts or galanin, respectively. Notably, Nts antagonists have no effect on sleep-wake states in orexin-ablated mice [26]. Thus, we propose that NtsLH neurons may act as a “master orchestrator” and modulate the activity of orexin and MCH neurons, depending upon the perceived stressors, and generate appropriate stress responses.
Because we did not study the specific inputs to NtsLH neurons, it is unclear how stress signals reach NtsLH neurons. However, previous studies have shown that several major mediators of stress responses such as the medial prefrontal cortex, PVH, BNST, and amygdala heavily project to the LH [67–69], and NtsLH neurons may receive these inputs. On the other hand, metabolic signals may directly target NtsLH because a subset of these neurons expresses leptin receptors and NtsLH neurons have been shown to be activated by leptin in brain slices [22, 70].
In addition to local LH circuits linking Nts neurons with orexin and MCH neurons, we observed direct projections from Nts neurons to other brain structures regulating sleep-wake, Tb, and LMA. Based on our tracing data, we predict that NtsLH could activate the VTA, LC, and PB, which are known as potent wake-promoting cell groups [27, 71–74]. Likewise, NtsLH neurons could inhibit the lateral preoptic area, which is sleep promoting [75, 76]. Moreover, activity in NtsLH neurons could increase Tb by activating RPa/Ppy and ventromedial medulla neurons and promote nonshivering and shivering thermogenesis, respectively [77–80]. Finally, NtsLH projections to the VTA may mediate the LMA responses, as increased LMA after NtsLH neurons were blocked by dopamine antagonists, and intra-VTA administration of Nts-antagonist blocked the dopamine release from VTA neurons [81]. While Nts can be excitatory or inhibitory depending upon the receptor expression in the postsynaptic neurons, a subset of NtsLH neurons also express GABA [81]. Thus, the hyperthermia, hyperactivity, and wakefulness after NtsLH activation could be due to a complex integration of inhibitory and excitatory signals. Future studies are required to identify the specific neurotransmitter and pathways involved in each of these responses.
Collectively, our results indicate that NtsLH neurons are capable of initiating and sustaining wakefulness and increasing Tb and LMA. While NtsLH neurons may not be necessary for spontaneous wakefulness or Tb maintenance under baseline conditions, they are necessary for modulating wake and hyperthermia after psychological or metabolic stressors. We show that NtsLH neurons reduce wake, LMA, and Tb in response to fasting, while they increase wake, LMA, and Tb in response to a novel environment. Moreover, NtsLH neurons may also be cold sensitive and potentially contribute to cold defense mechanisms, as they were unresponsive during REM sleep, and their activation induced strong hyperthermia. Considering the involvement of the LH in various physiological functions, including sleep-wake, feeding, and thermoregulation and the close-interrelationship between these functions, [1–3, 20, 82–85], our results suggest that NtsLH neurons may play a crucial role in modulating sleep-wake states, LMA, and Tb in response to a variety of physiologic and metabolic demands.
All experiments were conducted in accordance with the National Institutes of Health guidelines for the Care and Use of Laboratory Animals and were approved by the institutional animal care and use committee of Beth Israel Deaconess Medical Center (protocol #039–2016). All efforts were made to minimize the number of animals used and their suffering.
Prior to surgery, all mice were group-housed in a temperature (22 ± 1°C)–and humidity (40%–60%)–controlled animal room maintained on a 12:12-h light-dark cycle. All mice had ad libitum access to standard chow diet and water. After surgery, all animals were singly housed for 3–4 wk before the physiological data collection began. Male mice aged 8–12 wk and weighing between 20 and 24 g at the time of surgery were used for behavioral experiments, and 4-wk-old mice were used for ex vivo brain slice recordings. For this study, we used two transgenic mouse lines—mice expressing Cre recombinase under the Nts promoter (Ntstm1(cre) Mgmi/J mice; Jackson Laboratory, Stock No. 017525; “Nts-Cre mice” [22]) and a GFP-reporter mouse line (Rosa26-loxSTOPlox-L10-GFP; generated by Dr. Brad Lowell, BIDMC; “L10-GFP mice” [86]). Nts-Cre mice were crossed with L10-GFP mice to validate Cre expression in Nts neurons. For all experiments, we used heterozygous Nts-Cre mice on a mixed background.
Genomic DNA from mice was extracted from tail biopsies and analyzed via polymerase chain reaction using a REDE Extract-N-Amp Tissue PCR Kit (Sigma-Aldrich, US) (Nts-Cre: common forward, 5′-ATA GGC TGC TGA ACC AGG AA; WT reverse, 5′-CAA TCA CAA TCA CAG GTC AAG AA; Cre reverse, 5′-CCA AAA GAC GGC AAT ATG GT. Rosa26-loxSTOPlox-L10-GFP: WT forward, 5′-GAG GGG AGT GTT GCA ATA CC; mutant forward, 5′-TCT ACA AAT GTG GTA GAT CCA GGC; and common reverse, 5′-CAG ATG ACT ACC TAT CCT CCC).
Adult male Nts-Cre mice were anesthetized with a ketamine/xylazine mixture (100 mg/kg ketamine and 10 mg/kg xylazine) and were unilaterally microinjected with 60 nL of AAV-ChR2 (University of North Carolina Vector core) or AAV-hM3Dq (University of North Carolina Vector core) into the LH (anteroposterior, −1.7 mm from bregma; lateral ±1.1 mm; dorsoventral, −5.1 mm from dura [36]). Six weeks after the injections, all mice were killed for histological processing.
Adult male Nts-Cre mice were anesthetized with a ketamine/xylazine mixture (100 mg/kg ketamine and 10 mg/kg xylazine) and were microinjected bilaterally with 60 nL AAV-ChR2 or AAV8-EF1α-DIO-mCherry (AAV-mCherry; University of North Carolina Vector core) into the LH [15]. All mice were then implanted with (a) optical fibers targeting 0.2 mm dorsal to the LH for blue light/laser stimulation, (b) electrodes for recording EEG and EMG-EEG signals by using ipsilateral stainless steel screws and EMG signals by a pair of stainless steel wires inserted into the neck extensor muscles, and (c) i.p. radio transmitter (TA10TA-F20, Data Science International, MN) for measuring Tb and LMA [15, 34].
Four weeks after surgery and AAV microinjections, mice were connected to the recording cables and habituated for 3 d, after which we performed baseline sleep-wake, LMA, and Tb recordings. The EEG/EMG signals were amplified (AM systems, WA, US), digitized, and recorded using Vital recorder (Kissei Comtec, Nagano, Japan) [75]. Tb and LMA data were recorded using Dataquest ART 4.1 (Data Sciences International, MN). We examined the effects of laser stimulation (10 s of stimulation at 1, 5, and 10 Hz) after 30 s of stable NREM sleep (10 trials each stimulation frequency) or 10 s of stable REM sleep (10 trials each stimulation frequency) during the light period. For assessing changes in LMA and Tb, we applied 5-Hz laser stimulation for 30 min during the light period.
Nts-Cre mice were anesthetized with ketamine/xylazine (100 mg/kg and 10 mg/kg, i.p.) and injected with AAV8-hsyn-DIO-hM3Dq-mCherry, AV8-hsyn-DIO-hM4Di-mCherry, or AAV-mCherry bilaterally into the LH and were implanted with the telemetry transmitters (TL11M2-F20-EET; Data Science International, St. Paul, MN) that allow simultaneous recording of EEG, EMG, LMA, and Tb [15, 87]. Four weeks after surgery, mice were habituated to the recording room conditions for 3 d. For the interventions, mice were injected with the ligand for hm3Dq and hM4Di, CNO (0.3 or 1.5 mg/kg; Sigma, St. Louis, MO), or the vehicle (saline) at 9:50 AM (10 min before ZT3) or 6:50 PM (10 min before dark onset), and postinjection recordings of sleep-wake, LMA, and Tb (Dataquest ART 4.1, Data Sciences International, US) were performed for 24 h. The order of injections was counterbalanced and there were approximately 7 d between two CNO injections in the same mouse.
The EEG, EMG data were divided into 12-s epochs and scored manually into one of the three sleep-wake states, wake, NREM sleep, or REM sleep, using SleepSign 3 (Kissei Comtec, Nagano, Japan) [15, 34, 75]. Percentages of time spent in each sleep-wake state and their mean number and bout durations in 1- and 3-h bins were calculated, along with the total LMA and mean Tb for these periods. Latency to NREM sleep and REM sleep were calculated as time taken to that stage from the time of i.p. injections.
After the completion of physiological data collection, all mice were deeply anesthetized with 7% of chloral hydrate and were transcardially perfused with PBS (15 mL) followed by 10% formalin (50 mL). Mouse brains were harvested immediately and incubated with 10% formalin overnight, followed by incubation in 20% sucrose in formalin solution at 4°C. Brains were cut into three series of 40-μm coronal sections on a freezing microtome and processed for immunohistochemistry, immunofluorescence, and/or in situ hybridization.
For immunohistochemistry using diaminobenzidine (DAB) reactions, sections were incubated with the primary antibody for two nights (for cFos labeling) or overnight (all others), followed by incubation in the appropriate biotin-SP-conjugated secondary antibody (1:1,000; Jackson ImmunoResearch, West Grove, PA) for 2 h. Then, sections were incubated for 75 min in avidin-biotin-complex reagent (1:1,000; Vectastain ABC kit, Vector Lab, Burlingame), washed, and incubated in a 0.06% solution of DAB (Sigma-Aldrich) and 0.02% H2O2 for 2–5 min for staining in brown. CoCl2 (0.05%) and 0.01% NiSO4 (NH4) in PBS was added to the DAB solution for staining in black [15, 34, 87, 88].
For immunofluorescence, sections were incubated in primary antibodies overnight. After washes in PBS, the sections were incubated in the appropriate fluorescent secondary antibodies (1:1,000; Alexa Fluor Dyes, Life Technologies, Carlsbad, CA) for 2 h.
The following antibodies were used: primary antibody for cFos (1:20,000; PC38; MilliporeSigma, Darmstadt, Germany), ds-Red (1:10,000; 632496; Clontech Laboratory, Mountain View, CA), Orexin-A (1: 5,000; SC-8070; Santa Cluz Biotechnology, Dallas, TX), and MCH (1: 5,000; gift from Dr. Eleftheria Maratos-Filer, Harvard University, Boston, MA) [15, 34, 88, 89].
Sections were mounted on Superfrost glass slides, dehydrated, cleared, and coverslipped using Permaslip (Albose Scientific, MO) in case of DAB staining or Vectashield (Vector labs, CA) in case of fluorescent labeling.
All cell counting was performed by constructing a 500 × 500 μm box on the lateral hypothalamus. The dorsal border of the square was aligned with the dorsal edge of the third ventricle, while the lateral border was aligned with the lateral edge of the hypothalamus [89]. The Franklin and Paxinos mouse brain atlas [36] was used for determining anteroposterior coordinates. Abercrombie corrections were applied to all cell counts [90].
Under anesthesia, Nts-Cre mice (4 wk old) were injected with AAV-ChR2 (n = 3), AAV-hM3Dq (n = 4), or AAV-hM4Di (n = 3) into the LH and killed after 4 wk. Brains were removed and quickly transferred to ice-cold cutting solution consisting of 72 mM sucrose, 83 mM NaCl, 2.5 mM KCl, 1 mM NaH2PO4, 26 mM NaHCO3, 22 mM glucose, 5 mM MgCl2, and 1 mM CaCl2, carbogenated with 95% O2/5% CO2, with a measured osmolarity of 310–320 mOsm/L. Brains were cut into 250-μm slices and the slices containing the LH were used for current-clamp recordings. mCherry (ChR2/hM3Dq/hM4Di)–expressing Nts neurons in the slices were visualized using an upright microscope (SliceScope, Scientifica), and current-clamp recordings were performed using borosilicate glass microelectrodes (5–7 MΩ) filled with internal solution [33, 86, 91]. After achieving stable baseline recordings for 5–10 mins, the response to photo-illumination or CNO (depending upon the AAV injected) was investigated. To test the response of ChR2-expressing NtsLH neurons to photo-illumination, 477 nm light was applied for 5–10 s at various frequencies (1–10 Hz), and the recordings were continued for 1–2 min. Data from 10 s before, during, and after the stimulus were compared. To test the CNO effects of hM3Dq- or hM4Di-expressing NtsLH neurons, artificial cerebrospinal fluid (ACSF) solution containing CNO (500 nM) was perfused onto the slice preparation and recordings continued for 2–5 min, followed by ACSF perfusions to wash out the CNO. Data from 2 min just prior to bath application of CNO were considered as baseline; the response to CNO was measured during the last 1 min of CNO application. The resting membrane potentials before and during CNO were compared using paired t tests. Current (5–20 pA) was applied via the patch pipette if mCherry+neurons did not fire action potentials (for hM4Di inhibition experiments).
Statistical analysis was performed using GraphPad Prism version 7 (GraphPad Software, La Jolla, CA). For optogenetic experiments, data after photostimulations (in Nts-Cre mice injected with AAV-ChR2) were compared with prestimulation data as well as after sham stimulations using a one-way ANOVA followed by Tukey multiple comparisons. In chemogenetic experiments, post-CNO data from the experimental group (Nts-Cre mice injected with AAV-hM3Dq/AAV-hM4Di) were compared with post-saline data from the same mice and post-CNO data from negative controls (Nts-Cre mice injected with AAV-mCherry) using a two-way repeated measures ANOVA, followed by Sidak post hoc test. All data are presented as the mean ± SEM unless otherwise noted. Differences were considered significant at P values less than 0.05.
|
10.1371/journal.pntd.0005459 | Cost-effectiveness of meglumine antimoniate versus miltefosine caregiver DOT for the treatment of pediatric cutaneous leishmaniasis | Oral miltefosine has been shown to be non-inferior to first-line, injectable meglumine antimoniate (MA) for the treatment of cutaneous leishmaniasis (CL) in children. Miltefosine may be administered via in-home caregiver Directly Observed Therapy (cDOT), while patients must travel to clinics to receive MA. We performed a cost-effectiveness analysis comparing miltefosine by cDOT versus MA for pediatric CL in southwest Colombia.
We developed a Monte Carlo model comparing the cost-per-cure of miltefosine by cDOT compared to MA from patient, government payer, and societal perspectives (societal = sum of patient and government payer perspective costs). Drug effectiveness and adverse events were estimated from clinical trials. Healthcare utilization and costs of travel were obtained from surveys of providers and published sources. The primary outcome was cost-per-cure reported in 2015 USD. Treatment efficacy, costs, and adherence were varied in sensitivity analysis to assess robustness of results. Treatment with miltefosine resulted in substantially lower cost-per-cure from a societal and patient perspective, and slightly higher cost-per-cure from a government payer perspective compared to MA. Mean societal cost-per-cure were $531 (SD±$239) for MA and $188 (SD±$100) for miltefosine, a mean cost-per-cure difference of +$343. Mean cost-per-cure from a patient perspective were $442 (SD ±$233) for MA and $30 (SD±$16) for miltefosine, a mean difference of +$412. Mean cost-per-cure from a government perspective were $89 (SD±$55) for MA and $158 (SD±$98) for miltefosine, with a mean difference of -$69. Results were robust across a variety of assumptions in univariate and multi-way analysis.
Treatment of pediatric cutaneous leishmaniasis with miltefosine via cDOT is cost saving from patient and societal perspectives, and moderately more costly from the government payer perspective compared to treatment with MA. Results were robust over a range of sensitivity analyses. Lower drug price for miltefosine could result in cost saving from a government perspective.
| Cutaneous leishmaniasis (CL) is a tropical parasitic disease transmitted by sand flies that causes chronic skin and mucosal ulcers. Current standard of care therapy requires patients to travel to a clinic for twenty consecutive days for injections of meglumine antimoniate (MA). This may represent an economic burden, particularly for patients living far from healthcare services, especially children and their caregivers. We performed mathematical modeling to compare costs of the standard of care treatment with costs of miltefosine, an equivalently efficacious oral medication that allows pediatric patients to be treated at home under trained supervision of a caregiver. In our model, miltefosine led to substantially lower costs for patients and only slightly higher costs to the healthcare system. Importantly, the cost to society (combined patient and healthcare system costs) was lower for miltefosine compared to MA. Treatment of pediatric CL with miltefosine in the patient’s home could decrease overall cost of treatment, while diminishing the barriers and cost burden on patients, their caregivers, and society.
| Cutaneous leishmaniasis (CL) is a neglected tropical disease primarily affecting poor, marginalized populations. Worldwide incidence of CL has been estimated at 1–5 million cases per year [1,2]. In Latin America, over 57,000 annual cases were reported on average between 2001 and 2013 [3]. In Colombia, 7,000–18,000 cases of CL are reported per year [4]. Of the 7,777 cases reported in 2015, 71.5% were from rural areas, and 17% were younger than 15 years of age [5]. A recent retrospective study of the clinical and epidemiological profile of pediatric CL patients in Colombia indicates that children are increasingly affected by the disease due to population movements and environmental factors bringing vectors into closer contact with domestic settings [6].
CL patients in rural areas face economic and geographic challenges to securing treatment, including travel through zones of armed conflict. Pediatric populations are expected to incur particularly high costs, as their caregivers must accompany them to clinic for treatment. Additionally, MA has been shown to have a higher rate of renal clearance in pediatric patients, contributing to lower systemic exposure and a higher failure rate than adults [7,8]. Finally, MA has been associated with infrequent but serious adverse reactions, as well as intolerance of the intramuscular route of administration [9].
Miltefosine, a well tolerated medication for CL, has been shown to be non-inferior to MA in a clinical trial in pediatric patients in Colombia [10]. Miltefosine is administered orally and could be given via caregiver directly observed therapy (cDOT) at the patient’s home.
Directly observed therapy (DOT) describes any protocol in which a trained observer watches medication administration to ensure compliance in order to avoid treatment failure and microbe resistance. Traditionally, this observer has been a health care professional, but protocols in which the observation is done in the home and by lay observers have been shown to be non-inferior in the case of tuberculosis [11], the disease currently most commonly treated by DOT protocols. In our study, cDOT implies education of a pediatric patient’s caregiver in a manner that ensures course completion and appropriate use, including safe medication usage, storage and disposal. While cDOT has not yet been implemented for CL treatment, the efficacy [12–17] and cost-effectiveness [12,16] of the cDOT model for the treatment of tuberculosis have been established in a variety of contexts, including in pediatric populations [15,17].
Caregiver administration could ease the economic burden of CL treatment on families of affected children, as well as improve access and adherence to treatment in remote areas. Despite evidence demonstrating efficacy, the pediatric formulation of miltefosine is not widely available in Colombia.
This study describes the relative costs of pediatric CL treatments with MA and miltefosine treatment for patients in southwest Colombia. This information is intended to guide policy makers, health ministries, and healthcare providers in countries with endemic CL.
We developed a cost-effectiveness analysis study using a Monte Carlo simulation model of CL treatment to examine the potential clinical and cost impact of miltefosine cDOT versus MA for a pediatric population with CL. The model incorporated data from multiple sources including public health databases [18–20], primary surveys, expert opinion, and published data in order to project cost to stakeholders over the course of CL treatment.
Simulated strategies were based on the interventional arms of the RCTs undertaken in Colombia and Brazil between 2007–2010 comparing the efficacy of intramuscular MA and oral miltefosine. Independent parameters were subjected to random assignment along assigned probability distributions in order to represent uncertainty and heterogeneity in these parameters.
The study was conducted with public health data from four municipalities in two recognized endemic areas in Colombia [21] which have active leishmaniasis treatment programs—the lowland Pacific coastal municipalities of Buenaventura and Tumaco, and the Andean Central Cordillera municipalities of Chaparral and Rovira. These sites are among the municipalities with greatest transmission of CL in Colombia [5].
The University of Chicago Institutional Review Board and the Ethics Committee of CIDEIM and Universidad Icesi approved and monitored the study.
Cost-effectiveness analyses were performed from patient, government payer, and societal perspectives [22]. The patient perspective included out-of-pocket costs assumed by patient caregivers during the course of treatment, such as transportation to clinics, meals outside the home, lodging, childcare, and medical supplies. The patient perspective excluded drug costs as these are publically covered. The government payer perspective included costs assumed by the Colombian healthcare system in the course of treatment such as drug costs, clinical medical supplies, and treatment associated with adverse events. The societal perspective combined the patient and government payer perspectives to estimate total cost associated with treatment.
Costs are reported in 2015 USD [23], and no discounting was applied, as the time frame of treatment was under one month. Sensitivity analyses were performed to test the stability of model outputs with variation of parameters.
Primary outcomes were societal, patient, and government payer cost-per-cure for each treatment strategy. Cost difference is the MA cost-per-cure minus the miltefosine cDOT cost-per-cure. Cost neutrality is the point at which MA and miltefosine cDOT treatments incur the same cost-per-cure.
We employed a Monte Carlo simulation model (SimVoi v3.02 plugin for Microsoft Excel). Cases of clinically confirmed CL were simulated in a patient level probabilistic model in which unique patients entered the model and accrued costs to themselves and the government payer based on their treatment assignment, and left the model in either a cured or uncured state. Cure rate and adverse events during the course of treatment were modeled. Individuals failing treatment were not re-treated. Simulations of 100,000 patients were run for each potential intervention—meglumine antimoniate, miltefosine (availability of adult and pediatric formulations), and miltefosine (pediatric formulation only)—to ensure stability of results.
Baseline characteristics of children ages 2–12 with diagnosed CL were obtained from the pooled National Public Health Surveillance System (SIVIGILA) public health records from the municipality of Tumaco, Nariño from January 2012-May 10, 2014 [18] and Chaparral, Tolima from March 7, 2003-December 7, 2011 [19]. The average age was 7.12 years and 49.7% of patients were female (Table 1). Patient weights were based on means and confidence intervals presented in the National Survey of the Nutritional Situation in Colombia in 2010 [24]. Weights were plotted on a normal distribution for each year of age for each sex.
One- and multi-way deterministic analyses of selected parameters were performed to assess impact on base case results. The impact of treatment adherence was estimated by variation in treatment efficacy. Drug costs were varied, as was lost-time cost up to 100% of the Colombian minimum daily wage [35] during the course of MA treatment. Model inputs obtained from the survey were varied, including costs for supplies, treatment cost, transportation, meals, lodging, and childcare.
A multi-way sensitivity analysis explored variations in the efficacy ratio of miltefosine cDOT over MA. The base case was considered to be equivalent efficacy (efficacy ratio = 1.00), and lower and upper bounds were derived from the upper 95% CI of miltefosine efficacy divided by the lower 95% CI of MA efficacy (efficacy ratio = 1.19) and a reciprocal lower bound was calculated by subtracting the reciprocal change from the baseline assumption (efficacy ratio = 0.81).
Cost-per-cure ratios (miltefosine cDOT/MA), in which 1 indicates equivalent cost for miltefosine cDOT and MA, <1 indicates cost savings with miltefosine cDOT, and >1 indicates MA cost saving, were calculated.
From the societal and patient perspectives, miltefosine cDOT was less costly compared to MA. Mean societal cost-per-cure for MA and miltefosine cDOT were $531 (SD±$239) and $188 (SD±$100) respectively, a difference of +$343. Mean patient costs per cure were $442 (SD ±$233) for MA and $30 (SD±$16) for miltefosine cDOT, a difference +$412. Miltefosine cDOT cost savings were driven by high costs associated with patient travel required for MA treatment, including transportation, lodging, childcare, and meals ($259 (SD±$144) for MA and $21 (SD±$12) for miltefosine cDOT).
From the government payer perspective, MA was less costly than miltefosine cDOT due to the higher drug cost of miltefosine (mean drug costs were $41 (SD±$16) for MA and $116 (SD±72) for miltefosine cDOT). Mean government payer costs per cure were $89 (SD±$55) for MA and $158 (SD±$98) for miltefosine cDOT, a difference of -$69 (Table 4).
When availability of only the 10mg formulation of miltefosine was simulated, mean societal costs of $217 (SD±$93), mean patient costs were $30 (SD±$ 16), and mean government payer system costs were $187 (SD±$91) for miltefosine cDOT, with respective cost differences of +$344, +$412, and -$98.
In one-way sensitivity analysis of baseline assumptions (Fig 1), miltefosine cDOT remained cost saving compared to MA across a wide variation of parameters, including drug adherence varies between 50–100%. Cost superiority of miltefosine cDOT was also maintained as miltefosine and MA drug prices were varied from 50–200% of WHO negotiated prices. Increased cost-effectiveness was seen with the inclusion of up to 100% of daily minimum wage loss for the caregiver during the course of MA treatment.
One-way sensitivity analyses of cost parameters collected by survey instrument -
medical supplies cost, treatment cost, food, lodging, childcare, and transportation—showed no change in cost-per-cure superiority when varied between 50–200% of mean collected data (Fig 2).
In multi-way sensitivity analysis, cost-per-cure ratio remained below 1 over a wide range of miltefosine cDOT-associated government payer costs and MA-associated patient costs (Fig 3). Miltefosine cDOT remained a cost-saving option from a societal perspective when MA-related patient costs were above 18% of the base case, and miltefosine cDOT-associated government payer costs was less than 355% of the base case. Stability in cost-per-cure ratio over these ranges was also demonstrated with the availability of 50 and 10mg formulations, as well as 10mg formulation only.
Our analytic model of treatment of CL in pediatric patients with miltefosine by cDOT versus current first-line MA treatment indicates that the miltefosine regimen is cost saving from a societal perspective. This result reflects considerably lower travel-associated costs for patients treated with miltefosine cDOT versus MA, a savings that exceeded the increased drug cost of miltefosine versus MA to the government payer. These results were robust across wide variations in parameters including adherence, medication efficacy, MA patient costs, miltefosine government payer costs, lost-time cost, adverse events costs, and direct patients costs. The availability of 50mg and 10mg formulations was associated with lower costs than availability of 10mg alone, but did not affect conclusions regarding cost superiority.
It should be noted that previous studies have estimated higher government payer cost-per-cure for MA, which may indicate further cost advantage of the miltefosine cDOT protocol. A study of an outbreak in Colombia estimated costs at $345 per cure; with MA drug costs of 300% of the cost assumed in our analysis [36]. Government payer cost-per-cure for MA in Guatemala and Peru has been estimated at $280 and $300, respectively [37,38]. The MA cost from these studies would exceed that estimated for miltefosine cDOT, making miltefosine cDOT cost-per-cure superior from the government payer perspective. An analysis of government payer cost-effectiveness of miltefosine and MA for adult CL patients in Colombia showed that miltefosine costs were nearly equivalent to MA costs [39]. However, no other analysis has focused on pediatric populations nor included patient and societal viewpoints.
Conversely, it should also be highlighted that our use of the WHO pricing guidelines may represent a low cost for miltefosine in Latin America. Despite these guidelines, procurement costs in practice have been observed to be considerably higher [39,40]. We emphasize that acquisition of competitive drug prices by government actors is a priority in providing miltefosine cDOT therapy in a cost-sensitive budgetary context. Acquisition of drugs for NTDs bought in the absence of a national public health program are likely to be higher than drug prices achievable though centralized high volume ordering [41–43], and as such, coordinated purchasing represents an opportunity for improvement of cost to the government payer. Additionally, pricing guidelines are subject to eventual renegotiation, in which case it is imperative that national, international, and non-governmental actors push for advantageous pricing of drug that carry significant benefits for marginalized patients.
Burden of disease studies have demonstrated that leishmaniasis and other NTDs cause significant detriment to the lives and livelihoods of patients and caregivers in endemic areas [44–48]. Our study highlights that decisions on public health matters by government payers should consider more than direct expense, and incorporate value added and costs avoided by different options, as well as the ethical mandate of protection of vulnerable populations. As in many low-resource settings, direct cost saving at the level of drug purchasing is attractive from a budgetary standpoint. However, a systemic perspective of costs of disease and treatment may reveal reversals of treatment cost-effectiveness superiority when patient and societal points of view are considered.
The findings of this study should be interpreted in the context of certain limitations. Firstly, effectiveness data for the treatment of CL is unavailable in Colombia and scarce among all countries of the region [49]. Secondly, the strict compliance conditions under which clinical trials are conducted do not represent the typical clinical experience with unsupervised treatment [50]. A 2014 Pan American Health Organization epidemiological report on the state of leishmaniasis indicates that only 31.6% of cases entered in the trans-national SisLeish surveillance system included clinical course [49]. The baseline assumption that adherence to medication was as observed in RCTs and did not vary between treatment regimens is a conservative assumption that may underestimate the benefits of oral miltefosine cDOT. Oral treatment is intended to lower barriers to care versus intramuscular injection of MA. Given that literature has estimated adherence to unsupervised miltefosine treatment for visceral leishmaniasis in Asia at 83% [51] to 95% [52], we consider a high degree of adherence under a cDOT program achievable. Nonetheless, adherence will be a crucial consideration during the design and implementation of a cDOT program.
Thirdly, modeling of costs, rather than direct costing of study participants was necessary due to the inclusion of a to-date theoretical cDOT protocol for miltefosine administration. The establishment and testing of specific protocols for a cDOT protocol remains a crucial step for this use of miltefosine. Among other concerns, the protocol must address re-administration of medication in cases of vomiting, the provision of specific education materials, implementation of methods to ensure adherence and adverse event accounting, and prevention of the use of the medication by household members of childbearing potential, due to miltefosine’s known teratogenicity [26,53].
Fourthly, susceptibility of distinct Leishmania species to particular drugs was not taken into account; however, current public health protocols do not identify species before initiation of treatment. A recent in vitro study of prevalent Leishmania Viannia species indicates high levels of susceptibility to both MA and miltefosine [54]. L. panamensis is the predominant strain in the area of the study [55] and has been shown to have good in vitro susceptibility to miltefosine [54]. While early tests of miltefosine indicated poor susceptibility of L. braziliensis [56], subsequent testing has found greater susceptibility in South American strains [57]. Local species and susceptibility patterns will be an important consideration in adapting miltefosine cDOT programs in other areas. Concern for the emergence of resistant strains as a result of poor adherence has been described in L. Viannia species [58], and necessitates that any forthcoming cDOT protocol ensure close monitoring to ensure continued drug efficacy.
Fifthly, variation in clinical course was simplified. Simulated patients did not experience spontaneous resolution of CL within the timeframe of the primary outcome and did not experience progression of their disease to disseminated or mucocutaneous leishmaniasis, since these variations in natural history would be expected to be comparable for equivalently efficacious drugs. Super-infection or other complications occurring during CL were not considered in the analysis. Rare but serious (CTCAE grades 4 and 5) complications were not included in the model, as none were experienced in the course of the trial from which modeling parameters were derived.
Sixthly, dosing parameters did not take into account re-dosing in the case of vomiting, or potential changes in pediatric dosing regimens given evidence from pharmacokinetic studies showing inadequate drug plasma levels under current dosing guidelines [59,60], although the costs of such cases may be extrapolated from sensitivity analysis of drug costs.
Finally, assessment of patient and clinic costs in remote, often conflict-stricken zones necessitated the use of surveys of healthcare provider to gain local perspectives of the costs to patient patients and caregivers. We believe that their assessment reasonably approximates the costs and logistics of treatment, including transportation costs, which were among the elements of highest impact in cost determination.
In summary, CL is a NTD causing significant morbidity and social stigma among marginalized pediatric populations. As new drugs are proven efficacious in treating this disease [10,27,28], opportunities for novel treatment protocols that reduce cost to both patients and national healthcare systems may be possible and merit further exploration. Our analysis shows that treatment of pediatric patients with a miltefosine cDOT protocol is cost saving from patient and societal perspectives across a range of assumptions, and efforts to reduce miltefosine pricing could ultimately lead to cost neutrality or cost savings from a government perspective. Development of such treatment programs represents a critical opportunity to improve treatment and outcomes for pediatric CL patients.
|
10.1371/journal.pcbi.1006978 | Synchronization dependent on spatial structures of a mesoscopic whole-brain network | Complex structural connectivity of the mammalian brain is believed to underlie the versatility of neural computations. Many previous studies have investigated properties of small subsystems or coarse connectivity among large brain regions that are often binarized and lack spatial information. Yet little is known about spatial embedding of the detailed whole-brain connectivity and its functional implications. We focus on closing this gap by analyzing how spatially-constrained neural connectivity shapes synchronization of the brain dynamics based on a system of coupled phase oscillators on a mammalian whole-brain network at the mesoscopic level. This was made possible by the recent development of the Allen Mouse Brain Connectivity Atlas constructed from viral tracing experiments together with a new mapping algorithm. We investigated whether the network can be compactly represented based on the spatial dependence of the network topology. We found that the connectivity has a significant spatial dependence, with spatially close brain regions strongly connected and distal regions weakly connected, following a power law. However, there are a number of residuals above the power-law fit, indicating connections between brain regions that are stronger than predicted by the power-law relationship. By measuring the sensitivity of the network order parameter, we show how these strong connections dispersed across multiple spatial scales of the network promote rapid transitions between partial synchronization and more global synchronization as the global coupling coefficient changes. We further demonstrate the significance of the locations of the residual connections, suggesting a possible link between the network complexity and the brain’s exceptional ability to swiftly switch computational states depending on stimulus and behavioral context.
| In a previous study, a data-driven large-scale model of mouse brain connectivity was constructed. This mouse brain connectivity model is estimated by a simplified model which only takes in account anatomy and distance dependence of connection strength which is best fit by a power law. The distance dependence model captures the connection strengths of the mouse whole-brain network well. But can it capture the dynamics? In this study, we show that a small number of connections which are missed by the simple spatial model lead to significant differences in dynamics. The presence of a small number of strong connections over longer distances increases sensitivity of synchronization to perturbations. Unlike the data-driven network, the network without these long-range connections, as well as the network in which these long range connections are shuffled, lose global synchronization while maintaining localized synchrony, underlining the significance of the exact topology of these distal connections in the data-driven brain network.
| Structural neural connectivity and its implications for brain function have been a long-sought subject in neuroscience. Many previous studies have been limited either to small networks of few cells or coarser connectivity among larger brain regions [1–9], often binarized and without spatial information. Recent development of the Allen Mouse Brain Connectivity Atlas from anterograde fluorescent viral tracing experiments [10] provides us the unique opportunity to investigate precise weighted anatomical connectivity of the mammalian whole brain network. Combining the mesoscopic connectivity data with spatial information of the network, we seek a parsimonious representation of the weighted whole-brain network that captures salient network properties. Specifically, we investigate whether the network can be compactly represented solely based on the spatial dependence of the network topology.
Biological networks are inherently spatially constrained. Recent studies have shown that geographic constraints play a critical role in generating graph properties of real-world neuronal networks [5, 11–20], which cannot be fully captured by classical generative network models such as the small-world network [2] and the scale-free network [21]. Yet many of the studies are limited to binarized networks [11, 12, 17, 19, 20] and are focused explicitly on comparing graph theoretical measures [11, 13–20]. In this paper, we examine spatial embedding of a weighted whole-brain connectivity, and ask whether spatial dependence alone can depict the full computational capability of the brain network by studying dynamics of the network.
By analyzing the latest connectivity data from a new mapping algorithm, we find that the network connectivity strongly depends on its spatial embedding, with spatially close brain regions strongly connected and distal regions weakly connected. We study the precise relationship between connectivity and distance, and investigate possible computational roles of positive residual connection strengths that are not captured by the spatial dependence. To probe the possible implications of the residual connections on the network dynamics, we construct a network of phase oscillators with the data-driven adjacency matrix and compare its dynamics to those of the oscillator network with the connections strictly dependent on distance. We analyze spatial structures of synchronization by measuring the order parameter for varied amounts of global coupling coefficient. We further examine the strong connections between distal brain regions by studying network dynamics when fractions of the strong residual connections are added to the spatially constrained network. Finally, we relocate the positive residuals either to connections between nearby brain regions or to different fractions of longest-range connections, thus increasing the connection strengths for the spatially close or distal brain regions while eliminating sparse, strong connections spread across different edge lengths. The networks restructured this way maintain overall connection strength of the brain network but have a connectivity topology different from that of the brain network. By comparing dynamics of such restructured networks and the data-driven whole brain network, we show that the spatial locations of the strong positive residuals are important. Specifically, our study reveals that strong connections distributed over the brain network across many length-scales enhance the capability of the system to switch between asynchronous and synchronous states, underlining the significance of the existence of these connections. The network without these long-range connections, as well as the network in which these long-range connections are shuffled, when pushed by perturbations or low coupling coefficient, lose global synchronization but maintain local synchronization over small spatial scales. In the same conditions, the data-driven network loses synchronization over all spatial scales. It is interesting to speculate that this phenomenon is necessary for the integrative processes necessary for global cognitive functions.
The mesoscopic mouse whole-brain connectivity was constructed based on viral tracing experiments available on the Allen Mouse Brain Connectivity Atlas [10] with a recently developed interpolative mapping algorithm [22]. This produced a weighted and directed structural connectivity matrix with 244 brain regions as source nodes and 488 brain regions as ipsilateral (244) and contralateral (244) target nodes. By combining the ipsilateral and contralateral connections for each hemisphere, we constructed a whole-brain connectivity matrix with 488 nodes. The data-driven mouse brain network is shown in Fig 1A, left column.
We analyzed the relationship between connection strength and spatial distance between brain regions in the data set. In accordance with previous studies on brain networks [5, 15–20], the connectome strongly depends on the spatial embedding; connections are stronger between spatially close regions and weaker between distal regions. Specifically, the connection strengths decrease with distances between brain regions following a power law (Fig 1B) rather than an exponential relationship, in agreement with previous studies on Allen Mouse Brain Connectivity data [18, 22]. Additional details on the fitting are available in Methods (“Dependence of connection strengths on interregional distance”).
We constructed adjacency matrices for the ipsilateral and the contralateral networks based on the power-law relationship, as shown in Fig 1A, middle column. While the general trend of decrease in connection strength with distance is clear and well-predicted by a power law, there are also a number of residual connection strengths that are not captured by the power-law relationship (Fig 1A, right column).
To understand the structure and effects of the residual connection weights that are not captured by the power-law dependence on distance, we had a closer look at these residuals. For both ipsilateral and contralateral connections, a long, positive tail is observed in the distribution of residual connection weights, suggesting strong distal connections above the power-law dependence on distance (Fig 2A and 2B). The strongest 20 residual connections are plotted in Fig 2C. We observed that for the ipsilateral network, connections from preparasubthalamic nucleus (PST) to subthalamic nucleus (STN), laterodorsal tegmental nucleus (LDT) to Barrington’s nucleus (B), dorsal motor nucleus of the vagus nerve (DMX) to gracile nucleus (GR), cuneate nucleus(CU) to gracile nucleus (GR), and locus ceruleus (LC) to Barrington’s nucleus (B) are a few examples of the strong distal connections unexplained by the power-law dependence on distance. For the contralateral connectivity, on the other hand, many of the strongest residuals above the power-law relationship include connections between the same regions in different hemispheres as well as connections to and from hippocampal areas.
Do these positive residual connections between distal regions have any computational significance? In other words, can we capture the full computational capacity of the mesoscopic brain network with connectivity governed by strictly distance-dependent rules, with the residuals removed? To test this, we compare dynamics of the data-driven brain network to those of an artificial, strictly distance-dependent network generated by the power-law relationship. Specifically, we built a network of coupled phase oscillators whose coupling strengths are described by the weighted adjacency matrix of the data-driven brain network or the power-law distance-dependent connectivity. Each of these Kuramoto-type phase oscillators corresponds to a brain area. Kuramoto-type coupled phase oscillators have been widely used to model oscillatory brain dynamics [23–25]. The phase of region i, represented by θi, is described by:
θ ˙ i= ω i + k ∑ j = 1 N A i j sin ( θ j ( t - τ i j ) - θ i ( t ) ) + η i ( t ) (1)
where ωi denotes the natural frequency, and k describes the coupling coefficient. Aij is the adjacency matrix of the network. For the case of the data-driven brain network, Aij = Jij where Jij indicates the adjacency matrix obtained from viral tracing data, for both ipsilateral and contralateral connections. For simulations of the artificial, distance-dependent network, Aij = Kij indicates the adjacency matrix constructed by making the connection weights strictly follow the power-law dependence on distance. The last term ηi(t) represents an additive Gaussian white noise with zero mean (〈ηi(t)〉t = 0) and variance σ n 2 / T(〈 η i ( t ) η j ( t ′ ) 〉 t = δ i j δ ( t - t ′ ) σ n 2 / T), where δij is the Kronecker delta and δ(⋅) denotes the Dirac delta function. The standard deviation σn is in radians and T is a timescale, which is set to 1 second in our study. N denotes the number of nodes of the network, which is 488 in our whole-brain simulations. The natural frequencies ωi are randomly chosen from a symmetric, unimodal distribution g(ω). In this paper, we used a Gaussian distribution with the mean at 40 Hz and the standard deviation σd for g(ω), as done in other studies of modeling large-scale brain dynamics with phase oscillators [24–27]. Note that this falls within a frequency range of gamma rhythms (30-80 Hz) that are frequently observed in oscillatory brain dynamics.
Numerous previous studies have shown the importance of distance-dependent delays in networks of oscillators [28–33]. For example, time delays can destabilize synchrony in neuronal networks, leading to travelling waves [29–33]. To reproduce synaptic and axonal conduction delays dependent on connection distance, we incorporated distance-dependent time delays in our model as done in other studies [23–27, 34–37]. In the rodent brain, the conduction velocity ranges from values as low as 0.5 (m/s) to much higher speed around 10 (m/s) depending on various factors such as axonal myelination [32, 38, 39]. Experimental studies show that the propagation speed distributions peak in between 2-5 (m/s) [32, 39]. While time delays are heterogeneous over different regions in the brain, we simplified the model by using a fixed conduction speed at 3.5 (m/s) for the whole brain, which falls in the middle of the propagation speed distribution peak. In Eq 1, the distance-dependent time delay between areas i and j is denoted by τij, which is computed by dividing the Euclidean distance dij between nodes i and j by the fixed conduction speed.
We investigated the dynamics of the data-driven network and the power-law generated network using Eq 1, and measured the network coherence by calculating the “universal” order parameter r, recently proposed in [40] as following:
r≡1∑i=1Nki∑i,j=1NAij〈Re(ei(θi−θj))〉t=1∑i=1Nki∑i,j=1NAij〈cos(θi−θj)〉t (2)
where k i = ∑ j = 1 N A i j is the input strength of node i. Unlike the original order parameter which was proposed by Kuramoto [41, 42] for all-to-all coupled phase oscillators (see Eq 6 in Methods), the universal order parameter [40] was developed to quantify coherence in more general, weighted networks of oscillators. The universal order parameter accounts for the network topology and its influence on the phase coherence. Therefore, we can compare network coherence in topologically different weighted networks even when their total connections strengths are not the same. Furthermore, the universal order parameter captures partially phase-locked states accurately. To quantify different degrees of network coherence and to visualize localized and global synchrony, we measured the universal order parameter, both for the whole network of oscillators (Eq 7 in Methods) as well as for subnetworks of different spatial scales (Eq 8 in Methods). To compute order parameters of subnetworks on the spatial scale d, we measured the averaged phase difference for each node i, with all the other nodes that are within the given spatial distance d from the node i. Thus computed averaged phase difference for each node is weighted by the inverse of input strengths to the given node i provided by its neighbors within the distance d from the node, and summed over all regions in the whole network. By thus computing the order parameter for the subnetworks, we describe the order parameter as a function of distance.
Obtaining an explicit, analytical relationship between the order parameter and generalized network structures has been a challenging problem in studies of phase oscillators on complex networks [43, 44]. While analytical expressions for the order parameter as a function of the adjacency matrix have been derived in previous works, these mean-field approaches are based on strong assumptions of a large network with sufficiently high average degree, valid only near the onset of synchronization [41, 45–48]. Existing analytical approaches, therefore, are not applicable to the complex mesoscopic brain network of a finite size. We thus address the relationship between the network coherence and the network structure by computing the order parameter based on numerically obtained time series of the oscillators.
Phases were initialized randomly, and Eq 1 was integrated numerically using the Forward Euler method, with a sufficiently small time step Δt = 10−4 (s) for 4 seconds (Nt = 40000 steps), until a stationary state is reached. In our simulations, the time step size Δt = 10−4 (s) satisfies the condition Δ t ≤ 0 . 01 / max ( max ( k · A i j ) , 0 . 05 μ , σ n 2 2 T , 1 ), as in [37]. The data from the first Nt/2 steps are discarded in measuring the order parameter. The order parameter, representing network coherence, can be modulated by the global coupling coefficient k, the standard deviation σd of the intrinsic frequency distribution, and the standard deviation σn in the additive Gaussian white noise. In this paper, we computed the order parameter using Eq 8 in Methods for varied global coupling coefficient k, with the standard deviation of the natural frequency distribution fixed at σd = 0 (Hz) and the standard deviation of the Gaussian noise fixed at σn = 2 (rad). For each value of coupling coefficient k, we performed 10 independent runs, and plotted the average and the standard deviation of the order parameter as a function of distance between nodes (Fig 3B) as well as a function of global coupling coefficient k (Fig 3C). We also show the order parameter as a function of the coupling coefficient k, with a nonzero standard deviation in the natural frequency distribution in the Supporting Information S2(C) Fig. In this figure, the order parameter was averaged over 100 repeats with σd = 0.2 (Hz) and σn = 2 (rad), to offset different effects of each configuataion of the intrinsic frequencies due to the nonzero σd.
When the standard deviation of the white noise is held constant, increasing the coupling coefficient k with fixed σd has qualitatively the same effect as decreasing σd with k fixed, as the ratio of k/σd determines the network coherence. The same is true for decreasing the amount of σn. We show that varying σn and σd produces the same qualitative results as with varying k in the Supporting Information S2 Fig. When σn is varied, the intrinsic frequency distribution and the coupling coefficient are held constant, at σd = 0 and k = 3, and the order parameter was averaged over 10 repeats. When σd is varied, the other two parameters are fixed at σn = 0 and k = 2, and the order parameter was averaged over 100 repeats to account for the dependence of the time series on different configurations of the intrinsic frequencies. By computing the sensitivity of the network synchronizability on perturbation in each of these parameters—k, σn, or σd, we show that the observed trend in the data-driven brain network and the power-law approximated network is robust.
In Fig 3A, we show the phase difference cos(θi − θj) for pairs of nodes (i,j) plotted against time and distance between the nodes. Interestingly, for the same amount of change in coupling coefficient Δk, the data-driven brain network switches between an asynchronous state and near-global synchrony, while the power-law governed network fails to make such a drastic change in synchronization state. This difference is manifested in the order parameter. Fig 3B shows the universal order parameter (Eq 8) for subnetworks of different spatial ranges. When k is small, both the data-driven brain network and the power-law approximated network have overall low order parameters. In both cases, however, the order parameter is higher for small spatial scales, indicating that there is some spatially localized coherence in the networks due to the general trend of decreasing connection strength with distance between the connected regions.
At a finer scale, we also observe a small amount of initial increase in the order parameter for the shortest-range connections (110-346 μm) followed by a slow decrease in the order parameter as a function of distance in the data-driven brain connectome. Such an initial increase in the order parameter is not seen in the power-law estimated network. However, this initial rise at the very small length-scale should not be over-interpreted, because the experimental data are based on the mesoscopic measurements which are not accurate for distances less than 300-500 μm. In the viral tracing experimental data, the average distance to the closest injection is typically 500 μm at source level, which limits resolution [10, 22].
In the data-driven brain network, increasing the coupling coefficient k results in a transition from partial coherence to near-global synchrony, manifested by increased order parameters across a range of spatial scales (Fig 3B, left column, Data). However, in the artificially generated, strictly distance-dependent network, the same amount of change in global coupling coefficient does not induce such a leap in the network coherence state as in the real brain network (Fig 3B, right column, Power law).
Such trends can be also visualized in the order parameter for the whole network. The overall universal order parameter increases with global coupling coefficient in both the data-driven and the power-law networks (Fig 3C). However, the change in order parameter is significantly larger in the data-driven brain network. This trend appears in both the single hemisphere network with only ipsilateral connections (Supporting Information S1 Fig) and the whole brain network with both ipsilateral and contralateral connections (Fig 3C). For comparison, Kuramoto’s original order parameter (Fig 3C, dotted) is also plotted. Because the original Kuramoto’s order parameter does not account for different connection strengths among different pairs of nodes nor measure coherence scaled to the overall degree of the network, we see that the Kuramoto order parameter is lower than the universal order parameter for the power-law network. Nevertheless, for either type of the order parameter, we observe that the data-driven brain network spans a larger range of coherence states than the power-law governed network.
These trends are more clearly portrayed by plotting the sensitivity of the order parameter (Δr/Δk) as the coupling coefficient k is varied (Fig 3D). We observe that the sensitivity remains relatively constant throughout the range of the coupling coefficient in the power-law approximated network. However, the sensitivity of synchronizability is considerably more variable in the data-driven network, peaking around k = 2.5. As the coupling coefficient increases, the sensitivity in synchronizability thus increases and then drops after reaching the maximum in the data-driven network, while the power-law approximated network is marked by relatively invariant, low sensitivity of the order parameter.
This result on order parameter can be manifested by a couple of simple measures we use here. To compare the maximum sensitivity of the order parameter to changes in the global coupling coefficient k (or any other parameters that modulate synchronizability, such as σn and σd), we introduce a measure of the maximum sensitivity of synchronizability:
Γ k = max k , k + Δ k ( Δ r Δ k ) . (3)
For the power-law network, the averaged maximum sensitivity of the order parameter is Γk = 0.1144 ± 0.0214. The maximum sensitivity of the order parameter is higher in the data-driven mouse brain network, at Γk = 0.3172 ± 0.0829. The higher value of the sensitivity measure Γk for the data-driven brain network indicates that a small amount of change in the coupling coefficient can induce a significant change in the network’s coherence state, in particular, within the range of k where Δr/Δk is maximum.
To evaluate spatial dependence of the order parameter, we use another measure that quantifies the difference between the order parameter for short-range subnetworks and the order parameter for the whole network. This measure is defined as:
Γd=〈 r(d=dshortest)−r(d=dlongest)〉k, (4)
where dshortest is the distance less than 570 (μm) that generates the highest order parameter value r(d), and dlongest is 11955 (μm) which is the longest connection length in the mouse whole-brain network. 〈⋅〉k denotes averaging across varied coupling coefficient k, and r(d) is computed as defined in Eq 8. For the data-driven brain network and the power-law-driven network, Γd = 0.1851 ± 0.0706 and Γd = 0.5383 ± 0.0234, respectively. The larger Γd of the power-law network depicts a larger drop in coherence as the region of interest expands from the spatial vicinity to the whole network in the power-law network. In other words, the power-law network exhibits more localized coherence throughout a range of varied coupling coefficients.
We also confirmed that such difference between the data-driven brain network and the strictly distance-dependent, power-law network remains unchanged when the natural frequencies of the nodes are moved to 8 Hz and 20 Hz, which are in the ranges of theta (6-12 Hz) and beta (10-30 Hz) oscillations, respectively. Like gamma oscillations, theta and beta oscillations are frequently observed in the large-scale brain dynamics. While gamma oscillations are thought to be linked to cognitive processing and sensing, theta rhythms are observed in hippocampal LFP and thus believed to underlie memory formation. On the other hand, beta rhythms have been associated with movement preparation and motor coordination [23, 49, 50]. As with the natural frequencies at 40 Hz, simulations with intrinsic frequencies at 20 Hz and 8 Hz also predict that the synchronizability is more sensitive to changes in global coupling coefficient in the data-driven brain network than in the power-law approximated network (Supporting Information S3 Fig). With the realistic propagation speed 3.5 (m/s) and the longest connection at 11955 (μm) in the mouse whole-brain, the time delays in our model are quite small, and thus different intrinsic frequencies within the biologically realistic range induce qualitatively the same trend in synchronizability. It has been shown in previous studies that when the time delays multiplied by the intrinsic frequencies are sufficiently small compared to the coupling strengths in phase oscillators, the delay enters as a simple phase-lag [51, 52].
Our results indicate that in the real brain network, a small change in the global coupling coefficient induces a rapid transition between partial network synchrony and a more globally synchronized state, while in the network with connections strictly following a power-law dependence on distance, such a rapid transition to synchronization is not observed. We get qualitatively the same results when we vary parameters other than the coupling coefficient k, namely, σd and σn, to modulate the network synchronizability. The order parameter is more sensitive to changes in the dispersion of intrinsic frequencies (σd) and the standard deviation in the additive white noise (σn) in the data-driven brain network than in the power-law governed network as well (Supporting Information S2 Fig). Therefore, the residual connection strengths that are not explained by the simple spatial rule may have some computational significance, enabling even small perturbations in cognitive or behavioral states to induce a transition to synchronization.
We next examined what aspects of the residual connection strengths confer the network’s ability to span a wide range of coherence states. In previous studies on coupled oscillators, it has been found that even a small fraction of shortcuts in a small-world network significantly improves synchronization of the network [43, 53]. Motivated by this, we hypothesized that positive residual connections, namely, strong connections between distal brain regions, underlie the rapid transition in network synchronies. We tested this hypothesis by re-introducing small fractions of the positive residuals to the power-law distance-dependent network. As manifested in Fig 4A, adding just a small fraction (top 20 percentile) of the strongest positive residuals to the power-law generated network recovers the steep increase in order parameter with growing coupling coefficient (Fig 4, purple). As the fraction of positive residuals included in addition to the power-law network increases, the sensitivity of the order parameter as a function of the coupling coefficient resembles more of that of the real brain network (Fig 4B). This trend is also depicted by the maximum sensitivity measure which is at Γk = 0.1423 ± 0.0036, Γk = 0.1617 ± 0.1266, and Γk = 0.2785 ± 0.0506, respectively for top 5, 20, 40% of the positive residuals added to the power-law network, on the same edges as in the original data-driven whole-brain network.
Does the location of these strong connections have any significance in emergence of the rapid phase transition? To test whether the sensitivity of the network coherence to coupling coefficient can be recovered by simply adding the positive residuals anywhere to increase the overall connection strength of the power-law network, we studied the dynamics of the network constructed by relocating the positive residuals. We generated three networks with positive residuals relocated. In one of them, the positive residuals above the power-law relationship were positioned at random locations on the network (shuffled). In the other two, the positive residuals were preferentially relocated to the shortest 0.2% or to the longest 0.2% connections of the total edges. For the proximal-relocated network, the positive residual connections were added to connections between spatially close regions, by distributing the total positive residual connection strength among the connections between nodes within 570μm. For the distal-relocated network, the positive residuals were added to the connections between spatially distal regions, by distributing the total positive residual connection strength among the edges longer than 10500μm. The resulting networks thus maintain the total connection strengths of the real brain network, but have altered network structures.
When the locations of the positive residuals are randomized and thus there are strong connection weights across multiple spatial scales, the dependence of network synchronization on k remains similar to that of the data-driven network, as portrayed by the order parameter in Fig 5A, in gray and Fig 5B, left. In other words, although the precise network structure is different from that of the data-driven network, the network with shuffled residuals maintains its sensitivity to the global coupling coefficient, rapidly changing network coherence states. However, the spatial structure of the order parameter is dependent on the precise locations of these positive residuals. In the network with positive residuals randomly relocated, there is a steeper decrease in order parameter with distance (Fig 5B, left), compared to the data-driven brain network (Fig 3B, left). This trend is depicted by the higher spatial coherence measure, Γd = 0.4091 ± 0.00017 for the network with randomized positive residuals, compared to the data-driven brain network (Γd = 0.1851 ± 0.0706).
When the positive residuals are relocated to proximal connections, the network coherence is no longer as sensitive to small changes in the global coupling coefficient as in the whole-brain network (Fig 5A, dotted black; B, middle), in spite of the unaltered total connection strengths. This trend is robustly maintained when the standard deviation in the natural frequency distribution is varied instead of the global coupling coefficient (Supporting Information S4 Fig). Similarly, when the positive residuals are moved to distal connections, the network coherence loses sensitivity as well (Fig 5A, solid black; B, right). Unlike the network with randomly relocated residuals, the networks with positive residuals relocated only to proximal or distal connections lack strong connections distributed across a range of spatial scales. Thus, connections that are stronger than predicted by the distance-dependence should be spread over varied lengths of edges, for the network to switch between localized and global coherence states with a small change in the global coupling coefficient.
In addition, we observe that when positive residuals are relocated to proximal connections, the overall order parameters across the spatial scales are higher (Fig 5B, middle), compared to the network constructed by placing positive residuals to distal connections (Fig 5B, right). This effect arises from the definition of the universal order parameter (Eqs 7 and 8), where each of the time-averaged phase difference 〈cos(θi − θj)〉t is weighted by the connection strength between the pair of oscillators Aij. When positive residuals are placed on proximal connections, the influence of the phase differences between nearby nodes, which increases the overall network order parameter, is emphasized more by larger connection strengths Aij. On the other hand, when the positive residuals are relocated to distal connections, although distal nodes are now more strongly coupled than before, the phase differences between distal nodes are still quite large. Therefore, in this case, the large phase differences between distal nodes which lower the overall order parameter, are strongly weighted by Aij, and thus, the overall network order parameters are maintained at low values. We also note that the order parameter rapidly increases at large distances in the power-law network with the residuals preferentially added to the longest edges (Fig 5B, right). This rapid increase stems from the relatively high values of the connections strengths of these longest edges (Aij) which induce large values of 〈cos(θi − θj)〉t between distal regions i and j. Therefore, the order parameters at the large spatial scales are increased by large values of Aij〈cos(θi − θj)〉t terms.
To further examine the relationship between the spatial spread of the strong connections and the sensitivity of synchronizability, we measured the order parameter in networks generated from the power-law approximation by placing the positive residual strengths to different fractions of the longest edges. In Fig 5C, we show the order parameter as a function of the coupling coefficient k when positive residuals are preferentially added back on edges that have lengths greater than various cutoff values. As the spatial scale over which the residuals are added widens, the sensitivity of the order parameter gradually increases. Notably, the sensitivity and the growth of the order parameter become comparable to those of the data-driven brain network when the percentile of the longest edges with added positive residuals reaches 5 − 10% of the total connections. This indicates that while it is important to have a spread of strong connections above the power-law prediction over multiple spatial scales, the spread does not have to extend all the way to the shortest edges of the network in order to generate high sensitivity of the order parameter observed in the data-driven brain network.
Our results show that the location of strong connections above the power-law dependence on distance is critical for generating a steep change in the order parameter. While the precise positions of the strong connections do not have to match those of the data-driven network to produce highly sensitive order parameter to the coupling coefficient, there should be a sufficient amount of strong connections across a range of spatial scales. Precise locations of the strong residuals, however, determine the order parameter’s dependence on the spatial scale, modulating spatial coherence patterns. In sum, the spatial structure of the network connectivity plays a key role in maintaining the brain’s ability to change its computational states with small perturbations, and such sensitivity cannot be achieved by simply matching the total network connection strengths. The structure does not have to precisely match that of the real brain network to maintain the high sensitivity. What is critical to maintain, rather, is some connections stronger than the simple distance-dependence distributed over the network. However, the precise connectivity structure is important for generating specific spatial coherence patterns in the network dynamics.
In this paper, we studied synchronization of a spatially constrained model of a weighted whole-brain network at the mesoscale, constructed from viral tracing experiments. The importance of linking connectivity structure and large-scale brain dynamics have been noted in previous studies [54–56]. In particular, the heterogeneity in structural connectivity has been proposed as a key underlying mechanism for certain brain network dynamic features such as functional hubs in resting state dynamics [56]. However, additional complexities in the anatomically precise, weighted and directed whole-brain network that are not captured by spatially-defined connectivity have been often overlooked. In this work, we propose possible computational roles of these additional complexities by studying their effects on network synchronizability. We found that the connectivity has a significant spatial dependence, with the connection strength decreasing with distance between the regions following a power law. However, by studying the network dynamics of phase oscillators, we found that a network generated by the simple spatial constraints alone cannot reproduce the full computational versatility of the mesoscopic whole-brain network. Rather, we need to consider additional complexities of the network structure to capture their possibly significant roles in neural computation. Specifically, we found that residual connections not explained by the power-law dependence on distance have a long positive tail, corresponding to strong connections between distal brain regions. By computing the recently proposed universal order parameter, we showed that these strong distal connections underlie sensitive dependence of network synchrony on perturbations in coupling coefficient (or intrinsic frequency distribution/noise), potentially responsible for the brain’s exceptional ability to change its computational states depending on stimulus and behavioral context. Furthermore, our analyses on networks constructed by adding a small fraction of strong positive residuals to the spatially-constrained connectivity, as well as networks with the positive residuals relocated to random, proximal, or distal connections, reveal the key element underlying the rapid switch between global and partial synchronies—strong connections distributed over varied spatial distances. In other words, the network’s sensitivity to perturbation cannot be reproduced by simply manipulating the overall connection strengths, as locations of positive residual connections should be taken into consideration. A spatially-constrained model plus an idiosyncratic sparse matrix which features strong connections between distal regions provides a parsimonious representation of the measured connectivity.
We hypothesize that the sharp transition in synchronization in the data-driven network, which is absent in the spatially-constrained power-law model, may underlie the brain’s ability to rapidly switch computational states [57]. Such a feature is known to be impaired in the brain under pathological conditions such as Alzheimer’s disease, suggested by studies showing more modular structures and decreased global efficiency in brain connectivity constructed from EEG, MEG, fMRI, and diffusion tensor tractography [58–61]. Moreover, there is an experimental evidence for disruption of long-range connections in Alzheimer brain network [60], in agreement with our model results. Therefore, the strictly distance-dependent power-law network which maintains localized synchronization across a range of coupling coefficients may explain aberrant network dynamics and computational impairments in Alzheimer brains. A more detailed future study on genetically-controlled mouse models of Alzheimer’s disease will shed light on the possible link between changes in structural connectivity and impairment in rapid phase transitions of the whole-brain network.
The increased sensitivity of the network synchronizability induced by strong long-range connections further implicates a tradeoff between cost-efficiency and high functional capacity in the brain network. Such tradeoff between wiring cost and computational capacity has been suggested as a network-generating principle in a number of previous studies [18, 62–67]. The power-law dependence of connection strengths on inter-regional distance reflects spatial and energetic constraints in the brain network. Indeed, if the brain connectivity is designed to exclusively optimize the wiring cost, we will observe strong connections only between proximal regions. Yet, we observe some idiosyncratic, strong long-range connections which are expensive in the mouse brain connectome. By showing that these strong distal connections may serve to promote rapid transitions between network synchronization states and possibly, computational states, our work points to a possible functional role afforded by the presence of the long-range connections despite their high metabolic costs.
In this paper, we infer the dynamics of the mesoscopic brain network by constructing a network of phase oscillators with the coupling strengths determined by the structural connectivity obtained by viral tracing experiments. Thus, while the structural connectivity is based on actual data, the dynamics we conferred on the network are arbitrary. Building a more realistic, data-driven dynamic network based on imaging experiments such as calcium-imaging, ECoG, LFP, and MEG will be a crucial future extension of our study of connecting the network structures to the network dynamics. Furthermore, for future studies, more biophysically-motivated neural mass models [68] would be necessary to capture realistic dynamics of the brain network that are not predicted by simple phase oscillator models. However, our simulations with phase oscillators, despite their generality, still make valuable predictions on computational roles of spatial structures of the mesoscopic whole-brain network, underlining the importance of spatially distributed, strong distal connections on the network dynamics.
The mesoscopic mouse whole-brain connectivity was obtained from the Allen Mouse Brain Connectivity Atlas (http://connectivity.brain-map.org/), constructed based on anterograde viral tracing experiments in wild type C7BL/6 mice [10]. Based on the experimental data, a recently developed interpolative mapping algorithm was used to construct a model of whole brain connectivity at the 100 μm-voxel scale [22]. The voxel-based connection strengths were averaged over each brain region to produce a connectivity matrix with 244 brain regions per hemisphere as nodes, larger than the adjacency matrix of 213 pairs of nodes previously obtained from the linear model in [10]. For elements of the connectivity matrix, we use the normalized projection density, defined as the connection strength between two regions divided by the volume of the source and target regions. In order to account for the size of the source region, we also studied the relationship between the connection strength divided only by the size of the target region and the distance between two regions. In this case, however, the fit to either a power law or an exponential function was not very good which is not surprising given that the connection strengths that are not fully normalized with respect to the size of the source and the target is not an intrinsic quantity. For more details on the viral tracing experiments and the interpolative algorithm used to construct the connectivity matrix, see [10] and [22]. The connectivity matrix was first normalized to have values between 0 and 1. For the ipsilateral connection matrix, the diagonal entries were set to zeros removing self-connectivity, as done in [4, 20].
We fitted connection strengths as a function of interregional distance, where the distance between each pair of nodes was determined by computing the Euclidean distance in 3-dimensional coordinates between the centroids of the brain regions. Specifically, power-law functions for relationships between connection strength and interregional distance were fitted by using least squares on the log scale. For each of the ipsilateral and contralateral connectivity matrices, we found α and β by fitting the data to A ˜ i j = α · d i j - β + ∊ i j, where A ˜ i j denotes the connection strength from node j to node i, dij indicates the distance between nodes i and j, and ϵij is the residual error. We obtained α = 6.92 × 106 and β = 2.886 for ipsilateral connectivity, and α = 6.71 × 105 and β = 2.685 for contralateral connectivity (Fig 1B). In agreement with previous studies on Allen Mouse Brain Connectivity data [18, 22], we found that the power law explains the relationship slightly better than the exponential dependence (ipsilateral r-square: 0.264 vs 0.257, rmse: 1.089 vs 1.095; contralateral r-square: 0.167 vs 0.135, rmse: 1.124 vs 1.146).
We also investigated the power-law constrained network where the relationship between connection strength and interregional distance was found on the real scale, using nonlinear least squares (Levenberg-Marquardt algorithm), which has a poorer explanatory power than linear least squares on the log-scale (r-square: 0.264 vs 0.157 (ipsilateral) / 0.167 vs 0.131 (contralateral)). While this method generated a different power-law function from the one found by least squares on the log-log scale, the dynamics on the power-law network obtained by using nonlinear least squares maintained the same core characteristics, distinct from the data-driven brain network– the order parameter is less sensitive to changes in the global coupling coefficient.
In this section, we describe order parameters that were proposed previously [41, 42, 45, 46], demonstrating advantages of the recently developed universal order parameter [40] in our analysis.
In order to quantify network coherence in the original model of phase oscillators with all-to-all connectivity, Kuramoto introduced the complex order parameter [41, 42],
r ( t ) e i ψ ( t ) ≡ 1 N ∑ i = 1 N e i θ i , (5)
where ψ(t) gives the average phase of all oscillators and r(t) describes the degree of phase coherence at time t. The overall phase coherence is measured by the absolute value of the complex order parameter averaged over time. We denote this value rKuramoto, as the measure of the averaged phase differences of all pairs of oscillators:
r Kuramoto 2 ≡⟨ | r ( t ) e i ψ ( t ) | 2⟩ t = ⟨ 1 N 2 ∑ i , j = 1 N e i ( θ i - θ j )⟩ t = 1 N 2 ∑ i , j = 1 N ⟨ cos ( θ i - θ j )⟩ t . (6)< … >t denotes the average over time. However, this unweighted order parameter is not a good measure when comparing collective synchronizations in two networks described by different connectivity matrices, as it does not capture the topology of the networks.
To extend the use of order parameter to more general, weighted networks of oscillators, Restrepo et al [45, 46] proposed an order parameter which is defined as the average of local order parameters which measure the coherence of the inputs to each node. This parameter, however, does not capture partially phase-locked states well. Recently, Schroeder et al [40] proposed a new, “universal order parameter” to accurately measure phase coherence in weighted and directed networks of arbitrary topology, which overcomes the shortcomings of the previous order parameters. This newly proposed universal order parameter is defined as:
r ≡ 1 ∑ i = 1 N k i ∑ i , j = 1 N A i j ⟨ R e ( e i ( θ i - θ j ) )⟩ t = 1 ∑ i = 1 N k i ∑ i , j = 1 N A i j⟨ cos ( θ i - θ j )⟩ t (7)
where k i = ∑ j = 1 N A i j is the input strength of node i. Note that in unweighted binary networks, this measure represents in-degree [4]. This order parameter accounts for the network topology and its influence on the phase coherence, enabling a fair comparison between two topologically different weighted networks even when their total connection strengths are not matched. As this universal order parameter accurately captures partial synchrony within the network, different degrees of synchronization can be measured by order parameter of the whole network.
Furthermore, degree of coherence as a function of spatial extent can be obtained by computing the order parameter for subnetworks of different spatial scales. The order parameter r can be described as a function of distance d:
r ( d ) ≡ 1 ∑ i = 1 N k i ∑ i = 1 N ∑ j ∈ γ ( i , d ) A i j ⟨ R e ( e i ( θ i - θ j ) )⟩ t = 1 ∑ i = 1 N k i ∑ i = 1 N ∑ j ∈ γ ( i , d ) A i j ⟨ cos ( θ i - θ j )⟩ t (8)
where γ(i, d) indicates the set of nodes within spatial distance d from node i. ki = ∑j∈γ(i,d) Aij is the total connection strength of node i when the subnetwork composed of nodes within distance d from node i is considered. The order parameter of the whole network is obtained when d = size of the network (11752μm for ipsilateral and 11955μm for contralateral connectivity).
All of the MATLAB code used to numerically compute time-series data of coupled oscillators and the order parameters on the mouse whole-brain network from [10, 22] and the power-law approximated network are available at https://github.com/AllenInstitute/Choi2019_ConnectomeSynchrony.
|
10.1371/journal.pntd.0003932 | Histamine 1 Receptor Blockade Enhances Eosinophil-Mediated Clearance of Adult Filarial Worms | Filariae are tissue-invasive nematodes that cause diseases such as elephantiasis and river blindness. The goal of this study was to characterize the role of histamine during Litomosoides sigmodontis infection of BALB/c mice, a murine model of filariasis. Time course studies demonstrated that while expression of histidine decarboxylase mRNA increases throughout 12 weeks of infection, serum levels of histamine exhibit two peaks—one 30 minutes after primary infection and one 8 weeks later. Interestingly, mice treated with fexofenadine, a histamine receptor 1 inhibitor, demonstrated significantly reduced worm burden in infected mice compared to untreated infected controls. Although fexofenadine-treated mice had decreased antigen-specific IgE levels as well as lower splenocyte IL-5 and IFNγ production, they exhibited a greater than fourfold rise in eosinophil numbers at the tissue site where adult L. sigmodontis worms reside. Fexofenadine-mediated clearance of L. sigmodontis worms was dependent on host eosinophils, as fexofenadine did not decrease worm burdens in eosinophil-deficient dblGATA mice. These findings suggest that histamine release induced by tissue invasive helminths may aid parasite survival by diminishing eosinophilic responses. Further, these results raise the possibility that combining H1 receptor inhibitors with current anthelmintics may improve treatment efficacy for filariae and other tissue-invasive helminths.
| Filariae are tissue-invasive parasitic roundworms that infect over 100 million people worldwide and cause debilitating conditions such as river blindness and elephantiasis. One of the major factors limiting our ability to eliminate these infections is the lack of drugs that kill adult worms when given as a short course therapy. Additionally, the mechanisms by which adult worms are cleared from infected individuals remains poorly understood. In this study, we demonstrate that treatment of infected mice with fexofenadine, an inhibitor of histamine receptor 1, significantly reduces adult worm numbers through a mechanism dependent on host eosinophils. These findings suggest that histamine release induced by parasitic worms may aid parasite survival by decreasing eosinophilic responses. Further, as antihistamines are generally safe medications, these results raise the possibility that antihistamine therapy may be useful either alone, or potentially in combination with other antifilarial medications such as diethylcarbamazine (DEC), to eliminate adult filarial worms from infected individuals.
| Filariae are vector-borne tissue-invasive nematodes that infect over 100 million people worldwide and cause the debilitating conditions of river blindness and elephantiasis [1]. A major obstacle to ongoing efforts to control and potentially eradicate these diseases is the limited ability of anti-filarial drugs to kill adult worms, especially when given as single dose treatments.
One of the fairly unique aspects of helminth infections, in contrast to infection with most other pathogens, is the induction of histamine release in response to the parasites. Like other helminths, filariae induce the production of antigen-specific IgE, which then sensitizes basophils and mast cells to release histamine in response to parasite antigens. Histamine (2-[4-imidazolyl]ethylamine) is a short-acting biogenic amine that, in addition to having potent acute inflammatory properties, also has numerous immunomodulatory effects on chronic inflammation [2]. Histamine is synthesized by the enzyme histamine decarboxylase (HDC) and is either stored in cytoplasmic granules in basophils and mast cells or is immediately released into the periphery [3]. Histamine release from both basophils and mast cells in response to parasite antigen has been observed in numerous studies of helminth infections [2,4–8]. Although sensitivity to parasite antigens is primarily dependent on parasite-specific IgE [9], several helminths can also induce histamine release in the absence of parasite-specific IgE [10]. In this study we investigated the role histamine plays in the immune response to filariae and the effect antihistamine therapy has on filarial worm burdens. Using the Litomosoides sigmodontis/mouse model we observed that administration of fexofenadine, a histamine receptor 1 antagonist (HR1i), reduces adult worm numbers by over 50%. Additionally, clearance of adult worms in HR1i treated mice was found to be primarily eosinophil dependent, as HR1i administration did not enhance worm clearance in eosinophil-deficient infected mice.
All experiments were performed under protocols approved by the Uniformed Services University Institutional Animal Care and Use Committee.
Female BALB/c (NCI, Frederick, MD), and BALB/c eosinophil deficient (ΔdblGATA mice, The Jackson Laboratory, Bar Harbor ME), were maintained at the Uniformed Services University with free access to food and water. At study endpoints, all animals were euthanized using carbon dioxide followed by cervical dislocation. Blood was collected by cardiac puncture.
For Litomosoides sigmodontis infection, L3-stage larvae (L3s) were obtained from infected jirds (Meriones unguiculatus, TRS labs, Atlanta, GA) by pleural lavage with RPMI 1640. 40 L3s were collected and injected subcutaneously (dorsal neck) into 6–10 week old mice as previously described [11].
For microfilarial counts, 30 μl of blood was taken and mixed with 1 mL ACK lysing buffer (Quality Biological). To count microfilarial numbers, the lysate was pelleted and assessed for counts microscopically.
Fexofenadine HCl, an HR1 antagonist (HR1i), was dissolved in the drinking water at a concentration of 0.25 mg/ml for an average daily dosage of 20mg/kg/day. Cimetidine (Sigma-Aldrich), an HR2 antagonist, was prepared by dissolving in hydrochloric acid (HCl) and mixed with water. The pH was then adjusted to 7.0 with sodium hydroxide (NaOH). The final concentration of cimetidine was 2.5 mg/ml in drinking water, for an average daily dosage of 200 mg/kg/day. Drinking water bottles containing antihistamines were changed every other day. Antihistamine activity was confirmed by testing stomach pH at time of euthanasia (for HR1 antagonists) and by local anaphylaxis in response to a direct histamine challenge (for HR2 antagonists).
Blood was collected at different time points in heparinized plasma separator microfuge tubes (Starstedt, Nümbrecht, Germany). Samples were centrifuged at 15,000 X g for 1.5 minutes. Histamine in plasma was detected using a commercially available histamine ELISA assay according to the manufacturer’s instructions (Beckman-Coulter).
Adult worms were collected from the pleural cavity of infected animals at 8 weeks post infection. Adult worms were fixed overnight in 4% paraformaldehyde and were washed in 70% ethanol prior to histological processing by Histoserv, Inc (Rockville, MD). In brief, the fixed tissue was dehydrated through graded alcohols, cleared in xylene and infiltrated with paraffin. The processed tissue was then embedded in paraffin and sectioned on a microtome at 5 microns. The slides were then deparaffinized in xylene, hydrated through graded alcohols to water then stained with Carazzi’s hematoxylin. Following a water rinse, they were stained with eosin and dehydrated with graded alcohols. The slides were then cleared using xylene and coverslipped with permount.
RNA from whole blood was isolated according to the manufacturer’s instructions (Ambion, Mouse Whole Blood RNA isolation). cDNA synthesis was performed using random primers according to the manufacturer’s instructions (iScript cDNA synthesis kit, BioRad). RT-PCR was performed using a murine histidine decarboxylase (HDC) gene expression assay following manufacturer’s instructions. Samples were analyzed using an Applied Biosystems 7500 Real-Time PCR system and results calculated as fold change relative to an endogenous 18s rRNA control using the 2-ΔΔ CT method.
L. sigmodontis worm antigen (LsAg) was prepared as previously described [12].
L3 stage larvae were collected from infected jirds as previously described. For in vitro survival assays, 200 L3s were cultured in 5ml of RPMI 1640 medium supplemented with gentamicin. Cultures were supplemented with 200mM histamine or 1mM Fexofenadine HCl and observed daily for mobility to assess survival.
To assess eosinophil numbers at site of adult worm infection, pleural cells were collected by pleural lavage. Red blood cells were lysed using ImmunoLyse kit (Beckman Coulter) and then 2.0x106cells/mL were permeabilized with BD Permeablization/Wash buffer (BD Biosciences). For analysis, cells were blocked using CD16/CD32 (soluble FcεR III/II receptor, BD Pharmingen) and stained for flow cytometry using anti-SiglecF PE, anti-CD11c APC and anti-CD45 FITC (all from BD Pharmingen).
Flow cytometry was performed using a BD LSR II system and analyzed with FACSDiVa 6.1 software (BD Biosciences). Antibodies for all flow cytometry experiments were titrated prior to use. During analysis, cut-offs for CD45 positivity and Siglec F positivity were determined using the fluorescence minus one approach.
To assess levels of eosinophil peroxidase at the site of adult worm infection, pleural fluid was collected by pleural lavage using 1 mL of sterile RMPI. EPO in lavage fluid was assessed by ELISA according to manufacturers instructions (US Biological Life Sciences).
Statistical analysis was performed using GraphPad Prism software (GraphPad Software, San Diego, Ca). To determine differences between multiple groups, analysis was performed using Kruskal-Wallis test followed by Dunn multiple comparisons. To determine differences between two un-paired groups, Mann-Whitney analysis was performed. A p value of <0.05 was considered significant.
To determine if infection with L. sigmodontis results in detectable histamine release, BALB/c mice were infected with 40 L3 stage larvae by subcutaneous injection and circulating histamine levels assessed at 30 minutes, 1, 4, 8 and 12 weeks by competitive ELISA. In this model L3 larvae migrate to the pleural space where they mature to adult worms, start releasing blood-dwelling microfilariae by 7 weeks, and survive for 12–20 weeks. There was a significant peak of circulating histamine observed 30 minutes after injection of L3s and a second, higher peak correlating with the production of microfilariae at 8 weeks of infection (Fig 1A, p < 0.01 for both timepoints when compared to age-matched uninfected controls). Of note, histamine was not detected in the blood of mice 30 minutes after injection of vehicle (sterile RPMI) or peritoneal lavage fluid (S1 Fig).
Using RT-PCR we determined the levels of histidine decarboxylase (HDC) from whole blood RNA. In contrast to histamine, blood levels of HDC mRNA increased throughout the 12 weeks, indicating that histamine may be continually synthesized during infection (Fig 1B). Using ELISA, we determined circulating levels of Ls-specific IgE. Ls-specific IgE levels became detectable after 4 weeks of infection (Fig 1C). Development of detectable Ls-specific IgE thus preceded peak histamine levels in the plasma, suggesting peak histamine release may be due to IgE-mediated basophil and mast cell activation. In contrast, the immediate early release of histamine at the 30 minute time point is suggestive of parasite-specific antibody independent activation of basophils and mast cells.
We next sought to determine whether histamine plays a role in maintaining worm burdens during primary infection. To test this, BALB/c mice were infected with 40 L3s and treated with HR1 or HR2 antagonists administered in water for the duration of infection. At 8 weeks, mice were euthanized and adult worm burden was determined. Untreated infected mice had a mean recovery of 18 adult worms. Mice treated with HR1 antagonists had a mean recovery of 8 adult worms (58.1% reduction, p = 0.001) while mice treated with HR2 antagonists had a mean recovery of 13 worms (22.5% reduction, p = 0.0573) (Fig 2). This data indicates that signaling via HR1 may play a role in long-term survival of L. sigmondontis in the mammalian host.
Given that we observed a reduction of adult worms at 8 weeks post-infection in animals treated with an HR1 antagonist, we sought to determine if HR1 antagonism altered the circulating microfilaria load or the male-to-female ratio. HR1i administration did not alter the number of microfilariae circulating in the blood (S2 Fig) or the male-to-female ratio of recovered adult worms (S3 Fig).The lack of a decrease in microfilaria burden is not too surprising as microfilaria load does not correlate with adult worm numbers [13].
Due to the observed reduction in adult worm burdens at 8 weeks, we next sought to determine whether there was a particular timepoint during infection when HR1 blockade enhances worm clearance. In a typical course of infection, L3 stage larvae migrate from the subcutaneous tissues to the pleural space from days 1–5, molt to L4 stage worms by d 11, and then molt to adult worms between days 24–30. Mice were treated for 10, 35, and 56 days post-infection with HR1 antagonist. At 56 days (8 weeks), all groups of mice were euthanized and living adult worms collected. Recovered worms that were motile yet had parts that were covered with granulomas were classified as “encased” (Fig 3A and 3B). As previously demonstrated, mice treated with HR1 antagonists for 8 weeks demonstrated a significant reduction in adult worm burden compared to untreated mice (Fig 3C). While there was no difference in total worm burden at the 8 week timepoint between mice treated with HR1 antagonists for 10 days and untreated mice (mean worm burden of 17 vs 20), mice treated with 10 days of fexofenadine (HR1i) had significantly greater numbers of adult worms that were encased in granulomas (Fig 3D) at 56 days post infection. The trend towards lower worm burdens with longer fexofenadine treatment courses suggests that H1R blockade enhances worm clearance at numerous stages of worm development.
Given the reduction in adult worm burden observed in HR1 treated mice, we next evaluated whether fexofenadine is directly toxic to L. sigmodontis worms and whether exogenous histamine enhances worm survival. To test this, L3 stage worms were cultured in vitro and supplemented daily with 200nM histamine, 10mM of fexofenadine, or media alone and assessed daily for survival. As seen in Fig 4, there were no observed differences in survival times between L3s supplemented with histamine, L3s supplemented with fexofenadine, and those supplemented with media (Fig 4). These data indicate that exogenous histamine does not directly enhance worm survival and that fexofenadine is not directly toxic to worm viability.
Because fexofenadine did not appear directly toxic to L. sigmodontis worms, we next evaluated whether H1R antagonism alters the immune response that develops during infection. To assess this, humoral and cellular immunological studies were conducted on infected mice treated with 8 weeks of fexofenadine. Both total and LsAg-specific IgE were significantly decreased in mice treated with fexofenadine compared to untreated controls (Fig 5A and 5B). In terms of cellular immune responses, parasite antigen-driven production of IL-5 and IFN-γ from splenocytes was also significantly reduced in fexofenadine treated mice (Fig 5D and 5E), whereas IL-4 production was not (Fig 5C). This suggests that signaling via HR1 may enhance both type 1 and type 2 immune responses.
In contrast to the decreases in IgE, IL-5, and IFN-γ, the cellular infiltrate present in the pleural cavity, the site where adult L. sigmodontis worms reside, was dramatically increased in fexofenadine treated mice. Whereas infection of untreated BALB/c mice resulted in a median of 1.7 x 106 cells in the pleural space at study endpoint, fexofenadine-treated mice had 4.8 x 106 cells (p = 0.0080, Fig 6A). Flow cytometric analysis revealed that eosinophils comprised over half of the cells in the pleural infiltrate of fexofenadine-treated mice, increasing in numbers from a median of 4.7 x 105 cells in untreated infected mice to 2.5 x 106 cells fexofenadine-treated infected animals (p = 0.0043, Fig 6B)
A number of studies utilizing the L. sigmondontis model have demonstrated a significant role for eosinophils in immune-mediated clearance of worms [14,15]. As fexofenadine increased eosinophil numbers at the site of adult worm infection, we next tested whether fexofenadine mediated worm clearance is dependent on eosinophils. Eosinophil deficient mice (ΔdblGATA) and background control BALB/c mice were infected with L. sigmondontis, treated for 8 weeks with HR1 antagonists, and euthanized at 8 weeks for enumeration of adult worm burden. In contrast to fexofenadine-treated wild type mice, eosinophil deficient mice administered fexofenadine had no reduction in adult worm burden (mean recovery 18) when compared to untreated ΔdblGATA controls (mean recovery 15) or BALB/c background controls (mean recovery 14) (Fig 7). To further assess the activity of eosinophils in antihistamine mediated worm clearance, BALB/c mice were infected, treated with HR1 antagonists for 8 weeks, and then euthanized. A pleural lavage was performed and ELISA used to detect eosinophil peroxidase (EPO) as evidence of eosinophil degranulation. Fexofenadine treated mice demonstrated significantly highler levels of EPO in the lavage fluid (S4 Fig). Taken together, these findings demonstrate that H1R blockade enhances worm clearance through an eosinophil-dependent mechanism.
In this study we found that histamine is released throughout filarial infection, that antihistamine therapy reduced IgE levels and increased eosinophilic responses at the site of infection, and that administration of fexofenadine, a HR1 blocker, enhances clearance of adult worms in an eosinophil-dependent manner.
Our first experiment was a time course study of circulating histamine levels to determine the kinetics of histamine release during primary filarial infection. As the t1/2 of histamine in blood is approximately 60s [16,17], blood levels are representative of ongoing histamine release. We found that histamine was released throughout the course of primary L. sigmondontis infection. The 1st peak in circulating histamine occurred 30 minutes post infection in naïve mice. This finding has two important implications. First, as basophils and mast cells are the only cells carrying pre-formed histamine [18], it suggests that one or both of these cell types are activated within minutes of filaria infection. Early activation of these cells may be important for the shaping of the immune response to tissue invasive helminths. Second, early histamine release represents a non-specific mechanism of mast cell or basophil activation, since specific IgE is not present until weeks after infection. This is consistent with a number of studies that have demonstrated direct activation of basophils by helminth antigens (reviewed in [10]).
Timecourse studies next revealed that histamine release in infected mice peaks again at 8 weeks of infection. We speculate that the 8 week peak may be due to basophil and mast cell activation in response to circulating microfilariae, which appear starting 7 weeks post-infection. As detectable LsAg-specific IgE develops by 6 wks post-infection, this activation is likely occurring through IgE. This 2nd peak is then followed by a decrease in circulating histamine, even though histamine decarboxylase message in blood cells increases throughout infection. This data is consistent with previous findings that basophils become hyporesponsive over time, requiring more signal to achieve activation [19]. Therefore, even though histamine synthesis continues throughout the course of infection, basophils and mast-cells are releasing less histamine in the chronic stages of infection.
Perhaps the most striking finding of this study is the significant reduction of adult worm burden at 8 weeks in mice treated with a HR1 antagonist. To determine the timing of worm clearance, infected mice were treated for 10, 35, or 56 days with fexofenadine and assessed for adult worm burden at 8 weeks. We found that the longer mice were treated with fexofenadine the greater the reduction in adult worm burden at 8 weeks and that treatment with fexofenadine resulted in a significant increase in encased worms in all groups. Previous studies have suggested a role for histamine in early larval invasion into the lymphatics [20]. Taken together these data indicate that that HR1 blockade enhances worm clearance at numerous stages of development.
As in vitro studies revealed that fexofenadine is not directly toxic to Litomosoides sigmodontis, we next evaluated whether fexofenadine augments immune responses directed against the parasite. Although helminth-specific IgE, IL-5 and IFNγ responses were all decreased in fexofenadine treated mice, eosinophil numbers at the site of worm infection were significantly elevated. Experiments with eosinophil deficient ΔdblGATA mice demonstrated that eosinophils were required for fexofenadine-mediated helminth clearance. In contrast to fexofenadine treated wild type mice, which had 80% fewer adult worms than wild type controls, fexofenadine treated ΔdblGATA mice exhibited no decrease in worm numbers. These results are consistent with prior studies suggesting eosinophils are key effector cells against helminths [21–23]. Studies utilizing the L. sigmondontis model have shown that mice deficient in eosinophil peroxidase or major basic protein, key eosinophil granule proteins, have significantly higher filarial worm burdens than wild type controls [14].
Prior studies [24] have shown that mice deficient in IL-5 produce neutrophilic rather than predominantly eosinophilic granulomas around L. sigmodontis worms. In our study, IL-5 production was likely not the driving mechanism for larger eosinophilic granulomas in the setting of H1R blockade as splenocytes from fexofenadine-treated mice demonstrated less IL-5 production than splenocytes from untreated infected animals. Together, the results of these papers and this study suggest that IL-5 is required for eosinophilic granuloma formation, and that fexofenadine enhances this process.
There are multiple mechanisms by which HR1 blockade may have enhanced eosinophil responses in this study. One possibility is that HR1 blockade may have enhanced eosinophil survival. Data from one in vitro study suggests histamine signaling reverses IL-5 afforded eosinophil survival [25]. A second hypothesis to explain increased eosinophil numbers at the site of worm infection is enhancement of eosinophil chemotaxis by blockade of HR1 signaling. Histamine is a known chemoattractant molecule for eosinophils [26–28], and the recently discovered [29] histamine receptor 4 (HR4) has been demonstrated to play a significant role in eosinophil chemotaxis and activation [30–35]. As such, it is possible that blockade of histamine signaling through HR1 enhances the effects of histamine through HR4. Alternatively, HR1 blockade may indirectly enhance eosinophil chemotaxis by increasing production of eosinophil chemotaxins or augmenting eosinophil sensitivity to such agents. Of note, dblGATA mice are deficient in basophils as well as eosinophils. Thus, it is possible that worm clearance in fexofenadine-treated mice is due to the action of basophils rather than eosinophils. However, as we have previously found that depletion of basophils does not alter adult worm numbers ([36], [37]), and as eosinophils are known to have the ability to kill adult filarial worms [14,38], we believe it is most likely that worm clearance induced by fexofenadine is through enhancement of eosinophil numbers at the site of infection.
One of the most interesting findings of this study is the observation that fexofenadine treatment caused significant reductions in circulating IgE levels and splenocyte production of IL-5 and IFNγ as well as increased numbers of eosinophils at the site of infection. While we can only speculate on the mechanisms underlying these apparently contrasting findings, we expect that it may be related 1) to the concentrations of histamine locally (at the site of infection) vs systemically, and 2) to unknown effects of histamine on the function of various immune effector cells. Another possibility is that decreased IgE levels and IL-5 production may have been due to the decreased adult worm burdens observed in fexofenadine-treated mice. The concentrations of histamine at different body sites during infection and the effects histamine has type 2 responses from B cells, T cells, macrophages, and dendritic cells will be the focus of future investigations. Another possibility is that decreased IgE levels and IL-5 production may have been due to the decreased adult worm burdens observed in fexofenadine-treated mice.
These findings demonstrate that histamine, in addition to its immediate proinflammatory effects, also functions to shape the immune response to helminth infections. The exact mechanisms by which this occurs are not yet clear. Histamine is known to alter the immunological function of a variety of cell types, including epithelial cells, granulocytes, T-cells, B-cells, and dendritic cells [3,39–41]. Investigations combining HR1 deficient mice with airway hyperresponsiveness models are mixed [42–45]. Whereas one showed decreases in type 2 cytokines, no changes in IFNγ, decreased IgE levels, and increased blood eosinophil numbers [42], another showed increases in type 2 cytokines, decreased IFNγ, and decreased bronchoalveolar lavage eosinophil numbers [43]. We believe the differences in these studies demonstrate the complex role histamine plays in shaping immune responses. The exact effects of histamine are likely dependent not only on the cell types involved, but also on the cytokine environment in which histamine is acting and on the repertoire of histamine receptors displayed by individual cells.
The results of our study may have some important clinical ramifications. Currently there is a worldwide effort to control and potentially eradicate lymphatic filariasis and onchocerciasis by repeated mass drug administration (MDA) of anti-filarial medications, especially diethylcarbamazine (DEC) [44]. A major factor limiting success of MDA is the inability of anti-filarial drugs to kill adult worms when given as a short course [45]. Since antifilarial medications primarily clear microfilariae, ongoing mass drug administration programs require repeated administration of antifilarial agents s for years until natural death of adult worms occurs. [46,47]. One of the interesting aspects of DEC therapy is that DEC does not appear sufficient on its own to kill filarial worms. Numerous studies have shown that DEC-mediated clearance of filariae is dependent in large part on the host immune response [48,49].Since we have shown that fexofenadine can augment immune clearance of adult filarial worms, we hypothesize that addition of fexofenadine to DEC or other antifilarial medications may result in better adult worm eradication than current regimens. Discovering a short course therapy that can successfully eliminate adult filarial worms would greatly increase our ability to control and eradicate filarial infections. Further elucidating the mechanisms by which fexofenadine decreases adult worm burdens, and investigating whether combining fexofenadine with current antifilarial medications enhances adult worm clearance, will be the focus of future studies.
|
10.1371/journal.pgen.1002813 | Balancing Selection at the Tomato RCR3 Guardee Gene Family Maintains Variation in Strength of Pathogen Defense | Coevolution between hosts and pathogens is thought to occur between interacting molecules of both species. This results in the maintenance of genetic diversity at pathogen antigens (or so-called effectors) and host resistance genes such as the major histocompatibility complex (MHC) in mammals or resistance (R) genes in plants. In plant–pathogen interactions, the current paradigm posits that a specific defense response is activated upon recognition of pathogen effectors via interaction with their corresponding R proteins. According to the “Guard-Hypothesis,” R proteins (the “guards”) can sense modification of target molecules in the host (the “guardees”) by pathogen effectors and subsequently trigger the defense response. Multiple studies have reported high genetic diversity at R genes maintained by balancing selection. In contrast, little is known about the evolutionary mechanisms shaping the guardee, which may be subject to contrasting evolutionary forces. Here we show that the evolution of the guardee RCR3 is characterized by gene duplication, frequent gene conversion, and balancing selection in the wild tomato species Solanum peruvianum. Investigating the functional characteristics of 54 natural variants through in vitro and in planta assays, we detected differences in recognition of the pathogen effector through interaction with the guardee, as well as substantial variation in the strength of the defense response. This variation is maintained by balancing selection at each copy of the RCR3 gene. Our analyses pinpoint three amino acid polymorphisms with key functional consequences for the coevolution between the guardee (RCR3) and its guard (Cf-2). We conclude that, in addition to coevolution at the “guardee-effector” interface for pathogen recognition, natural selection acts on the “guard-guardee” interface. Guardee evolution may be governed by a counterbalance between improved activation in the presence and prevention of auto-immune responses in the absence of the corresponding pathogen.
| Pathogens have a negative impact on the fitness of their hosts and are responsible for drastic epidemics in humans, animals, and plants. In plants, it has been thought that natural selection acts predominantly on so-called “resistance genes,” which recognize pathogens following a key-lock interaction. In this study, we demonstrate that the arms race between hosts and pathogens extends to other components of the immune system. We discovered a signature of balancing selection at the tomato RCR3 gene, which serves as a target for pathogen-derived molecules and facilitates recognition of the pathogen via interaction with a tomato resistance gene. Functional assays of 54 RCR3 alleles reveal that the polymorphisms underlying the observed pattern of balancing selection do not play a role in pathogen recognition, but are responsible for fine tuning the defense response of infected cells upon pathogen recognition. Therefore, the optimal RCR3 allele depends upon a delicate balance between sufficient activation in the presence of, but avoidance of auto-activation in the absence of, the pathogen. The optimization of defense activation is likely a very important aspect of immune system evolution, especially when the selection pressure by the pathogen is variable in time and space.
| The coevolutionary arms race between hosts and pathogens is often described as a recurrent struggle for increased resistance in hosts and evasion of recognition by pathogens [1]–[3]. The coevolutionary dynamics can be driven by negative frequency-dependent selection, leading to the maintenance of allelic diversity at genes involved in interactions between hosts and pathogens [4]–[7]. In plants, the molecular perception of pathogens and activation of defense are well understood (reviewed in [8]–[10]) and provide an ideal means to study coevolutionary processes. For interactions of plants with biotrophic pathogens, two layers of pathogen recognition and defense are commonly described: 1) the basal defense is initiated following recognition of common pathogen-associated molecular patterns (PAMPs), such as bacterial flagellin or LPS (reviewed in [11]), and 2) a specific defense response that is activated upon pathogen recognition of host-specific pathogens via gene-for-gene interactions of pathogen effectors with their corresponding resistance (R) proteins [2], [9], [12], [13]. The specific defense response typically involves a localized cell death response, called the hypersensitive response (HR), which stops the course of infection [10], [14].
The latter specific interaction between effector and R protein can be direct or indirect. Direct interactions between pathogen effectors and R proteins have been demonstrated in remarkably few cases, for example between Pita and AvrPita in the rice-Magnaporthe pathosystem [15] and between L and AvrL567 proteins in the flax-flax rust pathosystem [16]. However, the majority of interactions appear to be indirect, following the ‘Guard-Hypothesis’ [17], [18]. In this scenario, the pathogen effector is recognized through detection of its activity in the host. Specific target molecules in the host plant, the ‘guardees’, are modified by the activity of secreted pathogen effector molecules. This modification is detected by the R protein, which serves as the so-called ‘guard’, triggering downstream defense responses including HR.
Due to the complex interaction between the guardee, its guard and the pathogen effector, the guardee is expected to be subject to contrasting evolutionary forces [19], [20]. For example if pathogen pressure is high, positive selection on the effector-guardee interface could act to improve the detection of the effector in presence of the guard, or curtail damage caused by the effector. Alternatively, positive selection on the guard-guardee interface may improve pathogen triggered activation and/or prevent auto-activation of the defense response resulting in auto-immune response [21]. Balancing selection may act on the guardee-effector interface (due to frequency-dependent selection for pathogen recognition [22]) or guard-guardee interface (due to selection for defense activation [22], [23]), if pathogen pressure or the allele frequency of the corresponding effector vary in time or space. Although it has been shown that guardees exhibit high inter- and intraspecific diversity [20], [24], it is still unknown which evolutionary forces shape their genetic diversity and genomic structure.
To decipher the role of the guardee in the evolution of the plant immune system, we quantified the natural genetic variation and investigated the functional consequences of this variation at RCR3, a secreted papain-like cysteine protease, which is thought to be guarded by the R protein Cf-2 in tomato. Cf-2 confers resistance to the leaf mold pathogen Cladosporium fulvum through recognition of the fungal protease inhibitor AVR2, which physically interacts with and inhibits RCR3 ([25], reviewed in [26]). The AVR2-RCR3 interaction is thought to cause conformational changes in RCR3, which are detected by the Cf-2 protein, leading to activation of the Cf-2-mediated defense response [25], which typically involves HR.
C. fulvum is a host specific pathogen of the tomato clade [27]. Tomatoes (Solanum section Lycopersicon) form a monophyletic clade within the Solanaceae family. The section Lycopersicon includes a total of 13 species representing all described wild tomato species and the cultivated tomato S. lycopersicum, which diverged within the last 6 million years [28]. The native geographical distribution of wild tomato species ranges from Ecuador to northern Chile and these species are found across a range of diverse habitats including temperate deserts, Andean highlands and tropical rainforests in the Amazon basin [29]. Each species displays a characteristic geographical distribution pattern, which is defined by its habitat preference [30]. Hence, wild tomatoes are suitable model organisms to study adaptation to biotic and abiotic stress. Within the tomato clade, the obligate outcrossing species Solanum peruvianum exhibits the highest level of morphological and genetic diversity and has the largest and most variable habitat range including both arid and mesic habitats. Since this species harbors the greatest variation of all species in the clade of Lycopersicon, it is an ideal starting point to understand the interplay of functional diversity and natural selection. S. peruvianum diverged from its closest relatives at least 500,000 years ago [31], [32]. Adaptation to biotic factors plays an important role in evolution of this species [30], [33]–[35]. Furthermore, since the habitat range of this plant species is large and infection and transmission of pathogenic fungi such as C. fulvum are likely affected by climatic conditions, pathogen pressure may be variable in time and space.
Even though documentation of C. fulvum in wild populations of tomato is lacking, empirical studies suggest that this fungus is a natural, coevolving pathogen of wild tomato species. Wild tomato species (S. peruvianum, S. pimpinellifolium and S. habrochaites) can be infected by C. fulvum and respond with different levels of resistance and susceptibility to pathogen challenge [36]. The observed differences vary within and between these three species suggesting variability in historical pathogen pressure. Furthermore, resistance genes to C. fulvum are present and functional in these wild tomato species [37] and have been introgressed from resistant accessions into the cultivated tomato [38]. A previous study reported high diversity at the RCR3 gene among different (wild) tomato species and suggested that the elevated amino acid variation at this locus might translate into functional diversity upon pathogen challenge [24].
Here we describe the natural variation occurring at the RCR3 locus in several wild tomato species with particular focus on a set of individuals originating from a population of the species S. peruvianum. Previous studies of other resistance genes in this species indicate that pathogen pressure is a significant evolutionary force, at least in some parts of the species range [33]–[35]. Moreover, high levels of polymorphism in this species provide sufficient power for population genetic analyses. In fact, we show that nucleotide and amino acid diversity at the RCR3 locus present in the Tarapaca population of S. peruvianum reflect the total diversity observed in interspecific comparisons across the whole tomato clade [24].
Combining a population genetic with a four-pronged functional approach, we show that the evolutionary history of the RCR3 locus is characterized by balancing selection, recent gene duplication and frequent gene conversion in S. peruvianum. The RCR3 gene forms a young gene family in this species and a closely related sister species. Two differentiated sequence types are maintained within and across RCR3 loci. In contrast to other studies that find variation in pathogen recognition segregating at resistance loci [33], [39], we find evidence for variation in the activation of the defense response. Our results suggest that coevolution between the guardee and its guard rather than with the pathogen effector is the major force in the evolution of the RCR3 locus.
To investigate the evolutionary history of the RCR3 locus, we cloned and sequenced RCR3 alleles from 28 individuals of multiple wild tomato species (S. chilense, S. chmielewskii, S. corneliomulleri, S. habrochaites, S. lycopersicoides, S. pennellii, S. peruvianum and S. pimpinellifolium) and the cultivated tomato S. lycopersicum (Table S2). This approach revealed that RCR3 forms a gene family with at least two paralogs in S. peruvianum and its sister species S. corneliomulleri. These paralogs are more closely related to RCR3 than to other cysteine proteases, including PIP1, which cluster in the same genomic region [40]. The duplication of the RCR3 locus appears to be restricted to S. peruvianum and S. corneliomulleri, since no evidence for a duplication event was found in the draft genome of the cultivated tomato or in the other tomato species investigated in this study. However, we cannot exclude the existence of more diverged paralogs in the other tomato species studied, which may not have been detected through our sequencing approach. The RCR3 paralogs detected in this study could not be unambiguously distinguished from one another based on sequence divergence in the RCR3 open reading frame (ORF). Therefore, we cloned and sequenced the flanking regions (FLRs) of 43 alleles from one S. peruvianum population (LA2744 from Tarapaca, Chile, described also in [34], [35]) and defined their genomic origin relative to the cultivated tomato through BLAST and phylogenetic analyses. These analyses showed consistent results: FLRs that corresponded to the orthologous RCR3 containing region of the cultivated tomato based on significant BLAST hits clustered together in the phylogenetic tree. FLRs that mapped to other genomic locations in S. lycopersicum formed distinct clusters (Figure S1). The analyses of the RCR3 FLRs revealed that the RCR3 gene was duplicated at least twice in S. peruvianum – the duplicates are named Locus A, Locus B and Locus C hereafter. All 5′ flanking regions matched the RCR3 locus from S. lycopersicum over the full sequenced length reaching 400 to 900 bp upstream of the gene. This indicates that the duplicated region extends further upstream of the RCR3 gene. In contrast, based on BLAST hits, only a portion of the 3′FLRs matched the RCR3 locus from S. lycopersicum. At approximately 580 bp downstream of the stop codon, Locus B diverges from both Locus A and the S. lycopersicum sequence (Figure 1C). This marks the likely insertion point of the duplicated RCR3 segment into a novel genomic location at the time of origin of this new duplicate. BLAST hits for the 3′FLR of Locus B alleles beyond this breakpoint mapped to a genomic region located approximately 8.2 kb downstream of the RCR3 locus in the tomato genome. Locus C is characterized by a large deletion in the 3′FLR relative to the S. lycopersicum sequence and it was not possible to map it using the draft tomato genome. The phylogenetic and BLAST analyses of the flanking sequences indicated that alleles from Locus A have the highest sequence similarity to RCR3 from other Solanum species (S. lycopersicum and S. pimpinellifolium) and sequence divergence lies within the range of the overall sequence divergence observed between S. lycopersicum and S. peruvianum (which ranges from 0.0039 to 0.0589 across 50 loci, [41]). It is therefore likely that alleles from Locus A are orthologous to the RCR3 gene in the other species, in which the RCR3 gene is not duplicated. This implies that Locus B and Locus C are more recently derived duplicates of Locus A in S. peruvianum.
Our approach allowed us to unambiguously match 27 RCR3 sequences with their corresponding 3′FLR and therefore assign 27 of 43 RCR3 sequences to the different loci: 14 alleles to Locus A, nine alleles to Locus B and four to Locus C (Table S2). The copy number of the gene varies between individuals of S. peruvianum and no individual seemed to carry all three RCR3 copies. However, all but two tested individuals carried alleles that were assigned to two different RCR3 loci and, in most cases, at least one allele originated from Locus A (Figure S1). For population genetic analyses, only alleles that could be unambiguously assigned to their corresponding locus were used. Due to small sample size (n = 4), alleles originating from Locus C were excluded from the analysis. The genomic origin of each assigned allele is indicated by the respective letter (A, B or C) in the nomenclature used in this study.
Gene duplication and subsequent (functional) divergence of duplicates are typical mechanisms generating diversity at genes involved in host-pathogen coevolution [42]–[44]. However, young duplicates that have not had time to diverge from one another can be homogenized by frequent intergenic gene conversion [42]. The high sequence similarity between the RCR3 ORFs and the presence of copy number variants within populations are consistent with the recent origin of the RCR3 gene family in S. peruvianum. We therefore developed an Approximate Bayesian Computation (ABC) method [32], [45] to evaluate whether gene conversion occurs and, if so, at what rate [46]. A model of evolution with gene conversion was largely favored over a model without gene conversion (Bayes Factor>1,000). The population gene conversion rate C between the RCR3 ORFs was consistently estimated to be significantly greater than zero (C = 1.08, credibility interval CI = [0.19–7.7], Figure S2, Table S3) and more than 100 times larger than the population mutation rate estimated at 14 reference loci in S. peruvianum (0.014; Table S4) or at the RCR3 gene (0.0085; Table S4).
A survey of the site frequency spectrum (SFS) of shared and private polymorphisms [47] also confirms this high rate of gene conversion (Text S1, Figure S3). We therefore suggest that functional divergence between the two copies on the protein level is unlikely at this stage of evolution because adaptive mutations appearing at one locus can be transferred to the other locus by gene conversion [46]. In contrast, signatures of gene conversion could not be detected at the 3′FLRs based on ABC analysis and the shape of the SFS (fewer shared polymorphisms, excess of fixed differences between loci, Text S2, Figure S3). This suggests that gene conversion does not happen as frequently in the 3′FLRs of the RCR3 gene as compared to the RCR3 ORFs.
We analyzed sequence variation within the population to evaluate which selective forces act on the RCR3 gene. Phylogenetic analyses of the coding sequence of all assigned RCR3 alleles revealed two differentiated sequence types (Figure 2 and Figure S4), which segregate within all three loci. The haplotypic structure of the sequence types is mainly due to two different intronic sequence types and variation in linkage disequilibrium with this intron. The two sequence types are highly differentiated from one another: The index of fixation at RCR3 (FST = 0.311) is higher than the average FST between populations of S. peruvianum at eight reference genes (FST = 0.198, minimum 0.081, maximum 0.352, [48]). However, polymorphism within each sequence type is low (πsequence type 1 = 0.007, πsequence type 2 = 0.005) consistent with the maintenance of the two sequence types via long-term balancing selection.
To evaluate whether natural selection contributed to the maintenance of the distinct sequence types at the RCR3 locus, several population genetic statistics were calculated for the alleles of RCR3 Locus A and Locus B. Putative pseudogenes (see below and Text S3) were excluded from these analyses. To rule out demographic effects, which could interfere with the detection of the signature of natural selection acting at the RCR3 locus, all statistics were compared to a set of 14 reference loci that had previously been sequenced in the same individuals of S. peruvianum [49], [50].
We computed Tajima's D (DT), which summarizes the SFS of mutations in a given dataset [51]. Positive DT values indicate an excess of polymorphism at intermediate frequency, a pattern indicative of balancing selection. A sliding window analysis depicting DT across the entire RCR3 ORFs revealed regions with highly positive values. To test whether DT at the RCR3 ORF would globally deviate from neutrality, we derived the expected distributions of DT for the studied population under neutrality, taking demography and the respective gene conversion rate into account (Texts S1 and S2). The observed values at the RCR3 ORFs do not deviate significantly from neutrality (Figure S5, Table S4). However, the 3′FLRs of both loci exhibit significantly positive DT values compared to the expected neutral distribution for this population (Figure 1B, Text S2, Figure S6). Taken together, our findings suggest the following evolutionary scenario for the RCR3 loci in S. peruvianum (Figure 1A). Since both sequence types segregate at each locus and the FLRs show positive Tajima's D values, the two sequence types most likely pre-date the formation of the gene family and have been maintained by balancing selection. At the initial time of duplication, only a single sequence type would have been transferred to the new genomic region (8.2 kb downstream of Locus A in the S. lycopersicum genome), for example sequence type 1 from Locus A to Locus B. Then, following the origin of Locus B, the second sequence type (e.g. type 2) was also introduced at Locus B by recombination events such as gene conversion. High levels of recombination within the coding region (perhaps via gene conversion as described above) have subsequently intermixed the two sequence types and likely obscured the signature of balancing selection in the coding region by whittling down the region targeted by natural selection. In contrast, the signature of balancing selection is apparent in the 3′FLRs, where gene conversion does not occur as frequently.
The presence of two distinct sequence types differentiated especially in their intron sequences suggests three potential targets of selection: 1) selection on different regulatory motifs in the intron, 2) selection for different splicing variants or 3) selection on one or more amino acid polymorphism(s) in linkage with the intron. In silico analysis did not reveal different regulatory motifs between the two intronic sequence types, although we cannot rule out the possibility that novel regulatory motifs have been overlooked. Nucleotide sequencing of mRNA from the two sequence types did not indicate the existence of different splicing variants at the RCR3 locus. Therefore, we reason that balancing selection is most likely acting on amino acid polymorphism(s) linked to the intron.
To evaluate functional differences between sequence types at the protein level, we took a four-pronged approach. Using an over-expression vector in planta, we first evaluated whether protein accumulated for all sequence types in apoplastic fluids (AFs) of Nicotiana benthamiana. In total, 54 different allelic protein variants were chosen for these assays as follows. Eleven of these protein variants were chosen from the set of 27 alleles of S. peruvianum that could be assigned to Locus A, B or C. These eleven variants represented the protein diversity found in this set of S. peruvianum alleles. These alleles originated from all three loci and included both sequence types. The remaining 43 variants were chosen from the set of S. peruvianum alleles that could not be assigned to their corresponding locus and from closely related tomato species to maximize the amount of amino acid variation assayed. Of the total number of tested alleles, 47 were detected in AFs by Western blotting (Figure S7, Table S5). The remaining seven RCR3 proteins did not accumulate in independent expression assays, although the accumulation of mRNA was confirmed by RT-PCR (class I alleles in Figure 2, Figure S8). One of these alleles originated from S. corneliomulleri and six alleles originated from S. peruvianum (one from Locus A, one from Locus B, and four could not be assigned to their corresponding locus). In all cases in which no protein accumulated, the causative mutations (frame shifts leading to premature stop codons in five of these alleles and point mutations in the remaining two) could be identified (Text S3, Figure S9, Table S5). Since these seven alleles appear to be pseudogenes, they were excluded from population genetic analyses described above.
The second assay was a protease enzymatic assay. The activity of the RCR3 proteins in AFs was detected by Activity-based Protein Profiling (ABPP) using fluorescent DCG-04. DCG-04 is an inhibitor of papain-like cysteine proteases and reacts irreversibly and covalently to the active site cysteine of proteases in an activity-dependent manner [52]. This assay has been applied frequently to detect the activity of plant proteases and their inhibition by pathogenic protease inhibitors [24], [25], [53]–[55]. All 47 expressed RCR3 proteins could be labeled by DCG-04 to similar levels, confirming that they all are active proteases (Figure S10, Table S5).
Our third and fourth functional assays were designed to detect differences among alleles in their sensitivity to AVR2 and in their ability to elicit HR upon co-infiltration with and without AVR2 into rcr3-mutant tomato plants (cv. Money Maker Cf-2/rcr3-3, [56]). Inhibition assays based on competitive ABPP were performed to determine which RCR3 can be inhibited by the fungal protease inhibitor AVR2. Of the 47 tested RCR3 proteins, 41 (including all tested alleles from Locus A, B and C) were inhibited by AVR2 resulting in the activation of the Cf-2 dependent defense response in planta (alleles in classes III and IV in Figure 2, Figure S11, Table S5). The six alleles that failed to be inhibited by AVR2 were isolated from individuals of S. peruvianum and S. chilense (alleles in class II in Figure 2). A single nonsynonymous substitution at position 692, resulting in a change from asparagine (N) to aspartic acid (D) at position 194 in the protein (N194D), is significantly associated with this phenotypic difference (at 1% after Bonferroni correction; R2 = 0.842, P = 1.05×10−26, Figure S12). This supports previous results using site directed mutagenesis by Shabab et al. (2008) [24], which found that RCR3 alleles carrying the N194D substitution are insensitive to inhibition by AVR2. Additionally we show here that alleles that carry the N194D mutation fail to activate the defense response in planta, in the presence of AVR2 (Figures S11 and S12). Due to the large sample size used in this study (54 alleles), we had power to detect epistatic interactions between amino acid variants, such as the substitution R151Q in an allele carrying the N194D mutation (peru1954_1). This allele with the Q variant at site 151 was inhibited by AVR2, contrary to other alleles with the D variant at site 194, implicating potential epistatic interactions between these two polymorphisms (Figures S9 and S13, Table S5). Among all tested alleles that do not carry the N194D substitution only a single allele, peru7233_2, did not induce HR despite being sensitive to inhibition by AVR2 (Table S5). This allele has one amino acid difference (R138I) compared to other alleles that induced HR upon co-infiltration with AVR2 (Figure S9).
In addition to the N194D polymorphism, nucleotide polymorphisms at positions 717 (synonymous mutation) and 750 (causes amino acid difference R213S) were associated with insensitivity to inhibition by AVR2 (Figure S12, R2 = 0.254, P = 2.9×10−6; and R2 = 0.336, P = 9.8×10−8). However, since alleles that have the polymorphism at bp 750 (R213S), but not N194D, can be inhibited by AVR2 and elicit HR in planta, it is likely that the association between phenotype and sequence variation for this polymorphism is due to linkage disequilibrium. In our data set encompassing nine Solanum species, the amino acid substitution N194D was found exclusively in six individuals of S. peruvianum and S. chilense and was only represented by two alleles in the dataset used for the population genetic study. Therefore, it is unlikely that this polymorphism alone can account for the signature of balancing selection at this locus.
According to the ‘Guard-Hypothesis’ the defense response relies upon two distinct events: modification of the guardee by the effector and activation of the defense signaling through the guard molecule [17], [18]. Our approach enabled us to investigate both events. The inhibition assays did not show differential sensitivity for modification by AVR2 among most of the tested RCR3 alleles besides the alleles carrying the N194D mutation. However, our functional assay for HR activation revealed differences in the strength of activation of the defense response. Despite similar levels of sensitivity to AVR2, the tested RCR3 variants differ substantially in the strength of the defense response they elicit, with 13 protein variants (two of which originating from Locus A and two originating from Locus B) showing weaker HR (slower HR, smaller extent of cell death) compared to the others (alleles in class III in Figure 2, Figure 3, and Figure S11, Table S5). Five SNPs are correlated with phenotypic variation in the strength of the HR; one (at nucleotide position 102) remaining statistically significant at the 1% level after Bonferroni correction (Figure 3, Figures S9 and S13). All five mutations associated with this phenotype are linked with one another and with the intron despite frequent gene conversion at the locus. The most likely polymorphisms targeted by natural selection are at nucleotide positions 728 (R2 = 0.1, P = 0.015), 775 (R2 = 0.132, P = 0.0044) and 1099 (R2 = 0.146, P = 0.0026) since all three polymorphisms result in amino acid changes (I206K, Q222E and S330A). One of these polymorphisms (Q222E) is identical to an amino acid substitution that has previously been suggested to be involved in auto-necrosis due to incompatibility between RCR3 and Cf-2 ([56], Figure S13). In a putative structural model of the RCR3 protease domain the remaining two amino acid polymorphisms are located on the same protein surface as four additional positions that may be involved in the incompatibility between RCR3 and Cf-2 ([56], Figure S13). All three amino acid changes result in dissimilar amino acid substitutions and could have an impact on the protein conformation and function, while the two polymorphisms at synonymous sites (Figure 3B) may affect RCR3 transcript stability and could also be targets of selection [57]. Together with the intron, all five mutations are located in the regions of positive Tajima's D values in the sliding window analysis and likely underlie the signature of balancing selection at the RCR3 locus (Figure 3).
Previous studies on R gene evolution demonstrated the maintenance of variation in pathogen recognition via balancing selection [4], [5], [34]. Our combination of functional assays, population genetic tools, computational and statistical methods allowed us to pinpoint specific amino acid polymorphisms affecting guardee function. We find that a balanced polymorphism is present at each copy of the young guardee gene family in S. peruvianum because (or in spite) of the homogenizing force of gene conversion. Balancing selection, gene duplication and gene conversion are mechanisms known to play a major role in R gene evolution [4], [42] and appear to be important in the evolution of the guardee RCR3, at least in S. peruvianum. The signature of balancing selection persists in this species and while there is no evidence that gene duplication and gene conversion are involved in evolution of RCR3 in other species, it is possible that balancing selection could play a role in the evolution of RCR3 also in other species. However, contrary to what is reported at R genes, variation in pathogen recognition does not seem to underlie the balanced polymorphism at RCR3. Instead, our results suggest that variation in the activation of the defense response, rather than effector recognition per se, underlies the balanced polymorphism. Two alternative scenarios of the evolution of RCR3 could explain these observations:
1) The diversity detected at the RCR3 locus could be due to coevolution with allelic types of AVR2 or with other pathogen effectors not included in this study. In our assay, the phenotypic response to a single allelic type of AVR2 was similar in all but six tested RCR3 alleles. Therefore, functional differences between the different RCR3 alleles regarding interaction with other allelic variants of AVR2 are improbable. Furthermore, the polymorphisms underlying the signature of balancing selection were not associated with phenotypic variation in AVR2 sensitivity. Therefore, balancing selection for differential recognition of AVR2 alone cannot account for the maintenance of the two functional types at RCR3.
RCR3 can be targeted by other pathogen effectors, including EPIC1/2B which are secreted by the oomycete Phytophthora infestans [58]. These effectors are protease inhibitors and are thought to target cysteine proteases similar to RCR3 close to the substrate binding groove [59]. Unlike the N194D mutation, which has been shown to have indeed an effect on the interaction between RCR3 and AVR2, the polymorphisms underlying the signature of balancing selection are not located close to the putative site of interaction between RCR3 and these effectors. It is therefore unlikely that the observed signature of balancing selection is due to coevolution between RCR3 and other protease inhibitors. Note however, that the molecular and structural details of the interaction between RCR3 and these other effectors and its role in disease resistance are not yet well-understood. It will be of great interest to study the coevolution between RCR3 and its full effector repertoire once their roles in disease resistance have been resolved.
2) Alternatively, variation could be maintained at the RCR3 locus through coevolution at the interface between guardee and interacting host molecules and involve balancing selection for resistance/susceptibility at the level of defense signal activation. Balanced polymorphisms for resistance and susceptibility due to a potential cost of resistance and/or ecological/epidemiological factors have been studied theoretically [22] and empirically [4], [60] at R genes. To our knowledge, this study provides first evidence that this mechanism can also drive guardee evolution. In the case of the RCR3 gene, a potential cost of resistance could be the result of coevolution with its guard Cf-2, which also exists as a gene family in wild tomatoes (S. pimpinellifolium; [61], [62]). The interaction between RCR3 and Cf-2 requires a precise matching between allelic variants. A mismatch between allelic variants from the closely related species S. lycopersicum (RCR3esc) and S. pimpinellifolium (RCR3pim) results in an auto-necrotic response [56] and may be an example of Dobzhansky-Muller incompatibility between tomato species [63]. One of the amino acid changes differing between RCR3pim and RCR3esc and potentially contributing to the reported incompatibility between RCR3 and Cf-2 (Q222E, [56]) is associated with the attenuated HR phenotype observed in this study. The remaining amino acid changes associated with this phenotype are not identical to, but are located on the same protein surface as the potential causative mutations of the RCR3-Cf-2 incompatibility (Figure S13). The amino acid changes underlying the balanced polymorphism at the RCR3 locus and causing differences in the strength of HR therefore likely play a role in the interaction between RCR3 and Cf-2. Different combinations of RCR3-Cf-2 allelic variants might thus result in differential activation of the defense response.
For practical reasons, we tested the RCR3 proteins in standard S. lycopersicum backgrounds. Since we did not conduct our assays in S. peruvianum, we cannot exclude the possibility that RCR3 and Cf-2 function differently in this species. However, the fact that RCR3 alleles from S. peruvianum do activate resistance in the presence of Cf-2 alleles in the S. lycopersicum background as expected from previous studies in S. lycopersicum and S. pimpinellifolium suggests that this interaction is conserved throughout the tomato clade.
Furthermore, even if the RCR3-Cf-2 interaction is not conserved in S. peruvianum, RCR3 may be coevolving with other host endogenous molecules which could explain the pattern of variation observed at RCR3.
Since in our study all RCR3 alleles were tested in an identical genetic background, some alleles may not be matched with their optimal Cf-2 partner, explaining the observed attenuated response for some pairings of RCR3s with Cf-2. However, we are confident that the different observed HR phenotypes are not an artifact of using a particular Cf-2 protein, because both RCR3 types are maintained by balancing selection. In nature, while an attenuated response due to weaker interaction between guard and guardee may result in reduced resistance in the presence of the pathogen, these alleles may be selectively advantageous in the absence of the pathogen, because they carry a lower cost and/or risk for activation of auto-immunity [23]. Therefore, the optimal RCR3 allele will depend upon this delicate balance between sufficient activation in the presence of the pathogen, but limited risk for auto-activation in the absence of the pathogen, explaining why both RCR3 types segregate within a single population. The optimization of defense activation may be a very important component of guard-guardee coevolution, especially when the selection pressure by the pathogen is variable in time and space.
For population genetic and functional analyses, the ORF of the RCR3 gene was cloned and amplified from genomic DNA from eleven heterozygous individuals of S. peruvianum (accession LA2744 from Tarapaca, Chile), collected by Charles M. Rick (Table S2). Seeds from different field collected plants were grown under standard greenhouse conditions in Davis, CA. DNA was isolated using the CTAB method (Doyle and Doyle, 1987). Alleles from single individuals from eight additional species of Solanum were cloned and sequenced and their RCR3 alleles were functionally tested (Table S2). These species included: S. peruvianum (accessions LA1954 and LA0446), S. chilense (accessions LA2748, LA1930 and LA1958), S. corneliomulleri (accessions LA1274 and LA1973), S. pimpinellifolium (accession LA0400), S. lycopersicum (cv. VFNT Cherry and cv. Rio Grande), S. chmielewskii (accession LA3653), S. habrochaites (accession LA1777), S. pennellii (accessions LA0716 and LA3791). For outgroup comparisons, the RCR3 gene from S. lycopersicoides (accession LA2951) was sequenced and tested. All accessions were obtained from the Tomato Genetics Resource Center (TGRC) in Davis, CA. Plant growth conditions and DNA extraction for these additional accessions (with exception of LA1954, LA0446, LA2748, LA1930, LA1958, LA1274 and LA1973) were identical as for S. peruvianum from Tarapaca (LA2744). DNA from these other accessions was extracted using the Dneasy DNA Extraction Kit (Qiagen).
The RCR3 gene was identified using the RCR3 reference sequence from S. lycopersicum cv. ‘Mogeor’ (GenBank, accession number AF493234). Restriction sites for cloning were introduced into the primer sequences, which were designed to cover the start and stop codon (Table S1). The gene was PCR amplified using the Phusion proofreading polymerase (Finnzymes, Espoo, Finland) and cloned into Zero Blunt TOPO vectors (Invitrogen, Carlsbad, CA). Direct sequencing of PCR products and sequencing of miniprepped plasmid DNA from clones were conducted in parallel (Big Dye Terminator v 1.1, Applied Biosystems). Sequencing was performed according to the Sanger sequencing protocol using the DNA analyzer ABI 3730 (Applied Biosystems & Hitachi). Multiple clones per gene per individual were sequenced and ambiguous positions were compared to the direct sequences from the original PCR products. When necessary, independent rounds of PCRs, cloning and sequencing were conducted to resolve ambiguities. Raw sequences were edited and aligned in Sequencher 4.8 (1991–2007 Gene Codes Corporation) and alignments were refined by hand with MacClade (Version 4.0, Maddison and Maddison 2000, Sinauer Associates).
Due to low sequence divergence at the RCR3 ORF, it was not possible to distinguish allelic and paralogous sequences using the ORFs exclusively. Paralogs and orthologs may be distinguished by their flanking sequences since allelic sequences originate from the same locus in a genome and should possess the same (or very similar) flanking sequences. Paralogs, which are located at different positions in the genome, should have different flanking sequences. To distinguish between paralogs and orthologs, fragments of 400–2000 bp of RCR3 flanking DNA (with a minimum of 200 bp overlap with the gene to identify the matching allele) were amplified, cloned and sequenced from individuals from the Tarapaca population of S. peruvianum (individuals 7232–7241). A three-step Tail-PCR protocol with a set of random and nested RCR3 specific primers was used (Table S1, [64]). The location of the amplified RCR3 flanking regions in the tomato genome was assessed using BLASTn searches (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and phylogenetic reconstruction (PAUP v. 4.0b10, Swofford 1999, Sinauer Associates). FLRs were assigned stringently to the different allele sequences of the RCR3 gene, and only unambiguous pairs of alleles and FLRs were retained. Subsequently, the genomic origin of alleles with matching FLR was defined according to the BLAST search. Only RCR3 alleles matched unambiguously to a given FLR were used for population genetic analysis.
The standard summary statistics including π, divergence, Tajima's D (DT) and Fu and Li's D test statistics were calculated using DnaSP v. 5.10 [65]. Sliding window analyses were also conducted using DnaSP. Phylogenetic analyses were performed using PAUP v. 4.0b10 (Swofford 1999, Sinauer Associates). The phylogenetic relationships between the sequences were determined using maximum parsimony (MP) and neighbor-joining (NJ) and these methods yielded similar topologies. To test whether gene conversion occurred between RCR3 copies, simulations were performed assuming a recent gene duplication event with copy number variation using the coalescent simulator by Thornton (2007) [66]. We then developed an ABC method using ABCest [67] to perform the model choice procedure (between models with and without gene conversion based on the code by Beaumont et al. (2002) [68]) and estimate the gene conversion rate (Text S1 and S2). Additionally, we surveyed the SFS of private and shared polymorphisms for the duplicated loci [47].
To investigate whether demographic effects could explain the pattern of sequence variation at the RCR3 locus, our results were compared to values obtained from 14 single-copy reference loci (CT066, CT093, CT099, CT114, CT143, CT148, CT166, CT179, CT189, CT198, CT208, CT251, CT268 and sucr), previously amplified from the same individuals of S. peruvianum [31], [49], [50]. A summary of their predicted gene products is found in Table 1 of [50]. Additionally, these loci were used to simulate expected neutral distributions of DT for comparison with the observed values at the RCR3 locus (Text S2).
A total of 54 RCR3 variants, which had been cloned into TOPO Zero Blunt for sequence analyses, were selected for functional testing. Cloning procedures of these variants were conducted according to the protocol described in [24]. Each RCR3 variant to be functionally tested was excised from the Zero Blunt TOPO vector using the restriction enzymes XhoI and NcoI, for which restriction sites resided in the PCR primers. Excised fragments were cloned into the pFK26 vector carrying the 35S overexpression promoter. 35S::RCR3 cassettes were shuttled into the binary vector pTP05 using the restriction enzymes XbaI and SalI [24]. All clones were verified by sequencing and electroporated into Agrobacterium tumefaciens strain GV3101. Agroinfiltration into leaves of N. benthamiana plants was performed as described previously [24]. After protein expression, RCR3 is secreted into the intercellular space outside the cytoplasm membrane. To recover expressed RCR3, infiltrated N. benthamiana leaves were harvested 72 h post inoculation, and the apoplastic intercellular fluids (AFs) of all infiltrated leaves were isolated according to [24]. Volumes of AFs containing equal concentrations of active RCR3 were used for all further experiments. Western Blot analysis was used to confirm the expression of RCR3 using RCR3 antibodies described previously [25].
Activity-based protein profiling (ABPP) using fluorescent DCG-04 [52] was used to detect RCR3 activity in the isolated AFs. 45 µl of AF were labeled with 2 µM fluorescent Bodipy-DCG-04 at pH 5.5 in the presence of 1 mM DTT for 5 h as described previously [24]. Concentrations of active RCR3 were adjusted based on the fluorescence signal. Accuracy of these adjustments was confirmed by independent ABPPs. Inhibition studies were performed by pre-incubation with 100 nM affinity-purified AVR2 [24], followed by ABPP. Proteins were separated via sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) and fluorescently-labeled proteins were detected by in-gel fluorescence scanning using a Typhoon 8600 scanner (GE Healthcare Life Sciences, http://www.gelifesciences.com) at ex/em 580 nm.
We investigated whether different RCR3 constructs could activate the hypersensitive response upon exposure to AVR2 in cultivated tomato plants (S. lycopersicum cv. Money Maker). For this purpose, 100 µl of AF containing equal concentrations of expressed, active RCR3 with or without 100 nM AVR2 were infiltrated into Cf-2/rcr3-3 and Cf0/RCR3 tomato leaves. Tissue collapse was monitored daily until six days post inoculation (dpi) and recorded photographically.
A general linear model algorithm implemented in TASSEL v. 3.0 (http://www.maizegenetics.net/) was used to evaluate correlations between phenotypic variation and sequence polymorphism at the RCR3 locus. The genotypic data was filtered such that only mutations that occurred in frequencies greater than 25% were included. Resulting P-values were Bonferroni-corrected for multiple testing. A structural model of the RCR3 protease domain was created as previously described in [24] using papain (PDB code 9papA) as a template.
Seven RCR3 constructs failed to be expressed in N. benthamiana leaves. To confirm that the construct was designed correctly and that the agroinfiltration was successful, RNA of infiltrated leaves was isolated and RT-PCR with RCR3-specific primers was performed. The extraction of RNA was conducted using the Rneasy Plus Mini Kit (Qiagen) starting with 40–80 mg of plant material. cDNA-banks were created by reverse transcription using SuperScript Reverse Transcriptase (Invitrogen). RT-PCR was conducted for the RCR3 gene and a portion of the Ribulose-bisphosphate-carboxylase-oxigenase as a RNA-extraction control (Table S1).
All sequences have been deposited on GenBank under accession numbers JQ927229–JQ927299.
|
10.1371/journal.ppat.1005874 | Screen of Non-annotated Small Secreted Proteins of Pseudomonas syringae Reveals a Virulence Factor That Inhibits Tomato Immune Proteases | Pseudomonas syringae pv. tomato DC3000 (PtoDC3000) is an extracellular model plant pathogen, yet its potential to produce secreted effectors that manipulate the apoplast has been under investigated. Here we identified 131 candidate small, secreted, non-annotated proteins from the PtoDC3000 genome, most of which are common to Pseudomonas species and potentially expressed during apoplastic colonization. We produced 43 of these proteins through a custom-made gateway-compatible expression system for extracellular bacterial proteins, and screened them for their ability to inhibit the secreted immune protease C14 of tomato using competitive activity-based protein profiling. This screen revealed C14-inhibiting protein-1 (Cip1), which contains motifs of the chagasin-like protease inhibitors. Cip1 mutants are less virulent on tomato, demonstrating the importance of this effector in apoplastic immunity. Cip1 also inhibits immune protease Pip1, which is known to suppress PtoDC3000 infection, but has a lower affinity for its close homolog Rcr3, explaining why this protein is not recognized in tomato plants carrying the Cf-2 resistance gene, which uses Rcr3 as a co-receptor to detect pathogen-derived protease inhibitors. Thus, this approach uncovered a protease inhibitor of P. syringae, indicating that also P. syringae secretes effectors that selectively target apoplastic host proteases of tomato, similar to tomato pathogenic fungi, oomycetes and nematodes.
| The extracellular space in the leaf (the apoplast) is colonized by a diversity of microbes that will have to deal with host-secreted hydrolytic enzymes, many of which accumulate during defence responses. We hypothesize that in addition to fungal and oomycete pathogens, the bacterial model plant pathogen Pseudomonas syringae also protects itself in the apoplast by secreting inhibitors targeting these apoplastic hydrolases. The genome of P. syringe harbours over 131 genes encoding putative small, non-annotated secreted proteins that have not been characterized previously. Here, we produced and purified 43 of these small proteins and tested them for their ability to inhibit the secreted immune protease C14 of tomato. We discovered a C14 protease inhibitor, coined Cip1, which carries chagasin-like motifs and contributes to virulence. Cip1 also effectively inhibits Pip1, another immune protease of tomato, known to suppress P. syringae infection. Interestingly, Cip1 has a lower affinity for the immune protease Rcr3, explaining why this protein, and PtoDC3000 producing Cip1, is not recognized in tomato plants carrying the Cf-2 resistance gene, which uses Rcr3 as a co-receptor to detect pathogen invasion.
| Pseudomonas syringae is an important model system for plant-pathogen interactions. Different pathovars of this Gram-negative bacterium can cause disease on a broad variety of plants. Most intensively studied is pathovar tomato DC3000 (PtoDC3000), which causes bacterial speck disease on tomato and Arabidopsis [1, 2]. This pathogen can enter the extracellular space (apoplast) of leaves through stomata and colonizes the apoplast, causing black specks, hence the name bacterial speck disease [1, 2]. P. syringae manipulates its host using effectors, which are secreted metabolites or proteins that manipulate the host cell. Most intensively studied are the type-III (T3) effectors that are injected into host cells through the T3 secretion system (T3SS) [3, 4]. These T3 effectors are collectively required but individually not essential to cause disease [5].
Filamentous tomato pathogens secrete dozens of apoplastic effectors with different functions, often contributing to pathogen virulence. The fungal tomato pathogen Cladosporium fulvum, for example, secretes Avr4 to prevent degradation of chitin in the fungal cell walls by secreted host chitinases [6]. C. fulvum also secretes Ecp6 to sequester chitin fragments and prevent their detection [7], and Avr2 to inhibit secreted host proteases [8]. Likewise, the oomycete tomato pathogen Phytophthora infestans secretes Epi and EpiC proteins inhibiting secreted host serine and cysteine proteases, respectively [9, 10]. In other pathosystems, apoplastic effectors include P. sojae Gip1, which inhibits a secreted glycosidase of its host, soybean [11] and Ustilago maydis Pep1, which blocks the apoplastic peroxidase of its host, maize [12]. Hence, secreted effectors are commonly used to manipulate the host apoplast. Importantly, all of these apoplastic effector proteins are small and often share no or low homology with annotated proteins.
The production of apoplastic effectors by filamentous pathogens suggests that also bacterial pathogens may employ apoplastic effectors to inhibit harmful enzymes in the apoplast. Here, we mined the genome of the model pathogen PtoDC3000 for genes encoding potential apoplastic effectors and found that many of these putative effectors are common to Pseudomonas species and expressed during apoplast colonization. We expressed over 40 of these non-annotated putative effectors as soluble proteins and screened them using competitive activity-based protein profiling (ABPP, [13]) for the inhibition of the C14 immune protease of tomato. Our results revealed that one of these proteins can inhibit immune proteases of tomato and contributes to virulence. This study investigates a repertoire of new putative effector proteins and describes the targets of the first apoplastic effector for this important model plant pathogen.
To identify non-annotated small secreted proteins of PtoDC3000, we analyzed the 5616 predicted proteins encoded by the PtoDC3000 genome ([14], S1 Table). First, we ranked the 5616 proteins on the length of the proteins, resulting in a histogram that visualizes that the majority of the PtoDC3000 proteins are 150–400 aa in length (Fig 1A). From this list we selected 2420 proteins with a length of 50–260 amino acids, which corresponds to protein sizes of 5–25 kDa. Most of the published apoplastic effectors fall in this size region.
Second, we analyzed these 2420 proteins for the presence of a Gram-negative bacterial signal peptide using SignalP [15]. SignalP predicts signal peptides using two algorithms: Hidden Markov (HM) and Neural Network (NN). Because the HM and NN algorithms predict signal peptides independently, we plotted each of the 2420 proteins against their scores in a dot plot (Fig 1B). To ensure that we select proteins that are likely to have a functional signal peptide, we selected the proteins that have an additive score of HM+NN>1.1. A total of 234 proteins were selected this way (Fig 1B). Third, the HM and NN algorithms also produce a significance score (HM (0–2) and NN (0–5)), which we used for further selection. By selecting proteins with a sum of both significance scores being 5 or higher, we selected 200 proteins having the highest confidence for secretion (Fig 1C).
Fourth, we investigated the 200 selected proteins for their annotation in the Pseudomonas genome database. Of the 200 small putative proteins, 69 are annotated, e.g. as components of secretion or motility systems (S1 Table). This group also includes chaperones, prolyl isomerases, various transporters, and a superoxide dismutase, carbonic anhydrase, and sorbitol dehydrogenase. None of these proteins are annotated as hydrolase inhibitor. The 131 remaining small proteins have previously only been annotated as ‘hypothetical proteins’ or ‘lipoproteins’, which means that they carry a lipobox after the signal peptide (Fig 1D).
To investigate if these 131 non-annotated proteins can be genuine proteins, we counted the number of cysteine residues in each mature protein, after omitting the signal peptide. The rationale being that secreted proteins frequently have disulphide bridges to increase their stability, and secreted effectors should therefore possess an even number of cysteine residues. Our analysis revealed that 47% of the 131 mature proteins have indeed an even number of cysteines, whereas only 11% have an odd number of cysteines (Fig 1E). The other 42% do not contain cysteines in the putative mature protein domain. These data suggest that our 131 non-annotated proteins represent a genuine set of secreted proteins and that a large portion of these putative proteins is equipped with putative disulphide bridges to provide stability in the apoplast.
Next, we investigated how common these putative proteins are amongst Pseudomonas species. We selected 24 other Pseudomonas strains of which genome sequences are publicly available. This collection included three other plant pathogenic P. syringae strains: pathovars syringae B728a (PsyB728a), phaseolicola 1448A (Pph1448A) and tabaci 11528 (Pta11528), which are pathogenic on snap bean, soybean, and tobacco, respectively [14, 16–17]. The genome collection also included genomes of the rice and rapeseed epiphytes (P. fulva and P. brassicacea, respectively), twelve human epiphytes and opportunistic pathogens (P. mendocina, P. stutzeri, and P. aeruginosa) and eight soil bacteria (P. fluorescence, P. entomophila and P. putida). BLAST searches of the 131 selected putative secreted small non-annotated proteins against these 24 Pseudomonas genomes revealed that most putative proteins have clear homologs in other Pseudomonas species (Fig 2 and S1 Fig). Interestingly, these homologies cluster in five groups. Group-1 consists of six putative proteins that are unique to PtoDC3000. Group-2 consists of 25 putative proteins that have close homologs only in other plant pathogenic bacteria. Group-3 consists of 26 putative proteins common with plant pathogenic bacteria and soil bacteria. Group-4 consists of nine putative proteins shared with opportunistic human pathogens, whilst the largest group (Group-5) consists of 65 putative proteins that are common to all Pseudomonas species. The high degree of conservation amongst Pseudomonas species suggests that these 131 putative proteins are genuine, bona fide proteins and are not incidentally generated by an occasional misannotation in the PtoDC3000 database. Of these, we randomly selected 43 putative proteins having relatively high SP confidence scores and we produced and purified these putative proteins for further studies (bottom of Fig 2).
To investigate if the genes encoding the selected 43 proteins are also expressed during infection, we mined gene expression databases for P. syringae infections. Infection with PsyB728a has been investigated for bacterial gene expression during epiphytic and apoplastic colonization of bean [18]. For the 38 of the 43 putative selected small non-annotated proteins, the ortholog was identified in the PsyB728a genome (Fig 3). For relative comparison, we also extracted the expression levels of the 25 type-III effectors of PsyB728a [19] from the same gene expression database [18]. The majority (21/38) of the selected genes encoding putative small, secreted non-annotated proteins are expressed with higher transcript levels than the average transcript levels of type-III effectors (Fig 3). It is therefore likely that many of the 43 selected genes are expressed during infection.
To also investigate if these proteins accumulate in the apoplast, we performed proteomic analysis of apoplastic fluids extracted from PtoDC3000-infected plants. This approach is challenging because small proteins may not produce two or more unique peptides for robust identification. Nevertheless, we could detect eight of the 43 selected proteins (dark gray bars and asterisks in Fig 3). Interestingly, these eight correspond to the genes with the highest transcript levels, often being higher than the highest expression level of type-III effectors genes, indicating that gene expression levels of the remaining genes are probably too low to detect the gene products by proteomics. In conclusion, the transcript levels and detection by proteomics indicates that several selected proteins are expressed by PtoDC3000 during apoplast colonization, and present in the apoplast.
To produce the selected proteins heterologously, we took advantage of the commercially available pFLAG-ATS expression system in Escherichia coli (Sigma-Aldrich). This expression system secretes N-terminally FLAG-tagged proteins into the growth medium using an N-terminal OmpA signal peptide to facilitate secretion. The growth medium of E. coli cultures has relatively low protein content and is easy to collect as supernatant after centrifugation, making this expression system ideal to produce small secreted bacterial proteins. Because of the large number of proteins, we decided to use the Gateway cloning strategy and add an N-terminal His tag to simplify the purification. We therefore generated a derivative of pFLAG-ATS that carries an extra fragment encoding a His-tag and the ccdB suicide gene located between the two attR1 and attR2 recombination sites (pTSGATE1, Fig 4A and S1 File). This construct was used for the expression and purification of 43 soluble proteins in a single step to a scale of over 100 μg per protein. Detection of the purified proteins on coomassie gels and anti-FLAG western blots confirmed the purity and molecular weight of these proteins (Fig 4B and S2 Fig).
Various tomato pathogens secrete apoplastic effectors that target papain-like cysteine proteases (PLCPs) that are secreted by the host during the immune response [8, 10, 20]. We therefore hypothesized that also P. syringae might employ this strategy and secrete protease inhibitors during infection. We first tested if any of the putative effectors would inhibit the secreted immune protease C14 of tomato. C14 is targeted by both EpiC1 and EpiC2B proteins of P. infestans [21] and is inhibited by Avr2 of C. fulvum [22]. C14 is also targeted by the RxLR-type effector AvrBlb2 of P. infestans, which blocks C14 secretion [23]. Furthermore, C14 knock-down dramatically increases P. infestans susceptibility [21], whilst C14 overexpression increases resistance [23]. Because C14 is a ‘hub’ for multiple effectors and because it plays a role in immunity, we decided to test if any of the 43 purified putative apoplastic effectors of PtoDC3000 would inhibit C14.
We screened the 43 proteins for their ability to block the activity of the mature C14 protease using competitive ABPP. Competitive ABPP is based on a preincubation of a protease with a putative inhibitor, followed by labeling of the non-inhibited proteases with an activity-based probe that reacts with the active site of the enzyme. Competitive ABPP has been routinely used to uncover the targets of Avr2, Epic1, Epic2B, Gr-Vap-1, CC9, and Pit2 [8, 10, 18–20, 24–29]. To facilitate quantification and medium-throughput screening, we used MV201, a fluorescent probe based on the papain inhibitor E-64 [30].
The C14 protease was transiently expressed by agroinfiltration of Nicotiana benthamiana [24]. Leaf extracts were labeled with MV201, separated on protein gels and fluorescently labeled C14 was detected by in-gel fluorescence scanning as a strong 30 kDa signal, representing the soluble, mature isoform of C14 (mC14 [21, 24]). This signal is absent in agroinfiltrated tissues that do not express C14 and in extracts pre-incubated with an excess of E-64 (S3 Fig). C14-containing extracts were diluted such that a robust fluorescent mC14 signal could still be detected upon MV201 labeling. Preincubation of the diluted mC14 with 1μg (50 μM) EpiC1 and EpiC2B blocked subsequent labeling by MV201 (Fig 5A), showing that mC14 inhibition by EpiCs can be detected using this competitive ABPP approach. To demonstrate that we could also detect interactions with weak inhibitors, we preincubated mC14 with 1 μg (33 μM) Avr2, a weak inhibitor of mC14 [21–22]. Importantly, Avr2 also prevents labeling of mC14 under these conditions (Fig 5A), indicating that we use conditions that allow us to detect even weak inhibitors of mC14.
We next screened the 43 proteins by preincubating 1μg (20–85 μM) of each protein with the mC14-containing leaf extract, followed by MV201 labeling. To select C14 inhibitors, signals were quantified and plotted for each PSPTO protein (Fig 5B). To exclude false positives, signals that were more than 0.5 times the signal of the no-inhibitor control were considered non-significant. Interestingly, this screen revealed one PtoDC3000 protein, PSPTO4211, that blocked labeling of mC14 by MV201 (Fig 5B). Hence, we named this protein C14-inhibiting protein (Cip1).
To determine the importance of cip1 for the virulence of PtoDC3000, we generated two independent knock-out mutants of PtoDC3000 through homologous recombination (Δcip1a and Δcip1b) and generated complemented strains by transforming the Δcip1 mutants with wild-type Cip1 using Tn7 transposition [31]. When infiltrated into tomato leaves, the Δcip1 mutants grow significantly less when compared to wild-type PtoDC3000 (Fig 6A and S4 Fig). By contrast, both Δcip1 mutant strains grow indistinguishable from wild type PtoDC3000 in liquid cultures (in vitro, S5 Fig). Importantly, the in planta growth defect is complemented in the Δcip1 +Tn7:cip1 strain (Fig 6A). The differential bacterial growth of the strains correlates with the severity of the disease symptoms: Δcip1 strains cause less bacterial spot symptoms than the wild-type or the complemented strains (Fig 6B). This experiment demonstrates that cip1 encodes an important virulence factor for PtoDC3000 on tomato.
The phenotype of the Δcip1 mutant suggests that the cip1 gene is expressed during infection. To confirm cip1 expression, we performed semi quantitative RT-PCR on RNA extracted from wild-type and a Δcip1 mutant PtoDC3000 grown in minimal medium (which mimics infection conditions) and isolated from the apoplast of infected plants, two days after infection. RT-PCR with cip1-specific primers amplified a gene product that is absent in the Δcip1 mutant and if no reverse transcriptase was added (Fig 6C), demonstrating that cip1 is expressed when bacteria are grown in minimal media or during infection. By contrast to cip1 itself (PSPTO4211), the expression of cip1-flanking genes PSPTO4210 and PSPTO4212 is similar in the Δcip1 mutant when compared to wild-type bacteria (Fig 6C), indicating that the flanking genes are unaffected in the Δcip1 mutant.
Cip1 has a predicted signal peptide for secretion of 21 amino acids. To investigate if Cip1 protein can also be detected in the apoplast during infection, we raised a Cip1-specific antibody against the Cip1 protein and performed western blot analysis of apoplastic fluids isolated from tomato plants infected with wild-type and Δcip1 mutant bacteria. Unfortunately, the affinity of the Cip1 antibody is not high enough to detect Cip1 in apoplastic fluids. We therefore also tested the supernatant of a centrifuged culture of wild-type and Δcip1 mutant bacteria grown in minimal medium. Western blot analysis of these samples displayed a signal of the expected molecular weight in both the WT strain and the ΔhopQ-1 mutant control that was absent in both tested Δcip1 mutants (Fig 6D). These data demonstrate that Cip1 protein is detected in the medium of PtoDC3000 cultures and suggest that Cip1 occurs in the apoplast during infection.
We have performed several experiments to also investigate the suppression of apoplastic PLCPs during infection but failed to detect a consistent suppression using MV201 labeling on apoplastic fluids isolated from infected and non-infected plants (S6 Fig). We believe this is caused by the relatively low amount of Cip1 produced locally during infection when compared to the active PLCPs that are present in apoplastic fluids isolated from whole leaves.
Analysis of the Cip1 protein sequence using PFAM revealed that this protein contains a chagasin motif, NPTTG. Chagasins are cysteine protease inhibitors initially described for the human protozoan parasite Trypanosoma cruzi, the causal agent of Chagas disease [32]. In T. cruzi, chagasin is an intracellular protein that controls the activity of cruzipain, an endogenous cysteine protease during the development of this parasite. Similar roles in regulation of endogenous proteases have been described for chagasin-like proteins from other human protozoan parasites [33–36].
Alignment of Cip1 protein sequence with homologs from other Pseudomonas species and chagasins of three human protozoan parasites shows that the homology of Cip1 is high to the homolog in other P. syringae strains (>90% identity) and moderately high (ca. 50% identity) to homologs from other Pseudomonas species, but Cip1 has less than 26% identity with the well-characterized chagasins from protozoans (Fig 7A). Nevertheless, in addition to the conserved NPTTG motif, two additional Chagasin motifs (GxGG and RPW) are conserved amongst these proteins. Importantly, the alignment also reveals that all Pseudomonas chagasins carry a putative signal peptide for secretion, whereas the protozoan chagasins do not (Fig 7A).
To confirm that Cip1 can inhibit papain-like proteases, we performed classical protease inhibition assays on purified papain using the chromogenic substrate BAPNA. Lineweaver-Burk plot analysis revealed that Cip1 is a classical, competitive inhibitor of papain with a Ki of 3.98 nM (Fig 7B).
To confirm that Cip1 is a chagasin-like protein, we mutated the conserved NPTTG motif by deleting the two conserved threonines (ΔT) or by substituting them into two alanines (AA). These mutations were previously shown not to affect the chagasin structure but significantly reduce the affinity of chagasin for papain [37]. The wild-type and mutant Cip1 proteins are soluble proteins and were purified using the N-terminal His-tag (Fig 7C). When compared to Cip1(WT), the Cip1(AA) substitution mutant has significantly less inhibitory activity, whereas the Cip1(ΔT) deletion mutant is even less able to inhibit papain (Fig 7D). This relative activity is consistent with the mutants described for chagasin [37]. To examine if this reduced activity is also displayed on C14, we preincubated apoplastic fluids of plants transiently overexpressing C14 with (mutant) Cip1 and then added MV201 to label the non-inhibited proteases. Papain was included as a control to compare with the traditional protease activity assay. Suppression of labeling shows that the Cip1(ΔT) deletion mutant has lost most of its inhibitory activity towards both papain and C14 whereas the Cip1(AA) substitution mutant is still able to partially suppress labeling of papain and C14 (Fig 7E). These data are consistent to the traditional substrate conversion assay (Fig 7D) and the previously described chagasin mutants [37]. In conclusion, the NPTTG chagasin motif is also important for Cip1 inhibitory activity to the same extend as chagasins, consistent with Cip1 being a chagasin-like protease inhibitor.
We next tested if, in addition to mature C14, also the intermediate isoform of C14 can be inhibited by Cip1. Intermediate C14 (iC14) migrates at 35 kDa and differs from the 30 kDa mature C14 (mC14) by carrying a C-terminal granulin-like domain [21]. Since iC14 tends to precipitate, the activity of this isoform can be monitored in extracts that have not been centrifuged. Preincubation of non-centrifuged extracts containing iC14 with Cip1 also blocks labeling of iC14 (Fig 8A), indicating that Cip1 inhibits both isoforms of C14.
Other tomato papain-like proteases, such as Rcr3 and Pip1 are often targets of the same set of pathogen-derived inhibitors [20, 21, 24]. We therefore tested if Cip1 also inhibits these proteases. Both Pip1 and Rcr3 were produced by agroinfiltration in N. benthamiana and isolated from agroinfiltrated plants in apoplastic fluids. Labeling of diluted apoplastic fluids with MV201 causes the characteristic 25 and 23 kDa signals for Pip1 and Rcr3, respectively (Fig 8A). Preincubation with 7.8 μM Cip1 blocks Pip1 labeling and suppresses Rcr3 labeling, respectively (Fig 8A), indicating that Cip1 also inhibits Pip1 and Rcr3. All our assays are performed at apoplastic pH (pH 5–5.5), which is important since some inhibitors (e.g. Avr2, [8]) only inhibit at acidic and not at neutral pH. By contrast, inhibition of mC14, Pip1 and Rcr3 by Cip1 occurs at both apoplastic and neutral pH (S7 Fig).
The consistently weaker suppression of Rcr3 labeling, however, suggests that Cip1 is a weak inhibitor of Rcr3 when compared to Pip1 and C14. To further investigate the relative strength of inhibition, apoplastic fluids of plants transiently overexpressing Rcr3, C14 and Pip1 were pre-incubated with a dilution series of Cip1 and then labeled with MV201. The fluorescence intensity was measured from protein gels and plotted against the Cip1 concentration. This inhibitor dilution experiment revealed that Cip1 is a strong inhibitor of both C14 and Pip1, and a weak inhibitor of Rcr3, requiring at least 200-fold more Cip1 to reach the same suppression level for Rcr3 when compared to C14 and Pip1 (Fig 7B). We also tested an additional 34 natural Rcr3 variants [26] but we were unable to identify a natural variant of Rcr3 with increased sensitivity for Cip1 inhibition (S8 Fig).
When leaves of tomato plants carrying Rcr3 and the Cf-2 resistance gene were injected with Avr2 and Gr-Vap-1, a hypersensitive cell death was triggered by their ability to interact with Rcr3 [20, 38]. To test if Cip1 is also recognized in Money Maker Cf2 (MM-Cf2) tomato plants, which carry Rcr3 and Cf-2, we injected Cip1 at protein concentrations up to 1 μM. However, neither hypersensitive cell death nor chlorotic responses were observed, even after prolonged incubation times (Fig 7C). By contrast, 100 nM Avr2 is sufficient to trigger hypersensitive cell death in MM-Cf2 tomato plants, but not MM-Cf0 tomato plants, which lack the Cf-2 resistance gene (Fig 7C). This indicates that Cip1 is not recognized by MM-Cf2 tomato plants.
We also tested if the presence of Rcr3 had an effect on bacterial growth of PtoDC3000. We therefore performed bacterial growth assays on MM-Cf2 plants carrying the wild-type Rcr3 gene (MM-Cf-2/Rcr3 plants) or the rcr3-3 null mutant gene (MM-Cf-2/rcr3-3 plants, [39]). These bacterial growth assays showed that PtoDC3000 grows equally well on MM-Cf2/Rcr3 as well as MM-Cf2/rcr3-3 tomato plants (Fig 7D), consistent with the absence of Cip1 recognition by the Cf-2/Rcr3 system.
Here we describe a diverse set of small secreted putative proteins encoded by the P. syringae genome that could act as effectors that manipulate the apoplast. We developed and employed an efficient expression system to produce these proteins heterologously and we used competitive ABPP to discover a novel chagasin-like protein that inhibits the activity of apoplastic immune proteases of tomato and contributes to virulence of PtoDC3000.
By choosing stringent selection criteria to select putative small non-annotated secreted proteins, we are relatively confident that these proteins are genuine proteins. First, the vast majority of these proteins are also predicted from genomes of other Pseudomonas species. Second, the observed high ratio of paired versus unpaired cysteines is not a random distribution and suggests that these proteins carry disulphide bridges, which would stabilize them upon secretion. Third, most genes are probably expressed during apoplast colonization at a higher level than T3 effectors and eight proteins with the highest expression levels were detected in the apoplast by mass spectrometry. Fourth, we were able to produce a large number of these putative proteins as soluble proteins, indicating that they were properly folded also upon heterologous expression.
The ability to produce these secreted proteins opens several research avenues to elucidate their functions. Although several of these proteins can have enzymatic or structural functions, we suspect, based on the small size, that many of these proteins are enzyme inhibitors that manipulate secreted host enzymes, such as glycosidases, proteases, peroxidases and lipases. For example, Gip1 from Phytophthora sojae, Avr2 from Cladosporium fulvum, Epi1 from Phytophthora infestans and Pep1 from Ustilago maydis, are secreted, pathogen-derived inhibitors that target host endoglucanases, cysteine proteases, subtilases or peroxidases, respectively [8–9, 11–12]. Some of the apoplastic effectors might interact with the components of the host or pathogen cell wall or membranes; retrieve metabolites or ions; or sequester elicitors to prevent perception. For example, C. fulvum secretes Avr4 which binds to the fungal cell wall to protect it against chitinases, and C. fulvum Ecp6, which sequesters chitin fragments to prevent their perception by the plant [6–7]. In addition to testing for inhibitors of apoplastic hydrolases, our FLAG-His-tagged proteins can be used for pull-down assays to identify interacting molecules, and to produce pure protein for crystallographic studies to elucidate their structures.
We employed a new approach to screen putative apoplastic effector proteins for novel functions by using competitive ABPP. Competitive ABPP is a powerful method to identify inhibitors as this assay can be performed in medium through-put and without the need to purify the enzyme and/or know their substrates. Cravatt and co-workers used library-to-library competitive ABPP screens to identify selective inhibitors for dozens of mammalian serine hydrolases [40]. Here, we used competitive ABPP to identify natural inhibitors secreted by a pathogen. Competitive ABPP assays are relatively quick and robust, and the throughput will increase further using broad range probes that label multiple enzymes simultaneously. The introduction of new activity-based probes for other apoplastic enzymes can now be used to identify PtoDC3000 apoplastic proteins that inhibit subtilases, lipases, acyltransferases and glycosidases [41–42].
Our competitive ABPP screen revealed that PtoDC3000 secretes the chagasin-like Cip1 that can inhibit tomato immune proteases. Chagasins have been mostly described for protozoan parasites. Protozoan chagasins are produced without signal peptide and control the activity of endogenous papain-like cysteine proteases that play essential roles during infection [32–36]. In recent years, however, these protozoan chagasins were found to be released during later stages of the infection and inhibit extracellular proteases of the host [43]. A role for bacterial chagasins has not yet been elucidated [44]. An argument supporting the hypothesis that Pseudomonas chagasins have extracellular targets lies in the fact that chagasins so far exclusively target C1A proteases, but that, even though all Pseudomonas species carry a chagasin ortholog, most Pseudomonas genomes do not encode C1A proteases [43–45]. Like Cip1, the chagasin ortholog of other pathogenic Pseudomonas bacteria may also act by suppressing secreted host proteases. Likewise, chagasins produced by soil Pseudomonas may act in controlling extracellular proteases that might be produced by the microbiome. The role of chagasins in other Pseudomonas species remains to be investigated.
Importantly, Cip1 contributes to virulence on tomato because cip1 mutants show significantly reduced bacterial growth on this host and this phenotype can be complemented by transformation with wild type Cip1. This is consistent with a role for secreted immune proteases being harmful to PtoDC3000 because transgenic tomato lines with reduced Pip1 protein and activity levels are significantly more susceptible for PtoDC3000 infection [46]. Thus, Cip1 provides protection against secreted host proteases that would otherwise suppress pathogen growth.
We found that Cip1 inhibits Rcr3 less efficiently when compared to Pip1 and C14. Pip1 and Rcr3 are close homologs and it is therefore remarkable that Cip1 is able to distinguish between these proteases. Previously described pathogen-derived inhibitors Avr2, Gr-Vap-1, Epic1 and Epic2B, do not show such a distinct discrimination between Rcr3 and Pip1, even though they showed distinct affinities for the less related C14 [20–25]. The ability to distinguish between Pip1 and Rcr3 could be a strategy of PtoDC3000 to prevent recognition by Rcr3 while suppressing the dominant proteolytic activities of Pip1 and C14 in the apoplast during infection [25]. This strategy is different when compared to that used by Phytophthora infestans, which uses ‘stealthy’ EpiC inhibitors, which do inhibit Rcr3, but evade recognition [25]. The molecular basis for the evolution of this evasive chagasin-like effector that can distinguish between Rcr3 and Pip1 is yet another exiting topic for future studies.
The predicted protein sequences of the PtoDC3000 genome and their annotations [14] were obtained from http://www.pseudomonas-syringae.org/. SignalP3.0 (http://www.cbs.dtu.dk/services/SignalP-3.0/) was used for signal peptide prediction [15] and both the scores for Hidden Markov and Neural Network (sum >1.1), as well as their significance scores (sum >4), were used to select secreted proteins. The annotation in the protein database was used to remove annotated proteins. The number of cysteines in each of these candidate proteins was counted after the removal of predicted signal peptide.
The list of 234 candidate effectors was blasted against a database of predicted proteins from 25 different Pseudomonas strains (144,359 sequences total) using Blastp 2.2.26 and an initial e-value cut-off of .001 (correlating for most candidates to roughly 35% similarity). The blast results were parsed and a matrix was generated in which each candidate protein was assigned a vector of 25 values reflecting the percentage similarity between the candidate and the best matching protein from each of the strains. When no hit was found with e < .001, the percentage similarity was set to -1. Both the rows (candidates) and columns (strains) were hierarchically clustered using Cluster 3.0 [47] using a Euclidean distance metric and centroid linkage and the resulting trees were visualized using java TreeView (http://jtreeview.sourceforge.net).
The closest PsyB728a-homologs of the 43 putative small secreted, nonannotated proteins of PtoDC3000 proteins were identified in the PsyB728a genome. The respective expression levels of these genes during colonization of PsyB728a in the apoplast were extracted from the microarray data (dataset GSE42544 from the GEO database at NCBI), published as supplemental data of [18]. Also the expression levels of 25 type-III (T3) effectors of PsyB728a were extracted.
Nicotiana benthamiana and tomato (Solanum lycopersicum Money-Maker) were grown in a climate chamber under a 14h light (22°C) and 10h dark (18°C) cycle. N. benthamiana plants were grown until four to six week old and used for Agrobacterium-mediated transient protein expression (agroinfiltration).
To detect the selected PtoDC3000 proteins during infection, we inoculated Nicotiana benthamiana with the ΔhopQ1-1 mutant of PtoDC3000 [48], which is fully pathogenic on this host. This pathosystem causes strong infections and sufficient apoplastic proteomes for proteomic analysis. Plants were hand-infiltrated with 108 bacteria/mL and two days after inoculation the apoplastic fluids were isolated by vacuum infiltration and centrifugation. Apoplastic fluids were concentrated by methanol/chloroform precipitation, digested with trypsin and analyzed by LC-MS/MS, against the annotated PtoDC3000 proteome.
Cloning primers of candidate proteins (S2 Table) were designed in three steps. The 5’ ends of cloning fragments were defined by SignalP predictions to remove the predicted signal peptide from the candidate amino acid sequence. The 3’ ends of cloning fragments were selected from the predicted amino acid sequences to include stop-codons at the end of the sequence. From the defined positions, 20 to 22 bases of nucleotide sequences of the Pseudomonas gene were selected and fused behind Gateway adaptor sequences 5’-gggacaagtttgtacaaaaaagcaggcttgatg-3’ (forward) and 5’ggggaccactttgtacaagaaagctgggta-3’ (reverse).
To produce a Gateway-compatible pFLAG vector carrying an additional His-tag, a PCR fragment of a Gateway cassette encoding an N-terminal 6x His-tag was generated using primers 5’-atgcctcgagcaccatcaccatcaccataagcttacaagtttgtacaaaaaagctgaacg-3’ and 5’-atgctctagataccactttgtacaagaaagctgaacg-3’ using a destination vector as template, and cloned into pFLAG-ATS (Sigma-Aldrich) using XhoI and XbaI restriction enzymes, resulting in pTSGATE1 (S1 File).
Genomic DNA isolated from PtoDC3000 was used as PCR template. Pfu-Ultra-II polymerase (Stratagene) was used for PCR reactions according to the guidelines of the manufacturer. PCR fragments were cloned into pEntry201 vector (Invitrogen) by BP reactions according to the guidelines of the manufacturer. Cloned fragments were verified by sequencing. The LR reaction was performed to transfer inserts from pEntry201 into pTSGATE1. A list of names for pENTRY clones and pTSGATE clones is provided in S1 Table. Candidate expression vectors were transformed in E. coli for protein expression, either in Rossetta or BL21 strains. pTSGATE1 was tested for protein expression by cloning and expressing Avr2, EpiC1 and EpiC2B [21].
Protein expression was induced with 1 mM IPTG and proteins were purified on Ni-NTA affinity resin (Qiagen) according to the instructions of the manufacturer. Protein levels and purity was verified by protein gel electrophoresis followed by coomassie staining or western blotting using anti-FLAG antibody (Sigma) and an HRP-conjugated secondary anti-mouse antibody (Pierce). Signals were generated by chemiluminescence using the ECL Super Signal West (Thermo) and visualized on X-ray films (Kodak).
Purified proteins were dialysed with protein storage buffer (50 mM NaCl, 10 mM NaH2PO4 (pH 8) and 20% glycerol) and further concentrated using Vivaspin spin columns (3 kDa MW cut-off, Sartorius). Protein quantity was measured using RCDC protein assay (Bio-Rad). Proteins were stored in aliquots at -80°C until used for the inhibition screen.
Cip1 was recloned in pFLAG-ATS using primers 5- atgcaagcttcatcaccatcaccatcacgactacgacattcctactacggaaaacctatacttccagggccaaacgcccaagaacatcgtttcg-3’ and 5’-gcatgaattctcagttcaccgtgattgcgcactcgaaggtctg-3’ using HindIII and EcoRI restriction sites, resulting in pSK8 encoding WT Cip1 with a N-terminal, TEV-cleavable FLAG-His purification tag. Mutant Cip1 protein were subsequently generated by site-directed mutagenesis of pSK8 using primers 5’-agcaacccgggctttcgctggctgacccag-3’ and 5’-cgaaagcccgggttgctgggcagcgtgagg-3’ for the deletion mutant (ΔT) and 5’-acccggctgccggctttcgctggctgaccc-3’ and 5’-aagccgggcagccgggttgctgggcagcgtg-3’ for the substitution mutant (AA). (Mutant) Cip1 proteins were produced and purified as described above.
An antibody was raised against the Cip1 protein in rabbit by Eurogentec. The secondary anti-rabbit antibody was from sheep and conjugated to horse radish peroxidise (HRP). An overnight-grown culture of PtoDC3000 grown in Minimal Medium was centrifuged and the supernatant was used for western analysis.
Agrobacterium-mediated transient expression of C14 (pTP41), Rcr3 (pTP36) and Pip1 (pTP43), was carried out as described previously [24]. Agrobacterium was grown overnight in Luria-Bertani (LB) medium containing 25 μg/mL Rifampicin and 10 μg/mL Kanamycin. The bacterial cultures were centrifuged at 4000g for 10 min and the obtained bacterial pellet was re-suspended in 10 mM MgCl2, 10 mM MES (pH 5) and 1 mM acetosyringone and diluted to OD600 = 2. The same procedure was applied to Agrobacterium carrying the p19 silencing inhibitor on a binary vector [49]. Agrobacterium cultures carrying protease genes were mixed with cultures carrying p19 to a 1:1 ratio. After 1-2h incubation in the dark, the cultures were injected into leaves of four-to-six-week-old N. benthamiana plants using a needleless syringe. The plant material was harvested at three days after injection. In case of C14 expressing leaves, leaves were frozen in liquid nitrogen, ground to frozen leaf powder and stored at -80°C. Protein was extracted from frozen leaf material in 1 mM DTT and used for ABPP assays. In case of Rcr3 and Pip1, apoplastic fluid was isolated as described previously [24]. Ice cold water was vacuum-infiltrated into the detached leaves transiently expressing Rcr3 or Pip1. Surface-dried leaves were placed into apoplastic fluid collection tubes and centrifuged for 10 min at 2000g. Collected apoplastic fluids were transferred to microtubes, flash-frozen in liquid nitrogen and stored at -80°C. Protein concentrations were measured using the RCDC protein assay (Bio-Rad). The level of active protease was evaluated by ABPP on a dilution series.
ABPP of cysteine proteases was performed as described in [30] with small modifications. A single 30 μL standard reaction contained 100 mM NaAc (pH 6.2), 2 mM DTT, 1 μM MV201 and total extracts or apoplastic fluids from agroinfiltrated plants over expressing various proteases. For each single inhibition assay, approximately 1 μg protein purified from E. coli was incubated with the plant proteome for 30 min before adding MV201. Controls contained 50 μM E-64 (positive control) or the same volume of protein storage buffer (negative control). After adding the probe, the reaction mixture was incubated in the dark for 1h at room temperature. The reaction was terminated by adding SDS-containing sample buffer and either immediately heated at 95°C or stored at -20°C until heat denaturation. The reaction mixtures were loaded onto large 50-well, 12% SDS polyacrylamide gels and separated by electrophoresis. Fluorescent signals were detected using the Typhoon FLA9000 scanner. Signal intensities were quantified using ImageJ (http://imagej.nih.gov).
Genes with the following accession codes were aligned: PtoDC3000, gi28871353 (Pseudomonas syringae pv. tomato DC3000); Pph1448A, YP_276078.1 (Pseudomonas syringae pv. phaseolicola 1448A); Pta11528, gi331010052 (Pseudomonas syringae pv. tabaci 11528); PsyB728a, gi66047172 (Pseudomonas syringae pv. syringae B728a); PpW619, gi170720265 (Pseudomonas putida W619); Pf0-1, gi77460801 (Pseudomonas fluorescens Pf0-1); PmNK-01, gi330502174 (Pseudomonas mendocina NK-01); PaPAO1, gi553899616 (Pseudomonas aeruginosa PAO1-VE13); LmICP, gi28625248 (Leishmania mexicana); Chagasin, gi14250894 (Trypanosoma cruzi); EhICP2, gi122082030 (Entamoeba histolytica); and EhICP1, gi68056711 (Entamoeba histolytica).
Apoplastic fluids were isolated at 2 days upon infiltration of leaves of 4-week-old N. benthamiana with PtoDC3000(WT/Δcip1) at OD = 0.0002. The RNA was isolated from the bacterial pellet and bacteria grown in minimal medium containing mannitol and glutamate using the RNeasy mini kit (QIAGEN) according to the manufacturers protocol with an in solution DNase digest (QIAGEN) and cDNA was generated with random hexamer primers (Invitrogen) in the absence or presence of SuperscriptII reverse transcriptase (RT) following the manufacturers protocol (Thermo Fisher). PCR was performed with the primers below using Phusion polymerase (NEB) according to the manufacturer’s protocol using the program 3’ 98°C; 32 cycles of 10 sec 98°C; 20 sec 66°C; 10 sec 72°C; then 5’ 72°C. PCR products were separated on a 1.5% agarose gel, stained with ethidium bromide and detected under UV. The used primers are for PSPTO4033(recA): 5’-cggcaagggtatctacctca-3’ and 5’-ctttgcagatttccgggta-3’; PSPTO4210(Lon): 5’-gcctggacctctccaaagtc-3’ and 5’-cacttccatccggtccaaca-3’; PSPTO4211(cip1): 5’-atgccccctgttcgttttct-3’ and 5’-gaccatctccttgctctcgg-3’; and PSPTO4212(methyltransferase): 5’-agcgatctggaaattgccca-3’ and 5’-cgttggcggtgttcttcaag-3’.
Two different types of PtoDC3000 Δcip1 mutants were made. UNL231 (Δcip1a) contains a polar mutation in cip1. It was made by PCR amplifying 2.0 kb upstream and downstream of cip1 using primer set 5’-agtcggtacccgtgcgcatccgcacctggctc-3’ (which contains a KpnI site) and 5’-agtcctcgagggcaagttccggttttgcgagacg-3’ (which contains a XhoI site) and primer set 5’-agtcggatccctgtttgcgcgcggcttgtccg-3’ (which contains a BamHI site) and 5’- agtctctagagaggtgtcgctgttcatcgatgc-3’ (which contains a XbaI site), respectively. These PCR products were cloned using the indicated restriction enzyme sites in the same orientation on either side of a Spr/Smr omega fragment from pHP45Ω [50] resulting in pLN3217. The insert from this construct was cloned into the suicide vector pRK415 [51] using KpnI and XbaI restriction enzymes resulting in pLN3272. pLN3272 was transformed into PtoDC3000 by electroporation. DC3000(pLN3272) transformants were grown in liquid KB medium containing spectinomycin for five consecutive days before the culture was plated onto KB plates containing spectinomycin. Homologous recombination was selected for by screening for the spectinomycin marker linked to the mutation and loss of the tetracycline marker carried on pRK415. The PtoDC3000 cip1 polar mutant (UNL231) was confirmed by PCR to show that cip1 gene was replaced by the omega fragment.
The second cip1 mutant UNL232 (Δcip1b), which contains a non-polar mutation in cip1 was generated using a similar strategy. The only difference was that the 2.0 kb fragments upstream and downstream of cip1 were cloned on either side in the same orientation of an nptII gene in pCPP2988 [52] resulting in pLN3218. This nptII gene confers resistance to kanamycin but lacks transcriptional terminators. The insert containing the cip1 mutation was cloned into pRK415 resulting in pLN3273 and this construct was electroporated into PtoDC3000. Homologous recombination of the non-polar cip1 mutation was done as described above except selection was for the kanamycin marker linked to the non-polar cip1 mutation. UNL232 was confirmed with PCR.
The wild type cip1 gene under its native expression was reintroduced into UNL231 and UNL232 using a Tn7 transposon strategy described by [31] with some modifications. Briefly, DNA regions containing the cip1 promoter and coding region were amplified with primer set 5’-caccgaattcctgccggattacctcaaaga-3’ and 5’-cataagctttcaagcgtaatctggaacatcgtatgggtagttcaccgtgattgcgcact-3’. The reverse primer contains nucleotides that encode a hemagglutinin (HA) tag. The resulting PCR fragment was cloned into pENTR/D-TOPO (Invitrogen, Carlsbad, CA) and then recombined into pUC18::Tn7-GATEWAY destination vector using LR Clonase according to the manufacturer’s instructions resulting in construct pLN6048. Construct pLN6048 was integrated into UNL231 and UNL232 by electroporation using rifampin (100 μg/ml) and gentamicin (1 μg/ml), the latter selected for strains that contained the Tn7 cassette. Primer set 5’-attagcttacgacgctacaccc-3’ and 5’-ttgaaaagagcctgccgagca-3’ was used to identify strains that contained Tn7::cip1-HA.
Inoculation and bacterial growth assays using PtoDC3000 were performed as previously described [53]. Briefly, for cip1 complementation experiments 4-week-old tomato (S. lycopersicum cv. Moneymaker) plants were blunt syringe-inoculated with 105 bacteria /mL and leaf disks were taken from surface-sterilized leaves at 4 h post inoculation (0 dpi), and at 3 days post inoculation. Three 1 cm2 leaf disks were combined and ground in 10 mM MgCl2 and a 1-fold dilution series was plated out on selection media. For other infection assays 4-week old tomato plants were spray-inoculated with 108 bacteria /mL as described previously [54]. Leaf disks were taken from surface-sterilized leaves at 4 h post inoculation (0 dpi), and at various days post inoculation. This was repeated four times per genotype per time point per assay.
Leaves of N. benthamiana plants were untreated or infiltrated with water (Mock, M), or with Agrobacterium tumefaciens (OD = 0.5) carrying the P19 silencing inhibitor alone (P19), or mixed with Agrobacterium carrying C14 (C14). Two days later (2dpi), PtoDC3000(ΔhopQ1-1) was infiltrated at OD = 0.001, or water was used as mock control. Apoplastic fluids were isolated two days later (4dpi) and preincubated with and without 100 μM E-64 for 30 minutes and then labeled with 2 μM MV201 for 5 hours. Proteins were separated on 14% SDS-PAGE and the gel was scanned for fluorescence (532nm excitation, 580BP filter, 600PMT) and stained by Sypro Ruby.
Wild-type and both Δcip1 mutants of PtoDC3000 were inoculated at OD = 0.05 in LB medium without antibiotics at 28°C and bacterial growth was measured every 30 minutes for 12 hours at OD600.
The inhibitory activity of Cip1 against papain (Sigma-Aldrich) was determined by assaying the proteolytic activity of 30 μl of 1 mg/ml papain in Tris-HCl buffer, pH 6.8 in the presence of 1 mM glutathione, using 1.5 mM Nα-Benzoyl-L-arginine 4-nitroanilide hydrochloride (BAPNA, Sigma-Aldrich) as the substrate in the presence or absence of Cip1. The kinetic parameters for substrate hydrolysis were determined by measuring the initial rate of enzymatic activity. The inhibition constant Ki was determined with the Lineweaver-Burk equation. The Ki value was also calculated from the double reciprocal equation by fitting the data into the computer software Origin 6.1. For the Lineweaver-Burk analysis, 1 μM papain was incubated with and without 3.98 nM and 7.46 nM inhibitor and assayed at increasing concentration of BAPNA (0.1–5 mM) at 37°C for 30 min. The reciprocals of substrate hydrolysis (1/V) for inhibitor concentration were plotted against the reciprocal of the substrate concentration, and the Ki was determined by fitting the resulting data. For comparison of Cip1 with mutant Cip1, 4.1 μM papain was incubated with 0.56 μM (mutant) Cip1 inhibitor or chicken cystatin and 0.2 mM BAPNA in a 100 μl volume. Substrate conversion was monitored by increased fluorescence at 410 nm over time using an Infinite M200 Tecan microtiter plate reader.
5 μg/ml papain (Sigma-Aldrich) or 10-fold diluted apoplastic fluids from N. benthamiana leaves overexpressing C14 (see above), were preincubated in a buffer containing 25 mM NaAc pH 5.0 and 2 mM DTT with 100μM E-64 or (mutant) Cip1 for 30 minutes, and then labeled with 1 μM MV201 for 1 hour. The labeling reaction was stopped by adding SDS loading buffer and boiling for 5 minutes.
Rcr3 (Solyc02g076980), Pip1 (Solyc02g077040), C14 (Solyc12g088670), and Cip1 (PSPTO4211)
|
10.1371/journal.pgen.1003458 | Dialects of the DNA Uptake Sequence in Neisseriaceae | In all sexual organisms, adaptations exist that secure the safe reassortment of homologous alleles and prevent the intrusion of potentially hazardous alien DNA. Some bacteria engage in a simple form of sex known as transformation. In the human pathogen Neisseria meningitidis and in related bacterial species, transformation by exogenous DNA is regulated by the presence of a specific DNA Uptake Sequence (DUS), which is present in thousands of copies in the respective genomes. DUS affects transformation by limiting DNA uptake and recombination in favour of homologous DNA. The specific mechanisms of DUS–dependent genetic transformation have remained elusive. Bioinformatic analyses of family Neisseriaceae genomes reveal eight distinct variants of DUS. These variants are here termed DUS dialects, and their effect on interspecies commutation is demonstrated. Each of the DUS dialects is remarkably conserved within each species and is distributed consistent with a robust Neisseriaceae phylogeny based on core genome sequences. The impact of individual single nucleotide transversions in DUS on meningococcal transformation and on DNA binding and uptake is analysed. The results show that a DUS core 5′-CTG-3′ is required for transformation and that transversions in this core reduce DNA uptake more than two orders of magnitude although the level of DNA binding remains less affected. Distinct DUS dialects are efficient barriers to interspecies recombination in N. meningitidis, N. elongata, Kingella denitrificans, and Eikenella corrodens, despite the presence of the core sequence. The degree of similarity between the DUS dialect of the recipient species and the donor DNA directly correlates with the level of transformation and DNA binding and uptake. Finally, DUS–dependent transformation is documented in the genera Eikenella and Kingella for the first time. The results presented here advance our understanding of the function and evolution of DUS and genetic transformation in bacteria, and define the phylogenetic relationships within the Neisseriaceae family.
| Through computational and biological methods, this work analyzes the function and evolution of short DNA sequences called DNA uptake sequences (DUS) that regulate genetic transformation of bacteria in the family Neisseriaceae. Previous studies show that DUS affects transformation favourably. Here, for the first time, we document the existence of eight distinct DUS dialects that display differences in their respective nucleotide sequence that limits genetic “communication” between species. This suggests that each DUS dialect represents a barrier to horizontal gene transfer of heterologous DNA contributing to genetic isolation. Single nucleotide analysis of DUS was used to identify a three nucleotide core sequence common to all DUS dialects and essential for transformation. The discovery of multiple DUS dialects emphasizes that homologous recombination, allelic reassortment, and bacterial “sex” play an important role in the evolutionary past of Neisseriaceae species.
| Transformation in bacteria is a complex process involving uptake of naked extracellular DNA followed by homologous recombination (HR). Different reproductive barriers have evolved in diverse transformation-competent bacteria, which distinguish in favour of acquisition and recombination of homologous DNA sequences and discriminate against heterologous and potentially hazardous DNA [1]. In particular, interspecies recombination with heterologous DNA in single cellular organisms could cause gene disruptions and/or disturb sensitive cellular processes, which could in turn have adverse phenotypic consequences. Adaptations that may contribute to sexual isolation and at the same time promote genetic stability include restriction modification systems, fratricide in streptococci and cannibalism in Bacillus subtilis, quorum-sensing, biofilm formation and HR regulation and suppression [2], [3], [4], [5]. Transformation in Neisseria sp. and members of the Pasteurellaceae family is unique in the requirement for short uptake sequences in the transforming DNA, named DNA Uptake Sequences (DUS) and Uptake Signal Sequences (USS), respectively [6], [7]. The genomes of these organisms harbour thousands of DUS and USS, constituting up to 1% of their entire chromosomes [8], [9], [10]. DUS has accumulated in the core genome, i.e. the set of common genes, of N. meningitidis, N. gonorrhoeae and N. lactamica and was found to maintain its sequence identity from frequent recombination [11]. DUS was first identified in N. gonorrhoeae as a 10-mer (5′-GCCGTCTGAA-3′) and has been documented functional in transformation of meningococci and gonococci [7], [12], [13]. Later, a revised 12-mer DUS (5′-AT-GCCGTCTGAA-3′, here named AT-DUS) was shown to elevate transformation further [14], [15]. High level expression of the competence and minor pilin protein ComP has been shown to increase DUS-specific uptake, and a definite association between DUS and ComP was published recently [16], [17], [18]. A linear relationship between the number of DUS and the ability to competitively inhibit the uptake of radio-labelled DNA in N. gonorrhoeae has been documented, suggesting initial surface binding of DUS [19]. An additive effect of DUS has been documented also in transformation experiments in N. meningitidis, although no linear relationship between the number of DUS and transformation frequencies was evident [20]. Importantly, DNA binding and uptake assays do not fully correlate with the outcome of transformation assays, indicating that more than one level of DUS specificity exist [15]. Recently an influence of DUS location relative to homologous and recombinogenic regions of transforming DNA was demonstrated, suggesting that DUS may initiate DNA processing by a yet undefined way [20]. Two versions of USS have been described in Pasteurellaceae: version A (5′-AAGTGCGGT-3′), named Hin-USS, is found in Haemophilus influenzae and Actinobacillus actinomycetemcomitans (now named Aggregatibacter actinomycetemcomitans) and USS version B (5′-ACAAGCGGT-3′), named the Apl-USS subtype, is found in Actinobacillus pleuropneumoniae [21]. DUS-like repeat sequences have been described for N. subflava and N. sicca [22] and recently also in N. elongata [23]. Different variants of DUS are here termed DUS dialects alluding to their role as nucleotide ‘words’ in genetic ‘communication’ and in concordance with the previous use of the term ‘dialects’ in genetic contexts [24].
Even though DUS seems to have disseminated in the genus Neisseria, virtually nothing is known about DUS repeats in the family Neisseriaceae genera Kingella, Eikenella and Simonsiella. To fill this knowledge gap, the work presented here examines DUS specificity and dialects within the family Neisseriaceae [11], [25]. The results reveal the presence of eight DUS dialects in different branches of the robust Neisseriaceae phylogenetic tree. In transformation assays, the DUS sequence divergence negatively influences inter-species transfer of DNA. A DUS core of only three nucleotides is present in all dialects and is strictly required for transformation. Assays with radiolabelled DNA show species specific relevance of DUS dialects for both binding and uptake of DNA. This work supports the idea that DUS specificity is a highly efficient barrier to interspecies transformation, that has great impact on the evolution of the Neisseriaceae.
Neisseriaceae genomes were obtained from online databases (i.e., the Human Microbiome Project [26] and other initiatives) and searched for overrepresented/highly-repeated sequences. Several very overrepresented 10-mers were identified in different genomes that displayed high degrees of similarity to the canonical DUS sequence first described in N. gonorrhoeae [7], [12], [13]. Eight distinct and abundant variants of DUS were identified and are shown in Figure 1, five of which are potential DUS as they were not previously functionally confirmed. The DUS variants are called dialects in concordance with previous use of the term for describing variants of short DNA motifs that probably are strongly affected by some DNA template-dependent processing proteins [24], and each DUS dialect was given a name according to the nomenclature scheme described in materials and methods. Every DUS dialect was found in exceptionally high numbers in their respective genomes and the exact occurrences are presented in Table S1 together with the number of degenerate DUS in which one nucleotide position were permitted to vary. AT-DUS was found in all available N. gonorrhoeae and N. meningitidis genomes (10-mer: n≈1900). In addition, AT-DUS was found as the most overrepresented repeat in the genomes of N. lactamica (n≈2200), N. cinerea strain ATCC 14685 (n = 943) and N. polysaccharea strain ATCC 43768 (n = 2183). AG-DUS was identified in Neisseria sp. oral taxon 014 (n = 3236), N. subflava strain NJ9703 (n = 2871), N. flavescens (n = 1196 and n = 2767 in strains NRL30031 H210 and SK114, respectively), N. mucosa strain C102 (n = 2964), N. bacilliformis strain ATCC BAA-1200 (n = 4265), N. weaveri strains ATCC 51223 and LMG 5135 (n≈2850), and N. elongata subsp. glycolytica strain ATCC 29315 (n = 3273). AG-mucDUS was the most prevalent repeat in N. mucosa strain ATCC25996 (n = 1543), N. sicca (n = 3770) and N. macacae strain ATCC 33926 (n = 3729) (Table S1).
A previous study showed that in N. subflava strain ATCC 19243, a 7 kb long sequence harbouring folP (GeneBank AJ581792.1) contained 7 mucDUS and 1 AG-DUS [22], whereas the genome of the N. subflava NJ9703 strain investigated here contained mainly the AG-DUS. The folP fragment is absent in the equivalent position in N. subflava NJ9703 and elsewhere in the genome. A BLAST search of all available Neisseriaceae genomes with the 7 kb fragment showed that the mucDUS positions around folP in N. subflava ATCC 19243 were present in N. sicca.
The genome of N. wadsworthii 9715 displayed a distinct DUS dialect, wadDUS (n = 2426), which is identical to AT-DUS with a T insertion after position +3 (Figure S1A).
In the genome of K. oralis ATCC 51147, yet another new dialect was discovered, the kingDUS (n = 5918). The occurrence of nearly six thousand kingDUS in a single small genome (2.4 Mb) is the highest density of any DUS dialect detected so far. By allowing a single nucleotide divergence in the kingDUS, the number of kingDUS-similar sequences increased by 21% to a total 7153 hits for the K. oralis genome. The completion, closure and annotation of the K. oralis genome may eventually alter the absolute numbers of kingDUS present, but approximately 2,5% of this particular genome will still remain occupied by the kingDUS which is very high compared to the approximate 1% DUS occupancy in N. meningitidis and N. gonorrhoeae genomes.
The genome of S. muelleri ATCC 29453 displayed the simultaneous presence of two DUS dialects. The kingDUS (n = 2257) described above and a new dialect simDUS (n = 2292) were in the S. muelleri genome detected in nearly equal numbers with a total count of 4549. SimDUS differed from the kingDUS by an A/T transversion at position +3 (Figure 1).
The genome sequences of K. kingae ATCC 23330 and K. denitrificans ATCC 33394 also revealed a new dialect, king3DUS, which differed from the kingDUS in an A/C transversion in position +9 (Figure 1) and a G/A transition in position −1 (Figure 1). The king3DUS was present in 2787 and 3603 copies in the genomes of K. kingae ATCC 23330 and K. denitrificans ATCC 33394, respectively.
Finally, the most divergent dialect of DUS relative to the AT-DUS was identified in the genomes of E. corrodens ATCC 23834 (n = 3269) and Neisseria shayeganii 871 (n = 2245), termed eikDUS. Notably, eikDUS was the only DUS with an A in position +4 (Figure 1). All the different dialects of DUS were conserved in positions +6, +7, +8 (CTG) as well as +10 (A) as demonstrated in Figure 1.
Based on the available genome sequences of genus Neisseria, no genome was devoid of any dialect of DUS. In the family Neisseriaceae, however, five genomes were found not to contain an abundant repeat that was an obvious DUS; these were the genomes of Laribacter hongkongensis HLHK9, Lutiella nitroferrum 2002, Pseudogulbenkiania sp. NH8B, Chromobacterium sp. C-61 and Chromobacterium violaceum ATCC12472. Their respective over-represented 10-mers are listed in Table S1D. We noticed, however, that the DUS core sequence 5′-CTG-3′ was found as the reverse complement sequence 5′-CAG-3′ in the most over-represented 10-mer sequences from Laribacter hongkongensis and C. violaceum (Table S1D).
Until now, a core genome phylogenetic tree for members of the Neisseriaceae was made only for the human genus Neisseria species [27], [28], [29], for all the available N. meningitidis genomes [25] and for collections of Neisseria strains [30], [31]. Here, a phylogenetic tree encompassing 23 representative members of the family Neisseriaceae was generated based on their common core genome containing 474 coding sequences (Figure 2). The 16SrDNA phylogenetic tree [32], [33] made for the Neisseriaceae differed from the core genome based tree (Figure 2). Notably, the DUS dialect distribution in the two trees differed considerably, and the core genome tree branches reflected the presence of different dialects in a congruent manner. Also a phylogenetic tree based on ComP, a recently reported DUS-specific binding protein [18], displays high degree of congruence with different DUS dialects, although some deviations are apparent (Figure S7). The robust phylogeny finds that N. shayeganii 871 is closely related to E. corrodens ATCC 23834, and S. muelleri ATCC 29453 is located among the three different Kingella species. Neisseria sp. oral taxon 014 is in the 16SrDNA tree wrongly placed close to the cluster containing N. lactamica. N. mucosa C102 is more closely related to N. subflava and N. flavescens than to N. mucosa ATCC 25996. The latter strain is the one in which the mucDUS was first described [28] and is located on the same branch in the 16SrDNA phylogenetic tree as the type strain N. mucosa ATCC 19696, based on the available partial sequence (data not shown). This observation separated N. mucosa C102 from the N. mucosa ATCC 25996 reference strain. C. violaceum served as the outgroup in Figure 2, based on its suitable genomic distance from the other Neisseriaceae members.
The evolutionary history of DUS in Neisseriaceae may be traced and depicted as follows: The DUS-based transformation system evolved after the split from the shared common ancestor with C. violaceum. Neither could a ComP be indentified by BLAST searches of the genome of C. violaceum. Among the DUS-containing bacteria, the eikDUS-group separated first from the main branch, and is also the only group with an A in position +4 of the DUS. N. shayeganii strain 871 clusters with E. corrodens and may erroneously have been taxonomically assigned to the genus Neisseria. Thereafter, the kingDUS- and king3DUS-groups branched off from the canonical DUS-group and S. muelleri might have been in the process of separating itself from the Kingella group. N. wadsworthii separated from the AT-DUS-group by a change in DUS specificity evident from the insertion of a T in position +3 of AT-DUS. N. macacae, N. sicca and N. mucosa ATCC 25996 were separated from the AT-DUS- and AG-DUS-groups by the C/T transition in position +2.
The new DUS dialects identified here exhibit several divergent positions (Figure 1) and we became interested in studying the discrete impact of the nucleotides that constitute a functional DUS. In a transversion mutation approach, the contribution of each individual nucleotide of the well-characterized AT-DUS was tested in quantitative transformation of N. meningitidis strain MC58 and the results are shown in Figure 3A. Also the effects of single transversion mutations in all twelve AT-DUS positions on DNA binding and uptake were measured and the results are summarized in Figure 3B. Any alteration of AT-DUS significantly reduced the transformation frequency (paired t-test, seven experiments, p≤0.02), although to a variable extent. The negative control lacking DUS does not transform at all. Our previous finding [14] was confirmed in that the two semi-conserved nucleotides in positions −2 and −1 at the 5′ end of the DUS, constituting the revised 12-mer AT-DUS, positively contribute to transformation efficacy, since both their respective transversions performed less than the complete AT-DUS. Furthermore, the transversions in individual positions of the 10-mer DUS were found to impair transformation performance. When the G in position +1 was transversed, the performance in transformation was reduced to 50% relative to the performance of the complete signal. Alterations in position +2 and +3 reduced the relative performance down to 20% and 28%, respectively. Alterations of position +4 (5%) and +5 (2%) had a more than one log reduction in relative transformation performance. The C, T and G at the 3′ half of the DUS (positions +6, +7 and +8) was shown to be particularly important for the DUS effect, since transformation was nearly abolished when the nucleotides in these positions were altered. A G/C transversion in position +8 gave rise to a total of only 2 CFU in seven experiments emphasizing the near complete loss of DUS-function. This functionally important 5′-CTG-3′ core is conserved in all dialects of DUS (Figure 1). The two adenines at the 3′ end of DUS (position +9 and +10) display minor contributions to the overall effect of DUS, and mutants perform at around 50% of the full AT-DUS. The A in position +10 is also conserved in all DUS dialects but contributes less significantly to the functionality of DUS than the 5′-CTG-3′ core. The distance-dependent gradual influence of the bases around the short 5′-CTG-3′ core sequence may reflect the strength of molecular interactions between DUS and the electropositive stripe on the surface of ComP [18] that warrants further investigations.
The effect of single transversion mutations on DNA binding and uptake in Figure 3B shows that DNA binding is high in N. meningitidis strain MC58 and that only DNA uptake is significantly affected. All the individual alterations of DUS bind better than the negative control lacking DUS. Relative DNA uptake was greatly reduced for the DNA without DUS and DUS with mutations in positions +4 to +8, being approximately 2% of bound DNA for the mutations in the core positions +6 to +8 (Figure 3B). In another strain, N. meningitidis 8013, DNA binding is lower overall and is together with the uptake negatively affected by the alterations of DUS. Again it is the alterations 5′-CTG-3′ that most dramatically affects DNA uptake and/or binding in both strains tested (Figure S4).
Potential commutation, defined as the interchange of DUS-linked genetic information, between different Neisseriaceae was first investigated by employing different DUS dialects in quantitative transformation experiments of N. meningitidis strain MC58, and the results are shown in Figure 4A and Table 1. Inversely, K. denitrificans, E. corrodens and N. elongata were tested for their respective DUS dependency by using PCR products of the rpsL gene conferring streptomycin resistance flanked by their own DUS or other DUS dialects. As shown in Table 1, the transformation frequency was always highest for their autologous DUS variant. The AG-DUS differs from AT-DUS in just a single nucleotide in position -1, and has 90% efficacy in N. meningitidis MC58. AG-mucDUS differs from AT-DUS in a T/G transversion in position −1 and a C/T transition in position +2. A 50% difference in transforming abilities of DUS and the 10-mer mucDUS in N. meningitidis was previously shown, although without statistical significance [22]. The 12-mer AG-mucDUS, which occurs 165 times in the genome of N. meningitidis strain MC58, displayed here a 66% reduced transformation efficacy relative to that of AT-DUS (Figure 4A and Table 1). For the more drastic C/A transversion in the second position (+2) in AT-DUS, a DUS-like sequence that occurred only once in the entire MC58 genome, the relative transformation was reduced by about 80% (Figure 3A). The AT-mucDUS was found 19 times, but the AG-mucDUS was found 88 times in the MC58 genome indicating previous interspecies transfer from the AG-mucDUS group. These transformation assays in N. meningitidis MC58 showed inter assay variations (Figure S3) but the Kendall's W test showed a very high concordance of gained orders (Kendall's W = 0.9145, χ2 = 44.8112, df = 7, p<0.0001).
AT-DUS containing DNA was completely unable to transform E. corrodens, the most phylogenetic distant DUS-containing species with the most divergent DUS dialect of the Neisseriaceae. E. corrodens was however readily transformed with its autogenic eikDUS documenting DUS-favoured transformation in the Eikenella genus for the first time. Also K. denitrificans transformed very poorly with AT-DUS and showed significant transformation with the autogenic king3DUS demonstrating DUS-favoured transformation in the Kingella genus for the first time. Low but significant transformation was achieved with AT-DUS in N. elongata harbouring the very similar AG-DUS. No biological transformation data was generated for Neisseria mucosa and N. sicca harbouring AG-mucDUS since the two strains tested, Neisseria mucosa type strain ATCC 19696 and N. sicca ATCC 29259, were not transformable with PCR-generated DNA or isogenic genomic DNA, both conferring streptomycin resistance.
Although the transformation efficiency is a measure of the final biological outcome, it is not useful for a quantitative measure of DNA binding to the cell and the DNA uptake. To assess the latter parameters in regard to the DUS dialects, the levels of binding and uptake of radiolabelled DNA was measured in different strains and species. In N. meningitidis MC58, binding of DNA with different DUS dialects was only reduced 1.7-fold and 3.3-fold for AG-DUS and AG-mucDUS, respectively, but about 60-fold for AG-kingDUS and about 95-fold for AG-eikDUS compared to AT-DUS (Figure 4B). DNA uptake was around 60% of bound DNA for AT-DUS, AG-DUS and AG-mucDUS but only around 10% of bound DNA for AG-kingDUS and AG-eikDUS. Another N. meningitidis strain, serogroup C strain 8013, was also tested and showed reduced overall sequence specific and dialect dependent DNA binding but did not show sequence specific DNA uptake (Figure S4 and Figure S5). ComP in these two meningococcal strains are identical, indicating that more factors influence DUS-dependent DNA uptake in these strains. N. mucosa ATCC 25996, N. elongata subsp. glycolytica ATCC 29315, K. oralis ATCC 51147 and E. corrodens ATCC 23834 were also tested for DNA binding and uptake (Figure 5). N. mucosa showed a DNA binding of >2% of added DNA while all other species tested displayed values <0.3%. Only N. elongata showed a DUS dialect-dependent DNA binding with 0.3% for its own AG-DUS and 0.01% for AG-eikDUS (Figure 5B). The binding and uptake performance of individual DUS-dialects in N. elongata mirrors those of N. meningitidis strain 8013 (Figure S5). The counts for K. oralis and E. corrodens were below 100 cpm and differences in performance of the different DNA templates were accordingly small. However, it is noteworthy that DNA binding of the autogenous DUS is significantly (p≤0.05) higher than the negative control in K. oralis. Similarly, the DNA uptake of the autogenous DUS is significantly (p≤0.05) higher than the negative control in E. corrodens. DNA uptake was around 60% of bound DNA for N. elongata, K. oralis and E. corrodens but only around 0.4% for N. mucosa. The results suggest that the investigated strains of these four bacterial species may not, or only to a small extent, carry out DUS-specific uptake of DNA (Figure 5) contrasting the clear transformation data (Table 1).
Genetic transformation in bacteria is a most common event in nature that requires a complex DNA uptake machinery and well-conserved recombination proteins [2], [34], [35], [36], [37]. Transformation distinguishes itself from other modes of horizontal gene transfer (conjugation and transduction) in that the recipient cell is actively taking part in the mobilization and integration of incoming DNA. Transformation in phylogenetically distant bacteria is thus adapted to ensure efficient uptake and recombination of homologous DNA [4]. This process relies on homologous recombination proteins whose processing functions are ubiquitously conserved from single celled bacteria to complex organisms, including humans. As such, transformation may be considered a low-complexity form of sex that is not firmly linked to reproduction but may have evolved to provide a similar selective advantage in breaking up associations among alleles [38]. The bacterial families Neisseriaceae and Pasteurellaceae are most suitable model organisms for the study of transformation and its role as a possible barrier to uptake of heterologous DNA. Here, we report the presence of eight distinct DUS dialects in the Neisseriaceae and correlate their distribution to the robust phylogeny of the family. This association emphasizes the influence of autogenic recombination on evolution and divergence of lineages. DUS-dependent transformation is documented for the first time in the genera Eikenella and Kingella. In a transversion mutant analysis the differential importance of each individual nucleotide that constitutes the AT-DUS was shown in N. meningitidis and these observations were found to relate to a conserved DUS core and the potential for interspecific DUS-mediated transformation.
There are highly overrepresented sequences in the genomes of bacteria in general and the skewed occurrence of di-, tri- and tetra-mers has been particularly well documented [39]. These repeat distributions have proven valuable for classification [40]. The crossover hotspot instigator (Chi) sequence differs between bacterial species and new Chi sequences have been identified by a bioinformatics search for motifs [41]. Although uptake sequences and Chi sequences both are closely linked to homologous recombination, Chi and USS are distinctly different sequences in H. influenzae with no functional overlap [42]. As previously demonstrated [40], the DUS sequence is the most abundant repeat in the genomes of N. meningitidis and N. gonorrhoeae with about 1900 occurrences in the 2.2–2.3 Mb genomes (Table S1). In the Neisseria sp. containing the AG-DUS dialect, the counts were generally higher, around 3000. The most frequent DUS dialect was found for the kingDUS in K. oralis ATCC 51147 with nearly 6000 kingDUS within its 2.4 Mb genome sequence. Despite the high numbers of accurate DUS hits, the numbers of DUS with a single nucleotide divergence were considerably higher in all species (Table S1), revealing a potential for the activation of even more DUS positions.
The difference in total DUS count and ratio of DUS to genome size may reflect an ultimate saturation-state of DUS or indicate that this state has not yet been reached. Also, if DUS specificity was for some reason lost, the DUS could be degenerating slowly but progressively, as observed for pseudogenes. Bacterial genomes with a high number of DUS had a relatively low number of DUS with a single divergence (DUS+1mut) and vice versa. For example, Neisseria bacilliformis ATCC BAA-1200 had 4265 DUS and 4914 DUS+1mut (ratio 1∶1.15) while Neisseria cinerea ATCC 14685 had 943 DUS and 1372 DUS+1mut (ratio 1∶1.45) (Table S1). These differences could reflect differences in DUS dependency, which is known to vary in different N. gonorrhoeae strains, and may therefore also vary between species and their respective dialects [15]. Future studies will seek to address the influence of sequence variation and regulation of ComP and its antagonist PilV [17] in this regard. DUS saturation of the chromosome may also be opposed by factors such as the degree of interference with coding ability for intragenic DUS. Notably, all DUS dialects, except king3DUS, harboured a stop codon (UGA) in one reading frame, which imposes an obvious limitation on the liberty of positioning DUS.
By exploiting two different DUS dialects simultaneously some flexibility may be achieved in regard to which amino acids that are encoded by intragenic DUS. The genome sequence of S. muelleri harboured both the kingDUS and the simDUS, allowing for the variation of Q↔L and S↔C at the protein level when a DUS is found within a coding sequence. However, other explanations for the co-occurrence of two DUS-dialects in a single genome are high frequency of commutation between species (simDUS and kingDUS differ in a single nucleotide only), or consecutive habitats in different mammalian hosts with access to variant DUS dialects. DUS specificity may be altered by mutations in comP or by the acquisition of alleles encoding a novel DUS dialect, or simply by the presence of two DUS-specific proteins with different affinities, which could have originated from a simple gene duplication. However, only a single copy of comP was identified in the S. muelleri genome. It is also noteworthy that 1294 occurrences of the simDUS and kingDUS in S. muelleri are arranged as overlapping pairs in a dyad symmetry structure (Figure S6), which may indicate a dimer-based mechanism of DUS recognition. No preference for a single reading frame or positioning inside or outside of coding sequences was obvious when using the preliminarily annotated genome sequence from S. muelleri. A similar symmetry is found also in Kingella species where king3DUS pairs with king2DUS. The latter showed positive influence on transformation (data not shown) but is not a commonly found DUS by itself (Figure S6).
DUS have been found to locate in permissive regions of the core genomes of N. meningitidis, N. gonorrhoeae and N. lactamica, and intragenic DUS positions are common allowing them to be transcribed [11]. Intergenic regions, on the other hand, are particularly permissive and DUS sequences have been found to associate with transcriptional terminators by having frequently adopted an inverted paired organization, able to form stem-loop structures on ssDNA [7], [14], [43]. This inverted pair organization was found in high numbers in all genomes harbouring DUS dialects, suggestive of their association to transcriptional terminators (Table S2). In contrast to the simDUS and kingDUS arrangement in the S. muelleri genome, the individual DUS in an inverted pair DUS do not overlap.
Another interesting observation is the occurrence of peregrine DUS, exemplified by the mucDUS in Table S1A. N. meningitidis and N. gonorrhoeae genomes had a very consistent mucDUS count of about 160 and 110, respectively, while Neisseria sp. oral taxon 014 st. F0314 had the highest mucDUS count (467), and the N. weaveri strains had the lowest counts (15) in the canonical DUS group. The N. meningitidis genome for example contained 8% mucDUS in addition to the canonical DUS, while the genomes of the mucDUS-containing group of bacteria harboured between 7% and 12% DUS. These numbers likely reflect recent commutation between these two groups sharing the same ecological niche. The exchange of highly selectable markers between N. meningitidis and the commensal Neisseriae is well established [44], [45], [46], [47], [48], [49]. The amount of mucDUS relative to the canonical DUS was particularly low in the N. weaveri genomes (0.5%, Table S1A), which is interesting since N. weaveri is a canine commensal and only an opportunistic pathogen to humans [31], [50], [51].
In contrast to a previous report on the abundance of mucDUS in N. subflava strain ATCC 19243, we found that AG-DUS is the most abundant dialect in N. subflava NJ9703 [22]. Possibly, N. subflava strain ATCC 19243 acquired the folP sequence fragment from N. sicca and therefore harbours the mucDUS in this region, or alternatively, N. subflava strain ATCC 19243 is more closely related to N. sicca than to N. subflava NJ9703. We identified more canonical DUS in the N. elongata subsp. glycolytica strain ATCC 29315, 3273 as opposed to 2142 than in the former study by Higashi et al. [23], and more mucDUS, 174 as opposed to 117. These differences could possibly be due to recent updates of the genome sequence files available. Comparison of the predominantly mucDUS-containing bacteria is difficult, as Higashi et al. did not specify which strains were analysed, both were, however, reported to contain >3400 copies of the mucDUS. In contrast to this observation, we identified only 1543 mucDUS in N. mucosa strain ATCC 25996 while N. mucosa C102 had only 155 mucDUS and 2964 canonical DUS. These discrepancies indicate that strain C102 may erroneously be assigned N. mucosa, also since both the core genome phylogeny (Figure 2) and that of ComP demonstrated the close genetic relationship between this strain and N. subflava and N. flavescens.
Neisseriaceae are highly recombinogenic yielding a polyphyletic family structure, and resolving the family into distinct species was achieved by including large amounts of sequence data in the analysis. Initially, such analyses were based on sequence divergence of a single gene (16SrDNA) or on a small number of housekeeping genes as in multi locus sequence typing (MLST) [52].
The evolution of distinct DUS dialects in this phylogenetically compact family is a striking example of how preference for homologous DNA in highly transformable bacteria affects evolution. Differences between the dialects are expected to be mirrored in the amino acid sequence(s) of the recently confirmed DUS-specific binding protein ComP, and warrants further functional investigation. The congruence between DUS-dialect and phylogeny and the presence of ComP suggests that those dialects that remain to be confirmed functional DUS are true DUS. This is also further emphasized by the exceptional overrepresentation and conservation of each dialect in their respective genomes. The most plausible hypothesis explaining these observations is DUS-dependent bias in frequent transformation/recombination [11].
The differential influence of each nucleotide in AT-DUS on N. meningitidis transformation and DNA binding/uptake was tested by employing donor DNA harbouring altered DUS. Single nucleotides were altered to be the transverse (purine↔pyrimidine) and the least common nucleotide at that position in the N. meningitidis genome. A similar analysis, based on the uptake of radioactive labelled DNA, has previously been reported in H. influenzae [53], [54]. Here the first steps of transformation were investigated by a DNA binding and uptake assay. The quantitative transformation method employed here measures the outcome of both uptake and recombination of DNA. This gradual analysis is important since it has been documented that DUS may influence multiple steps during transformation [15]. The most significant 5′-CTG-3′ core identified in transformation was conserved in all dialects of DUS (Figure 1). In contrast to the transformation experiments with N. meningitidis MC58, DNA binding did not display differential binding of AT-DUS and mutated AT-DUS versions, but clearly showed that binding discriminated against the DUS-less negative control (Figure 3B). This observation suggests that DNA binding in this strain is not very strict in terms of DUS specificity, and that DUS and single nucleotide mutated DUS can contribute to binding. Also the observation that the DUS dialects most similar to AT-DUS bound better than the more distant dialects emphasizes this point (Figure 4B).
It has been hypothesized that DUS specificity may function at more than one level during transformation [15] and one may speculate that initial binding by ComP [17], [18] could display weak DUS specificity and that the influence of the core 5′-CTG-3′ first become influential during uptake or later during the transformation process. The DNA uptake data from N. meningitidis (Figure 3B and Figure 4B) are corroborating the transformation data in that AT-DUS outperforms the other dialects tested. The differences in relative uptake of the close and distant DUS dialects in N. meningitidis (Figure 4B) suggest that DNA uptake is not only a relative function of binding, but may be influenced by DUS specificity.
The transformation performance of individual DUS-dialects from separate phylogenetic branches (Figure 2) was tested in N. meningitdis, N. elongata, E. corrodens and K. denitirificans (Table 1). The degree of similarity between the DUS-dialect of the recipient species, AT-DUS in N. meningitidis, and that in the donor DNA, AG-DUS, AG-mucDUS, AG-kingDUS and AG-eikDUS, directly correlate with the level of transformation. The potential for high levels of commutation when DUS dialects are similar is reflected in reports describing interchange of DNA between pathogenic and commensal Neisseria in vivo [45], [55].
This correlation is also evident in transformations of the AG-DUS species N. elongata, since AT-DUS outperforms AG-mucDUS. These observations suggest further that a nucleotide change in position −1 of the DUS is less influential than a change in position +2 in concordance with the results in N. meningitidis (Figure 3A). Transformations in the genera Eikenella and Kingella show strict autologous DUS-dependency in transformation indicating that AT-DUS is too divergent to allow transformation. It is well established that general sequence divergence between recipient chromosome and transforming DNA is strongly affecting homologous recombination, the last step in transformation [3]. Based on these observations one may anticipate that the phylogenetic distance correlates with the potential for commutation since DUS dialect distribution is reflected in the orthology of the Neisseriaceae. Furthermore, no significant transformation of N. meningitidis was observed when transforming DNA carried USS, which is the DUS of the Pasteurellaceae. Since H. influenzae and N. meningitidis share the same habitats, and are likely to encounter each other's DNA in e.g. oropharyngeal biofilms, the establishment of a functional barrier to commutation between these species may be important for the preservation of genome integrity. Genetic exchange between N. meningitidis and H. influenzae is rare [56] while the frequent commutation within the Pasteurellaceae is well documented [21], [57].
N. elongata subsp. glycolytica was previously shown to be transformable with a GT-mucDUS, but with an 8-fold reduced efficacy compared to a GT-DUS [23]. In our analysis, using a similar reporter construct, this factor was higher for the AG-mucDUS when compared to the AT-DUS (25-fold) and when compared to the AG-DUS (35-fold) (Table 1). These differences could relate to the employment of the non-ideal GT-DUS in the initial study. The DNA binding and uptake assays show that N. elongata subsp. glycolytica, like N. meningitidis, binds DNA in a DUS-specific manner with preference for the most similar DUS sequence corroborating the transformation results discussed above.
The influence of AG-DUS and AT-DUS in transformation of N. mucosa or N. sicca could not be tested since these strains were incompetent for transformation. The molecular basis for this remains unexplored, but the DNA uptake data show that transformation deficiency can be linked to the reduced ability to take up DNA, suggestive of a malfunction in this initial step of transformation. It is curious that N. mucosa binds DNA exceptionally well in a DUS-independent manner and this observation warrants further investigation. The functionality of AG-mucDUS in transformation has also been verified in other laboratories (N. Weyand, personal communication, [22]) but mucDUS-dependent transformation of species in the mucDUS-group has not yet been demonstrated. Recent observations in our lab confirm that also wadDUS is a true DUS affecting transformation in N. wadsworthii (unpublished data). The simDUS found in S. muelleri and kingDUS found in K. oralis are the only DUS that remains to be functionally verified, but the presence of comP genes in their respective genomes [18], the high overrepresentation of each DUS dialect and intragenomic DUS conservation in addition to their high similarity to the other DUS strongly suggests that they are, or at least have been, genuine DUS.
E. corrodens and K. denitrificans have previously been shown competent for transformation with homospecific DNA [58], [59]. Here, we established that this specificity is DUS-dependent. The DNA binding and uptake results did not reflect the differences observed in transformation since DNA binding was relatively uniform irrespective of the DUS-dialect used. However, it must be noted that binding was very low in both K. oralis and E. corrodens although a weak but statistically significant preference for autologous DUS in DNA binding or uptake, respectively, is evident (Figure 5).
Here, we expanded the number of bacteria that utilize a DUS-dependent mechanism for transformation of homologous DNA. Eight distinct dialects of DUS in the family Neisseriaceae were described, and the ability to overcome transformation barriers was assayed by both transformation and DNA binding and uptake for five of these. Furthermore, the evolution of DUS dialects corresponds with the evolution of the distinct core genomes of each phylogenetic clade. The DUS signal was analyzed by single nucleotide mutational analysis and an essential three nucleotide core sequence was found to be strictly required for transformation. This functional DUS core is conserved in all eight DUS dialects. The level of commutation was found to correlate with the phylogenetic distance and also to the similarity of the DUS sequences themselves. Future studies will explore the evolution of DUS dialects in regard of the recently confirmed association between DUS and ComP.
Plasmids containing the dialects and transversion variants of DUS were based on the plasmid p0-DUS [14]. Oligonucleotides listed in Table S3 were used to amplify the pilG::ermC fragment from p0-DUS by PCR whereby the oligomer OH3 was always used as reverse primer. The PCR products were digested with XhoI and SacII and inserted into the multiple cloning site of the vector pBluescript II SK+ (Stratagene, USA). E. coli strain ER2566 (NEB, USA) was used for cloning and the strain XL-1 Blue (Stratagene, USA) was used for large scale purification of plasmids due to higher yields. Plasmids were purified using the QIAGEN Plasmid Plus Midi Kit (Qiagen, Germany). Plasmids are listed in Table S4. The DNA was diluted to a concentration of 100 ng/µl in 10 mM Tris, pH 8, and stored at −20°C until used.
Bacteria used in this study are listed in Table S4. Escherichia coli strains were grown on LB medium. E. coli XL-1 blue was used for the quantitative production of the pDV plasmids. N. meningitidis and N. elongata were grown on blood agar plates, on GC medium plates or in liquid GC medium supplemented with IsoVitaleX. K. denitrificans and E. corrodens were grown on blood or chocolate agar plates or, when in liquid, in brain heart infusion broth. Antibiotics were added to the media when appropriate.
Quantitative transformation was performed as previously described [14] using plasmid or genomic DNA carrying an antibiotic resistance marker. Briefly, for N. meningitidis, cells grown over night at 37°C were suspended in 5% CO2 saturated GC medium containing IsoVitaleX and 7 mM MgCl2. 5 µl of DNA (100 ng/µl) were provided in 15 ml tubes, 500 µl cell suspension was added, shortly mixed by vortexing and incubated at 37°C for 30 min without agitation. Each sample was diluted by adding 4.5 ml GC medium and incubated for 4.5 h at 37°C on a tumbler (60 rpm). The cultures were then mixed and serial dilutions prepared in GC medium. Of each undiluted sample, 50 µl aliquots were spread on blood agar plates containing 8 µg ml−1 erythromycin or 50 µg ml−1 streptomycin and 50 µl of the 10−7 dilution were spread on blood agar plates without antibiotics. At least 2 agar plates were inoculated from each sample and experiments were repeated at least three times. Colonies were counted following over night incubation in a 5% CO2 atmosphere at 37°C. Individual transformation frequencies were calculated as the number of antibiotic-resistant colony forming units (CFU) per total CFU. The absolute transformation frequencies varied between experiments, possibly due to the difficulty in reproducing bacterial suspensions with identical fractions of competent bacteria. This problem was also reported earlier [22], and was resolved here by the use of relative values based on an internal standard. A Kendall's W test showed a very high concordance of gained orders (Kendall's W = 0.9346, χ2 = 85.0506, df = 13, p<0.0001). The absolute transformation frequencies are plotted in Figure S2.
In order to transform N. elongata, K. denitrificans and E. corrodens, small alterations of the protocol were required. The N. elongata subsp. glycolytica strain employed here was the type strain originally isolated by Henriksen et al. [68]. Agglutinating P+ colonies of N. elongata were pre-selected since previous experiments had documented a positive correlation between agglutination and transformation [69]. The type strain of K. denitrificans and E. corrodens strain 31745 were used for quantitative transformation. The E. corrodens strain 31745 is a reference strain that was previously shown to be transformable [58]. To generate donor DNA for the quantitative transformation experiments, spontaneous streptomycin-resistant mutants of N. elongata, K. denitrificans and E. corrodens were isolated and the genomic DNA from these strains used as a template for a PCR of the rpsL gene using primers listed in Table S5. The rpsL genes were sequenced, confirming the Lys to Arg mutation at amino acid position 43 conferring streptomycin resistance [70]. rpsL PCR products were adjusted to 50 ng/µl and 10 µl thereof was used in transformation of 500 µl cell suspension. K. denitrificans was transformed in Brain heart infusion medium supplemented with 7 mM MgCl2 and the second incubation time was extended to 5 hours.
For radiolabeling the pDV plasmids (Table S4) were linearized by digestion with ScaI, purified using QIAquick columns (Qiagen) and treated with exonuclease III (Fermentas). After heat inactivation 10 µg of the partially single stranded DNA were incubated with 3 µM dNTPs, 20 µCi [α-32P]dCTP and 10 units Klenow fragment (3′→5′ exo–) (NEB) for 90 min at 37°C. The fill-in reaction was finished after increasing the dNTP concentration to 30 µM and additional incubation for 30 min. The products were purified as before and showed specific activities of 4×105 to 2×106 CPM µg−1. The Neisseriaceae were grown in liquid medium to an optical density at 660 nm of approximately 1. MgCl2 was added to 7 mM and one ml aliquots were incubated with DNA with about 5×106 CPM activity and rotated at 37°C for 45 min. The samples were then split into two 500 µl samples and 12.5 units Benzonase (Merck) was added to one of these. After additional incubation for 15 min the cells were washed three times by centrifugation for 3 min at 5000× g and resuspension in liquid medium including 7 mM MgCl2. The final pellets were resuspendet in 3 ml scintillation fluid (Ultima Gold MV, PerkinElmer) and measured twice for 3 min in a Tri-Carb 2900TR (PerkinElmer) using an energy window LL-UL = 50–1700. DNA binding and uptake are reported as the percentage of DNA added and percentage of cell-bound DNA, respectively, and the results are presented as the means of for 3 to 5 replicates.
|
10.1371/journal.pcbi.1003902 | Relationships between Th1 or Th2 iNKT Cell Activity and Structures of CD1d-Antigen Complexes: Meta-analysis of CD1d-Glycolipids Dynamics Simulations | A number of potentially bioactive molecules can be found in nature. In particular, marine organisms are a valuable source of bioactive compounds. The activity of an α-galactosylceramide was first discovered in 1993 via screening of a Japanese marine sponge (Agelas mauritanius). Very rapidly, a synthetic glycololipid analogue of this natural molecule was discovered, called KRN7000. Associated with the CD1d protein, this α-galactosylceramide 1 (KRN7000) interacts with the T-cell antigen receptor to form a ternary complex that yields T helper (Th) 1 and Th2 responses with opposing effects. In our work, we carried out molecular dynamics simulations (11.5 µs in total) involving eight different ligands (conducted in triplicate) in an effort to find out correlation at the molecular level, if any, between chemical modulation of 1 and the orientation of the known biological response, Th1 or Th2. Comparative investigations of human versus mouse and Th1 versus Th2 data have been carried out. A large set of analysis tools was employed including free energy landscapes. One major result is the identification of a specific conformational state of the sugar polar head, which could be correlated, in the present study, to the biological Th2 biased response. These theoretical tools provide a structural basis for predicting the very different dynamical behaviors of α-glycosphingolipids in CD1d and might aid in the future design of new analogues of 1.
| To modulate the natural immune response toward aggressive (Th1) or protective (Th2) profiles remains a difficult challenge, but can also offer great therapeutic opportunities, particularly for the treatment of cancer or auto-immune diseases. It has been demonstrated that a particular type of cells, named invariant Natural Killer T (iNKT) cells, are able to induce both protective and aggressive response profiles, depending on the antigen that is presented to it by CD1d proteins. Since this discovery, efforts have been made to find synthetic compounds that would selectively induce Th1 or Th2 immune response. KRN7000 was the first to selectively induce a Th1 response, and for two decades many analogues of this compound were synthesized. Some of them effectively induce Th1- or Th2-biased responses. But, unfortunately, the Th1/Th2 selectivity mechanism remains unclear. That is the reason why we have undertaken large-scale molecular modeling (molecular dynamics) simulations of various CD1d-ligand systems with the aim to find out correlation between chemical modulation and the orientation of the biological response for a variety of known ligands.
| Compound 1, [1], [2] also referred to as α-GalCer, is a synthetic glycolipid that has shown promising bioactivity against diverse pathologies (atherosclerosis, malaria, auto-immune diseases…). [3]–[7] This compound is presented to the iNKT cells via a MHC class 1-like protein, named CD1d, associated to β2-microglobulin. 1 can be readily loaded onto both mouse and human CD1d. The resulting binary complex is carried by Antigen-Presenting Cells (APCs) such as dendritic cells and macrophages. Upon recognition of the CD1d-glycolipid complex by the T Cell Receptor (TCR) (Figure 1), the iNKT cells rapidly initiate response that leads to a release of cytokines implied in Th1 (interferon-γ, IFN-γ) and Th2 (Interleukin-4, IL-4) immune response profiles, which yields opposing results from a medicinal point of view. Much research has been focused on being able to control the cascade by attempting to bias the cytokine release profile Th1/Th2. Many biological and synthetic studies have been undertaken by a variety of research groups aimed at understanding the mechanism of 1 recognition in regards to both CD1d and TCR proteins with the hope of finding novel analogues with improved biological response (magnitude and profile selectivity of the iNKT cell stimulation). [8]–[10] But, for the time being, the relationship between glycolipid pharmacomodulation and cytokine polarization is not completely understood. However, principles were established from earlier studies, based on the stability of the CD1d-ligand-TCR trimolecular complex. Mc Carthy et al. demonstrated experimentally that modifications of the lipid chain buried in the F′ channel of human CD1d molecules (Figure 1) can modulate the TCR affinity. [11] However, they also showed that even though the length of the acyl chain controls the stability of the binary complex it does not automatically affect the CD1d-glycolopid complex affinity to TCR. In order to affect the binding affinity for TCR, it seems that a ligand chemical variation must additionally induce conformational changes of CD1d, which propagate to the TCR recognition surface. For example, these authors demonstrated that the incomplete occupation of the human CD1d F′ channel by the chain-shortened analogue 2 (OCH) [12] results in a less stable binary complex but also suggested that this causes conformational differences at the TCR recognition surface. In other respects, Porcelli et al. [13] have shown that the sugar head group of the ligand contacts the TCR in the initial phase whereas CD1d contacts with the TCR contribute to the stability of the whole complex. Since the IL-4 production was shown to require shorter TCR stimulation than IFN-γ, it has been thought to generate less stable CD1d-ligand complexes in order to impair the interaction at the ternary interface and then to elicit the cytokine profile toward a Th2 response. Conversely, a biased Th1 response was predicted through increasingly stable CD1d-glycolipid complexes. Hence, all attempts to design new ligands that polarize the cytokine profile were based on this principle of stability of the binary and ternary complexes, however not taking into account directly CD1d conformational changes induced by ligand modulations. Many modulations have been envisaged. Derivatives were obtained by changing at least one of the four distinct portions describing 1: the sugar part, the osidic bond, the polar linker, and the two lipid chains. Throughout our manuscript, the term “polar head” will refer to the distinct fragment of the ligand that is (α-anomerically) linked to the ceramide, regardless of the analogue. This designates the group that protrudes out the binding groove of the CD1d, towards the TCR, in contrast to the two more deeply anchored alkyl chains.
In order to explain observed biological evaluation of analogues of 1, molecular modeling supplements have sometimes been given by authors. However, these theoretical approaches are often limited to molecular docking or local optimizations of the ligand into the CD1d pockets. According to the huge number of degrees of freedom of 1 and its analogues such docking simulations are not appropriate to understand the mechanisms involved in such complex recognition process. Pipelier et al. [14] recently employed ab initio QM/QM′ level of theory to estimate electron withdrawing effect induced by the introduction of mono or difluoro substituent at C3. Such CPU extensive quantum mechanical study is however limited to a part only of the entire system and to a single structure. In order to explore the impact of a single-amino acid variation at position 93 of iNKT on the conformational stability of Complementarity Determining Region (CDR) 3α, Gadola et al. [15] recently carried out molecular dynamics (MD) simulations of the ternary complex. Though very interesting, several questions arise from such simulation. Can tools for measuring conformational stability of CDRs be restrained to a single Cα-Cα distance probability distribution function? Will the expected conformational change be observed during a unique 10 ns simulation? Unlike experimental studies, only a few molecular modeling studies were fully dedicated to the investigation of the binary (CD1d-glycoliplid) or ternary (CD1d-glycolipid-TCR) complexes. To our knowledge only two previous studies have been fully devoted to the theoretical study of these systems. Nadas et al. [16] performed molecular dynamics of the ternary complex that allowed for the in-depth statistical and visual analysis of the H-bond network between CD1d, TCR and a set of 12 ligands during the simulation. The study was however limited to a single 3 ns simulation of a truncated complex, and tools for analysis were limited to hydrogen bond monitoring and visual inspection. In their theoretical study, E. Henon et al. [17] addressed the influence of three modulations on the dynamic behavior of the CD1d-glycolipid complex. However, only one 10 ns trajectory was produced for each of the four envisaged binary systems. In this previous study, the influence of the ligand modulations on the dynamic behavior of the CD1d-glycolipid complex was addressed by distance analysis and mainly focused on the so-called OTAN H-bond network built up from 2-OH, Thr154, Asp151, and NH. To be able to predict the strength of the Th1/Th2 polarization, very recently, De Spiegeleer et al. have presented multi-linear regression (MLR) and partial least squared (PLS) models based on a set of chemical descriptors of the ligand. [18] Though simple and easy to implement, these statistical methods partially failed to explain Th2 biased responses in vitro, and the use of numerous chemical descriptors prevents us from truly understanding the underlying correlation between chemical alterations and the cytokine-responses.
Besides, for now, no MD simulations have been performed for characterizing differences between recognition of glycolipid by human or mouse CD1d. This point is important since there may be difference between antigen recognition by mouse and human iNKT cell. Nor is there any study of the influence of spacer lipids on the conformational behavior of the protein. Actually, sometimes, the CD1d protein has got non-specific lipid into its pockets, even in presence of a ligand (for instance the shortened glycolipid 2).
Clearly, molecular dynamics is one of the most appropriate tools to study interactions in these complexes and to examine how chemical variation can affect their properties. However, as previously explained, since the ligand binding affinity to CD1d is not systematically correlated to the affinity of TCR to the binary complex, this reduces considerably the interest of ligand binding affinity predictions (such as relative MMGB-SA calculations [19]). Moreover, relative binding free energy calculations via alchemical transformations [20] were ruled out due to the very large number of degrees of freedom in the two lipid chains in 1. Simulating the ternary TCR-ligand-CD1d complex would be a very interesting study, but it would require still larger sampling compared with the binary system. Indeed, such a ternary complex is a very different system from the binary one. The TCR binding involves many additional interactions, compared to the binary complex, some of which stabilize the polar head at the binding interface (Phe29, Ser30 of CDR1alpha, and Gly96 of CDR3alpha). Thus, the impact of chemical alterations of the ligand involved in the ternary “lock and key” recognition might be observed but at a much larger time scale. Therefore, we have chosen another route. Since the TCR recognition process requires that the binding footprint onto CD1d to be maintained, instability of the ternary complex can occur only if this binding footprint is deteriorated. Figure 1 shows the non-covalent interactions[21], [22] at this interface in the X-ray structure of human CD1d-1-TCR ternary complex (PDB reference 2PO6). Any deformation of this interface may disable the interaction with TCR, or at least makes it less effective. Hence, the idea is to focus on the binary complex (CD1d-ligand) and to determine how a chemical modulation of the ligand loaded into CD1d affects the interface part of this binary complex. Since binary complex-TCR contacts involve both the ligand polar head and the α1/α2 helices of CD1d, these two portions of the complex have to be particularly monitored during the simulations. Most of the SAR approach assumes that the structure of the ligand alone contains the features responsible for its biological properties. Here, using molecular dynamics, we took another step forward by studying the propagation of conformational changes (induced by a chemical alteration) from within the binding grove to the surface of the binary complex. In other words, in our study, we consider the whole binary complex as a “ligand” in correlating biological activity to chemical space.
The reason why the TCR recognition process is altered may be due to an irreversible structural deformation of a portion of the interface or it may also result from an unusual dynamical behavior of this interface caused by abnormal fluctuations with larger amplitude for example. From MD trajectories, detecting these fluctuation deviations can be quite difficult because they can occur at different scales: small scale (hydrogen-bond, residue fluctuation, polar head rocking) or at even a larger scale (secondary structure motions). That is the reason why, beyond simple distance monitoring, employing additional appropriate analyzing tools is important to reveal such unusual dynamic behavior. The question is whether chemical modulations known to induce a Th2 profile (or Th1) will give MD simulations with similar features and readily detectable with post processing tools. This methodological and interpretative task is made more difficult since some ligands generally induce simultaneously both Th1 and Th2 responses (only a bias is experimentally observed) and also because experimental protocols that measure this bias can be quite different making it difficult to rigorously compare the Th1/Th2 bias values. Biological evaluations from different data sources can also sometimes be contradictory. [23] Furthermore, other factors such ligand solubility, biodisponibility, or stability in biological systems, may also play a role, which cannot be handled with our simple MD models.
In our work, we carried out 48 MD simulation (11.5 µs in total) involving eight different ligands (conducted in triplicate) (Figure 2) in an effort to probe if a ligand modulation, which is known to lead to a Th2 bias, impacts on the conformational stability of the system, and how this ligand alteration is reflected in the dynamic behavior of the whole molecular structure. To what extent computational tools are able to predict a Th2 bias is very challenging for the design of new ligands. The main aim is then to test whether such a simple rule: complex instability-Th2/complex stability-Th1, is really reliable or not. From our simulations, human and mouse CD1ds are found to exhibit different structural dynamics. Moreover, specific dynamical features haven been identified, which could be correlated here to Th2-biased systems exclusively.
Overall, 16 systems have been simulated and analyzed (Table 1, Figure 2). Whenever it was possible, the short name chosen by the authors for the ligand has been followed. But additionally, the prefixes “H” or “M” have been inserted and stand for human or mouse, respectively. This set has been established in order to allow three types of comparison: human CD1d against mouse CD1d simulations, Th1 versus Th2 response and simulations with or without spacer lipid. The 2PO6 and 3SDA PDB structures[24], [25] were used for the human and mouse CD1ds, respectively.
Seven analogues of 1 have been chosen so as to account for modifications on the four portions of the glycolipid: the sugar part, the osidic bond, the polar linker, and the two lipid chains (Figure 2). The ligand 4 (7DW8-5) has got a phenyl group at the end of the shortened acyl chain. This analogue was shown to induce a Th1 biased response against human iNKT cells in vitro with binding affinity to human and mouse CD1d molecules. [26] The H_LIP designates simulations where the ligand has been replaced by two free lipid chains in the pockets F′ (C12) and A′ (C16) of the CD1d protein, thus, with no polar head present. These lipid structures have been taken from PDB files 3ARB [27] and 1Z5L. [28] The analogue 6 [29] has its acyl lipid chain truncated to eight carbon atoms and a Th2 polarization response of iNKT cells has been determined for this glycolipid. The 5 (NU-α-GalCer) ligand presents an ureido-naphtyl-group at position 6″ and exhibits pronounced Th1-biased cytokine production. [30] No interaction is observed between this ureido-naphtyl-group and the TCR part in the X-ray structure (PDB 3QUZ). The analogue 2 is a glycolipid with a shortened sphingosin chain (9 carbon length) and was found to induce a pronounced Th2-biased cytokine release compared to 1. The molecule 3 (OCH9) is almost identical to 2, only differing from it by the addition of two methylene groups on the acyl chain. In the study by Mac Carthy et al. [11] the ratio of IL-4/IFN-γ found for this compound showed clearly a Th2-bias compared to the ratio obtained for 1. 7 (α-S-GalCer) is a thioglycoside analogue of 1 and did not activate murine iNKT cells in vivo. [23], [31], [32] But, there are conflicting studies for the biological evaluation of 7 for human iNKT cells that is predicted to elicit a preferential Th2-biased response [32] or no real bias compared to 1. [23] Substitution of the amide function by a 1,2,3- triazole group, compound 8, induces Th2 cytokine production. [33] Finally, the human CD1d alone has been simulated (H_CD1d) without any ligand and free lipid chains in its pockets.
A crystal structure is not available for each of the 16 studied systems. For some of them, the X-ray data are available but unfortunately they leave some protein structures incomplete. That is the reason why we selected only two CD1d structures: 2PO6 (human CD1d) and 3SDA (mouse CD1d) with full CD1d structures. Then, starting from the geometry of the KRN7000 ligand from PDB file 2PO6 we generated the OCH9, AZOL, SaGAL, and 7DW analogues. For this, the program molden [34] was used to achieve chemical alterations. Truncating the sphingosine chain to obtain the OCH9 compound was straightforward. Using the Z-matrix editor of Molden allowed us substituting the amide function by a 1,2,3-triazole group (AZOL compound), substituting an oxygen atom by a sulfur atom (SaGAL) and adding a phenyl group at the end of the shortened acyl chain (analogue 7DW). In this way, the derived analogues fitted naturally into the hydrophobic pockets of human CD1d (2PO6). In order for these analogues to fit also in the mouse (3SDA) CD1d we performed the superposition of the 3SDA structure on the 2PO6 one (using sequence alignment). The geometries of the analogues GOF, NUaGAL and OCH were taken from X-ray structures 1Z5L, 3QUZ and 3ARB, respectively. These protein structures were previously aligned to the 2PO6 one in order for the associated ligands to fit in the two binding pockets. Concerning the NUaGAL ligand, we had to modify the ureido-naphtyl group position in order to avoid steric clash with Trp153 in human CD1d. All these generated structures were then minimized in a first step before molecular dynamics simulations. Whenever a spacer lipid was added, it was taken from crystal structures 3ARB (spacer lipid simultaneously present with the sphingosine chain) and 1Z5L (spacer lipid simultaneously present with the acyl chain).
In the primary sequence of CD1d, the mouse protein contains an insertion of two residues between residues 89 and 90 of human CD1d. This does not concern the binding domain. The preparation of the protein (disulfide bond linkages, protonation state of ionisable side chains) was achieved using the same protocol as in our previous study. [17] In particular, as specified in the X-Ray structures, three disulfide covalent bonds were set between cysteine residues. We employed the package program xLeap of the AMBER11 package [35] to add the hydrogen atoms to the protein structure. The Propka [36] application was used to examine the protonation state of ionizable side chains with a focus on residues asparte, cysteine, histidine and glutamate in proximity to the ligand. No protonation state change was required compared to the state proposed by xLeap. Counterions (Na+) were added such that to neutralize the unit system. The ff99SB force field was used for the protein. The general amber force field GAFF was used in conjunction with the antechamber program to describe the ligands. [37] The respgen procedure was employed to derive atomic charges from HF/6–31G* electrostatic potential calculations obtained with the Gaussian package [38]. The ligand and protein CD1d were fully explicitly solvated in a truncated octahedral box using TIP3P water molecules with a buffer distance of 7.0 Å and under periodic boundary conditions. We have carefully checked that the initial water molecules buffer remains during our simulations, and the CD1d protein does not interact with its neighbors in periodic images. The entire system consisted of about 40700 atoms (depending on the ligand).
All the dynamics calculations were carried out with the programs sander and pmemd of the AMBER11 package. The particle mesh Ewald procedure was employed to handle long-range electrostatic interactions. The default value of 8.0 Å for the non-bonded cutoff was set to calculate van der Waals and electrostatic energies. No switching function was used for the van der Waals interactions. The system was then carefully prepared. At first, the energy of the entire system was minimized with 1000 cycles of steepest descent followed by 1000 cycles of conjugate gradient minimization. This process was repeated twice. Classical Langevin NVT molecular dynamics simulations (collision frequency = 2 ps−1) employed a 2 fs integration time step (bonds involving hydrogen atoms were constrained). Firstly, the water molecules were heated from 0 to 300K during 20 ps, and equilibrated for 20 ps. A force constant of 100 kcal.mol−1.Å−2 was used to restrain all other atoms during these steps. After cooling the solvent molecules to 0 K during 20 ps, this heating-equilibrating procedure was re-run twice for the entire system with no constraint and followed by 240 ns of data collection. Prior to production, in order to allow the density to equilibrate, our system has been equilibrated using the NPT ensemble (isothermal-isobaric ensemble) at 300K and 1 atm for 40 ps.
Three 240 ns trajectories have been produced in parallel for each of the 16 systems (0.72 µs each). Coordinates of the system were written every 1 ps.
As said above, the polar head is known to protrude out the binding groove of the CD1d and is a key element in the TCR recognition. Its orientation is characterized by three dihedral angles: φx, φy, φz. Therefore the ability of the GAFF force field to reproduce static ab initio quantum chemistry calculations was tested on these three degrees of freedom. Focusing on these three parameters, rigid potential energy scans for internal rotations about the three axes φx, φy, φz in the 8 compounds were performed to compare GAFF energies to electronic structure calculations (at the DFT/6-31G* level of theory using the M06 hybrid functional, which is recommended by Truhlar and col. [39] for non-covalent bonds descriptions). Single points energy calculations were performed every 5°. It can be clearly seen from Figure S1 that the potential energy profiles for these three rotational degrees of freedom are reproduced in a satisfactory manner by the force field GAFF. The rotations around the two first angles (φx, φy) emphasize wide regions of steric hindrance (several tens or even several hundreds of kcal/mol) corresponding to steric clashes between the head group and the osidic bond, the sphingosine and acyl chains. Outside these very repulsive zones, GAFF correctly describes the minimum energy regions. Our tests were limited to these three internal coordinates. Indeed, it is to be noticed that in these binary complexes, the lipid anchors and the hydrogen bonds considerably reduce the other internal degrees of freedom of the ligand.
A 2D-RMSD graph represents the root mean square deviation (RMSD) of every conformation to all other conformations of a simulation, as a function of time during a 240 ns simulation. Each point of the two-dimensional plot corresponds to the RMSD between two trajectory structures, and its value is encoded into a color. The diagonal elements (black) represent self-comparison (zero RMSD) and the yellow dots show the largest pairwise RMSD. RMSD values have been computed based on the binding region, i.e., using an atom selection including the alpha carbons of helices α1 and α2 only (residues in the range 58–92 and 137–184 in human and in the range 58–92 and 139–186 in mouse). The average of all RMSD values of this matrix was also computed for each graph. The resulting number represents the level of fluctuation of the whole system during the simulation. In parallel, the root mean square fluctuation (RMSF) of the CD1d protein was computed as a measure of the average atomic mobility. It was calculated on a residue basis using the RMSF of the positions of the Cα atoms of all the protein.
A so-called dihedral footprint was computed, based on 9 dihedral angles made by the Cα atoms of 9 CD1d residues that form a contact with TCR. The considered residues are: 76, 79, 80, 83, 84, 87, 99, 147(149 in mouse) and 150(152 in mouse). Sine and cosine transformed dihedral angles were used to avoid problems arising from the circularity of angles. The human 2PO6 X-ray structure was chosen as the reference in all simulations for the calculation of the dihedral values. For one given dihedral angle θi, the squared deviation from its reference position θ0 is measured as follows: . This formula derives from the sum of the two squared differences for sine and for cosine. The final index was obtained by calculating the root mean square deviation over all the 9 dihedral angles. This index varies in the range 0–2 and the percentage change is reported in the figures.
The change in inter-helix distance was monitored by computing the distances between centroid pairs along the two portal helices α1 and α2. Each centroid is formed by four successive residues. Helices α1 and α2 consist of 7 and 10 centroids, respectively. The chosen procedure is based on a modified Hausdorff distance calculation. For every centroid of helix α1 we determine the smallest distance to any centroid of helix α2. The sum of all these distances is computed. The procedure is repeated for every centroid of helix α2 relative to all points of helix α1 and a second sum is deduced. Finally, the inter-helix separation was assessed by summing these two distances and by averaging over the total number of centroids of helices α1 and α2.
Each dihedral angle θi formed by 4 bonds joining 4 successive Cα atoms along the main chain of the protein CD1d is a probe of the free-energy landscape (FEL) along the primary sequence. The 1D Free Energy Landscape (1D_FEL) [40] of these coarse-grained dihedral angles are obtained based on the logarithmic relation at 300K: G = −RT ln P(θi) using the one-dimensional probability distribution function P(θi) derived from our simulations. For a given dihedral angle, in order to compare the 1D_FEL between two simulations i and j, the following procedure was carried out. First, the 1D_FEL curve j was shifted on the energy axis so as to align the minima of i and j. Next, the curve j was shifted on the angle axis so as to minimize the Hausdorff-like distance between the two curves i and j. In this way, the resulting distance value (with unit mixing the energy and angle axes) measures shape dissimilarity between the two 1D_FELs. Points with energy above 20 kBT in the 1D_FEL were excluded from this procedure.
Similarly, for the polar head, three-dimensional histograms were constructed from values of dihedrals φx, φy, φz and converted to free energies (3D_FEL) based on an analogous logarithmic relation at 300K using the probability distribution function P(φx, φy, φz) given by our simulations. Each of the isosurfaces shown in our paper corresponds to points of the 3D-space (φx, φy, φz) with a constant free energy isovalue (in kcal mol−1). Nine isovalues were considered from 1 to 9 kBT.
Our study disclosed the existence of specific states with lifetime ranging from a few nanoseconds to 30ns. It takes only a few picoseconds to switch from one of these states to the other. Due to this very short timescale, no dihedral principal component analysis (dPCA) could be carried out to correlate the occurrence of these transitions to specific structural and dynamical features of the system (coordinates of the system were written every 1 ps during the MD).
To achieve our objectives, a large conformational sampling is required. Such a large amount of data is difficult to analyze. Our trajectory analyses use a wide range of tools to extract the most relevant and unanticipated events. Since we are interested in detecting conformational behavior differences between systems, these tools are now exemplified in the case of the two systems Sy1 and Sy16, expected to behave quite differently.
First, 2D-RMSD (α1/α2 interface) can be helpful to detect different conformational “substates”. While replicas I and II of H_aGAL show a relative homogenous 2D-RMSD plot (Figure 3), replica III appears to move into two “wide” different conformations. Trajectories of H_CD1d give more contrasted patterns, with replica III showing three dissimilar clusters. The matrix average of RMSD values accurately reflects the level of granularity observed by visual inspection (numbers reported in Figure S2). These plots clearly show that the two CD1d helices can be very stable during 240 ns or in contrast go through several distinct conformational states with a life of a few tens of nanoseconds. Therefore, it seems that one cannot just perform a single 240 ns MD simulation but rather running multiple MD with different starting initial conditions is better to study the interface of the binary complex. Surprisingly, at this stage, though H_CD1d displays slightly more contrasted 2D-RMSD figures, the α1/α2 interface of H_aGAL and H_CD1d does not appear to have greatly differing dynamical behavior.
Despite the presence of ligand 1 in the case of H_aGAL, a comparison of the per-residue RMSF calculated from the binary complex trajectories shows that the two systems (with or without ligand) undergo very similar fluctuations (Figure 4, panel A), in particular no appreciable difference is observed in the ligand-binding pocket. For both systems the linker domain (amino acid sequences separating the helices domain from the beta sheet part of the CD1d protein) is the most flexible region, as expected. The extremities of the helices show the highest mobility, beta sheets are very stable. The same conclusions arise for the three replicas. The advantage of this tool over 2D-RMSD is to localize the regions of high mobility. However, it does not permit to distinguish between two situations: (a) regular but larger fluctuations due to higher mobility of some residues at the local scale in specific region of the interface but maintaining the secondary structure or (b) direct loss of secondary structure elements occurring at one point of the simulation, involving a larger amplitude motion and impacting RMSF values.
The analysis of the 1D Free Energy Landscape (1D-FEL) of the Coarse-Grained Dihedral Angles of the protein (CGDA, defined by four successive alpha carbon atoms) was used with a view to detecting the presence of such possible structure deformation. This methodology has been recently applied to studies of proteins. [40] In order to capture the largest deformations in the protein resulting from the absence of ligand we decided to compute a Hausdorff-like distance (see Materials and methods section) that measures how far two Free Energy Profiles (FEP) are from each other for a given dihedral angle in H_aGAL or in H_CD1d. The deformations can be well appreciated in panel B of Figure 4 where the color of the ribbons is directly proportional to this distance acting as a dissimilarity index. We report at the bottom of Figure 4 the FEP of the six residues the most influenced by the presence of 1. As can be seen, significant anharmonicity appears at the end of the helix α1 (on the side towards the F′ pocket) when the ligand is lacking. This comes in conjunction with the middle of the helix being more rigid. The lack of ligand clearly produces a bend in the middle of helix α1 combined with the destructuration of its extremity.
One may now question how this deformation will affect the binding footprint between the TCR and CD1d. The evolution of this interface during the simulation was monitored according to two complementary indexes calculated with reference to the X-ray structure (PDB 2PO6). In Figure 5, the first index (the distance between the centroids of α1 and α2 helices) reveals the closure of the cavity entrance in the lipid-free system (H_CD1d). Secondly, a so-called dihedral footprint was computed, based on the dihedral angles made by the successive alpha carbons of the 9 CD1d residues that are known to form a contact with TCR. This index varies in the range 0–2 and the percentage change is also reported Figure 5 (right). According to this index, the ligand-free CD1d interface greatly deviates from the reference (X-ray structure), by about 50% in comparison with the H_aGAL system (35%). All these results are in agreement with the theoretical results of Garzón et al. [41] who observed the same trend.
Finally, the conformational space explored by the polar head of the ligand during the simulations of H_aGAL was described using the three torsion angles of the three successive rotatable bonds shown in Figure 6. In the following, the first one (φx) will be referred as the anomeric pivot, the second one (φy) as middle pivot and the last one (φz) will be denoted the amide axis. Theoretically, rotations around φx and φz yield conformations in which the polar head has rotated horizontally at the top of the cavity entrance. In the CD1d environment, these two motions should be hardly sterically hindered by the two helices. In contrast, rocking about the middle pivot will get the polar head striking the helices. In all cases, such individual rotations would require the loss of the OTAN H-bond network. The free energy landscape of the MD trajectory along these three coordinates is illustrated in Figure 6 for the system Sy1. The top panel shows all points in the (φx, φy, φz) space with isovalues of G being one to nine kBT (0.6 to 5.3 kcal.mol−1) for replica I of H_aGAL. For all simulations, the conformational space does not grow any more for energies above 9 kBT. For replica I, the 3D-FEL emphasizes only one conformational state, showing that the OTAN network is strong enough to maintain the orientation during all the 240ns simulation. This state, hereafter referred to as “OTAN state”, is centered about the φx, φy, φz coordinates: (50°, 157°, 51°) in replica I. Comparison with the three replicas of H_aGAL at G = 5.3 kcal.mol−1 (9 kBT, bottom panel of Figure 5) reveals the presence of the OTAN state in every case. But in addition, two secondary conformational states can be visited, mainly limited to the (φx, φz) plane. It is very important to note that, as with the OTAN state, these two complementary states maintain the polar head interacting with helix α2. Actually, the second state observed in replica II is a combination of two rotations about φx and φz that limits the displacement of the sugar ring (see Figure S3). The third state observed in replica III corresponds to a rotation of about 180° around φz, which brings again the polar head in contact with helix α2 (see Figure S4). In this last state, the Asp151 residue is now hydrogen-bonded to the 4′-OH group (sphingosine chain) and the Trp153 side chain is again in van der Waals (VDW) contact with the hydrophobic side of the polar head. But most importantly, no conformations are observed involving rotations about the middle pivot φy, points that would lie inside the 3D-FEL box, involving the “y” axis.
Considerable structural similarities are apparent between mouse and human CD1d molecules. Focusing on the α1 and α2 helices-binding domain, human and mouse amino acid sequences show a similarity of 81.2% and an identity rate of 65.2% in this portion of the protein. The amino acid sequence is highly conserved in particular the residues Asp80, Asp151(153 in mouse) and Thr154(156 in mouse), which are of crucial importance to bind the ligand through the OTAN network, are present in both proteins. There are 25 amino-acid variations located in the TCR-interface domain. One of these seems to be particularly important. At position 153, the crystal structures show the presence of a bulky tryptophan side chain in human CD1d (Figure 1) in contrast with the glycine (no side chain) in mouse CD1d. This Trp153 side chain is in VDW contact with the hydrophobic side of the sugar in a face-to-face configuration (see Movie S1) and one may question whether the absence of this residue in the mouse CD1d may exert an indirect effect on TCR binding or not. Actually, as can be seen in Figure 1, the galactose head group acts as a mechanical stop restricting the TCR approach to only one side: the F′ part of the CD1d groove.
Both human and mouse CD1d proteins are able to induce a balanced Th1/Th2 response depending on the loaded ligand. However, the biological activities of analogues of 1 loaded into mouse or human CD1d can be sometimes quite different. [14], [32]
Hence, four “human versus mouse” comparisons are conducted for systems: Sy1/Sy2, Sy3/Sy4, Sy6/Sy7 and Sy8/Sy9. These systems were chosen because, with ligand unchanged, human and mouse CD1ds produce the same polarization (Th1 or Th2). Moreover, to ensure a thorough comparison, the ligand 2 is supplemented in both human and mouse simulations with a “spacer”, i.e. a linear hydrophobic compound taken from the CD1d-2 PDB (3ARB). From the analyses of these 24 trajectories, the outstanding results are the following. From a structural point of view, in all CD1d mouse simulations, we clearly observe an increased inter-helix distance localized on the A′ pocket side. On average, the inter-helix distance is about 1.3 Å larger in mouse simulations than in human CD1d simulations (see Table S1). This is not surprising since this portion of the two helices concentrates ten amino acid variations (8 residues on α2 helix and 2 on α1 helix) involving residues with physico-chemical properties very different between human and mouse proteins. This effect is however counterbalanced on the F′ pocket side where the portions of helices α1 and α2 come closer such that overall, the total inter-helix distance is almost the same for human and mouse CD1ds during the simulations (see Figure 7 and Table S1). Curiously, the M_NUaGAL/H_NUaGAL comparison exhibits a still larger increase of the inter-helix distance: 2.1 Å larger in mouse simulations than in human. This is likely to be correlated with the fact that the naphthyl group introduced on the head part of this analogue facilitates the interaction with two residues in a small pocket between helices α1 and α2 of CD1d (as demonstrated by Trappeniers et al. [30]). One of these residues is Ile69 in humans, replaced by Me69 in mice. Finally, it is to be noted that this inter-helix variation observed between mouse and human CD1ds is not accompanied by a dihedral footprint deviation (measured as a percentage value from our analysis tool based on CD1d contacts).
From a dynamical point of view, a higher mobility is found for the human protein CD1d compared to mouse CD1d. This is slightly noticeable from the 2D-RMSD (see Figure S2) and from RMSF pictures where more fluctuations are observed for the human linkers. From the 3D-FEL analysis tool, comparing the motions of the polar head in mouse or human CD1d, all simulations emphasize more flexibility when the ligand is loaded into the human CD1d rather than in the mouse CD1d. Whether for human or mouse, there is predominantly only one conformational free energy well explored during the simulations, which we term the “OTAN state” (Figure 6). However, the conformational space sampled within the mouse simulations is significantly smaller than the one for human simulations. This is evidenced by estimating the corresponding volumes enclosed by the free energy isosurface (9 kBT) in the 3D representation (φx, φy, φz) for both mouse and human simulations (see Table S2).
Unexpectedly, we observe a very intriguing free energy landscape for the replica II simulation of human 2, indicating an ensemble of five conformational states. This will be discussed in more details in the next section.
The goal is now to find out correlation at the molecular level, if any, between chemical modulation of the ligand and the orientation of the known biological response, Th1 or Th2. Analogue 1 (Th1) is compared to 2 (Th2), 7 (Th2) and 8 (Th2) within three comparisons. Firstly, two comparisons are made with the ligand loaded into the human CD1d: (a) H_aGAL/H_OCH or (b) H_aGAL/H_SaGAL. Additionally, in mouse CD1d, the compared behavior (c) M_aGAL/M_AZOL is then addressed. This study focuses on the systems Sy1, Sy2, Sy3, Sy12 and Sy14 of Table 1.
The key results are the following. From a structural point of view, H_OCH and M_AZOL, simulations revealed significant structural changes of the CD1d protein with regard to the other simulations. The replica II (mainly) of the H_OCH simulation shows a modification of the α1 helical structure involving residues 74 to 82 (on the F′ pocket side) as revealed and evidenced by the 1D-FEL analysis (it can be seen on Figure 8, middle panel). Overall, human CD1d in complex with 2 appears to be slightly more fluctuating compared to H_aGAL simulations. More specifically, the residues on the F′ pocket side display a higher mobility in the RMSF analysis of H_OCH. All these changes are very likely due to the truncated sphingosin chain of 2, even though a spacer lipid complements the F′ pocket in this case. The presence of the ligand 8 in the human CD1d protein causes a pronounced enlargement of the 1D-FEL for dihedrals around the residue 153 (residues 151 to 155 of helix α2). This was strongly observed for all three replicas of the M_AZOL system, compared to the M_aGAL one. This clearly shows that the replacement of an amide function with a triazole group significantly increases the flexibility of the α2 helix and consequently disturbs the OTAN hydrogen bond network, which involves the residues Asp151 and Thr154 (helix α2, human numbering). A polar head destabilization is then expected (discussed hereafter). Concerning the ligand 7, no CD1d structural change has been observed here. In spite of the two aforementioned structural changes, absolutely no CD1d difference has been observed concerning the binding footprint and the inter-helix distance between all these systems, which could have explained a Th2 bias.
We turn now to the discussion about the dynamical features of the polar head. Very interestingly, the three ligands (2, 7, 8), which are known to induce a Th2 bias, display multiple well 3D-FEL (Sy3, Sy12, Sy13 and Sy14 on Figure 8) compared with ligand 1 (Th1 bias in human and mouse CD1d, Sy1 and Sy2 on Figure 8). H_OCH simulations exhibit this feature only once, for replica II. This suggests that at least three trajectories or more are needed to reveal and confirm such a conformational behavior. A common pattern to these three ligands is a new conformational state, which appears about 1.8 kcal.mol−1 above the OTAN state, designated as the “aside” state in Figure 8. The polar head has rotated by about 110° relative to the φz axis in the resulting conformations. In these conformations, the OTAN network and the hydrogen bond with Asp80 are lost, or partially lost. The polar head points now toward the α1 helix but no direct contact has been observed between the sugar and this helix. In fact, the two residues Val72 and His68 of helix α1, which face the polar head, are positioned too far away to attract the ligand. In other words, the source of the new observed state is not helix α1. Thus, it seems that a chemical alteration is able to make accessible the “aside” state, thanks to the CD1d interaction.
It should be noticed that the VDW contact initially made by the hydrophobic side of the sugar with Trp153 in the OTAN state (Figure 1), is also lost in the “aside” state. The new polar head orientation clearly penalizes a TCR recognition process. But this reorientation towards helix α1 is found to be reversible. The lifetime of this state ranges from a few nanoseconds to 30 ns. It takes only a few picoseconds to change from the OTAN state to the “aside” state and vice versa. It is difficult to correlate the occurrence of this transition to specific structural and dynamical features at such a short scale, but obviously, the new state arises in conjunction with the partial or total loss of the OTAN network and the lost of hydrogen-bond between 3′-OH and Asp80. Also, rare and slight rotations around the middle pivot φy can occur. From an energetic point of view, the OTAN state remains the ground state in all our simulations (except in the case of M_SaGAL that will be discussed below). For the H_SaGAL simulations, the “aside” state is also present. But moreover, the free energy landscape shows unstructured features above 2 kcal.mol−1, indicating that all conformations are frequently visited along the “x” axis in a broad basin (Figure 8). In other words, in these high-energy states, the polar head is almost free to move around φx.
Extending this 3D-FEL analysis of the polar head dynamics to all studied systems leads to very interesting results. All the binary systems known to have a Th1 biased response (Sy1, Sy2, Sy6, Sy7, Sy8, Sy9) never show the “aside” state, nor such multiple-well energy landscapes in the (φx, φy, φz) space, for any replica. In contrast, all the other seven loaded CD1d systems (but Sy4 and Sy10) for which a Th2 biased polarization has been demonstrated, are characterized (at least once) by a multiple-well landscape including the specific state above-mentioned. This concerns the systems Sy3, Sy5, Sy11, Sy12 and Sy14 in Table 1. One may question whether it is a pure coincidence. Obviously, despite our very large-scale MD simulations, a more representative statistical sample of MD trajectories is required to further reinforce this trend (typically, more than three replicas would be needed). But, our results are consistent with the accepted model that correlates a Th2 biased response to chemical modulations yielding a less stable CD1d-glycolipid complex. More precisely, we show here that the complex instability results from increased sugar head fluctuations above the CD1d binding groove that will hinder TCR recognition. All the simulations showing an “aside” state are associated with a Th2 response exclusively.
More interesting is the simulation of the M_SaGAL system, which displays a very particular 3D-FEL illustrated at the bottom of Figure 8. All three replicas also reveal multiple energy wells with broad basins and the so-called “aside” specific state is present. But it is the only system for which the OTAN state is not the ground state. Actually, the specific “aside” state is observed to be the lowest one in the 3D-FEL analysis of M_SaGAL, the OTAN state being about 1.2 kcal.mol−1 above it. Furthermore, in our study, it is the only system that displays conformations involving large rotations of the polar head around the middle pivot φy, resulting in points distinctly inside the 3D-FEL box. A typical conformation is illustrated in Figure 8. It is associated to a third distinctive state we call the “flip” state in which the polar head hydrogen bonds to the α1 helix. Let us recall that rotations about this φy axis would normally drive the polar head in contact with the helices. Such motions are clearly hindered in all other studied systems. This means that in this case, the polar head of the thio-galactoside derivative slightly pulls out of the CD1d host during the M_SaGAL simulation. Relative to the C–O, the C–S bond is longer. Combined with a greater separation inter-helix distance observed on the A′ pocket side of mouse CD1d, this is probably the reason why we observe this specific dynamical behavior for the M_SaGal simulation. Obviously, these slightly “extricated” conformations will prevent the TCR binding. This theoretical result fully agrees with biological evaluations [23], [31], [32] of this compound that show no activity against mouse iNKT cells.
In this general analysis, it is to be noted that, although they are known to give a Th2 biased response, the two binary systems Sy4 (M_OCH) and Sy10 (H_GOF) do not display the “aside” state. Additional replicas might be necessary to observe such behavior. However, in our simulations, these two systems with a truncated glycolipid chain (acyl or sphingosin) involve a strong change in the 1D-FEL of dihedrals on the helix α1 bearing the contacts, which are critical for CD1d-TCR recognition. Clearly, anharmonic flat or double-wells appear in the 1D-FEL of these dihedral angles. This suggests that under chemical modulation not only the polar head flexibility but also the deformation of helix α1 itself can deteriorate the TCR recognition process.
When short chain lipids (in analogues) are bound to the protein, spacer lipids may be simultaneously present in the pockets of CD1d. For example, two X-ray structures (PDB references 3ARB and 1Z5L) show the presence of A′ and F′ pockets spacer lipids when CD1d is partially occupied by the sphingosin truncated 2 and the acyl truncated 6 analogues, respectively. In the absence of the glycolipid ligand, endogenous ligands can also fill the entire volume in order to maintain stability and prevent protein denaturation [28]. It was then interesting to examine how the presence or the lack of spacer lipid impacts the polar-head dynamics and the conformations of the CD1d surface.
Firstly, the influence of the presence of a spacer is addressed by comparing the trajectories of the systems Sy3 (with 2+ lipid) and Sy5 (with 3). The two ligands are almost identical, the molecule 3 having only two CH2 more in its acyl chain. The tremendous difference is that in the system Sy3, an additional linear hydrophobic compound C12H26 is present in the F′ pocket in addition to 2. As expected, 2D-RMSD and RMSF analyzing tools indicate a higher fluctuation for the system lacking the spacer lipid. This increased flexibility concerns the CD1d residues in the vicinity of the F′ pocket. Also, in the absence of spacer lipid (Sy5) the inter-helix distance becomes slightly larger (by about 0.3 Å, see Table S1). In addition, the absence of a spacer lipid causes a slightly larger conformational space of the polar head in the 3D representation (φx, φy, φz) of 3 compared to 2 (see Table S2). Both systems reveal the specific “aside” state on their 3D-FEL figure, but the simulation without complementary free lipid displays broad basins with the polar head almost free to move around the φx axis (see H_OCH and H_OCH9 3D-FELs in Figure S5). All these results are consistent with the experimental findings of Garcia et al. [42] who concluded that the spacer lipid appears to work in concert with the ligand to stabilize the binding groove.
How the presence of the polar head impacts on the molecular dynamics was addressed based on the comparison between simulations Sy1 and Sy16 (human CD1d with two headless lipids filling simultaneously pockets A′ and F′). As expected, the observation of RMSF figures shows that fluctuations are more important in the middle of helix α2 when the polar head is missing. But further, fluctuations also tend to affect helix α1 in this case. This shows the importance of the presence of the OTAN hydrogen bond network for stabilizing the binding groove and then the CD1d surface in view of recognition by TCR.
From the analysis of 48 trajectories, it appears that the α1/α2 inter-helix distance differs in mouse and human loaded-CD1d simulations. A greater separation distance is observed on the A′ pocket side of mouse CD1d where strong residue dissimilarities appears between human and mouse helices. This point may be interesting in view of the known differences in cytokine profile production that sometimes appear between human and mouse systems. For the lipid-free CD1d simulations, we observe the spontaneous closure of the binding domain entrance, accordingly with previous theoretical results. [41] In complement, a spacer lipid simultaneously present with the ligand stabilizes the F′ pocket. In a similar manner, we show the key role of the polar head in stabilizing the binding groove at the CD1d surface.
The goal of this study was to get insight into the impact of a chemical variation of molecule 1 on the structure of the binary complex formed between the ligand and protein CD1d. A crucial point was how this could affect the CD1d binding footprint with possible deterioration of the TCR recognition. The major result of our study is that the dynamical behavior of the polar head seems to be a key factor when trying to correlate a ligand modulation with the orientation of the known biased biological response, Th1 or Th2. Considering three successive dihedral degrees of freedom (φx, φy, φz), which govern the polar head rotation above the CD1d binding groove, our simulations permitted to identify three model situations. In the first one, the conformations visited by the polar head during the simulation mainly fall in a portion of the free energy landscape we call the “OTAN” state. It corresponds to the polar head hydrogen-bonded to the CD1d through the well-known H-bond network built up from 2-OH, Thr154, Asp151, and amide-NH. Only a few higher free energy states can appear, but all of these maintain the polar head in contact with helix α2. Sampled conformations are exclusively restricted to well-separated (φx, φz) minima corresponding to rather structured states. A second situation corresponds to the emergence of a specific conformational state in the energy subspace (φx, φz), about 1–2 kcal.mol−1 above the OTAN state. This new “aside” state is characterized by the polar head pointing toward the α1 helix, but without direct interaction with CD1d. It arises in conjunction with the partial or total loss of the OTAN network and the loss of the hydrogen bond between 3′-OH and Asp80. The new polar head orientation clearly penalizes the presentation to TCR. The lifetime of this state ranges from a few nanoseconds to 30 ns. The ligands associated with Th1 biased response never displayed this “aside” state in their free energy landscape in any simulation replica. By contrast, all the simulations showing an “aside” state are associated with a Th2 response exclusively. Obviously, the previous 10 ns simulations of Henon et al. [17] were missing this conformational space. Sampling has been improved here with longer trajectories but also mainly by running several independent simulations, thus exploiting different starting conditions. Three 240 ns trajectories have been produced for each of the 16 systems (0.72 µs each). Only on this time scale and using multiple replicas could the emergence of these specific states be disclosed. But a still more representative statistical sample of MD trajectories would be required to further reinforce our findings. The interesting thing is that our model holds for very different chemical modulations affecting the anomeric bond as well as the polar linker, or either the sphingosin lipid chain. A third model situation arises when the specific “aside” state become the ground state, below the OTAN state. Then, at higher energies, new conformations appear in which the polar head turns upside-down and hydrogen bonds to the helix α1, a situation incompatible with TCR recognition. We observe then broad basins in the 3D free energy landscape of the polar head that must necessarily involve the slight extrication of the sugar head from the cavity entrance, correlated with no biological activity. In this case (thio-analogue of 1), our findings indicate that mouse simulations behave differently than human ones.
The existence of a high-energy but populated “aside” state in the simulations of some analogues of 1 very likely contributes to reduce the stability of the ligand-CD1d binary complex. Binding free energy calculations using molecular dynamics tools could have potentially provide information. However, we presume that the very large number of degrees of freedom in these glycolipids would have prevented obtaining accurate results. Moreover a change in the binary complex stability does not systematically affect the affinity to TCR. Therefore, such binding affinity calulations are certainly not the best way to find out relationships here. The route we have chosen allowed unambiguous identification of these states.
Of course, it would be naive to believe that a molecular model can capture such a complex biological response. The mechanisms by which analogues govern the cytokine profile are multifactorial. For example, Sullivan et al. [43] showed that a critical parameter for a glycolipid to influence the cytokine response is its stability in cells. Overall, our results are consistent with the often-invoked model that correlates a Th2 biased response to chemical modulations yielding a less stable CD1d-glycolipid complex and hence a less stable ternary complex. Additionally, the 1D free energy landscape analysis tool permitted to show that not only the polar head but also modifications of α1 and α2 helical structures could result from chemical variations of molecule 1.
Even though the chemical alterations seem large and significant, the set of molecules studied here have common structural features, what could limit the application of our methodology to other class of analogues. First, chemical alterations that prevent defining the three dihedral angles φx, φy, φz cannot be studied within our procedure. This concerns analogues that derive from specific osidic link variations. Moreover, it is clear that our results here only apply to analogues having a galactose residue. Another sugar group (glucose, …) would likely change the free-energy profiles in a way that cannot be predicted by our results obtained with the galactose component. Furthermore, our methodology only allows checking that preconditions for interaction of the binary complex with the TCR exists. From our study, these requirements for an efficient association are: a polar head dynamics with significantly populated OTAN state, and a limited deformation of helices α1 and α2. But our model (based on the CD1d-ligand system) cannot explicitly handle the interactions between TCR and the binary complex.
However, our polar head 3D-FEL tool combined with the 1D-FEL analysis of CD1d dihedrals in the binary complex provides a structural basis for predicting the very different dynamical behaviors of α-glycosphingolipids in CD1d and might aid in the future design of new analogues of 1.
|
10.1371/journal.pntd.0006957 | Accuracy of the WHO praziquantel dose pole for large-scale community treatment of urogenital schistosomiasis in northern Mozambique: Is it time for an update? | A pioneering strategy developed by the World Health Organization (WHO) for the control of schistosomiasis was the concept of a height-based dose pole to determine praziquantel (PZQ) dosing in large-scale treatment campaigns. However, some recent studies have shown variable accuracy for the dose pole in terms of predicting correct mg/Kg dosing, particularly for treatment of adults. According to the WHO, 91 million adults in 52 countries are targeted to be treated by 2020.
The present study aimed to test the accuracy of the dose pole in determining PZQ dosage by comparing the number of tablets determined by the dose pole with the number of tablets determined according to total body weight. The analysis included height-for-weight data from 9,827 school-aged children (SAC) and adults from 42 villages in the province of Cabo Delgado in Mozambique. The results revealed that of the 7,596 SAC, 91.8% has received an appropriate dose (30-60mg/Kg), 6% received an insufficient dose (<30mg/Kg) and 2% an excessive dose (> 60mg/Kg). On the other hand, 13.7% out of 2,231 adults were treated inaccurately with 13.5% receiving an insufficient dose and 0.2% an excessive dose. When the percentage of insufficient dosing was disaggregated by gender, the frequency of adult females who were underdosed reached 18.3% versus 10.8% of adult males. Of note, Adult females aged 21–55 years were found to have an underdose frequency of 21.3%, compared to 11.8% of adult males in the same age range. The performance of a proposed modified dose pole was compared using the same dataset of adult Mozambicans. The results showed that the modified dose pole reduced the underdose frequency among adults from 13.5% to 10.4%, and subsequently increased the percentage of optimal dosing from 33.7% to 45.3%.
Our findings highlight the need to update the WHO-dose pole to avoid administration of insufficient PZQ doses to adults and therefore minimize the potential emergence of PZQ-resistant strains.
International Standard Randomized Controlled Trial registry under ISRTC number 14117624
| Schistosomiasis control is currently based on large-scale, preventive chemotherapy treatment with praziquantel (PZQ), and this has resulted in a significant reduction in its prevalence and associated morbidity worldwide. Recommended PZQ dosing is 40–60 mg/Kg body weight, but to facilitate large-scale drug administration programs, a dose pole to estimate PZQ doses based on height has been developed by WHO, as height has been shown to accurately correlate with body weight among children in endemic areas. WHO estimates that 206 million people, including 91 million adults, will need to be treated annually across 52 countries by 2020. Published literature indicates that the dose pole has a satisfactory accuracy for children, however it seems to deliver an insufficient dose (<30mg/Kg body weight) to approximately 20% of adults and overweight children. This paper evaluated the accuracy of the PZQ dose pole for the large-scale treatment of urogenital schistosomiasis in Mozambique and discusses the necessity for an update of the current WHO-dose pole based on modifications recommended in recent literature. Preventing underdosing among adults and overweight children should improve treatment outcomes and minimize the potential emergence of PZQ-resistant strains.
| Human schistosomiasis is an acute and chronic neglected tropical disease (NTD) caused by six species of parasitic blood flukes of genus Schistosoma: S. guineensis, S. haematobium, S. intercalatum, S. japonicum, S. mansoni and S. mekongi. Together, they affect about 250 million people worldwide, with more than 779 million people are at risk of infection [1, 2]. Despite being endemic in 52 countries around the world, approximately 91% of schistosomiasis prevalence is concentrated in Africa, causing severe morbidity, mainly in school-aged children (SAC)[3].
For almost 20 years, schistosomiasis control has been based on large-scale, repeated, preventive chemotherapy treatment with praziquantel (PZQ), with an optimal dose of 40-60mg/Kg bodyweight. The drug is recommended due its safety, acceptable cure rate in low intensity infection, and its currently low cost [4]. However, Taylor et al. [5] and King, et al. [6] demonstrated that PZQ may have similar effects in slightly lower doses, therefore, a single dose between 30-60mg/Kg has been considered acceptable for schistosomiasis preventive chemotherapy. On the other hand, others have shown variation in the schistosome sensitivity to PZQ in different endemic regions [7, 8]. The current World Health Organization (WHO) guideline for schistosomiasis preventive chemotherapy recommends annual treatment of all SAC as well as at-risk adults in areas with >50% prevalence, treatment of SAC every two years (school-based treatment) as well as high-risk adults in areas with 10–50% prevalence, and treatment of SAC twice during primary school years in areas with 1–10% prevalence [9].
A part of the mass drug administration (MDA) strategy recommended by WHO is the use of the height-based dose pole to estimate appropriate PZQ doses. This low-technology approach is field-applicable and suitable for rapid, large-scale implementation of treatment [10]. In the early 2000s, when the WHO dose pole for schistosomiasis was developed, several studies demonstrated that it could determine an appropriate dose between 30-60mg/Kg in more than 98% of SAC [10–12]. However, recent studies have found a questionable accuracy for the dose pole in determining correct dosage for adults. In South Africa, for example, 34.6% of adult females (16–23 years old) received an insufficient dose of PZQ based on dose pole recommendations [13]. Another study, which evaluated 5,614 rural adult Zimbabweans (15–49 years old), showed a 15% risk of inadequate treatment (< 30mg/Kg) for adults [14]. Interestingly, a common aspect between these two studies was an elevated prevalence of obesity among adults; 40.9% in South Africa and 19.4% in Zimbabwe.
Between 2011 and 2015, a Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) treatment project [15] was carried out in Mozambique [16, 17]. This longitudinal study compared the impact of different treatment strategies in study areas having prevalence of ≥ 21% for urogenital schistosomiasis (S. haematobium) among SAC. The primary goal of SCORE project was to gain and sustain low levels of infection through preventive chemotherapy. In different study arms, treatment was given in schools or in communities or both to evaluate the impact of alternative treatment approaches [17]. In the final and the fifth year of this study, data were additionally collected in some villages to evaluate the accuracy of the WHO-dose pole for both SAC and adults.
This study was performed as part of the SCORE project in Mozambique, which was registered with the International Standard Randomized Controlled Trial registry under ISRTC number 14117624 for Mozambique. Informed written consent was obtained from all individuals ≥18 years of age and from parents or legal guardians of children less than 18 years of age. The purpose of the study was explained to all school children and verbal assent was obtained from the children. Permission was also obtained from school headmasters. Ethical clearance was obtained from the National Bioethical Committee for Health of Mozambique (NBCHM) [Comitê Nacional de Bioética para a Saúde (CNBS)], and the survey was conducted according to CNBS guidelines (reference no. IRB00002657). The study protocol was also approved by Imperial College London (ICREC_10_2_2).
This research was conducted during the preventive chemotherapy phase of the SCORE study conducted among SAC and adults from 42 communities in five districts of Cabo Delgado province, Northern Mozambique (Fig 1) [16, 17]. Individuals were assessed in the school or at the household level during treatment whereby demographic and physical data (age, sex, height, and weight) were collected. All villagers were treated for schistosomiasis using the height-based dose pole. Weight was measured using a body digital scale (accuracy resolution 0.1Kg), calibrated according to the manufacturer’s recommendation (Tian Shan, China). In total, 10,611 individuals were surveyed. The WHO AnthroPlus software (WHO) was used to flag erroneous height/weight readings. 9,827 individuals’ data for height and weight, including SAC and adults (Fig 2), were included for analysis. Statistical analysis was performed using the statistical language R (version 3.3) (Lucent Technologies, USA). The package ‘tidyverse’ was used for data cleaning and summary statistics and ‘lme4’ for specifying generalized linear mixed models of the data. For the statistical comparison among the groups, we assessed the likelihood of underdosing by creating a binary variable, which was 0 for appropriate dose (as given by use of the WHO dose pole) and 1 for insufficient dose. This was specified as the dependent variable in a binomial generalized linear mixed model (GLMM), with ‘logit’ link function. Sex (m/f) and age group (5 to 8, 9 to 12, 13 to 15, 16 to 20, 21 to 55 and 56 to 95) were included in the model as fixed effects, and village and district were included as random effects to account for geographic clustering effects.
In 2014, Palha de Souza, et al. [14] developed a modified dose pole by adding two height/dose intervals to correct for potential underdosing in adults. The new height cuts-offs of 156 cm and 164 cm allow for dosing of 3.5 tablets and 4.5 tablets, respectively.
For our secondary analysis, the relative accuracy of the standard and modified dose poles was calculated in terms of their capacity to provide a recommended dose between 40-60mg/Kg, considered to be optimal; 30-40mg/Kg, considered to be acceptable; < 30mg/Kg, considered to be insufficient; and finally, > 60mg/Kg, considered to be excessive. An appropriate dose was considered to be between 30-60mg/Kg. Moreover, the modified-dose pole results were also analyzed in combination with a body mass index (BMI) adjustment. For this analysis, adults with BMI > 25 kg/m2 were estimated to require an additional 25% of the average adult dose (2,400 mg), translating to one extra 600 mg tablet, as suggested by the modified-dose pole developers.
Of the total 9,827 individuals surveyed in northern Mozambique, 7,596 were SAC from 5 to 15 years old (4,141 male and 3,545 female), and 2,231 were adults between 16 to 95 years old (1,428 male and 803 female). As presented in Table 1, among 7,596 SAC, a total of 54.7% received an optimal dose (40-60mg/Kg) of praziquantel using the WHO dose pole, and 37.1% received a lower, but acceptable dose (30-40mg/Kg). Thus, 91.8% of the students in school surveys received what is considered an appropriate dose (30-60mg/Kg). 6.2% of the students received an insufficient praziquantel dose (< 30mg/Kg) whereas 2% received an excessive dose (> 60mg/Kg). Among the 7,596 SAC, the average and median doses were 41.4 mg/Kg and 41.2 mg/Kg respectively, and the minimum and maximum doses administered by the dose pole were 16.9 mg/Kg and 77.9 mg/Kg, respectively (Table 1).
When the performance of the dose pole was evaluated among 2,231 adults, it was found that 13.5% received an insufficient dose. When the percentage of insufficient dosing was disaggregated by gender, the frequency of adult females who were underdosed was 18.3%, in contrast to 10.8% for adult males (p = 0.005, Table 2). Ultimately, the most substantial difference by gender was found among adults aged 21 to 55 years (p < 0.001, Table 2), wherein 21.3% of females and 15% of males were underdosed. Among the adults, only 30% received an optimal dose (40–60 mg/Kg) of PZQ using the WHO dose pole, 51% received an acceptable dose (30–40 mg/Kg), and 5% an excessive dose (> 60mg/Kg). Among adults, the average and median doses were 37.3 mg/Kg and 37.0 mg/Kg, respectively, and the minimum and maximum doses administered by the dose pole were 15.3 mg/Kg and 63.1 mg/Kg, respectively.
When frequency of overweight/obesity (Body Mass Index (BMI) > 25.0) was evaluated among the Mozambican adults, the overall analysis revealed that 333 out of 2,231 adults (14.9%) could be classified as overweighed/obese. The peak rates of overweight among adults were among females aged between 21 and 55 years (23.5%) (Table 3).
Next, the performance of the WHO-dose pole was compared to that of the modified dose pole proposed by Palha de Souza and collaborators [14], using our database of Mozambican adult anthropometrics (Table 4). This modified dose pole was specifically developed to address the potential for underdosing in adults posed by the WHO dose pole. To re-study this problem, we used the height and weight data collected to determine the dosing prescribed by a modified dose pole and assess whether this would provide appropriate or insufficient doses (> 30g/kg or < 30g/kg). We created a binary variable for underdosing (0 for appropriate dose, 1 for underdose/insufficient), the dataset was filtered to consider only adults aged 16 and older, and the data were stacked so that there were two observations for each individual: One for insufficient dose (0 = appropriate /1 = insufficient) with the WHO dose pole, and one for insufficient dose (0 = appropriate /1 = insufficient) with the modified dose pole. We specified a GLMM with insufficient dose as the dependent variable. Fixed effects were sex, age group, and type of dose pole (WHO or modified). Individual ID was included as a random effect (variances from village and district level were accounted for by the individual ID random effect). The model indicated that the WHO dose pole was significantly more likely to result in underdosing for adult’s treatment than the modified dose pole (Table 5).
The performance of the modified dose pole in reducing the underdosing of Mozambican adults was to decrease its frequency from 13.5% to 10.4%. In addition, it markedly increased the percentage of optimal dosing (40-60mg/Kg) from 33.7% to 45.3%. On the other hand, the modified dose pole combined with a recommended BMI correction adjustment did not provide any improvement in in reducing the number of potential underdose treatments (10.4% vs 9.8%).
Schistosomiasis control is based on large-scale, repeated, and preventive chemotherapy treatment with PZQ, which has resulted in a significant reduction in its prevalence and associated morbidity worldwide. The WHO PZQ dose pole represents one of the key advances in the global initiative for schistosomiasis control. However, although it has a satisfactory accuracy for SAC treatment, it was not necessarily designed for accurate dosing of adults.
Currently, in some countries of sub-Saharan Africa, there are still reports of high endemicity areas for schistosomiasis (>50% of prevalence), requiring treatment of high-risk adults. [16]. The accuracy of the standard dose pole has been extensively tested for SAC treatment, showing very satisfactory effectiveness in delivering correct doses for 95% of SAC treated [10, 12]. Although it has satisfactory performance for SAC, there are discrepancies in efficiency in its current format, with much reduced performance in accuracy amongst adults [13, 14]. This is problematic, because adults are now identified as an important target for the control and elimination of schistosomiasis in highly endemic areas because of their contribution in the maintenance of active transmission within communities [23]. Moreover, although not assessed in this study, Recently studies have suggested a high risk of Schistosoma infections in early life [18], suggesting the need for inclusion of children ≤6 years in MDA programs. Coulibaly et al. [19] have reported the efficacy and safety of different dosages of praziquantel in preschool children, whereby there was a 72% cure rate with the 40 mg/kg dose and therefore endorsed preventive chemotherapy programs in children younger than 5 years of age. To include infants and PSAC in the MDA programs, an extended tablet pole was proposed, with two more height intervals: 60-83cm for ½ tablet and 83-99cm for ¾ tablet, and an upward revision of the 94cm threshold to 99cm [20, 21, 22]. This adaptation should lead to a new accuracy of 95.4% for all children (including infants), yielding an appropriate dose (30-60mg/Kg), with only 1.6% and 3% receiving an insufficient, or an excessive dose respectively [22].
Our study’s data collected in northern Mozambique supports the findings of other studies performed in South Africa and Zimbabwe [13, 14] that have suggested a need to implement an alternative way to deliver PZQ to adults on a mass scale. In 2007, WHO performed field-testing PZQ dosage for the treatment of opisthorchiasis in Lao PDR, comparing the accuracy of the dose pole with bathroom scales [24]. Using the Lao population’s height-for-weight data, and extrapolating it for the treatment of schistosomiasis, the study revealed that 18.7% of the adults (aged >15 years) would have received an insufficient dose of 30mg/Kg of PZQ using the dose pole. This is in line with the 10–20% dose pole under dosage for schistosomiasis that we and other have found among adults. In Mozambique, approximately 21% of adult females from 21 to 55 years old were found to have been underdosed. In light of the fact that several countries have conducted multiple rounds of MDA and are now moving towards potential elimination of schistosomiasis, this underdosing of an important subgroup of people may represent a significant risk for persistence of schistosomiasis. Considering their daily occupational exposures, such as collecting water, bathing, and washing in open water bodies, adults significantly contribute to the transmission potential in a community [17]. The results from the SCORE project in 150 villages in the province of Cabo Delgado in Mozambique in 2011, demonstrated that 44.8% of the surveyed adults were infected with S. haematobium, and out of the 44.8% infected, 7.1% were heavily infected, expelling ≥ 50 eggs per 10mL of urine [16]. This group of individuals must not be neglected by MDA programs because they may represent a key factor in maintaining transmission, while they are also in danger of developing advanced forms of Schistosoma-associated morbidity. In the Mozambique study villages, among 4,154 adults randomly surveyed in a knowledge, attitudes, and practices (KAP) survey, 3,971 (95.6%) stated that farming was their main occupation, washing (91.3%) and bathing (86.7%) in open water source were common practices, as was open defecation (83.3%) [17].
A common aspect in the dose pole studies in South Africa, Zimbabwe, Lao PDR, and Mozambique was the percentage of overweight and obese adults (BMI > 25), which occurred among 40.9%, 19.4%, 19% and 14.9% of adults, respectively. Adult females surveyed in Mozambique with an age range between 21 and 55 years demonstrated a peak of overweight/obesity of 23.5%. As reviewed by Ng and collaborators [25], the global prevalence of obesity has increased, affecting some countries more than others. According to the WHO [26], in 2016, more than 1.9 billion adults, 18 years and older, were overweight. Of these, over 650 million were obese. Thus, in affected countries, the accuracy of the height-based WHO dose pole for weight-adjusted schistosomiasis treatment might be dramatically reduced, because it functions under the presumption that a patient’s height and weight are closely correlated.
The resolution for this issue suggested by Palha de Souza et al. [14], and later supported by Baan et al. [13], was to modify the current dose pole by adding two height/dose intervals. Their modified dose pole was aimed at reducing underdosing in adults. Moreover, the authors also suggested a simple adjustment in the PZQ dose for individuals with overweight or obesity. These individuals, who could be classified by the community drug distributors using a pictogram [27], would require an additional 25% of the average adult PZQ dose (2,400mg), translating to one extra 600mg tablet. The use of the modified dose pole would have reduced the proportion of adult Zimbabweans receiving insufficient PZQ doses (< 30 mg/Kg), from 15% to 8.7%, and ultimately to 1.4% when combined with the BMI correction adjustment. Among South-African female students, underdosing would be reduced from 27% to 20%, and finally to 3.4% by adding the BMI correction. In the present study, use of the modified dose pole would have reduced underdosing in Mozambican adults from 13.5% to 10.4%, with a significant reduction from 21.3% to 15.5% among adult females between 21 and 55 years old. The reduction by the modified dose associated with BMI correction adjustment was not as great as in South Africa and in Zimbabwe, which could be explained by a lower rate of obesity among adults in Mozambique. Whereas concurrent food intake is known to affect PZQ uptake and ultimate blood levels [28], the impact of patient obesity (or pregnancy) on drug uptake and peak blood levels remains unknown. This will be a useful area for research in the near future.
In summary, particularly in the current climate where countries are moving towards elimination and community-wide preventive chemotherapy against schistosomiasis, we recommend a validation, based on an assessment of PZQ plasma concentrations, of the dosing recommended by the WHO standard dose pole. If significant underdosing is detected, to avoid large-scale administration of insufficient PZQ and its potential for emergence of drug-resistant strains, the WHO dose pole should be modified considering the implementation of a universal dose pole for all age groups to deliver praziquantel in endemic areas for schistosomiasis (Fig 3). Whilst we make this recommendations for a modified dose pole, we are similarly cognizant of the increasing burden for PZQ production that expansion to community-wide treatment programs is putting on pharmaceutical companies and donors alike. This should not however, take away from the need for accuracy particularly in the move towards elimination.
|
10.1371/journal.pcbi.0030007 | Large-Scale Discovery of Promoter Motifs in Drosophila melanogaster | A key step in understanding gene regulation is to identify the repertoire of transcription factor binding motifs (TFBMs) that form the building blocks of promoters and other regulatory elements. Identifying these experimentally is very laborious, and the number of TFBMs discovered remains relatively small, especially when compared with the hundreds of transcription factor genes predicted in metazoan genomes. We have used a recently developed statistical motif discovery approach, NestedMICA, to detect candidate TFBMs from a large set of Drosophila melanogaster promoter regions. Of the 120 motifs inferred in our initial analysis, 25 were statistically significant matches to previously reported motifs, while 87 appeared to be novel. Analysis of sequence conservation and motif positioning suggested that the great majority of these discovered motifs are predictive of functional elements in the genome. Many motifs showed associations with specific patterns of gene expression in the D. melanogaster embryo, and we were able to obtain confident annotation of expression patterns for 25 of our motifs, including eight of the novel motifs. The motifs are available through Tiffin, a new database of DNA sequence motifs. We have discovered many new motifs that are overrepresented in D. melanogaster promoter regions, and offer several independent lines of evidence that these are novel TFBMs. Our motif dictionary provides a solid foundation for further investigation of regulatory elements in Drosophila, and demonstrates techniques that should be applicable in other species. We suggest that further improvements in computational motif discovery should narrow the gap between the set of known motifs and the total number of transcription factors in metazoan genomes.
| In contrast to the genomic sequences that encode proteins, little is known about the regulatory elements that instruct the cell as to when and where a given gene should be active. Regulatory elements are thought to consist of clusters of short DNA words (motifs), each of which acts as a binding site for sequence-specific DNA binding protein. Thus, building a comprehensive dictionary of such motifs is an important step towards a broader understanding of gene regulation. Using the recently published NestedMICA method for detecting overrepresented motifs in a set of sequences, we build a dictionary of 120 motifs from regulatory sequences in the fruitfly genome, 87 of which are novel. Analysis of positional biases, conservation across species, and association with specific patterns of gene expression in fruitfly embryos suggest that the great majority of these newly discovered motifs represent functional regulatory elements. In addition to providing an initial motif dictionary for one of the most intensively studied model organisms, this work provides an analytical framework for the comprehensive discovery of regulatory motifs in complex animal genomes.
| Essentially complete sequences of metazoan genomes have now been available for nearly ten years, and in that time considerable progress has been made towards annotation of their best-known features, the protein coding genes. Computational pipelines such as that run by Ensembl [1] provide automated annotation of most protein-coding genes, which for some genomes have been augmented by manual curation to improve the accuracy and completeness of gene sets: examples of this approach include the Vega database of vertebrate annotation [2] and some popular model organism databases such as Wormbase [3] and FlyBase [4].
Annotation of other functional genomic features—notably the sequences responsible for regulating gene transcription—has lagged behind. Regulatory elements can be broadly divided into two classes: proximal or core promoter elements that occur close to the initiation site of transcription, and enhancer/silencer elements that act at distance to regulate basal levels of transcription. Both classes of regulatory elements consist of clusters of transcription factor binding sites (TFBSs) [5]. This common architecture suggests that a first step towards regulatory element annotation should be to define a dictionary of motifs that reflects the full repertoire of transcription factor binding specificities. Classically, the binding specificity of transcription factors can be identified using data compiled from DNase I footprinting [6] or in vitro binding site selection experiments [7]; however, data of this kind is only available for a limited subset of transcription factors. For example, just over ten percent of 753 candidate transcription factors in the D. melanogaster genome [8] have annotated binding site data [9]. Moreover, in many cases only one or two sites have been annotated for a given protein, making it hard to build a reasonable model of a factor's binding specificity. Computational methods have existed for more than twenty years that can identify overrepresented motifs in a set of sequences (reviewed in [10]). However, the problem of inferring transcription factor specificity remains intrinsically challenging even for small, well-defined datasets of functionally characterized binding sites [11]. On a genome-wide scale, computational motif discovery methods have typically been applied to 5′ flanking regions of genes grouped by similar expression patterns, with the aim of discovering one or a few factors responsible for controlling coregulated expression. Applying computational motif finders to large sets of unrelated promoter regions from a single genome is a much more challenging task, and previous work in well-studied systems such as Drosophila has yielded only a relatively small set of core promoter motifs [12]. If we aim to build a comprehensive motif dictionary for metazoan genomes, it is necessary to scale up the motif discovery process and identify much larger sets of motifs.
Here, we describe the application of NestedMICA [13], a sensitive new computational motif finder, to a large set of D. melanogaster promoter regions. Using this new method, we have so far discovered 120 distinct overrepresented motifs, including good matches to previously reported transcription factor binding motifs (TFBMs) as well as many novel putative motifs. An important feature of our strategy is that the dictionary of motifs is inferred purely from sequence fragments selected from a single genome on the basis of gene annotation (which itself is supported primarily by the alignment of cDNA and EST evidence to the genome sequence). No gene expression or comparative genomic data is used in the selection of motif-discovery input data or in the motif discovery process itself. The latter is particularly important since, in contrast to previous large-scale motif inference efforts [14–17], it means we can assess the quality of the discovered motif set by evaluation against comparative genomic data. Such comparisons offer strong supporting evidence that many of the motifs we have discovered are biologically significant.
The first step in computational motif discovery is to define a good set of search regions. One strategy would be to focus on the most highly conserved non protein-coding portions of the genome, in the expectation that these would be enriched for regulatory elements [18]. However, we chose to avoid this approach, at least for a first round of motif discovery, since the use of comparative genomic data at this point would prevent its use as an independent source of information when we validate our discovered motif set. Instead, we took a more traditional approach of focusing on a set of presumed proximal promoter sequences. To do this, we extracted up to 200 bases of sequence flanking the 5′ ends of annotated genes on D. melanogaster chromosome arm 2L, with some special treatment for very closely spaced genes as described in the Materials and Methods section. Large tracts of low-complexity sequence, such as mononucleotide and dinucleotide repeats, were masked using the dust program (R. Tatusov and D. J. Lipman, unpublished data) with default options. No other preprocessing of the sequences was performed.
In total, this procedure yielded 422 kb of putative promoter sequence from 2,424 genes. Seventy-six percent of genes have annotated UTRs, and we assume that we have obtained true 5′ flanking sequence for most of these. Many D. melanogaster 5′ UTRs are fairly short, with 66% of UTRs less than 200 bases long, so even for the 24% of genes without an annotated UTR, we expect that our set will include at least some 5′ flanking sequence in many cases. Less than 0.7% of promoter regions in this dataset contain a transposable element repeat [19], so it is unlikely that motifs in transposable element sequences contribute strongly to the results presented here. These data represent more than a 2-fold increase in amount of sequence, and a 25% increase in the number of genes analyzed, relative to the primary dataset in [12]. We also note that in contrast to Ohler et al. [12], who investigated motifs on the leading strand from −60 to +40 relative to the transcription start site using MEME [20], we investigated the presence of TFBMs on both strands from −200 to −1.
Our motif discovery strategy was based on the NestedMICA method [13]. NestedMICA is a probabilistic motif finder: it models motifs as position-weight matrices (PWMs) rather than as consensus sequences. PWMs are an established way of modeling the specificity of molecules that interact with nucleic acids [21] and have been shown to be more powerful than simpler representations such as consensus sequences when detecting TFBSs [22].
NestedMICA infers multiple motifs simultaneously. This is distinct from many previous probabilistic motif finders, which have adopted a stepwise approach: finding one motif, masking its occurrences, then finding the next (e.g., [20]). In this regard, NestedMICA shares affinity with methods that perform simultaneous inference of multiple motifs such as the Gibbs Recursive Sampler [23] and CisModule [24]; however, there are no published reports applying these methods to the genome-wide discovery of large numbers of TFBMs in metazoans. Simultaneous motif discovery is likely to maximize sensitivity (see [13]), and it also contributes to the good scalability of the method, since the program does not need to restart the analysis for each additional motif.
NestedMICA is also distinctive in terms of its inference strategy: while previous probabilistic motif finders have used expectation maximization or traditional Monte Carlo methods such as Gibbs Sampling to parameterize their probabilistic models, NestedMICA uses a recent and distinctive Monte Carlo strategy called Nested Sampling (J. Skilling, unpublished manuscripts at http://www.inference.phy.cam.ac.uk/bayesys). This strategy was chosen because we have found it to be effective at finding globally good solutions to the motif-inference problem without requiring heuristics to choose good starting states (see [13]). Another interesting property of Nested Sampling is that it can provide reliable estimates of the evidence term of a Bayesian computation. Bayesian evidence has historically been very hard to calculate, but can be used to perform model comparison (for example, “is this set of sequence data best modeled by a 10-base or an 11-base PWM”) in a manner that correctly penalizes the extra parameters in more complex models [25]. We took advantage of these evidence estimates in the refinement step of our motif discovery pipeline.
While the primary aims in developing NestedMICA were sensitivity and statistical rigor, we also worked hard to maximize performance and scalability. NestedMICA can run on large volumes of sequence data (up to several megabases), and while the run time on large datasets can still be high, this can be made manageable by running the program in a distributed mode that spreads the workload across several machines connected by a fast network. Depending on dataset size and exact configuration, NestedMICA can effectively utilize 10–20 CPUs.
Effective motif-finding strategies require an appropriate background model against which to assess motif overrepresentation. D. melanogaster upstream sequences are known to have compositional biases [26], and it is important that the background model does a good job of capturing these biases. Many motif finders model the background sequences using a single Markov process or, equivalently, a single oligonucleotide frequency table. Where this strategy has been adopted, high-order Markov processes generally give the best results: Thijs et al. recommended a 5th-order (hexanucleotide) model [27]. However, such a model is complex (3,072 free parameters for a 5th-order Markov chain) and potentially hard to train: biologically meaningful regulatory motifs could be captured by such a high-order background model, preventing their detection in a subsequent motif inference step, and therefore it would be necessary to select truly nonfunctional sequences for background model training. An alternative approach is to relax the assumption that the sequence is generated by a single Markov chain. NestedMICA implements a family of background models where each base of the sequence is generated by one of several possible Markov chains. We call these mosaic models, since they treat large sequences as mosaics of compositionally distinct regions. We have previously shown that a mosaic of four order chains can better model mammalian promoter sequences than a single higher-order chain, while requiring fewer free parameters [13]. We used a similar strategy here and randomly split the set of 5′ flanking sequences on chromosome arm 2L in half to give independent “test” and “training” sets, then used the training portion to optimize a range of background models—with between two and eight classes—using the makemosaicbg program from the NestedMICA package. The results were very similar to those shown in [13], except that on these D. melanogaster sequences the optimal model consisted of six order classes. We selected this six-class model as the basis for our large-scale motif inference. As in the mammalian case, we saw classes modeling neutral, purine-rich, and pyrimidine-rich regions. We also saw Drosophila-specific classes for A/T-rich, C/A-rich, and G/T-rich regions. There was no equivalent of the mammalian G/C-rich sequence class, which associated primarily with CpG islands.
For the analysis presented here, we inferred 120 motifs (Figure S1, statistics also included in Table S1) from the chromosome arm 2L 5′ flanking sequences. The NestedMICA method currently requires that the motif length be specified a priori. For this initial production run, we requested 12 base motifs: long enough to represent the core length of most known Drosophila TFBMs. Motif inference, following the procedure under Materials and Methods, took approximately four weeks on eight Pentium IV processors (2.8 GHz clock speed). We have derived several other sets of motifs, from the same set of chromosome arm 2L sequences with slightly different NestedMICA parameters as well as also from other D. melanogaster chromosome arms. Using the motif comparison strategy described in the Materials and Methods section, we always see a large overlap between independently trained sets of motifs (unpublished data). Discovery and refinement of the motif dictionary is an ongoing process, and in the future we plan to scale up the procedures described here and explore strategies for merging overlapping motif sets to produce a single comprehensive set of TFBMs.
All the initially discovered PWMs were 12 bases long, but in many cases several positions towards the edge of the motif had very low information content, suggesting that the PWM was a model of an underlying motif less than 12 bases long. Therefore, we individually retrained each motif, following the refinement and trimming procedure under Materials and Methods. In 54 out of 120 cases, this led to a shorter final PWM. Finally, in six cases where we were confident that the discovered PWM correspondended to a previously reported, named motif (as discussed at length below), we chose to reverse-complement the inferred motif PWM to correspond with the previously reported orientation. None of our subsequent analyses are sensitive to motif orientation, so this manipulation should have no effect except in terms of motif display.
The motif dictionary contained many motifs with specificity towards A/T rich sequence: 54 out of 120 motifs preferred to match sequences that are more than 66% A/T. This was not due to inability of our motif inference strategy to find G/C-rich motifs: there are several motifs in the set with very strong preferences towards G/C rich sequences. It is possible that there might have been some bias in our inference procedure that leads to preferential detection of A/T rich motifs, but we doubt this explanation: noncoding D. melanogaster sequence as a whole is A/T rich (61.6% A/T on average across our set of 5′ flanking regions), so it would not be too surprising to find that many of the most common regulatory elements might also be A/T rich. Also, our background model was trained on the same set of 5′ flanking sequences and contains a class that modeled regions of sequence with a high A/T content. Thus, the overall high A/T content in 5′ flanking regions was accounted for during the motif inference process, which searched for overrepresented motifs relative to this background model.
We attempted to define an optimal score cutoff to use when scanning bulk sequence with each of our inferred PWMs. To do this, we scanned the training set of 5′ flanking sequences as described under Materials and Methods, and subdivided the matches by score. For each 1-bit score interval (bin), we assessed overrepresentation of motif matches relative to what might be expected given our mosaic background model (see Materials and Methods). We saw significant (p ≤ 0.05 in a binomial test) overrepresentation in high-scoring bins for 118/120 motifs, and were thus able to select a score cutoff. One limitation of this method is that its resolution is limited to the width of the bins used to subdivide the matches—in this case 1 bit. Narrower bins might help, but would reduce the power of the statistical tests used to assess overrepresentation. Applied to our motif set, this method suggested cutoffs between −1 and −8 bits relative to the maximum possible score for each motif. This wide variability implies that using a common threshold for all motifs would not be optimal, and also suggests that some DNA-binding factors might be more tolerant of variation in their binding sites than others. We use these suggested cutoffs in several of our subsequent analyses. It is worth noting that this procedure also indicates that all but two of our inferred PWMs are significantly overrepresented relative to the background model. This should not be surprising—overrepresentation was, after all, the criterion for motif inference—but it does give us additional confidence in the inferred motif set. It also suggests that the two motifs that weren't significantly overrepresented (TIFDMEM0000001 and TIFDMEM0000043) should be treated with some caution.
All our discovered PWMs are also available in release 1.2 of Tiffin, a database of sequence motifs. Tiffin can be browsed via a web interface, which also permits export of the motif PWMs in a variety of machine-readable formats: http://servlet.sanger.ac.uk/tiffin.
Tiffin assigns IDs to computationally discovered motifs in the same spirit as the IDs assigned to Ensembl gene predictions. As with Ensembl, the intention is to maintain IDs wherever possible as the motif collection grows and improves. We use Tiffin IDs throughout this paper when referring to specific motifs in our discovered set. A typical Tiffin ID is TIFDMEM0000040. This is made up of a database identifier (TIF for Tiffin), a three-letter species code (DME for D. melanogaster), a one-letter object type (M for motif), and finally a numerical identifier. This syntax closely matches that used by Ensembl, and where possible Tiffin will use the same species codes.
Given that little is currently known about the binding specificities of most transcription factors, it is difficult to evaluate whether discovered motifs are indeed authentic. Here, we consider four distinct lines of evidence: comparison with previously reported sequence motifs, analysis of cross-species conservation, analysis of motif position in the genome, and association between motifs and gene expression pattern. All four of these analyses offer support for a substantial fraction of our discovered motifs, suggesting that our motif-discovery approach can successfully recover motifs that are predictive of functional sequences. Alongside the comparison step, we also address issues of possible redundancy within the motif collection.
The pioneering study by Ohler et al. (2002) demonstrated the possibility of large-scale probabilistic promoter motif inference in metazoans, and generated a list of top ten motif PWMs that are overrepresented in D. melanogaster core promoter regions [12]. We assessed whether this limited set of highly abundant motifs could be recovered by NestedMICA while simultaneously searching for a much larger set of motifs. To do this, we measured the divergence between each of the ten reported promoter motifs and each of the 120 motifs discovered by our strategy, using the divergence function described in the Materials and Methods section. We then searched the resulting divergence matrix for best reciprocal hits: pairs of motifs where each is the others' best match. This is closely analogous to the strategy used to define orthologous genes between two genomes. Finally, we assessed the statistical significance of each match by repeating the comparison using shuffled PWMs: the fraction of cases where a shuffled PWM can give an equal or better score gives an empirical p-value for the comparison. We found reciprocal best matches for eight out of ten of the Ohler et al. (2002) motifs (Figure 1), including well-established promoter sequences such as the TATA, DRE, and INR motifs. All eight of these matches were highly significant (p ≤ 0.001) In seven out of ten cases, visual inspection leaves little doubt that the motifs are essentially identical, while the final and most divergent case (TIFDMEM0000057 versus Motif 8) shows some differences. Both of the previously discovered motifs with no best reciprocal match in our set (DPE and MTE) have been shown to be located primarily downstream of the transcription start site [12], so it is not surprising that we did not find them in our set. We note that an independent set of 15 consensus motifs derived from positionally biased octomers in Drosophila promoters recently reported by FitzGerald et al. (2006) [26] overlaps substantially with the Ohler et al. (2002) set and therefore also with the subset of NestedMICA motifs in Figure 1.
For a subset of developmentally regulated transcription factors in Drosophila, a reasonable amount of experimental evidence from SELEX-like methods [7] and DNase I footprinting assays [6] is available to infer their binding specificities. From such empirical data, it is possible to derive PWMs that should be a good reflection of the binding specificity of the protein in question, at least in vitro. However, it is important to note that the motifs learned from in vitro binding of purified protein to naked DNA may not accurately represent those obtained from in vivo conditions. We expect motifs inferred directly from genomic sequences to differ slightly from in vitro sequences, and in some cases may better reflect the in vivo binding specificity of a transcription factor.
For this analysis, we used two reference collections of experimentally supported PWMs. The first is a set of PWMs we have learned from the FlyReg database of DNase I footprints [9]. For each factor in the FlyReg dataset with at least five footprints, we attempted to infer a single optimal PWM using NestedMICA, as described in the Materials and Methods section. From an original set of 52 factors, we obtained a set of 30 optimal PWMs for known Drosophila transcription factors (Figure S2) that we can compare to the set of 120 motifs learned from bulk genomic DNA. The second dataset used for evaluation is the JASPAR CORE collection of PWMs [28], which are derived primarily from SELEX experiments or other compiled experimental data. This set includes 123 motifs (database accessed 05/07/2006) from a variety of species, including some from Drosophila, but also many vertebrate, yeast, and plant motifs. Many additional Drosophila motifs exist in the literature that are not present in JASPAR, and thus we extended the JASPAR CORE set of motifs to include an additional 49 SELEX and consensus motifs for Drosophila transcription factors curated from primary publications (http://bioinf.man.ac.uk/bergman/data/motifs). This resulted in a set of 172 motifs of which 62 are derived from Drosophila transcription factors (Figure S3).
We performed a reciprocal-best-hits assignment between 120 discovered motifs and both of these known TFBM sets, using the same divergence measure and significance-testing procedure as before. We applied a significance threshold of p ≤ 0.05 for all comparisons. As shown in Figures 2 and 3, we see a number of very good matches in both sets of experimentally derived PWMs. One striking match shown in Figure 2 is between TIFDMEM0000009 and Trithorax-like (Trl), the gene encoding the GAGA factor, a protein involved in activating gene expression by influencing chromatin structure (reviewed in [29]) and which is known to bind a large number of genomic regions [30]. A second striking match shown in Figure 3 is between TIFDMEM0000040 and serpent (srp), a GATA factor necessary for the development of the amnioserosa, fat body, endoderm, and blood cells (reviewed in [31]). Known binding sites for both Trl and srp are found within 200 bp of the TSS of their respective genes [9], and therefore binding sites for these developmentally regulated transcription factors might be expected to be enriched in our dataset. In total, we saw seven matches to FlyReg, 14 matches to the extended JASPAR CORE, and eight matches to the motifs from Ohler et al. (2002) [12]. Accounting for redundancy between these sets, we obtained a set of 25 inferred motifs which significantly match a known TFBM.
We further investigated the similarities between discovered and previously known motifs by considering the fraction of common predictions between two similar PWMs. Given sets of base positions covered by predictions from two motifs, B1 and B2, we define the overlap between them as:
i.e., O = 0 when the two motifs match completely distinct sets of positions, while O = 1 if one motif is matching a subset of the other's predictions. This latter property should mitigate the effect of any potential errors made when setting score cutoffs for each of the motifs. We used this measure to compare each of the 25 known motifs in our set with the closest existing PWM from the three sets above. Inferred PWMs were scanned across the whole sequence of D. melanogaster chromosome arm 2L using the optimal score cutoffs we defined previously. We set score cutoffs for the corresponding existing PWMs using the same strategy, then calculated prediction overlaps for 22 pairs of PWMs (for three of the existing PWMs, we were unable to define an optimal cutoff). In 12 out of 22 cases, the predictions matched rather closely (O ≥ 0.5, and in three cases O ≥ 0.9). In eight out of 22 cases, 0.1 <O ≤ 0.5. This still represents a substantial overlap, but the majority of predicted bases are different. This suggests that the inferred PWM in these cases might not be doing a good job of modeling the same binding specificity as the existing known motif, which could represent a failure of the motif inference process. Alternatively, these cases might also suggest the existence of a family of similar, but not identical, motifs—perhaps targeted by a family of related DNA-binding proteins. Finally, two out of 22 pairs show little or no overlap between sets of predicted sites.
These results confirm that NestedMICA can simultaneously recover many good PWMs from large genomic datasets, for both core promoter motifs and motifs for developmentally regulated transcription factors with characterized binding specificities. Nevertheless, 95 of the motifs we have discovered were not assigned to a known TFBM by this analysis. This result may not be surprising: characterization of a transcription factor's binding specificity is a complex and laborious process, and only a relatively small subset of known factors have been fully studied. Therefore, it seems reasonable to propose that many of the remaining discovered motifs will be good models for the binding specificities of as-yet-uncharacterized transcription factors.
Our set of 120 inferred motifs contains several PWMs that are visually quite similar (for examples, see Figure 4). To address the question of whether our motif set includes possibly redundant motifs, we applied a similar comparison strategy to that described above to perform an all-against-all comparison of the 120 PWMs. Using a significance threshold of p ≤ 0.05, we found 25 significant similarities between 31 of the 120 motifs. These 31 motifs formed 13 clusters, suggesting that 18 out of 120 motifs might be redundant. The largest of these clusters had four members, which are shown in Figure 4. We also analyzed the overlaps between predicted sites for our motifs, using the same strategy as above. As in the comparison of known motifs, significantly similar PWMs do not always predict strongly overlapping sets of sites: indeed, only one of the 25 pairs had an overlap score of O > 0.5. Therefore, it seems possible that some of these similar motifs are not truly redundant but might instead represent binding specificities for related—but not identical—transcription factors.
The question of motif redundancy is important when classifying our inferred motifs as known or novel. While 25 out of 120 PWMs are significant best-reciprocal-matches to known motifs from one of the existing motif sets described above (and are therefore classified as “known”), an additional eight related PWMs show significant similarity to one of these known motifs. This leaves 87 out of 120 PWMs (80 out of 102 if we assume that all similar motifs are in fact duplicates) that we classify as novel.
Functional binding sites are likely to be subject to purifying selection and thus should exhibit a reduced rate of sequence evolution. This is based both on the observation of increased levels of conservation in known TFBSs relative to their background sequences [32,33] and the intuition that losing elements responsible for gene regulation may often be deleterious [34]. Of course this does not mean that all regulatory elements are under strict purifying selection, and indeed there are good examples of divergence in regulatory element function [35], as well as conservation of regulatory function with underlying binding site turnover at the sequence level [36]. Nevertheless, increased conservation of predicted TFBSs provides evidence for functional constraint [33].
To test whether motifs in our set show signatures of evolutionary constraint among Drosophila species, we studied patterns of motif conservation in a large set of orthologous non protein-coding alignments. Alignments were available genome-wide, but to avoid any possible overfitting artifacts, we discarded the subset of alignments matching D. melanogaster chromosome arm 2L. Since we are more confident of the non protein-coding sequence alignment between closely related species [37], we concentrated on testing conservation between D. melanogaster and two closely related species, D. simulans and D. yakuba. For each match to the D. melanogaster genome, we looked for matches of the same motif to orthologous positions in all three genomes. We then stratified all the D. melanogaster matches of a given motif by decreasing bit-score. In each bin, we calculated the fraction of sites where a prediction was present in all three species (score ≥7 bits for all motifs. This cutoff is less stringent or equal to the optimal cutoffs chosen for all but one of our inferred motifs). In many cases, we saw striking correlations between motif score and degree of conservation, as shown in Figure 5. While the most common pattern is for high-scoring motifs matches to be more conserved, in a few cases we saw a strong inverse correlation (e.g., TIFDMEM0000087), with the strongest matches being substantially less conserved. These underconserved motifs are intriguing, since such a distribution of conservation seems improbable if the motif wasn't associated with some function.
In total, 78 out of 120 motifs showed statistically significant (p ≤ 0.001 in a test) excess conservation for the highest-scoring matches compared with the lowest-scoring matches. At this significance level, we expect a false discovery rate of less than one motif to show an excess of conservation for high-scoring matches, and thus we interpret this result as strong evidence for the majority of discovered motifs being preferentially maintained by purifying selection. A further 22 motifs showed significant underconservation, and the remaining 20 had little or no correlation between score and conservation. This does not conclusively show that they do not reflect functional binding sites, but does suggest that perhaps a relatively small fraction of the total matches to the genome are functional for these motifs. Taking a conservative interpretation of these results (i.e., that only motifs positively correlated with increased sequence conservation have any power to recognize functional sites), this suggests that 65% of our discovered motifs may reflect functional recognition sequences.
It is widely expected that motifs that are functional constituents of promoters should be specifically overrepresented in promoter regions (i.e., close to, and especially upstream of, the transcription start sites) relative to the rest of the genome. This was our main justification for using 5′ flanking sequences when inferring the motif dictionary. Positional bias in the distribution of motifs relative to TSSs has been used in the past in Drosophila for characterizing discovered motifs [12] and as an objective function for motif discovery [26].
Since we used only one chromosome arm for the initial motif discovery process, we had access to many independent promoter sequences for testing purposes. To investigate positional biases of our predicted TFBMs, we scanned the whole sequence of D. melanogaster chromosome arm 2R, using the score cutoffs defined on the basis of overrepresentation tests. For each motif, we recorded the number of matches in 100-base windows relative to the starts of all annotated transcripts. Many motifs showed nonuniform distributions relative to transcription start sites, but the exact distribution varied, with some motifs located very close to the TSS while others showed much broader peaks (c.f., [26]). In many cases, there was also a trough just downstream of the TSS, but again the magnitude and width varied. Several representative examples are shown in Figure 6.
To evaluate the significance of positional biases, we counted motif matches overlapping 400 (200 upstream, 200 downstream) 100-base windows relative to the transcription start sites. In 70 out of 120 cases, the highest peak occurred in the 400 bases immediately upstream of the TSS. We consider these motifs to have significantly nonuniform position distributions (p ≤ 0.01). We would expect a false discovery rate of slightly more than one of our 120 motifs to show a positional bias in this analysis.
There are known compositional biases in D. melanogaster promoters [38]: in particular, an overrepresentation of A/T both upstream and (to a slightly lesser extent) downstream of the TSS. It is possible that this mononucleotide frequency bias might explain at least some of the observed motif position bias, so positional bias is not in itself compelling evidence that a motif does in fact represent the binding specificity of a transcription factor. However, not all positionally biased motifs are A/T-rich (e.g., TIFDMEM0000026, see Figure 6B). It is also possible that causality runs in the opposite direction: promoters could be A/T-rich primarily because they are enriched with large numbers of A/T-rich motifs. In any case, we believe that a strong association between any motif and transcription start sites suggests that a motif is biologically interesting, even if it does not confirm its status as a TFBM.
We would like to discover the biological function for each of our motifs. Functional annotation of motifs offers an extra line of evidence that they are biologically relevant, and may also prove useful in understanding the contribution that individual motifs make to regulatory elements in the genome. Previous work has attempted to associate discovered motifs with Gene Ontology terms or with microarray expression data for genes containing motif instances [12,26]. Here we base our functional annotation on a dataset of whole mount in situ hybridization experiments, which includes annotated expression patterns of about 3,000 genes in developing Drosophila embryos [39]. Although the primary content of this database is a set of images showing where each target gene is expressed, this image atlas is accompanied by a curated set of labels that use terms from a controlled vocabulary (ImaGO) to describe where each gene is expressed at each developmental stage where it was seen.
Genome-wide scanning of PWMs is widely considered to give many false positive matches [40]. For this analysis, it poses an additional problem: given a PWM match to bulk genomic sequence—especially in intergenic regions—it is hard to predict which of the nearby genes is the most likely regulatory target. We therefore focused once again on probable promoter regions: 200 bases upstream of each gene. We scanned 200 bases of 5′ flanking sequence for all annotated D. melanogaster genes with all 120 motifs, using the score thresholds defined previously, then counted the number of times each motif was associated with each term in the ImaGO vocabulary, either directly or via the hierarchy of terms in the ImaGO ontology. For this analysis, we counted all co-occurrences, including cases where multiple matches to a given motif occur upstream of a given gene. We tested the significance of each positive association by removing the motif ID labels from all the motif match records, shuffling all the labels, then reattaching them randomly to match records. By repeating this shuffling process many times (in this case, 5,000,000 iterations) and recording the occasions when a given motif-to-expression association is equally or more abundant in the shuffled data than in the real data, we can obtain empirical p-values that describe how often each observed association would have occurred by chance.
The p-values calculated by this method are not directly useful, since we have performed repeated testing of each motif against each term in ImaGO. One way to correct for this would be to apply a Bonferroni correction by multiplying each p-value by the number of terms in ImaGO. However, such a correction would give an overly conservative picture, since many ImaGO terms are highly correlated with one another, so in practice not every test is independent. Instead, we performed another run of the association process using shuffled motif IDs rather than the real labeling giving an empirical view of the false discovery rate we would expect from this method if motifs matched randomly around the genome. For each of these two runs—real and shuffled motif labeling data—we found the lowest p-value for each motif, out of all the ImaGO terms. These are plotted in rank order in Figure 7.
From this analysis, we see that many motifs associate strongly with at least one ImaGO term. We take this as strong evidence that a large fraction of the discovered motifs are involved in time- or tissue-specific transcriptional regulation during the embryonic stages 1 through 16. Moreover, a subset of these have associations that are very much stronger than any associations seen in the randomized set. Setting a threshold of 3.5 × 10−4—at which we would expect approximately one false discovery—25 motifs have at least one association that we consider significant, including eight novel motifs. These 25 motifs are shown in Figure 8. The largest sets of associations are observed for the TIFDMEM0000009/Trl motif (n = 131) and Adf-like motif TIFDMEM0000076 (n = 196), suggesting potentially widespread roles in embryonic development for the GAGA factor and a putative factor that may bind TIFDMEM0000076. We found that the TATA and INR core promoter motifs discovered here do not have significant associations with ImaGO terms, which is consistent with previous results that these motifs are used preferentially by genes that are active in the adult [26]. The full set of ImaGO terms for each of these 25 motifs can be browsed interactively at http://servlet.sanger.ac.uk/tiffin.
A number of interesting associations emerge from this analysis. Foremost is a set of associations between TIFDMEM0000040 and nine ImaGO terms that fall into two categories relating to the development of the fat body (fat body specific anlage, fat body/gonad primordium, embryonic/larval fat body, fat body, embryonic/larval adipose system, and adipose system) and development of the amnioserosa (extraembryonic structure, amnioserosa anlage in statu nascendi, amnioserosa). Since the fat body is mesodermal in origin and the amnioserosa is an extra-embryonic tissue, we interpret this result as two independent biological associations of the TIFDMEM0000040 motif with two independent sets of related ImaGO terms. As noted above, TIFDMEM0000040 is an almost perfect match to the PWM for the gene srp, which is required for the ongoing differentiation and maintenance of the fat body [41] and amnioserosa [42]. If the TIFDMEM0000040 motif reflects srp specificity, the association of TIFDMEM0000040 with many genes is consistent with the hypotheses that srp activates a “large battery of early and late fat-body genes” [41] and may function as a “selector gene” [31].
Finding regulatory elements in large genomes remains one of the hardest, and also one of the most exciting, problems in contemporary genome biology. Here we have shown that simultaneous probabilistic motif inference using NestedMICA can successfully be applied to large sets of unrelated promoter sequences in metazoan genomes. We have produced a dictionary of 120 motifs in an attempt to capture a significant fraction of the common promoter motifs in Drosophila. While our set still falls well short of the estimated 753 transcription factors in the D. melanogaster genome [8], we expect that scalability improvements in the NestedMICA algorithm, and the use of larger datasets—should close that gap over time.
We offer several lines of independent evidence that support the validity of many of the 120 motifs discovered here. First, we find significant matches to eight out of ten core promoter motifs found previously in a smaller dictionary of computationally derived motifs [12]. The two unmatched motifs were not expected to be recovered in our analysis of upstream flanking regions, since they have been shown to be preferentially located downstream of the TSS [12]. Thus, we can recover all these previously discovered upstream promoter motifs while simultaneously inferring a much larger dictionary. Some of these additional motifs in our dictionary are quantitatively and qualitatively very similar to experimentally derived binding transcription factors motifs (Figures 2 and 3), including matches to developmentally regulated transcription factors, such as srp. Together, these results demonstrate that running NestedMICA on large sets of sequences is an effective way of simultaneously recovering valid binding motifs for both basal and developmentally regulated transcription factors.
In addition to recovering known motifs, 87 of the motifs discovered in this study have no significant match—directly or indirectly—to any of the three reference motif sets used here, suggesting that NestedMICA can also discover novel motifs. We believe that a large proportion of these novel motifs are predictive models of functional sequence elements, as the majority of the motifs discovered here are either preferentially conserved (78/120, 65%, including 22/25 known and 49/87 novel motifs) or preferentially located upstream of transcription start sites (70/120, 58%, including 14/25 known and 50/87 novel motifs). In total, 106 out of 120 motifs (88%) are supported by at least one of these two analyses. Furthermore, many motifs are preferentially located near genes with similar expression patterns in D. melanogaster embryos, and for 25 of these (14 of which were previously known and eight of which we believe are novel) we can make statistically significant associations to one or more ImaGO controlled vocabulary terms used to annotate the in situ gene expression atlas. A summary of the sources of evidence supporting each of our predicted motifs is shown in Figure 9.
Integrating results from our study with previous analyses of promoter motifs in D. melanogaster [12,26] reveals that diverse computational strategies yield nearly identical sets of core promoter motifs, including classical promoter motifs such as the TATA box and INR. In addition, all three studies identify the DRE and E-box motifs, suggesting that proteins that bind these motifs play an important role in regulating a large number of Drosophila genes. All three studies also consistently recover three uncharacterized motifs (TIFDMEM0000116/Motif 1/Dmv4, TIFDMEM0000091/Motif 6/Dmv5, TIFDMEM0000042/Motif 7/Dmv3) that are likely to be widely used binding sites for as-yet unknown transcription factors. Intriguingly, we find strong associations for all three of these unknown motifs (and for DRE) with ImaGO terms, including maternal expression. These associations support previous results indicating that the presence of these motifs positively correlate with female germline expression [26]. Moreover, we find one of these unknown motifs (TIFDMEM0000116/Motif 1/Dmv4) is strongly associated with multiple ImaGO terms, implicating a role in mesoderm and muscle development. Discovering the factors that bind these putative TFBMs may uncover new core promoter selectivity factors in Drosophila.
The strong association of TIFDMEM0000040 with genes expressed in the fat body and amnioserosa demonstrates that we may be able to discover and annotate the function of developmentally regulated TFBMs in upstream flanking regions using NestedMICA in conjunction with the ImaGO controlled vocabulary. However, it is difficult to unambiguously interpret these associations as deriving solely from the srp gene, even though the TIFDMEM0000040 shows a best hit to the srp PWM, and srp is known to be involved in the development of both the fat body and amnioserosa [41,42]. The Drosophila genome contains five recognized GATA factors, which are likely to share similar binding specificities, as has been shown directly for two genes, srp [43] and pannier (pnr) [44]. In addition, pnr has been shown to be expressed in the amnioserosa [45], and cells of the amnioserosa die in pnr mutant embryos [46]. Thus, the TIFDMEM0000040–ImaGO association may in fact derive from a composite signal of both srp (in the fat body and amnioserosa) and/or pnr (in the amnioserosa). Likewise, all five GATA family members may contribute to the signal of TIFDMEM0000040 overrepresentation in promoter regions. This example highlights a general problem in any large-scale motif inference effort—resolving the many-to-one mapping of factors with related specificities to individual motifs—a problem that should be less severe in model organisms such as Drosophila that have fewer paralogues per transcription factor gene family [47]. The discovery in our motif dictionary of a number of apparently similar PWMs that nevertheless match to substantially distinct sets of genomic sites suggests that computational methods may be able to distinguish between the exact binding specificities of related transcription factors, but this remains a topic for future research.
We acknowledge that our strategy of discovering motifs purely from 200 bp of 5′ flanking regions is a significant limitation: in particular we recognize that binding sites for proteins that interact exclusively with distal enhancer/silencer elements rather than proximal/core promoter regions are likely to be overlooked in this analysis. Underscoring this point is the fact that the majority of known motifs for developmentally regulated factors are not recovered here, suggesting that these factors do not bind preferentially to the 200 bp upstream of their target genes. Given the computational challenge presented by a whole-genome motif discovery experiment, we believe that the approach taken here is a simple and robust intermediate strategy for obtaining sequences enriched in a variety of TFBMs, and our resulting motif dictionary—including 87 novel motifs, eight of which have significant associations to embryonic expression patterns—appears broadly to support this decision.
Looking to the future, the most significant question remains that of how best to use a motif dictionary to scan the genome and to annotate functional binding sites. Simply scanning bulk DNA with PWMs tends to yield many false positive matches, even when using relatively stringent score thresholds. Searching for clusters of predicted binding sites has been shown to improve regulatory element detection [48], but does not itself solve the problem of annotating individual binding sites. It is well-known that comparative data generally can improve functional genomic predictions, and sequence conservation specifically has been shown to enhance TFBS annotation [49]. At the time of writing, genome sequences are available for 12 Drosophila species, so they should offer a good platform to investigate comparative approaches to TFBS annotation. Improving the specificity of TFBS annotation should reduce the false discovery rate when performing analyses such as the comparison with ImaGO terms presented here, and indeed such analyses may represent a good initial in silico test for new TFBS annotation methods.
We used versions 3 and 4 of the Drosophila melanogaster genome sequence from BDGP [50] and corresponding curated gene annotation from FlyBase [4]. Sequence and annotation was extracted from the Ensembl database [1].
Genome sequence from other drosopholids was obtained via the Drosophila Assembly/Alignment/Annotation portal http://rana.lbl.gov/drosophila/
Multiple sequence alignments between the noncoding genomic sequences of Drosophila species were obtained from http://rana.lbl.gov/drosophila/alignments_eisenlab.html.
Briefly, these alignments were produced by using synteny information plus the results of BLAST [51] comparisons of exon sequences to define orthologous exons, then using MLAGAN [52] to align the regions between each adjacent pair of exons from all available genomes.
We considered all gene starts on chromosome arm 2L in version 3 of the D. melanogaster genome annotation. Where multiple transcripts were annotated for a single gene, the start of the most upstream transcript was considered to be the gene start. For each gene start site, we attempted to extract 200 bases of 5′ flanking sequence. However, if this region overlapped another gene, it was truncated such that no sequence was extracted from within annotated genes. If this criterion meant that the length of the 5′ flanking region fell below 101 bases, it was discarded. In some cases, two gene start sites in opposite orientations fell close to one another (“divergent genes”). When this meant that two regions would overlap, they were merged into a single region such that no sequence was duplicated in the final set.
Large sets of motifs were inferred from 5′ flanking sequences using the motiffinder program from version 0.7.0 of the NestedMICA package. The following options were used: (1) -numMotifs 120: desired number of motifs; (2) -targetLength 12: desired PWM length, (3) -expectedUsageFraction 0.1: specifies a prior belief that many motifs will be relatively rare in the input sequence set; (4) -revComp: allow motifs to occur in either orientation; (5) -mixtureUpdate weakResample: enables an optimization for faster estimation of NestedMICA's internal matrix that indicates which motifs are present in which input sequences
In addition, NestedMICA has various options that control communication between nodes on a network, and the creation of checkpoint files that periodically store the state of the ongoing computation. None of these options directly affect the final outcome, but they are important for fast and reliable completion of long-running processes. We refer interested readers to the NestedMICA documentation (see Availability section).
A motif PWM is a generative model for a small fragment of sequence; therefore, the natural score for a PWM W at position p in sequence S is:
For convenience, we perform two transformations on this basic scoring function. First, in common with most workers in the field, we use (bit) scores. Second, to compensate for the wildly different magnitudes of score from different motifs, we subtract the highest possible bit score (i.e., the score that would be received by a sequence fragment that consisted of the most likely symbol at each position), so that the highest possible renormalized score for each motif is always 0.
Each motif in the initial discovered set was scanned across the 5′ flanking sequences from chromosome arm 2L. At each position where the motif matched with a score ≥−6, we extracted a short sequence containing the motif match (always 12 bases in this case) plus ten bases of flanking sequence on either side.
NestedMICA requires that the target motif length be specified when the program is run, and does not attempt to optimize PWM length internally. NestedMICA can, however, output the Bayesian evidence for training a particular model configuration (including PWM length) on a given set of sequences, estimated using the Nested Sampling method. We therefore performed multiple runs of NestedMICA with target motif lengths varying from four to 12 (otherwise using default options) and selected the model with the highest evidence. For this analysis, we used the same background model as originally used for large-scale motif discovery.
PWMs were scanned across the training set of chromosome arm 2L 5′ flanking sequences as described above, using an extremely lenient threshold of −10 bits. We subdivided matches by score, binning them into 1-bit intervals, then calculated frequency distribution of the ten different score bins. In parallel, we exhaustively enumerated all words that could match the PWM with a score of −10 or greater, then calculated the likelihood of each of these words under the same background model we used in the motif discovery process. By weighting each word with its background likelihood, we can obtain an expected histogram of PWM match scores. For each score bin, we compared the observed and expected frequencies using a binomial test. We then picked a score cutoff equal to the highest score of the first bin that was not significantly overrepresented relative to the background model (significance defined as p ≤ 0.05). In a few cases, the highest-scoring bins were not significantly overrepresented—generally because of a very small total number of matches in these bins—but slightly lower-scoring bins did show overrepresentation. In these cases, we skipped over up to two high-scoring nonoverrepresented bins and set a threshold based on the first nonoverrepresented bin following an overrepresented bin.
As positive controls for our analyses, we generated a set of PWMs of the binding specificity for developmentally regulated transcription factors characterized by DNase I footprinting. Footprint data was obtained from version 2.0 of the FlyReg database [9]. To ensure that we observed a representative sampling of binding sites for each factor, we concentrated on the 52 factors with at least five reported footprints. For each of these factors, we extracted genomic sequences for all the footprints plus ten bases of flanking sequence on either side. PWMs were built using NestedMICA [13], requesting a single motif from each set of sequences. Since we expect the concentration of true binding sites in the footprint sequences to be very high, choice of background model is less important than when learning motifs from bulk sequence, so we used a simple, uniform zeroth order background model.
Since we did not know the size of each factor's binding site, we used a similar procedure to that described under the Motif refinement and trimming section. For each factor, we performed several runs with motif lengths between four and 12. We then plotted the evidence for each motif length. In 30 of 52 cases, we saw a clear peak in the evidence plot, and used the corresponding weight matrix as the optimal model for that factor. For the remaining 22 cases, there was not an obvious peak, so we were not able to confidently choose a single model. This may be due to insufficient data (supported by the observation that the factors with no clear optimum had a median of 7.5 sites, compared with 18.5 for those with a sharp evidence peak), contamination of the footprint data with some false positives, or because a single PWM is not a good model of the binding specificity for that factor (perhaps because it binds in several different conformations). The 30 optimal PWMs can be found in Figure S2.
We define a divergence measure between two distributions, P and Q, over alphabet A as:
When the exponent, ε, is 1.0, this gives a Cartesian distance. In practice, we found that this gave too much weight to small differences. In this paper, we use an exponent ε = 2.5.
The distance between two PWMs can then be measured as the sum of distances between corresponding positions. Since it is possible that two PWMs could reflect the same motif offset by a few bases in either direction, we consider each motif to be flanked (to infinity in both directions) by uniform distributions over A, then consider all possible ungapped alignments between two motifs. The lowest score (i.e., the best alignment) is reported.
We note that there is not an obviously principled way to compare PWMs. This is an ad hoc distance metric selected on the basis that pairs of motifs with low divergences under this metric tended to look similar by eye. Other possible distance metrics include Cartesian distances with a different exponent, KL divergence, and correlation coefficient. These have all been tested but were judged to be inferior to this measure. Better methods for motif comparison, and also techniques for evaluating motif comparison methods, remain an interesting topic for future research.
This comparison measure—along with the ability to identify best matches and reciprocal best matches between two sets of motifs—has been integrated into the MotifExplorer tool, which can be downloaded from http://www.sanger.ac.uk/Software/analysis/nmica/mxt.shtml.
Where a measure of the significance of a match between two motifs is required, we repeatedly shuffle the columns of the query PWM and compare the resulting random motif with the same target motif database. An empirical p-value for the significance of the initial match can be obtained by counting the number of times a random motif matches with a score less than or equal to the best match of the query motif.
Source code for the NestedMICA motif-discovery system is freely available under the GNU Lesser General Public License (LGPL) from http://www.sanger.ac.uk/Software/analysis/nmica.
The motif PWMs and annotation can be downloaded from the Tiffin database of computational motif finding results http://servlet.sanger.ac.uk/tiffin. |
10.1371/journal.ppat.1006547 | Different rates of spontaneous mutation of chloroplastic and nuclear viroids as determined by high-fidelity ultra-deep sequencing | Mutation rates vary by orders of magnitude across biological systems, being higher for simpler genomes. The simplest known genomes correspond to viroids, subviral plant replicons constituted by circular non-coding RNAs of few hundred bases. Previous work has revealed an extremely high mutation rate for chrysanthemum chlorotic mottle viroid, a chloroplast-replicating viroid. However, whether this is a general feature of viroids remains unclear. Here, we have used high-fidelity ultra-deep sequencing to determine the mutation rate in a common host (eggplant) of two viroids, each representative of one family: the chloroplastic eggplant latent viroid (ELVd, Avsunviroidae) and the nuclear potato spindle tuber viroid (PSTVd, Pospiviroidae). This revealed higher mutation frequencies in ELVd than in PSTVd, as well as marked differences in the types of mutations produced. Rates of spontaneous mutation, quantified in vivo using the lethal mutation method, ranged from 1/1000 to 1/800 for ELVd and from 1/7000 to 1/3800 for PSTVd depending on sequencing run. These results suggest that extremely high mutability is a common feature of chloroplastic viroids, whereas the mutation rates of PSTVd and potentially other nuclear viroids appear significantly lower and closer to those of some RNA viruses.
| Spontaneous mutations are the ultimate source of genetic variation and their characterization provides fundamental information about evolutionary processes. The highest mutation rate so far described corresponds to a hammerhead viroid infecting plant chloroplasts. Viroids are plant-exclusive parasites constituted by 250–400 nt-long, non-protein-coding RNAs, and are divided into two families with distinct mechanisms of replication and localization: chloroplastic (Avsunviroidae), and nuclear (Pospiviroidae). Here, we have used high-fidelity ultra-deep sequencing to compare side by side the mutation rates of one representative member of each viroid family in the same host. We found that the mutation rate of the nuclear viroid was several fold lower than that of the chloroplastic viroid.
| Spontaneous mutations are pivotal to evolution as they constitute the ultimate source of genetic variation. The biochemical and genetic bases of replication fidelity have been extensively studied, and it is well-established that spontaneous mutation rates vary by orders of magnitude across biological systems [1, 2]. Whereas bacteria and other microorganisms show highly accurate replication, RNA viruses replicate with frequent errors [3]. Yet, the lowest replication fidelity reported so far corresponds to chrysanthemum chlorotic mottle viroid (CChMVd), a chloroplastic viroid in which a mutation is incorporated approximately every 400 bases copied [4]. Viroids are small (250–400 nt), circular, highly-structured RNAs that do not encode proteins and are copied by nuclear or chloroplastic DNA-dependent RNA polymerases forced to accept RNA templates [5–7]. They infect plants exclusively and their pathogenicity has been linked to RNA silencing [6], although other mechanisms cannot be excluded. Chloroplastic viroids encode in both polarity strands hammerhead ribozymes that play an essential role in their replication cycle. Together with their extreme simplicity, the presence of ribozymes makes these viroids reminiscent of the primordial replicons postulated by the RNA world hypothesis for the origin of life [5, 8].
Although the genetic diversity of some representative viroids has been characterized in previous work [9–11], CChMVd is the only viroid for which the rate of spontaneous mutation has been determined experimentally [4]. As such, it remains to be shown to what extent extremely high mutation rate is a more general property of viroids or, in contrast, is specific to CChMVd and closely-related viroids. CChMVd belongs to the family Avsunviroidae, the members of which replicate in plastids (mostly chloroplasts), where their single-stranded circular RNA is copied by a bacteriophage-like nuclear-encoded RNA polymerase (NEP) through a rolling-circle mechanism to yield linear oligomers [12, 13]. The latter are cleaved co-transcriptionally [14] by the embedded hammerhead ribozymes to yield monomers [15, 16], which are circularized by a tRNA ligase [17]. By convention, the (+) polarity is assigned to the most abundant strand, but the replication cycle is fully symmetric, i.e. identical for both polarities (Fig 1).
In contrast, members of the family Pospiviroidae are copied by RNA polymerase II in the nucleus, where rolling-circle replication of the circular (+) strand produces (–) oligomers, which are used directly for a second round of copying to yield (+) oligomers [18–20]. These replicative intermediates are then cleaved into (+) monomers by RNAse III [21] and circularized by DNA ligase I accepting RNA substrates [22] (Fig 1). Although analysis of genetic diversity suggested differences in replication fidelity between nuclear and chloroplastic viroids [9–11], recent work based on previously published deep sequencing data has posited that potato spindle tuber viroid (PSTVd), the type species of the family Pospiviroidae, might show extremely high copying error rates similar to those of CChMVd [23]. It therefore remains to be elucidated whether the two viroid families show different rates of spontaneous mutation. Importantly, these previous works have not disentangled the various factors contributing to genetic diversity, which include selection, mutational robustness, but also genetic drift and population structure among others, precluding an unbiased inference of mutation rates.
Here, we sought to quantify in vivo the mutation rate of one representative viroid of each family replicating in a common host. Typically, mutation rate estimation methods require highly controlled laboratory conditions as well as detailed information about the number of replication cycles elapsed [24], which makes them often unsuitable for measuring mutation rates in vivo. To circumvent this problem, it is possible to quantify the population frequency of lethal mutations [4]. This is because lethal mutations cannot be inherited, and hence, their frequency should equal the rate at which they are produced. As formalized by classical mutation-selection balance models, in haploid populations the equilibrium frequency of a deleterious mutation is q = μ / s, where μ is the rate of spontaneous mutation and s the selection coefficient [25]. Hence, whereas for slightly deleterious or neutral mutations (s → 0) the observed mutation frequency may strongly deviate from mutation rate (q >> μ), for highly deleterious mutations (s → 1) q will approach μ. Analysis of lethal or quasi-lethal mutations has been previously used for inferring the mutation rates of hepatitis C virus [26, 27], poliovirus [28], and human immunodeficiency virus 1 [29], in addition to CChMVd [4]. A complication of this approach, though, is that since lethal mutations have low population frequencies, sequencing must be carried out with both high depth and accuracy. Sanger sequencing is highly accurate but has limited depth, whereas standard next-generation sequencing (NGS) has extreme depth but low per-read accuracy. This problem has been solved recently by the development of methods that increase the accuracy of NGS by orders of magnitude, such as CircSeq and Duplex Sequencing (DS) [30, 31], which now permit a better characterization of viroid genetic diversity and mutation rates.
Here we have focused on DS because CirSeq demands high amounts of starting material and is thus impractical for viroid RNA obtained from infected tissue. DS reduces considerably sequencing mistakes by tagging and sequencing independently each of the two DNA strands multiple times, wherein true mutations are detected in the same position. Whereas DS does not allow removal of errors associated with reverse transcription and PCR, it can nevertheless strongly increase accuracy compared to conventional NGS by removing errors associated with sequencing. We have applied DS to the chloroplastic eggplant latent viroid (ELVd) and the nuclear PSTVd infecting eggplant to exclude possible biases caused by using different hosts. We found that, while ELVd showed an extremely high mutation rate similar to that of CChMVd [4], the mutation rate of PSTVd was lower and fell closer to the typical range of RNA viruses.
For each ELVd and PSTVd, three eggplant seedlings were agro-inoculated with infectious plasmids containing head-to-tail dimeric inserts of the corresponding viroid cDNAs and total nucleic acids from upper non-inoculated leaves were extracted six months post-inoculation (mpi) for PSTVd and 12 mpi for ELVd. Subsequent fractionation with non-ionic cellulose [32] resulted in preparations enriched in RNAs with a high content in secondary structure, including viroid RNAs. The six individual preparations were separated by denaturing PAGE and the RNAs migrating between the markers of 400 and 600 nt containing the monomeric circular RNAs were eluted and recovered (Fig 2). To assess whether levels of genetic diversity varied among different viroid RNA forms, we also recovered the strands migrating between the markers of 600 and 1000 nt, which correspond to oligomeric viroid RNAs. The extracted RNAs were used for high-fidelity RT-PCR and sequenced by the DS method. The RT-PCR was performed with adjacent primers of opposite polarities to generate full-length products from the monomeric viroid circular (+) strands and the (–) oligomers, hence allowing us to sequence the entire viroid except for the primer regions. To control for errors associated with reverse transcription, PCR, and sequencing, we also performed DS of the PCR product obtained directly from a plasmid with a PSTVd insert, as well as of the RT-PCR products from (+) PSTVd and (+) ELVd RNAs transcribed in vitro using T7 or T3 RNA polymerases. The PCR products from plant extracts and the RT-PCR controls were tagged and analyzed in the same run, using a MiSeq Illumina sequencer. DS of the direct PSTVd PCR product with an average depth of 8071 reads per site yielded 16 total mutations in 2,808,781 bases read. Hence, assuming there was no variation in the plasmid template, the joint technical error rate of PCR and DS was 5.7 × 10−6. This rate increased to 3.9 × 10−5 and 4.4 × 10−5 in RT-PCR products from the in vitro transcripts of PSTVd and ELVd, respectively, showing the important contribution of RT to sequencing errors, although mutations arising during in vitro transcription were also probably present in these controls.
Analysis of the three ELVd-infected plants yielded an average mutation frequency of (1.5 ± 0.3) × 10−2 for circular (+) strands and of (1.8 ± 0.3) × 10−2 for (–) oligomers (Table 1; S1 Dataset), confirming the extremely high genetic diversity of chloroplastic viroids. At an average sequencing depth of 764 reads/site/run, 264 of the 295 ELVd nucleotide sites examined (89%) were polymorphic. The genetic diversity was not uniformly distributed along the ELVd sequence, with peaks in regions encompassing sites 120–140 and 240–245, which map to a hairpin and a loop, respectively, in the secondary structure proposed in vivo for the monomeric ELVd (+) strand [33] (Fig 3). In contrast, relatively low diversity values were found in regions delimited by sites 69–79 and 95–103, which form the two strands of a base-paired stem, as well as an adjacent bulge. We also found lower-than-average diversity in the region encompassing sites 18–50, which maps to the (+) hammerhead ribozyme. Sites 152–180 and 188–200, which map to the (–) hammerhead ribozyme, also showed low diversity in circular (+) strands. Overall, there was an excellent correlation between per-site mutation frequencies in circular (+) strands and (–) oligomers of the same plant at the analyzed sampling time (Pearson correlation: r ≥ 0.952; P < 0.001). We also found that per-site mutation frequencies were significantly correlated among plants (r ≥ 0.785; P < 0.001), suggesting that genetic diversity was mainly driven by a deterministic mutation-selection balance.
In PSTVd, we found variation in 262 of the 304 sites examined (86%) at an average sequencing depth of 4471 reads/site/run. Hence, a greater depth was required to sample a number of polymorphic sites similar to that found for ELVd. The average frequency of mutations was two orders of magnitude lower in PSTVd than in ELVd, with values of (1.5 ± 0.2) × 10−4 and (3.0 ± 2.0) × 10−4 for circular (+) strands and (–) oligomers, respectively (Table 1). The higher diversity of (–) oligomers was driven by an anomalously high mutation frequency in plant 1 (7.1 × 10−4). Importantly, mutation frequency was significantly higher in sequences obtained from PSTVd-infected plants than in control sequences derived from the in vitro transcript, indicating that most sequence variants detected in vivo were real and not RT-PCR-sequencing artifacts (Fisher test: P < 0.001 in all six runs). As for ELVd, we found significant correlations between the per-site mutation frequencies of circular (+) strands and (–) oligomers in plant 2 (r = 0.551; P < 0.001) and plant 3 (r = 0.735; P < 0.001). This correlation, albeit still significant, was much lower in plant 1 (r = 0.228; P < 0.001), with the (–) oligomers of this plant showing a distribution of mutations across sites discordant with the other five runs (Fig 3). Excluding this anomalous set, we also observed correlations between per-site mutation frequencies from different plants, albeit lower than those found for ELVd (r ≥ 0.468; P < 0.001). This result could be explained by the greater difficulty of reproducibly sampling rarer genetic variants, or could indicate that random genetic drift has a stronger influence on genetic diversity in PSTVd than for ELVd. Finally, no major diversity peaks were found in PSTVd except for position 117 at the terminus of an A-rich sequence in the pathogenicity domain of the secondary structure proposed in vivo for the monomeric PSTVd (+) strand [34, 35] (Fig 3). This region had a marked tendency to point insertions/deletions, possibly resulting from polymerase slippage.
The most abundant types of mutations in ELVd sequences from circular (+) strands were transitions (75.4 ± 4.6% of all mutations), followed by transversions (21.7 ± 4.5%) and point insertions (2.5 ± 0.1%), whereas point deletions were the rarest type (0.4 ± 0.1%). C-to-U, G-to-A, and U-to-C substitutions were found at similar frequencies, whereas A-to-G changes were slightly less frequent. A very similar pattern was found for sequences derived from (–) oligomers (Table 2). In contrast, the mutational spectrum was markedly different in PSTVd, with 50.5 ± 2.0% transitions, 40.8 ± 1.6% transversions, 3.6 ± 1.1% insertions, and 5.2 ± 0.6% deletions in circular (+) strands, versus 56.5 ± 3.5%, 31.5 ± 6.2%, 6.4 ± 2.7%, and 5.6 ± 0.6% in (–) oligomers, respectively. Contrarily to ELVd, we found clear differences among transition types in PSTVd, such that C-to-U > G-to-A > U-to-C > A-to-G in circular (+) strands, whereas (–) oligomers showed a different pattern (G-to-A > A-to-G > U-to-C ≈ C-to-U; Table 2). The different mutational spectrum of PSTVd (+) and (–) strands was explained in part by reverse complementarity, i.e. C-to-U > G-to-A in (+) strands as opposed to G-to-A > C-to-U in (–) strands, and U-to-C > A-to-G in (+) strands as opposed to A-to-G > U-to-C in (–) strands.
To estimate the ELVd mutation rate by the lethal mutation method we focused on the hammerhead ribozymes, which mediate self-cleavage of the linear oligomers and are hence essential for completing the replication cycle. The hammerhead ribozyme consists of a central catalytic core of 15 nucleotides flanked by three double helices [16, 36] (Fig 4). The core nucleotides are required for the catalytic activity of the ribozyme [36–38] and, since the vast majority of mutations at these positions inactivate self-cleavage activity, the 15 sites can be used for mutation rate inference using the lethal mutation method, as shown previously [4]. In circular (+) strands average mutation frequencies were (1.8 ± 0.3) × 10−3 for the (+) hammerhead and (0.6 ± 0.3) × 10−3 for the (–) hammerhead, whereas in (–) oligomers the frequencies were (1.1 ± 0.3) × 10−3 for the (+) hammerhead and (1.6 ± 0.3) × 10−3 for the (–) hammerhead ribozymes. Such reduction of one order of magnitude in diversity compared with the rest of the sequence was expected, because mutations falling at these essential domains should tend to be lethal and hence leave little or no progeny. Therefore, mutation frequencies in these domains should resemble the rate of spontaneous mutation, as opposed to those in the rest of the sequence. Averaging the above values, the estimated rate of spontaneous mutation of ELVd was (1.3 ± 0.3) × 10−3, or roughly one mutation every 800 bases.
We adopted the same mutation rate estimation approach for PSTVd. To do this, we focused on specific sites forming the central conserved region (CCR) and the terminal conserved region (TCR). These regions mediate key functions including replication for the CCR, or are presumed to play alternative essential roles (yet unknown) for the TCR [39]. Hence, most mutations at these sites should have highly deleterious or lethal effects. We also focused on a set of 23 different single-base substitutions previously reported to impair PSTVd infectivity [40–44]. In circular (+) strands, average mutation frequencies were (1.4 ± 0.1) × 10−4 for the CCR/TCR regions and (1.4 ± 0.3) × 10−4 for the set of previously described lethal mutations (Fig 4). We found similar values in (–) oligomers, except that variance was higher and that plant 1 showed a higher value, as discussed above. Therefore, the estimated rate of spontaneous mutation of PSTVd was 1.4 × 10−4 or roughly one mutation every 7000 bases. Unexpectedly, mutation frequencies at these essential sites were not lower than those obtained for the rest of the PSTVd sequence. It is possible that most mutations in PSTVd are highly deleterious or lethal, regardless of whether they map or not to these specific regions, meaning that PSTVd would show very low mutational tolerance. It is also possible that genetic diversity accumulated at low rates due to slow replication, such that few polymorphisms were produced at 6 mpi. Alternatively, the actual PSTVd mutation rate might be lower than 1.4 × 10−4 and the noise introduced by sequencing errors could have precluded us from measuring this lower value. However, our overall sequencing error rate as determined using a PSTVd RNA transcribed in vitro was 3.6-fold lower (3.9 × 10−5) than the estimated PSTVd mutation rate (the error rates estimated specifically for the CCR/TCR and the set of 23 predefined mutations being 2.0 × 10−5 and 6.6 × 10−5, respectively). Notice that this probably represents an upper-limit to the actual sequencing error rate, because in vitro transcription is an error-prone process that may have contributed mutations in our control assays, but not in actual sequences from plants.
We performed a second set of experiments from the same plants at 18 mpi to address whether viroid diversity depended on sampling time and/or sequencing run. The RNA extraction procedure was identical except that we focused only on monomeric circular RNAs (i.e. migrating between the markers of 400 and 600 nt; Fig 2C and 2D). As above, the RT-PCR was performed with adjacent primers of opposite polarities producing full-length products from monomeric circular (+) RNAs, but annealing at different regions in order to cover the portions of viroid sequence that were not analyzed in the first run (see Fig 3). This new run included six PCR products (three plants, two viroids) as well as controls of RT-PCR products from (+) PSTVd and (+) ELVd RNAs transcribed in vitro. The three ELVd-infected plants yielded an average mutation frequency of (1.45 ± 0.04) × 10−2, which is nearly identical to the value obtained in the first run at 12 mpi (Table 3). Furthermore, the distribution of mutations along the ELVd sequence was also highly similar between the two time points (Fig 3; within-plant Pearson correlation between per-site mutation frequencies at 12 and 18 mpi: r ≥ 0.795), confirming that ELVd reached a deterministic mutation-selection balance.
At 18 mpi, the average frequency of mutations in PSTVd was (2.8 ± 0.1) × 10−3, a value an order of magnitude higher than at 6 mpi but still five times lower than for ELVd (Table 3). In addition to position 117 at the terminus of an A-rich sequence (which already showed a high frequency of point insertions/deletions at 6 mpi) at 18 mpi, we found other single-nucleotide polymorphisms (G143A, U161C, C167A, U209del, and U309A) that independently arose at high population frequencies (> 5%) in the three plants (Fig 3). Removal of these few sites from the analysis reduced the average mutation frequency by threefold, i.e. (9.0 ± 1.1) × 10−4. We also found that the distribution of mutations along the sequence was markedly different at 6 and 18 mpi (within-plant correlation between per-site mutation frequencies: r ≤ 0.131; Fig 3). Contrarily, the per-site mutation frequencies from the three plants were highly correlated at 18 mpi (r ≥ 0.917; Fig 3). Hence, in contrast to the earlier analysis in which random processes such as genetic drift appeared to play an important role in PSTVd genetic diversity, sequences obtained at 18 mpi were more consistent with a deterministic mutation-selection balance, similar to the pattern found for ELVd. Therefore, PSTVd accumulated relatively low levels of diversity at 6 mpi and showed an unexpectedly slow onset of mutation-selection balance.
As above, we estimated the spontaneous mutation rate of ELVd by focusing on the 15 central catalytic core nucleotides of each hammerhead ribozyme. Average mutation frequencies were (1.2 ± 0.3) × 10−3 for the (+) hammerhead and (0.9 ± 0.1) × 10−3 for the (–) hammerhead, or approximately one mutation every 1000 bases copied. These values are similar to those obtained at 6 mpi (Fig 4). For PSTVd, we again focused on CCR/TCR, as well as on the set of 23 single-point mutations previously reported to impair PSTVd infectivity. Average mutation frequencies were (4.1 ± 0.2) × 10−4 for the CCR/TCR and (4.5 ± 1.3) × 10−4 for the set of previously described lethal mutations (Fig 4). As opposed to the results obtained at 6 mpi, these values were six-fold lower on average than those obtained for the rest of the PSTVd sequence (Fig 3). This supports the conclusion that the accumulation of diversity was restricted specifically at these sites by the strongly deleterious/lethal effects of mutations. On the other hand, mutation frequencies at the CCR/TCR and for the set of 23 predefined mutations were threefold higher at 18 than at 6 mpi, which was unexpected assuming that these mutations were lethal. This discrepancy could be in part explained by a higher sequencing error rate in this run. The in-vitro transcribed PSTVd control showed a mutation frequency of 1.7 × 10−4 (172 mutations in 1,014,425 bases read) in the CCR/TCR and of 1.6 × 10−4 (44 mutations in 271,663 bases read) for the set of 23 predefined mutations, versus 2.0 × 10−5 and 6.6 × 10−5 in the previous experiment, respectively. By subtracting the corresponding error rates obtained in the 18 mpi run, the estimated net mutation frequencies were 2.4 × 10−4 and 2.9 × 10−4 for the CCR/TCR and the predefined set, respectively, suggesting approximately one mutation every 3800 bases.
Owing to their lethality or quasi-lethality, mutations in the catalytic core of ELVd hammerhead ribozymes as well as in the CCR/TCR and some specific sites of PSTVd should have a very small number of rounds of copying to accumulate. Specifically, mutations falling at the central catalytic core of the hammerhead ribozyme should be able to survive for 0 to 2 rounds of copying, depending on the polarities of the sequenced strand and of the mutated hammerhead ribozyme (Fig 1). This makes them an excellent target for mutation rate inference by the lethal mutation approach, and a similar argument should hold for the CCR/TCR and the set of PSTVd mutations previously shown to inactivate infectivity. For instance, changes inactivating the ELVd (+) hammerhead ribozyme should prevent production of circular (+) RNA, implying that these mutations should not be found in the catalytic core of the (+) hammerhead ribozyme in sequences derived from the circular (+) strand template. In contrast, we found mutations at a frequency in the order of 10−3 in these sequences, a value not attributable to RT-PCR-sequencing errors because the latter were two orders of magnitude less frequent. Yet, at least two other explanations are possible. First, some mutations may have resulted from RNA editing or spontaneous RNA damage (in vivo, or during the extraction process). RNA damage appears more likely in the single-stranded circular (+) RNA than in the in (–) oligomers, a fraction of which could be forming double-stranded complexes. Second, the hammerhead ribozyme located in the 5´-end repeat of the (+) oligomer should not be required for cleavage and hence may incorporate loss-of-function mutations, as opposed to the other oligomer repeats.
We have found a mutation rate for ELVd (1/100 to 1/800) relatively similar to that of CChMVd (1/400) and, hence, our results suggest that an extremely fast mutation is shared by at least two of the four chloroplastic viroids. In contrast, the mutation rate of PSTVd was 4–8 times lower and more similar to those of RNA viruses [3]. This marked difference is probably at the origin of the higher genetic diversity of chloroplastic viroids compared with their nuclear counterparts. RNA polymerase II has proofreading capacity [45] and its estimated misincorporation rate in Caenorhabditis elegans is 4 × 10−6, the most frequent errors being C-to-U, followed by G-to-U [46]. Interestingly, these were also the two most frequent mutation types in PSTVd (+) circular strands, although the overall mutation rate of PSTVd was much higher than the estimated RNA polymerase II misincorporation rate. This suggests that the fidelity of RNA polymerase II is strongly reduced when the enzyme is forced to accept viroid RNA as template instead of nuclear DNA. In contrast to PSTVd, chloroplastic viroids are thought to be copied by a bacteriophage-like NEP with a lower fidelity than RNA polymerase II [47, 48]. In addition to differences in replication fidelity, we cannot discard other factors that could contribute to explaining differences in mutation rates, such as RNA editing [49] and spontaneous RNA damage. The more open secondary structure of chloroplastic viroids could increase susceptibility to RNA damaging agents. Furthermore, mutagenic free radicals resulting from electron transduction during photosynthesis, as well as unbalanced nucleotide pools, may also contribute to increased mutation rates in the chloroplast. Finally, in addition to differences in replication fidelity and/or RNA damage, chloroplastic and nuclear viroids may also exhibit different tolerance to mutations [50]. We found that, at 6 mpi, PSTVd mutation frequencies showed low heterogeneity along the sequence, with few peaks of diversity and no apparent differences between the CCR/TCR and other viroid regions. However, higher diversity was apparent in some PSTVd regions at 18 mpi.
According to the RNA world hypothesis, RNA preceded DNA as the carrier of genetic information during early stages of life. Indirect evidence supporting the existence of an RNA world is provided by ribozymes, which include the hammerhead structures found in chloroplastic viroids and in viroid-like satellite RNAs. It has been suggested that their small size, circularity, high G+C content, lack of protein-coding ability, and, specially, the catalytic activity associated to ribozymes, make these minimal replicons candidates for being relics of early life-forms [5, 8]. An important consequence of error-prone copying in early replicons is the existence of a limit to genome complexity, as genomes over a certain size would incur in an excessive mutational load. This limit would prevent the evolution of new functions, including repair mechanisms, thereby trapping RNA genomes in an evolutionary dead-end, a problem known as Eigen’s paradox [51]. This constraint predicts a negative correlation between mutation rate and genome size, although such correlation may have other explanations, including random genetic drift [52] and mutation rate optimization [53]. Whereas there is strong evidence for such a negative correlation among viruses and bacteria [2, 3, 53, 54], viroids, are not self-replicating entities and hence should be subject to different contraints. Except for the possible role of secondary structure, factors determining the mutation rate of nuclear viroids are mainly controlled by the host, implying that lower mutation rates may not be evolutionarily accessible to them. Whereas extremely high mutation rates may situate chloroplastic viroids close to Eigen´s error threshold and may hence impose limits to the evolution of larger sequences, this does not seem to be the case for PSTVd and, probably, other nuclear viroids.
Eggplant seedlings (Solanun melongena cv. ‘Redonda morada’) were PSTVd- or ELVd-agro-inoculated 6 and 12 months, respectively, before RNA extraction (run 1), and 18 months before RNA extraction (run 2). Total nucleic acids were extracted by grinding systemic leaves in buffer-saturated phenol, and then fractionated on non-ionic cellulose (CF11; Whatman) with STE (50 mM Tris-HCl, pH 7.2, 100 mM NaCl, 1 mM EDTA) containing 16% ethanol [32]. The resulting preparations, enriched in RNAs with a high content in secondary structure including viroid RNAs, were electrophoresed in denaturing 5% polyacrylamide gels containing 89 mM TBE (Tris-Borate-EDTA) and 8 M urea. The gels were stained with ethidium bromide and the viroid circular RNA (migrating between the linear RNA markers of 400 and 600 nt) and the viroid oligomeric forms (migrating between the linear RNA markers of 600 and 1000 nt), were excised, eluted overnight with 10 mM Tris-HCl, pH 7.5 containing 1 mM EDTA and 0.1% SDS, and recovered by ethanol precipitation.
The substrate for the ELVd control was the dimeric product resulting from in vitro transcription driven by the T7 promotor of a recombinant plasmids containing a dimeric head-to-tail ELVd-cDNA insert of the reference variant ELVd-2 (GenBank AJ536613). For the PSTVd control, the substrate was the monomeric product resulting from in vitro transcription driven by the T3 promotor of a recombinant plasmid containing a monomeric PSTVd-cDNA insert of variant RG1 (GenBank U23058) opened between positions C1-G2 flanked by a modified version of the hammerhead ribozyme of tobacco ringspot virus satellite RNA and a modified version of the ribozyme of hepatitis delta virus minus RNA strand [20, 22]. The resulting unit-length transcripts, purified by denaturing PAGE and subsequent elution, were added to leaves of healthy eggplant homogenized in buffer-saturated phenol and the RNA extraction was continued as indicated in the previous section in order to prepare the controls under conditions mimicking those of infected samples.
Prior to reverse transcription, all samples and controls were treated with the TURBO DNA-free kit (Ambion) to remove any DNA contamination following manufacturer´s instructions. ELVd circular (+) and oligomeric (–) RNAs purified from infected tissue were reverse transcribed for 1 h at 42°C with AccuScript Hi-Fi reverse transcriptase (Agilent) and primer RF-1298 (run 1) or RF-1405 (run 2) for the circular forms, and with primer RF-1299 for the oligomers (see S1 Table for details). The cDNA products were PCR-amplified with Phusion High-Fidelity DNA Polymerase (Thermo Scientific) and adjacent primers RF-1298 and RF-1299 (run 1) or RF-1404 and RF-1405 (run 2), using the following program: 1 min at 98°C, 35 cycles of 15 s at 98°C, 20 s at 66°C or 62°C (run 1 or 2, respectively), and 30 s at 72°C, with a final extension of 2 min at 72°C. The ELVd control RNA was reverse transcribed with primer RF-1298 (run 1) or RF-1405 (run 2) and PCR-amplified with this primer and primer RF-1299 (run 1) or RF-1404 (run 2). PSTVd RNA purified from infected tissue was reverse transcribed with primer RF-1242 (run 1) or RF-1406 (run 2) for circular forms and RF-1359 for oligomers. The cDNA products were PCR-amplified with adjacent primers RF-1242 and RF-1359 (run 1), or RF-1406 and RF-1407 (run 2) using the following program: 1 min at 98°C, 35 cycles of 15 s at 98°C and 20 s at 72°C, and a final extension of 2 min at 72°C. For run 1, the PSTVd RNA control was reverse transcribed with primer PSTVd-rev and PCR-amplified with this primer and primer PSTVd-fw of the PSTVd RG1 variant. For run 2, this control was reverse transcribed with primer RF-1406 and PCR-amplified with this primer and primer RF-1407. For PSTVd extracts taken at 18 mpi, we observed a minor additional PCR band and we excised the band of interest by running a 5% non-denaturing polyacrylamide gel.
This technique increases per-read accuracy by orders of magnitude compared to standard Illumina sequencing, using adapters that have random yet complementary double-stranded nucleotide sequences [55]. Since the probability of two molecules being labeled with the same adapter sequence is vanishingly small, these molecular tags can be used to identify reads originating from each individual strand of DNA in the sequencing output and calculation of a consensus sequence for each of these individual strands, hence allowing removal of sequencing errors. DS adapters were constructed by annealing two oligonucleotides, one of which contained a 12-nt single-stranded randomized sequence tag. Annealed oligonucleotides were extended using the Klenow fragment, digested with a specific restriction endonuclease to produce cohesive ends, and annealed to viroid RT-PCR products for library preparation, following previously described protocols [31]. Given the small size of viroids, no template fragmentation was required. A library was prepared to identify each RT-PCR product and run on an Illumina MiSeq machine sequencer. Sequencing of direct PCR controls was made on a separate run. FastQ files were processed with the DS software pipeline (https://github.com/loeblab/Duplex-Sequencing) using BWA 0.6.2, Samtools 0.1.19, Picard-tools 1.130 and GATK 3.3–0, and GenBank accessions AJ536613 and AJ634596 as reference sequences for ELVd and PSTVd, respectively. After parsing of tags, the first 200 bases of each read were selected to increase accuracy, and initial alignment and single stranded consensus sequence (SSCS) were assembled, followed by duplex consensus sequence (DCS) assembly. The DCS outputs were finally realigned to the reference sequence to count mutations. Previously defined default parameters were used for this process [31].
As described previously, the frequency of lethal mutations in a population should equal the rate at which these mutations are produced [4, 26–29]. For ELVd, we assumed that all mutations in the hammerhead ribozyme core sites defined in Fig 4 should be lethal [4], and the same assumption was made for the PSTVd CCR/TCR sites analyzed. The overall mutation rate was simply estimated as μ=∑i=1TNi∑i=1TCi, where Ni is the number of mutations at site i, Ci is sequencing coverage at site i, and T is the number of sites analyzed. For the 23 mutations that were previously reported to have lethal effects in PSTVd [40–44], the estimation was more complicated because for most of the sites only one or two of the three possible base substitutions could be used for mutation rate estimation, as the other substitutions were not reported lethals. To account for this, the mutation rate was estimated as μ=∑i=1TNi∑i=1T∑j=1kρjρiCi, where ρj is the contribution of the specific mutation considered to the total mutation spectrum of PSTVd, k is the number of different lethal substitutions at site i (k = 1, 2, or 3) and ρi is the contribution of the three possible base substitutions at this site to the total mutational spectrum. These coefficients are provided in Table 2 for sequences obtained from circular (+) and oligomeric (–) forms. For instance, if at a given site only C-to-U substitutions were lethal, k = 1 and, for sequences from circular (+) forms, we used ρj = 1 = 25.7 (i.e. the percentage of C-to-U mutations in the total spectrum) and ρi = 25.7 + 4.9 + 3.3 (i.e. the percentages of C-to-U, C-to-A, and C-to-G mutations). Notice that this formula could also be used for estimating mutation rates in the PSTVd CCR/TCR and ELVd hammerhead ribozymess more precisely, but this was not necessary as long as base composition and mutational spectra are similar for these regions and the rest of the viroid sequence.
|
10.1371/journal.ppat.1006383 | Detection of a microbial metabolite by STING regulates inflammasome activation in response to Chlamydia trachomatis infection | The innate immune system is a critical component of host defence against microbial pathogens, but effective responses require an ability to distinguish between infectious and non-infectious insult to prevent inappropriate inflammation. Using the important obligate intracellular human pathogen Chlamydia trachomatis; an organism that causes significant immunopathology, we sought to determine critical host and pathogen factors that contribute to the induction of inflammasome activation. We assayed inflammasome activation by immunoblotting and ELISA to detect IL-1β processing and LDH release to determine pyroptosis. Using primary murine bone marrow derived macrophages or human monocyte derived dendritic cells, infected with live or attenuated Chlamydia trachomatis we report that the live organism activates both canonical and non-canonical inflammasomes, but only canonical inflammasomes controlled IL-1β processing which preceded pyroptosis. NADPH oxidase deficient macrophages were permissive to Chlamydia trachomatis replication and displayed elevated type-1 interferon and inflammasome activation. Conversely, attenuated, non-replicating Chlamydia trachomatis, primed but did not activate inflammasomes and stimulated reduced type-1 interferon responses. This suggested bacterial replication or metabolism as important factors that determine interferon responses and inflammasome activation. We identified STING but not cGAS as a central mediator of interferon regulated inflammasome activation. Interestingly, exogenous delivery of a Chlamydia trachomatis metabolite and STING ligand—cyclic di-AMP, recovered inflammasome activation to attenuated bacteria in a STING dependent manner thus indicating that a bacterial metabolite is a key factor initiating inflammasome activation through STING, independent of cGAS. These data suggest a potential mechanism of how the innate immune system can distinguish between infectious and non-infectious insult and instigate appropriate immune responses that could be therapeutically targeted.
| Innate responses to bacterial infection such as Chlamydia trachomatis activate inflammasomes to enable the processing of IL-1β, IL-18 and the induction of an inflammatory form of cell death termed pyroptosis. Inflammasomes are crucial to host defence but require tight regulation in order to prevent inappropriate inflammation and immunopathology. Here, we demonstrate that the pro-inflammatory potential of an attenuated strain of Chlamydia trachomatis, that fails to activate the inflammasome, can be rescued by the addition of a bacterial metabolite. The requirement for this metabolite, highlights a novel mechanism of inflammasome regulation and reveals a crucial role for STING mediated interferon signalling independent of cGAS. These findings further our understanding of how the innate immune system can differentiate between potential infectious and non-infectious threats and mount appropriate immune responses.
| The obligate intracellular pathogen Chlamydia trachomatis is a major cause of infectious disease world-wide and can initiate inflammatory pathology such as pelvic inflammatory disease, reactive arthritis and infectious blindness (trachoma). Significantly, murine models of Chlamydia infection demonstrate that host inflammatory mediators, particularly the inflammatory cytokine IL-1β, type-1 interferons, caspase-1 and caspase-11 account for a significant proportion of infection associated pathology [1–3].
Inflammasomes are molecular scaffolds that facilitate the activation of inflammatory caspases resulting in the proteolytic processing of the cytokines IL-1β and IL-18 in addition to the induction of a form of programmed necrosis termed pyroptosis [4, 5]. Recently, a non-canonical inflammasome requiring caspase-11 (caspase-4/5 in humans) has been identified that responds specifically to LPS contamination of the cytosol, independent of TLR4 [6–9]. Uniquely, activation of the non-canonical inflammasome does not require an upstream sensor that is required for canonical caspase-1 activation, and occurs as a consequence of LPS being directly recognised by Caspase-11 (or Caspase-4/5) [10] and is critically dependent on the acylation status of the lipid-A moiety [9]. The non-canonical inflammasome is essential for pyroptosis and IL-1β maturation in response to infection with certain gram-negative bacterial pathogens, or the delivery of cytoplasmic LPS, and occurs as a consequence of the caspase-11 dependent cleavage of gasdermin-D (GSDMD) [11, 12] and pannexin-1 [13]. Activation of the non-canonical inflammasome in response to bacterial pathogens can occur as a direct result of cytosolic invasion by bacteria that occurs during infection with Burkolderia thailandensis, resulting in a rapid execution of caspase-11 mediated pyroptosis, essential to host defence [14]. Alternatively, non-canonical inflammasome activation can occur as a consequence of LPS being released from bacterial pathogen containing vacuoles (PV), via the action of Immune Related GTPases (IRG’s) and Guanylate Binding Proteins (GBP’s) [15].
Despite type-1 interferons being reported to inhibit canonical inflammasome activation [16], non-canonical inflammasomes require interferon signalling for up-regulation of caspase-11 [17], GBP and IRG expression [18].
Mechanisms controlling inflammasome activation in response to Chlamydia sp infection are still poorly understood, although NLRP3, reactive oxygen species, mitochondrial damage and oxidised mitochondrial DNA have been implicated in the process [19, 20] [21, 22]. In addition, recent work using interferon-primed macrophages has demonstrated that GBP’s, NLRP3 and AIM2 participate in IL-18 maturation in response to Chlamydia trachomatis infection, while cell death occurs via either caspase-1 or caspase-11 activity, suggesting engagement of the non-canonical inflammasome [23]. Importantly, Chlamydia muridarum infection of macrophages induces type-1 interferon expression via Stimulator of Interferon Gene (STING) activity, implying that Chlamydia sp infections of macrophages are ‘self-priming’ and exogenous interferon is not necessary [24]. However, mechanisms controlling this process are poorly understood.
Here, we report that inflammasome activation in un-primed bone marrow derived macrophages (BMDM) required replication or metabolic activity of Chlamydia trachomatis. We also provide evidence that production of a chlamydial metabolite, but not detection of host or microbial DNA by cGAS, activated STING initiating autocrine, type-1 interferon signalling leading to canonical and non-canonical inflammasome activation. We propose that intracellular Chlamydia trachomatis replication or metabolism, and the subsequent detection of cyclic di-AMP by STING, are key events initiating inflammasome activation in response to Chlamydia trachomatis infection.
We sought to identify whether canonical and non-canonical inflammasomes regulate cytokine maturation and cell death in response to C. trachomatis infection of un-primed BMDM using cytosolic delivery of LPS as a control for non-canonical inflammasome activation. A time course infection of BMDM with C. trachomatis resulted in significant increases in IL-1β secretion as early as 8–10 hours post infection in wild-type and caspase-11 deficient but not caspase-1 deficient BMDM indicating that the non-canonical inflammasome was not required for IL-1β maturation in response to C. trachomatis infection (Fig 1A). We next analysed LDH release as an indicator of lytic cell death, associated with pyroptosis and necrosis over the same time course of C. trachomatis infection (Fig 1B). As expected C. trachomatis infection induced significant cytotoxicity in wild type BMDM but surprisingly, this was delayed compared to IL-1β release and individual deletion of either caspase-11 or caspase-1 did not confer resistance to cell death suggesting a potential non-pyroptotic cell death mechanism. However, deletion of both caspase-1 and caspase-11 resulted in a dramatic protection of cell death indicating that both inflammatory caspases contribute to the process and the form of cell death is therefore likely to be classical pyroptosis as both caspase-1 and caspase-11 have been demonstrated to cleave and activate the pore forming protein gasdermin D in order to initiate pyroptotic cell death [11, 12]. Intriguingly, despite the ability of both caspase-1 and caspase-11 to induce cell death in response to C. trachomatis infection, death appeared to be delayed in the caspase-1 ko / caspase-11 transgenic cells, particularly at the 20-hour time point. This suggests that caspase-1 may induce early cell death while caspase-11 contributes later. This is an attractive answer to the conundrum of why caspase-11 does not affect IL-1β but contributes to cell death. Late activity of caspase-11 may not be required for the NLRP3 dependent activation of caspase-1 that controls cytokine maturation, as this may have already occurred through canonical inflammasome activation. We confirmed the functional status of the caspase deficient cells by specific activation of the non-canonical inflammasome using cytosolic delivery of LPS. As expected, intracellular LPS induced IL-1β secretion was dependent on both capase-1 and caspase-11 (Fig 1C) while cell death was only dependent on caspase-11 (Fig 1D) as previously reported [6, 8, 9].
We next examined which canonical intracellular sensor activated caspase-1 in response to C. trachomatis infection. Both NLRP3 [19, 20] and AIM2 [23] are reported to be activated in response to C. trachomatis infection enabling caspase-1 activation and the processing of IL-1β. We therefore confirmed the requirement for both inflammasomes in our study (Fig 2A, 2B, 2C and 2D). Both NLRP3 and AIM2 deficient macrophages displayed reduced IL-1β responses to C. trachomatis infection as expected (Fig 2A and 2C). However, the absence of AIM2 but not NLRP3 had a small but significant effect on cell death in response to infection potentially accounting for the small reduction in cell death we observed with caspase-1 deficiency alone, but also reinforcing our findings that in contrast to IL-1β release, cell death was not solely dependent on caspase-1 activity (Fig 2B and 2D).
Mitochondrial dysfunction is known to contribute to inflammasome activation by sterile agonists [25] and Chlamydia sp infection [21, 22]. We therefore investigated whether canonical activation of caspase-1 resulting in IL-1β processing was inhibited by the presence of a mitochondrial anti-oxidant (Mito-Q) (S1A, S1B and S1C Fig). Mito-Q, in a pattern similar to NLRP3 deficiency, inhibited C. trachomatis induced IL-1β release, but had no effect on cell death, most likely due to the overriding effect of caspase-11 and further supporting our findings that C. trachomatis induced death could be regulated by both caspase-1 and caspase-11. Importantly, MitoQ did not block pro IL-1β upregulation by C. trachomatis indicating that MitoQ was blocking caspase-1 mediated IL-1β processing and not priming (S1B Fig). Infection of macrophages with Brucella abortus activates the NLRP3 inflammasome in a mechanism requiring IRE-1 mediated ER stress, mitochondrial damage and caspase-2 activity [26]. We have previously demonstrated that C. trachomatis infection of human dendritic cells results in IRE1α activation [27]. We therefore examined if IRE1α was relevant to C. trachomatis induced inflammasome activation by utilising IRE-1 deficient macrophages or the IRE-1 inhibitor; 4μ8c. IRE-1 deficient BMDM or BMDM pre-treated with 4μ8c demonstrated normal IL-1β release in response to C. trachomatis infection (S1D and S1E Fig) indicating that ER stress driven inflammasome activation was not relevant to C. trachomatis infection and the mitochondrial component differs from that of B. abortus. Furthermore, our data suggest that canonical activation of caspase-1 through release of DNA (either host or pathogen derived) in to the cytosol activating AIM2 is amplified by mitochondrial ROS to activate NLRP3 and induce maximal inflammasome responses to C. trachomatis. Both NLRP3 and AIM2 share the requirement for the adaptor ASC to provide a scaffold for the recruitment of caspase-1. The fact that deletion of either AIM2 or NLRP3 only partially attenuated inflammasome responses and that only one ASC speck is formed per cell suggests that maximal inflammasome response to C. trachomatis requires the co occupation of the ASC filament by both receptors. However, it remains unknown whether recruitment of both receptors occurs simultaneously to initiate the aggregation of the ASC complex resulting in an active inflammasome during infection.
We next investigated whether inflammasome responses to C. trachomatis infection required factors produced by the live-replicating organism. We attenuated C. trachomatis using gamma-irradiation or used a Chlamydia Protease-like Activity Factor (CPAF) deficient mutant strain that exhibited delayed growth kinetics in Hela cells [28]. Attenuated C. trachomatis (both gamma-irradiated and the CPAF mutant) failed to induce IL-1β secretion and cell death (Fig 3A and 3C and S2 Fig) but still induced pro IL-1β expression indicating that the attenuated C. trachomatis particles were not biologically inert and pathogen recognition receptor (PRR) and NFκB responses were intact, but an inflammasome activatory signal was absent (Fig 3B). Importantly, we confirmed that lack of inflammasome activation was not due to reduced internalisation of attenuated C. trachomatis particles as intracellular staining demonstrated equivalent levels to live C. trachomatis (S3 Fig) to the attenuated forms. We also noted that although pro IL-1β priming was normal in response to attenuated C. trachomatis, expression of caspase-11 was markedly reduced in cells stimulated with irradiated but not CPAF deficient C. trachomatis. This indicated that replication or metabolism of C. trachomatis intracellularly, provided additional, essential signals required for inflammasome activation and caspase-11 expression in addition to CPAF function.
Our experiments utilising attenuated C. trachomatis indicated that intracellular replication could be contributing to signals that activate the inflammasome. We therefore wished to identify macrophages that were permissive to C. trachomatis replication to investigate this hypothesis further.
The NADPH oxidase system is a key component of host defence against microbial pathogens via the generation of anti-microbial reactive oxygen species (ROS). Individuals with inherited, disabling mutations of genes that contribute to the NADPH oxidase system, develop a primary immunodeficiency termed chronic granulomatous disease (CGD) that is characterised by recurrent bacterial and fungal infections [29]. Generation of ROS by NADPH oxidase contributes to the anti-microbial action of macrophages in response to infection with bacteria such as Salmonella [30]. Previous studies using chemical inhibition of NADPH oxidase in Hela cells suggested that ROS amplify C. trachomatis replication through a caspase-1 mediated mechanism [31]. However, the role of NADPH oxidase in the control of C. trachomatis replication within primary macrophages is unknown.
We therefore investigated Chlamydia trachomatis growth in murine macrophages or human monocyte derived dendritic cells obtained from CGD patients deficient in gp91 Phox (Cybb); a critical component of the NADPH oxidase system in phagocytic cells.
Intracellular staining of C. trachomatis LPS and analysis of Chlamydia 16s RNA expression was employed to determine the bacterial burden within infected cells Significantly, infection of macrophages deficient in gp91 phox exhibited a marked increase in the burden of C. trachomatis compared to wild-type controls (Fig 4A). The intracellular staining of Chlamydia LPS in murine macrophages (Fig 4B) or human dendritic cells (Fig 4C) was also quantified using FACS and further confirmed by analysis of Chlamydia 16s RNA in BMDM (S4 Fig), a technique employed by others to analyse Chlamydia replication [31]. Both murine and human cells deficient in gp91 phox displayed a significant increase in intracellular C. trachomatis burden compared to control (gp91 phox sufficient) cells, indicating that NADPH oxidase activity was a key regulator of C. trachomatis replication and survival within myeloid cells in both mice and humans.
Given that we had observed an apparent requirement for C. trachomatis replication or metabolism in the induction of inflammasome activation, we tested the hypothesis that increased C. trachomatis replication observed in NADPH oxidase deficient cells, would correspond with increased inflammasome activation. C. trachomatis infection of dendritic cells from CGD patients (Fig 5A and 5B) or gp91 phox deficient murine macrophages (Cybb-/-) (Fig 5C, 5E and 5F) resulted in increased IL-1β maturation and pyroptosis indicating increased inflammasome activation. Furthermore, expression of IL-1β mRNA was equivalent in wild-type and gp91 phox deficient cells in response to infection indicating that the increased bacterial burden was having a direct effect on inflammasome activation, and was not just due to increased priming (Fig 5D). Crucially, inflammasome responses to sterile agonists such as LPS/ATP or gamma attenuated C. trachomatis, were not elevated in the absence of NADPH oxidase activity, indicating that the increased bacterial burden was the critical factor leading to inflammasome activation and that this was not simply a consequence of elevated background inflammasome activation in the absence of NADPH oxidase. These data also reinforced our hypothesis that aspects of C. trachomatis metabolism or replication were factors influencing inflammasome activation.
Recently, type-1 interferon signalling has been demonstrated to play an important role in host defence and inflammasome activation to bacterial pathogens [32] and is crucial for the expression of caspase-11 in response to a range of gram-negative bacteria [17]. We therefore investigated whether intracellular metabolism or replication of C. trachomatis within macrophages contributed to type-1 interferon expression. Significantly, gamma-irradiated attenuated C. trachomatis or the CPAF deficient mutant (S5 Fig) failed to induce significant interferon-β expression compared to non-attenuated bacteria (Fig 6A). Furthermore, gp91 phox deficient macrophages, harbouring increased bacterial burdens also displayed elevated interferon-β expression compared to wild-type controls (Fig 6B). These data indicated that type-1 interferon responses could be critical mediators of inflammasome activation in macrophages during C. trachomatis infection. We therefore tested the hypothesis that autocrine, type-1 interferon signalling through the type-1 interferon receptor, IFNAR, contributed to inflammasome activation during C. trachomatis infection. To do this, we infected macrophages with C. trachomatis in the presence of an IFNAR blocking antibody or isotype control. We confirmed the effectiveness of blocking IFNAR by examining STAT-1 phosphorylation in response to interferon-β stimulation of wild-type BMDM (Fig 6C). Blocking IFNAR during C. trachomatis infection resulted in reduced IL-1β secretion (Fig 6D), IL-1β maturation, caspase-11 expression (Fig 6E) and cell death (Fig 6F) but, as expected, did not significantly affect canonical inflammasome activation with LPS/ATP. Furthermore, the effect on IL-1β secretion was of reduced proteolytic processing and not priming of pro IL-1β, as blocking IFNAR during C. trachomatis infection actually resulted in increased pro IL-1β, presumably due to reduced maturation by caspase-1 and accumulation of the un-processed substrate and the inhibitory effect of interferon signalling on pro IL-1β synthesis [16] (Fig 6E). These data suggested that C. trachomatis replication or metabolism induced a type-1 interferon signature that controlled both canonical and non-canonical inflammasome activation.
Stimulator of interferon gene (STING) is a critical mediator of type-1 interferon expression in response to Chlamydia sp infection [24, 33, 34]. Given that we have demonstrated a crucial role for autocrine type-1 interferon signalling in C. trachomatis induced inflammasome activation, we tested the hypothesis that STING was a central regulator of inflammasome activation. Infection of macrophages from STING deficient mice with C. trachomatis failed to induce interferon-β expression (Fig 7A) and exhibited reduced IL-1β secretion (Fig 7B), cell death (Fig 7C) and reduced IL-1β processing and caspase-11 expression (Fig 7D) but had equivalent responses to LPS/ATP stimulation. Thus, STING is a critical mediator of interferon dependent inflammasome activation in response to C. trachomatis infection of macrophages.
Activation of STING to induce type-1 interferon responses can occur through two distinct pathways: 1. the conversion of cytosolic DNA to cGAMP catalysed by cGAS which is then recognised by STING or 2. the direct recognition of cyclic di-nucleotides (cyclic di-AMP/GMP) produced by certain bacteria. Recently, cGAS was shown to be crucial for STING dependent inflammasome responses to Francisella tularensis infection of macrophages [35, 36]. We therefore analysed whether cGAS was required for STING mediated inflammasome activation in response to C. trachomatis infection using cGAS deficient BMDM. In contrast to inflammasome responses to F. tularensis, cGAS deficient BMDM produced elevated IL-1β in response to C. trachomatis infection indicating that STING dependent canonical caspase-1 activation through NLRP3 and AIM2 activation was independent of cGAS conversion of host or microbial DNA to cGAMP (Fig 8A). Surprisingly however, there was a small but significant protection from C. trachomatis induced pyroptosis in cGAS deficient cells (Fig 8B). Given that we demonstrate cell death responses are delayed compared to IL-1β release and death is governed by both canonical activation of caspase-1 and non-canonical caspase-11 activation, suggests that later release of microbial DNA to the cytosol may amplify STING responses that are not required for canonical inflammasome activation.
STING induced interferon responses during C. trachomatis infection, have been demonstrated to occur as a consequence of recognition of a metabolite; cyclic di-AMP, that is produced by metabolically active C. trachomatis [33]. Given that we identified STING as a critical regulator of canonical and non-canonical inflammasome responses to C. trachomatis infection and that cGAS was not required for canonical caspase-1 activation to induce IL-1β release in addition to finding a requirement for C. trachomatis metabolism or replication in the induction of interferon-β expression and subsequent inflammasome activation, we investigated whether cyclic di-AMP was a critical factor in STING mediated activation of the inflammasome. To test this hypothesis, we stimulated wild-type BMDM with attenuated C. trachomatis that failed to induce inflammasome activation and transfected the metabolite cyclic di-AMP in to the cells. Crucially, transfection of titrated amounts of cyclic di-AMP in to cells stimulated with attenuated gamma irradiated (Fig 9A) or CPAF deficient (Fig 9B) C. trachomatis resulted in partial recovery of IL-1β secretion that was dependent on STING (Fig 9C). These data provide an attractive explanation as to how STING mediated inflammasome responses could be regulated in the absence of a requirement for cGAS conversion of DNA to cGAMP during infection with C. trachomatis that requires a replicating or metabolically active organism.
Innate host defence relies upon a repertoire of pathogen recognition systems that detect a diverse array of conserved microbial components in order to initiate inflammatory responses. Here, we provide evidence that inflammasome activation in response to Chlamydia trachomatis infection is dependent on detection of a microbial metabolite produced by the live organism. The inflammasome is an important component of innate host defence against bacterial pathogens including Chlamydia sp, but excessive or inappropriate activation contributes to infection associated immunopathology. Therefore, inflammasome activation must be tightly regulated. Our data demonstrate a novel mechanism of inflammasome regulation that allows the host to differentiate between live (a potential infectious threat) or attenuated Chlamydia trachomatis and mount appropriate responses. We suggest that this distinction is critical for effective host defence, but also regulates inappropriate or excessive responses by only responding to live organisms. Conversely, these data also suggest a potential mechanism of immune evasion that could be adopted by C. trachomatis by modulating production of cyclic di-AMP. Bacterial pathogens are known to evade non-canonical inflammasome activation by modulating the acylation status of their lipid-A moieties [9], and we suggest that modulating production of inflammasome activating metabolites could also be a strategy utilised by pathogenic bacteria to evade this response.
We have shown that STING, independent of cGAS, is a critical component of interferon mediated inflammasome activation in response to C. trachomatis infection, and differs from recent studies utilising F. tularensis that required conversion of bacterial DNA to cGAMP, rather than detection of a microbial metabolite to induce STING mediated AIM2 inflammasome activation [35, 36]. These previous studies raise an important paradoxical question because AIM2 activation also requires the release of DNA from the pathogen in an interferon dependent process, yet the interferon signal also requires conversion of this same DNA by cGAS to induce the STING mediated interferon signal. This would suggest that either cGAS is a more sensitive sensor of DNA than AIM2 and responds to low levels within the cytosol that would not initiate AIM2 activation or that something else is providing the initiating interferon signal leading to loss of compartmentilisation and amplification of responses. We envisage that in our model—microbial metabolism and the production of cyclic di-AMP precedes any involvement of cGAS that is only partially required for cell death responses. Instead, cyclic di-AMP induction of STING activation results in small, but significant increases in type-1 interferon expression that facilitates the up-regulation of IRG’s and GBP’s that are known to be recruited to Chlamydial containing vacuoles [37–39]. Damage to the Chlamydia containing vacuole, mediated by IRG’s and GBP’s, could then result in the further release of vacuole contents including microbial DNA that further enhances STING activation through cGAS [34] and also activates the AIM2 inflammasome as we and others have reported recently [23]. In this regard, we would not discount cGAS from C. trachomatis induced inflammasome activation, but would suggest it enhances cell death responses, rather than initiates, the response. Furthermore, our experiments using attenuated C. trachomatis support this hypothesis, as attenuated organisms did not activate the inflammasome despite containing a full complement of unaltered DNA and LPS that would be immunostimulatory if detected by cytosolic receptors. This suggests that a loss of compartmentalisation of the Chlamydia containing vacuole is a key event that delivers inflammasome activatory ligands to the cytosol. Thus release of cyclic di-AMP activates STING which in turn mediates type-1 interferon signalling.
We also provide evidence that cell death observed during C. trachomatis is true pyroptosis. Deletion of both caspase-1 and caspase-11 prevented cell death and this observation would rule out other forms of lytic cell death such as necroptosis. It is unusual that both caspases could induce pyroptosis, but this is not without precedent [40], and has also been reported for Chlamydia infection by others independently of our studies [23]. We think it is also important that cell death was delayed compared to IL-1β release and suggests IL-1β release is controlled via an active process rather than accidental release as a consequence of cell death, although we cannot rule out that sensitivity of the assays employed could also contribute to these findings. However, in support of our findings, Recent studies have identified that cleavage of GSDMD by inflammatory caspases is a critical mediator of pyroptosis [11, 12, 41] and IL-1β release [12, 41]. It is also hypothesised that the number of gasdermin pores formed with the plasma membrane of the cell could control the balance between release of mature cytokine and commitment to cell death so that fewer pores allow release of IL-1β without cell death [42]. This would be an attractive explanation of the asynchronous behaviour of IL-1β release and cell death responses observed during C. trachomatis infection. Given that C. trachomatis induced pyroptosis could utilise both caspases, the role of GSDMD would be an interesting line of enquiry and will form part of our future studies. Furthermore, although inflammatory caspases are known to play a role in Chlamydia induced immunopathology [3], the role of pyroptosis has yet to be defined.
Finally, we have demonstrated a striking role for the control of intracellular growth of C. trachomatis by the NADPH oxidase system in both human and murine cells. Increased intracellular growth correlated with elevated inflammasome activation reinforcing our hypothesis that C. trachomatis replication or metabolism is a critical factor inducing inflammasome activation. There are very few published data investigating the role of NADPH oxidase during Chlamydia infection. Previous studies have utilised chemical inhibition to study ROS function during Chlamydia infection of Hela cells which suggested that ROS were important for C. trachomatis replication within these cells [31]. However, Hela cells provide a wholly different cellular environment to primary macrophages and C. trachomatis exists within a specially constructed pathogen containing vacuole that may well protect the organism from the toxic effects of ROS generation within non-phagocytic cells. It is not well defined which compartment C. trachomatis occupies within macrophages. We speculate from our work that this compartment is likely to be formed as a consequence of phagocytic activity as uptake of the bacteria was independent of high speed centrifugation often employed to achieve infection of non-phagocytic cells such as Hela. We would therefore suggest that Chlamydia trachomatis is likely to be existing within a different compartment in macrophages compared to Hela cells and therefore may be more susceptible to the toxic effects of ROS generated through NADPH oxidase activity which could account for the intriguing differences between our findings and others. Intriguingly, other members of the Chlamydiaceae family have been shown to express functional catalase activity [43] suggesting that ROS is an important host defence mechanism against Chlamydia infection and may limit the replicative ability of C. trachomatis within phagocytic cells such as macrophages. Importantly, a recent report has linked macrophage pyroptosis and neutrophil oxidative killing of bacteria as an important axis in innate host defence against bacterial infection [44]. Neutrophils from CGD patients display reduced bactericidal effects on Chlamydia trachomatis [45]. We would therefore predict that CGD patients may be at risk from more severe immunopathology during C. trachomatis infection due to elevated inflammasome responses of macrophages and reduced bacterial killing by neutrophils. This would also be in agreement with the consensus that CGD patients have a pro-inflammatory phenotype [46].
In summary, we propose that intracellular replication of Chlamydia trachomatis and production of the metabolite cyclic di-AMP is a key pathogen associated molecular pattern detected by STING that is crucial for activation of both canonical and non-canonical inflammasomes. This requires type-1 interferon and allows the host to initiate appropriate immune responses. We also suggest that modulation of cyclic di-AMP production by Chlamydia trachomatis could provide a mechanism of immune evasion and contribute to mechanisms of infection latency. Finally, targeting of the STING/interferon pathway may provide useful vaccine adjuvant and therapeutic targets to aid the treatment of Chlamydia trachomatis infection and its associated inflammatory pathology.
All studies involving human subjects were performed in accordance with the Declaration of Helsinki, with approval of the Cambridge Regional Ethical Committee (01/363). All donors gave written informed consent.
All animal related work was conducted by trained and appropriately licensed staff, under the authority and conditions of a Home Office project licence. This licence is issued under the UK Animal (Scientific Procedures) Act 1986 (ASPA), following local ethical approval. Home Office Inspectors provide governmental supervision of the work. Local ethical approval and supervision of standards of work and animal husbandry is carried out by the University of Cambridge Animal Welfare Ethical Review Body (AWERB) and its delegated representatives. Animals used for tissue were culled by fully trained personnel using approved humane methods under Schedule 1 of ASPA.
Bone marrow derived macrophages (BMDM) were prepared from femurs of littermate wild type and genetically deficient animals by culturing bone marrow isolates for 7-days in RPMI containing 10% v/v HIFCS, L-glutamine, 5% v/v L929 conditioned medium with gentamicin. Of note, we became aware that macrophages produced from mice housed in different facilities displayed extremely variable responses to stimulation. It was therefore imperative that comparisons between wild-type and knock-out animals were made using littermates or, as a minimum, using age and sex matched mice housed in the same facility, and not between mice housed at different facilities. Unprimed macrophages were infected with live Chlamydia trachomatis at a multiplicity of infection (MOI) of 20. Attenuated Chlamydia trachomatis was achieved by γ-irradiating live C. trachomatis elementary bodies purified by high speed density centrifugation for 6-hours in a Gammacell-1000 irradiator (Atomic Energy of Canada Ltd) or using a Chlamydia Protease Activity like Factor (CPAF) deficient C. trachomatis mutant (ΔCPAF) described previously [28]. Attenuated C. trachomatis were used at a multiplicity of infection of 20 and attenuation was determined by the absence of replication in Hela cells. Intracellular LPS stimulation of macrophages was achieved by priming BMDM with 100ng/ml LPS for 4-hours in L929 conditioned RPMI. After priming, the supernatants were removed and replaced with serum free optiMEM (Gibco) containing 5μg/ml LPS with 5μl/ml Fugene (Promega). Cells were incubated for 24-hours before supernatants were harvested for ELISA and LDH assays. Recovery of inflammasome activation with attenuated C. trachomatis was achieved by stimulation of BMDM for 4-hours with γ-attenuated or CPAF deficient C. trachomatis mutant (ΔCPAF) in L929 conditioned RPMI as per intracellular LPS stimulation. The supernatant and non-internalised C. trachomatis were then removed and the media replaced with optiMEM containing cyclic-di-AMP (Invivogen, France) at indicated concentrations with Fugene. BMDM were then incubated for 24hrs before supernatants were harvested for ELISA.
Human monocyte derived dendritic cells (mDC) were obtained by IL-4/GMCSF differentiation [27, 47] of peripheral blood monocytes from age and sex matched chronic granulomatous disease (CGD) patients (recruited at the Royal Free Hospital, London) or healthy donors, (recruited at the University of Cambridge Department of Medicine, Cambridge) conforming to ethical guidelines of each institution.
Assessment of relative amounts of intracellular C. trachomatis was determined by fluorescent staining of Chlamydia LPS. BMDM or mDC were plated on coverslips (microscopy) or without coverslips (FACS) in 24-well plates at 2.5x105 cells/well and infected as described above for 24-hours. Intracellular staining of LPS was achieved by washing cells three times in PBS before fixation and permeabilisation using CellFix (BD Bioscience, USA) and incubating cells at room temperature with 1μg/ml FITC conjugated monoclonal mouse anti-chlamydia LPS (SourceBioscience U.K.). Cells were washed a further three times and mounted on slides using DAPI counterstain for fluorescent microscopy or scraped and assayed by FACS (FacsCalibur, BD Bioscience, USA). Uninfected cells were stained as described and acted as a negative control.
Detection of mature IL-1β in cell culture supernatants was achieved by ELISA following the manufacturer’s instructions (Mouse or human Ready-SET-Go IL-1β ELISA, Ebioscience, USA). Macrophages were seeded in 96-plates at 1x105 cells/well and cultured for 24hrs in the presence of cell stimulations. For assessment of cell death, a lactate dehydrogenase (LDH) release assay was employed (promega U.K.) following the manufacturer’s instructions. The percentage cytotoxicity was calculated from the absorbance of the test well divided by the absorbance of the corresponding 100% lysis control wells.
For analysis of protein expression cytosolic protein extracts and cell culture supernatants were assayed by SDS PAGE and western blotting as described previously [27]. Briefly, equal amounts of cytosolic protein, determined by Bradford Assay (Thermo U.K.) was mixed with an appropriate volume of 6x reducing buffer and boiled for 10 minutes before loading on to 4–20% gradient SDS pre-cast gels (BioRad U.K.). Separated protein was then transferred to PVDF membrane (BioRad U.K.) using a Transblot Turbo (BioRad U.K.). Membranes were blocked for 1-hour at RT in 5% w/v milk protein before incubation with appropriate primary antibody overnight at 4°C. Membranes were washed 3 times in TBS-TWEEN (0.05%v/v) before incubating with appropriate secondary HRP conjugated antibodies for 1-hour at RT. Membranes were developed via ECL (Lightning super signal, Perkin Elmer, USA) and visualised using a G-box (Syngene U.K.) or chemiluminescence film (Amersham Hyperfilm GE Healthcare U.K). Membranes were stripped using low pH stripping buffer for 30-minutes at RT followed by blocking and incubation with antibody. The following primary antibodies were used in this study: Monoclonal rat anti-mouse Caspase-11 (clone 17D9) (Sigma U.K.), Goat anti-mouse IL-1β (R&D Systems U.K.), Goat anti-human IL-1β (R&D Systems U.K.) and monoclonal mouse anti-actin (Abcam U.K.).
Macrophages were cultured at 0.5x106 cells per well of 24-well plate and were infected with an MOI of 20 live C. trachomatis, γ-irradiated C. trachomatis or a CPAF deficient C. trachomatis for 8hrs. Cells were lysed via addition of RNA lysis buffer (Norgen, Canada) directly to the well. RNA was purified using the Norgen RNA kit according to manufacturer’s instructions. For interferon-β and IL-1β expression qRT-PCR was employed using commercial probe/primer sets (LifeTechnologies U.K.) and analysed using the Taqman ‘one-step’ system (LifeTechnologies U.K.). Analysis of C. trachomatis 16s expression was determined by quantitative PCR of cDNA prepared from total RNA extracted as above.
All numeric data were analysed using Graphpad Prism (USA). Analysis of multiple data groups with single variables was analysed using 1-way ANOVA with Dunnet’s post-test while multiple data groups with two variables was analysed using a 2-way ANOVA with Bonferonni’s post-test. Comparison of data between two-paired data sets was analysed using a paired Student’s t-test. All data are represented as ±SEM of the mean performed on BMDM obtained from at least three individual mice or mDC obtained from at least three individual human donors unless stated otherwise. A p value <0.05 was deemed significant.
|
10.1371/journal.pntd.0002086 | A Neglected Aspect of the Epidemiology of Sleeping Sickness: The Propensity of the Tsetse Fly Vector to Enter Houses | When taking a bloodmeal from humans, tsetse flies can transmit the trypanosomes responsible for sleeping sickness, or human African trypanosomiasis. While it is commonly assumed that humans must enter the normal woodland habitat of the tsetse in order to have much chance of contacting the flies, recent studies suggested that important contact can occur due to tsetse entering buildings. Hence, we need to know more about tsetse in buildings, and to understand why, when and how they enter such places.
Buildings studied were single storied and comprised a large house with a thatched roof and smaller houses with roofs of metal or asbestos. Each building was unoccupied except for the few minutes of its inspection every two hours, so focusing on the responses of tsetse to the house itself, rather than to humans inside. The composition, and physiological condition of catches of tsetse flies, Glossina morsitans morsitans and G. pallidipes, in the houses and the diurnal and seasonal pattern of catches, were intermediate between these aspects of the catches from artificial refuges and a host-like trap. Several times more tsetse were caught in the large house, as against the smaller structures. Doors and windows seemed about equally effective as entry points. Many of the tsetse in houses were old enough to be potential vectors of sleeping sickness, and some of the flies alighted on the humans that inspected the houses.
Houses are attractive in themselves. Some of the tsetse attracted seem to be in a host-seeking phase of behavior and others appear to be looking for shelter from high temperatures outside. The risk of contracting sleeping sickness in houses varies according to house design.
| To explore the nature of houses as venues for the contact between humans and tsetse flies, and hence for the transmission of sleeping sickness, we studied the sex and species composition and physiological condition of samples of tsetse caught in various types of house throughout the day and at different seasons. These aspects of the catches were intermediate between those for traps which caught host-orientated flies and artificial refuges that sampled flies seeking a cool dark resting site. This suggested that some flies entered houses in search of food, and others entered for shelter. Windows seemed about as effective as doors as entry points. Several times more tsetse were found in a large thatched house, compared to smaller houses with asbestos or metal roofs. Many of the tsetse in houses were old enough to be potential vectors of sleeping sickness. Some of the tsetse inside alighted on people that inspected the houses.
| Sleeping sickness, or human African trypanosomiasis, is caused by two species of trypanosome, i.e., Trypanosoma brucei gambiense and T. b. rhodesiense, that are transmitted by tsetse flies (Glossina spp.) when taking blood from hosts [1]. It seems to have been assumed that the risk of humans being bitten by tsetse is by far the greatest when people enter the normal woodland habit of the flies. In keeping with this, almost all of the data available for the nature of the contact between humans and tsetse relate to humans in woodland, especially to people walking through it [2]. Such data indicate that the samples of tsetse caught from humans usually contain high proportions of males which appear to be seeking a mate rather than food [3]. Hence, while many tsetse can occur in the vicinity of humans, the risk of a human being bitten is usually very low.
However, a recent investigation of the numbers of Glossina morsitans morsitans and G. pallidipes that actually attempted to feed on humans in various situations indicated that the risk of humans being bitten in woodland was less than the risk occurring when the humans were in or near houses and offices located in large clearings [4]. Moreover, the same work showed that the proportion of females among the tsetse probing humans in the buildings was consistently higher than among tsetse probing people in woodland settings away from buildings. The upshot is that buildings seem to be important, distinctive and neglected venues for the transmission of sleeping sickness, and this leads to many questions. Why are tsetse found in buildings? Do they enter only at certain seasons and times of day? Are some types of building more important than others? How does the sex, species and age compositions of samples of tsetse from buildings compare with those from traps designed to catch host-seeking [5] or resting [6] tsetse?
Present work addressed such questions by studying the catches of G. m. morsitans and G. pallidipes in houses and at other baits in Zimbabwe. To focus on the attractiveness of the houses themselves, none of the houses studied was occupied by humans.
All studies were performed at Rekomitjie Research Station, in the Zambezi Valley of Zimbabwe. The station and its seasonal meteorology are described by [4].
The procedures for sampling tsetse followed long-standing protocols practiced at Rekomitjie. All persons used as catchers or baits in the experiments were permanent pensionable employees of the Division of Tsetse Control, Government of Zimbabwe and given regular updates on the purpose and results of the studies. Before recruitment, the Division explains the nature of the work, the risks associated with tsetse, other disease vectors and wild animals, and warns of the social hardships attending life on a remote field station. Recruits sign a document indicating their informed consent to perform the work required. This document is held by the Division. All experiments were given ethical approval by the Division's Review Committee for Rekomitjie.
All houses (Fig. 1) were 20–30 years old and were situated near the centre of the 30 ha clearing of the station, that contained short grass and only a few trees and bushes. Semi-evergreen and deciduous woodland occurred outside the clearing. Each of the houses was unoccupied during the studies, having been vacant for at least a year previously. The walls of the houses were 25 cm thick, made of cement blocks with air cavities, and painted inside and out with white PVA. The roofs were of gabled thatch (House 1) or consisted of corrugated and gently sloping sheets of asbestos (House 2) or galvanized iron, henceforth called tin (House 3) – the latter two “houses” were in fact unused kitchens about 3 m from large thatched houses, but they simulated the types of small building commonly used for field accommodation in central and southern Africa. For some studies the corrugated sheets were covered externally with a 15 cm layer of compressed grass to simulate thatching. Doors on all houses were windowless, hinged and wooden, 2 m tall and 0.8 m wide. Windows were of various width, extending between about 1 m to 2 m above floor level, steel framed and clear-glazed, with the exception of the large mosquito-netted windows along the veranda of House 1. About half of the area of each glazed window could be opened. The netted windows were permanently closed.
Four treatments of each type of house were made, involving changes to the windows and doors that opened to the outside: (i) windows and doors shut, (ii) only windows open, (iii) only a door open, and (iv) windows and a door open. Items opened were fully open. Any internal doors and windows were always open. Whereas House 1 had two exterior doors, only the one on the West front was ever opened. At all houses the four treatments on windows and the exterior door were operated for 24 h, starting just after 1700 h, with subsequent inspections of the house at 2 h intervals from 0700 h to 1700 h the next day. For each inspection, three hand-net catchers stopped just outside the door and closed it quickly. They then caught and discarded any flies seen around them; entered the house, re-closed the door and closed any open window rapidly. Thereafter, the men walked slowly through the house for a few minutes, catching and recording any fly that alighted on them. Afterwards, any flies in the house were captured, most being taken at the windows after being disturbed by swishing hand-nets and long sticks to disturb flies on the walls or roof. The whole inspection took about 5 min, after which the men left the house and reset the windows and doors to the treatment conditions of the day. While separate records were kept of flies caught from the house structures and from the men, the numbers from the men were always relatively small. The catches from the men were pooled with those from the house structures when the intention was to assess the overall number of tsetse in the houses.
An Epsilon trap [5], baited with artificial ox odor was employed to give samples of host-seeking tsetse [7]. The odor consisted of 200 mg/h of acetone, 1 mg/h of 4-methyl phenol, 0.5 mg/h of 1-octen-3-ol and 0.1 mg/h of 3-n-propyl phenol [7], dispensed as described by [8]. Three Box refuges [6] provided samples of tsetse seeking a cool dark place to rest during hot weather. The trap and refuges were operated all day at 25–100 m from the houses, in a predominantly cross-wind direction from them, and were sited to maximize catches. This involved putting the trap in a sunny position [9], and placing the refuges next to boles of shady trees [6], although the absence of many such trees from the general surroundings of the refuges would have reduced their performance [6]. Tsetse were removed from the trap cage and the refuges a few minutes before the inspection of the houses. The removal of flies from a refuge involved quickly closing the entrance with netting sheet, and disturbing the flies inside so that they presented themselves to a cage at the end of a conical part of the sheet.
Dry bulb temperatures were measured in a Stevenson screen near the centre of the station. Inside the houses, thermometers were at head height on walls not in direct sunlight. In Box refuges the thermometers were at the back of the insulated drum, i.e., where most tsetse rested.
Female tsetse were dissected to determine their ovarian category, which offers an index of age [10]. Flies that had ovulated at least once, i.e., in ovarian categories ≥1, had their uterus examined and classed as either empty, or containing an egg or a first to third instar larva (L1–L3). Females with no undigested blood, i.e., those roughly equivalent to hunger stage IV for males [11], were distinguished from those with blood, i.e., stages I–III.
With each house the four window/door treatments were allocated in randomized 4-day blocks of consecutive or nearly consecutive days, but the number of flies of each sex and species caught daily in the houses and at some of the other baits were often zero, making it impossible to perform reliable statistical analyses of mean daily catches. To avoid this problem, the analyses were performed on the combined catches of males and females of both species; the catches from the three refuges were pooled, and unless stated otherwise, the catches from the four window/door treatments were also pooled. Chi-squared tests were performed for the homogeneity of the distributions of catches between various categories, with pooling of categories in some cases to ensure expected values ≥5. The term “significant” implies P<0.05.
Catches were made from House 1 for four or five 4-day blocks per calendar month between Aug 2009 and Aug 2010. The total catches (Table 1, House 1) indicated no gross effect of the door plus windows open, as against just the door or windows. Not surprisingly, when the windows and doors were closed, i.e., for the Nil treatment, the catches were reduced greatly, by an average of 88%. Perhaps more surprisingly, the catches with this treatment were not zero. Some of the flies may have entered the house via the gaps of about 10 cm that occurred under the eaves. Others may have followed the observers un-noticed into the house – according with the observation that a relatively high proportion of the catch with the Nil treatment consisted of male G. m. morsitans, the sex and species that predominates grossly in samples from walking men [4]. In the case of the treatments with an open door, some of the flies following the men may have entered the house when the men arrived outside the door, and before the door was closed. Nevertheless, the compositions of catches with all of the house treatments did not show the huge bias normally expected in catches from men [4], suggesting that the men caused no more than a few flies to enter. Hence, an intriguing point emerged: the house itself seemed attractive in its own right.
The elements in the attractiveness of the house are suggested by considering the percent of G. pallidipes in catches from the various baits. The proportion in the trap was very high, at 91%, and significantly different (P<0.001) from the 36% evident at the refuges. With the house treatments the percents were intermediate, at 61–81% (average 76%). This suggested the hypothesis, henceforth termed the “mixed sample” hypothesis, that the catches from House 1 consisted of two segments, one comparable to refuge catches and the other comparable to trap catches. The implication is that House 1 functioned as both a trap and a refuge, attracting some flies that were host-seeking and others looking for shelter. It seemed that House 1 did indeed offer a good refuge since in the middle of the day, when screen temperatures were greatest, the temperatures in the house were about two degrees lower than screen temperatures – much like the Box refuges but in sharp contrast to the asbestos-roofed House 2 and particularly the tin-roofed House 3 (Fig. 2).
In some of the months in which catches were made from House 1, simultaneous catches were also made from the other houses. In the first experiment (Table 2, Expt 1) the mean catches from the small houses as a percent of those from the large thatched house were only 17% for the small asbestos-roofed House 2 and even lower at 13% for the small tin-roofed House 3, i.e., the hotter the house (Fig. 2) the lower the catches. Moreover, the hotter the house, the lower the proportion of G. pallidipes in the total catches – the percents being 87%, 36% and 28% for Houses 1, 2 and 3, respectively. These proportions were significantly heterogeneous (P<0.001). In the next experiment the asbestos or tin roofs of the small houses were covered in grass, so that the temperatures in them became cooler and more like those of the thatched House 1, with temperatures at 1100 h–1700 h in Houses 2 and 3 being less than screen temperatures by an average of 1.4°C (95% CL 1.3–1.6) and 0.9°C (0.8–1.1), respectively. The mean catches at these houses then increased slightly to 14–36% of the House 1 catch. However, House 3 still gave the fewest tsetse, and the proportion of G. pallidipes in catches from Houses 1 and 2 was still lower than in House 1 (P<0.001) (Table 2, Expt 2).
Having failed, above, to demonstrate any gross effect of temperature and roof type on the magnitude and composition of catches from houses, it was suspected that the distinctive samples from the different houses were associated with window type. In Houses 2 and 3 the windows consisted only of glass, whereas in House 1 much of the “window” space was netting, i.e., on the veranda, so encouraging ventilation. Hence, the following study of window type was made.
On some days in Aug–Sep 2010 the windows of Houses 1 and 2 were closed, so that exit via them was completely barred by glass. On other days the opening parts of the windows were fully open, but covered in netting, so that tsetse could not enter or leave via the windows. The doors were open for both of the window treatments, and the roofs were covered with grass. Catches were compared with simultaneous catches from House 1 with the door open and windows closed, i.e., the way the small houses were operated.
The total catches (Table 3) showed that even with the all-glass windows, i.e., the type of treatment used in previous months, the numbers of tsetse caught from the small houses relative to House 1, and the proportions of G. pallidipes in catches from the small houses, were now increased substantially. This was associated with the onset of the hot-dry season, so perhaps the rising temperatures outside the house caused G. pallidipes to disregard those features of the small houses that previously reduced the availability to such houses. In any event, the main point of the experiment, i.e., the investigation of any effect of window type on the formerly very low proportions of G. pallidipes from small houses, was somewhat undermined. Nevertheless, the results did show that, during the hotter weather at least, there was no gross effect of window type in the small houses, and that the total catches from the small houses were still less than from the large, and still contained relatively low proportions of G. pallidipes. The heterogeneity in the proportions of G. pallidipes in samples remained significant (P<0.001).
In House 1 the total catches from the men consisted of 23 males and 16 female G. m. morsitans, and two males and one female G. pallidipes. In the smaller Houses 2 and 3 the figures were 24, 3, 2 and 0, respectively. The percents of male G. m. morsitans in the samples was therefore 53% with House 1 and 83% with the other houses, and the difference was significant (P<0.05).
The monthly catches at the trap and refuges (Fig. 3, A and B) followed the patterns typically observed at Rekomitjie, with the refuge catches being by far the greatest in the hot-dry season of Sep–Nov and smallest in the cool-dry season of mid-year, and with the trap catches being more evenly distributed [6]. The pattern with the house catches was intermediate, giving support to the mixed sample hypothesis, above.
The general patterns of the availability to traps and refuges was as usually found at Rekomitjie [6]. Thus, with both species of tsetse the catches from the traps (Fig. 4, A) were greatest in the morning and late afternoon, but there were seasonal distinctions. The mid-day trough in trap catches was most pronounced in the hottest months of Sep–Nov (Fig. 4, A1) and least marked in the coolest months of May–Aug (Fig. 4, A3). Moreover, while the morning peak of trap catches was greater than the afternoon peak in Sep–Nov, the afternoon peak became more pronounced as the weather cooled. The refuge catches (Fig. 4, B) were concentrated in the middle of the day and early afternoon, and so differed markedly from trap catches. Again there were seasonal variations in that during Sep–Nov (Fig. 4, B1) the refuge catches started to rise earlier than in the cooler conditions of Dec–Aug (Fig. 4, B2 and B3), presumably because during the hotter months the need to avoid high temperatures occurred sooner in the day.
Catches from House 1 (Fig. 4, C) differed from trap catches (Fig. 4, A) in being large in the morning and/or the afternoon, i.e., somewhat like trap catches. However, the house catches differed from trap catches in showing no trough in the late morning and early afternoon. In general, the diurnal pattern at the house seemed to be a hybrid of the patterns at the trap and refuge, as expected on the mixed sample hypothesis. Nevertheless, there was a slight departure from expectation in that the catches from the house were not as great as predicted at the 0700 h inspection, when the presence of many flies in traps should have been associated with many flies being caught at the house. This could be due to the fact that the trap was baited with odor, whereas the house was not. However, the more likely explanation is associated with the observation that some of the tsetse in the house were attacked by ants, as evidenced by the presence of half-eaten carcasses or wings, found mainly after the long overnight delay between the 1700 h inspection of one day and the 0700 h inspection on the next. The was little or no evidence of trap catches being attacked overnight.
Few flies were caught at certain times of day with all baits, making it impossible to identify any clear diurnal variations in the reproductive condition of samples, so the data for all times of day were pooled. Such pooling led to no evidence of a seasonal change in the distributions of uterine contents of females of either species. However, the proportion of old flies was relatively low in the latter half of the dry season. For example, in Aug–Nov the percent of G. m. morsitans in ovarian categories ≥4 was 20% (N = 56) in the houses and 24% (25) the traps, as against figures of 48% (42) and 30% (61), respectively, in other months. For G. pallidipes the figures were 49% (166) and 47% (128), respectively, in Aug–Nov and 65% (141) and 58% (499), respectively, in other months. The seasonal heterogeneity in the proportion of old flies was significant (P<0.01 to <0.05) in all cases except for G. m. morsitans from the trap. With the latter bait some of the catches of G. m. morsitans were small, making it difficult to find a significant difference
Despite the seasonality in some aspects of the results, the pooled data for ovarian categories (Fig. 5) and uterine contents (Fig. 6) in the whole study period illustrate two matters that applied at all seasons. First, with each bait the samples of G. pallidipes were older than for G. m. morsitans and contained a lower proportion of flies with larvae as against eggs. Second, the samples of G. m. morsitans from all baits were older, and with higher proportions of larvae, than the samples taken from men during other work performed at Rekomitjie in parallel with the present investigations [4]. In that other work the catches of G. m. morsitans from the men in various situations inside and outside houses throughout the year showed only 18% (N = 257) in ovarian categories 4–7, and only 23% (189) of the flies in categories ≥1 carried larvae. These compositions are significantly different from the figures of 32% (N = 98, P<0.01) and 51% (N = 85, P<0.001), respectively, for the present catches of G. m. morsitans from houses over the year (Figs. 5 and 6).
Of the 19 female G. m. morsitans caught from men in houses in the present work, only five were dissected. Two were in category, 0, one was in category 1 and two were in category 2. All three flies in categories 1 and 2 had an egg in the uterus.
Since the age structure and uterine contents of samples from the trap and refuges where closely similar (Fig. 5, A and B; Fig. 6, A and B), the mixed sample hypothesis required, as observed, that the age structure and uterine contents of the house catches (Fig. 5, C; Fig. 6, C) were much the same as for the trap and refuge catches.
For the refuge catches, most of which were in Sep–Nov, the percent of females with undigested blood was fairly high with each species, averaging 34.9% (N = 109) for both species combined. For traps at all times of year, and for the houses in months other than Sep–Nov, the percent of the catches with blood was very low, averaging 2.7% (N = 713) for the traps and 4.4% (252) for the houses. However, the percent in the catches from houses increased significantly (P<0.05) to 10.5% (N = 153) in Sep–Nov, consistent with the evidence (Fig. 3, C) that many refuge-seeking flies entered the houses in these months.
We recorded the sex and species composition, age structure, pregnancy condition and hunger stage of samples of tsetse caught in various types of unoccupied houses at different times of day throughout the year, and compared these data with those of catches from artificial refuges and host-like traps nearby. In general, the character of catches from the houses was intermediate between those from the refuges and traps. Our results suggest that the structure of a house is itself attractive to tsetse, so that the flies enter even when no humans are inside, but that if humans then enter the house some of the tsetse already in it can go to the people.
The methods of the present work offer valid indications for the numbers of tsetse that entered houses and then remained inside for up to two hours during the day, even if the numbers staying inside overnight and found at 0700 h may have been reduced by ant predation. However, the methods provide only crude measures of the numbers entering since, strictly speaking, the work showed only the numbers found in the houses at each inspection, rather than addressing the entry responses themselves. Thus, many flies may have entered the houses and left before the inspections were made. In particular, when many openings allowed tsetse to enter the house they might also have facilitated a quick exit, so it is hardly surprising that having both the door and windows open had no great effect on house catches. Moreover, the reason for the seasonally low proportions of G. pallidipes in catches from small houses might have been that many G. pallidipes entered the small houses at all seasons, but sometimes they left them rapidly, perhaps because the houses were insufficiently large and lofty to offer the right microclimates. These matters could have been investigated more critically by placing electrocuting grids [12] over the openings of the doors and windows, to catch flies at the instant of entry, but this would have precluded an important aspect of the present work, i.e., assessment of the number of tsetse that remained in the houses for some while, so that they would have had a good opportunity to contact any humans that entered.
Despite the above problems, the results do suggest that there were two main reasons why tsetse entered houses. First, in all months a house acts like a trap that attracts tsetse in the host-seeking phase of behavior that they exhibit in the early morning and/or late afternoon. Second, in hot weather other tsetse enter houses to find a cool shady refuge during the late morning and early afternoon. The indication that the flies identify the doors and windows as entrances to refuges fits with the fact that some natural refuges consist of openings into very large objects – for example, rot-holes in baobab trees and hollows in tall river-banks [6]. Why tsetse appear to mistake a large white house for a host is less clear, but then it is hardly clear why tsetse seem to regard a bright blue trap as a host. In any event, the fact that traps [13] and artificial refuges [6] of various color differ greatly in their efficacy suggest that house color could also be important. In particular, by analogy with various types of artificial refuges operated at different seasons and situations [6], one could expect that a large dark-colored house in shady riparian woodland would attract many refuge-seeking tsetse in the hot season – far more than found in present houses.
There is no direct evidence in present work to indicate what proportion of the overall catch at the houses was represented by refuge-seeking flies, especially given the caveat that the catches at the three Box refuges might have underestimated substantially the numbers of tsetse seeking refuge in the very much larger houses. However, taking together the data for catch compositions (Table 1), and for diurnal and seasonal patterns of catches (Figs. 3 and 4) and hunger stage, the proportion of refuge-seeking flies seems to have been substantial – about a quarter to three-quarters on hot days. Despite the apparent importance of houses as refuges, many of the flies in houses at all times of the year appeared to have entered in direct search of food, and the blood reserve of many of these flies and some of the refuge seekers seemed so low that, had they been left in the houses, they might have sought food from resident humans there once the temperatures declined sufficiently in the evening to obviate any need for refuge. Thus, present results accord with the indication [4] that houses can be at least as important as other venues for contact between humans and hungry tsetse.
In respect of the risk of being bitten by tsetse, it might seem fortunate that the samples of flies in houses contained relatively high proportions of old tsetse, and high percents of females and of G. pallidipes, and so were very different from those normally associated with probing on humans [4]. In particular, the fact that old flies usually avoid human hosts is important because only such flies can be effective vectors of sleeping sickness [14]. However, it is worrying that humans in houses can be in very close proximity to tsetse old enough to be potential vectors. The big question, therefore, is to what extent different types of building, and different patterns of human occupation, might induce a broader spectrum of the flies in the buildings to attack humans there. The fact that conditions inside buildings can change the normal behavior of tsetse is indicated by previous work [4] which found that female G. m. morsitans formed a relatively high proportion of the flies probing men in the mainly large buildings at Rekomitjie. Current data for the numbers of G. m. morsitans taken from men in the large House 1 accord with that result, although in the smaller houses the proportion of females in catches from men was low.
Present work considered only three types of house, each of which was unoccupied by people except for the presence of the observers at the brief inspections during the day. The number of flies in a house, and their propensity to attack humans, might be expected to change if, for example, the humans remained in or near the house for many hours, if the flies in the house were not removed frequently during the day, if other animals were kept in or near the house to attract or distract flies [4], and if domestic cooking generated wood smoke that can be repellent [15]. These matters are currently under investigation at Rekomitjie.
|
10.1371/journal.pntd.0006475 | The protein family TcTASV-C is a novel Trypanosoma cruzi virulence factor secreted in extracellular vesicles by trypomastigotes and highly expressed in bloodstream forms | TcTASV-C is a protein family of about 15 members that is expressed only in the trypomastigote stage of Trypanosoma cruzi. We have previously shown that TcTASV-C is located at the parasite surface and secreted to the medium. Here we report that the expression of different TcTASV-C genes occurs simultaneously at the trypomastigote stage and while some secreted and parasite-associated products are found in both fractions, others are different. Secreted TcTASV-C are mainly shedded through trypomastigote extracellular vesicles, of which they are an abundant constituent, despite its scarce expression on culture-derived trypomastigotes. In contrast, TcTASV-C is highly expressed in bloodstream trypomastigotes; its upregulation in bloodstream parasites was observed in different T. cruzi strains and was specific for TcTASV-C, suggesting that some host-molecules trigger TcTASV-C expression. TcTASV-C is also strongly secreted by bloodstream parasites. A DNA prime—protein boost immunization scheme with TcTASV-C was only partially effective to control the infection in mice challenged with a highly virulent T. cruzi strain. Vaccination triggered a strong humoral response that delayed the appearance of bloodstream trypomastigotes at the early phase of the infection. Linear epitopes recognized by vaccinated mice were mapped within the TcTASV-C family motif, suggesting that blockade of secreted TcTASV-C impacts on the settlement of infection. Furthermore, although experimental and naturally T. cruzi-infected hosts did not react with antigens from extracellular vesicles, vaccinated and challenged mice recognized not only TcTASV-C but also other vesicle-antigens. We hypothesize that TcTASV-C is involved in the establishment of the initial T. cruzi infection in the mammalian host. Altogether, these results point towards TcTASV-C as a novel secreted virulence factor of T. cruzi trypomastigotes.
| Trypanosoma cruzi is the kinetoplastid parasite that causes Chagas’ disease, a neglected infection endemic in Latin America and emerging worldwide. Being vaccines currently unavailable and treatments not completely effective, identification and characterization of parasite molecules that can be target for these interventions are urgently needed. Of particular interest are surface anchored and secreted proteins involved in parasite—host interplay. Recently, extracellular vesicles released from protozoan pathogens have been shown to alter host cell function favoring the establishment of infection. Trypomastigotes are the disseminating stage of T. cruzi, being their presence in peripheral blood a hallmark of early acute infection in mammals. While the most abundant proteins of the trypomastigote surface are fairly well characterized, little is known about other, less abundant and more recently discovered multigenic families, which could have critical functions in the parasite—host interaction. The T. cruzi Trypomastigote Alanine, Valine and Serine rich proteins (TcTASV) belong to a medium-size multigene family of ~40 members that remained unobserved until a few years ago when it was identified through a trypomastigote-enriched cDNA library. Almost simultaneously, an expression library immunization approach designed to discover novel vaccine antigens in T. cruzi, spotlighted the TcTASV-C subfamily, as a fragment of a TcTASV-C gene was identified in a pool of protective clones. A distinctive feature that characterizes TcTASV proteins–and particularly the TcTASV-C subfamily- is their predominant expression in trypomastigotes. Recent transcriptomic and proteomic studies uphold our previous observations that the TcTASV family is over-represented in the trypomastigote stage, and therefore could represent an interesting target for rational intervention against T. cruzi infection. Here show that TcTASV-C is mainly secreted through extracellular vesicles (EVs) of trypomastigotes, and is a major cargo of its content. We have also shown that TcTASV-C is much more expressed in trypomastigotes purified from blood from infected mice than in trypomastigotes harvested from in vitro cultures, suggesting that host molecules should trigger TcTASV-C expression in vivo during the infection. The immunization of mice with TcTASV-C interfered with the early acute phase of T. cruzi infection through a strong humoral immune response. TcTASV-C should be considered as a novel secreted virulence factor of T. cruzi trypomastigotes and -although its biological function is still unknown- we hypothesize its participation in the early steps of T cruzi infection in the mammalian host.
| Trypanosoma cruzi, is the kinetoplastid pathogen that causes Chagas’ disease. There are about 10 million people currently infected and more than 50–60 million people living in endemic areas, at risk of infection. Chagas’ disease is a chronic debilitating illness with symptoms generally appearing 10 or more years after the initial infection [1–3]. At this stage, anti-parasitic drugs are poorly effective and patients are treated according to their cardiac, digestive or neurological compromise [4–5]. Acute infection is usually undetected because of its mild and unspecific symptoms and, therefore, the patients are not diagnosed. The acute phase of the infection is characterized by the presence of high levels of trypomastigotes in blood. These nonreplicative trypomastigotes invade nucleated cells, where they differentiate to the amastigote stage that replicates in the cytoplasm and differentiates again to trypomastigotes. Then, the infected cell bursts and trypomastigotes are released once again to circulation. During the acute phase this cycle repeats itself actively and the trypomastigote disseminates the infection to several organs and tissues [6]. Considering the absence of preventive or chemoprophylactic vaccines as well as the life cycle of the parasite, uncharacterized molecules differentially expressed in the infective trypomastigote stage can be interesting novel targets for rational intervention against Chagas’ disease [7,8].
The T. cruzi Trypomastigote Alanine, Valine and Serine (TcTASV) rich proteins belong to a medium-size multigene family of ~40 members that was identified from a library of trypomastigote-enriched mRNAs [9]. The TcTASV family is conserved among all the T. cruzi lineages analyzed so far and has no orthologues in other species, including the closely-related trypanosomatids T. brucei, T. rangeli and Leishmania sp. [9]. TcTASV proteins, whose function is still unknown, are expressed mainly in the trypomastigote stage. The N- and C-terminal regions of the TcTASV proteins possess a signal peptide and a consensus for a GPI anchor addition, respectively, and display the highest level of conservation, while the central region presents more variability [9]. TcTASV family can be distinguished by the common amino acid motif tasv_all that starts approximately at amino acid 42 (Vx1x2x3[CES]x4x5TDGx6Lx7Wx8x9x10x11Ex12x13Wx14x15Cx16x17x18P). The TcTASV family is comprised of 4 subfamilies -TcTASV-A, B, C and W- defined by the primary amino acid sequence and length of polypeptides. Further, each subfamily presents certain amino acids at the indeterminate positions (x1, x2, etc) of the tasv_all motif. For example, subfamilies TcTASV-C and TcTASV-A both have proline and glycine at positions X4 and X5, while TcTASV-B contains serine and arginine, and TcTASV-W has alanine at X4 and glutamic acid at X5.
The TcTASV-C subfamily includes approximately 15 genes (small variations are found in different strains) with protein products of 330–360 amino acids [9, 10]. A few years ago, in the search for novel vaccine candidates by a genetic immunization approach, a fragment of a TcTASV-C gene (TcCLB.511675.3; ID TritrypDB, [11]) was identified among a pool of antigens that protected mice from a parasite challenge with a highly virulent T. cruzi strain [12]. In a first characterization of the TcTASV-C subfamily we found that TcTASV-C is a thickly glycosylated ~60 kDa polypeptide, expressed in trypomastigotes and absent in all other parasite stages [10]. TcTASV-C presents a characteristic distribution pattern of scattered dots along the parasite surface and flagellum, and is spontaneously secreted to the medium. While anti-TcTASV-C antibodies are detected in about 30% of chronically-infected patients, the seroprevalence in reservoir dogs with active infection rises to 75% [10,13]. In the experimental murine T. cruzi model, TcTASV-C specific antibodies can be detected early from the beginning of the infection [10]. Although TcTASV-C proteins are not major components of the parasite in trypomastigotes derived from in vitro cultured cells [10], several TcTASV-C peptides have been recently identified in secretomes of T. cruzi trypomastigotes [14–16]. Interestingly, TcTASV peptides were found in the bloodstream trypomastigote proteome, but not in proteomes from in vitro cultured cell derived trypomastigotes [17]. Also, a recent analysis of an overall transcriptome of the host cell and T. cruzi during the course of infection confirmed that the TcTASV family is extensively over represented in trypomastigotes, relative to all the other stages of the parasite, and several TcTASVs mRNAs are among the most abundant in the trypomastigote stage [18]. After being unnoticed for several years, and in agreement with our previous reports, these findings also outpoint towards TcTASVs as potential virulence factors and as interesting targets for study and rational intervention.
Here we present results leading to a deeper understanding of the TcTASV-C subfamily and its performance as a vaccine antigen.
First, a bioinformatic analysis was carried out to determine whether there was a common pattern among all TcTASV-C members. A distinctive and conserved tasv_c motif of 50 amino acids was identified in all TcTASV-C proteins (Fig 1). In most proteins, the tasv_c motif starts at amino acid 42–43, including and expanding the previously reported tasv_all motif, common for all TcTASV proteins irrespectively of their subfamily (asterisks in Fig 1). Only TcTASV-C proteins are retrieved by searching any database with the tasv_c motif.
We have already described that TcTASV-C is expressed both at the parasite surface and secreted to the medium [10]. To investigate whether–among the 15 TcTASV-C genes of CL Brener strain- several genes (or only one) are simultaneously expressed, and to clarify if surface-located and secreted TcTASV-C proteins are expressed from the same genes, we undertook a 2D gel based approach (Fig 2). More than one TcTASV-C product was observed both in the parasite-associated and in the secreted fractions (Fig 2), suggesting that more than one TcTASV-C gene are simultaneously expressed. On the other hand, there were common and differential TcTASV-C spots detected in both fractions. There was a clear band of higher molecular weight only present in the parasite fraction (arrow in Fig 2, upper panel) and two bands of more acidic isoelectric point (pI) in the secreted one (arrowheads in Fig 2, lower panel). Two other bands seemed to be shared by both samples.
These results show that the expression of different TcTASV-C genes occurs simultaneously at the trypomastigote stage and suggest that while some secreted and parasite-associated products are found in both fractions, others are different.
TcTASV-C is secreted and also detected–by immunofluorescence microscopy- at trypomastigote surface in spots that are compatible with detergent resistant domains [10]. These domains are often associated with secretion of molecules through extracellular vesicles (EVs) [20,21]. In this context, we investigated whether TcTASV-C proteins were released associated with EVs or as soluble factors (VF: vesicle-free fraction). In a first set of experiments, trypomastigote- derived conditioned media was resolved by ultracentrifugation in density gradients; TcTASV-C was detected in fractions corresponding to extracellular vesicles (S1 Fig).
We next investigated whether TcTASV-C proteins were released associated with large (V2) or small (V16) EVs, whose presence and purity was confirmed by transmission electron microscopy (TEM; Fig 3A). All samples showed vesicles of 30–130 nm in size after 2 h of ultracentrifugation, while vesicles obtained after 16 h were smaller (~50 nm; Fig 3A). TcTASV-C was secreted in both EVs populations in the CL Brener strain (Fig 3B); in the small EV fraction (V16, Fig 3B, upper panel) TcTASV-C appeared as a highly abundant component, while it was almost undetectable on parasite pellets at these conditions. This agrees with the already reported low level of expression of TcTASV-C on parasite body in cell derived trypomastigotes [10]. Longer exposure times were needed to evidence the TcTASV-C expression on trypomastigotes but render overexposed and unclear images for V2 and V16 fractions. As expected, the heat shock protein 70 (HSP70), a secretome marker, was detected in all fractions and, TcSR62, a nucleo-cytoplasmic and non-secreted RNA binding protein, only in the parasite pellet (Fig 3B) [10,22,23].
A proteomic analysis of V2 and V16 secreted extracellular vesicles of CL Brener strain was also carried out, to confirm our western blots results and, besides, because there are no proteomic analysis of V2 and V16 cargo proteins of trypomastigotes. Indeed, the data currently available from exoproteomes of trypomastigotes were derived from total secreted material or from a mixture of vesicles from parasites and host cells purified together [14–16]. High confidence peptides of four TcTASV-C genes were found both in V2 and V16 samples (TcCLB.508741.440, TcCLB.509123.10, TcCLB.509147.40, TcCLB.508737.10), which is in line with the 2D western blot results. Although TcTASV-C peptides were detected in both EV fractions, is noteworthy that each EV population presented a differential set of major proteins (V2: n = 271; V16: n = 189), and only a minor core of 142 common proteins (among which are the TcTASV-C peptides) (S2 Fig). This suggests that both fractions of vesicles correspond to different populations. Surface, intracellular as well as a considerable percentage of hypothetical proteins were identified by proteomics in trypomastigote EVs (S2 Fig). In context, the picture obtained could indicate that–at least at the assayed conditions- the paucity of TcTASV-C in the parasite’s body probably reflects that most of TcTASV-C produced is delivered to the secretory pathway.
We then analyzed the secretion profile of TcTASV-C in T. cruzi strains that encompass a wide spectrum of virulence and also represented the major T. cruzi lineages [24]. Overall, TcTASV-C was secreted in EVs in all the strains analyzed (Fig 4). Only mild differences in TcTASV-C secretion profile were detected among strains. In the low-virulent SylvioX10 strain, TcTASV-C seemed to be poorly represented in small (V16) EVs, while in 173 strain (DTU TcI)–middle-virulence- the secretion profile of TcTASV-C was quite similar to that found in CL Brener strain (Fig 3), which is also of intermediate virulence in the murine model. In culture-derived trypomastigotes from highly virulent strains (i.e. Y, TcII and RA, TcVI) a more dynamic pattern of secretion was observed, and TcTASV-C was detected alternatively in different fractions (Fig 4 and S3 Fig). Particularly for the Y strain, trypomastigotes released on the 1st day after cells began to lyse usually showed the profile depicted on Fig 4, while TcTASV-C expression shifted to V16 and VF fractions, on EVs secreted from parasites harvested the 2nd and 3rd day, showing a dynamic expression pattern (S3 Fig). As a whole, in all the analyzed strains, TcTASV-C was significantly more represented in the secreted than in the parasite associated fraction. Of note, the total protein content secreted in EVs was similar for all strains thus allowing to discard that changes in TcTASV-C expression by each parasite strain had a correlation with the amount of total EVs secretion.
Strikingly, the first proteomic evidences of TcTASV-C were registered few years ago with the publication of the proteome of bloodstream trypomastigotes [17]. In contrast, no evidence of expression of TcTASV-C was noticied when proteomics of culture-derived trypomastigotes were analized. Therefore, we decided to investigate the TcTASV-C expression profile in bloodstream trypomastigotes from different T. cruzi strains, in connection with its expression on culture-derived trypomastigotes.
TcTASV-C was detected in 1x10^6 (or even less) bloodstream parasites (Fig 5A, upper panel, BT lanes) while it was necessary to load more than tenfold of culture-derived trypomastigotes to be weakly detected (Fig 5A, Cult lanes, upper panel). This finding was observed in all T. cruzi strains assayed, belonging to different lineages, and proved that TcTASV-C is upregulated in bloodstream forms. Importantly, this differential expression pattern between both types of trypomastigotes was not observed with other proteins of T. cruzi, which were detected accordingly to the parasite amount loaded on the gel (Fig 5A, middle and lower panels). As well as detected for culture-derived trypomastigotes, bloodstream trypomastigotes were also able to strongly secrete TcTASV-C. In the highly virulent RA strain (TcVI) TcTASV-C was essentially identified in all secreted fractions, even from 1-2x10^6 trypomastigotes (Fig 5B and S5 Fig).
As molecules released by the parasite could potentially interact with host cells, we evaluated this possibility for TcTASV-C on Vero (Fig 6A) and J774 (Fig 6B) cells. TcTASV-C (but not the control protein GST) exhibited a dose-dependent adhesive capacity, suggesting a ligand-receptor interaction. Similarly, EVs derived from trypomastigotes also interacted with mammalian cells (S4 Fig). This interaction was only observed with freshly isolated EVs, but not with EVs that had been previously purified and stored at -80°C. We also evaluated a potential role of TcTASV-C on T. cruzi cell infection, employing as model two T. cruzi strains from different lineages and obtained from in vitro cultures (CL Brener, Fig 6C) or purified from blood of infected mice (Tulahuen strain, expressing β-galactosidase, Fig 6D and 6E) [25,26]. Neither pre-incubation of recombinant TcTASV-C with mammalian cells before infection (Fig 6C and 6D) nor pre-incubation of trypomastigotes with anti-TcTASV-C sera (Fig 6E) interfered with parasite internalization or cellular infection.
The delivery of molecules in vesicles is a well known immune evasion mechanism exploited by parasites [27]. We therefore investigated whether infected hosts could recognize the protein content of secreted vesicles (Fig 7). Pooled sera from humans, mice or rabbits chronically infected with T. cruzi failed to efficiently detect EVs antigens, although they strongly reacted with trypomastigote antigens and with “naked” secreted proteins, both from RA and CL Brener strains (Fig 7 and S6 Fig). These findings led us to hypothesize that immunization with TcTASV-C–which is highly expressed in bloodstream trypomastigotes and also secreted- would be a good target for immunotherapy control. This hypothesis was encouraged by our previous finding of a TcTASV-C gene fragment among a pool of protective antigens [12]. Besides, being TcTASV-C expressed at early stages of the infection [10], we hypothesize that humoral immune response could be mediating TcTASV-C neutralization.
To evaluate the performance of TcTASV-C as vaccine antigen, we designed a DNA-prime protein-boost schedule of immunization. The first 2 doses consisted of plasmid DNA of an eukaryotic expression vector carrying a fragment of TcTASV-C [12] adjuvanted with a plasmid coding for GM-CSF [28]. In the 3rd and 4th doses, mice were boosted with TcTASV-C recombinant proteins (two different genes fused to GST or histidine tags, rTcTASV-CGST and rTcTASV-CHIS) adjuvanted with aluminium salts.
As expected, immunization was effective to induce high levels of total anti-TcTASV-C IgGs (Fig 8). Most animals presented a mixed Th1/Th2 response, with strong IgG2a and IgG1 responses (Fig 8B and 8C). However, the cellular and cytokine response in splenocytes obtained from mice 15 days after immunization, showed a low proliferative response and negligible levels of IFN-ɣ and IL-10 after rTcTASV-C restimulation in culture.
Two weeks after the last dose, animals were challenged with parasites of the highly virulent RA strain (DTU TcVI). TcTASV-C vaccinated mice exhibited a delayed appearance of circulating trypomastigotes and lower parasitemia peaks (Fig 9A–9C). Bloodstream trypomastigotes were detected from the day 9th on, in all controls while were nearly unnoticeable until day 12th in all TcTASV-C vaccinated (Fig 9A–9C). Besides, TcTASV-C vaccinated mice presented reduced trypomastigote numbers at the peak of parasitemia (Fig 9C) and–overall- lower bloodstream parasite levels than the control group (Fig 9C; p<0.05, at 9 and 12 pdi, Mann-Whitney test). Immunized mice also showed higher survival rates than controls (Fig 9D; TcTASV-C, ~30%; control group, 0%; p<0.05 Log-Rank test). All animals that survived at the end of experiments belonged to TcTASV-C vaccinated groups.
After T. cruzi challenge, vaccinated mice showed a strong response against parasite antigens, with an IgG2a-biased isotype, similar to that found in infected animals (Fig 10A, 10B and 10C). Besides, sera from vaccinated mice were able to lyse trypomastigotes in the presence of an external complement source (Fig 10D); the lytic response was increased in mice after infection. Focusing on the anti-EV response after T. cruzi challenge, vaccinated animals reacted not only with TcTASV-C but also with other EV antigens, that were unseen by infected but non-vaccinated animals (Fig 10E and 10F). Therefore, the vaccination was able to trigger an EV-focused immune response after challenge that can’t be mimicked by unvaccinated infected animals.
As the main immune response detected in immunized animals was humoral, we mapped the TcTASV-C epitopes detected. Peptides covering putative TcTASV-C epitopes were designed by weighing linear B-cell epitope predictions (bioinformatic approach) and those epitopes previously discovered in a high-density peptide microarray screened with human sera [29]. We selected 5 peptides of 15–20 amino acids to evaluate the reactivity of sera from TcTASV-C vaccinated, control and infected unvaccinated mice (Fig 11). Eighty-six percent (86%; 19/22) of the sera from vaccinated mice reacted with at least one peptide, and 45% reacted with 2 or more peptides (S1 Table). In contrast, only 30% (4/13) of sera from unvaccinated infected mice (with previously reported reactivity against TcTASV-C) recognized any of these peptides (S1 Table).
P46-62 and P172-189 were the peptides most detected by the sera from vaccinated group, with 84% (16/19) of sera reacting with one or both of them (Fig 11 and S1 Table). P46-62 was the peptide most detected by sera from vaccinated mice while it was not detected by any of the sera from the infected group. Interestingly, P46-62 is part of the tasv_all motif, but is only partially present in the rTcTASV-Cs employed in the vaccination schedule (TcTASV-CHIS: KLSWRLRGEEEW; TcTASV-CGST: SWRLQGEEEW). Even more striking, the reactivity to peptide P46-62 seems to be driven by the RLR triplet or the second arginine, since an identical peptide with an unique substitution that changes the RLR motif to the RLQ, turned it into an unrecognized peptide (P47-63; S1 Table). Both RLR and RLQ sequences are present in TcTASV-C genes (see Fig 1), and represented by the rTcTASV-Cs employed in the vaccination scheme (RLR in TcTASV-CHIS, RLQ in TcTASV-CGST).
Altogether these results support the idea that the broad anti-peptide reactivity of immunized mice is probably mediating the partial resistance and/or the delay in the appearance of circulating trypomastigotes in challenged mice.
The T. cruzi TcTASV gene family remained unobserved until a few years ago when it was identified by our group through a trypomastigote-enriched cDNA library [9]. Almost simultaneously, an expression library immunization approach designed to discover novel vaccine antigens in T. cruzi, spotlighted the TcTASV-C subfamily, because a fragment of a TcTASV-C gene was identified in a pool of protective clones [12]. A distinctive feature that characterizes TcTASV proteins–and particularly the TcTASV-C subfamily- is their predominant expression in bloodstream trypomastigotes. Recent transcriptomic and proteomic studies uphold our previous observations that the TcTASV family is over-represented in the trypomastigote stage [17,18], and therefore could represent an interesting target for rational intervention in T. cruzi infection.
Here the TcTASV-C expression and secretion dynamics and its performance as an individual vaccine candidate were analyzed. We demonstrate that, despite its scarce expression on culture-derived trypomastigotes, TcTASV-C is strongly secreted, and is a major component of trypomastigote’s EVs, at least in the T. cruzi reference strain CL Brener. This was observed both by western blot and proteomics on large (V2) and small (V16) EVs. It is a novelty, although not unexpectedly, that a parasite-associated and low-expressed protein (or protein family) is actually a highly abundant component of the trypomastigote secretome. The secretion of EVs by parasites has been proposed as a pathogen-driven mechanism aimed to generate -in the host- an environment that favours the initial infection [30–34]. Indeed, in most of the tested T. cruzi strains, TcTASV-C was mainly secreted contained into EVs. Of note, the more virulent strains (i.e. RA and Y) presented also a more dynamic secretion pattern (Fig 4, Fig 5, S3 Fig and S4 Fig). On the other hand, we have shown that TcTASV-C expression is upregulated in bloodstream parasites, suggesting that some molecules present in the host trigger TcTASV-C expression.
The potential of TcTASV-C as an individual vaccine candidate, however, was somehow limited to the acute phase. In our model, TcTASV-C immunized mice achieved an enhanced control of parasitemia at the beginning of the infection. The delayed appearance of bloodstream trypomastigotes was along with the presence of functional antibodies in sera from TcTASV-C vaccinated mice, with ability to lyse trypomastigotes by ADCC. This is also consistent with the detection of TcTASV-C early upon infection and suggests that TcTASV-C could have a role during this phase of infection [10,13] (unpublished results). We hypothesize that the window of time with lower bloodstream parasites, gives a handicap to TcTASV-C primed mice to launch effector mechanisms against the parasite. However, it has to be said, the humoral response induced by vaccination was not strong enough to completely protect and clear parasites from a lethal challenge, and mortality rates were only mildly improved. Likewise, Ramirez et al (2017) [35] have recently reported that EVs derived from the interaction between mammalian cells and trypomastigotes potentiated parasitemia, particularly in the early acute phase (3–6 days) of infection. This effect was stage-specific since it was not observed with EVs derived from the interaction of mammalian cells with metacyclic trypomastigotes or epimastigotes, suggesting that stage-specific EVs components might play a role in survival and dissemination of this parasite stage in the vertebrate host [35].
Secretion of virulence factors contained in extracellular vesicles has also been understood as a parasite strategy to deliver long distance effector molecules that should act in concert [36]. In particular, T. cruzi trypomastigotes release EVs that can interact with the host and modulate immune responses. The first communication in this way was in 2009, when Trocoli-Torrecilhas et al [37] demonstrated that inoculation of mice with naked extracellular vesicles predisposed them to a more virulent infection, along with a strong inflammatory tissue damage and higher parasitic loads in heart. In fact, the effect observed with whole EVs had been observed several years before with an EV cargo molecule, the trans-sialidase (TS). Chuenkova and Pereira (1995) [38] reported that mice sensitized with TS were more susceptible to T. cruzi infection, displaying enhanced parasitemia and mortality. Here, by analyzing the proteome from CL Brener EVs, we found peptides of both TS and TcTASV families, suggesting that both components of trypomastigotes are secreted as part of the same cargo and can act in a concerted fashion. Actually, in retrospective, we found several EV cargo proteins employed as vaccine antigens with promising results [12,39–46]. Interestingly, peptides of most of these proteins were found in our EV proteome (Tc24, SA85, CRP, MASP, TS, tryparedoxin-peroxidase, paraflagellar rod proteins, etc). We propose that immunization with some of the molecules delivered into EVs with proper adjuvanticity, could allow the host to develop an adequate immune response against T. cruzi.
The prime and boost vaccination scheme employed here mostly triggered a humoral mediated immune response able to block or neutralize surface anchored and/or secreted TcTASV-C. Although yet unknown, we speculate that the possible function of the TcTASV-C subfamily is exerted through its most conserved motif (tasv_c), which encompasses a 50 amino acid long sequence at the amino terminus of the protein. Also, the shorter tasv_all motif common to all TcTASV subfamilies can be found within the tasv_c motif, but with specific amino acids at certain positions for each subfamily (see Fig 1). Interestingly, a linear B-cell epitope located within the tasv_all-tasv_c motif was exclusively recognized by sera from TcTASV-C vaccinated mice (P46-62, Fig 11). This reactivity seems to be specifically prompted by the prime and boost vaccination scheme since sera from infected unvaccinated mice are unable to react with this peptide, suggesting that antibodies against this motif are mediating the TcTASV-C neutralization achieved–at least partially- by this vaccination scheme during the early infection.
Packaging molecules into EVs can also be considered as a parasite driven strategy to escape from the host immune surveillance or extracellular degradation until they reach the target cells or tissues. In our hands, EV proteins contained in trypomastigote-secreted EVs were not detected by sera from infected hosts from different species, in contrast with trypomastigote-associated and freely secreted antigens, that were recognized by sera from infected hosts. Indeed, this finding suggests that ~30% of sera from infected hosts that do recognize TcTASV-C actually reacted against proteins attached to the parasite’s surface or freely-secreted to the environment, but not against the TcTASV-C genes that are secreted contained into EVs. We support the hypothesis that secretion of cargo in EVs (and particularly the secretion of TcTASV-C) is another parasite-driven immune evasion mechanism. In a recently published work, Bautista-Lopez et al (2017) [14] looked for “Trypomastigote Excreted Secreted Antigens” (TESA, because the whole secreted population was analyzed) that are exposed to the host immune system. They carried out an immune capture assay with T. cruzi-infected patient’s antibodies to screen for novel and secreted antigens, which could be useful markers of disease status. In accordance with our results, and although TcTASV-C peptides were found in the TESA proteome, none of TcTASV proteins were revealed by patient’s sera. Altogether, these results reinforce the idea that most of the proteins delivered into EVs are hidden from the host or, at least, are hard to be detected in the way they are presented to the host immune system.
In 2016, Queiroz et al [15] published the first proteomic analysis of the trypomastigote secretome (which included both free and EV-secreted proteins) from the Y strain (DTU TcII), and soon after Bautista Lopez et al (2017) [14] presented the proteome of EVs derived from the culture of both cells and trypomastigotes (Tulahuen strain; DTU TcVI). In both of these proteomes TcTASV peptides were eventually identified, supporting the results presented here that demonstrated that, despite being a medium-size gene family, TcTASV proteins are an important component of the trypomastigote secretome and EVs. Regarding the expression of TcTASV-C in the trypomastigote EVs, it is notable that TcTASV-C is easily detected by western blot, suggesting it as a major EV component, especially in contrast with its weak expression on parasite’s body of culture-derived trypomastigotes. Although TcTASV-C is hard to be detected in culture-derived trypomastigote homogenates (undetectable for the conditions of western blot in Fig 3B, upper panel), it is revealed as a major component of EVs in CL Brener strain. In fact, the identification of peptides from 4 different TcTASV-C genes in our EV proteome corresponds with this observation and also with the 2D gel results, where 4 spots were detected as TcTASV-C proteins in the secreted fraction. The picture obtained for TcTASV-C could indicate that the paucity of TcTASV-C in the parasite’s body probably reflects that most of the protein produced is delivered to the secretory route, thus suggesting that its putative function is related somehow to the development of a permissive environment for early T. cruzi settlement. It is well known, but poorly documented, that parasites–and particularly T. cruzi trypomastigotes- express a very different set of molecules when isolated from the host (i.e. in vivo infection) than from culture (i.e. in vitro infection), basically in response to the pressure of the immune system. Here, we demonstrate that TcTASV-C expression is much higher in bloodstream than in culture-derived trypomastigotes. We show that this is true for different T. cruzi strains and also that it is specific for TcTASV-C, but not for other antigens or virulence factors of trypomastigotes. We still do not known what factors of the vertebrate host trigger this expression, which is a matter of our current research. Besides, these results highlight the relevance of working with trypomastigotes obtained from in vivo sources to study the T. cruzi biology, especially when research involves parasite stages that are under immune system pressure in the vertebrate host.
The delivery of virulence factors in exosomes or extracellular vesicles is also a strategy to interfere with host cell signaling pathways required to control infection. Exposure of mice to exosomes of L. infantum resulted in higher parasitic loads in spleen, which was linked to a suppressive T cell phenotype [30,31]. As well as Leishmania exosomes display immunomodulatory properties, T. cruzi extracellular vesicles also do. In fact, Nogueira et al (2015) [47] found that different T. cruzi strains secreted different concentration of vesicles. This variability could not be associated with the current T. cruzi DTU classification because -for example- two TcI strains presented polar secretion levels (Col vs. YuYu). Protein concentration and alpha-galactosyl residues in secreted EVs also varied among the different strains, and without a lineage specific association [47]. Focusing on the modulation of immune responses by EVs in the acute phase of infection, only after stimulation with EVs from YuYu (DTU TcI) and CL-14 (DTU TcVI), peritoneal macrophages from C57BL/6 mice produced high levels of proinflammatory cytokines (TNF-alpha) and NO, via the TLR-2. This profile was not stimulated by EVs from other T. cruzi strains. Similar findings were recently reported by Clemente et al (2017) [48] employing EVs secreted by metacyclic trypomastigotes from other strains. In the present work, we registered a variable expression of TcTASV-C in the different secretory fractions (i.e. V2, V16 and soluble factors) among the different strains analyzed. Although we found similar levels of total protein content in EVs derived from the different strains, it should be stated that the protocols employed to isolate EVs and the strains analyzed were different. As in previous works, we could not link a particular secretion profile with a certain T. cruzi DTU or strain. This complex scenario led us to speculate that differences in EVs cargo could reflect the broad spectrum of clinical manifestations observed in Chagas’ disease. Our opinion is that we are still building a puzzle from somehow complementary but still fragmented data, showing currently a complex and not very well understood picture.
In brief, we have demonstrated that TcTASV-C is a major component of bloodstream trypomastigotes, and that TcTASV-C is mainly secreted, either contained into EVs or free. Besides, although with the prime and boost strategy employed TcTASV-C did not result a promising vaccine candidate, it was possible to interfere with the early acute phase of T. cruzi infection. Indeed, the strong anti-TcTASV-C humoral immune response elicited by immunizations allowed to understand–partially- the TcTASV-C functionality; we hypothesize that TcTASV-C is involved in the establishment of the initial T. cruzi infection in the mammalian host. Although we highlight TcTASV-C as a potential antigen to bit the parasite in the early acute phase, we bear in mind that an effective vaccine to control Chagas’ disease should include other antigens and/or trigger also other arms of the host immune response. Ultimately, results presented here strongly highlight TcTASV-C as a novel secreted virulence factor of T. cruzi trypomastigotes.
All protocols conducted with animals were designed and carried out in accordance with international ethical standards for animal experimentation (Helsinki Declaration and its amendments, Amsterdam Protocol of welfare and animal protection and National Institutes of Health, USA NIH, guidelines) and were approved by the Institutional Animal-Care Ethics Committee of the University of Buenos Aires (CICUAL, res number: 2846/2013) and from University of San Martin (CICUAE, protocol number: 01/2012 and 08/2016).
TcTASV-CGST (amino acids 65 to 330 of ORF Tcruzi_1863-4-1211-93) was already cloned in our laboratory in pGEX-3X and was expressed and purified as we previously described [10]. The same procedure was used with GST.
Amino acids 52 to 342 of the TcTASV-C gene AM492203 (GenBank; emb.CAM33606.1) were cloned between BamHI and KpnI restriction sites, fused in the N-term to a Histidine tag into pQE-30 (Qiagen). A point mutation was introduced to change the amino acid H56R. TcTASV-CHIS was expressed and purified by standard methodologies for histidine-tagged proteins (The QIAexpressionist). Purity of proteins was analyzed by SDS–PAGE, followed by staining with Coomassie Brilliant Blue. Proteins were quantified (Bradford assay and/or Picodrop) and dialyzed against PBS. Recombinant proteins were stored aliquoted at −80°C until use.
For mice immunizations, purified recombinant proteins were incubated with a Polymyxin B resin in a column format (Detoxy-Gel Endotoxin Removing Gel Thermo Scientific). Endotoxin levels were quantified by Amebocyte lysis assay (Limulus Amebocyte Lysate Test, Lonza). Only preparations with endotoxin levels <100 U/mg were used.
Recombinant TcTASV-CGST was digested with Factor Xa (GE Healthcare) and the purified TcTASV-C fragment was used to produce specific anti-TcTASV-C sera in mice [49]. The specificity of the anti-TcTASV-C sera was verified by competition assays and western blot, both against trypomastigotes lysates and recombinant proteins.
Recombinant TcTASV-CGST was used to produce complete anti-TcTASV-C-GST serum in mice, following the same immunization protocol described above. The sera obtained reacted both with TcTASV-C and GST.
Vero and J774 cells were grown at 37°C in a 5% CO2 humidified atmosphere in MEM or RPMI (Gibco), respectively, supplemented with 10% fetal bovine serum (Natocor), 10 μg/mL streptomycin (Sigma), 100 U/mL penicillin (Sigma).
Cell-derived T. cruzi trypomastigotes were cultured by passages in Vero cells at 37°C and 5% CO2 humidified atmosphere in MEM (Gibco Life Technologies) supplemented with 10% fetal bovine serum, 10 μg/mL streptomycin, 100 U/mL penicillin. Trypomastigotes were harvested from supernatants of infected cells as previously described [10]. As a rule, T. cruzi stocks are kept in liquid nitrogen and all strains are regularly thawed twice a year to preserve their biological characteristics. Parasites from Sylvio (TcI), 193–173 (TcI), K98 (TcI), Y (TcII), Tul (TcVI), VD (TcVI), CL Brener (TcVI) and RA (TcVI) were employed [50–54].
Bloodstream trypomastigotes of the RA strain (DTU TcVI) were maintained in vivo by weekly passages in CF1 mice with 105 trypomastigotes, at IMPaM (School of Medicine, University of Buenos Aires-CONICET) and at the BLS3 laboratory at UNSAM. The Tulahuen strain expressing E. coli β-galactosidase (Tul-β-gal) was also maintained in vivo by passages on CF1 mice at UNSAM [25]. Purification of bloodstream trypomastigotes was essentially carried out by a Ficoll gradient with a swinging bucket rotor, essentially as previously described [55]. Bloodstream RA trypomastigotes used for EV purification, were either purified as stated above or by swimming (2 x 40 min at 37°C), essentially as described by Miranda et al (2015) [56]. Briefly, heparinized blood was diluted with 3 volumes of PBS and, after centrifuged at 300 x g for 5 min, the sample was incubated for 40 min at 37°C. The supernatant containing parasite forms was then carefully harvested–to exclude the erythrocyte containing phase–and the procedure was repeated twice. Then, trypomastigotes were pelleted, washed with PBS-1% BSA, resuspended in MEM and incubated for shedding assays, as described below. Similar volumes of blood from non-infected mice were processed in parallel and used as controls.
Cell-derived trypomastigotes, were washed with MEM without serum and incubated at a concentration of 108 parasites/ml in MEM at 37°C, during 6 hours, in a 5% CO2 humidified atmosphere. Trypomastigote-secreted products (soluble plus vesicles) were isolated as previously described [10]. A similar procedure was carried out for bloodstream trypomastigotes.
Extracellular vesicles were purified by an iodixanol density gradient (Optiprep, Sigma) ultracentrifugation as described by van Deun J et al (2014) [57] or by sequential ultracentrifugation as described by Bayer-Santos et al (2013) [22]. Briefly, for both procedures, after shedding, parasites were removed by centrifugation and the cell-free supernatant filtered through a 0.45-μm syringe filter (Micron Separation Inc.).
A discontinuous gradient was created by layering 2,7 mL of 40%, 20%, 10% and 2,3 mL of 5% Optiprep solutions from bottom to top in a 13,2 mL polyallomer tube (Beckman Coulter). The cell-free supernatant of trypomastigotes (EVs plus free secreted fraction, 2 ml) was overlaid onto the top of the gradient, which was then centrifuged for 18 hours at 100,000 x g without brake at 4°C in a SW 41 Ti rotor in an Optima XL 100k ultracentrifuge (Beckman Coulter). Gradient fractions of 1 mL were collected from the top of the gradient. Density was determined weighing on an electronic balance a known volume of each fraction.
Alternatively, to isolate large and small extracellular vesicles, EVs plus the free secreted fraction were centrifuged at 100.000 x g for 2 h at 4°C to obtain the first pellet, enriched in large extracellular vesicles (V2), and the resulting supernatant was centrifuged again at 100.000 x g for 16 h at 4°C, to obtain the second pellet, enriched in small extracellular vesicles (V16), and the EV-free supernatant fraction (VF). All ultracentrifugation steps were carried out in a 70Ti fixed angle rotor in an Optima XL 100k ultracentrifuge (Beckman Coulter).
All relevant methodological data of our EV’s isolation procedures have been submitted to the EV-TRACK knowledgebase (EV-TRACK ID: EV170020) [58].
Extracellular vesicles were resuspended in Hepes Buffer, pH 6.5, and fixed with paraformaldehyde (4% in Hepes). Negative staining was carried out on grids coated with acrylic membranes and graphene oxide. Extracellular vesicles were stained with 5 μl of 0.5% ammonium molybdate at pH 7.5, and observed using a Zeiss EM 109T transmission electron microscope operating at 80kV; the images were acquired with a Gatan ES1000W (11 Mpx) digital camera.
For 2D gel electrophoresis, trypomastigotes were incubated in serum-free DMEM for 2 h a 37°C (or at 0°C for controls). The medium containing the secreted antigens and the parasites were separated by centrifugation at 4000 x g for 10 min at 4°C. Pelleted parasites were washed twice in 10 mM Tris-Cl, pH 7.0, 25 mM sorbitol and, after being pelleted by centrifugation, were lysed by vortexing for 30 s in 250 μl of IEF rehydration buffer (9M urea, 2M thiourea, 2% CHAPS, 65 mM DTT, 0.5% IPG buffer [Amersham Pharmacia] and 0.002% bromophenol blue) with protease inhibitor cocktail (Roche). The secreted material was also mixed with IEF rehydration buffer and both the trypanosome lysate and the secreted antigens were incubated at R.T for 1 h, with vortexing for 30 s every 15 min, as described by van Deursen et al (2003) [59]. Samples were loaded into Immobiline DryStrip (pH 4–7, 13 cm; GE Healthcare) and isoelectric focusing carried out in an IPGphor Isoelectric focusing System for 24 h. Second-dimension SDS-PAGE was carried out in a Hoeffer SE 600, and gels were transferred to nitrocellulose membranes in a semi-dry TE 70 PWR (Amersham Biosciences). Blocking and washing solutions and antibodies used were similar to those described below for conventional western blot.
Thirty million (30x106) of in vitro cell-derived trypomastigotes, or its secretion equivalent from small (V16), large (V2), total EVs or the soluble EV-free fraction (VF) were electrophoresed on 10% denaturing polyacrylamide gels, and transferred to nitrocellulose membranes by standard methodologies [60]. The correct transfer was verified by reversible membrane staining with Ponceau Red (5% w/v) in 1% (v/v) acetic acid. The membrane was blocked with PBS-3% non-fat milk for 1 hour, washed with PBS-0.05% Tween and incubated with primary antibodies. Anti TcTASV-C (mouse, 1/400), anti HSP-70 (rabbit, 1/1000) and anti-SR62 (rabbit, 1/1000) were employed. Then, washes were repeated and membranes were incubated with a peroxidase-conjugated secondary antibody (anti-mouse or anti-rabbit, both from Thermo Scientific) for 1 hour and the washes repeated. For the detection, we used a chemiluminescent reagent (SuperSignal West Pico, or SuperSignal West Femto, Thermo Scientific). The emitted signal was detected by exposure on radiographic plates (AGFA). For bloodstream trypomastigotes, or its secretion equivalent from EVs, an additional blocking step with non-labelled anti-mouse IgG (Sigma-Aldrich) before incubation with the primary antibodies was included. Western blots were developed as indicated above or with Alexa Fluor 590 goat anti-mouse IgG or Alexa Fluor 680 goat anti-rabbit IgG as secondary antibodies (Invitrogen) at a 1:20000 dilution and visualized with an Oddysey Infrared Imager (Li-Cor).
Purified EVs (20 μg) were diluted in 50 mM ammonium carbonate. Mass spectrometry analysis was carried out at Centro de Estudios Químicos y Biológicos por Espectrometría de Masa (CEQUIBIEM), Argentina, in a Q Exactive HESI-Orbitrap coupled to a nano HPLC Easy-nLC 1000 (Thermo Scientific). MS/MS data were used to search the all the available Trypanosoma cruzi databases at Tritrypdb (version 30) [11].
Interaction assays were carried out by an ELISA-like assay, as described by Baida et al (2006) [61]. Briefly, macrophage (J774) or epithelial (Vero) cells were cultured overnight, washed with PBS-3% BSA and fixed with 1% paraformaldehyde in PBS for 15 minutes. The fixed cells were blocked with PBS-3% BSA-1% normal goat serum for 1 hour at room temperature and washed again. Recombinant proteins (TcTASVGST or GST) were incubated for 1 hour at 37°C. The cells were washed and then incubated for 1 hour at 37°C with complete anti-TcTASVGST sera, which recognizes both TcTASVGST and GST proteins. Normal mouse serum was used as background control. Detection continued as for conventional ELISA technique. Three replicates per condition and three independent tests were carried out. Data were analyzed by Student t-test. A similar protocol was employed to assay the interaction EVs with Vero cells. Briefly, cells were incubated with freshly isolated EVs for 1.5 h at 37°C, washed and the interaction detected by a pool of sera developed against soluble and membrane antigens of trypomastigotes. Normal mouse sera and frozen EVs were used as controls.
The ability of rTcTASV-C to interfere with parasite infection on Vero cells was assessed in vitro by two different methods and with two T. cruzi strains. In both set ups cells were incubated with rTcTASV-C or GST (as a control), before infection. On one hand, 20000 Vero cells/well were incubated in p24 Wells (Costar) for 24 hs at 37°C. Then the cells were washed and incubated with recombinant proteins (TcTASV-CGST or GST) in MEM 4% FBS at 37°C for 30 min. CL Brener trypomastigotes (10:1) were added to the cultures, and 18 h later uninternalized parasites were washed and infection proceeded for additional 48 hs. Cells were then fixed and stained with May-Grünwald Giemsa. At least 500 cells were counted in each technical replicate, and the presence of amastigotes registered. Data were normalized to infected (untreated) cells; 3 independent experiments were performed with 3 technical replicates each one. On the other hand, T. cruzi bloodstream trypomastigotes (Tulahuen strain) expressing E. coli β-galactosidase were used to infect treated Vero cells in p96 (Costar) in a relation of 10:1 [25,26]. After an overnight incubation (37°C, 5% CO2), cells were washed with PBS to remove non-infecting trypomastigotes and the culture maintained for additional 72 hs. Cells were then lysed with Igepal (1% v/v) and β-galactosidase activity was spetrophotometrically measured with the chromogenic substrate chlorophenol red β-D-galactopyranoside (CPRG). Reaction was read at 595 nm in a multi-plaque reader FilterMax F5 (Molecular Devices).
Purified T. cruzi bloodstream trypomastigotes (Tul- β-gal) were pretreated for 30 min at 37°C with anti-TcTASV-C sera (1/10) and then co-incubated (37°C, 5% CO2, 18 h) with Vero cells (ratio 10:1) in MEM–5% FBS in a 96-well plate format. Parasites pretreated with anti-GST or normal sera were used as controls. Cell culture and quantification of infection were the same as stated above. Untreated parasites were used to determine 100% of infection; 3 independent experiments were performed with 3 technical replicates each one.
For the preparation of plasmid DNA used in immunizations, E. coli DH5a containing a fragment of the TcTASV-C gene TcCLB.511675.3 (amino acids 233 to 305) cloned in pCI_Not_32 [12] or the plasmid VR1019 that contains the murine GM-CSF gene [26] were first grown as starter cultures in LB containing ampicillin at 37°C for 8 hours, then inoculated into a larger culture and grown O.N. and, finally, incubated additional 8 h in the presence of chloramphenicol (170 μg/ml) for amplification of plasmid copy number.
Plasmid DNA was purified with the QIAGEN EndoFree Plasmid Mega Kit (QIAGEN, GmbH, Germany) according to manufacturer’s instructions. Purified DNA was resuspended in TE endotoxin-free buffer and DNA concentration was estimated and stored at -20°C. For mice immunization, DNA was precipitated with ethanol and reconstituted at 1 μg/μl with sterile endotoxin-free PBS. The VR1019_GM-CSF plasmid was gently provided by Dr. Walter R. Weiss of the "Malaria Program and Pathology Division, Naval Medical Research Center," Maryland, United States.
C3H/He mice (n = 10 per group) were vaccinated with a prime (plasmid DNA) and boost (recombinant proteins) immunization protocol. Briefly, the first two doses consisted in intramuscular injections of plasmid DNA containing 100 μg of pCI_Not-TcTASV-C and 25 μg of VR1019_GM-CSF [12, 28]. The third and fourth doses consisted in subcutaneous injections of mixed TcTASV-CGST and TcTASV-CHIS (12.5 μg each one) with a colloidal suspension of aluminum hydroxide (Sigma). Control groups were immunized with 100 μg of the empty plasmid backbone pCI_Not_32 plus 25 μg of VR1019_GM-CSF (doses 1 and 2) and 25 μg of GST along with aluminum hydroxide (doses 3 and 4). Fifteen days after the last dose, 3 mice per group were sacrificed to evaluate cellular responses in spleen cells (cytokine production after culture) and the remainder 7 mice challenged with 100 bloodstream trypomastigotes of the RA strain by the intraperitoneal route [12]. Parasitemia was determined from day 7-on every 2–3 days until day 35. Mortality was daily monitored.
Spleens of immunized mice were aseptically removed and homogenized. Red blood cells were lysed and cells cultured in RPMI 1640 supplemented with 2 mM L-glutamine, 100 U of penicillin/ml, 50 μg of streptomycin/ml, and 10% FCS at a concentration of 4 × 106cells/ml in 24 well plates (Nunc). Cells were stimulated with TcTASV-CGST, GST (10 μg/ml), anti-CD3 (0.2 μg/ml) or solely maintained with culture medium (basal control) at 37°C in a humidified atmosphere of 5% CO2. After 72 h, supernatants were collected, and production of gamma interferon (IFN-γ) and interleukin-10 (IL-10) was evaluated by sandwich ELISA according to manufacturer's instructions (BD OptEIA, Pharmingen, San Diego, CA).
Serology against recombinant antigens or whole T. cruzi trypomastigote lysates was determined by ELISA, as we previously described [10,12]. Mice were bled by submandibular puncture to take serum samples 5 days before the immunization schedule start, 15 days after the last dose and at 42–62 days post infection. ELISA plates were sensitized with 50 ng of recombinant proteins or 100 ng of T. cruzi trypomastigote homogenates. Goat anti-mouse IgG, anti-IgG1 or anti-IgG2a conjugated to peroxidase (Thermo Fisher Scientific) were used as secondary antibodies. Reaction was revealed with 3,3’,5,5’-Tetramethylbenzidine (Sigma) and H2O2 in citrate buffer and read at 450 nm in a multi-plaque reader FilterMax F5 (Molecular Devices).
Parasites (500000/assay) were incubated with inactivated sera (53°C, 40 min) from vaccinated or control mice, for 1 h at 37°C, followed by treatment with fresh human sera (1:4) either with active or inactivated complement for an additional 1 h at 37°C [43]. Trypomastigote lysis was calculated by counting living, motile and unstained parasites in a Neubauer chamber after staining with Trypan blue.
Putative B-cell epitopes in TcTASV-C proteins were predicted by Bepipred software [62]. The TcTASV-C epitopes identified in a previous work (peptide microarray screened with human antibodies) were also considered [29]. Peptides were purchased from Genscript and screened by ELISA. Briefly, plates were sensitized with 1 or 0.33 μg of peptide (for 96 or 384 plate format, respectively) in PBS O.N. Sera from vaccinated or infected mice were assayed at 1/100 dilution by triplicate. After incubation with a peroxidase- conjugated secondary antibody the reaction was developed as described above. The cut off was set up for each peptide as the media of the O.D. of the negative (uninfected unvaccinated) sera plus 3SD plus 10%. Reactivity of an experimental serum was classified positive for an X peptide, if the ratio between its O.D. for the peptide X and the cut-off value for peptide X, resulted in a value higher than 1 (i.e. O.D. sera for peptide X/ cut-off peptide X >1).
For ELISA comparisons, we employed one-way ANOVA. Differences in the parasitemia between groups were determined by the Mann–Whitney U test. In survival analysis, groups were compared by the Log-rank test. In all cases, Graph Pad Prism version 5.01 (GraphPad Software, USA) was used and a P value below 0.05 was considered significant.
|
10.1371/journal.pgen.1006047 | Enhancement of Transcription by a Splicing-Competent Intron Is Dependent on Promoter Directionality | Enhancement of transcription by a splicing-competent intron is an evolutionarily conserved feature among eukaryotes. The molecular mechanism underlying the phenomenon, however, is not entirely clear. Here we show that the intron is an important regulator of promoter directionality. Employing strand-specific transcription run-on (TRO) analysis, we show that the transcription of mRNA is favored over the upstream anti-sense transcripts (uaRNA) initiating from the promoter in the presence of an intron. Mutation of either the 5′ or 3′ splice site resulted in the reversal of promoter directionality, thereby suggesting that it is not merely the 5′ splice site but the entire splicing-competent intron that regulates transcription directionality. ChIP analysis revealed the recruitment of termination factors near the promoter region in the presence of an intron. Removal of intron or the mutation of splice sites adversely affected the promoter localization of termination factors. We have earlier demonstrated that the intron-mediated enhancement of transcription is dependent on gene looping. Here we show that gene looping is crucial for the recruitment of termination factors in the promoter-proximal region of an intron-containing gene. In a looping-defective mutant, despite normal splicing, the promoter occupancy of factors required for poly(A)-dependent termination of transcription was compromised. This was accompanied by a concomitant loss of transcription directionality. On the basis of these results, we propose that the intron-dependent gene looping places the terminator-bound factors in the vicinity of the promoter region for termination of the promoter-initiated upstream antisense transcription, thereby conferring promoter directionality.
| Eukaryotic genes differ from their prokaryotic counterparts in having intervening non-coding sequences called introns. The precise biological role of introns in eukaryotic systems remains unclear even more than forty years after their initial discovery. One function of intron that has been remarkably conserved during evolution is their ability to enhance the transcription of genes that harbor them. How does the intron regulate transcription, however, is not known. Here we show that the intron enhances gene expression by affecting direction of the promoter-initiated transcription. In the presence of an intron, polymerase tends to transcribe the downstream coding region producing mRNA, while in the absence of a splicing-competent intron polymerase starts transcribing promoter upstream region producing upstream antisense RNA (uaRNA). Intron-mediated promoter directionality was dependent on gene looping, which is the interaction off the promoter and terminator region of a gene in a transcription-dependent manner. We show that the intron-dependent gene looping facilitates the recruitment of termination factors in the promoter-proximal region. The recruited termination factors stop uaRNA synthesis thereby conferring directionality to the promoter-bound polymerase.
| Although introns were discovered more than four decades ago, their precise physiological role in biological systems still remains an enigma [1, 2]. One of the evolutionarily conserved functions of introns in eukaryotes is in regulation of the mRNA output of a gene [2–6]. The promoter-proximal introns often stimulate transcription of genes that harbor them [4, 7–13]. This phenomenon of enhancement of transcription by a splicing-competent intron is called ‘intron-mediated enhancement of transcription’ (IME) [1, 2, 4, 6, 14].The discovery of IME coincided with the development of cDNA technology. It was observed that the expression of the cDNA version of a gene is much less efficient than its native intron-containing counterpart in transfected mammalian cell lines [12, 15]. It was soon realized that the effect of an intron on transcription is a general feature of all eukaryotic organisms, including yeast, flies, worms, plants and humans [4]. Despite the ubiquity, the molecular mechanism underlying the phenomenon remains elusive even more than 25 years after its initial discovery.
Although less than 5% of genes in budding yeast contain introns, the intron-containing genes contribute nearly 28% of mRNA produced in yeast cells [3, 16]. We previously demonstrated that the intron-mediated enhancement of transcription in yeast involves gene looping, which is the physical interaction of the promoter and terminator regions of a gene in a transcription-dependent manner [10]. How the intron-facilitated looped gene architecture brings about enhancement of transcription, however, was not clear. A clue came when our laboratory and others demonstrated that gene looping confers directionality to the promoter-initiated transcription [17, 18]. The eukaryotic promoters and terminators are generally located in nucleosome free regions. Genomewide analysis has revealed that the promoters of most RNAPII-transcribed genes are bidirectional [19–27]. The transcription initiates in both the sense and upstream antisense directions from these promoters. Transcription in sense direction produces mRNA, while upstream antisense transcription generates non-coding transcripts called uaRNA (upstream antisense RNA) or PROMPT (Promoter upstream transcript) [19, 21, 26]. Transcription of the mRNA continues till the polymerase reaches the 3′ end of the gene, whereas synthesis of uaRNA is terminated when the transcript is just a few hundred to a thousand kilobase long [20]. This phenomenon is referred to as ‘promoter directionality’ [28]. It is generally believed that the uaRNA synthesis in yeast is terminated in a poly(A)-independent manner by the Nrd1-Nab3-Sen1 complex [29, 30]. In mammalian systems, however, uaRNA transcription is terminated by the same cleavage and polyadenylation machinery that stops mRNA synthesis at the 3′ end of a gene in a poly(A)-dependent manner [19, 21]. A number of reports suggest that the components of cleavage and polyadenylation machinery are involved in the termination of yeast uaRNA transcription as well [17, 18, 29–32]. In both yeast and higher eukaryotes, uaRNA is immediately degraded by the RNA surveillance machinery. Because of their short half-life, yeast uaRNA species are often referred to as cryptic unstable transcripts (CUTs) [23, 24]. In mammalian cells, the asymmetric distribution of poly(A) site and U1 snRNA-binding sites in the promoter-proximal region is believed to contribute to transcription directionality [19, 21]. In budding yeast, however, gene looping has been shown to confer promoter directionality [17, 18]. How gene looping enhances transcription directionality, however, is not clear.
Here we show that the transcription directionality of a subset of genes in yeast is dependent on the presence of a splicing-competent intron. The intron facilitates the recruitment of CF1, CPF and Rat1 termination complexes in the vicinity of the promoter region. We provide evidence that the intron-dependent gene looping facilitates the recruitment of termination factors near the promoter region. The recruited termination factors selectively terminate uaRNA synthesis, thereby conferring directionality to the promoter-initiated transcription.
Research conducted during last eight years has confirmed that the nucleosome free region located at the 5′ end of most RNAPII-transcribed genes contains two unidirectional promoters [25, 26]. Each of these promoters assembles its own preinitiation complex (PIC) and is competent to initiate transcription [25, 27]. Mechanisms are in place in the cell to limit upstream antisense transcription and promote transcription in the sense direction. In mammalian systems, asymmetric distribution of 5′ splice sites and poly(A) sites in the promoter-proximal region has been shown to play a crucial role in conferring promoter directionality [19, 21]. The presence of 5′ splice sites in the promoter downstream region inhibits poly(A)-dependent termination of transcription, while the absence of 5′ splice sites in the promoter upstream region allows poly(A)-dependent termination of transcription in that region. The net result of this arrangement is that the synthesis of mRNA is favored over uaRNA. There is no such asymmetric distribution of U1 binding sites near yeast promoters [30, 33], but a subset of yeast genes contain introns. Since a promoter-proximal intron enhances transcription of mRNA, we hypothesized that the intron-mediated enhancement of mRNA synthesis could be, at least in part, due to the effect of the intron on promoter directionality.
To test this hypothesis, we examined transcription of three intron-containing genes, IMD4, ASC1 and APE2, in the promoter-proximal upstream antisense and downstream sense direction in the presence and absence of an intron. We constructed strains with the intron-less version of these three genes following the strategy described in Moabbi et al., (2012) [10]. The mRNA and uaRNA levels were then compared by reverse transcription-polymerase chain reaction (RT-PCR) in the presence and absence of the intron. RT-PCR analysis revealed that the mRNA level of IMD4, ASC1 and APE2 deceased by 2.5 to 10 fold upon deletion of the intron (S1B Fig). In contrast, the uaRNA content of all three genes registered an increase in the absence of intron. The uaRNA level of APE2 increased by about 15 fold, while that of ASC1 and IMD4 by about 1.6 fold upon deletion of intron (S1B Fig). These results suggested that the intron could be playing a role in regulating the direction of promoter-initiated transcription. The presence of an intron favored synthesis of mRNA over uaRNA, while the absence of intron switched direction of transcription so as to favor the uaRNA synthesis.
There was, however, a possibility that the observed alteration in the steady state level of mRNA and uaRNA was not due to the effect of intron on transcription, but on the stability of transcripts. The presence of intron could somehow stabilize mRNA, but facilitate the degradation of uaRNA by exosomes. To rule out this possibility, we performed strand-specific transcription run-on (TRO) analysis as described in Medler and Ansari (2015) [34]. Briefly, the technique involved labeling the nascent transcripts with Br-UTP, purifying Br-UTP labeled RNA using anti-Br-UTP affinity beads, and then subjecting affinity purified nascent RNA to RT-PCR analysis as described above. Strand-specific TRO analysis revealed about a 2 to10 fold decrease in nascent transcription of IMD4, ASC1 and APE2 in the sense direction (mRNA synthesis) upon deletion of intron (Fig 1B). Simultaneously, there was about a 3–10 fold increase in upstream antisense transcription (uaRNA synthesis) in the absence of the intron (Fig 1B). These results confirmed the findings observed above in (S1 Fig), and corroborated the role of the intron in promoter directionality for the three genes examined here.
To gain an insight into the role of the intron in promoter directionality, we calculated the directionality index by dividing nascent mRNA level with nascent uaRNA level for each tested gene in the presence and absence of an intron. The directionality indices in the presence of intron for these three genes ranged from 4 to 25 (Fig 1C). Upon deletion of the intron, the directionality index registered a decline by about 50–250 fold (Fig 1C).
Having demonstrated the role of intron in promoter directionality in budding yeast, we next asked if it is just the 5′ splice site as has been shown in mammalian systems or the whole splicing-competent intron that confers directionality to the promoter-initiated transcription in yeast cells. We therefore inserted a wild type, a 5′ splice site mutated, and a 3′ splice site mutated ACT1 intron into an intron-less IMD4 gene as previously described (Moabbi et al., 2012) [10]. The 5′ splice site was mutated from GT to CA, while 3′ splice site region was mutated from AG to GC (Moabbi et al., 2012) [10]. Both mutations abolished splicing as a longer mRNA was produced (S2B Fig). The strand-specific TRO analysis was then carried out to detect transcription of mRNA and uaRNA of IMD4 in the presence of a wild type native intron, in the presence of a wild type ACT1 intron, in the absence of an intron, in the presence of 5′ splice site mutated and a 3′ splice site mutated ACT1 intron. The insertion of a wild type ACT1 intron brought about a 10-fold increase in transcription of mRNA (Fig 2B). The enhancement of IMD4 transcription by the ACT1 intron was almost to the extent conferred by its native intron. As expected, the mutation of either the 5′ or 3′ splice site failed to enhance transcription of IMD4 (Fig 2B). Simultaneously, we compared transcription of uaRNA. There was little detectable uaRNA signal in the presence of the native or wild type ACT1 intron (Fig 2B). A 6–8 fold increase in nascent uaRNA signal was observed in the presence of a 5′ splice site mutated intron (Fig 2B). The mutation of the 3′ splice site gave similar results (Fig 2B). The drop in directionality index upon mutation of the 5′ or 3′ splice site was almost to the same extent (70–100 fold) as in the absence of an intron (Fig 2C). A logical conclusion of these results is that it is not the 5′ splice site alone, but the whole splicing-competent intron that confers transcription directionality to a subset of yeast genes.
In mammalian cells, uaRNA synthesis is terminated in a poly(A)-dependent manner by the cleavage and polyadenylation machinery [19, 21]. In contrast, promoter-initiated upstream antisense transcription in yeast is believed to be terminated by the Nrd1-Nab3-Sen1 complex in a poly(A)-independent manner [29, 30]. We hypothesized that the presence of a splicing-competent intron facilitates the recruitment of termination factors in the vicinity of the promoter region. The recruited termination factors stop upstream antisense transcription, thereby providing directionality to the promoter-initiated transcription. To test this hypothesis, we examined the recruitment of termination factors in the promoter-proximal region of a gene in the presence and absence of an intron. Although uaRNA in yeast belongs to the category of CUTs, which are predominantly terminated by the Nrd1-Nab3-Sen1 complex, the recent studies have also implicated CPF subunit Ssu72 and CF1 subunit Pcf11 in the termination of uaRNA transcription [17, 29, 31, 32]. We therefore checked for the presence of all four termination complexes; CF1, CPF, Rat1 and Nrd1 complexes, in the promoter-proximal region of IMD4 and ASC1 genes by ChIP as described in Al Husini et al., (2013) [18]. The termination factor ChIP was performed in strains with intron-containing or intron-less versions of the gene under investigation. The promoter occupancy of CPF complex was monitored in terms of recruitment of its Pta1 subunit, while CF1 complex recruitment was detected using its Rna15 subunit. Similarly, Rat1 complex recruitment was monitored using its Rat1 subunit, and Nrd1 complex was tracked using its Nab3 subunit. The strains carrying epitope-tagged version of these termination factors were generated to facilitate ChIP.
We first examined the recruitment of Nrd1 complex subunit Nab3 at IMD4 and ASC1 in the presence and absence of an intron. Nab3 was recruited at both the 5′ and 3′ ends of IMD4 with almost equal intensity in the presence of an intron (Fig 3B). Upon deletion of intron, there was no appreciable change in the Nab3 occupancy of either the 5′ or the 3′ end of IMD4 (Fig 3B). Although Nab3 crosslinking to the 5′ end of ASC1 was about 60% less than that at the 3′ end of the gene, still no significant change in Nab3 crosslinking was observed at the 5′ end of gene in the absence of the intron (Fig 3C). These results suggest that the recruitment of the Nrd1 complex at the 5′ end of IMD4 and ASC1 genes is not dependent on the presence of an intron. We then checked for the recruitment of CF1, CPF and Rat1 complexes in the vicinity of the promoter of IMD4 and ASC1 by ChIP. The CF1 subunit Rna15 was found crosslinked to both the ends of IMD4 and ASC1 in the presence of an intron (Fig 3B and 3C). These results are in agreement with our published results that the CF1 complex occupies distal ends of a number of yeast genes in a transcription-dependent manner [18, 34, 35]. In the absence of an intron, however, both the 5′ and 3′ occupancy of Rna15 decreased. The Rna15 signal at the promoter of IMD4 and ASC1 decreased by about 3.5 fold and 2.2 fold respectively upon deletion of the intron (Fig 3B and 3C). A similar reduction in the promoter occupancy of Pta1, which is a subunit of CPF complex, and Rat1 was observed for both IMD4 and ASC1 in the absence of an intron (Fig 3B and 3C). The promoter crosslinking of Pta1 decreased by about 2–12 fold, and that of Rat1 by about 2–4 fold in the intron-less versions of these two genes (Fig 3B and 3C). The overall conclusion of these results is that the recruitment of CF1, CPF and Rat1 cleavage-polyadenylation/termination complexes at the 5′ end of IMD4 and ASC1 occurs in an intron-dependent manner. Furthermore, the promoter occupancy of these termination complexes coincides with the enhanced directionality of promoter-initiated transcription. A corollary of these observations is that the intron-dependent recruitment of termination factors near the 5′ end of genes could be playing a critical role in transcription directionality.
The experiments described above clearly demonstrated an increase in the promoter recruitment of factors required for poly(A)-dependent termination of transcription in the presence of an intron. It was, however, not clear if the recruited termination factors were enhancing transcription directionality by affecting uaRNA transcription. We therefore examined nascent uaRNA and mRNA levels in the promoter-proximal region of IMD4, ASC1 and APE2 in temperature-sensitive mutants of RNA15 (rna15-2) and PTA1 (pta1-td) by strand-specific TRO approach. The results show that uaRNA transcription increased by about 5-fold upon shifting of rna15-2 cells to elevated temperature (Fig 4B). A similar increase in nascent uaRNA level was observed when pta1-td cells were shifted to non-permissive temperature (Fig 4D). The increase in uaRNA level in pta1-td mutant, however, was to a lesser extent (about 3-fold). No such increase in uaRNA transcription was observed in the isogenic wild type cells at elevated temperature (S3B Fig). In contrast, mRNA transcription registered a decline upon shifting of mutants to 37°C for all three genes (Fig 4B and 4D). The enhanced uaRNA synthesis in rna15-2 and pta1-td mutants at elevated temperature was accompanied by a concomitant decrease in directionality indices (5–10 fold) (Fig 4C and 4E). These experiments strongly suggest that the termination factors at the promoter are enhancing transcription directionality by preventing uaRNA transcription.
Next we asked how the presence of an intron facilitates the recruitment of termination factors in the promoter-proximal region. A clue came from our previous observation that a gene assumes a looped conformation in the presence of an intron (Moabbi et al., 2012) [10]. A gene loop is formed due to the physical interaction of the terminator region of a gene with its cognate promoter in a transcription-dependent manner [36]. Gene looping has been shown to affect promoter directionality in budding yeast [17, 18]. We hypothesized that it is the looped gene structure formed in the presence of a splicing-competent intron that facilitates the recruitment of terminator-bound factors to the promoter end of a gene owing to the close physical proximity of the promoter and terminator regions. We have already demonstrated the intron-dependent gene looping of INO1 and ASC1 [10], but it was not clear if IMD4 and APE2 also exhibit a similar intron-dependent change in gene conformation. We therefore performed ‘Chromosome Conformation Capture’ (CCC) analysis of IMD4, ASC1 and APE2 in the presence and absence of their native wild type intron in the same batch of cells that were used for measuring transcription directionality in Fig 1 above, following the protocol described in El Kaderi et al., (2012) [37]. The CCC assay measures gene looping in terms of a PCR product obtained using P1T1 primer pair that flanks the promoter and terminator regions as shown in Fig 5A. A robust P1T1 looping signal was observed for IMD4, ASC1 and APE2 in the presence of the native wild type intron (Fig 5B). The looping signal decreased by about 3 fold in the absence of an intron (Fig 5B). Thus, the promoter-proximal recruitment of termination factors at all three genes used in this analysis was accompanied by the gene assuming a looped architecture.
Intron-dependent gene looping, however, is different from transcription-dependent looping of non-intronic genes. It is characterized by additional interactions of the intron with gene ends and requires functional 5′ and 3′ splice sites [10]. To corroborate the role of intron-dependent gene looping in the promoter-recruitment of termination factors, we measured conformation of IMD4 gene in the presence of a wild type, a 5′ splice site mutated as well as a 3′ splice site mutated ACT1 intron. A robust P1T1 looping signal was observed for IMD4 in the presence of the wild type ACT1 intron, which was almost to the same extent as in the presence of the native intron (Fig 5D). Mutation of either the 5′ or 3′ splice sites resulted in a decrease in looping signal, almost to the same extent as in the absence of intron (Fig 5D). We reasoned that if gene looping was responsible for the recruitment of termination factors at the promoter region of intron-containing genes, then loss of looping upon mutation of either 5′ or 3′ splice site will adversely affect the promoter occupancy of termination factors. ChIP analysis revealed that the promoter occupancy of CF1 subunit Rna15 and CPF subunit Pta1 was indeed reduced in the splice site mutants of IMD4. The promoter Rna15 signal was reduced by about 9–12.5 fold, while that of Pta1 declined by 5–10 fold in the splice site mutants (Fig 6B). The correlative nature of the gene looping and the promoter occupancy of termination factors support the idea that the looped gene architecture could be playing a crucial role in the loading of termination factors to the 5′ end of yeast genes.
To further explore the role of gene looping in the promoter recruitment of termination factors, we examined the promoter occupancy of termination factors in the looping-defective sua7-1 strain. The sua7-1 is an allele of the general transcription factor TFIIB with the glutamic acid at position 62 replaced with lysine (E62K) [38]. We have previously used this strain to show the role of gene looping in the intron-mediated enhancement of transcription [10]. This strain has also been used to demonstrate the role of gene looping in transcription memory and termination of transcription [34, 39–41]. We expected that if gene looping was responsible for the recruitment of termination factors at the 5′ end of genes, then the crosslinking of termination complexes in the promoter-proximal region will be compromised in the looping defective strain. We found that the promoter occupancy of all three termination complexes, CF1, CPF and Rat1, registered a decline in the looping defective mutant. The promoter recruitment of CF1 subunit Rna15, CPF subunit Pta1 and Rat1 complex subunit Rat1 decreased by about 1.5–3.5 fold in the looping defective sua7-1 strain (Fig 7B and 7C). To rule out the possibility of loss of promoter recruitment of termination factors in the looping defective strain being an indirect effect of defective splicing in the sua7-1 strain, we examined splicing of IMD4 and ASC1 pre-mRNA. The looping defect did not affect the splicing efficiency of either IMD4 or ASC1 transcripts (S4B Fig). If it was the looping-mediated recruitment of termination factors that conferred promoter directionality, then we expected that the transcription directionality will be adversely affected in the looping-defective sua7-1 cells. Our results show that the transcription directionality is indeed compromised in the sua7-1 strain (S5B Fig). These findings strongly support the idea that gene looping determines promoter directionality by facilitating the recruitment of termination factors to the 5′ end of genes.
Our published results suggest that it is either the transcription activator or the presence of an intron that facilitates transcription-dependent gene looping [10, 42]. The non-intronic genes are dependent on gene specific transcription activators for gene looping [42]. The intron-containing genes, however, require a splicing-competent intron to assume a looped gene conformation during transcription [10]. The promoter directionality of RNAPII-transcribed genes in budding yeast has been shown to be dependent on gene looping [17, 18]. Employing looping defective mutants, it was demonstrated that there is an increase in synthesis of uaRNA at the expense of mRNA in the absence of gene looping. These experiments, however, were performed with genes that exhibited activator-dependent gene looping. Here we show that the intron-dependent gene looping, which is characterized by additional interaction of the promoter and terminator regions with the intron, also confers directionality to the promoter-bound polymerase. How gene looping conferred transcription directionality was, however, not clear from any of the previous studies [17, 18]. On the basis of results presented here, we suggest a possible molecular mechanism underlying the enhancement of transcription directionality by the looped gene architecture. We propose that the proximity of the promoter and terminator regions in the gene loop allows the terminated polymerase along with the termination factors to be released from the 3′ end in the vicinity of the promoter of the gene. This leads to an increase in the local concentration of the termination factors near the 5′ end of a gene. These termination factors can now be recruited by the polymerase engaged in upstream anti-sense transcription leading to termination of uaRNA synthesis. Our hypothesis is supported by multiple experimental analyses. First, we observed enhanced crosslinking of the components of the CF1, CPF and Rat1 termination complexes near the 5′ end of several genes in the presence of an intron when the gene is in looped conformation (Fig 3B and 3C). Second, the recruitment of termination factors in the promoter-proximal region was compromised in the looping defective sua7-1 strain (Fig 7B and 7C). This mutant effects gene looping and transcription directionality without any adverse effect on splicing (S4 and S5 Figs). Third, the promoter occupancy of the termination factors exhibited a declining trend in the absence of an intron and in the presence of a mutated introns (Fig 6B). The mutation of splice sites in the intron selectively abolishes looping of the gene under investigation without any adverse effect on global gene looping [10]. It corroborates the finding with the looping defective sua7-1 mutant, which affects gene looping on a genomewide scale and therefore can potentially have an indirect effect on promoter directionality. The overall conclusion of these results is that it is looped gene architecture that facilitates the recruitment of termination factors near the 5′ end of a gene, and the termination factors then terminate the transcription of uaRNA thereby conferring promoter directionality.
In mammalian cells, asymmetric distribution of poly(A) sites and U1-binding sites has been shown to influence the recruitment of termination factors in the promoter-upstream region, which in turn terminates uaRNA synthesis [19, 21]. The transcription directionality of the mammalian β-globin gene, however, was found to be compromised in a looping defective mutant of the gene [17]. This invokes the possibility of gene looping playing a similar role in the recruitment of termination factors in the promoter-proximal region of at least a subset of mammalian genes during transcription.
If gene looping enhances transcription directionality by facilitating the recruitment of termination factors in the promoter-proximal region, the next logical question is why the promoter-recruited termination factors selectively terminate uaRNA synthesis, while mRNA transcription continues unabated. We have already shown that the activator-dependent gene looping facilitates recycling of polymerase from the terminator to the promoter for transcription in sense direction [18]. The intron-dependent gene looping may have a similar effect. A logical question is why the promoter recruited polymerase in a looped gene tends to preferentially transcribe in the sense direction, while it is terminated in the upstream anti-sense direction. We hypothesize that the differential effect of termination factors on the promoter-initiated divergent transcription could be due to differential chromatin structure in the vicinity of the promoter region. It has been shown that the histone modification pattern in the regions upstream and downstream of the bidirectional promoter is markedly different. In mammalian cells, the promoter downstream region is characterized by H3K79 dimethylation [22], which is the mark of elongating polymerase. In contrast, the promoter upstream region is deficient in H3K79 dimethylation [22]. Furthermore, H3K4 is trimethylated in the promoter downstream sense direction, while the upstream antisense region is marked by H3K4 monomethylation [27]. A similar differential modification of H3K27 was recently reported around the bidirectional promoter region in a murine cell line [27]. The H3K27 was found preferentially acetylated in the promoter upstream region near the antisense transcription start site in a subset of bidirectional promoters in murine macrophages. These differential chromatin marks around the bidirectional promoter region may inhibit elongation of uaRNA transcript and facilitate their termination by the termination factors.
The emerging view is that a vast majority of RNAPII-transcribed genes in yeast and mammalian systems have bidirectional promoters. The regulation of promoter directionality is critical for optimal transcription of these gene. We show a novel role of introns in yeast in regulating promoter directionality through looping-mediated recruitment of termination factors at the promoter. Less than 5% of yeast genes contain introns. In contrast, a vast majority of genes in higher eukaryotes contain introns. It is, therefore, tempting to speculate that introns might have a similar mechanistic impact on transcription directionality in higher eukaryotes as well.
Yeast strains used in this study are listed in S1 Table.
Cultures were started by inoculating 5 ml of YP-dextrose medium with colonies from a freshly streaked plate and grown at 30°C with gentle shaking. Next morning, overnight grown cultures were diluted (1:100) to appropriate volume and grown to A600 ~0.6. Equal number of cells were used for strand-specific RT-PCR, CCC, ChIP or strand-specific TRO assays. The rna15-2 and pta1-td mutants were grown at permissive temperature (25°C) till A600 reached 0.5. Cells were then shifted to non-permissive temperature (37°C) for 90 minutes and processed for strand-specific TRO analysis.
CCC experiments were performed as described previously [37]. The primers used for 3C analysis are shown in S2 Table. The enzyme used for chromatin digestion of IMD4 gene were Alu1 and Dra1, were obtained from New England Biolabs. Each experiment was performed with at least four independently grown cultures. The P1T1 PCR signals were normalized with respect to F1-R1 PCR signals.
ChIP experiments (crosslinking, cell lysis and isolation of chromatin) were performed as described previously [42]. Different strains were constructed by tagging subunits of all four termination complexes (CPF, CF1, Rat1 and Nrd1) as mentioned in S1 Table. Anti-HA antibodies were used to pull down HA-tagged subunits, were obtained from Thermo-scientific. Anti-Myc antibodies were used to pull down Myc-tagged subunits were obtained from Upstate Biotechnology, and IgG-Sepharose beads purchased from GE Healthcare were used to pull down TAP-tagged subunits. For ChIP analysis, primers used for ChIP PCR are shown in S2 Table. Each experiment was repeated with at least four independently grown culture.
Transcription analysis was performed by RT-PCR approach as described previously [35]. The primers used for RT-PCR analysis are shown in S2 Table.
The strand-specific ‘Transcription Run-On’ (TRO) assay was performed as described previously in [34]. The primers used for making cDNA and PCR for all the genes are mentioned in S2 Table.
The data shown in Figures is the result of at least four biological replicates. The quantification and statistical analysis was performed as described in [37]. Error bars represent one unit of standard deviation. P-values were calculated by two-tailed student t-test.
|
10.1371/journal.pgen.1000381 | Sporadic Infantile Epileptic Encephalopathy Caused by Mutations in PCDH19 Resembles Dravet Syndrome but Mainly Affects Females | Dravet syndrome (DS) is a genetically determined epileptic encephalopathy mainly caused by de novo mutations in the SCN1A gene. Since 2003, we have performed molecular analyses in a large series of patients with DS, 27% of whom were negative for mutations or rearrangements in SCN1A. In order to identify new genes responsible for the disorder in the SCN1A-negative patients, 41 probands were screened for micro-rearrangements with Illumina high-density SNP microarrays. A hemizygous deletion on chromosome Xq22.1, encompassing the PCDH19 gene, was found in one male patient. To confirm that PCDH19 is responsible for a Dravet-like syndrome, we sequenced its coding region in 73 additional SCN1A-negative patients. Nine different point mutations (four missense and five truncating mutations) were identified in 11 unrelated female patients. In addition, we demonstrated that the fibroblasts of our male patient were mosaic for the PCDH19 deletion. Patients with PCDH19 and SCN1A mutations had very similar clinical features including the association of early febrile and afebrile seizures, seizures occurring in clusters, developmental and language delays, behavioural disturbances, and cognitive regression. There were, however, slight but constant differences in the evolution of the patients, including fewer polymorphic seizures (in particular rare myoclonic jerks and atypical absences) in those with PCDH19 mutations. These results suggest that PCDH19 plays a major role in epileptic encephalopathies, with a clinical spectrum overlapping that of DS. This disorder mainly affects females. The identification of an affected mosaic male strongly supports the hypothesis that cellular interference is the pathogenic mechanism.
| Severe epilepsies associated with cognitive impairment in children are multifarious and most affected patients are sporadic cases. Thus, there is a challenge to identify which of these epilepsies are genetically determined, since their sporadic status excludes the use of classical genetic approaches. We have used microarrays, which are new technological tools to investigate the whole genome of an individual, to search for small genomic abnormalities and identify novel genes in 41 patients with a clinically well-characterized severe infantile epileptic disorder called Dravet syndrome. We have identified PCDH19, a new gene on chromosome X, which was recently found in a familial epileptic syndrome known as female-limited epilepsy and cognitive impairment. This gene was mutated in 12 out of 74 patients with clinical features compatible with Dravet syndrome. Eleven of these patients were females. The single male with a PCDH19 deficiency was mosaic in his skin; i.e., some of his cells express PCDH19 and others do not. This finding suggests that a new pathogenic mechanism—cellular interference—is associated with an unusual X-linked mode of inheritance in which females are more frequently affected than males.
| Epileptic encephalopathies are a group of rare disorders in which impairment of cognitive, behavioural and other brain functions is caused by the same underlying disease process. This heterogeneous group of disorders has multiple aetiologies such as symptomatic brain lesions, metabolic causes and diverse genetic syndromes. Much progress has been made in the past few years in the identification of genes responsible for genetic infantile epileptic encephalopathies. Among the genetic syndromes that have been characterized are: Dravet syndrome (DS), also called severe myoclonic epilepsy of infancy (SMEI, MIM# 607208) [1], CDKL5/STK9 Rett-like epileptic encephalopathy [2],[3], ARX-related epileptic encephalopathies [4], SRPX2-related rolandic epilepsy associated with oral and speech dyspraxia and mental retardation [5], and very recently, female-limited epilepsy and cognitive impairment (EFMR) associated with mutations in PCDH19, the gene encoding the protocadherin 19 on the X chromosome [6].
Dravet syndrome is characterized by the occurrence of generalized or unilateral clonic or tonic–clonic seizures, usually triggered by fever, in the first year of life of a previously normal infant. Later on, other types of seizures occur, including myoclonus, atypical absences and partial seizures [7]. Development is progressively delayed starting from the second year. Susceptibility to febrile seizures persists over time, and status epilepticus is frequent. Epilepsy generally persists despite appropriate anti-epileptic therapy (polytherapy including sodium valproate, clobazam or topiramate and stiripentol). Children with DS typically have poorly developed language and motor skills, learning disabilities and variable degrees of mental retardation [8]. They are usually sporadic cases; however, sib pairs with SMEI, or patients with a family history of epilepsy, have occasionally been reported [9].
Heterozygous de novo mutations in SCN1A, the gene encoding the voltage-gated neuronal sodium channel alpha 1 subunit (Nav1.1), are a major cause of DS [1]. All types of mutations [10] and rearrangements [11]–[15] in SCN1A have been observed in SMEI patients. However, no point mutations or rearrangements have been found in a fraction of patients, now estimated to 20–25% [15]–[18], strongly suggesting that DS is a genetically heterogeneous disorder.
Our aim was to identify the gene(s) involved in SCN1A-negative patients with Dravet syndrome. Most of our patients were isolated, excluding the use of classical genetic approaches. Our hypothesis was that genomic micro-rearrangements, which are increasingly identified as causes of human genetic disorders, might be found in a subset of the SCN1A-negative patients with DS, thus identifying new causal genes.
In this study, we have searched for genomic rearrangements in 41 SCN1A-negative patients using high-density SNP microarrays (Illumina, 370K). Genes located in the rearrangements were then considered to be candidate genes and were analysed for point mutations by direct sequencing in the remaining negative patients with DS.
An initial series of 41 probands (18 females and 23 males), referred for genetic analysis of Dravet syndrome but negative for point mutation and intragenic rearrangement of SCN1A [15], was screened for genomic rearrangement using Illumina 370CNV microarrays. A hemizygous deletion on chromosome Xq22.1 was identified in a male patient (patient 1 from family 1). This deletion spanned approximately 1 Mb and encompassed a single gene, PCDH19 (Figure 1A). A duplication of the same region was previously reported in one of 776 healthy controls (506 unrelated healthy individuals from Northern Germany and 270 HapMap subjects) [19], but no deletions in healthy individuals have been recorded in the database of genomic variants. Patient 1 and his mother were then analyzed with high-resolution CGH arrays (Nimblegen). This analysis confirmed that the deletion spans 890 Kb, between genomic positions g.98731380 and g.99618794 on chromosome X, and showed that it has occurred de novo since it was not found in the mother of the patient (Figure 1B).
PCDH19 encodes protocadherin 19, a transmembrane protein of the cadherin family of calcium-dependent cell–cell adhesion molecules, which is strongly expressed in the central nervous system. In the postnatal brain, protocadherins might be involved in the modulation of synaptic transmission and the generation of specific synaptic connections [20]. PCDH19 was therefore an attractive candidate gene for epilepsies and mental retardation. To test whether a PCDH19 deficiency might be implicated in some epileptic encephalopathies resembling Dravet syndrome, we sequenced the coding region of this gene in 73 SCN1A-negative probands (the remaining 40 patients of the initial series plus 33 additional patients, for a total of 45 females and 28 males).
Ten different variants were identified in 11 unrelated female probands at the heterozygous state (Figure 2A). All but one were located in exon 1: three were nonsense mutations (c.142G>T/p.Glu48X, c.352G>T/p.Glu118X, c.859G>T/p.Glu287X), two were small deletions and insertions creating a frameshift (c.506delC/p.Thr169SerfsX43 and c.1036_1040dup/p.Asn347LysfsX23) and the remaining five were missense mutations (c.361G>A/p.Asp121Asn, c.595 G>C/p.Glu199Gln, c.1019A>G/p.Asn340Ser, c.1628 T>C/p.Leu543Pro and c.3319 C>G/p.Arg1107Gly). Glu48X was present in two affected sisters of family 2; Glu118X was identified in an isolated patient (family 3) and Glu287X was found independently in a patient with family history of epilepsy and mental retardation (family 4) and in an isolated patient (family 5). Interestingly, the c.3319 C>G/p.Arg1107Gly missense variant, located in exon 6, was associated with the p.Glu287X mutation in the proband of family 5. In family 6, cytosine 506 (c.506delC) was deleted in a patient whose parents were unaffected, but whose female cousin also had epilepsy and moderate mental retardation. The 5-bp duplication (c.1036_1040dup) was present in the index case of family 7. The p.Asp121Asn mutation was identified in the index case of family 8, who had a sister with epilepsy and psychotic disturbances. Finally, p.Glu199Gln, p.Asn340Ser and p.Leu543Pro variants were identified in the 4 remaining isolated patients (families 9 to 12); Asn340Ser was found in two independent patients (families 10 and 11). These 4 missense variants (p.Asp121Asn, p.Glu199Gln, p.Asn340Ser and p.Leu543Pro) all affected amino-acids in the extracellular domain of protocadherin 19, which are highly conserved in orthologs and in paralogs of PCDH19 in the delta protocadherin family (Figure 2B). Interestingly, p.Arg1107Gly, associated with the de novo Glu287X mutation in the proband of family 5, affected a residue of the protein that is conserved in mammalian orthologs, but not in other species or in paralogs (Figure 2B). To confirm that the variants are pathogenic, we screened 180 healthy Caucasians. Only Arg1107Gly was found in a healthy female individual and was thus considered to be a rare polymorphism. None of the other variants was found in the control population, confirming that they are causal mutations.
The parents and relatives of PCDH19-positive patients were also analysed when possible (Figure 3). The p.Glu48X mutation, found in two affected sibs in family 2, was inherited from their asymptomatic father. Likewise, the c.1036_1040dup5, p.Asp121Asn and p.Leu543Pro mutations were inherited from the healthy fathers of the index cases in families 6, 8 and 12. In family 6, the 5-bp duplication was inherited from the paternal grandmother who also had epilepsy and cognitive impairment, and transmitted to the half-brother of the father and his affected daughter (i.e. the index case's cousin, Figure 3). In family 4, the mother of the proband had mental retardation associated with adult-onset epilepsy, a clinical feature also present in the maternal grandmother and maternal aunt; the proband's father also presented with moderate mental retardation but without epilepsy. The Glu287X mutation in this family was also inherited from the father. In contrast, in families 5, 7, 10 and 11, the mutations (p.Glu287X, c.506delC and p.Asn340Ser, respectively) occurred de novo in the index cases, since they were not found in either parent. Interestingly, Arg1107Gly was inherited from the asymptomatic father in family 5. In family 4, only the mother and sisters of the index case were available for genetic analyses. Both sisters, who were monozygous twins, had mild psychomotor and cognitive impairment but never had seizures. Neither the mother nor the sisters had the p.Glu287X mutation. Analysis of the haplotypes in Xq22.1 (PCDH19 locus) with microsatellite markers confirmed that the three sisters (the two twins and the affected proband) received the same X chromosome from their father, with and without the p.Glu287X mutation, which indicates that the mutation also occurred de novo in this family. Finally, in family 9, the mother did not have the p.Glu199Gln mutation but the father remained unavailable for genetic analyses.
Recently, PCDH19 mutations were shown to cause epilepsy and mental retardation limited to females (EFMR), a familial disorder associating childhood-onset epilepsy and a variable degree of cognitive impairment with an unusual mode of inheritance: this X-linked disorder is found in females with heterozygous mutations but not in males with hemizygous mutations [6]. How, then, can we explain the affected male in our series with a deletion of the entire PCDH19? Random X-inactivation in mutated females normally leads to tissue mosaicism in which two cell populations, one expressing normal PCDH19 and the other expressing the mutated allele, co-exist. To explain why only females are affected, it might be hypothesized that the co-existence of PCDH19-positive and PCDH19-negative cells would be pathogenic whereas homogeneous cell populations (PCDH19-positive in normal individuals but PCDH19-negative in mutated males) would not [6]. A mechanism of this type was previously termed “cellular interference” [21]. Two cell populations would also be found in mosaic males, who, according to this hypothesis, would be affected like mutated females.
To test whether our male patient was mosaic for the PCDH19 gene deletion, we compared peripheral blood lymphocytes (PBL) and cultured fibroblasts from the patient by FISH with a probe specific to the PCDH19 genomic region. Although no signal corresponding to PCDH19 was detectable in PBL, a normal PCDH19 allele was found in 53% of the fibroblasts (Figure 4), confirming that the patient was mosaic, in his skin, for the PCDH19 deletion. This result confirms that mutations in PCDH19 can be responsible, in mosaic males, for epileptic encephalopathy phenotypes that are usually limited to females, and strongly supports the hypothesis that cellular interference is the main pathogenic mechanism of the disease.
The clinical features of the male patient with the PCDH19 deletion and the female patients with PCDH19 point mutations are summarized in Table 1. These patients fulfil the main criteria for DS (see material and methods' section), with a mean age of seizures onset of 9.5 months (ranging from 7.5 to 12 months). Nevertheless, contrary to SCN1A-positive patients (SCN1A-DS), myoclonic jerks, atypical absences, and photosensitivity were unfrequent in PCDH19-positive patients (PCDH19-DS) (3, 3 and 1 patients out of 13, respectively). Only 6 patients presented status epilepticus. The mental delay was mild in 6 patients, moderate in 4 and only 3 patients presented with severe delay. Although much delayed, the language was present in all patients, with 12 out of 13 able to formulate short sentences.
In this study, we used SNP microarrays to search for microrearrangements in patients with clinical features suggestive of Dravet syndrome but without mutations in SCN1A in order to identify new causative genes. The identification of a de novo hemizygous deletion of PCDH19, encoding protocadherin 19, in a male patient led us to screen the coding region of this gene in the remaining patients. Eleven unrelated probands with point mutations in PCDH19, all females, were found. While this study was ongoing, PCDH19 was reported to be the causative gene for female-limited epilepsy and cognitive impairment (EFMR), a disorder characterized by seizure onset in infancy or early childhood and cognitive impairment, which is found only in females in multi-generational families [6]. Since all of our patients with point mutations in PCDH19 were females as previously reported, we investigated the possibility that the male patient in whom the gene was deleted might be mosaic for the deletion. FISH analysis confirmed this latter hypothesis.
The thirteen patients with PCDH19 mutation or deletion (12 probands and one sib, family 2) all fulfilled the main criteria for DS and were all negative for mutation or rearrangement in SCN1A after direct sequencing and multiplex ligation-dependent probe amplification (MLPA) [15]. The proportion of PCDH19-DS probands in our series of SCN1A-negative patients was 16% (12/74), or even 25% (11/45) if only female patients were included in the calculation. Considering that approximately 25% of all patients with DS are SCN1A-negative [15], PCDH19 might overall account for 5% of DS patients. PCDH19-DS patients and SCN1A-DS patients have many features in common including: normal psychomotor development before seizures onset, early onset of seizures (before age one year), association of febrile and afebrile seizures, with a high susceptibility of the seizures to fever for all 13 patients, occurrence of hemiclonic or unilateral seizures (11/13), and association of generalized tonic-clonic and focal seizures (12/13), a high proportion of seizures occurring in clusters (12/13), prolonged seizures, a proportion of which lead to status epilepticus, secondary progressive appearance of mental and motor regression and language delay, accompanied, in some cases, with ataxia (Table 1).
However, PCDH19-DS patients slightly differ on average from the classical pattern reported in SCN1A-DS. PCDH19-DS patients were slightly older at onset than SCN1A-DS patients (9.5 months, with a range from 7.5 to 12 months, versus 6.3 months, calculated from our series of SCN1A-positive DS patients, p<0.0001) [15]. Less than half (6/13) of the PCDH19-DS patients had status epilepticus although this is a highly frequent feature in SCN1A-DS (93/113, p<0.007). Photosensitivity, frequently reported in SCN1A-DS, was exceptional in PCDH19-DS and was reported in only one patient but the difference with SCN1A-DS in our series of patients was however not significant. Seizures were, on average, less intractable than in SCN1A-DS, and patients above six years of age (9/12 patients) had less than 4 seizures a year with one patient who was free of seizures at the time of the study. Although all patients were on tri- or poly-therapy, seizures were relatively well-controlled, a situation rarely achieved in SCN1A-DS. Intellectual and language delay were constant but were less severe than the classical outcome of SCN1A-DS [8] (mostly with important speech and mental delay) although the difference was not significant. Finally, myoclonic jerks and atypical absences were present in only 2 and 3 patients, respectively, whereas they are frequent features in SCN1A-DS (myoclonic jerks: 55/110, p<0.018; atypical absences: 92/108, p<0.0001). Patients with SMEI but without myoclonia have been previously referred as SMEB (borderline severe myoclonic epilepsy in infancy) [22], but SMEI and SMEB are currently grouped together under the term DS. In addition, the same types of mutation, and even the same mutations, are found in patients with DS and patients with other infantile epileptic encephalopathies (such as cryptogenic generalized or focal epilepsies), which has extended the clinical spectrum of SCN1A and the definition of DS [15]. Therefore, in individuals, these divergent clinical characteristics are not sufficient to distinguish between patients with SCN1A or PCDH19 mutations, and the two clinical spectrums largely overlap. They can be useful, however, to prioritize molecular diagnosis although they must be first confirmed on larger series.
Mutations in PCDH19 were recently reported to cause EFMR, which also associates mental retardation and epilepsy exclusively in females. EFMR was differentiated from DS by the authors on both clinical and genetic grounds [6],[23]. The clinical features of EFMR, unlike those of DS, are highly variable, even in members of the same family: onset of seizures is between 6 and 36 months, the patients present with a combination of febrile and afebrile seizures of various types and a variable degree of psychomotor delay and cognitive impairment, ranging from mild to severe mental retardation [23]. Dibbens et al. reported PCDH19 mutations in six large families and one small family with two affected sib pairs [6]. All the patients were familial cases that were, for the most part, already adults at the time of examination, and appeared socially integrated in that most of them were married and had children. In the present study, on the contrary, the patients were essentially young, had a severe epileptic encephalopathy, and 8 of the 12 were isolated cases. In 6 patients out of 11 in whom inheritance could be assessed, the mutation occurred de novo. In the 5 remaining patients, the mutation was inherited from fathers who were healthy, had no cognitive impairment, and never had febrile seizures or epilepsy (families 2, 6, 8 and 12), or had mild mental retardation but no epilepsy (family 4). The global clinical pictures of PCDH19-DS and EFMR appear therefore to differ. The difference in the phenotypes might be due to the modes of recruitment (familial versus sporadic cases). It might also be hypothesized, that patients with PCDH19-DS have a better final outcome than in those with SCN1A-DS despite the severity of their disease in childhood and that the two disorders are different clinical expressions of the same disease. Both hypotheses are not mutually exclusive. Interestingly, the variability in the severity of epilepsy and cognitive impairment in EFMR is reminiscent of what is observed in GEFS+ families (generalized epilepsy with febrile seizures plus, # MIM# 604233), an autosomal dominant condition that also associates febrile seizures with epilepsy of variable types and severity, and which is associated in ∼10–15% of the families with missense mutations in SCN1A [10]. Although patients with GEFS+ are usually responsive to treatment and generally have a benign outcome, some family members may be more severely affected, and even present with DS. The clinical spectrum of PCDH19 mutations could be as broad as the spectrum of GEFS+. Random X inactivation could contribute to this variability by generating variable proportions of mutated to normal cells in the brains of the mutated females.
Although the mutations in EFMR families and in PCDH19-DS patients are distinct, the spectra of mutations are comparable, and include nonsense mutations, small deletions/insertions introducing a frameshift as well as missense mutations affecting highly-conserved amino-acids in the protein (Figure 5), which would probably cause loss-of-function of the mutated allele. Messenger RNAs with mutations introducing premature termination codons (PTC) have indeed been shown to be degraded via the nonsense-mediated mRNA decay (NMD) surveillance system of the cell in fibroblasts from EFMR patients [6]. The identification of a whole gene deletion in the mosaic male patient with PCDH19-DS also supports the loss-of-function as the main consequence of the mutations. However, all the point mutations identified so far are clustered in the large exon 1 of the gene corresponding to the extra-cellular cadherin domain of the protocadherin 19 protein, as previously reported by Dibbens and collaborators [6]. Further studies are needed to determine whether PTC mutations can be found in other exons; this would be expected if the loss-of-function assumption is correct.
EFMR and PCDH19-DS are paradoxical X-linked disorders in which mutated females are severely affected whereas males carrying the mutation are phenotypically unaffected: they have normal cognitive function and no seizures although a subtle psychiatric carrier status was evoked [6],[23]. All affected patients with point mutations identified in this study were also females. In families 2, 4, 6, 8 and 12, the mutation was inherited from the father. Five males (families 2, 6 and 8) were asymptomatic carriers of PCDH19 mutations, they were healthy, had no cognitive impairment or epilepsy, and none had histories of febrile seizures. In family 4, however, the father who transmitted the mutation to his daughter had moderate mental retardation but no epilepsy. The link between the mutation and his cognitive impairment remains, however, uncertain. The only definitely affected male was, therefore, the patient who was mosaic for the PCDH19 deletion. There was no molecular evidence of mosaicism in the blood of the father in family 4.
Several mechanisms have been suggested to account for the unusual mode of inheritance observed in EFMR. 1) A dominant negative effect of the mutant protein in females (as for mutations in STK9/CDKL5 and MECP2) is unlikely, since it is usually associated with lethality in males. 2) Compensatory factors may exist in males; in particular, a protocadherin gene on the Y chromosome (PCDH11Y) is specifically expressed in males and could play a role in a sex-dependent compensation; a paralogous gene is located on the X chromosome (PCDH11X), but the proteins encoded by the two genes are not identical [6]. In addition, the protocadherin family contains more than 80 genes scattered throughout the human genome [24], supporting the hypothesis of molecular compensation. 3) Another explanation for the unusual mode of inheritance associated with PCDH19 mutations is cellular interference, a mechanism reminiscent of metabolic interference [6],[21],[25]. It postulates that random inactivation of one X chromosome in mutated females generates tissue mosaicism (i.e. co-existence of PCDH19-positive or PCDH19-negative cells), which would be pathogenic by altering cell-cell interactions; normal individuals and mutated males, who are homogeneous for PCDH19-positive or PCDH19-negative cells respectively, would not develop the disease (Figure 6). The identification of an affected male who was mosaic for the PCDH19 deletion in his fibroblasts, and therefore had PCDH19-positive and PCDH19-negative cells in this tissue, strongly supports the hypothesis of cellular interference as the main pathogenic mechanism associated with PCDH19 mutations. The co-existence of normal and mutated cells and the proportion of each population in the brain of this patient cannot, however, be extrapolated from fibroblasts or lymphocytes. To definitely establish that cellular interference is the pathogenic mechanism, it is necessary i) to demonstrate that neuronal cells are mosaic, but also that ii) females who are homozygous for PCDH19 mutations or deletions are also unaffected, like hemizygous males. Although pathogenesis in cells that express the mutated allele after inactivation corresponds to a loss-of-function, cellular interference would result in a gain-of-function at the tissue level, because of abnormal interactions between mutated and normal cells. This hypothesis supposes that the loss of protocadherin 19 is compensated for, but by a mechanism that is relatively independent of gender.
The same X-linked pattern of inheritance and has been observed for craniofrontonasal syndrome (CFNS), a disorder in which females have multiple skeletal malformations. The gene responsible for CFNS is EFNB1, located in Xq12 and encoding Ephrin B1, a transmembrane protein that is a ligand for Eph receptors [21]. The Ephrin B1/Eph interaction plays a role in cell migration and pattern formation during developmental morphogenesis [26]. Cellular interference, also proposed as the pathogenic mechanism for CFNS [21], had previously been demonstrated in female mice heterozygous for Ephrin B [27]. Although homozygous female and hemizygous male mice showed comparable perinatal lethality due to major skeletal abnormalities, heterozygous females were even more affected, and they alone had polydactyly. Ephrin B1-EphB receptor signaling was shown to regulate skeletal development by controlling cell movement. Mosaic expression of Ephrin B1, caused by random X inactivation in heterozygous females, results in ectopic interactions between the Ephrin B1 ligand and EphB receptors, sufficient to induce the skeletal defects [27].
Protocadherin 19 is an 1148 amino-acids transmembrane protein belonging to the protocadherin delta2 subclass of the cadherin superfamily, which is highly expressed in neural tissues and at different developmental stages [6],[28],[29]. The precise functions of the protein remain so far unknown. However, Delta protocadherins were reported to mediate cell-cell adhesion in vitro and cell sorting in vivo, and could regulate the establishment of neuronal connections during brain development [24],[30]. Ephrin B1 and protocadherin 19 could therefore share major characteristics. Several isoforms of protocadherin 19 have been reported to result from alternative splicing of exon 2 and the existence of two acceptor sites for intron 4 which adds a residue at the beginning of exon 5. The isoform(s) implicated in the physiopathology of EFMR and PCDH19-DS are still not known. Functional studies as well as the development of mouse models are now needed to confirm and unravel the molecular mechanisms of cellular interference in these diseases.
In conclusion, these results extend the clinical spectrum associated with PCDH19 mutations: we demonstrated that mutations in this gene are not limited to familial female patients, but can also account for isolated cases. Our results suggest that isolated mosaic male patients are also susceptible to the disease. Finally, mutations in PCDH19 can cause an early and severe epileptic encephalopathy mimicking DS, a major problem for differential diagnosis. The high frequency of patients with PCDH19 found in this study justifies the molecular testing of this gene in SCN1A-negative patients, especially females, diagnosed as having Dravet syndrome. This study also validates the use of SNP microarrays to identify novel genes in isolated patients with severe genetic pathologies. This strategy will hopefully identify new genomic regions or genes that would account for the ∼15–20% of DS patients that do not have SCN1A and PCDH19 mutations.
A total of in 74 probands (45 females and 29 males), referred by specialized neuropediatric centres as having Dravet syndrome but who were negative for point mutations or rearrangements in SCN1A, were included in the study [15]. Forty-one of these patients were initially selected for the microarray analysis and 33 were later on included for sequencing of PCDH19. The referring physicians filled out detailed clinical questionnaires for every patient. Clinical histories were also obtained when possible to assess the evolution of the disease. All clinical reports and questionnaires were re-examined by the same neuropediatrician (RN). Intellectual assessment was based on psychological evaluation when available. Psychomotor skills and cognitive delay were clinically evaluated in all patients. The clinical diagnosis of DS included: normal cognitive and motor development prior to seizures onset, onset of the seizures before the age of one year, seizures mainly triggered by fever, long-lasting seizures (>15 min, that might evolve to status epilepticus), later occurrence of other types of seizures (febrile and afebrile) and cognitive regression. The presence of myoclonic jerks and/or ataxia was considered to be a highly characteristic, although inconstant, feature of the disease that could reinforce a diagnosis; however, their absence did not exclude the clinical diagnosis of Dravet syndrome, since they were not previously observed in all patients with DS [7],[31]. Informed written consent was obtained from the patients' parents before blood sampling. This study was approved by the ethical committee (CCPPRB of Pitié-Salpêtrière Hospital, Paris, n°69-03, 25/9/2003).
Patients were screened using Illumina 370CNV-Duo genotyping BeadChip arrays (370 K). The Infinium II Genotyping reaction steps were performed according to the manufacturer's specifications (Illumina, San Diego, CA) on the P3S platform (Pitié-Salpêtrière Hospital). Briefly, 750 ng of genomic DNA were isothermally amplified at 37°C overnight. The amplified products were fragmented by a controlled enzymatic process then precipitated with isopropanol. The dried precipitated pellet was resuspended, hybridized to 370CNV-Duo beadchips in a capillary flow-through chamber and incubated overnight at 48°C. The amplified, fragmented DNA samples anneal to locus-specific 50-mers during the hybridization step. Each bead type corresponds to one allele per SNP locus. After hybridization, allelic specificity was conferred by enzymatic single-base extension and fluorescent staining. Arrays were washed and dried for 1 h before imaging using a BeadArray Reader (Illumina). Image data analysis and automated genotype calling was performed using Beadstudio 3.1 (Illumina). All genomic positions were based on the UCSC and Ensembl Genome Browsers. Each copy number variant (CNV) identified in patients was searched in the database of genomic variants (http://projects.tcag.ca/variation/), which repertories the structural variation in the Human genome, to determine whether this CNV is normally present in a control population.
Genomic DNA from the patients was analysed by microarray-based comparative genomic hybridization with the HG18 WG Tiling 385 K CGH array v2.0 (Roche NimbleGen, Madison, WI), according to the NimbleGen hybridization Kit Protocol. Briefly, DNA samples from patients and controls were labelled by random priming: the DNA (1 µg) was denatured in the presence of 5′Cy3- or Cy5-labeled random nanomers (Trilink Biotehcnologies, San Diego, CA) and incubated with 100 units of exo-klenow fragment (NEB, Beverly, MA) and dNTP mix [6 mM each in TE buffer (10 mMTris/1 mM EDTA, pH 7.4, Invitrogen)] for 2 h at 37°C. Reactions were terminated by addition of 0.5 mM EDTA (pH 8.0), precipitated with isopropanol and resuspended in water. The Cy-labelled test sample (Cy3) and the reference sample (Cy5) were combined in 13 µL of Nimblegen Hybridization solution (Roche Nimblegen). After denaturation, hybridization was carried out on a MAUI Hybridization System (BioMicro Systems, Salt Lake City, NE) for 18 h at 42°C. The array was washed with the NimbleGen Wash System (Roche NimbleGen), dried by centrifugation and scanned with the genePix 4000B scanner (Axon Instrument, Union City, CA). Fluorescence intensity (raw data) was obtained from the scanned images of the oligonucleotide tiling arrays with NIMBLESCAN 2.0 extraction software (Nimblegen Systems). For each spot on the array, log2 ratios of the Cy3-labeled test sample versus Cy5 reference sample were calculated. Regions were considered to be duplicated or deleted when result exceeded the +/−0.25.
Eleven specific primer pairs were designed to amplify the 6 exons and adjacent intron-exon boundaries (∼100 bp from each side of the exons) of the PCDH19 gene (transcript reference EF676096). Primer sequences are available on request. Forward and reverse sequence reactions were performed with the Big Dye Terminator Cycle Sequencing Ready Reaction Kit (PE Applied Biosystems) using the same primers. G50-purified sequence products were run on an ABI 3730 automated sequencer (PE Applied Biosystems) and data were analyzed with the Seqscape 2.5 software (Applied Biosystems). Mutations identified in the patients were looked for directly in the DNA of available parents by sequencing the corresponding amplicon. If neither parent had the mutation, the parents were tested with microsatellite markers at the Xq22.1 locus to ensure that the mutation occurred de novo. In addition, 180 European controls (90 males and 90 females) were included to test new variants in the PCDH19 gene.
FISH experiments were performed on peripheral blood lymphocytes (blood samples) and fibroblasts (skin biopsies). Fibroblasts were grown in Dulbecco's modified Eagle's medium containing 4.5 mg/ml glucose and 110 µg/ml pyruvate (DMEM) supplemented with 10% fetal calf serum (FCS), 0.03% glutamine, 1000 U/ml penicillin/streptomycin in a 5% CO2 atmosphere for 2 weeks before FISH. Lymphocytes were grown in PB-Max medium (Invitrogen) for 3 days. Metaphase chromosome spreads were obtained by standard hypotonic treatment and methanol/acetate (3/1) fixation. The slides were washed with the cytology FISH accessory kit (Dako). A FISH DNA probe, specific for the Xq22.1 region covering PCDH19, was labeled with rhodamine by nick-translation after amplification of the RP11-99E24 BAC (Invitrogen) and cohybridized with a commercial subtelomeric control probe (Cytocell), specific for the pseudo-autosomal region 1 (chromosomes X/Y) labeled with fluorescein isothiocyanate (FITC). The slides were then washed and counterstained with 4,6-diamino-2-phenylindole (DAPI) for chromosome identification. Metaphase cells were examined under a motorized reflected BX61 Olympus fluorescence microscope with filters for separate detection of DAPI, FITC and rhodamine. One hundred metaphase cells were counted to determine the degree of mosaicism in fibroblasts and lymphocytes. Metaphase chromosomes from a karyotypically normal female were used as a control.
Frequencies were compared with the Chi-Square test or the Fisher exact test when appropriate. Means were compared using Mann-Whitney Rank Sum Test. Statistical analysis was performed using SigmaStat 3.5 software.
|
10.1371/journal.pgen.1001360 | The Toll-Like Receptor Gene Family Is Integrated into Human DNA Damage and p53 Networks | In recent years the functions that the p53 tumor suppressor plays in human biology have been greatly extended beyond “guardian of the genome.” Our studies of promoter response element sequences targeted by the p53 master regulatory transcription factor suggest a general role for this DNA damage and stress-responsive regulator in the control of human Toll-like receptor (TLR) gene expression. The TLR gene family mediates innate immunity to a wide variety of pathogenic threats through recognition of conserved pathogen-associated molecular motifs. Using primary human immune cells, we have examined expression of the entire TLR gene family following exposure to anti-cancer agents that induce the p53 network. Expression of all TLR genes, TLR1 to TLR10, in blood lymphocytes and alveolar macrophages from healthy volunteers can be induced by DNA metabolic stressors. However, there is considerable inter-individual variability. Most of the TLR genes respond to p53 via canonical as well as noncanonical promoter binding sites. Importantly, the integration of the TLR gene family into the p53 network is unique to primates, a recurrent theme raised for other gene families in our previous studies. Furthermore, a polymorphism in a TLR8 response element provides the first human example of a p53 target sequence specifically responsible for endogenous gene induction. These findings—demonstrating that the human innate immune system, including downstream induction of cytokines, can be modulated by DNA metabolic stress—have many implications for health and disease, as well as for understanding the evolution of damage and p53 responsive networks.
| Among the most prominently studied regulators of gene function is the p53 tumor suppressor, which has many roles in human biology. The transcriptional master regulator p53 directly targets expression of >200 genes. Previously, we sought to define the p53 network in terms of functionality, specifically the ability of target response element sequences (REs) to support p53 transactivation. Here we identify p53 target canonical and noncanonical REs in the family of Toll-like Receptor (TLR) innate immune response genes and establish p53 regulation of most TLR genes. We address p53 responsiveness in primary human lymphocytes and alveolar macrophages collected from healthy volunteers. Notably, all TLR genes show responses to DNA damage, and most are p53-mediated. However, there is considerable variability between individuals, suggesting that DNA and p53 metabolic stresses can markedly differ in impact on the innate immune system as well as downstream appearance of cytokines. Indeed, we report a SNP in a p53 RE within the TLR8 promoter that alters p53 responsiveness in primary human cells. Furthermore, the p53-mediated expression of TLRs is unique to primates. Overall, these findings identify a new, pivotal role for the well-known human tumor suppressor p53, namely, integration of DNA damage and innate immune responses.
| The p53 master regulator is responsive to a variety of DNA metabolic stresses resulting in induction or repression of over 200 genes as well as several LINC- and micro-RNAs [1], [2]. In its role as tumor suppressor and “guardian of the genome” many of the target genes in humans influence cell cycle progression or apoptosis. Over the past decade the p53 network has been extended to transcriptional regulation of genes associated with a wide variety of biological functions including DNA repair, angiogenesis, cellular metabolism, autophagy, stem cell renewal, fertility, differentiation and cellular reprogramming. To better understand the broad role that p53 can play in human biology, we have pursued “functionality rules” for identifying target response element (REs) sequences where p53 can directly influence transactivation. Using in vivo transactivation systems based in yeast and human cells as well as binding in human cell extracts [3], [4], we recently found that functionality of the binding consensus RRRCWWGYYYnRRRCWWGYYY (R, pyrimidine; Y, pyrimidine; W, A or T; n, spacer of 0 to 13 bases) is greatest when there is at most a single base spacer and if “WW” is “AT.” Furthermore, we defined functionality for half-sites (RRRCWWGYYY) and found in cis synergy when another transcription factor, estrogen receptor, was bound nearby (for a description of noncanonical REs including half-sites see [3], [5] and summary in [6]). Based on these functionality rules, we found that the evolution of p53 control of at least one gene family, DNA metabolism and repair, is limited to primates [7]. Furthermore, we identified single nucleotide polymorphisms in REs that are predicted to modify responsiveness of genes to p53 mediated stress [8].
Our functionality studies of canonical and noncanonical promoter p53 REs suggested that p53 may have a role in human Toll-like receptor (TLR) expression. We had reported that a single nucleotide polymorphism (ChrX:12923681, rs3761624 A/G) in the promoter of the TLR8 gene creates an RE that can be targeted by p53 in yeast and cell line reporter systems [8] although we could not detect endogenous expression of the TLR8 gene. In addition, the TLR3 gene in epithelial cancer cell lines was found to be induced by p53 following exposure to 5-fluorouracil (5FU) [9].
Innate immunity is paramount to infection and tissue injury responses [10]. The TLR gene family mediates innate immunity to a wide variety of pathogenic threats. The TLRs recognize conserved exogenous pathogen-associated molecular patterns (PAMPs) and endogenous danger-associated molecular patterns (DAMPs) [11], while adaptive immune receptors ensure specific antigenic responses through clonal expansion. TLR function is associated with circulating immune cells such as monocytes and dendritic cells, tissue phagocytes that include alveolar macrophages and with nonimmune cells (such as epithelial cells of gut, skin, lung, etc.) that are exposed to environmental injury or pathogens. The roles of TLRs in mammalian biology are continuously expanding and they are now understood to function in such diverse processes as inflammation, cell differentiation and cell survival [11]. Altered TLR function is implicated in human diseases such as systemic lupus erythematosus, inflammatory bowel disease (IBD) and cancer [11], [12], and agonist/antagonist manipulations of the TLR system are being pursued to alleviate various diseases (reviewed in [13]).
Given the varied functions of TLRs, factors that regulate their expression are expected to shape immune responses [14]. However, there are few examples of TLR gene induction and these are limited to specific TLR-stimulus interactions [15], [16]. The TLR-dependent innate immune response is thus generally considered to be “hard-wired.” Although environmental stress can influence gene expression indirectly, e.g., through epigenetic mechanisms, to our knowledge there are no reports of TLR induction by environmental factors in primary human cells [15], [17]–[20].
These collected observations led us to investigate the responsiveness of TLR genes as a class to common DNA stressors. We have examined expression of the entire TLR gene family following exposure to anti-cancer agents that induce the p53 network in primary immune cells obtained directly from human subjects as well as the impact on downstream cytokines. Agents were applied to T-lymphocytes and alveolar macrophages ex vivo. Prior to this there were no reports that we are aware of addressing DNA damage-induced responses of any TLR genes in cells directly associated with innate immunity. As part of this study, we establish that in the evolution of the TLR responses, the p53-mediated expression of TLRs is unique to primates.
We selected ionizing radiation (IR), 5FU and Doxorubicin (Doxo) because these agents represent a cross-section of DNA metabolic stressors, such as damage or replication inhibition, that may occur endogenously or environmentally and are well-known to activate p53 and its network of target genes. Furthermore, these agents are often employed in cancer treatments. To address TLR responses ex vivo in primary human immune cells, T-lymphocytes were expanded by phytohemagglutinin (PHA) stimulation of peripheral blood mononuclear cells (PBMC) freshly isolated from the blood of healthy human volunteers (see Figure S1A; for demographics of volunteers see Table 1); alternatively, T-lymphocytes were obtained from the PBMC fraction using anti-CD3 antibody-based purification as described in Figure S1B.
Presented in Figure 1 are the expression responses of the entire family of TLR genes in stimulated lymphocytes for 18 subjects (except where noted) following exposure to IR (4 Gray), 5FU (300 µM), and Doxo (0.3 µg/ml) (Figure 1A, 1B, and 1C, respectively). These responses are relative to no treatment controls and normalized to 18S ribosomal DNA. As additional controls, we also examined the expression of the beta-glucuronidase, GUSB, and actin genes (Figure S2) that are considered to be nonresponsive to chromosomal and/or p53 stress; they showed little variation after doxorubicin and nutlin exposure. The doses chosen were similar to therapeutic doses or doses commonly used in the literature. By way of comparison, we also examined the response of the cyclin-dependent kinase inhibitor gene p21WAF1 (CDKN1A), a prototypical p53 target gene induced by these agents in a variety of human cells. Notably, these results establish that expression of all TLRs can be responsive to DNA metabolic insults, even exceeding p21 induction. However, there is considerable variability in the individual responses between subjects, TLRs, and treatments as summarized in the “box and whiskers” presentation of Figure S2. For example, the IR induction of the ten TLR genes varies from 1- to over 4-fold for each TLR except TLR3 (Figure 1 and Figure S2), which is not detected in the PHA-stimulated lymphocytes of most subjects. This agrees with the large differences in expression of IR-inducible genes between human lymphoblastoid cell lines [21]. The variability in gene expression among the 18 subjects can represent a continuum of responses (e.g., TLR4 expression after IR), or exhibit more of a binary induction pattern (e.g., TLR8 response after IR; see Figure 1). Specifically, for the TLR8 gene, approximately half the subjects respond strongly and half exhibit a much lower response, a finding that appears to be genetically determined [8], as discussed below. Different agents also elicit differential TLR gene expression responses. For example, TLR1 is responsive to IR by more than 2-fold in most subjects, but generally there is only a small level of induction in response to 5FU or Doxo.
To better assess each TLR gene response across subjects, we portrayed the induction of TLRs for each subject in a format akin to a heat map, as described in Figure 1E for IR, where 2.5-fold induction is indicated in red and <2.5 fold in black (this value corresponds to the minimal p21WAF1 response for nearly all agents and subjects; see Figure S3 for other agents). Among three subjects (BS4, 5, 21), all TLRs were induced at least 2.5-fold by IR exposure, while only 1 (BS2), 2 (BS9) or 3 (BS8, 13) TLRs were induced in four others. There is no obvious pattern to the differences between TLRs or subjects for IR as well as for 5FU and Doxo, although it is clear that TLR1 is generally much less responsive to the last two agents (Figure 1, Figures S2 and S3). Consistent with TLR3 induction by DNA damage in cancer cell lines [9], its expression was induced in 5 out of 7 subjects; however, TLR3 gene expression was not detected in 11 subjects (Figure 1 and Figure S3). Even though all TLRs are responsive in at least one subject, only subjects BS4, 5 and 21 exhibited high responses for most TLRs for all agents tested (Figure S3).
We also addressed statistically the responsiveness of the population as a whole (18 subjects, Figure 1A, 1B, 1C, except TLR3 which was expressed only in 7 subjects) to IR, 5FU, and Doxo employing a t-test (see Material and Methods) to assess the ability of each agent to induce expression of each TLR gene in the population. As expected, induction of the p21 gene in the population was highly significant for all the treatments (p<0.0001). While there was variation between individuals, all the TLR genes in the population were responsive to IR (p<0.0001). A similar likelihood of responsiveness (p<0.0001) was observed for 5FU and Doxo treatment for all but the TLR1 and 7 genes (p<0.003) and TLR9 (p = 0.011 for 5FU and 0.053 for Doxo); however, there were substantial responses for the TLR 9 genes of several individuals. Collectively, these results suggest that multiple pathways may influence TLR expression after DNA damage, and these are specific to the mode of chromosomal stress.
Since p53 mediates many DNA damage responses and given our finding that all TLRs respond to at least one DNA metabolic disruptor, we sought to examine in more detail the ability of p53 to induce TLRs. PHA-stimulated lymphocytes were exposed to the p53 activator nutlin (10 µM) to increase p53 levels. Stabilization of p53 normally occurs through stress-induced post-translational modifications affecting both p53 and MDM2 [22]. Nutlin can directly prevent p53 destruction by interfering with the MDM2-p53 interaction [23]. As expected, all treatments activated the p53 pathway in the lymphocytes of all individuals as assessed by immunodetection of p53 and p21 proteins (Figure S4; Figure 1F is a representative example). As shown in Figure 1D, Figures S2 and S3, nearly all TLR genes in most subjects are induced over 2.5 fold after nutlin treatment, except for TLRs 1 and 7. The TLR1 gene much less responsive while the TLR7 gene is generally repressed by p53 induction. Interestingly, TLR7 is induced by the DNA damaging treatments, suggesting p53-independent induction mechanisms. While TLR3 is not detected in 11 subjects, it is induced by nutlin in 5 of the remaining 7 subjects. Although there was variation between individuals, we found that most of the TLR genes in the population were responsive to nutlin. There was no statistically significant induction of the TLR9 gene (p = 0.11); however, there was statistically significant repression of the TLR7 gene (p = 0.006). The expression of all the remaining TLR genes was significantly induced in the population (p<0.003).
A role for p53 in TLR induction is further indicated by the finding that co-treatment with the p53 inhibitor pifithrin-alpha [24] represses the induction of TLRs by IR and other agents (Figure 1G for IR; Figure 2 for Doxo, 5FU and nutlin). Notably, samples taken on two occasions from the same individuals in a 4 month-interval, blind study revealed a striking consistency in TLR gene expression patterns (Figure 3), thereby excluding technical or temporal variables as significant sources of variability in findings.
The p53 inducibility of most TLRs led us to investigate if p53 could act directly on transcription. As discussed in the Introduction, the commonly accepted consensus target RE consists of two decamers composed of (RRRC A/T A/T GYYY) separated by up to 13 bases (reviewed in [25], [26]). We had established rules for predicting in vivo functionality of p53 REs including separation of <2 bases and dependence on a core CATG (summarized in [3], [6]). Among several potential p53REs identified by in silico search and bioinformatic tools, we found that each TLR has at least one p53 target sequence within ±5 kb of the transcription start site that is predicted to provide at least a weak-to-modest p53 responsiveness (Table 2 contains the sites predicted to have the greatest functional responses). As shown in Figure 4A, each of these target sequences (TLR7 was not examined because we did not find any p53-like binding sequences) could support p53-driven transcription of a luciferase reporter co-transfected, along with a p53 expression plasmid, into p53 null H1299 cells. For half of the TLRs, the induction levels were comparable to those obtained with the moderately responsive p53 target response element of AIP.
Based on these results, we examined whether the functional REs of Figure 4A are indeed targets of induced p53 using chromatin immunoprecipitation (ChIP) analysis. We chose a representative subject BS4 whose lymphocytes exhibited strong Doxo-induced expression (Figure 1) for all TLRs except TLR1 and TLR7. As shown in Figure 4B, there is clear binding following exposure of cells to Doxo at the established p53RE on p21WAF1 as well as at the TLR2, 4, 5, 6, 8, 9 and 10 sequences described in Figure 4A. Thus, all TLR genes are subject to DNA-damage associated transcriptional regulation and most are directly targeted by p53. (We also identified other p53-like sequences in the analyzed regions; however, these were predicted to be less functional than the sequences examined). Furthermore, a direct role for p53 in TLR gene expression was demonstrated using p53 null SaOS2 osteosarcoma cells that have p53 under the control of a tetracycline inducible promoter [27]. Expression of wild-type p53, but not the transcriptionally inactive G279E mutant protein, results in induction of all TLRs except TLR7 (TLR8 was not tested because it is not expressed by SaOS2 cells; Figure S5).
We also determined whether p53 induction of TLR genes by nutlin can lead to a corresponding increase in protein. (Nutlin was chosen because it typically led to the largest increase in p53, as shown in Figure S3.) Using western blot analysis with TLR specific antibodies (see Materials and Methods), the levels of TLR2 and TLR5 proteins were examined in the membrane fraction from stimulated lymphocytes of volunteers BS25 and B26 (sufficient cells were obtained from these subjects to enable the protein measurements; the TLR2 and TLR5 proteins were only detected in the membrane fraction as noted in the Material and Methods). Nutlin treatment resulted in a substantial increase in both proteins in the membrane fraction which corresponded well with induced expression of the TLR 2 and TLR5 genes, as described in Figure 5.
To address the generality of DNA damage-induced TLR expression and the role of p53, we also examined alveolar macrophages since they are well-known to play a pivotal role in pulmonary innate immunity and are susceptible to DNA damage from environmental exposures including cigarette smoke and particulate matter [28]–[30]. The cells were collected by bronchoalveolar lavage of healthy human subjects and treated ex vivo. Among 6 subjects, there were considerable differences between TLRs with little or no induction of TLR1 and TLR6 by Doxo or nutlin, as described in Figure 6, and a dramatic induction of TLR9 by Doxo, suggesting cell-specific factors that determine damage-TLR response profiles. In general, fewer of the TLR genes in the macrophages responded to Doxo and nutlin as compared to lymphocytes and the response levels were not as large.
Induction of cytokines is a prototypical functional response that is downstream of TLR activation [11], [31], [32], and increases in TLR expression can enhance PAMP or DAMP-induced signaling and innate-immune mediated effects. As shown in Figure 7, pretreatment of freshly isolated CD3+ T-lymphocytes (see Figure S1) with nutlin to induce TLR2 by p53 resulted in 2- to 5-fold increased expression of interleukin IL-1 and IL-8 by the TLR2 ligand PAM3CSK4, suggesting that TLR induction by p53 can directly affect innate immune function. (The TLR2 induced cytokine response was examined because of our observation that T-lymphocytes from all subjects experience nutlin-induced TLR2 expression, as shown in Figure 1 and Figure S3). The magnitude of cytokine induction varied between subjects (see Figure S6), suggesting that other factors, besides p53-induced gene expression, affect innate immune responsiveness.
Our previous report [8] of a SNP in a potential p53 response element of the TLR8 promoter (AAACAT(G/A)TCa; see Table 1) provided a unique opportunity to directly assess the relationship between p53 and TLR expression. While we had established large differences in the potential p53-responsiveness of the two SNP sequences [8]; see Figure 4A for the “positive”, p53-responsive G-allele), their cellular impact could not be assessed because the TLR8 gene was not expressed in the human cell systems previously examined. Unlike the results with cell lines, we observed a dichotomous response in TLR8 expression in lymphocytes following IR and nutlin treatment (Figure 1), and in alveolar macrophages following nutlin exposure (Figure 6), with some subjects showing low TLR8 induction and others robust induction. We, therefore, determined which p53 response element alleles were present and their relationship to TLR8 induction. Since TLR8 is located on the X-chromosome, males carry only a single copy and females 2 copies (the specific alleles for each volunteer are described in Table 1). As shown in Figure 8 and Figure S3, the ability of nutlin and IR to induce TLR8 correlates well with the presence of the G-allele. The frequency of this allele in our study is 0.43 (13/28, which corresponds to the frequency in the general population (Table 1). Importantly, TLR8 induction was always high when only G-alleles were present (both alleles in females or the single allele in males) and absent if there were only A-alleles. However, among the 3 female subjects that were heterozygous for these alleles, only one responded poorly to both nutlin and IR. Possibly, the variability between female subjects heterozygous for the SNP is related to X chromosome inactivation.
While these results demonstrate differences in the impact of the polymorphic alleles, the differences do not extend to 5FU or Doxo, suggesting that other p53-related target sequences are responsible for the induction or involvement of additional p53-independent mechanisms (Figure 8 and Figure S3). Notably, these findings with TLR8 provide the first direct demonstration, to our knowledge, in humans of the ability of a specific RE in a p53 target gene to drive transcription. The SNP-associated differences in expression may provide opportunities to identify other factors in human primary tissue that determine the ability of specific REs to support p53-driven transcription [33], [34]. Although there are SNPs in the REs of TLR5 and TLR6, as described in Table 2, they are predicted to have no functional impact (see [6]–[8]).
Our results with primary human cells have been confirmed for DNA damage induction of TLRs (to be presented elsewhere) in several human cell lines including the previously reported TLR3 [9]. However, as shown in Figure S7, they do not extend to murine cells. The TLR responses to Doxo, 5FU and IR were found to be small in mouse peritoneal elicited macrophages, bone marrow-derived macrophages, and embryonic fibroblasts (MEFs; except for TLR9 in MEFs). The low induction level appears to be p53-independent based on results with nutlin and on the similar responses in p53-positive and -null MEFs. These results are consistent with the lack of sequence conservation between humans and rodents of functional p53 response elements in TLR genes (Figure S8) even though the coding sequences are well-conserved. Although we were able to identify p53 RE-related sequences in the vicinity of the transcription start site of several mouse TLRs, none were predicted to be functional p53 targets.
Cellular stress and the inflammatory response are intricately linked pathways subject to endogenous and exogenous challenges. Here, we provide the first evidence that DNA stressors may also modulate inflammatory responses at a fundamental level in primary human cells, namely, by altering TLR expression. All members of the human TLR gene family tested (TLR1-10) are responsive to at least one disruptor of chromosome metabolism. While there are considerable variations between individuals, for each TLR gene there is a group of several subjects that is responsive to at least one of the stressors (5FU, doxorubicin, IR and/or nutlin). Note, for example, that there is at most a low level of TLR1 induction by 5FU and nutlin in the primary cells from 18 subjects and there is even a general repression of TLR7 by nutlin. We thus propose that, contrary to the paradigm that innate immunity is hard-wired, the TLR system in humans actually has a complex, robust responsiveness to environmental conditions that challenge the integrity of the genome. We establish that induction of TLRs by DNA damage is a class effect that in many cases can be mediated by p53 (including repression of TLR7) and prevented by the p53 inhibitor pifithrin. In support of these findings, several potential p53 binding sites were identified in the proximity of the transcription start sites for all TLRs, except TLR7. The new p53REs were identified using functionality rules to predict p53RE responsiveness [6], [7]. These were confirmed with reporter assays and by the binding of p53 in human lymphocytes following stress activation. In addition the functional TLR8 SNP in the p53 target response element directly confirms a role for p53. Interestingly, for TLR2 and TLR10, the most functional sites were predicted to be noncanonical, containing only a half–site p53RE (i.e., one decamer). Previously, the genes FLT1 [35] and RAP80 [36] were shown to be directly controlled by p53 through half-site REs and several other target genes have been identified that are predicted to be regulated by noncanonical p53 RE (summarized in [6]).
The inclusion of TLR genes expands the universe of genes and biological functions that fall within control of the p53 master regulator. The addition of the set of TLR genes identifies a biological gene niche regulated by the p53 master regulator that is distinct from rodents whose TLR genes appear to lack functional p53 REs. It is interesting that a similar niche was identified for all the DNA metabolic genes that are p53-responsive in humans in that none of them respond to p53 in rodents [7], [36]. p53 provides a rapid, integrated signaling mechanism for increasing or maintaining gene responses to acute and chronic DNA damage stress. In a more general sense, we suggest that the evolutionary inclusion of sets of genes with related functions may provide an efficient means of dealing with sudden, temporary challenges that can result from DNA damage and/or DNA damage itself may be a modulator of broader biological threats, as for the case of infection.
We speculate that there may be feedback loops that integrate this newly identified role for p53 in DNA damage and innate immune responses, as described in Figure 9. In this scheme, DNA damage from environmental agents or potentially from TLR-elicited reactive oxygen species (ROS) may amplify the responsiveness of the innate immune system by promoting TLR upregulation. p53 itself displays a variety of roles in mediating ROS signals [37], [38]. On the other hand, TLR upregulation may also sensitize tissues to maladaptive aseptic inflammation in the setting of environmental injury. For example, tissue injury is reported to induce inflammation through release of DAMPs that act upon TLR2, TLR4, and TLR9 [39], [40]. We speculate that, during tissue injury, upregulation/activation of TLRs may serve as a cell-fate counterbalance to p53-mediated pro-apoptotic responses by leading to activation of the pro-survival factor NF-κB [39]. This may be particularly relevant to cancer therapy, as stimulation of TLR5, 7, 8, and 9 have been shown to modify cellular radio- and chemoresistance [41], [42]. These findings demonstrating that p53 can increase an inflammatory response differ from the generally held view relating to the antagonistic affect of p53 on inflammation directed by NF-κB [41]. However, the mechanism here is quite different in that it involves the p53-mediated increase in a receptor that translates ligand interactions into cytokine responses. Our results may be particularly relevant to diseases in which variations in T-cell function can impact pathogenesis, such as autoimmune disease, asthma, and IBD. For example, recent reports suggest that intestinal inflammation can induce genotoxicity in circulating leukocytes [43], while increased DNA damage is also detected in lymphocytes obtained from rheumatoid arthritis patients [44]. The heightened pro-inflammatory status in such patients, as well as the common finding of systemic or multi-organ inflammation during exacerbations of autoimmune disease, might be mediated by circulating immune cells which have suffered DNA damage during passage through inflamed tissues.
Beyond the TLR gene and cell subset variability in response to DNA damage and p53 activation, we also demonstrate considerable inter-individual variation. This variability may be relevant to inter-individual differences in susceptibility to a wide spectrum of diseases and therapies. Our results suggest that various anti-cancer agents may yield different patterns of responses across TLRs and between subjects. Since TLR ligands are increasingly used as adjuvants for vaccines (TLR4 and TLR9) and cancer treatment (TLR3/7/9) (reviewed in [45]), the ability to detect and predict genetically determined inter-individual variability in TLR induction may prove a useful therapeutic tool. Future studies are warranted to determine whether single agents such as nutlin, or even factors that induce chromosome stress, may serve as useful immune adjuvants through manipulation of TLR expression in human subjects.
Healthy adult volunteers were recruited to the NIEHS Clinical Research Unit and underwent phlebotomy. Subjects were excluded if they had a history of recent infection, were on anti-inflammatory medications, or tested positive for hepatitis B, C or HIV. Up to 300 ml of whole blood were withdrawn from an antecubital vein into citrated tubes. Lymphocytes were isolated using percoll (Sigma) and anti-CD3-coupled Magnetic Beads (Miltenyi Biotec) as per manufacturer's protocol. Cell purity was >98% after percoll/magnetic bead isolation based on flow cytometry. We maintained lymphocytes in RPMI supplemented with 10% FBS. For T cell stimulation, cells were activated with phytohemagglutinin-M (PHA, Invitrogen, 3% vol/vol) for 72 h. The total number of lymphocytes available per treatment conditions after PHA was typically around 10 million cells or less. Cells were treated starting at 48 h post PHA addition and cell cultures were harvested 24 h later. Freshly isolated CD3+ cells were treated with nutlin for 20 h or DMSO as a vehicle control, then washed and exposed to TLR1/2 ligand PAM3CSK4 (1 µg/ml) at 1×106 cells/ml. Total yield of CD3+ cells from a single subject was typically around 10–15 million cells or less. Protocol and procedures were approved by the NIEHS Institutional Review Board.
Healthy, nonsmoking male volunteers, 18 to 40 yr of age, underwent fiberoptic bronchoscopy with lavage to procure alveolar macrophages. The screening procedures for each subject included a medical history, physical examination, and routine hematologic and biochemical tests. None of the subjects had a history of asthma, allergic rhinitis, chronic respiratory disease, or cardiac disease. Subjects were excluded from the study if they had suffered a recent acute respiratory illness and were asked to avoid exposure to air pollutants such as tobacco smoke and paint fumes. A fiberoptic bronchoscope was wedged into a segmental bronchus of the lingula. Six 50-ml aliquots of sterile saline were instilled and immediately aspirated. The procedure was repeated on the right middle lobe, again using 300 ml of saline. Samples were put on ice immediately after aspiration and centrifuged at 300 x g for 10 min at 4–8°C. Cells from all aliquots were pooled, washed twice with RPMI 1640, and re-suspended in RPMI 1640 at 1.0×106/mL. Total yield was typically around 10–15 million cells or less. Around 2.0×106 cells per well were seeded in a 12 well plate. After 2 hr, cells were washed twice with warm PBS and 2 ml of growing media was added to each well. Cells were then treated with nutlin or doxorubicin. Cells were harvested 24 hr post-treatment. The protocol and consent form were approved by the University of North Carolina School of Medicine Committee on the Protection of the Rights of Human Subjects. Prior to participation in the study, subjects were informed of the procedures and potential risks and each signed a statement of informed consent.
H1299 lung cancer cells (American Type Culture Collection) were routinely maintained following standard conditions and procedures for culturing mammalian cells. All cultures were incubated at 37°C with 5% CO2. p53 tetracycline inducible SaOS2 TET-off cell lines expressing the wild-type or the G279E mutant protein were cultured as described previously [27]. p53 expression was kept “off” by 2 mg/ml doxycycline (Clontech). To induce the p53 expression, cells were washed 3X with phosphate-buffered saline (PBS) and placed in medium lacking doxycyline during 24 h.
For drug treatment and p53 activation, cells were incubated with doxorubicin (Sigma 0.3 µg/mL), nutlin3 (Sigma, 10 µM) and 5-fluorouracil (Sigma, 300 µM). For ionizing radiation treatment, cells were irradiated at 1.56 Gy/min from a Shepherd cesium irradiator in PBS at room temperature at final dose of 4 Gy. Where indicated cells were also pretreated 2 h with p53 inhibitor pifithrin-alpha (Sigma, 40 µM).
Total RNA was isolated by RNEasy kit (Qiagen). Real-time PCR was performed in triplicate with Taqman PCR Mix (Applied Biosystems) in the 7000 ABI sequence Detection System (Applied Biosystems). All human and mouse primers were purchased from Applied Biosystems (information available upon request). For PHA stimulated lymphocytes and alveolar macrophages expression of TLR genes was normalized to 18S ribosomal RNA gene while for freshly isolated CD3+ T cells, the glucuronidase-beta gene was used for normalization.
Pairs of complimentary oligonucleotides for the desired p53RE from selected TLRs and containing restriction sites were cloned into the open reading frame of firefly luciferase pGL4.26 plasmid (Promega) previously double digested by Xho I/Kpn I restriction enzymes. The identity of the inserts was confirmed by DNA sequencing. Luciferase activity was measured 48 h after Fugene6- mediated co-transfection of the TLR p53RE constructs in the presence of p53 (pC53-SN3) or empty vector pCMV NEO-BAM3 along with pRL-TK Renilla as a transfection efficiency control into p53 null H1299 cells, as previously described [5]. Forty-eight hours post-transfection extracts were prepared using the Dual Luciferase Assay System (Promega) following the manufacturer's protocol and luciferase activity was measured on a Victor Wallac multilabel plate reader (PerkinElmer). Relative luciferases activities for each construct was defined as the mean value of the firefly luciferase/Renilla luciferase rations obtained from 4 independent experiments performed in triplicate.
Whole cell extracts were quantified using the Bradford protein assay kit and gamma globulin as a reference standard (BioRad). For TLR protein detection, cellular pellets were subjected to subcellular protein fractionation (Thermo Scientific) following the manufacturer's instructions and protein was quantified using BCA protein assay kit (Thermo Scientific). For TLR western blot analysis ∼30 µg of total membrane fraction was used, while for the analysis of other proteins ∼25 µg of total cell extract were used. As expected, we did not detect TLR2 and TLR5 in the cytosolic fractions; therefore, those data are not included. Proteins were resolved on 4–12% BisTris NuPAGE and transferred to polyvinylidene difluoride membranes (Invitrogen) and were visualized with primary antibodies followed by horseradish peroxidase–conjugated goat anti–mouse or donkey anti-goat immunoglobulin (Santa Cruz Biotechnology) through the use of enhanced chemiluminescence reagents (Amersham Biotechnology). The primary antibodies used in these studies were against p53 (DO1, Santa Cruz Biotechnology,), p21 (SXM30, BD Biosciences Pharmigen) and Actin (C-11 Santa Cruz Biotechnology). The following is the list of TLR antibodies tested in this study in order to detect TLR protein expression in whole cell extracts as well as membrane and cytosol protein fractions: TLR8 ab24185 and TLR10 ab45088 from Abcam, Inc.; TLR1#2209, TLR2#2229, TLR7#2633 and TLR9#2254 from Cell Signaling. We also used a Toll-like receptor detection kit that includes antibodies for all human TLRs (TLR1 to TLR10 antibodies from ProSci, Inc., as well as TLR3-4H270 and TLR5-H1-27 antibodies from Santa Cruz Bitoechnology. The TLR3-IMG-315A, TLR4-IMG6370A antibodies were from IMGENEX. Only the TLR2 (Cell Signalling) and TLR5 (Santa Cruz) gave clear results. Attempts to detect other TLRs with these antibodies were unsuccessful and appear to be a general problem with TLR antibodies from our collective experience. One of the antibodies enabled us to detect induction of full length TLR4 by nutlin; however, those results are not presented due to the appearance in both untreated and treated samples of nonspecific bands.
ChIP assays were done as previously described [35] using ChIP kits (Millipore). Approximately 40×106 PHA stimulated lymphocytes were used for each experimental sample. Cell lysates were sonicated using conditions that yield chromatin fragments 200–500 bp long. One microgram of DO-7 p53-specific monoclonal antibody (BD Biosciences Pharmigen) was used per ChIP assay. As a negative control, mouse Ig (Santa Cruz Biotechnology) was used. PCR amplifications were performed on immunoprecipitated chromatin using primers to amplify specific regions on the TLRs promoters (sequence information available upon request). The PCR cycles were as follows: initial 10 min Taq polymerase (Invitrogen) at 95°C followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. The PCR products were then run on a 1.8% agarose gel.
Cells were resuspended in 100 µl of PBS and incubated with 5 µl of fluorescent antibody per sample for 30 min, then washed and fixed with 0.5% paraformaldehyde. The fluorescence intensity was evaluated using a Becton Dickinson LSR II Flow Cytometer. All antibodies used for FACS were from BD Pharmigen.
SNPs were assessed by three different approaches. In RFLP assays, genomic DNA was extracted from Percoll-isolated lymphocytes by DNeasy kit (Qiagen). For the SNP in the TLR8 p53RE#6 (AGGCAAGATGAAACAT(G/C)TCA), the G-SNP creates a unique restriction cutting site for NspI (R CATG Y). PCR was performed with 100 ng of DNA, 50 pmol of each primer, 1.5 mM MgCl2, 1 µL 10× PCR buffer, and 0.0125 U of Taq (Invitrogen). After 10 min at 94°C, 35 cycles were repeated as follows: 94°C 30 s, 60°C 30 s, and 72°C 35 s; this cycling was followed by a final extension at 72°C for 10 min. The PCR product was digested with 5 U Nsp I (New England Biolabs, Ipswich, MA), at 37°C for 4 h. Since Nsp I recognizes the polymorphic sequence, a G allele is demonstrated by the presence of two fragments 109 and 69 bp in a gel. The A allele is revealed by the presence of a single 177 bp band. The following primers were used for amplifying the region containing the p53RE on TLR8 promoter region:
Forward: 5′TCATAACAAGGTGTTCCACAGTC-3′
Reverse: 5′-ATCTGGCCCTTTACAGAAAAAGTT-3′.
The status of this SNP was determined also by using a Taqman SNP genotyping assay. All primers were purchased from Applied Biosystems (Assay ID:C_27497635_10). For direct sequencing the region containing the p53RE was first PCR amplified using the following pairs of primers:
Forward: 5′-TTGAATTCCCTTAGGGTGTGA-3′,
Reverse: 5′-AAACTGCCTTCGATTATTATTATTACA-3′
This was followed by running the samples on a TEA-agarose gel. The expected product (397 bp) was cut out and cleaned using QIAquick gel extraction kit (QIAGEN). The sequencing reactions used Big dye (Applied Biosystems) per manufacturer recommendations and the following primers:
Foward: 5′TCATAACAAGGTGTTCCACAGTC-3
Reverse: 5′-ATCTGGCCCTTTACAGAAAAAGTT-3′
p53+/+ and p53−/− mouse embryonic fibroblasts (MEFs) were cultured in DMEM media and 10% of FBS. Female C57BL/6 mice were purchased from Jackson Laboratories. All experiments were performed in accordance with the Animal Welfare Act and the U.S. Public Health Service Policy on Humane Care and Use of Laboratory Animals after review of the protocol by the Animal Care and Use Committee of the National Institute of Environmental Health Sciences. For murine peritoneal macrophage harvests, mice were injected i.p. with 2 ml of 4% Brewer's thioglycollate and euthanized 96 h later. The peritoneum was washed with 10 ml ice cold PBS three times. Cells were centrifuged (1,000x RPM, 6 minutes, 4°C) and washed twice with sterile PBS. Peritoneal exudate macrophages were resuspended in DMEM/0.1% FBS, counted, and plated at 2×106 cells/well in a 12-well plate. Cells were allowed to settle for 2 h (37°C/5% CO2) before replacing media with DMEM complimented with 10% FBS.
For bone marrow-derived macrophages (BMM), marrow was flushed from femoral and tibial bones using bone marrow media (DMEM/2 mM L-glutamine/10% L929-conditioned medium/10% FBS). Cells were spun down (2200x RPM, 5 min, 4°C), brought up in 1 ml sterile ACK buffer, incubated 4°C for 1 min after which 10 ml PBS was added. Cells were spun as above, resuspended in bone marrow medium, counted, and plated at 1×106 cells/well in a 12-well plate. Cells were cultured at 37°C and 10% CO2 in 2 ml bone marrow medium/well and fed on Day 5 with addition of 1 ml medium/well. Experiments were performed on Day 6. At 24 h post-treatment, cells were harvested for RNA extraction.
To examine statistically whether the average mRNA fold-change at each locus in the population sampled differed from 1 for the various exposures, we applied one-sample Student's t tests to log-transformed values of mRNA fold change. The logarithmic transformation helps the data meet the distributional assumptions for the t test. This procedure, in effect, tests the null hypothesis that the geometric mean mRNA fold change at the locus is equal to 1 against the two-sided alternative that the geometric mean differs from 1.
|
10.1371/journal.pbio.1001569 | Rare Species Support Vulnerable Functions in High-Diversity Ecosystems | Around the world, the human-induced collapses of populations and species have triggered a sixth mass extinction crisis, with rare species often being the first to disappear. Although the role of species diversity in the maintenance of ecosystem processes has been widely investigated, the role of rare species remains controversial. A critical issue is whether common species insure against the loss of functions supported by rare species. This issue is even more critical in species-rich ecosystems where high functional redundancy among species is likely and where it is thus often assumed that ecosystem functioning is buffered against species loss. Here, using extensive datasets of species occurrences and functional traits from three highly diverse ecosystems (846 coral reef fishes, 2,979 alpine plants, and 662 tropical trees), we demonstrate that the most distinct combinations of traits are supported predominantly by rare species both in terms of local abundance and regional occupancy. Moreover, species that have low functional redundancy and are likely to support the most vulnerable functions, with no other species carrying similar combinations of traits, are rarer than expected by chance in all three ecosystems. For instance, 63% and 98% of fish species that are likely to support highly vulnerable functions in coral reef ecosystems are locally and regionally rare, respectively. For alpine plants, 32% and 89% of such species are locally and regionally rare, respectively. Remarkably, 47% of fish species and 55% of tropical tree species that are likely to support highly vulnerable functions have only one individual per sample on average. Our results emphasize the importance of rare species conservation, even in highly diverse ecosystems, which are thought to exhibit high functional redundancy. Rare species offer more than aesthetic, cultural, or taxonomic diversity value; they disproportionately increase the potential breadth of functions provided by ecosystems across spatial scales. As such, they are likely to insure against future uncertainty arising from climate change and the ever-increasing anthropogenic pressures on ecosystems. Our results call for a more detailed understanding of the role of rarity and functional vulnerability in ecosystem functioning.
| In ecological systems most species are rare—that is, represented by only a few individuals or restricted to particular habitats—and are vulnerable to being lost. Yet the ecological consequences of such biodiversity loss are often overlooked and remain controversial. In the best-case scenario, the functions that these rare species provide to their ecosystems might be insured by more common species, which share combinations of functional traits with the rare species, thereby helping to maintain ecosystem functioning despite rare species loss. In the worst-case scenario, rare species would have functional traits that are distinct from those of common species; thus, the functions they support would also be vulnerable to extinction. We examined three highly diverse ecosystems (coral reefs, alpine meadows, and tropical forests) and addressed whether common species would insure against the loss of functions carried by rare species. We demonstrate that highly distinct combinations of traits are supported predominantly by rare species. It is thus not only the quantity but also the quality of biodiversity that matters. Thus, our findings highlight that we need to change how we think about biodiversity in general, and about conservation strategies in particular, by moving beyond the protection of biodiversity per se and beyond focusing on iconic, charismatic, or phylogenetically distinct species, to protecting species that support irreplaceable functional roles and associated services.
| The vast majority of species are rare—that is, comprising few individuals—and often have restricted geographic distributions [1]. Although several forms of rarity have been defined with respect to the trajectories by which species become extinct [2],[3], rare species are all seen as highly vulnerable to overexploitation [4], habitat loss [5], competitive interactions with exotic species [6], and climate change [7]. Rare species have thus received important consideration from conservation biologists because their extirpation contributes disproportionately to the ongoing sixth extinction crisis [8]. This biotic impoverishment may, in turn, alter the biogeochemical and dynamic properties of ecosystems [9]. Beyond aesthetic, cultural, and moral arguments, the maintenance of ecosystem functioning has thus become a powerful justification to limit biodiversity erosion [10]. Indeed, most key ecosystem processes, such as organic matter degradation, bioturbation, bioerosion, and productivity, are threatened by the loss of functions performed by particular species [11],[12],[13], some of which may be rare.
It has long been assumed that the loss of rare species will have a limited impact on ecosystem functioning at short terms and local scales, given their low abundance within communities [14]. However, this hypothesis has been challenged because the loss of rare species can affect local ecosystem processes [15],[16] and rare species can contribute significantly to long-term and large-scale ecosystem functioning [17], eventually providing ecological insurance in variable environments where species abundances vary in time [18]. Indeed, rare species may perform functions complementary to those delivered by other, even closely related, species as a result of their distinct functional traits [19]. In turn, those rare species may increase the functional diversity of local communities [20], sustain ecosystem functioning [21], and provide functional traits able to support the main ecosystem processes under future environmental conditions [18].
Ecosystems depend on the maintenance of multiple processes [13] across space and time under environmental-change scenarios [22]. This requires species with complementary functions [23]; however, current knowledge is still far from being able to assess the roles played by individual species, especially in highly diverse regions where data are lacking even for common species. Rather, current practice is to assess the ecological role of species indirectly via their functional traits. Here, we assume that species with distinct combinations of functional traits are more likely to support functions that cannot be delivered by species with more-common traits. This assumption is based on experiments showing that species with traits that are not present in others (thus minimizing functional redundancy) regulate ecosystem processes [24], and that trait dissimilarity within species communities, favored by the presence of species with distinct trait combinations, increases ecological process rates [21],[25]. A modeling study further showed that the covariance between species extinction risks and their functional traits mediates bioturbation, with species possessing the most distinct traits having the highest impact [12]. In practice, this assumption needs to consider multiple functional traits to embrace the range of potential roles that species may play [26]. In this respect, some species play unique roles in the ecosystem according to the distinctiveness of their functional traits relative to the rest of the species pool [27]. The loss of species with such distinctive traits may thus affect ecosystem functioning [12], especially when multiple functions are considered [21]. Conversely, functional redundancy, where different species sustain similar functions, may insure against the loss of ecosystem functioning following biodiversity erosion [28],[29]. It is therefore critical to know the degree to which rare species share combinations of functional traits with common species.
In the best-case scenario, common species would share combinations of functional traits with rare species, thereby maintaining ecosystem functioning despite the loss of rare species. The protection of common species would thus become the primary focus for the maintenance of ecosystem processes [30]. In the worst-case scenario, rare species would have functional traits markedly distinct from those of common species; hence the functions they support would be vulnerable to extinction. Vulnerable functions are, therefore, defined by having low insurance—that is, there are few species and few individuals with similar combinations of traits that provide this particular function. In this case, the loss of rare species would have greater ecosystem impacts than expected simply as a result of numerical species loss. The conservation of rare species would thus be a priority for the maintenance of ecosystem functioning, beyond the classic motivations of preserving the diversity of life and the precautionary principle [31]. This issue is even more critical in species-rich ecosystems where high functional redundancy among species is likely [32],[33] and where it is thus often assumed that ecosystem functioning is buffered against species loss.
Recent studies that investigated the contribution of rare species to functional diversity reached inconsistent conclusions, but were restricted to local samples of a limited number of species [20],[34]–[36]. The question of whether species with unusual combinations of functional traits, which are likely to support vulnerable ecological functions, are overwhelmingly rare is still unresolved in species-rich regional assemblages and at large scales. An extensive body of literature has looked at why some species are specialists and searched for suites of traits underpinning the link between rarity and specialization [37]. In our study, we adopted an alternative approach by focusing on whether distinct trait combinations, which could be irreplaceable, were likely to be supported by rare species. Using extensive datasets of species local abundances, regional occurrences, and functional traits from three highly diverse ecosystems (846 coral reef fishes, 2,979 alpine plants, and 662 tropical trees), we demonstrate that highly distinct combinations of traits are supported predominantly by rare species both at the local and regional scales. Moreover, we show that the species that are likely to support the most vulnerable functions—that is, those that might be supported by poorly insured functional trait values—are rarer than expected by chance in all three ecosystems, again at both local and regional scales.
For each of the three datasets we estimated two complementary aspects of rarity: (i) local abundance as the abundance in communities where the species was found and (ii) regional occupancy as the proportion of communities in which the species was recorded. For simplicity, we use “common” as the antithesis of “rare” regardless of the scale considered. Rarity is a continuous measure, so we defined two thresholds to classify species. At a local scale, we defined “rare” species as those with a local abundance (number of individuals for fish and trees, surface cover for plants) less than 5% of the most abundant species, whereas the “rarest” species were those represented by a single individual (for fish and trees) or less than 1% of most abundant species (for plants). At the regional scale, we defined “rare” and “rarest” species as those having less than 5% of the occupancy of the most common species in the dataset, and as those having only one occurrence, respectively. We estimated the functional distinctiveness of each species using its functional distance from the rest of the species pool based on multiple traits. We then regressed functional distinctiveness against regional occupancy, both being measured on a standardized scale to allow comparisons among ecosystems.
Functional distinctiveness was negatively and significantly related to commonness, whether estimated as local abundance or regional occupancy (Figure 1). Reef fishes and tropical trees show a consistently triangular relationship: the most unusual combinations of functional traits—that is, those with high functional distinctiveness—were invariably supported by rare species, whereas species with low functional distinctiveness were either common or rare. For alpine plants, the slopes of the 95th and 99th quantile regressions were not significant at both scales, but the two species with the highest functional distinctiveness values (Saxifraga mutata and Rosa sempervirens) were rare at local and regional scales. Across all three ecosystems, the most functionally distinct species (having a functional distinctiveness value higher than that predicted by the 99th quantile regression) all had a regional occupancy less than 50% of the maximum value and most of them were rare (Figure 1).
We then estimated the potential vulnerability of the functions supported by each species. Vulnerability is inversely related to the extent of insurance provided by functionally similar common species. If a species shares a similar combination of traits with common species, it is more likely to support functions with a high insurance and low vulnerability to extinction. Vulnerability is therefore estimated based on the commonness of species that share similar combinations of traits. At both scales, in all three ecosystems, functional vulnerability significantly decreased with commonness, resulting in concordant triangular relationships (Figure 2). The most vulnerable functions, those that might be supported by poorly insured combinations of functional traits, were mainly supported by rare species, whereas common species never supported highly vulnerable functions.
The association of rarity and functional vulnerability could result from a sampling effect, given the many rare species in our datasets. Therefore, we tested whether the rare or rarest species, at two different scales, were over- or underrepresented in different levels of functional vulnerability. We compared the observed percentages of rare and rarest species for different levels of functional vulnerability with those expected if rarity and functional vulnerability were independent. At the local scale (Figure 3A), the rarest species (only one individual by sample) were significantly overrepresented among reef fishes (47% against 12.5% expected) and tropical trees (54% against 36% expected) that are the most likely to support highly vulnerable functions (top 5%). Rarest species were consistently and significantly underrepresented among species supporting the least vulnerable functions (last 50%) in all three ecosystems. Rare species (less than 5% of local abundance) also contributed more than expected to the pool of species supporting highly and moderately vulnerable functions whatever the ecosystem, reaching a value up to 80% for tropical trees. At a regional scale (Figure 3B), in all three ecosystems, the rarest species were significantly overrepresented among those most likely to support highly vulnerable functions (top 5%) and underrepresented among species supporting the least vulnerable functions (last 50%). Rare species were even more overrepresented among those supporting highly vulnerable functions, whereas they were consistently underrepresented among those supporting the least vulnerable functions. For instance, 98% of fish species that were likely to support highly vulnerable functions in coral reef ecosystems were rare. This percentage was 89% and 52% for alpine plants and tropical trees, respectively.
The overrepresentation of rare and rarest species among those that support highly and moderately vulnerable functions could potentially result from the inclusion in our datasets of species from neighboring biogeographic regions. One would expect such “marginal” species to have combinations of traits adapted to other ecosystems, and to colonize only the edges of the studied ecosystems. If these “marginal” species were generating the observed rarity–vulnerability relationships (Figure 3), then we would predict that the species supporting highly and moderately vulnerable functions would occur farther from the geographic center of each ecosystem than would randomly chosen species. After calculating the marginality of each species, we performed randomization tests. They show that species supporting highly and moderately vulnerable functions were no more marginal than expected by chance (Figure S1). This result refutes the hypothesis that the most vulnerable functions were mainly supported by rare but geographically marginal species.
The link between species rarity and functional vulnerability is critical to understand the implications of biodiversity erosion for the decline of ecosystem functioning. Our study tackles this issue using three species-rich ecosystems at two different scales and offers a clear result: the combinations of traits with the highest distinctiveness values are all supported by rare species (Figure 1). We also assessed to what extent some “functional insurance” against the loss of rare species would be provided by regionally common species sharing similar combinations of traits. Since the relationships are triangular (Figure 2) we do not suggest that all rare species support distinct and vulnerable functions; indeed, most rare species probably support common and redundant functions. However, our results unambiguously show that rare species, those that have low local abundance and are regionally sparse, consistently carry the least-redundant combinations of traits.
If the distinctiveness of species-trait combinations does indeed map to distinct ecological functions, then such functions are likely to be the most vulnerable, given the ongoing threats to the rare species that sustain them [6]. This may be particularly important in areas with intense human impacts [5],[38]. We therefore suggest that the conservation of rare species offers more than taxonomic, aesthetic, cultural, or ethical value and must be also considered, in the addition to that of common species, when planning for the long-term maintenance of ecosystem functioning. For instance, some coral reefs can maintain processes and deliver services with a fraction of the species seen on reefs elsewhere [29],[39], but our results indicate that rare species may be functionally important and cannot be discounted. Indeed, our remarkably consistent results across scales highlight that, beyond protecting species with a low area of occupancy at a regional scale, it would be equally important to protect species that are locally rare, since they tend to support the more vulnerable functions and increase the level of functional diversity within communities, which in turn sustains local ecosystem processes [21],[40]. This latter argument is in agreement with a recent study showing that, using a global survey of reef fish assemblages, ecosystem functioning (as measured by standing biomass) scales in a non-saturating manner with biodiversity (measured either as species richness or functional diversity using the same fish traits as in our study) [41]. This precautionary principle applies in highly diverse ecosystems, characterized by high functional redundancy among species [32],[33], and even more so in lower diversity ecosystems where the potential for functional redundancy is limited [42],[43].
The functional importance of species carrying the most vulnerable combinations of traits is underlined by a closer examination of some of their roles in each ecosystem. On coral reefs, for example, the giant moray eel (Gymnothorax javanicus), ranked with the fifth-highest functional vulnerability value, is a large sedentary nocturnal benthic predator with few potential challengers to this role (Figure 4A). Likewise, the batfish (Platax pinnatus), supporting the 20th most vulnerable function, was recently identified as a key species in reef regeneration following a phase-shift to macroalgae—a role that many common herbivorous species were unable to play [44]. For plants, some functionally distinct rare species might seem unimportant at first glance but can have critical roles. For instance, the Pyramidal Saxifrage (Saxifraga cotyledon), a spectacular plant inhabiting cliffs (Figure 4C), occupies the 3rd rank for functional vulnerability. It has thick and dense leaves with long life span, indicative of slow plant growth and an adaptation to highly stressful environments [19]. S. cotyledon also possesses exceptionally long flowering stems, which makes it easy to detect and provides a locally important resource for pollinators in those species-poor habitats. Cytisus polytrichus, ranked 5th for functional vulnerability, is one of the few myrmechorous species in the region (i.e., dispersed by ants) and thus likely to be a principal resource for ant species. Among tropical trees, Pouteria maxima (Sapotaceae), which has the highest functional vulnerability value, is a recently described species known only from three collections in eastern French Guiana (Figure 4E). This tree grows to more than 40 m in height and at least 75 cm in diameter, with buttresses rising to 8 m in height. Its functional distinctiveness hinges on its very thick, dense leaves coupled with very thick plate-like bark and low-density wood. These traits provide it with the potential for exceptional resilience to the increasing frequency and intensity of fires that are likely to occur in the region [45], making the species an important potential buffer maintaining both forest structure and functioning during global climate change.
Our results thus call for new approaches that will specifically address the role of rarity and functional vulnerability in ecosystem functioning with, for example, experiments using species for which we have information (abundance and traits) in controlled designs where species richness and relative abundances would be kept constant. An important step forward will be to scale up our results from the one-trait one-function perspective to a more sophisticated multifunctionality perspective [21], to disentangle the relative contribution of rare and common species traits to complex ecosystem properties. As a complementary investigation, and since the species functional traits that determine ecosystem functioning may also drive their extinction risk, the level of covariation across species between the susceptibility to decline and the contribution to ecosystem functioning needs to be known [12],[46]. The loss of vulnerable functions, those that are overwhelmingly supported by rare species, may also render communities and hence ecosystem processes more unstable in the face of fluctuating environmental stressors at longer time scales. For instance, the salinity stress may change the hierarchy of successful functional traits in phytoplankton communities and compensatory growth of rare species may sustain primary productivity [18]. The conservation and restoration of communities may thus need to maintain or re-establish both dominant species that provide high levels of target functions and rare species, which might provide additional key functions under future conditions [47]. At a longer time scale, it remains crucial to know whether observed macro-evolutionary patterns of species functional traits would lead to niche filling and recovery of functions that were lost following selective species extinctions [48].
In the end, it is the functional abilities of species that are critical in maintaining ecosystems. Our results indicate that rare species may deliver more unusual and important functions than their local abundance or regional occupancy may suggest. We also show that such species are not geographically marginal, highlighting their potential importance to resilient ecosystem functioning particularly given future environmental uncertainty. Thus, even in highly diverse systems, we can no longer assume that rare species can be discounted by the high probability of functional redundancy. In these high-diversity systems, rare species may be as important as their more common counterparts.
Rarity can be considered at different spatial scales with several metrics being used depending on the species' geographic range size, habitat specificity, and local abundance [3]. Although these three components tend to be correlated across species [53], a joint consideration aids in depicting the scales of a species' extinction risk. For instance, species well adapted to a particular habitat may be regionally rare but abundant in appropriate habitat [50],[54]. Considering a range of spatial scales allows evaluation of a range of cases in which climate change, harvesting, or habitat degradation may threaten species 1. Accordingly, we defined two categories of rarity (rare and rarest species) using thresholds and two scales (regional occupancy and local abundance). For the three datasets, regional rare species were defined as those with a regional occupancy of less than 5% of the maximum observed value across the species pool, while the regional rarest species were those with only one occurrence. For reef fish and tropical trees, the local rarest species were defined as those with an average of one individual by sample where present, while for alpine plants the rarest threshold was set at less than 1% of the maximum observed cover (88%), thus at 0.88%. The locally rare species were defined as those with less than 5% of the maximum observed local abundance—that is, those with less than 1.5 individuals by transect for reef fish (using a log scale due to the large magnitude in observed values), less than 4.4% maximum cover for alpine plants, and less than 2.4 individuals by plot for tropical trees.
For the three datasets, we estimated species' geographic marginality. For reef fish and tropical trees, the marginality value was calculated as the mean distance from samples where the species occurred to the barycenter of the ecosystem—that is, the geographic center of all the samples. However, this method cannot be applied when the area has an irregular and concave shape since species can be close to the barycenter, thus having a low marginality value, while sampled on the edge. This was the case for the alpine geographic domain. As an alternative, we considered the mean distance from samples where the species occurred to the closest edge of the domain as a measure of geographic marginality. To compare geographic marginality values across ecosystems, those values were standardized by dividing species values by the maximum value observed across species from each ecosystem.
The relationship between the commonness of species over the region and their functional distinctiveness—that is, how different a species is from the other species in the assemblage in terms of ecologically significant functional traits—is triangular, with a weak relationship between the means of the two variables, and the variance of the response variable changes with values of the independent variable in all three ecosystems. Since conventional regression-correlation analyses are inappropriate for testing such relationships, we performed, in addition to classical ordinary least square regressions, quantile regressions (95th and 99th quantiles) that are able to detect constraints of an independent variable on the upper limit of a response variable while assuming a linear relationship between the maximum possible value of a response variable along the gradient of the independent variable [61].
We used the rq function from the quantreg package to build quantile regressions. Confidence intervals for each quantile regression were obtained using a kernel estimate implemented in the function summary.
To test whether rare and rarest species were disproportionately represented along the gradient in functional vulnerability, we classified species by their degree of functional vulnerability (High, [0–0.05]; Moderate, [0.05–0.25]; Low, [0.25–0.5]; Least, [0.5–1]). We chose this irregular binning to focus on species supporting the most vulnerable functions—that is, those of primary conservation concern—in the same vein as the classification of biodiversity hotspots focuses on the top 5% regions. Then we implemented a geometric series (0.25, 0.5, and 1) to define the other thresholds in order to discriminate species with a moderate degree of functional vulnerability—that is, those with a medium conservation concern—from the others—that is, with a low or very low conservation concern, without inflating the number of categories. For each level, we observed the percentage of rare and rarest species. We then randomized species among functional vulnerability levels (without replacement) to test whether the observed percentages were greater or less than expected using unilateral thresholds (5% and 95%) given the patterns observed in Figure 1.
To test whether the level of species geographic marginality was similar among functional vulnerability levels as previously defined, we first calculated the observed mean species marginality by level. We then used a first null model where marginality values were randomly distributed among all species to test whether the observed means were greater or less than expected by chance using unilateral thresholds (5% and 95%) given the patterns observed in Figure 3. Indeed, common species, which are underrepresented among species supporting highly and moderately vulnerable functions, cannot have high marginality values as present in many samples over the ecosystem, while rare species, which are overrepresented in those functional vulnerability levels, are more likely to be marginal. Since this test is highly conservative and does not account for the distribution of commonness among functional vulnerability levels, we implemented a second null model where we removed common species (those with a commonness value higher than the median) and where we shuffled marginality values among uncommon species from different functional vulnerability levels.
We provided the number of species in each category (rare, rarest, functional vulnerability levels), for each ecosystem and for each statistical test in Table S1.
|
10.1371/journal.pgen.1002918 | Loss of Axonal Mitochondria Promotes Tau-Mediated Neurodegeneration and Alzheimer's Disease–Related Tau Phosphorylation Via PAR-1 | Abnormal phosphorylation and toxicity of a microtubule-associated protein tau are involved in the pathogenesis of Alzheimer's disease (AD); however, what pathological conditions trigger tau abnormality in AD is not fully understood. A reduction in the number of mitochondria in the axon has been implicated in AD. In this study, we investigated whether and how loss of axonal mitochondria promotes tau phosphorylation and toxicity in vivo. Using transgenic Drosophila expressing human tau, we found that RNAi–mediated knockdown of milton or Miro, an adaptor protein essential for axonal transport of mitochondria, enhanced human tau-induced neurodegeneration. Tau phosphorylation at an AD–related site Ser262 increased with knockdown of milton or Miro; and partitioning defective-1 (PAR-1), the Drosophila homolog of mammalian microtubule affinity-regulating kinase, mediated this increase of tau phosphorylation. Tau phosphorylation at Ser262 has been reported to promote tau detachment from microtubules, and we found that the levels of microtubule-unbound free tau increased by milton knockdown. Blocking tau phosphorylation at Ser262 site by PAR-1 knockdown or by mutating the Ser262 site to unphosphorylatable alanine suppressed the enhancement of tau-induced neurodegeneration caused by milton knockdown. Furthermore, knockdown of milton or Miro increased the levels of active PAR-1. These results suggest that an increase in tau phosphorylation at Ser262 through PAR-1 contributes to tau-mediated neurodegeneration under a pathological condition in which axonal mitochondria is depleted. Intriguingly, we found that knockdown of milton or Miro alone caused late-onset neurodegeneration in the fly brain, and this neurodegeneration could be suppressed by knockdown of Drosophila tau or PAR-1. Our results suggest that loss of axonal mitochondria may play an important role in tau phosphorylation and toxicity in the pathogenesis of AD.
| Abnormal phosphorylation and toxicity of a microtubule-associated protein tau are involved in the pathogenesis of Alzheimer's disease (AD). Tau is phosphorylated at multiple sites, and phosphorylation of tau regulates its microtubule binding and physiological functions such as regulation of microtubule stability. Abnormal phosphorylation of tau occurs in the AD brains and is thought to cause tau toxicity; however, what pathological conditions trigger abnormal phosphorylation and toxicity of tau in AD is not fully understood. Since a reduction in the number of mitochondria in the axon has been observed in the AD brains, we investigated whether and how loss of axonal mitochondria promotes tau phosphorylation and toxicity. Using transgenic flies expressing human tau, we found that knockdown of milton or Miro, an adaptor protein essential for axonal transport of mitochondria, enhanced human tau-induced neurodegeneration. This study demonstrates that loss of axonal mitochondria caused by milton knockdown increases tau phosphorylation at an AD–related site through partitioning defective-1 (PAR-1), promotes detachment of tau from microtubules, and enhances tau-mediated neurodegeneration. Our results suggest that loss of axonal mitochondria may play an important role in tau phosphorylation and toxicity in the pathogenesis of AD.
| Mitochondria are principal mediators of local ATP supply and Ca2+ buffering. In neuronal axons, these requirements need to be addressed locally, and the proper distribution of mitochondria is essential for neuronal functions and survival [1]. Defects in mitochondrial distribution have been observed in the brains of patients suffering from several neurodegenerative diseases including Alzheimer's disease (AD) [2]. Recent studies have shown that the localization of mitochondria to the axon is reduced in neurons in the AD brain, as well as in cellular and animal models of AD [3]–[14]. The reduction in mitochondria in the axon may be due to alterations in mitochondrial fission/fusion [3], [5], [6] and/or due to defects in the axonal transport of mitochondria [4], [6], [8], [10]–[12]. However, how it contributes to the pathogenesis of AD remains elusive.
Tau is a microtubule-associated protein that is expressed in neurons and localizes predominantly in the axons, where it regulates microtubule dynamics. Tau is phosphorylated at a number of sites, and a fine-tuned balance between phosphorylation and dephosphorylation of tau is critical for its physiological functions, such as microtubule stabilization, in the axons [15].
Hyperphosphorylated tau is found in neurofibrillary tangles, the intracellular protein inclusions that are associated with a range of neurodegenerative diseases including AD [15]. In AD brains, tau phosphorylation is abnormally increased at several specific sites, and these changes are associated with tau toxicity [15], [16]. However, the effects of loss of axonal mitochondria on abnormal phosphorylation and toxicity of tau has not been fully elucidated.
Mitochondrial transport is regulated by a series of molecular adaptors that mediate the attachment of mitochondria to molecular motors [17]. In Drosophila, mitochondrial transport is facilitated by milton and Miro, which regulate mitochondrial attachment to microtubules via kinesin heavy chain [18], [19]. In mammals, two isoforms of milton (OIP106 and GRIF1) and Miro (Miro1 and Miro2) have been identified and are proposed to act in a similar manner [20]. In Drosophila, in the absence of milton or Miro, synaptic terminals and axons lack mitochondria, although mitochondria are numerous in the neuronal cell body [18], [21].
In this study, using Drosophila as a model system, we investigated the effects of knockdown of milton or Miro, an adaptor protein essential for axonal transport of mitochondria, on tau phosphorylation and toxicity. We demonstrate that loss of axonal mitochondria caused by milton knockdown increases tau phosphorylation at Ser262 through PAR-1, promotes detachment of tau from microtubules, and enhances tau-mediated neurodegeneration.
To test whether loss of axonal mitochondria enhances human tau toxicity in vivo, we used transgenic Drosophila expressing human tau [22]. Wild-type human 0N4R tau, which has four tubulin-binding domains (R) and no N-terminal insert (N), was expressed in fly eyes using the GAL4/UAS system [23] with the pan-retinal gmr-GAL4 driver. Expression of human tau causes age-dependent and progressive neurodegeneration in the lamina, the first synaptic neuropil of the optic lobe containing photoreceptor axons: degeneration in the lamina is undetectable or very mild in flies at 3-day-old, while it is prominent at 10-day-old (Figure S1A and S1B; S1D, quantification).
It has been reported that overexpression of tau alone can reduce anterograde transport of a variety of kinesin cargos, including mitochondria [4], [9], [11], [12], [24]. We examined whether tau expression alone causes the loss of mitochondria at the synaptic terminals of young tau flies by electron microscopy. Mitochondria were observed in the synaptic terminals of photoreceptor neurons expressing tau at 3-day-old (Figure S2), suggesting that, under our experimental conditions, severe defects in microtubule-dependent transport are not occurred in the young flies expressing human tau.
Milton is a component of an adaptor complex that links mitochondria to kinesin heavy chain and is essential for axonal transport of mitochondria (Figure 1A) [19]. Previously, we have shown that milton RNAi expression effectively reduces milton protein levels, reduces the axonal distribution of mitochondria and increases the mitochondrial localization to the cell body in the fly brain [7]. We confirmed that expression of milton RNAi in fly eyes caused loss of mitochondria in the synaptic terminals of the photoreceptor neurons by electron microscopy analysis. Mitochondria were seldom observed in the synaptic terminals of photoreceptor neurons expressing milton RNAi, while mitochondria were abundant in the synaptic terminals of control flies (compare Figure 1B and 1C). In addition, the presynaptic terminals contained vesicles with a wider range of sizes in the milton knockdown flies than in controls (Figure S3), as previously observed in milton mutant flies [25].
To investigate tau toxicity under the condition in which mitochondria are chronically depleted from the axon, we co-expressed milton RNAi with human tau. We confirmed that milton knockdown caused loss of axonal mitochondria in the neurons expressing tau by electron microscopy (Figure S4). Co-expression of tau with milton RNAi (milton RNAiGD) dramatically enhanced neurodegeneration in the lamina at 3-day-old compared to fly eyes expressing tau alone (Figure 1E and 1F; 1M, quantification). In 3-day-old flies, knockdown of milton alone did not cause neurodegeneration (Figure 1H) [21]. To limit the possibility of off-target effects of RNAi, another independent transgenic fly line carrying a milton RNAi that targets a different region of milton (milton RNAiTRiP) was used. Expression of this RNAi in neurons reduced milton mRNA levels in the fly brain (Figure S5A) as well as the axonal distribution of mitochondria (Figure S5B). The enhancement of tau-induced neurodegeneration was also observed with milton RNAiTRiP (Figure 1G; 1M, quantification).
We also tested the effect of knockdown of Miro, which is another critical component of the adaptor complex that controls mitochondrial trafficking in the axons [18], [19] (Figure 1A), on tau-mediated neurodegeneration. Expression of Miro RNAi (Miro RNAiKK) [26] reduced the axonal distribution of mitochondria in the fly brain (Figure S6) and significantly enhanced tau-induced neurodegeneration (Figure 1I; 1M, quantification). To limit the possibility of off-target effects of RNAi, we generated another independent transgenic fly line carrying Miro RNAi (Miro RNAiiai) that targets a different region of Miro. Expression of Miro RNAiiai reduced Miro mRNA levels (Figure S7) and significantly enhanced tau-induced neurodegeneration (Figure 1J; 1M, quantification). Similar to a previous report [21], knockdown of Miro alone did not cause neurodegeneration in 3-day-old flies (Figure 1K).
The enhancement of tau-induced neurodegeneration by milton RNAi or Miro RNAi is not due to non-specific effects of RNAi overexpression, since the expression of an RNAi against firefly luciferase (Figure 1L), as well as the expression of many other RNAis (Figure S8), did not enhance tau-induced neurodegeneration.
Expression of human tau in Drosophila eyes reduces the external eye size (Figure S9B), which is due to apoptosis during the larval stage [27]. Genetic screens assessing changes in this phenotype have identified a number of modifiers of tau toxicity [28]–[30]. However, neither milton nor Miro was identified in the previous screens [28]–[30]. We found that knockdown of either milton or Miro did not enhance the tau-induced reduction in external eye size (Figure S9C and S9D). These results indicate that the modifier screen using tau-induced lamina degeneration as a read-out phenotype yields new genes involved in tau-induced neurodegeneration.
Taken together, these results demonstrate that the knockdown of milton or Miro enhances human tau-mediated neurodegeneration.
Neurodegeneration in the lamina in flies expressing tau alone (Figure 2A) or expressing tau and milton RNAi (Figure 2B, 2E, 2F and 2G) at 3-day-old was examined at the ultrastructural level. Axon pathology, including the formation of vacuoles in the axons (asterisks in Figure 2A, 2B and 2E) and swollen axons (arrows in Figure 2E), were observed. In the presynaptic terminals, vacuoles (Figure 2F, asterisks) and the accumulation of autophagic bodies and multivesicular bodies (Figure 2F and 2G, arrows) were observed. These pathological changes were more severe and prominent in the lamina of flies expressing tau and milton RNAi than in flies expressing tau alone. Neurofibrillary tangles were not detected in flies expressing human tau and milton RNAi, indicating that milton knockdown enhances the tau-induced axonopathy without the formation of large tau aggregates. Axonal or presynaptic degeneration was not observed in the control flies (Figure 2C and 2H) or in flies with milton knockdown alone at 3-day-old (Figure 2D and 2I).
A group of Ser/Thr phosphorylation sites in tau is abnormally phosphorylated in the AD brain [31]. Using well-characterized phospho-tau-specific antibodies, we examined whether milton knockdown affects tau phosphorylation at AD-related sites by Western blotting. The level of tau phosphorylated at Ser262 was significantly increased by milton knockdown (Figure 3A). Knockdown of Miro also increased the levels of tau phosphorylated at Ser262 (Figure 3B). In contrast, tau phosphorylation at the AT8 epitope (phospho-Ser202) or the AT180 epitope (phospho-Thr231) was not significantly altered by milton knockdown (Figure 3A). The levels of total tau were not significantly changed (Figure 3A), indicating that milton knockdown does not cause tau accumulation.
Tau phosphorylation at Ser262 has been reported to reduce tau binding to microtubules [32], [33]. We tested whether milton knockdown alters the binding of tau to microtubules by using microtubule binding assay. Microtubules and microtubule-bound proteins were recovered as the pellet by centrifugation from brains of flies expressing human tau alone or co-expressing human tau and milton RNAi. The pellet (microtubule-bound fraction) and supernatant (microtubule-free fraction) were separated by SDS-PAGE, and tau levels in these fractions were analyzed by Western blotting. Milton knockdown caused a significant reduction in the amount of tau in the pellet and an increase in tau in the supernatant (Figure 3C). This result indicates that milton knockdown reduces tau binding to microtubules and increases the levels of microtubule-unbound, free tau in the fly brain.
Drosophila partitioning defective-1 (PAR-1) and the mammalian homolog of PAR-1, microtubule affinity-regulating kinase (MARK), are reported to phosphorylate tau at Ser262 in vivo [34], [35]. RNAi-mediated knockdown of PAR-1 in fly eyes caused a significant reduction in tau phosphorylation at Ser262, suggesting that PAR-1 is the major Ser262 kinase in the fly eye (Figure 4A) [35]. We examined whether blocking PAR-1 activity suppresses the increase in tau phosphorylation at Ser262 caused by milton knockdown. In the PAR-1 knockdown background, milton knockdown did not increase tau phosphorylation levels at Ser262 (Figure 4B), indicating that PAR-1 mediates the increase in tau phosphorylation at Ser262 caused by milton knockdown.
We investigated the role of tau phosphorylation at Ser262 in the enhancement of tau-induced axon degeneration caused by milton knockdown. We first examined whether PAR-1 knockdown enhances or suppresses tau-induced axon degeneration in the milton knockdown background. RNAi-mediated knockdown of PAR-1 significantly suppressed neurodegeneration in the lamina of flies expressing human tau and milton RNAi (Figure 5A and 5B; 5D, quantification). This effect is not due to titration of the effectiveness of RNAi, since the expression of an RNAi against firefly luciferase did not significantly suppress tau-induced neurodegeneration in the milton knockdown background (Figure 5A and 5C; 5D, quantification).
Next, to determine whether the Ser262 site is required for the knockdown of milton to enhance tau-induced axon degeneration, transgenic flies carrying human tau with the S262A mutation (S262A tau) expressed at the levels similar to the expression of wild-type tau ([36] and Figure 5E) were used. It has been reported that introduction of the S262A mutation dramatically rescues tau-induced reduction in external eye size [35], [36], which is due to apoptosis during the larval stage [27]. Interestingly, we found that expression of S262A tau caused age-dependent neurodegeneration in the lamina similar to that caused by wild type tau: in the flies expressing S262A tau, degeneration in the lamina was undetectable or very mild in flies at 3-day-old, while it was prominent at 10-day-old (Figure S10). Using S262A tau flies, we found that the introduction of the S262A mutation suppressed the enhancement of tau-induced axon degeneration caused by milton knockdown (Figure 5F and 5G; 5H, quantification). Taken together, these results suggest that tau phosphorylation at Ser262 and PAR-1 play a critical role in the enhancement of tau-induced neurodegeneration caused by milton knockdown.
Our results demonstrate that knockdown of milton or Miro enhances tau-induced neurodegeneration and increases tau phosphorylation at Ser262. PAR-1 mediates the increase in tau phosphorylation at Ser262, and tau phosphorylation site Ser262 and PAR-1 are critical for the enhancement of tau-induced neurodegeneration caused by milton knockdown. To further investigate the relationship between loss of axonal mitochondria and PAR-1, the effect of knockdown of milton or Miro on PAR-1 activity was examined.
To detect active PAR-1, a phospho-specific antibody that recognizes phosphorylated Thr408 of PAR-1, which is important for PAR-1 activity [37], was used. The titer of the antibody is not sufficient to detect endogenous PAR-1 in fly eyes [37], but the antibody recognizes the active form of PAR-1 when PAR-1 is overexpressed [37]. Co-expression of milton RNAi increases the levels of Thr408-phosphorylated PAR-1 in the fly eye (Figure 6A). In addition to the levels of Thr408-phosphorylated PAR-1, we observed an increase in total PAR-1 levels with knockdown of milton (Figure 6A). Similar results were obtained with another milton RNAi line (milton RNAiTRiP) (Figure 6B). Furthermore, expression of Miro RNAi also caused an increase in the levels of Thr408-phosphorylated PAR-1 as well as total PAR-1 in the fly eyes (Figure 6C). These effects are not due to non-specific effect of RNAi expression, since the expression of an RNAi against firefly luciferase did not increase either the levels of Thr408-phosphorylated PAR-1 or total PAR-1 (Figure 6D).
A previous report showed that total PAR-1 level increased when it was phosphorylated at Thr408 in Drosophila [37]. To examine whether the increase in PAR-1 levels with milton knockdown is Thr408-dependent, transgenic flies carrying PAR-1 with unphosphorylatable alanine mutation at Thr408 (PAR-1 T408A) [37] were used. Milton RNAi coexpression did not increase PAR-1 T408A protein levels (Figure 6E), indicating that Thr408 is important for the increase in PAR-1 levels caused by milton knockdown.
Milton knockdown did not cause non-specific activation of kinases, since it did not increase the level of phosphorylated, active p44mapk in fly eyes (Figure S11). Moreover, milton knockdown did not cause non-specific accumulation of overexpressed proteins, or an increase in the expression of genes under the control of GAL4/UAS system, since it did not increase the levels of total p44mapk, GFP, or amyloid precursor protein expressed via the GAL4/UAS system (Figure S11). Taken together, these results demonstrate that milton knockdown specifically increases the level of active PAR-1.
We also observed that the phenotype induced by PAR-1 overexpression was enhanced by milton knockdown. Overexpression of PAR-1 in fly eyes has been reported to cause eye degeneration [35], and we found that overexpression of PAR-1 caused age-dependent, mild neurodegeneration in the lamina: neurodegeneration in the lamina is undetectable in flies overexpressing PAR-1 at 3-day-old, while it is detectable at 10-day-old (Figure S12A–S12C, S12D, quantification). Co-expression of PAR-1 and milton RNAi caused prominent neurodegeneration in the lamina at 3-day-old, when neither PAR-1 overexpression alone or knockdown of milton alone caused neurodegeneration (Figure S12, S12E–S12G; S12H, quantification). These phenotypic analyses further suggest that milton knockdown increases PAR-1 activity in the eye.
Although knockdown of milton or Miro without human tau overexpression did not cause neurodegeneration in the young flies (Figure 1H and 1K), we found that knockdown of milton or Miro alone caused late-onset neurodegeneration in the fly brain. Expression of milton RNAi in the fly eyes and brains with a combination of two GAL4 drivers, the pan-retinal gmr-GAL4 driver and pan-neuronal elav-GAL4 driver, caused age-dependent neurodegeneration (Figure 7A, 21 days-after-eclosion (day-old)). Interestingly, although milton RNAi was expressed in the all neurons in the eye and brain, degeneration was the most prominent in the optic lobe (Figure 7A). No degeneration was observed in the brain in flies expressing an RNAi against firefly luciferase at the same age (Figure 7B, control).
We quantified age-dependent progression of neurodegeneration caused by milton knockdown in the lamina, where neurodegeneration was the most prominent. Degeneration in the lamina is undetectable at 3-day-old, which is in line with the previous observation in milton mutant flies [25]. In contrast, degeneration was observed at 20-day-old and was more prominent at 30-day-old, indicating that milton knockdown causes late-onset and progressive neurodegeneration (Figure 7, compare 7C (3-day-old), 7D (20-day-old) and 7E (30-day-old); G, quantification of vacuole areas). No degeneration was observed in the lamina in control flies at 30-day-old (Figure 7F).
To limit the possibility of off-target effects of RNAi, another independent transgenic fly line carrying a milton RNAi that targets a different region of milton (milton RNAiTRiP) was used. Expression of milton RNAiTRiP also caused age-dependent neurodegeneration in the lamina (Figure 7H–7J; 7K, quantification of vacuole areas). Knockdown of Miro in the fly brains also caused neurodegeneration in the lamina in aged flies (Figure 7L–7M; 7N, quantification of vacuole areas). Collectively, these results suggest that loss of axonal mitochondria is sufficient to cause late-onset neurodegeneration.
While our paper is under review, it was reported that knockdown of milton led to progressive axon degeneration in the Drosophila wing neurons [38].
The Drosophila tau exhibits a high degree of similarity with the human tau protein and shares a numbers of important features such as the microtubule-binding domains [39] and several phosphorylation sites critical for its functions and toxicity [40]. Overexpression of Drosophila tau has been reported to be capable of inducing neuronal dysfunction and neurodegeneration [41]–[43]. We examined whether Drosophila tau is involved in neurodegeneration caused by milton knockdown. We found that expression of RNAi targeting Drosophila tau reduced tau mRNA levels in the fly brain (Figure S13) and significantly suppressed neurodegeneration in the lamina caused by milton knockdown (Figure 8A and 8B; 8D, quantification). This effect is not due to titration of the effectiveness of RNAi, since the expression of an RNAi against firefly luciferase did not suppress neurodegeneration caused by milton knockdown (Compare Figure 8A and 8C; 8D, quantification). We further examined whether PAR-1 is involved in neurodegeneration caused by milton knockdown. RNAi-mediated knockdown of PAR-1 significantly suppressed neurodegeneration in the lamina of flies expressing milton RNAi (Figure 8E and 8F; 8G, quantification). These results suggest that Drosophila endogenous tau and PAR-1 contribute to milton knockdown-induced neurodegeneration and further support the conclusions of this study.
Abnormal phosphorylation and toxicity of tau are thought to play a critical role in the pathogenesis of Alzheimer's disease (AD). Accumulation of amyloid-β peptides is thought to be causative for AD and has been suggested to cause tau abnormality [44]–[53], however, the underlying mechanisms are not clear. A reduction in the number of mitochondria in the axon is observed in the brains of AD patients [3], and we and others previously reported that amyloid-β peptides reduce the number of mitochondria in the axons [3]–[11]. In this study, we examined whether and how loss of axonal mitochondria increases phosphorylation of human tau at AD-related sites and enhances tau toxicity. Our data demonstrate that loss of axonal mitochondria caused by knockdown of milton or Miro increases tau phosphorylation at an AD-related site Ser262 through PAR-1, promotes detachment of tau from microtubules, and enhances tau-mediated neurodegeneration. These results suggest that loss of axonal mitochondria may play an important role in tau phosphorylation and toxicity in the pathogenesis of AD.
It has been reported that, in non-neuronal cultured cells or primary-cultured hippocampal neurons with virus-mediated overexpression of human tau, an excess of microtubule-bound tau blocks microtubule-dependent transport of vesicles and organelle including mitochondria and causes synaptic degeneration [54], [55]. These works have demonstrated that tau phosphorylation at Ser262 by a PAR-1 homolog MARK2 removes tau from the microtubule tracks, which restores microtubule-dependent transport of vesicles and organelle, and rescues accompanied synaptic degeneration [54]. Thus, tau phosphorylation at Ser262 plays a protective role against tau-induced toxicity in their models in which an excess of microtubule-bound tau blocks traffic of vesicles and organelle.
In contrast, this study examined whether and how specific loss of axonal mitochondria promotes tau phosphorylation and toxicity. To address this question in vivo, we used a Drosophila model of human tau toxicity [22]. In this model, we did not observe severe defects in microtubule-dependent transport under our experimental condition, since mitochondria are present at the synaptic terminals of neurons expressing human tau in the young flies (Figure S2). To chronically deplete mitochondria from the axon, we used knockdown of milton or Miro. The most critical difference between the models used in Thies and Mandelkow [54] and our model is that, in the models of Thies and Mandelkow, mitochondrial transport defects were depending on tau binding to microtubules, while in our model, mitochondria were chronically depleted from the axon by milton knockdown.
Using this model system, we found that milton knockdown significantly enhanced tau-mediated neurodegeneration. Milton knockdown increased the levels of active PAR-1 and tau phosphorylation at Ser262, and promoted detachment of tau from microtubules. If the enhancement of tau toxicity caused by milton knockdown in our model is due to an additive reduction in the number of axonal mitochondria, blocking tau phosphorylation at Ser262, which increases tau binding to microtubules and blocks microtubule-dependent transport, would enhance neurodegeneration. However, our results have shown that blocking tau phosphorylation at Ser262 by PAR-1 knockdown rescues the enhancement of tau-mediated neurodegeneration in the milton knockdown background. These results suggest that the enhancement of tau toxicity in the milton knockdown background is not likely to be due to an additive reduction of axonal transport of mitochondria caused by an excess of microtubule-bound tau. Rather, this study suggests that, when axonal mitochondria are chronically depleted, increased free, microtubule-unbound, Ser262-phosphorylated tau promotes neurodegeneration.
A fine-tuned balance of microtubule-binding of tau is critical for its physiological functions. It has been suggested that both an excess of microtubule-bound tau and an excess of free, microtubule-unbound tau can cause toxicity [56]. Since tau phosphorylation at Ser262 promotes tau detachment from microtubules [32], [33], Ser262 phosphorylation by MARK/PAR-1 plays critical roles under both physiological and pathological conditions. Thus, when axonal distribution of vesicles and organelle are reduced by tau, detachment of tau can play a protective role by temporarily enhancing microtubule-dependent transport [54]. However, our results suggest that, under pathological environments in which axonal mitochondria are chronically depleted, microtubule-unbound, free, Ser262-phosphorylated tau in the axons may become toxic and cause neurodegeneration.
We found that the levels of active PAR-1 are increased by milton knockdown (Figure 6). PAR-1 is activated by various stress stimuli such as high osmolarity and amyloid precursor protein accumulation in Drosophila [37]. Our results suggest that loss of axonal mitochondria may trigger a process that increases the levels of active PAR-1. However, the detailed mechanisms by which milton knockdown increases the levels of active PAR-1 require further investigations. PAR-1 activity is regulated by various kinases including LKB1, aPKC, and Death-associated protein kinase (DAPK) [37], [57], [58]. A recent report demonstrated that PAR-1 protein levels were regulated by the Drosophila homolog of adducin, a cytoskeletal protein involved in regulating actin filament dynamics [59]. Milton knockdown may act through one or a combination of the mechanisms to increase the level of active PAR-1.
Detachment of tau from microtubules has been suggested to initiate its abnormal metabolism and toxicity of tau in AD pathogenesis [56], however, the underlying mechanisms are not fully understood. This study shows that loss of axonal mitochondria promotes detachment tau from microtubules and enhances tau-mediated neurodegeneration, and tau phosphorylation at AD-related Ser262 by PAR-1 plays a critical role in this process. Our results also suggest that an increase in Ser262-phosphorylated, microtubule-unbound tau may contribute to neurodegeneration under pathological conditions in which axonal mitochondria is depleted. An important question is how free, microtubule-unbound, Ser262-phosphorylated tau causes neurodegeneration under such pathological conditions. Loss of axonal mitochondria would disrupt multiple signaling pathways in the axon, and those changes may further enhance toxicity of tau. Elucidation of such mechanisms will further our understanding of tau-mediated neurodegeneration in the pathogenesis of AD.
In summary, this study highlights a potential role of loss of axonal mitochondria in tau phosphorylation and toxicity in AD pathogenesis. Reductions in the function and number of mitochondria in the axon have also been implicated in several neurodegenerative diseases such as Parkinson's disease, Huntington's disease, and amyotrophic lateral sclerosis [60]. Our study raises an interesting possibility that mitochondrial mislocalization may cause abnormal metabolism and toxicity of other disease-related, aggregation-prone proteins.
Transgenic fly lines carrying UAS-luciferase RNAi were established following the method described previously [61]. Target sequences were amplified by PCR from luciferase cDNA using primers (for, 5′-CCGGAATTCGATATGGGCTGAATACAAATCACAGAATCG-3′, rev, 5′-CTAGTCTAGATTCATTAAAACCGGGAGGTAGATGAGATGT-3′ ), and the resulting constructs were subcloned into the pUAST Drosophila transformation vector and microinjected into fly embryos of the w1118 genotype. Transgenic fly lines carrying UAS-Miro RNAi were established by microinjecting the Miro RNAi construct (a kind gift from Dr. Barry Dickson (Research Institute of Molecular Pathology, Austria)) into fly embryos of the w1118 genotype. The transgenic fly lines carrying S262A mutant tau was described previously [36]. Other fly stocks are listed in Table S1. Fly genotypes used in each experiment are listed in Table S2.
Probosces were removed from decapitated heads, which were then immersion-fixed overnight in 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M sodium cacodylate buffer at 4°C. Samples were post-fixed 1 hr in 1% osmium tetroxide in 0.1 M sodium cacodylate buffer on ice. After washing, samples were stained en bloc with 0.5% aqueous uranyl acetate for 1 hr, dehydrated with ethanol and embedded in Epon. Thin-sections (70 nm) of laminas, in which photoreceptor axons were cut longitudinally, were collected on copper grids. The sections were stained with 2% uranyl acetate in 70% ethanol and Reynolds' lead citrate solution. Electron micrographs were obtained with a VELETA CCD Camera (Olympus Soft Imaging Solutions GMBH) mounted on a JEM-1010 electron microscope (Jeol Ltd.).
Preparation of paraffin sections, hematoxylin and eosin staining, and analysis of neurodegeneration were described previously [53]. To analyze internal eye structure, heads of female flies were fixed in Bouin's fixative (EMS) for 48 hr at room temperature, incubated 24 hr in 50 mM Tris/150 mM NaCl, and embedded in paraffin. Serial sections (6 µm thickness) through the entire heads were prepared, stained with hematoxylin and eosin (Vector), and examined by bright-field microscopy. Images of the sections that include the lamina were captured with Insight 2 CCD Camera (SPOT), and vacuole area was measured using Image J (NIH). Heads from more than five flies (more than 10 hemispheres) were analyzed for each genotype.
Twenty fly heads for each genotype were homogenized in SDS-Tris-Glycine sample buffer, and the same amount of the lysate was loaded to each lane of multiple 10% Tris-Glycine gels and transferred to nitrocellulose membrane. The membranes were blocked with 5% milk (Nestle), blotted with the antibodies described below, incubated with appropriate secondary antibody and developed using ECL plus Western Blotting Detection Reagents (GE Healthcare) or imaging with an Odyssey system. One of the membranes was probed with anti-tubulin, and used as the loading control for other blots in each experiment. Anti-tau monoclonal antibody (Tau46, Zymed), and anti-tau pSer262 (Biosource and Calbiochem), phospho-Thr231 (AT180, Thermo and Endogen), anti-HA (Santa Cruz), anti-myc (Millipore), anti-active p44mapk (Promega), anti-tubulin (Sigma), anti-GFP (Clontech) were purchased. Anti-tau pS202 (CP13) was a kind gift from Dr. Peter Davis (Albert Einstein College of Medicine). Anti-PAR-1 pT408 was described previously [37]. The signal intensity was quantified using ImageJ (NIH) or an Odyssey system. Western blots were repeated a minimum of three times with different animals and representative blots are shown. Flies used for Western blotting were 3–5 day-old after eclosion.
Quantitative real-time PCR analysis was performed as described previously [53]. More than thirty flies for each genotype were collected and frozen. Heads were mechanically isolated, and total RNA was extracted using TRIzol (Invitrogen) according to the manufacturer's protocol with an additional centrifugation step (11,000× g for 10 min) to remove cuticle membranes prior to the addition of chloroform. Total RNA was reverse-transcribed using Superscript II reverse transcriptase (Invitrogen), and the resulting cDNA was used as a template for PCR on a 7500 fast real time PCR system (Applied Biosystems). The average threshold cycle value (Ct) was calculated from five replicates per sample. Expression of genes of interest was standardized relative to actin, rp49 or TBP. Relative expression values were determined by the deltaCt method according to quantitative PCR Analysis User Bulletin (Applied Biosystems). Primers were designed using Primer-Blast (NIH).
milton for 5′-GGCTTCAGGGCCAGGTATCT-3′
milton rev 5′-GCCGAACTTGGCTGACTTTG-3′
Miro for 5′-AAAAGCACCTCATTCTGCGT-3′
Miro rev 5′-CCTCAGGTGAGGAAACGCT-3′
dTau for 5′-AAGCCCGGTGGCGGTGAGAA-3′
dTau rev 5′-GCGCCAGAAGCCGTCATGGA-3′
Actin for 5′-TGCACCGCAAGTGCTTCTAA-3′
Actin rev 5′-TGCTGCACTCCAAACTTCCA-3′
rp49 for 5′-GCTAAGCTGTCGCACAAATG-3′
rp49 rev 5′- GTTCGATCCGTAACCGATGT-3′
TBP for 5′- GCGGCTGTGATTATGCGAAT-3′
TBP rev 5′-AGGGAAACCGAGCTTTTGGA-3′
Microtubule binding assay was performed using a previously reported procedure with a modification [62]. Fifty heads from adult flies expressing the human tau protein with the gmr-GAL4 driver were collected and homogenized in 150 µl of Buffer-C+ [50 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid pH 7.1, 1 mM MgCl2, 1 mM ethylene glycol tetraacetic acid, protease inhibitor cocktail (Roche), and phosphatase inhibitor cocktail (Sigma-Aldrich)] in the presence of taxol 20 µM (Sigma-Aldrich) diluted in dimethylsulfoxide. After centrifugation at 1,000× g for 10 min, aliquot of supernatant was subjected to Western blotting as the “total fraction”. The remaining supernatant was layered onto a 2 volume cushion of buffer-C+ with 50% sucrose. After centrifugation at 100, 000× g for 30 min, the upper fraction containing soluble tubulin was collected as the microtubule-free fraction and the pellet containing microtubule polymers and proteins bound to microtubules was resuspended in 150 µl of SDS-Tris-Glycine sample buffer. Protein concentration in each fraction was measured using the BCA Protein Assay Kit (Pierce). The same amount of protein was loaded to each lane of Tris-Glycine gels and analyzed by western blotting using anti-tau antibody (Tau46, Zymed) or anti-tubulin (Sigma). For quantification, the signal intensity in each lane was quantified with an Odyssey system.
Statistics was done with the JMP software (SAS) with Student's t or Tukey-Kramer HSD. Values are given as mean ± standard deviation or standard error.
|
10.1371/journal.ppat.1000449 | In Vivo Transcriptional Profiling of Listeria monocytogenes and Mutagenesis Identify New Virulence Factors Involved in Infection | Listeria monocytogenes is a human intracellular pathogen able to colonize host tissues after ingestion of contaminated food, causing severe invasive infections. In order to gain a better understanding of the nature of host–pathogen interactions, we studied the L. monocytogenes genome expression during mouse infection. In the spleen of infected mice, ≈20% of the Listeria genome is differentially expressed, essentially through gene activation, as compared to exponential growth in rich broth medium. Data presented here show that, during infection, Listeria is in an active multiplication phase, as revealed by the high expression of genes involved in replication, cell division and multiplication. In vivo bacterial growth requires increased expression of genes involved in adaptation of the bacterial metabolism and stress responses, in particular to oxidative stress. Listeria interaction with its host induces cell wall metabolism and surface expression of virulence factors. During infection, L. monocytogenes also activates subversion mechanisms of host defenses, including resistance to cationic peptides, peptidoglycan modifications and release of muramyl peptides. We show that the in vivo differential expression of the Listeria genome is coordinated by a complex regulatory network, with a central role for the PrfA-SigB interplay. In particular, L. monocytogenes up regulates in vivo the two major virulence regulators, PrfA and VirR, and their downstream effectors. Mutagenesis of in vivo induced genes allowed the identification of novel L. monocytogenes virulence factors, including an LPXTG surface protein, suggesting a role for S-layer glycoproteins and for cadmium efflux system in Listeria virulence.
| The facultative intracellular bacterial pathogen Listeria monocytogenes is the etiological agent of a severe foodborne disease. In humans it causes a variety of manifestations ranging from asymptomatic intestinal carriage and gastroenteritis to invasive and disseminated severe diseases. Septicemia, meningoencephalitis, and infection of the foetus in pregnant women are the most serious clinical features of listeriosis. Virulence is a trait that only manifests in a susceptible host, involving a highly coordinated interaction between bacterial factors and host components. This article reports the use of in vivo genome expression profiling as a powerful approach to gain a detailed understanding of the Listeria responses and the molecular cross-talk taking place in infected mice. We showed that, during infection, L. monocytogenes shifts the expression of its entire genome to promote virulence, subverting host defenses and adapting to host conditions. This first analysis of L. monocytogenes gene expression in vivo significantly enhances our understanding of the means by which intracellular pathogens promote infection.
| Listeria monocytogenes is an intracellular food-borne pathogen that causes listeriosis, an infection characterized by gastroenteritis, meningitis, encephalitis, and maternofetal infections in humans. It has one of the highest fatality rate among food-borne infections (20%–30%) [1]. Our knowledge of the infectious process in vivo mostly derives from infections in various animal models, in particular the mouse model. It is considered that bacteria after crossing the intestinal barrier reach, via the lymph and the blood, the liver and the spleen where they replicate actively. Then, bacteria via hematogenous dissemination, can reach the brain and the placenta. The disease is thus due to the original property of L. monocytogenes to be able to cross three host barriers: the intestinal barrier, the blood brain barrier and the materno-fetal barrier. It is also due to the capacity of Listeria to resist intracellular killing when phagocytosed by macrophages and to invade many non-phagocytic cell types. In the murine model, within minutes after intravenous inoculation, most bacteria can be found in the spleen and the liver [2].
L. monocytogenes ranks among the best-known intracellular pathogens and, until now, 50 genes have been shown to be involved in virulence in the mouse model (Table S1). However, whereas the different steps of the cell infectious process and the virulence factors specifically involved are well described [3], our knowledge of the in vivo infectious process is still fragmentary. Virulence is by definition expressed in a susceptible host, and involves a dynamic cross talk between the host and the pathogen. A detailed understanding of this interaction thus requires global approaches in the context of an in vivo infection. Analysis of the pathogen whole genome expression within the host should allow the identification of new bacterial genes critical for the infectious process, and lead to a better understanding of the molecular events responsible for Listeria infection.
The technology of DNA arrays allows to both study the gene content of different strains and measure gene expression levels on a genome-wide scale under different conditions. The genetic basis of L. monocytogenes pathogenicity was addressed by comparative genomics [4] and transcriptomics [5] using Listeria DNA arrays and various L. monocytogenes strains. Listeria arrays were also used for the analysis of the in vitro global gene expression of Listeria mutants for PrfA, the central regulator of virulence genes [6], and for other transcriptional regulators important for stress response and virulence (σB, σ54, HrcA, CtsR, VirR) [7]–[13]. Recently, this approach was applied to the determination of the intracellular gene expression profile of L. monocytogenes in epithelial and macrophage cell lines [14],[15]. In vivo genome profiles of other pathogens (Streptococcus pneumoniae, S. pyogenes, Mycobacterium tuberculosis, Borrelia burgdoferi, Yersinia pestis) infecting different mouse organs (dermis, soft tissue, lung, blood) were previously performed [16]. However, to our knowledge, the genome expression of a pathogen was never studied in infected mouse spleen.
Here, we present the first “in vivo” transcriptome of L. monocytogenes. We compared expression profiles of L. monocytogenes grown in standard culture medium in exponential phase vs. bacteria recovered from mouse spleens 24, 48 and 72 hours after intravenous infection. We determined the detailed expression kinetics of the complete L. monocytogenes genome in the course of the infection, and identified new Listeria virulence factors whose expression was highly up regulated in vivo.
We used the DNA macroarray technology to profile the transcriptome of Listeria during mouse infection. We used the previously described L. monocytogenes whole-genome arrays containing 500-bp-long PCR products specific for each gene [6]. Ninety-nine per cent of the 2853 predicted ORFs of the L. monocytogenes EGDe genome are represented on the arrays. They were used to analyze Listeria transcription profiles under in vitro growth in BHI in exponential phase at 37°C under aerobic conditions with shaking (pH 7) (Figure S1), and under in vivo growth conditions (mouse spleen) at 1, 2 and 3 days post intravenous infection (p.i.). Listeria present in spleen were analyzed because this organ is with the liver one of the major sites of L. monocytogenes infection. For unknown reasons, we never succeeded to prepare good quality bacterial RNAs from infected mouse livers. The time points chosen reflect key steps in the Listeria infectious process.
Culture in BHI in exponential phase at 37°C with shaking was chosen as reference conditions because BHI is the Listeria reference growth medium where bacteria divide in exponential growth phase at rates that are comparable to those observed for intracellular growth [17]. In addition, these are the in vitro reference conditions used in all previous studies analyzing the genome expression of L. monocytogenes in vitro or intracellularly [6]–[15]. However, in order to analyze the potential impact of the in vitro culture conditions used as reference on the relative gene expression in vivo, we first analyzed the results obtained comparing transcriptome from in vivo grown bacteria to transcriptomes from bacteria grown in vitro in exponential or stationary phase (Table S2). In addition, expression of known and potential virulence genes was analyzed by quantitative real time-PCR (qPCR) on RNAs extracted from bacteria cultured in BHI at 37°C in exponential or stationary growth phase, or in defined minimal medium [18], and compared to in vivo expression (Figure 1A). Results indicated that culture in exponential growth phase are closer conditions to those met by Listeria in vivo (Table S2). In addition, even if the expression of tested genes behaved differently in function of the in vitro conditions, expression of all the genes was always lower in vitro as compared to in vivo, independently of the in vitro growth conditions (Figure 1A). These experiments supported the choice of exponential growth phase in BHI as reference conditions and minimized the impact of the in vitro growth conditions on the identification of genes differentially expressed in vivo.
The reliability of the macroarray expression data was further assessed by qPCR. We selected a subset of 10 genes and performed qPCR on cDNA from bacteria grown in either standard medium or extracted from mouse spleens 48 h p.i.. qPCR results and array data exhibited a high correlation coefficient (0.7) (Figure 1B). This strong correlation was also observed for other infection time points (Figure S2). However, the differences in gene expression, as measured by qPCR, were generally higher, indicating that in vivo transcriptome data rather underestimate changes in gene expression.
The procedure used for bacterial RNA extraction from infected mouse spleens is an adaptation of the standard procedure originally used for transcriptional analysis of RNA extracted from pure culture. In order to test the effect of the RNA extraction method on gene expression, RNAs from bacteria grown in pure culture were extracted using the two methods. The relative expression of known virulence genes, cold shock genes and potential virulence genes was analyzed by qPCR in the two RNA pools. The results showed that the relative expression of the genes tested is not significantly affected by the RNA extraction procedure (Figure 1C).
For bacteria cultured in BHI at 37°C in exponential phase or extracted from infected mouse spleen at the different times p.i., two different RNA preparations from independent cultures (or infections) were used for cDNA synthesis and subsequent hybridization to two sets of arrays. To identify statistically significant differences in gene expression, we used the Statistical Analysis for Microarrays (SAM) program [19]. Subsequently, all the genes showing statistically significant changes in the expression level and an at least two-fold change in their level of expression were considered in our analysis.
Overall, a total of 568 genes representing ≈20% of the total genome exhibited a differential expression during infection as compared to growth in BHI at 37°C in exponential phase. Among these 568 genes, 457 were up regulated (≈80%) and 111 (≈20%) were down regulated during mouse infection as compared to exponential growth in BHI medium (Table S3).
In order to identify genes potentially implicated in virulence, all the genes differentially regulated in vivo were analyzed for the presence of an ortholog in the nonpathogenic close relative Listeria innocua strain CLIP11262 [20]. This analysis revealed that only 30 of the in vivo regulated genes (25 up and 5 down regulated) were absent from L. innocua (Table 1). Of these 30 genes, 20 were L. monocytogenes “specific” (i.e. also present in L. monocytogenes 1/2a F6854, L. monocytogenes 4b F2365 and H7858 [21], and absent from L. innocua). Interestingly, of these 20 genes, 16 were up regulated in vivo. Among these 16 genes, 11 have been previously implicated in Listeria virulence. The remaining 10 in vivo regulated genes, among which 9 up- and 1 down-regulated in vivo, appeared lineage specific, i.e. present only in the sequenced serovar 1/2a strains (Table 1).
To identify genes regulated during different stages of listeriosis, gene expression levels of spleen-recovered bacteria at different time points p.i. were compared. This analysis revealed a core regulon of 106 genes (68 up and 38 down regulated) whose expression was significantly differentially regulated at all the time points of the infection as compared to bacteria grown in pure culture (Figure 2). No gene appeared specifically differentially regulated at 24 h p.i. At two days p.i., a large proportion (245/457) of genes was up regulated. The largest number of down regulated genes was observed 72 h p.i..
As compared to Listeria grown in BHI at exponential phase, bacteria extracted from mouse spleens showed a differential expression of genes belonging to various functional categories (Figure 3). In particular, analysis of the expression profile of the 50 genes previously implicated in Listeria virulence in the mouse model revealed that 29 were up regulated during infection, and two (stp and fbpA) down regulated in vivo (Figure 4 and Table S1).
We observed that the entire virulence gene cluster of L. monocytogenes comprising the genes prfA, plcA, hly, mpl, actA and plcB was highly activated during the 3 first days of infection (Table 2). In addition to the virulence gene cluster, genes encoding the two major L. monocytogenes factors implicated in entry into eukaryotic cells (inlA and inlB) [22], and uhpT, a gene encoding a sugar phosphate transporter that mediates rapid intracellular proliferation [23] were also activated during infection. PrfA is the principal regulator of the expression of not only these key virulence genes, but also of most other L. monocytogenes genes involved in intracellular survival and virulence [6]. The 12 genes previously reported to be preceded by a PrfA box and positively regulated by PrfA in a transcriptional analysis of the PrfA regulon [6], were all highly up regulated in mouse spleens (Table 2). From the 53 other genes already shown as positively regulated by PrfA [6], 20 were more expressed in vivo. As previously shown [8], 19 of these 20 genes are also controlled by SigB, including the LPXTG internalin-like protein inlH known to be involved in Listeria virulence [24]. Two genes, lmo0206 and lmo0207, recently shown as regulated by PrfA and implicated in L. monocytogenes intracellular survival [14] were also activated in infected mice. Importantly, no gene previously shown under the PrfA positive regulation appeared down regulated during mouse infection.
VirR, another key Listeria virulence regulator that mainly controls genes involved in the modification of bacterial surface components, is the response regulator of a two-component system (TCS) implicated in cell invasion and virulence [13]. Using a transcriptomic approach, 17 genes were previously identified as regulated by VirR in vitro [13]. In our study, 13 of the 17 VirR regulated genes, including the dlt operon and mprF, were up regulated in vivo (Table 3). The dlt operon is necessary for D-alanylation of lipoteichoic acid (LTA) and was reported to be important for L. monocytogenes virulence [13]. The VirR regulated mprF encodes a protein shown to be required for lysinylation of phospholipids in the Listeria cytoplasmic membrane and to confer Listeria resistance to cationic antimicrobial peptides (CAMPs) [25]. The virR and virS genes were themselves up regulated, constituting the only TCS whose expression of both components was induced in mouse spleens.
In addition to VirRS, the L. monocytogenes genome contains 15 additional predicted TCS systems [26]. Genes encoding one component of three of these TCS (degU, resD and phoR) were also up regulated in vivo. DegU is an orphan response regulator (absence of the sensor kinase DegS in the L. monocytogenes genome) and a pleiotropic regulatory system previously involved in Listeria virulence [27],[28]. In particular, DegU has been implicated in the regulation of some Listeria secreted proteins (gap, tsf, sod, lmo0644) [26]. Interestingly, the expression of these four genes was also increased in mouse spleens.
Finally, OhrR a transcriptional regulator controlling OhrA, a hydroxyperoxidase implicated in intracellular survival of Listeria [14], as well as several predicted transcriptional regulators were up regulated in infected mouse spleens.
In addition to genes already mentioned and involved in LTA modification (dltABCD), we observed that several genes implicated in peptidoglycan (PG) biosynthesis (lmo0516, lmo0540, lmo1438, lmo1521, lmo1855, lmo2522, lmo2526 and pbpB), cell shape determination (mreBC, lmo1713), cell wall peptide synthesis (murC) were up regulated in bacteria growing in mouse spleens (Table 4).
The expression of 3 genes encoding virulence factors involved in bacterial cell wall modifications (murA, iap, and pgdA) [29]–[31] was also increased in vivo. MurA and P60, the iap gene product, are two SecA2-secreted autolysins required for Listeria full virulence [29],[30]. pgdA encodes for the PG N-deacetylase of L. monocytogenes that was demonstrated as playing an important role in virulence and evasion from host defenses [31]. In addition, spl [32] and lmo2203 are two other autolysins encoding genes up regulated in vivo, but until now never implicated in virulence.
Moreover, prsA2, a gene encoding a surface protein involved in protein folding and previously shown as implicated in Listeria intracellular survival and virulence [14],[33] was up regulated in vivo. Interestingly, the gene encoding the sortase SrtB that covalently links proteins to the Listeria peptidoglycan, and two genes encoding SrtB substrates (svpA and lmo2186) [34], were also over expressed in vivo (Table 4).
Whereas a total of 44 genes encoding potential surface proteins were up regulated in vivo, only three were observed as down regulated during infection (lspA, lmo1851 and lmo2642) (Table 5). In addition, among the 55 proteins previously identified in the cell wall subproteome of L. monocytogenes [35], we found that 23 were up regulated in vivo (Table S4). The L. monocytogenes genome encodes 41 LPXTG surface proteins [20],[36],[37]. This class includes proteins containing leucine rich repeats (LRRs) and belonging to the internalin family. Four LPXTG-protein encoding genes were up regulated in vivo. In addition to InlA and InlH, lmo1290 and lmo2714 are the two other LPXTG encoding genes activated during infection (Table 5). Four genes encoding proteins associated to the cell wall via GW modules were also more expressed in vivo: inlB, the known invasion protein [38], and lmo1521, lmo2203 and lmo2713. actA [39] was the only gene encoding a protein with a carboxyl-terminal hydrophobic tail up regulated in vivo. Genes encoding lipoproteins previously implicated in Listeria virulence (TcsA and OppA) [33],[40] or in cell invasion (LpeA) [41], were over expressed in mouse spleens. In addition, 10 genes predicted to encode other lipoproteins were activated in vivo (Table 5).
Protein secretion is of key importance in both the colonization process and virulence of Listeria [42]. Besides L. monocytogenes virulence factors with a signal peptide (ActA, LLO, InlA, InlB, InlC, InlH, Mpl, MurA, PlcA, PlcB, P60 and SvpA), three other virulence proteins (Fri, TcsA and Sod) were also found secreted in the Listeria culture supernatant [43]. All the genes encoding these secreted virulence factors appeared activated in our in vivo approach (Table S5). The analysis of the products present in the Listeria culture supernatant after growth in vitro allowed the identification of 89 additional proteins [43]. 29 of the genes encoding these secreted proteins were up regulated in vivo (Table S5). Most of the Listeria secreted proteins are presumed to be secreted through the Sec translocation system. A gene encoding one component of the predicted Sec system, secE, was observed up regulated in vivo. SecA2 is an auxiliary secretory protein required for persistent colonization of host tissues, and responsible for the secretion of several Listeria virulence factors (MurA, P60, Sod, OppA and TcsA) [29],[30],[44]. We observed an in vivo up regulation of the majority of the genes encoding SecA2-secreted proteins, including all the SecA2-secreted virulence factors (Table S5).
We observed an in vivo up regulation of several genes involved in DNA synthesis (dnaX and lmo0162), DNA restriction/modifications and repair (mutL, uvrB, lmo1639 and lmo1782), DNA recombination (recFRX, codV and lmo2267), and DNA packaging and segregation (gyrA, hup, lmo1606 and lmo2794) (Table S6). In addition, the expression of 25 genes encoding ribosomal proteins, as well as genes involved in protein synthesis initiation (infAC), elongation (fus, tsf, lmo1067) and termination (frr) was up regulated during infection. Genes encoding proteins implicated in chromosomal replication and segregation (dnaABC, ssb and divIVA), and cell elongation and division (mreBC, ftsHX and lmo0196) were also up regulated in mouse spleens (Table S6).
In our study, genes belonging to the three principal classes of stress genes were up regulated in the host. Class I genes encode classical chaperones and are controlled by the HrcA repressor. Nine of the 25 genes previously shown as HrcA repressed [11] were activated in vivo, including genes encoding the molecular chaperones DnaK and GroEL respectively also shown as induced in macrophages and required for survival following phagocytosis [45],[46] (Table S7). Inversely, 17 of the 36 genes shown to be indirectly positively regulated by HrcA [11], were up regulated in mouse spleens. This list includes genes encoding ribosomal proteins, as well as a number of DNA replication, transcription or translation related genes.
The class II stress response is mediated by sigma B (SigB). A total of 30 genes that have been recently classified as SigB activated [8] appeared here up regulated in vivo (Table S7). In particular we detected the up regulation of inlH [24], ltrC implicated in response to cold shock [47], and lmo1601 similar to general stress proteins. Interestingly, 40 genes previously classified as down regulated by SigB during the stationary growth phase [8] were detected as activated in vivo (Table S7). These include kat, a catalase involved in the oxidative stress response [48], a large proportion of genes encoding ribosomal proteins or implicated in translation, cell division and cell wall biogenesis. Furthermore, iap, the P60 gene [29], is part of this group. Finally, rsbU and rsbX, two components of the complex regulation system of SigB [49] were also up regulated in mouse spleens (Table S7).
CtsR is a transcriptional repressor involved in the control of class III stress proteins and previously shown to be responsible for the repression of 42 genes [12], 15 of which appeared up regulated in the host (Table S7). In particular, CtsR regulates the expression of Clp proteases required for the degradation of abnormal proteins and implicated in bacterial escape from macrophage vacuoles and virulence in mice [50]. Expression of clpBCE was activated during infection, as well as mcsA and mcsB the modulators of the CtsR regulon.
In some host cells, bacteria are confronted with severe oxidative stress due to the release of reactive oxygen intermediates. We observed the in vivo activation of an important number of oxidative stress resistance mechanisms. The qoxABCD operon that encodes a quinol oxidase important for oxidative stress response, and two major proteins implicated in protection against superoxides and reactive oxygen species (ROS), Kat and Sod, were highly up regulated in vivo (Table S7). Sod was previously shown as required for Listeria full virulence and is a target of Stp, a serine-threonine phosphatase also involved in L. monocytogenes virulence [44],[51], and detected down regulated in the host. A decrease in the level of Stp was previously associated to an increase in phosphorylated Sod, accompanied by the secretion of active non-phosphorylated Sod by the SecA2 system [44],[51]. Furthermore, genes encoding a thioredoxin and two thioredoxin reductases involved in the response to oxidative stress (lmo2152, trxB and lmo2390) were up regulated in our study (Table S7). The ferritin protein Fri, that also provides protection against reactive oxygen species, is essential for virulence and is required for efficient bacterial growth at early stages of the infection process [52],[53]. Fri transcription is directly regulated by Fur, the ferric uptake regulator. The expression of fri and fur was activated during infection. In addition, ohrA and gap were up regulated in vivo and encode two proteins respectively involved in hydroperoxide resistance [54] and in resistance against reactive oxygen species produced by host phagocytic cells in Leishmania [55] (Table S7).
Remarkably, 30% of the in vivo regulated genes are involved in L. monocytogenes metabolism (99 metabolism-related genes were up and 72 were down regulated) (Table S8). As described above, uhpT is an in vivo highly up regulated virulence gene, regulated by PrfA and that promotes the uptake of phosphorylated hexoses (glucose-1-phosphate and glucose-6-phosphate) [23],[56]. Phosphorylated glucose is the product of glycogen hydrolysis in eukaryotic cells and there is experimental evidence that the PrfA-dependent utilization of this compound has a role in L. monocytogenes virulence [23],[56].
We observed an in vivo up regulation of several genes encoding enzymes involved in the glycolysis, like gap, pgi, fbaA, and pgm. Inversely, we found a down regulation of the expression of four genes involved in the non-oxidative phase of the pentose phosphate pathway (lmo2660, lmo2661, lmo2662 and lmo2674). The final step of glycolysis leads to pyruvate, which is then converted to acetyl-CoA by the pyruvate dehydrogenase complex. We found this complex partly up regulated in vivo, as well as one of its activator, the lipoate ligase protein LplA2 [57],[58]. The citric acid cycle is continuously supplied with acetyl-CoA during aerobic respiration. We observed an up regulation of three citric acid cycle genes (citBCG) (Table S8). The citric acid cycle is followed by oxidative phosphorylation. In this study, we found the up regulation of several genes implicated in biosynthesis and assembly of components of the respiratory chain (menD, lmo1677, qoxABD, ctaA, cydA, cydD, atpD). In addition, genes encoding resD, a regulator of aerobic and anaerobic respiration [59] and rex, a redox-sensing transcriptional repressor [60], were also up regulated in vivo. Genes encoding the pyruvate-formate lyase (pfl) and pyruvate-formate lyase activating enzymes (pflCB) are required for the anaerobic metabolism of pyruvate and were activated in the host (Table S8).
Genes implicated in amino acid biosynthesis were also induced in vivo, in particular aroA and pheA, two genes responsible for aromatic amino acid biosynthesis. Mutations in aroA and pheA were previously shown to induce an attenuation of virulence in the mouse model [61],[62]. Furthermore, the expression of genes implicated in the biosynthetic pathways of branched amino acids (alsS, ilvN and lmo0978), and amino acids of the aspartate and glutamate families (ansB, lmo0594, lmo1006, lmo1011, lmo2413 and glnA, lmo2770, respectively), was also increased in vivo (Table S8).
Significantly, mannose (lmo0781–lmo0784), maltose (lmo0278) and cellobiose (lmo0301 and lmo0915) -specific PTS encoding genes [63] were up regulated in vivo. Inversely, fructose (lmo2733), galactitol (lmo2665) and mannitol (lmo2649) -specific PTS encoding genes appeared down regulated.
Among the genes involved in bacterial ion uptake systems, a potassium-transporting ATPase encoding gene (kdpB) was down regulated in vivo. Cobalt (lmo1207), manganese (lmo1424) and calcium (lmo0841) transporter systems were, inversely, up regulated. As indicated above, the ferritin and ferric uptake protein encoding genes, fri and fur, shown to be activated under low iron concentration [64],[65], appeared highly up regulated in vivo (Table S8).
A major goal of this work was the identification of genes that encode proteins that may play a role in the infectious process. To identify such virulence genes and in order to establish a short list, we arbitrarily used several criteria. The gene should be preferentially 1) highly activated during infection; 2) absent in the non pathogenic strain L. innocua and present in other L. monocytogenes strains from diverse serotypes; 3) a member of a specific protein family encoding gene (surface, secreted, stress) possibly involved in virulence; 4) controlled by virulence regulators (PrfA, VirR, CtsR, HrcA, SigB). Several candidates emerged, matching, at least, some of the above criteria (Table 6).
lmo0206, lmo0257, lmo0915, lmo1290 and lmo2157 are genes that, as eleven already known virulence factors, are L. monocytogenes species-specific and induced in vivo. lmo0206 and lmo2157 are the only two genes activated in vivo, controlled by PrfA, absent from L. innocua and whose role in virulence was never investigated. lmo0206, orfX [66], is in addition located at the end of the Listeria virulence cluster and was recently implicated in intracellular survival [14]. The expression of lmo2157 was shown to be controlled by PrfA and SigB [6],[8].
lmo1081, lmo1082, lmo1099 and lmo1102 are L. monocytogenes EGDe species-specific genes highly up regulated in vivo over the three time points of the infection (Table 6). Interestingly, these genes encode proteins potentially involved in cell wall metabolism and heavy metal detoxification.
Only two uncharacterized genes encoding LPXTG surface proteins (lmo1290 and lmo2714) and three encoding GW surface proteins (lmo1521, lmo2203 and lmo2713) were up regulated within the host (Table 6). lmo1521 and lmo2203 are in addition predicted autolysins. lmo2713 and lmo2714 seem to be part of a genomic region over expressed at all time points of the infection and Lmo2714 was found in the Listeria culture supernatant [43]. Four genes (lm0540, lmo1438, lmo1855 and lmo2522) predicted to be involved in cell wall metabolism were up regulated in vivo, and similar to pgdA, iap, and murA [29]–[31], could participate in Listeria infection.
Twenty-five uncharacterized genes activated in vivo encode secreted proteins that may interact with the host cells, including Lmo2201, a Tat-secreted protein [42], and GAPDH. GAPDH was previously shown to be part of the Listeria cell wall subproteome [35], and to impair Listeria phagosome maturation [67]. GAPDH seems, in addition, to be implicated in the virulence of several other pathogens [68]–[70].
lmo0788 is highly activated in mouse spleens during infection and is the only gene of the group I PrfA-regulated genes (i.e. preceded by a PrfA-box and positively regulated by PrfA) [6] whose role during infection has never been addressed (Tables 2 and 6). lmo0788 encodes a protein similar to subunits (BadFG) of the benzoyl-CoA reductase used by facultative aerobes in absence of oxygen for reductive aromatic metabolism [71].
VirR appears as the second main regulator of virulence genes and controls lmo0604, lmo1742, lmo2114, lmo2115, lmo2177 and lmo2439, whose expression was activated in the host (Tables 3 and 6). lmo2114 and lmo2115 are in addition part of a transcriptional unit co-regulated by CtsR and SigL [7].
Several stress protein encoding genes that are under the control of different stress regulators were up regulated in vivo. In particular, lmo2048 is a stress protein-encoding gene that is co-controlled by CtsR and HrcA (Table 6). The 19 genes up regulated in vivo and co-controlled by PrfA and SigB (Table 2) could also be important for the infectious process. Among these, lmo1601 and lmo1602 are furthermore regulated by SigL [7].
The use of such arbitrary criteria obviously not guaranteed that a selected gene was a virulence factor, and conversely probably excluded many virulence genes. In particular, it is worth mentioning that 91 genes encoding proteins similar to unknown proteins, and 31 encoding putative proteins with no similarity in public databases were differentially expressed in vivo (Table S3), representing a large reservoir of potential new virulence factors. Of these genes, those highly regulated all over the infectious process could be of special relevance for virulence.
In order to validate our transcriptomics approach and identify new L. monocytogenes virulence factors, 6 genes (lmo1081, lmo1082, lmo1102, lmo2713, lmo2714 and gap) were selected for mutagenesis using the criteria presented above. As we were unable to produce a gap deletion mutant (probably because GAPDH is an essential protein), we constructed a GAPDH secretion mutant.
To analyze the potential role of the selected genes in virulence, we performed intravenous inoculations of BALB/c mice with wild type (wt) and mutant strains, and the number of bacteria in the mouse liver and spleen was determined 72 h after infection (Figure 5). Mutants can be classified with respect to their virulence potential. Bacterial counts for lmo1081 and lmo2713 mutants were not significantly changed as compared to the wt strain, suggesting the non-implication of these genes in Listeria virulence in mice. For the lmo1082 mutant, bacterial counts were significantly affected (≈1 log) in mouse livers and at a lesser extent in the spleens. Interestingly, for lmo1102, lmo2714 and gap mutants we observed a remarkable decrease of bacterial counts in both mouse organs as compared to the wt. In particular, the number of bacteria was dramatically impaired in the liver 72 h after inoculation (≈2,5 to 4,5 log). The gap mutant appeared as the most attenuated mutant of our analysis with a considerable virulence decrease in both organs reaching 3,5 log in the spleen and 4,5 log in the liver as compared to the wild type (Figure 5).
In order to better characterize virulence attenuated strains, mutants for lmo1082, lmo1102, lmo2714 and gap were complemented. The corresponding wild-type gene was inserted as a single copy under the control of its own promoter on the chromosome of the mutant strain, at the PSA bacteriophage attachment site using the pPL2 integration vector [72]. Wild type, mutant and complemented strains were tested for growth in BHI at 37°C and for intracellular behavior after internalization in the murine macrophage cell line J774 (Figure 6).
The growth rate observed in BHI at 37°C for the majority of the strains tested was comparable to that of the wild type (Figure 6A). However, the gap secretion mutant exhibited an important in vitro reduced growth rate and reduced density at the stationary phase. The growth defect observed for the gap mutant was even accentuated in the complemented strain (Figure 6A). This is most probably the result of an over expression of intracellular GAPDH, expressed at the same time from the bacterial genome and from the plasmid harbored by this strain. Surprisingly, the prsA2 mutant presented also a notable growth delay. This growth defect was not mentioned in previous studies implicating PrsA2 in intracellular behavior and virulence [14],[33].
Wild type, mutant and complemented strains were also tested for intracellular behavior. As shown in Figure 6B, all the strains grew with similar multiplication rates after internalization in J774 cells, indicating that the slight growth delay observed in BHI at 37°C for some strains has no consequences on intracellular multiplication.
In addition, complemented strains were analyzed after intravenous inoculations of BALB/c mice as compared to wt and mutant strains, and the number of bacteria in the mouse liver and spleen was determined 72 h after infection (Figure 7). The virulence phenotype was restored, albeit partially in the case of lmo2714, in complemented strains, except for the gap mutant. The virulence defect of the gap complemented strain was even more severe in the spleen as compare to the corresponding mutant (Figure 7A). This was in correlation with the increased growth defect observed in BHI at 37°C for the gap complemented strain.
These results revealed a role for lmo1082 and lmo1102, and at less extent for lmo2714 and gap in Listeria virulence, validating our in vivo transcriptomics approach.
Identification of bacterial gene expression patterns during host-pathogen interactions has long been a goal for understanding infectious processes of intracellular pathogens [73]. Here, we undertook the first time course study of the L. monocytogenes in vivo transcriptome by comparing the genomic transcriptional patterns of bacteria grown under laboratory conditions (BHI, 37°C, exponential growth phase, pH 7) with that of in vivo-grown bacteria over three days of infection (mouse spleen). This constitutes also the first genome expression analysis of a pathogen in infected mouse spleens. Our results indicate that a significant part of the Listeria genome is differentially expressed for adaptation to the host environment, essentially through gene activation. We showed an in vivo over expression of an impressive number of genes involved in virulence and subversion of the host immune systems, together with genes involved in adaptation of the bacterial metabolism to host conditions and stress responses. We revealed that all these expression modifications are controlled by a complex regulatory network.
Whereas metabolic genes represent only 22% of the in vivo activated genes, they constitute the major part of the down regulated genes (65%). These observations reveal that the modification of the Listeria genome expression during infection is dominated by the activation of virulence specific genes.
A major finding of our study was the demonstration that the majority of genes previously reported as implicated in virulence were highly up regulated in the host. We observed a peak of activation for these genes 48 h p.i., preceding the peak of bacterial loads that occurs at 72 h p.i. when mice are intravenously inoculated with a sub-lethal dose. Our results support the idea that Listeria uses in vivo a complex and coordinated regulatory network that includes virulence and stress regulators in order to tightly control genes that contribute to its survival and progression of the disease. Our study definitively establishes PrfA as the major Listeria virulence regulator and VirR as the second one. These two regulators, as well as a large proportion of the genes they regulate, including known virulence factors and potentially new virulence genes, were strongly activated in vivo. In addition, the co-control by PrfA and SigB of several genes activated in vivo strongly highlights the importance of the interplay between these two regulons during the infectious process. The hypothesis on an in vivo intersecting regulation of SigB and PrfA is in agreement with a very recent study demonstrating the contribution of SigB and PrfA to a regulatory network critical for appropriate regulation of virulence gene expression [74]. The large number of additional regulons and predicted transcriptional regulators differentially regulated inside the host underlines the high degree of regulation required for adaptation of Listeria to the host environment.
Another major aspect observed when Listeria interacts with its host is the active remodeling of the bacterial envelope through activation of the cell wall metabolism and enhanced exposure of virulence proteins at the bacterial surface.
Pathogens have evolved various systems for the secretion of bacterial factors that contribute to the progression of the disease. Listeria uses different secretion systems and a significant number of their products were activated in vivo. It is particularly the case of the SecA2 system, itself activated in vivo, and responsible for the secretion of several proteins lacking a signal peptide and also up regulated in the host, including known virulence factors.
Although competence genes have been found in L. monocytogenes genome [20], Listeriae have never been shown to be naturally competent. Interestingly, we observed several competence genes (comEA, comEB, comGF, comGE, clpC, mecA and degU) up regulated in vivo. This is the first report of a simultaneous activation of a great number of competence genes in Listeria, suggesting that this bacterium could be competent during infection and use this system to incorporate DNA from the host environment in order to acquire new potentialities.
In addition to the up regulation of a number of virulence factors, L. monocytogenes activates mechanisms of subversion of the host defenses. Lysinylation of phospholipids in Listeria membranes by MprF, and D-alanylation of cell wall TAs and LTAs by the dlt complex lead to a reduced negative charge of the bacterial surface. One of the consequences of this process is the repulsion of cationic antimicrobial peptides [13],[25]. In vivo activation of both dlt and mprF by VirR strongly suggests a regulation of L. monocytogenes resistance to cationic peptides in the host, as previously proposed [25]. Furthermore, surface components e.g. LTAs and PG play a role in the innate immune response through receptors like Nods and Toll-like receptors [75]. LTA modification by VirR regulated factors could thus be used by Listeria to escape the host innate immune response.
SecA2-dependent secretion has been proposed to coordinate PG digestion by the activity of secreted autolysins [30]. The muramyl glycopeptide predicted to be generated by the combined activities of p60 and MurA is known to modify host inflammatory responses [76],[77]. Thus, the strong in vivo induction of SecA2 and SecA2-secreted proteins may activate release of specific peptides that interfere with host pattern recognition.
N-deacetylation by PgdA was shown to be a major modification of Listeria PG, conferring the ability to survive in the gastrointestinal tract, in professional phagocytes, evade the action of host lysozyme, and modulate the inflammatory response [31]. The over expression of pgdA during infection appears as an additional strategy used by Listeria to subvert host pattern recognition and control the host inflammatory responses to promote its own survival.
In the same way, in accordance to the potent proinflammatory activity of flagellin [78], Listeria down regulates flagella related genes (lmo0681 and lmo0697) during the infection by the activation of mogR, a repressor of motility and chemotaxis gene expression [79].
Once inside the host, bacteria need to adapt to nutritional changes, including carbon and nitrogen sources. This is illustrated by the high number of metabolic genes differentially regulated in vivo.
UhpT and several enzymes involved in glycolysis were up regulated whereas enzymes implicated in the non-oxidative phase of the pentose phosphate pathway were repressed. These data suggest that phosphorylated glucose transported by UhpT is being metabolized through glycolysis. Glucose or phosphorylated glucose seems thus one of the major carbon sources in vivo, the pentose phosphate cycle appearing not essential for the generation of necessary intermediates and for gluconeogenesis. In addition to genes involved in glycolysis, we observed the in vivo up regulation of numerous genes implicated in the citric acid cycle and in oxidative phosphorylation, indicating that L. monocytogenes is using oxidative phosphorylation to generate energy. However, the activation of genes involved in fermentation, suggests that, even though L. monocytogenes is using oxidative phosphorylation to generate energy, it might be experiencing some level of oxygen starvation in spleen cells. The activation of the glycerol kinase and glycerol-3-phosphate dehydrogenase, indicates that glycerol, probably deriving from the activity of the phospholipases A and B on cellular lipids, is an additional carbon source for intracellular growth.
Metal ions are essential cofactors for functional expression of many proteins in bacterial systems. Thus, alterations in Listeria ion transport genes in vivo reflect the accessibility/inaccessibility of those ions in the intracellular environment, i.e presence of potassium and lack of cobalt, manganese, calcium and iron.
Due to defense mechanisms developed by the host to limit bacterial multiplication, it could be expected a growth rate decrease for invading pathogens in their host [16]. A very interesting discovery resulting from our in vivo transcriptomics analysis was the observation of an active growth status of Listeria in infected mouse spleens. This was demonstrated by the increased expression of numerous genes encoding proteins involved in bacterial growth and multiplication, including genes implicated in DNA replication and cell division. Listeria thus seems to have 24 h p.i. overcome organism defenses and being engaged in an active multiplication phase.
Bacterial responses to environmental changes are often characterized by the induction of specific stress responses. The in vivo induction of Listeria stress genes indicated that bacteria are faced with stress within the host. The alternative sigma factor B is the master regulator of stress. Even in the absence of a significant regulation of sigB itself in vivo, an impressive number (70) of genes previously shown to be under SigB-regulation were up regulated in vivo, including numerous virulence factors. In addition, genes up regulated in vivo and under the control of HrcA and CtsR seem also to be particularly relevant for virulence. During the process of host colonization, L. monocytogenes induces a host inflammatory response [80]. This defense is accompanied by the generation of ROS presented to the persistent pathogen. In addition to the traditional ROS combating enzymes like catalase and superoxide dismutase, this transcriptomic analysis revealed the in vivo activation of panoply of genes implicated in the response to oxidative stress, suggesting a special relevance of this response for Listeria pathogenesis/persistence.
Our study highlights that the analysis of a host-pathogen interaction in its real context (i.e. the living host) is highly informative. Indeed, it is not currently feasible to reconstruct in vitro the exact environment faced by Listeria in the host. This is illustrated by the strong difference observed between our in vivo transcriptome and previous transcriptomes of Listeria growing inside epithelial cells or macrophages [14],[15], with only 15% and 29% overlap for the up regulated genes, and no more than 3% for the down regulated genes. Furthermore, 25% and 44% of the genes down regulated in epithelial cells and macrophages respectively, were up regulated during mouse infection. One of the main differences between in vitro intracellular growth and in vivo infection was the much higher number of previously known virulence genes activated in vivo (29 against 17), or down regulated intracellularly (8 against 2). Genes identified in this study as up regulated in vivo and implicated in virulence are not regulated in epithelial cells, and not regulated (lmo1102) or even repressed (lmo1082, gap) in macrophages. lmo2714 is the only of these genes that appeared activated both in vivo and in macrophages. Other significant differences between “in cultured cells” and in vivo approaches were observed at the level of genes involved in cell division and cell wall metabolism, repressed in macrophages but activated in the host, suggesting a more active multiplication status of Listeria in mouse organs. Listeria metabolism in the two environments appeared also significantly different, in particular concerning glycolysis and the pentose phosphate pathway that, in contrast to what was observed in cultured cells [14],[15], were respectively activated and repressed in bacteria growing in mouse spleens. Finally, we observed a strong in vivo down regulation of flagella related genes, in accordance to the potent proinflammatory activity of flagellin [78], and inversely to what observed during intramacrophagic growth [14],[15].
In addition to the global analysis of the expression of the entire Listeria genome during infection, a major goal of this study was the identification of new Listeria virulence factors. The in vivo differential expression of a remarkable number of genes previously implicated in Listeria intracellular survival and virulence underscores the relevance of our approach. Our analysis allowed the detection of several potential novel virulence genes. Mutagenesis of 6 of these genes demonstrated the implication of a majority of them in virulence, thus definitively establishing the value of our strategy.
lmo2714 is a gene up regulated during infection and that encode a LPXTG surface protein [35],[36]. The probable implication of this LPXTG protein in L. monocytogenes virulence in mice confirms the importance of this protein family for the Listeria-host interaction and underlines the complexity of the mechanisms developed by this pathogen to reach a maximal infectious capacity. As other known surface virulence determinants [81],[82], Lmo2714 could interact with a specific cellular receptor or ligand that remains to be identified. Lmo2714 was also shown as present in the Listeria supernatant [43], and could thus also act as secreted factor.
gap encodes GAPDH, a glycolytic enzyme involved in bacterial energy generation that is essential for growth in the absence of neoglucogenic substrates. In Listeria, GADPH was previously described as a surface protein present in the cell wall, as well as a secreted protein [35],[43]. As a secreted product, GAPDH was shown to impair Rab5a mediated phagosome–endosome fusion [67]. Interestingly, GAPDH was also recently shown to be a key virulence-associated protein of Streptococcus suis type 2 up regulated in vivo [70]. The impossibility of constructing a gap deletion mutant confirmed the essential role of this protein in the bacterial metabolism as previously shown [70]. Using a gap secretion mutant, we showed that, whereas required for full growth in BHI, secreted GAPDH was not essential for intracellular multiplication. In addition, mouse infection indicated a role for secreted GAPDH in Listeria virulence, probably in part through its ability to retain and inactivate phagosomal Rab5a as previously described [67]. However, the gap secretion mutant exhibited an important in vitro growth defect. In addition, the complemented strain showed an accentuated growth delay in vitro and a more pronounced virulence decrease in vivo. These effects, even most probably due to an over expression of intracellular GAPDH, should be corrected in order to definitively prove the critical role of secreted GAPDH in virulence.
lmo1082 is homolog to rmlC that encodes a dTDP-dehydrorhamnose epimerase potentially implicated in the surface layer (S-layer) glycoprotein synthesis [83]. lmo1082 is furthermore part of a L. monocytogenes EGDe specific chromosomal region that contains two other genes that are homologous to S-layer biosynthesis genes in other organisms, several genes involved in TA biosynthesis, and the autolysin Auto previously implicated in Listeria virulence [84]. S-layers are two-dimensional crystalline arrays that completely cover bacterial cells. In addition to the impaired survival of the lmo1082 mutant in mouse organs, the high in vivo activation of lmo1082 could suggest a role for S-layer glycoproteins in Listeria virulence. S-layers have been shown to be virulence factors of several pathogens. In particular, the S-layer glycoproteins were implicated in mechanisms evolved by pathogenic bacteria to evade host immune systems. However, the examination of several strains using different techniques has so far never demonstrated the presence of S-layers in Listeria [85]. Preliminary experiments by electron microscopy did not allow to confirm the presence of a Listeria S-layer in vivo (data not shown). This aspect thus requires further investigation.
Lmo1102 is similar to CadC, a protein required for cadmium resistance. Cadmium is a heavy metal and its cation is toxic for microbes in the environment. Cadmium resistance in Listeria is an energy-dependent cadmium efflux system, involving two proteins, CadA and CadC. Listeria cadA and cadC genes for cadmium resistance were previously located, in the strains analyzed, on a transposable element (Tn5422) closely related to Tn917 and capable of intramolecular transposition [86],[87]. In the L. monocytogenes EGDe strain, the cadA and cadC genes are part of an EGDe specific chromosomal region located downstream an integrase encoding gene and containing 13 genes similar to Tn916 genes. This seems to indicate that L. monocytogenes EGDe has also acquired resistance to cadmium by transposon insertion. The strong in vivo activation of cadC and the significant impaired virulence of the cadC mutant suggest that this heavy metal resistance system constitutes an advantage for in vivo Listeria survival. The real function of CadC in vivo reserves further investigation.
In vivo, bacteria are challenged with unique cues that are difficult to reproduce under in vitro conditions. The molecular analysis of in vivo infectious processes by means of this approach provides the first comprehensive view of how L. monocytogenes adapts to the host environment in the course of the infection. We showed that the remarkable shift of the Listeria genome expression during infection is characterized by the activation of a number of genes involved in virulence and subversion of the host immune systems, and is associated with the adaptation of the bacterial metabolism to host conditions. All these mechanisms are under the control of a complex regulatory network. As confirmed here by the identification of several new virulence genes, this analysis provides a powerful tool for the detection of novel virulence determinants and a better understanding of the complex strategies used by pathogens to promote infections.
It would be now particularly interesting to perform the same in vivo genome profile analysis on different infected organs (intestine, liver, brain) and using different animal models in order to identify organ- or host-specific virulence factors.
L. monocytogenes EGDe was grown in Brain Heart Infusion (BHI) medium (BD-Difco) or in a defined minimal medium (modified Welshimer's broth [18]) at 37°C, under aerobic conditions with shaking. Erythromycin was included at 5 µg/ml when the bacteria carried pMAD and pAUL-A derivatives. Chloramphenicol was included at 7 µg/ml when the bacteria carried pPL2 derivatives. E. coli strains were grown in LB medium at 37°C, with shaking. Ampicilin and erythromycin were added at 100 µg/ml and 300 µg/ml, respectively, when required.
The macroarrays used here are described in [6]. Briefly, specific primer pairs were designed for each of the 2853 ORFs of the L. monocytogenes EGDe genome, in order to amplify a fragment of ∼500 bp specific for each ORF. For macroarray preparation, nylon membranes were soaked in 10 mM TE, pH 7.6. Spot blots of ORF-specific PCR products and controls were printed using a Qpix robot. Immediately after spot deposition, membranes were neutralized for 15 min in 0.5 M NaOH, 1.5 M NaCl, washed three times with distilled water and stored wet at −20°C until use.
cDNA synthesis, labeling and hybridization were performed as previously described [6]. Briefly, cDNA was reversely transcribed in the presence of [α-33P]-dCTP. Labeled cDNA was purified using a QIAquick column (Qiagen). Hybridization and washing steps were carried out using SSPE buffer. Macroarrays were pre-wet in 2× SSC and pre-hybridized in hybridization solution (5× SSPE, 2% SDS, 1× Denhardt's reagent, 100 µg of sheared salmon sperm DNA/ml) at 65°C. Hybridization was carried out for 20 h at 65°C. After hybridization, membranes were washed twice at room temperature and twice at 65°C in 0.5× SSC, 0.2% SDS. For each condition, two independent RNA preparations were tested, and two cDNAs from each of the RNA preparations were hybridized to two sets of arrays and analyzed.
Array results are available at the GEO database under the accession number GPL7248 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=jvqvnqwicgkoajy&acc=GSE13057).
Membranes were scanned using a 445SI PhosphoImager. The ARRAYVISION software was used for quantification of the hybridization intensities. The intensity of each spot was normalized according to the median value of the total intensities of all spots on each array. The global background was calculated from the median intensity of 610 “no-DNA” spots homogeneously distributed throughout the membrane. For spots whose intensity value was lower than the median background intensity, the intensity value was replaced by the median background intensity, for analysis purposes.
The significance analysis of microarrays (SAM) program was used for identification of genes with statistically significant changes in expression [19]. SAM was conducted with the following log2 ratios of gene expression values: 1) 24 h post-infection versus pure culture, 2) 48 h post-infection versus pure culture, and 3) 72 h post-infection versus pure culture. One-class responses were chosen to test if the mean level of gene expression differed from a hypothesized mean. A delta value corresponding to a false discovery rate <5% was chosen. Genes with at least a twofold expression change that were significant according to this analysis in at least one time point were taken into account.
For clustering analysis, data was log transformed, median centered and an average-linkage clustering was carried out using CLUSTER software and the results were visualized by TREEVIEW [88].
Up to 1 µg of total RNA was reverse-transcribed by using the iScript kit (Bio-Rad). Forward and reverse primers (Table S6) were designed using Primer3 software (http://frodo.wi.mit.edu/) to produce an amplicon length of 70–200 bp. A standard curve was generated for each primer pair by using four ten-fold dilutions of cDNA from L. monocytogenes EGDe, to ensure that PCR efficiency was 100%. Quantitative PCR was performed for 45 cycles with 2 µl of cDNA, 10 µl of 2× SYBR green PCR master mix (Bio-Rad) and 0,25 pM (each) forward and reverse primers in a final volume of 20 µl. For each primer pair, a negative control (water), was included during cDNA quantification. After PCR amplification, a melting curve was generated for every PCR product to check the specificity of the PCR reaction. Data were analyzed by the ΔΔCt method which provides the target gene expression value as unitless fold changes in the unknown sample compared with a calibrator sample [89]. Both unknown and calibrator sample target gene expression data were normalized by the relative expression of 16S rRNA.
For complementation, the entire gene and flanking regions were amplified using primers described in Table S9. PCR products were digested as described in Table S9 and ligated to the site-specific phage integration vector pPL2 [72]. Plasmid DNA of pPL2 bearing the fragments was transformed into E. coli S17-1 and the resulting strain was mated into each mutant strain. Chloramphenicol-resistant transconjugants were tested by PCR for pPL2 integration at the appropriate chromosomal site using primers PL102 (5′-TATCAGACCAACCCAAACCTTCC-3′) and PL95 (5′-ACATAATCAGTCCAAAGTAGATGC-3′). Primers described in Table S9 were used to confirm the presence of each gene in the respective complemented strain.
Animal experiments were performed as previously described in [94]. Bacterial growth in mice was studied by injecting 6-week-old specific pathogen-free female BALB/c mice (Charles River) intravenously with a sublethal bacterial inoculum, 104 CFUs, of wild type or mutant strains. At 72 h after infection the liver and spleen were sterilely dissected and the number of CFUs was determined by plating serial dilutions of organ (liver and spleen) homogenates on BHI agar medium (five animals for each strain).
All animals were handled in strict accordance with good animal practice as defined by the relevant national and local animal welfare bodies, and all animal work was approved by the Direcção Geral de Veterinária (FCT-POCI/SAU-MMO/60443/2004, FCT-PTDC/SAU-MII/65406/2006).
Listeria strains were grown to OD600 = 0.6, washed and diluted in DMEM such that the MOI was about 10 bacteria per cell. Bacterial suspensions were added to J774A.1 cells for 45 min. Cells were then washed and non-phagocytosed bacteria were killed by adding 20 µg/ml gentamicin for 1 h15 min. After washing, cells were lysed in 0.2% Triton X-100, at 2 h, 5 h, 7 h and 20 h post-infection and the number of viable bacteria released from the cells was assessed after serial dilutions of the lysates on BHI agar plates. Experiments were repeated two times in triplicate.
|
10.1371/journal.pcbi.1005630 | Optimal structure of metaplasticity for adaptive learning | Learning from reward feedback in a changing environment requires a high degree of adaptability, yet the precise estimation of reward information demands slow updates. In the framework of estimating reward probability, here we investigated how this tradeoff between adaptability and precision can be mitigated via metaplasticity, i.e. synaptic changes that do not always alter synaptic efficacy. Using the mean-field and Monte Carlo simulations we identified ‘superior’ metaplastic models that can substantially overcome the adaptability-precision tradeoff. These models can achieve both adaptability and precision by forming two separate sets of meta-states: reservoirs and buffers. Synapses in reservoir meta-states do not change their efficacy upon reward feedback, whereas those in buffer meta-states can change their efficacy. Rapid changes in efficacy are limited to synapses occupying buffers, creating a bottleneck that reduces noise without significantly decreasing adaptability. In contrast, more-populated reservoirs can generate a strong signal without manifesting any observable plasticity. By comparing the behavior of our model and a few competing models during a dynamic probability estimation task, we found that superior metaplastic models perform close to optimally for a wider range of model parameters. Finally, we found that metaplastic models are robust to changes in model parameters and that metaplastic transitions are crucial for adaptive learning since replacing them with graded plastic transitions (transitions that change synaptic efficacy) reduces the ability to overcome the adaptability-precision tradeoff. Overall, our results suggest that ubiquitous unreliability of synaptic changes evinces metaplasticity that can provide a robust mechanism for mitigating the tradeoff between adaptability and precision and thus adaptive learning.
| Successful learning from our experience and feedback from the environment requires that the reward value assigned to a given option or action to be updated by a precise amount after each feedback. In the standard model for reward-based learning known as reinforcement learning, the learning rates determine the strength of such update. A large learning rate allows fast update of values (large adaptability) but introduces noise (small precision), whereas a small learning rate does the opposite. Thus, learning seems to be bounded by a tradeoff between adaptability and precision. Here, we asked whether there are synaptic mechanisms that are capable of adjusting the brain’s level of plasticity according to reward statistics, and, therefore, allow the learning process to be adaptive. We showed that metaplasticity, changes in the synaptic state that shape future synaptic modifications without any observable changes in the strength of synapses, could provide such a mechanism and furthermore, identified the optimal structure of such metaplasticity. We propose that metaplasticity, which sometimes causes no observable changes in behavior and thus could be perceived as a lack of learning, can provide a robust mechanism for adaptive learning.
| To successfully learn from reward feedback, the brain must adjust how it responds to and integrates reward outcomes, since reward contingencies can unpredictably change over time [1,2]. At the heart of this learning problem is a tradeoff between adaptability and precision. On the one hand, the brain must rapidly update reward values in response to changes in the environment; on the other hand, in the absence of any such changes, it must obtain accurate estimates of those values. This tradeoff, which we refer to as the adaptability—precision tradeoff [3,4], can be easily demonstrated in the framework of reinforcement learning [5]. According to this framework, larger learning rates result in higher adaptability but lower precision, and smaller learning rates give rise to lower adaptability but higher precision. In recent years, the failure of conventional reinforcement learning (RL) models to capture the level of adaptability and precision demonstrated by humans and animals has led to alternative explanations for how we deal with uncertainty and volatility in the environment [1,6,7,8]. However, most of these solutions for adjusting learning require complicated calculations, and their underlying neural substrates are unknown.
Given the central role of synapses in learning, we asked whether there are local synaptic mechanisms that can adjust the level of plasticity according to reward statistics and, therefore, allow the learning process to be adaptable. A candidate mechanism for such adjustment is metaplasticity, defined as changes in the synaptic state that shape the direction, magnitude, and duration of future synaptic changes without any observable change in the efficacy of synaptic transmission [9,10,11,12]. Extending our recent heuristic model of reward-dependent metaplasticity, which enables adjustment of learning to reward uncertainty [3], we examined a general class of metaplastic models to identify features that are beneficial for mitigating the adaptability-precision tradeoff (APT) during the estimation of the probability of binary reward.
Using the mean-field and Monte Carlo simulations, we identified optimal metaplastic models that can substantially overcome the APT. These models, which we refer to as ‘superior’ models, achieve both adaptability and precision by forming two separate sets of meta-states: reservoirs and buffers. Synapses in reservoir meta-states do not change their efficacy upon reward feedback, whereas those in buffer meta-states can change their efficacy. In superior models, rapid changes in efficacy are limited to synapses occupying buffers, creating a bottleneck that reduces noise without significantly decreasing adaptability. In contrast, more-populated reservoirs can generate a strong signal without manifesting any observable plasticity. Comparison of the behavior of our model and a few competing models during a dynamic probability estimation task revealed that superior metaplastic models perform close to optimally for a wider range of model parameters. However, superior models were suboptimal when precision was defined in the absolute term (absolute value of the difference between estimated and actual probabilities) rather than relative (the ability to distinguish neighboring values of probability), indicating that the brain could use different objectives to deal with reward uncertainty. Finally, we showed that metaplasticity provides a robust mechanism for mitigating the APT, and that metaplastic transitions are crucial, since replacing these transitions with plastic ones reduces the ability of the model to mitigate the APT. Altogether, our results illustrate how metaplasticity can mitigate one of the most fundamental tradeoffs in learning and, moreover, reveal the critical features of metaplasticity that contribute to adaptive learning.
To study the relationship between adaptability and precision, we considered a general problem of estimating reward probability from a stream of binary outcomes (reward, no reward). We defined adaptability and precision in the context of this estimation task in order to quantify the adaptability-precision tradeoff (APT). We assumed that the estimation of reward probability is performed by a set of synapses and, thus, reward probability is stored in the strength of these synapses. As a result, adaptability in estimation of reward probability requires these synapses to change their strengths quickly whereas precision requires that the strength of synapses can discriminate between neighboring values of reward probability (see below).
The estimation of reward probability can be preformed with a model consisting of synapses that follow a stochastic reward-dependent plasticity rule [13–15] (Fig 1a). In this model, which we refer to as the ‘plastic’ model, synapses are binary so they can be in the weak or strong state [16,17]. Weak synapses can be potentiated on rewarded trials with a probability t+ (potentiation rate), whereas strong synapses can be depressed on unrewarded trials with a probability t− (depression rate). The difference between the fractions of synapses that are in the strong and weak states determines the signal stored in these synapses, reflecting the model’s estimate of reward probability. This model is equivalent to a simple RL model based on reward prediction error and can provide an unbiased estimate of reward probability when potentiation and depression rates are equal (S1 Text).
For a given value of reward probability, pr, the steady state of the model can be used to calculate the signal, and the weighted average change in signal due to single potentiation and depression events (‘one-step’ noise) provides a good proxy for noise (see Methods). We measured precision with the ability of the model to differentiate between adjacent reward probabilities instead of a more conventional definition based on the difference between the estimated and actual probabilities. We adopted the former definition because encoding and representation of reward information are inherently relative in the brain, since the probability estimated by a set of synapses can be easily scaled and biased by changes in the input neural firing to these synapses. Therefore, we defined the ‘precision’ (ℙ) as equal to the sensitivity of the model’s signal to changes in reward probabilities (‘sensitivity’), divided by noise in the signal. The ‘adaptability’ (A) of a model in estimating reward probability was defined as the rate at which the fractions of meta-states approach their final values (see Methods).
Because of the simplicity of the plastic and RL models, adaptability and precision can be analytically computed for these models (S2 Text). Both these models show a strict APT, since the product of adaptability and precision is independent of model parameters and only depends on reward probability (Fig 1a and 1b). Importantly, adopting different values for the potentiation and depression rates (or equivalently the learning rates in RL) cannot improve the APT. Rather, adoption of different values slightly alters the average values of adaptability and precision over a set of reward probabilities (Fig 1d and S1 Fig).
Here, we considered a general model of metaplasticity and used optimization to identify the superior metaplastic models for mitigating the APT. Our general model of metaplasticity consisted of multiple meta-states associated with one of the two values of synaptic efficacy (weak and strong), and all possible transitions between these meta-states (Fig 1c; see Methods). In this model, the difference between the fractions of synapses that are in the strong and weak meta-states determines the signal (S) stored in these synapses, reflecting the model’s estimate of reward probability. Importantly, we assumed that metaplastic transitions have a consistent order, and thus, within the set of weak and strong meta-states, there are multiple meta-states with different levels of depth (Fig 1c). Since we were interested in conditions under which metaplasticity can improve the APT, we examined ‘superior’ metaplastic models (i.e. those which optimized A × ℙ for a given value of ℙ).
We found that for many model parameters, the APT can be mitigated by superior metaplastic models that consist of as few as four meta-states (Fig 1d; S2 Fig). These superior models overcame the APT by exhibiting three important characteristics: differential adjustments of learning based on reward probability; matching of sensitivity to noise; and optimal adaptability. Firstly, the learning on rewarded and unrewarded trials was differentially adjusted according to reward probability. Secondly, the sensitivity of the signal to reward probability matched the level of noise (sensitivity-to-noise matching), and this matching was improved with larger numbers of meta-states. Thirdly, the adaptability of the models was optimized for a given level of noise (see below).
The first characteristic of superior models is that learning was naturally adjusted according to reward probability without any changes in the model’s parameters. To show this adjustment, we computed the ‘effective’ learning rates for potentiation and depression events for a given value of pr (see Methods). The effective learning rate assigned a single rate to transitions between the weak and strong meta-states or vice versa (plastic transitions, Fig 1c), which are the only transitions that can change synaptic efficacy and thus the signal. We found that the effective learning rate on rewarded trials (t˜+) was close to zero for small values of pr but monotonically increased as pr increased (Fig 2a). At the same time, the effective learning rate on unrewarded trials (t˜−) was large when pr was close to zero and decreased as pr increased. The effective learning rates on rewarded and unrewarded trials crossed over at 0.5 due to the symmetry in models parameters with respect to reward and no reward.
To understand why these adjustments are beneficial for mitigating the APT, one should note that in the RL model, the convergence to the final estimate of reward probability (when pr is small) slows down as more negative outcomes (no reward) are observed, since reward prediction error becomes smaller for unrewarded trials. This property limits adaptability. At the same time, the response to a positive outcome (reward) increases since reward prediction error increases on rewarded trials, and this property increases noise. In contrast, metaplastic models increase t˜− as the models receive more negative outcomes allowing them to slow their convergence to the final value of probability estimate to a lesser degree. On the other hand, decreasing t˜+ makes the estimate more robust against sporadic positive outcomes (noise). The opposite happens when pr becomes closer to one.
These complementary adjustments in learning resulted in a sigmoid-shape signal for superior metaplastic models (Fig 2b), which in turn, gives rise to the second characteristic of superior models, the match between the sensitivity and the noise level (Fig 2c). More specifically, the maximum sensitivity (dS/dpr) for superior models occurred at pr = 0.5, such that the steepest part of the signal matched the maximum level of noise (Fig 2c). Additionally, for a given level of precision, the signal became a steeper function of reward probability and the maximum sensitivity increased as the number of meta-states increased (Fig 2b). Importantly, the slope of the signal (i.e. sensitivity) at pr = 0.5 was linearly proportional to the ratio of effective learning rates around pr = 0.5, indicative of a direct relationship between the sensitivity-to-noise matching and adjustment of learning to reward probability. The adjustments occur in metaplastic models without any changes in parameters; as reward probability deviates from 0.5 (say when pr > 0.5), more synapses move to shallower weak meta-states, increasing the effective potentiation rate above the effective depression rate (Fig 2a). As the ratio of effective potentiation to depression rates increases, however, the fraction of synapses in weak meta-states decreases. Consequently, sensitivity to reward probability decreases as pr becomes larger or smaller than 0.5.
As noted above, in addition to the sensitivity-to-noise matching, adaptability of the superior models was optimized for a given level of noise. This optimization occurred because metaplasticity enabled superior models to form two separate sets of meta-states: reservoirs and buffers. Reservoirs, which are unique to metaplastic models, are the deepest sets of meta-states that cannot change their efficacy upon potentiation or depression events; they can only undergo metaplastic transitions (Fig 3a). Buffers, on the other hand, are the shallowest meta-states, and are able to undergo plastic transitions that change their synaptic efficacy. We refer to the remainder of the meta-states as ‘transient’. Because the superior models had reservoirs and buffers, they were able to keep a large proportion of their synapses in the weak or strong reservoirs (Fig 3b). Synapses within reservoirs were protected against changes in efficacy upon potentiation or depression events, and as a result, the signal could increase without increasing the level of noise.
The adaptability in the model depends on the rates of transitions between all subsets of meta-states, whereas noise (in reward estimation) depends on the flow across the plastic boundary (i.e. transitions between weak and strong meta-states and vice versa). Therefore, to understand how the model’s adaptability is optimized for a given level of noise, we computed the ‘effective transition rate’ for all subsets of meta-states. The effective transition rate was defined as the outward flow of synapses out of that subset divided by the fraction of synapses in that subset. This quantity, which is closely related to the concept of conductance in Markov chains [18], measures how easily synapses could leave a subset of meta-states (Fig 4a; see Methods). Importantly, the model’s adaptability is constrained by its slowest effective transition rate.
In superior models, to reduce noise with a minimum cost to the adaptability, the slowest transition rates should be at the plastic boundary. We found that this was the case for all superior models (Fig 4). Interestingly, having the minimum effective transition rates at plastic transitions created a ‘bottleneck’ for the flow between weak and strong meta-states. This bottleneck helped reduce noise without significantly reducing the adaptability. The superior models with N > 4 also contained transient meta-states, with the fastest effective transition rates between buffers and reservoirs, resulting in improved adaptability (Fig 4c and 4d).
This specific arrangement of meta-states and transitions between them, as well as the adjustment of the metaplastic model to reward probability, enabled metaplastic models to be more adaptable than corresponding binary plastic models. To demonstrate this superior adaptability, we used the effective learning rates for a given value of pr to define an equivalent binary plastic model (N = 2 model) for any metaplastic model. We found that metaplastic models showed larger sensitivity to reward probability than equivalent plastic models (Fig 5). Moreover, metaplastic models were more adaptable and more precise than their equivalent plastic models. These results demonstrate that the dynamic adjustment of learning in metaplastic models is crucial for improving the APT, and that this adjustment cannot be achieved by simply replacing the learning rates in corresponding plastic models with the effective learning rates based on the superior metaplastic models.
To further study the characteristics of superior metaplastic models, we next examined the transition probabilities in these models. We found that most transition probabilities were very close to zero, allowing for the creation of reservoirs and buffers, while non-zero transition probabilities varied proportionally to create models with different levels of adaptability and precision (Fig 6a–6c). For example, in metaplastic models with four meta-states (Fig 6a), three of six transition probabilities for potentiation were zero, two others were equal, and the last one was very close to those two other non-zero probabilities.
Based on these observations, we constructed a superior family of metaplastic models using a single parameter (transition probability). This was done to test whether such metaplastic models with only a single transition probability can significantly mitigate the APT. We found that even such simple metaplastic models can overcome the APT, and this ability was improved with additional meta-states (Fig 6d–6f). Overall, these results show that metaplastic models outperform plastic models, not because they have more parameters, but because they have a structure that allows for strong adjustment of learning.
These results illustrate that having more meta-states can improve the ability of metaplasticity to overcome the APT even for superior one-parameter models. The basic mechanism for this improvement is the existence of reservoirs, buffers, and a bottleneck for changing synaptic efficacy. Additional meta-states provide intermediate transitions between reservoirs and buffers that could increase signal and reduce noise without significantly decreasing the adaptability (Fig 6d–6f). As a result, models with larger numbers of intermediate meta-states show better matching of sensitivity to noise as well as more optimized adaptability for a given level of noise. Essentially, the specific structure for changing synaptic efficacy allows the models with a large number of meta-states to collect evidence (by transitioning synapses to shallower meta-states) before making a change.
The results above were obtained using the mean-field (MF) approach. Although the MF approach could accurately estimate the signal, there are two components of the MF approach that could yield different results from the Monte Carlo (MC) simulations: adaptability and noise. As we show below, only the estimation of noise based on the MF approach is significantly different than noise based on the MC simulations. Nevertheless, our main findings based on MF also hold using MC simulations.
In the MF approach, adaptability is measured by the eigenvalue of the slowest decay mode of the transition matrix. However, what influences the estimation of reward probability is the synaptic strength (signal), since the synaptic efficacies of all weak meta-states or all strong meta-states are the same. That is, the asymptomatic rate of convergence to the new equilibrium or steady state of the synaptic strength could be different. Reaching steady state based on meta-states provides a lower bound for adaptability, since such equilibrium guarantees reaching steady state based on the synaptic strength but not vice versa. Comparing adaptability computed by the two methods, however, revealed only a small difference due to finite-size effects in the MC simulations (Fig 7a).
The only difference between the MF approach and MC simulations was the estimation of noise. Using the MF approach, the estimated noise was set to one-step noise, which is equal to the weighted average of changes in the steady state of synaptic strength due to a potentiation and depression event. The one-step noise converges to the actual noise if the adaptability is equal to 1. When the adaptability is different from 1, one-step noise underestimates the actual level of noise measured by real simulations (Fig 7b). Intuitively, this underestimation occurs because of the extra noise in the MC simulations due to fluctuation of the fractions of synapses in different meta-states around their steady-state values. While underestimation of noise in the MF approach increases with the number of meta-states, this effect is not strong enough to change the sensitivity-to-noise matching (Fig 7c). Moreover, the MC simulations showed the same order of models in their ability to overcome the APT (compare Figs 7d, 1d and 6d).
In order to obtain the optimal structure of metaplasticity based on a general model, independently of a given task or set of task parameters, we measured adaptability as the rate at which the signal in the model approaches its steady state. Moreover, we measured precision as the ability of the model to differentiate between adjacent values of probability while considering noise. This “relative” definition of precision was adopted because any information stored at the synaptic level can be amplified (and thus be biased) by a change in the input firing rate.
Although superior metaplastic models are optimal in mitigating the APT based on the adopted definitions, a certain task or set of task parameters could favor certain models or certain model parameters. Alternatively, one could measure precision in an absolute fashion, for example as the difference between the estimated and actual reward probability. Therefore, we tested the performance of one-parameter superior models and a few competing models during a dynamic probability estimation task using both relative and absolute definitions of precision (see Methods). More specifically, we computed the performance of various models using the average difference between the transient signal and the steady state of the signal based on the actual value of reward probability at each time point in the task. We also computed the estimation error as the difference between the estimated and actual reward probability.
As expected, for a simple environment defined by the value L (the number of trials before reward probability is changed), the estimation error depends on the model parameter (transition probability or learning rate), except for when using the Bayesian model (Fig 8). For a large value of L, the RL model can achieve its optimal performance for a small value of learning rate, but the estimation error increases sharply for other values of the learning rate above that of the Bayesian model (Fig 8a). In contrast, the one-parameter superior family shows a small estimation error for a wide range of transition probability values. The cascade model shows larger estimation error since this model was designed to preserve its signal [19]. The performance of our previous heuristic metaplasticity model (RDMP) [3] falls between the superior and cascade models for larger values of the transition probability. Qualitatively, similar behavior was observed for performance in a simple environment with a smaller value of L (Fig 8b) and in a complex environment in which the value of L changes between blocks of trials (Fig 8c). Overall, these results illustrate that superior metaplastic models, which are identified by optimizing for the APT, can perform near-optimally over a wide range of parameter values during a dynamic estimation task.
In contrast, performance based on the absolute measure of estimation error (i.e. difference between the estimated and actual reward probability) revealed that one-parameter superior models perform worse than competing models (S3 Fig). This suboptimality of one-parameter superior models, however, stems from the steady-state signal (i.e. estimated reward probability) that strongly deviates from the actual reward probability (S4 Fig). This deviation, which allows the superior models to be very sensitive to changes in reward probability near 0.5, was much less pronounced in the cascade and RDMP models (signal in the RL model is equal to the actual reward probability). Overall, these results illustrate that superior metaplastic models, which are identified by optimizing for the APT, can perform close to optimally over a wide range of parameter values during a dynamic probability estimation task. However, these models can be suboptimal when an absolute metric is used for measuring precision (see Discussion).
In order to test the robustness of the metaplasticity solution, we examined how sensitive the superior solutions were with respect to changes in transition probabilities. To do so, we randomly perturbed the non-zeros elements in the potentiation and depression matrices of the superior models by a specific amount. We found that superior models show a high degree of robustness in their adaptability and precision against changes in their potentiation and depression transfer matrices as long as their transition topologies are not altered (i.e. the zero elements of transition matrices are kept zero) (Fig 9).
The ultimate test for whether metaplastic transitions are crucial for mitigating the APT is to replace these transitions with plastic ones (transitions that change synaptic efficacy) while keeping the same number of states and transitions. Therefore, we examined the APT in the simple family of metaplastic models, but with different values of synaptic efficacy assigned to different meta-states (Fig 10a; see Methods). This ‘graded’ plastic model could be reduced to the metaplastic model by setting equal values of synaptic efficacy for different weak or strong states. We found that A × ℙ monotonically increased as the graded plastic model became more similar to the metaplastic model, reflecting the importance of metaplasticity to overcome the APT (Fig 10b–10d). Nevertheless, additional states in the plastic models improved the ability of these models to mitigate the APT beyond binary plastic synapses (S5 Fig) similar to improvement of memory storage capacity with more states [20]. Overall, these results demonstrate that metaplastic transitions are crucial for mitigating the APT.
The demands of learning in a changing world require a high degree of adaptability, which comes at the cost of low precision [4]. Here we show how metaplasticity, which is reflected in the unreliability of synaptic plasticity, can provide a solution for substantially overcoming the APT. More specifically, by optimizing the APT for a given level of precision, we identify crucial characteristics of superior metaplastic models. The superior models contain reservoir and buffer meta-states; synapses in reservoir meta-states do not change their efficacy upon reward feedback, whereas those in buffer meta-states can change their efficacy. Moreover, rapid changes in efficacy are limited to synapses occupying buffers, which provides a bottleneck that reduces noise without significantly decreasing adaptability. In contrast, more-populated reservoirs can generate a strong signal without manifesting any observable plasticity. The generation of reservoirs and buffers by metaplastic synapses results in the adjustments of learning, or the degree of plasticity, according to recent reward history. For example, when synapses occupy reservoir meta-states, which occurs with consecutive rewarded or unrewarded trials in a stable environment, the behavior should become less adaptable. However, when reward history changes over time, synapses mainly occupy buffer meta-states, causing more adaptable behavior. Overall, the model predicts that learning should be more sensitive to the reward sequence than what has previously been assumed (also see [3]).
Importantly, the results of one-parameter superior models and the MC simulations show that more meta-states can improve ability to overcome the APT and, in addition, give rise to more robust models for adaptive learning. These results are compatible with similar improvements in memory storage capacity with larger numbers of states [20]. Interestingly, the signal in the metaplastic models is a sigmoid-like function of reward probability. This illustrates that, for intermediate values of pr (around 0.5), learning based on metaplastic synapses is more sensitive to changes in reward probability than learning based on plastic synapses. This sensitivity increases with the number of meta-states. Future experiments that can measure the sensitivity of probability learning can test this prediction of metaplasticity.
The basic mechanism for improvement with more meta-states is also related to the generation of reservoirs and buffers that create a bottleneck for changing synaptic efficacy, and not merely because of having a larger number of states as in graded synapses [21]. More specifically, additional meta-states provide intermediate transitions between reservoirs and buffers that could reduce noise without significantly compromising adaptability. Interestingly, it has been shown that in the framework of Markov chains, the eigenvalues and eigenvectors of models with bigger spectral gaps (i.e. more adaptable models) are less sensitive to perturbation of transition probabilities [22–25]. In other words, more adaptable models can produce signals without fine-tuning. Superior metaplastic models require only a few parameters, and their behavior is not very sensitive to these parameters.
As a higher-order form of plasticity, metaplasticity has been successfully used to explain paradoxical observations regarding synaptic plasticity by considering prior synaptic activity [12]. At the cognitive level, however, the computational power of metaplastic synapses has been mainly explored to address memory retention [19,26,27]. For example, Fusi and colleagues proposed the so-called cascade model to explain how memory could be protected from synaptic changes due to ongoing activity over large timescales [19]. By having multiple timescales associated with different meta-states, the cascade model can achieve high levels of both memory storage and retention time, and therefore, mitigate the ‘storage-retention’ tradeoff (i.e. a system which is good at storage would be poor at retention, and vice versa). A more recent study has shown that this tradeoff in memory systems can be further improved by having a large number of states that initially store memory quickly and then transfer memories to slower states [28]. This storage-retention tradeoff is exactly the opposite of the adaptability-precision tradeoff studied here, since memory systems are concerned with maintaining the signal whereas learning systems need to be adaptable. Nevertheless, it is encouraging that metaplasticity can mitigate two very different tradeoffs. Moreover, these results suggest that metaplasticity can be useful for estimating signals other than reward probability and is generalizable to other domains of learning for which adaptability and precision are both important.
Our results could also explain why plasticity protocols are unreliable and outcome plasticity is heterogeneous [29]. As we showed, superior metaplastic models create bottlenecks for changing synaptic efficacy, since such a property can reduce noise with minimal decrease in adaptability. However, limiting plastic transitions to those that occur from buffers would make many transitions invisible to measurement of change in synaptic efficacy. Therefore, until such a structure is specifically tested, plasticity protocols will be perceived as noisy and unreliable. Besides recent behavioral evidence [3], there is no direct electrophysiological evidence for the structure of metaplasticity proposed here. We hope that our study stimulates experimentalists to investigate this structure.
Using the difference between the estimated and actual reward probability as a measure of precision, we find that superior metaplastic models are suboptimal. This occurs because the signal in these models is biased to allow maximal sensitivity to reward probability for intermediate values of reward probability. However, if reward estimates have to be ultimately used for making choices as in binary decision-making tasks, it is more desirable to have a higher accuracy near such values of probabilities (0.5) where the outcome of the decision is more sensitive to the estimated value. Moreover, reward information has to be encoded and represented in the brain in a relative fashion in order to deal with a limited range of neural firing rates. Accordingly, we adopted a relative definition for precision that measures the ability to distinguish neighboring values of reward probability. Therefore, our results suggest that some of the suboptimality in the estimation of reward probability could be due to the biophysical limitations of the nervous system in encoding values.
Our proposal provides a new approach for studying synaptic plasticity and its contribution to brain computations. Our model predicts that a previous reward outcome (learning experience) not only contributes to learning and behavioral changes, but also affects subsequent induction of such changes within a specific time window. On the one hand, certain sequences of reward feedback cause the nervous system to become more receptive to subsequent similar feedback. On the other hand, consecutive feedback can shape future learning such that it is not responsive to feedback in the opposite direction. Understanding such propensity for and unresponsiveness to reward feedback could provide new insights into habit and addiction, respectively. Therefore, further investigations into metaplasticity, both at the behavioral and synaptic levels, could help researchers discover tools for improving learning, especially with respect to habits and addiction [30,31]. Overall, our work highlights an overlooked contribution of synaptic mechanisms to solving complex cognitive problems [32].
Our general model of metaplasticity consisted of multiple meta-states associated with two values of synaptic efficacy (weak and strong) and all possible transitions between these meta-states (Fig 1a). The metaplastic models have N distinct meta-states, half of which are associated with strong synaptic efficacy and half with weak. The model is completely specified with two transition matrices, one for a potentiation event (Tij+) and one for a depression event (Tij−) corresponding to rewarded and unrewarded trials, respectively. Here, we assumed that metaplastic transitions have a consistent order such that potentiation and depression events (on rewarded and unrewarded trials, respectively) create flows in opposite directions. This assumption also establishes weak and strong meta-states with different ‘depths’ such that deeper states are further from the plastic boundary (Fig 1a). Moreover, we assumed symmetry between information by reward and no-reward feedback, and thus only focused on mirror-symmetric flows. This assumption put another constraint on the potentiation and depression matrices:
Ti,j+ =TN−i,N−j−
(1)
Based on these assumptions, transition matrices for potentiation and depression events can be represented by lower-triangular and upper-triangular matrices:
T+=[T110T12T22⋯0⋯0⋮⋮T1NT2N⋱0⋯TNN], T−=[TNN⋯0⋱T2NT1N⋮⋮⋮⋯0⋯T22T120T11]
(2)
There are N(N − 1)/2 unique transition probabilities for models with N meta-states. The probability conservation was dictated by the transition flows out of any meta-state summing up to 1.
We assumed that the estimation of reward probability was performed by a set of synapses and, thus, reward probability was stored in the strength of these synapses. At any point in time, the signal (S) was defined as the difference between the fractions of synapses in the strong and weak meta-states,
S(t) = Ψ+(t) − Ψ−(t)
(4)
where Ψ− = ∑i=1N/2Ψi and Ψ+ = ∑i=N2+1NΨi are fractions of synapses in the weak and strong meta-states, respectively. In the mean-field (MF) approximation approach, the average system dynamics is fully described by the average transition matrix for a given value of reward probability (T¯ij = prTij+ + (1 − pr)Tij−). The eigenvector, Ψ, with an eigenvalue λ = 1 (the largest eigenvalue according to Perron-Frobenius theorem) of average transition matrix, T¯ij, provided the steady state of the model from which the average signal was calculated using Eq 4.
As a proxy for signal fluctuations around its average value, we introduced the concept of ‘one-step noise’ as the mean magnitude deviation from the average signal due to one potentiation or depression event:
η ≡ pr|〈S〉 − S+| + (1 − pr)|〈S〉 − S−|
(5)
where 〈S〉 is the average signal based on the steady-state solution, and S+ and S− are the signal values after the application of the potentiation or depression transition matrices on the steady-state solution, respectively. In general, noise at time (t + 1) is a combination of several components: (1) the attenuated transferred noise from the state of the system at time t; (2) the amount of noise generated in one step, from t to (t + 1); (3) the inherent noise involved in translating p(t) to a binary representation with potentiation and depression events; and finally (4) a finite size effect when dealing with a limited number of identical synapses. The one-step noise measures the second component and always underestimates the level of the noise in the model. The Monte Carlo simulations, however, contain the sum of the first three components mentioned above and thus capture the overall noise.
We defined precision as the ratio of the signal sensitivity and the one-step noise:
ℙ = (dS/dpr)/η
(6)
Therefore, precision measures the discriminability between two adjacent reward probabilities based on their resulting signals. We chose this measure instead of the difference between the estimated and actual reward probability because the firing rate of neurons, which represents reward values, can be differentially scaled by their input firing rates. Therefore, the absolute difference may be irrelevant for the nervous system.
Finally, the adaptability of the model was defined as the rate of the decaying mode in the system, and was estimated using the difference between the second-largest eigenvalues (slowest decaying mode) of the average transition matrix and 1 (A = 1 − λ2), also known as the spectral gap in the Markov chains literature. We chose this definition because it is not possible to reduce the dynamics of metaplasticity to arrive at one equation for the synaptic strength. As a result, adaptability measures the lower bound for the rate of convergence to the final steady state of the synaptic strength. Nevertheless, we found that our definition provides a good approximation for this rate (Fig 7a).
By focusing on the steady-state solution, the concept of learning rates in the binary plastic models (N = 2) can be generalized to higher N as the effective learning rates, t˜±. The effective learning rates were defined as the relative change in the fraction of synapses in the weak or strong meta-states after a potentiation or depression event:
(T± Ψ)±=Ψ±+t˜±Ψ∓
(7)
where (T± Ψ)± is the sum of the fraction of strong/weak meta-states. To examine transitions from a given subset of meta-states, we also defined the ‘effective transition rate’ as the outward flow of synapses from that subset, divided by the fraction of synapses in that subset (Fig 3a). The effective transition rate (T˜ab) assigns a single rate for outward transition from a set of meta-states a to a set of meta-states b. There are (2N − 2) non-trivial ways that N meta-states can be partitioned into two disjoint, complementary subsets.
A closely related concept of conductance, C(S), for a given subset S in a Markov chain is defined as the outward flow from that subset divided by the minimum of occupancy in that subset, π(S), and occupancy in its complementary set π(Sc). The magnitude of one-step noise is directly related to the effective transition rate when the two subsets are chosen based on their synaptic efficacy. The value of spectral gap (i.e. the difference between the second-largest eigenvalues of the average transition matrix and 1) is constrained by the minimum conductance among all possible subsets of meta-states [18].
The Monte Carlo simulations were performed by running multiple trials starting from a given initial state in environments with identical reward statistics (reward probability was the same but the reward sequence varied across different simulations). Data from an initial relaxation period was discarded to remove dependence on the initial state, and the relevant quantities were computed by averaging over the ensemble at a given time step or across time. Moreover, to further reduce the relaxation time, we started from the steady-state solution of the mean-field equation for the initial environment.
To measure the decay rate in the Monte Carlo simulations, we simulated the dynamics of signal in one-parameter superior models (with N = 4, and 6) in response to a sudden jump from reward probability of 0.3 to 0.8. We averaged over 100000 different instances of such simulations to obtain the asymptotic convergence of the signal. The asymptotic signal was then fit to an exponential function and the best fit for the time constant was computed using minimum squared error methods. We performed these simulations for different values of model parameters and transition probabilities between 0.05 and 0.25 (with 0.01 increments). The results of the Monte Carlo simulations were then compared with the models’ slowest mode using the mean-field approach.
The optimal solutions (i.e. upper-boundary in adaptability × precision vs. precision plot) were found in two stages. An initial upper envelope in the A × ℙ vs. ℙ (using discretization for ℙ) was constructed by random sampling of 107 transition matrices. The transition matrices were divided into n bins according to their precision, ℙ, and the transition matrix with the highest value of A × ℙ in each bin was selected. These transition matrices were then used as the initial points for our optimization process. To avoid local minima, at the beginning of each iteration, a duplicated copy of the initial transition matrix with added small jitters was generated. All the resulting 2n transition matrices were used as the starting point of our optimization. At the end of each optimization iteration, the best solutions in each bin were selected out of all initial transition matrices, and the final outcome of our optimization procedure was used for the initial samples of the next iteration. For models with a large number of meta-states (N > 4), we conducted multiple iterations of the optimization process. The higher dimensional solutions are more robust against fluctuations, and optimized solutions can be found by increasing the bin numbers (initial points) and the number of optimization iterations. The optimization was constrained by keeping the sum of every column in transition matrices with positive elements to one. We used MATLAB’s ‘fminsearch’ function for the basic optimization process.
To compare superior metaplastic models and a few competing models, we measured the performance of these models in a dynamic probability estimation task. In this task, the reward is provided on each trial based on a fixed probability. This probability, however, increases or decreases (with equal probability) by 0.1 every L number of trials, resulting in 11 different values of reward probability ([0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]). Therefore, parameter L defines the level of volatility in this task. We simulated the behavior of various models in simple environments (environments with a fixed value of L) and in a complex environment where L can also change. For the complex environment, after each change in the reward probability, the value of L was selected randomly from the following set of values: [10, 20, 30, 40, 50, 60, 70, 80, 90, 100], subject to the constraint that overall lengths of blocks with a given value of L are similar.
We used two methods to compare the performance of various models (see below) in this task in terms of estimation error. In the first method, we computed the average difference between the transient signal and the steady state of the signal based on the actual value of reward probability at each time point during the task (relative estimation error). This was done because the signal (or the estimate of reward probability) in each model is relative, and there is a one-to-one mapping between the signal in a given model and the actual reward probability. In the second method, we computed the estimation error by the absolute value of the difference between the estimated and actual reward probability at each time point (absolute estimation error).
In order to compare the performance of our model with competing models, we simulated four models and measured their performances in a dynamic probability estimation task. The first model was an RL model based on reward prediction error (see S1 Text). This model is equivalent to the binary plastic model (N = 2) and can be quantified with one or two parameters (learning rates). The second model was the so-called cascade model of Fusi et al. (2005)[19]. This cascade model also assumes metaplasticity and order for transitions between different meta-states similarly to our superior models. However, the cascade model has a different structure for transition than our simple family model and, moreover, transition probabilities become smaller for deeper meta-states. The third model was a heuristic model of reward-dependent metaplasticity (RDMP), which we have proposed to capture behavioral data during a dynamic learning and decision-making task [3]. Finally, we also simulated a hierarchical Bayesian model that directly estimates volatility and change in volatility in order to determine the amount of update based on reward feedback [1].
|
10.1371/journal.pbio.1001298 | Cytoneme-Mediated Delivery of Hedgehog Regulates the Expression of Bone Morphogenetic Proteins to Maintain Germline Stem Cells in Drosophila | Stem cells reside in specialised microenvironments, or niches, which often contain support cells that control stem cell maintenance and proliferation. Hedgehog (Hh) proteins mediate homeostasis in several adult niches, but a detailed understanding of Hh signalling in stem cell regulation is lacking. Studying the Drosophila female germline stem cell (GSC) niche, we show that Hh acts as a critical juxtacrine signal to maintain the normal GSC population of the ovary. Hh production in cap cells, a type of niche support cells, is regulated by the Engrailed transcription factor. Hh is then secreted to a second, adjacent population of niche cells, the escort cells, where it activates transcription of the GSC essential factors Decapentaplegic (Dpp) and Glass bottom boat (Gbb). In wild-type niches, Hh protein decorates short filopodia that originate in the support cap cells and that are functionally relevant, as they are required to transduce the Hh pathway in the escort cells and to maintain a normal population of GSCs. These filopodia, reminiscent of wing disc cytonemes, grow several fold in length if Hh signalling is impaired within the niche. Because these long cytonemes project directionally towards the signalling-deficient region, cap cells sense and react to the strength of Hh pathway transduction in the niche. Thus, the GSC niche responds to insufficient Hh signalling by increasing the range of Hh spreading. Although the signal(s) perceived by the cap cells and the receptor(s) involved are still unknown, our results emphasise the integration of signals necessary to maintain a functional niche and the plasticity of cellular niches to respond to challenging physiological conditions.
| The Drosophila ovary contains a well-defined stem cell niche that hosts 2–3 germline stem cells (GSCs). The Hedgehog (Hh) family of signalling proteins mediates cellular homeostasis in several adult tissues, and here we decipher the detailed mechanism of action of Hh in the adult female GSC niche. We demonstrate that Hh acts in a juxtacrine manner (i.e., it requires physical contact between the cells involved) to maintain the normal pool of GSCs in the ovarian niche. Hh is produced in one type of niche support cell (the cap cells), and it is received, upon secretion, by a second, neighbouring population of niche cells (the escort cells). In the latter, we show that the Hh signalling pathway regulates the expression of the Drosophila Bone Morphogenetic Protein (BMP) homologues and essential stem cell factors decapentaplegic (dpp) and glass bottom boat (gbb). We also find that Hh distribution in the GSC niche is mediated by short cellular projections, reminiscent of wing disc cytonemes, although they grow from the (Hh) signal-producing cells towards the receiving cells. Under conditions of low levels of Hh protein and/or Hh signalling within the niche, cap cells emit up to 6-fold longer Hh-decorated cytonemes towards the signalling-deficient area of the niche. Our data reveal that stem cell niches are dynamic structures that can sense, and react to, changes in the activity of essential stem cell factors to prevent stem cell differentiation.
| Stem cells are responsible for the integrity of tissues during growth, ageing, and repair. They reside in specialised microenvironments, or niches, which frequently comprise support cells that control stem cell self-renewal, proliferation, and differentiation [1],[2]. Stem cell niche regulation often involves short-range signalling between stem cells themselves and the surrounding microenvironment. One such short-range signal is the Hedgehog (Hh) family of proteins, which mediates homeostasis in several adult tissues, including the gastrointestinal tract, the hematopoietic system, and the vertebrate central nervous system [3]–[7]. In fact, Hh signalling dysfunction can lead to stem cell depletion or proliferative disorders such as tumourigenesis [8],[9]. However, the detailed mechanisms by which Hh acts in stem cell maintenance remain elusive.
In Drosophila females, germline stem cells (GSCs) are located at the apex of the ovary, in a structure termed the germarium that constitutes a well-defined stem cell niche. The germarium hosts three types of somatic niche cells: terminal filament cells (TFCs), cap cells (CpCs), and escort cells (ECs), which support two to three GSCs and which can be labelled with specific markers such as the bab1-Gal4 and patched-Gal4 drivers (Figure 1) [10]. The spatial organisation of the GSC niche permits direct contact between two to three CpCs and one GSC, which is anchored to the CpCs by adherens junctions [11]. In addition, approximately two ECs almost completely surround a given GSC [12]. The coordinated action of GSCs and their support cells allows continuous egg production during adulthood. Thus, GSCs normally divide asymmetrically to produce a differentiating cystoblast and a lineage-renewing GSC daughter [13]. Cystoblasts divide four times to give rise to 2-, 4-, 8-, and 16-cell cysts. ECs transfer the differentiating germline cystoblasts and cysts down the germarium using dynamic cytoplasmic processes [14],[15]. Germline cells in the germarium contain specialised organelles rich in membrane skeletal proteins that adopt a spherical (called spectrosome) appearance in GSCs and cystoblasts. Upon germline differentiation, the spectrosome grows in size and becomes a branched structure, termed fusome, characteristic of differentiating cysts. Hence, GSCs can be unambiguously identified by their location within the niche (in direct contact with CpCs) and by the presence of spectrosomes (Figure 1).
Reciprocal crosstalk between GSCs and support cells shapes the niche. Firstly, the size and organisation of the CpC cluster depends on proper Notch signalling between GSCs and CpCs [16]. Secondly, both the CpCs and the adjacent ECs play an important role in GSC maintenance, as they transduce the Janus kinase/Signal transducer and activator of transcription (Jak/Stat) pathway to induce the production of the Bone Morphogenetic Protein (BMP) protein Decapentaplegic (Dpp) [12],[17],[18]. Thirdly, the germline lineage activates the epidermal growth factor receptor pathway in the ECs to repress dally expression, thus limiting Dpp movement and stability [19]. Because Dpp (and another BMP homologue called Glass bottom boat [Gbb]) [20],[21] act directly on GSCs to repress differentiation and promote self-renewal [22],[23], the control of BMP activity is of the utmost importance for correct GSC niche homeostasis.
Here, we demonstrate a key role for the Hh pathway in the regulation of BMP signalling in the Drosophila female GSC niche. In addition, we found that wild-type niche support cells grow short Hh-coated filopodia that are functionally relevant for GSC maintenance. Furthermore, support cells sense dysfunctional Hh signalling within the niche and react by growing up to 6-fold longer cytonemes that help increase the range of Hh ligand spreading.
In a number of tissues, the Engrailed (En) transcription factor regulates hh expression. Because both en and hh are expressed in TFCs and CpCs (Figure 1B and 1C), and considering the importance of the Hh signalling cascade in stem cell maintenance in insects and vertebrates [24],[25], we tested whether the en/hh connection played a role in the GSC niche. To generate en-deficient germaria, we cultured adult females bearing a thermosensitive en allele (enspt) in combination with an en deficiency (enE) for 7 or 14 d at restrictive temperature (28°C; hereafter referred to as ents germaria). Compared to control germaria (enspt/CyO), which contained an average of 2.3±0.8 GSCs and 10.2±1.3 developing cysts (n = 62; 7 d) and 2.1±0.9 GSCs and 9.1±3.1 developing cysts (n = 49; 14 d), ents germaria showed a significant decrease in the average number of GSCs and cysts (1.4±1.1 GSCs and 4.3±2.9 developing cysts, n = 52, 7 d; 1.2±0.8 GSCs and 3.9±2.5 developing cysts, n = 41, 14 d). Interestingly, 28.6% of ents germaria analysed after 7 d at restrictive temperature were devoid of germline cells, which emphasised the importance of en gene function in GSC maintenance (Figure 2A–2D; Table S1). To distinguish between a requirement for en in the germline versus in the niche support cells, we abolished en function from either GSCs or niche cells by utilising two genetically null alleles, enE and en54. The removal of en from the germline did not affect oogenesis, even 3 wk after gene inactivation (n>30 for each genotype; Figure 2E). To eliminate the activity of en in TFs, CpCs, or ECs we utilised the bab1-Gal4 driver (Figure 1D). Similar to the removal of en from the germline, elimination of en from all ECs in contact with a given GSC did not yield a visible phenotype (100% of cases, n = 23; Figure 2F). However, in 67.7% (n = 37) of mosaic germaria where en function was removed from at least three clustered CpCs, we observed differentiating cysts that contained branched fusomes and showed the accumulation of the differentiation marker Orb in contact with CpCs (Figures 2G and S1), a phenotype never found in wild-type germaria. Because we did not detect increased apoptosis in mosaic germaria contaning en mutant CpCs and since these mutant cells still expressed CpC markers (Figures S1 and S2), we conclude that en is required in CpCs to prevent GSC differentiation.
The effect on the germline of removal of En from CpCs suggested the existence of one or more En-dependent niche cell signals that act on GSCs to promote their maintenance. Hh expression in TFCs and CpCs has been shown to be required for germline development [26] (Figures 1 and S3), which made Hh an excellent candidate to mediate En function in GSC maintenance. We examined the distribution of Hh in mosaic germaria that contained en mutant cells and found that en was required in a cell-autonomous fashion for strong membrane accumulation of Hh in TFCs and CpCs (81.8% of mutant cells, n = 98; Figure 3A and 3B). In addition, we established that the removal of Hh from at least three adjacent CpCs induced GSC differentiation (51.3% of cases, n = 39; Figure 3C). It has been shown that the release of the cholesterol-modified form of Hh requires the activity of the dispatched (disp) gene [27]. Interestingly, we found that the removal of disp from CpCs was also associated with the appearance of differentiating cysts within the mosaic niche, albeit at a lower frequency (31.6% of germaria with clusters of ≥3 mutant CpCs, n = 19; Figure 3D).
The incomplete penetrance of GSC differentiation in en and particularly in hh or disp mosaic niches was most likely due to non-autonomous Hh release from the remaining wild-type cells present in the niche. In fact, the larger the number of hh mutant CpCs, the fewer GSCs remained in the niche (see below and Table S2). Alternatively (or in addition), disp mutant CpCs may still be able to sustain a certain level of Hh signalling to adjacent ECs, as shown for the wing disc [27],[28]. Because the absence of either hh or disp from other niche cells, such as TFCs or ECs, did not cause a visible GSC phenotype (data not shown), and considering the requirement for Disp in cholesterol-modified-Hh release, these results strongly suggest that Hh needs to be produced in, and secreted from, CpCs to support a stable GSC population.
Hh signalling is transduced intracellularly by Hh ligand binding the Patched (Ptc) receptor in receiving cells, allowing the phosphorylation and activation of Smoothened (Smo), a G-protein-coupled receptor normally inhibited by Ptc [29]. In the germarium, Hh ligand produced in the CpCs might act on GSCs directly, indirectly via ECs, or a combination of the two. To distinguish between these possibilities, we studied the expression pattern of ptc, itself a target gene of Hh signalling, as a readout of pathway activation. Analysis of a reporter of ptc expression (ptc-lacZ) showed expression in ECs but not in CpCs or TFCs (Figure 4A). To corroborate that activation of the ptc reporter responded to the canonical Hh pathway, we removed smo from ECs to abrogate the Hh response and found that ptc-lacZ expression was largely eliminated (100% of cases, n = 43; Figure 4B). These results indicate that the Hh pathway is active only in ECs and not in CpCs or TFCs. In fact, the generation of smo− CpC clones showed no effect on GSC loss by differentiation (100% of cases, n = 25; Figure 4C), whereas the removal of smo function from larval/pupal or adult ECs induced GSC differentiation, as visualised by the appearance of branched fusomes within the mosaic niches (69.56% of cases, n = 23; Figure 4D). Finally, the generation of mutant smo germline clones using two different null alleles did not result in any visible phenotypes 7, 14, or 21 d after clone induction (100% of cases, n>39 for each genotype and time point; Figure 4E). From these observations, we conclude that Hh produced and secreted by CpCs activates Smo in ECs to elicit a response that is responsible for GSC maintenance.
In an attempt to identify the nature of this response, we measured the mRNA levels of the essential stem cell factors dpp and gbb in Hh-depleted germaria. Real-time quantitative PCR analysis of ents germaria showed that the levels of dpp, gbb, and hh mRNAs were reduced by more than 60% when compared to control samples (Figure S4). Because en is not expressed in ECs, and since dpp and gbb are transcribed in CpCs and ECs [17],[18],[22], our data indicate that en could regulate dpp and gbb transcription in ECs via Hh signalling. However, dpp has also been shown to be a target of En [30]. To test the possibility that en is regulating dpp and gbb transcription via hh, we analysed the amounts of dpp and gbb mRNAs in ovaries in which the Hh pathway was blocked specifically in ECs for 7 d (ptc-Gal4; UAS-smo RNAi/tub-Gal80ts ovaries). In this experimental condition, the levels of dpp and gbb mRNAs are diminished by half (Figure 4F). Furthermore, these germaria also show a significant decrease in the number of GSCs per niche (Figure S5; control, 2.7±0.5 GSCs/germarium, n = 40; experimental, 1.4±0.6 GSCs/germarium, n = 44). Finally, in order to demonstrate that the expression of dpp in ECs is essential for GSC maintenance, we analysed niches in which dpp levels were diminished specifically in ECs for 14 d (ptc-Gal4; UAS-dpp RNAi/tub-Gal80ts ovaries). We found a strong reduction in the number of GSCs due to their precocious differentiation (Figures 4G and S5; control, 2.5±0.8 GSCs/germarium, n = 28; experimental, 1.4±0.6 GSCs/germarium, n = 36). Considering that these BMP molecules are essential for GSC survival [22],[23] and that dpp is a target gene of the Hh pathway [31], our results support a model in which female GSC self-renewal requires the en-dependent production of Hh in CpCs. Upon secretion by CpCs, Hh juxtacrine signal is transmitted to the adjacent ECs, which in turn control Dpp and Gbb production to sustain GSC maintenance. The fact that the removal of hh from CpCs or smo from ECs induces a decrease in phospho-Mad levels in the germline, a direct reporter of Dpp signalling, supports this hypothesis (Figure S6). Thus, in addition to the proposed role for CpCs in ovarian niche signalling [32], ECs emerge as important regulators of niche signalling, as they not only are responsible for controlling the Jak/Stat and the EGFR pathways [12],[19] but also exert a key role in the regulation of Hh signalling.
Morphogens exert their effects over long distances, which, in the case of Hh, can be as long as 300 µm in the vertebrate limb bud [33]. In contrast, in the Drosophila ovarian niche, the Hh-receiving cells adjoin the Hh-producing cells, as ECs directly contact the CpC rosette, which limits the spread of this ligand. To investigate the mechanism by which Hh is transported within the ovarian niche, we analysed in detail the distribution of Hh in the CpCs. The Hh protein is strongly localised to the cell membrane, and in 30.1% of germaria analysed (n = 149; Figure 5A), it decorated short cellular projections 0.53 to 1.11 µm in length (0.93 µm on average) and 0.1 to 0.3 µm in diameter that formed at the CpC–EC boundaries. These narrow, filiform structures were reminiscent of the thin filopodial membranes, called cytonemes, that were initially described in the wing disc. Cytonemes are actin-rich cytoplasmic extensions thought to mediate specific morphogen signalling and to prevent inadequate diffusion of ligands [34]–[36].
In order to test the biological significance of these structures, we analysed two different experimental scenarios. First, we investigated whether these processes would respond to challenging physiological conditions such as deficient Hh niche signalling. To this end, we analysed the distribution of Hh-coated cytonemes in mosaic germaria harbouring en mutant CpCs and found that in 54% of these germaria one or two of the remaining wild-type CpCs displayed thin, Hh-labelled filopodia significantly longer than those of the controls (average size 3.1 µm, n = 50; Figure 5B and 5D). We then blocked the ability of adult ECs to respond to Hh signalling by generating smo− ECs, and we looked for long cytonemes in these mosaic niches. We found that the absence of Hh pathway transduction in ECs provoked a response from signalling CpCs in the form of long, Hh-coated cellular extensions detected in 50% of the cases analysed (average size 3.3 µm, n = 28; Figure 5C). To discard the possibility that the presence of differentiated cysts within the niche, such as those generated after removing smo from ECs, induces long, Hh-positive cytonemes, we generated CpCs mutant for the Jak/Stat kinase hopscotch, which also causes GSC differentiation [17],[18], and measured cytoneme lengths. In this condition, the cellular processes were not significantly different from those of wild-type controls (average size 1.1 µm, n = 20; Figure 5D). Altogether, these results clearly show that the GSC niche can react specifically to decreased Hh levels and/or to impaired Hh signalling by increasing the range of ligand spreading. Moreover, the extended cytonemes found in en− or smo− mosaic germaria projected towards the signalling-deficient area of the niche (Figure 5E), demonstrating that niche support cells sense, and respond directionally to, spatial signalling cues. Finally, to determine whether these cytonemes are specialised structures developed to mediate niche signalling, we studied the distribution of the adherens junction components DE-Cadherin and Armadillo in cytonemes. These proteins labelled the periphery of wild-type CpCs, delineating their round, regular shape, but were absent from their short cytonemes. Similarly, in mosaic en− or smo− niches, long cytonemes did not contain DE-Cadherin or Armadillo, which suggests that cytonemes are Hh-coated filopodia grown specifically to deliver a stem cell survival factor rather than a reflection of mere changes in cell shape (Figure S7).
Next, we wished to study cytoneme functionality by affecting their formation. Because cytonemes are rich in actin filaments [34], we reasoned that disturbing actin polymerisation in adult CpCs could have an effect specifically on cytoneme production and/or kinetics. Thus, we utilised the bab1-Gal4 driver to express modified versions of two known regulators of actin polymerisation in TFCs and CpCs of adult ovaries. We induced the expression of either a constitutively activated form of the Drosophila Formin homologue Diaphanus (Dia), DiaCA [37], or a myristoylated form of the Arp2/3-complex regulator Wasp, WaspMyr [38]. While interfering with actin polymerisation may affect other cellular processes rather than cytoneme formation, we performed several controls to make sure that the observed results where as specific as possible. First, we measured the mean value of fluorescence intensity per area unit in control (tub-Gal80ts/+; UAS-diaCA/+ or tub-Gal80ts/+; UAS-waspMyr/+) or experimental ovaries (tub-Gal80ts/+; UAS-diaCA/bab1-Gal4 or tub-Gal80ts/+; UAS-waspMyr/bab1-Gal4) kept at 31°C for 5 d upon eclosion to confirm that overexpression of UAS-waspMyr or UAS-diaCA in adult germaria affected significantly neither the overall amounts of Hh protein in the niche cells nor the expression of CpC markers such as bab1 or Lamin C (Figure 6 and data not shown). Second, we manipulated only post-mitotic cells to prevent unwanted effects during mitosis, as we induced ectopic gene expression in adult CpCs. Third, we utilised an experimental setting that did not affect visibly niche morphology or CpC viability. In this scenario, we found that ectopic expression of WaspMyr or DiaCA for 5 d in niche cells halved the number of germaria growing short cytonemes (from over 30% in controls to 13.6% and 15.4%, respectively, n>36 for each genotype; Figure 6C). Interestingly, this condition also produced a significant decrease in the number of GSCs per niche (from 2.45±0.7 in controls to 1.8±0.55 and 1.7±0.6, respectively, n>36 for each genotype; Figure 6D). Since we did not observe apoptosis above control levels in germline cells (data not shown) and because we could detect differentiating cysts in these experimental niches (Figure 6B), the formation of short Hh-decorated filopodia in CpCs is an essential step to prevent GSC differentiation. We next tested whether diminishing the number of cytonemes per CpC would affect Hh signalling. To this end, we overexpressed DiaCA in adult TFCs and CpCs utilising the bab1-Gal4 driver and monitored the activation of the Hh-signalling reporter ptc-lacZ. We found that, in contrast to controls, experimental females grown for 5 d at 31°C largely failed to activate the ptc-lacZ reporter in the germarium (Figure 6E and 6F). These results, together with our previous finding that Hh is produced in CpCs and received in ECs, strongly suggest that the Hh-coated cytonemes regulate Hh signalling in the germarium by facilitating Hh delivery to the target ECs to ensure that a normal pool of stem cells is maintained.
Niches are dynamic systems often containing stromal cells that provide physical support and survival factors to nurture a population of stem cells. The data presented here demonstrate that the heterotypic association of support cells is crucial for niche function. In the case of the Drosophila ovarian niche, it has been previously described that the Jak/Stat pathway regulates the expression of dpp in CpCs [17],[18]. Our results show that the maintenance of a stable population of GSCs relies also on the coordinated action of the CpCs and the ECs, which allows the production and release of the GSC survival ligand Hh in the CpCs and its reception in the ECs. As a consequence of the transduction of the Hh pathway, ECs produce the stem cell factors Dpp and Gbb (see model in Figure 7). The recent finding of a similar partnership between mesenchymal and haematopoietic stem cells that operates in the bone marrow niche [39] indicates that such collective regulatory interactions within support cells may be a common feature of cellular niches.
The study of the mechanisms behind Hh signalling in the Drosophila ovary has allowed the identification of Hh-coated cytonemes in a cellular stem cell niche, emphasising the idea that cytonemes mediate spreading of the activating signal from the producing cells. Recently, it has been reported that the Hh protein localises to long, basal cellular extensions in the wing disk [40]. In addition, filopodial extensions in the wing, eye, and tracheal system of Drosophila have been shown to segregate signalling receptors on their surface, thus restricting the activation of signalling pathways in receiving cells [36]. Hence, cytonemes, as conduits for signalling proteins, may be extended by receiving cells—and so are involved in uptake—or may be extended by producing cells—and so are involved in delivery and release.
Interfering with actin polymerisation in adult niches leads to a significant reduction in the number of CpCs growing Hh cytonemes, concomitant with precocious stem cell differentiation, demonstrating that these actin-rich structures are required to prevent stem cell loss and thus are functionally relevant. Importantly, because we disturbed actin dynamics in post-mitotic CpCs that still produce wild-type levels of Hh protein and express CpC markers (but fail to activate the Hh pathway in ECs), the observed effects on stem cell maintenance are most likely specific to Hh delivery from CpCs to their target ECs via short cytonemes. This interpretation is further reinforced by the observation that CpCs can sense decreased Hh levels and/or a dysfunction in the transduction of the Hh pathway in the niche and respond to it by growing Hh-rich membrane bridges up to 6-fold longer than in controls. In this regard, it is interesting to note that the two lipid modifications found in mature Hh act as membrane anchors and give secreted Hh a high affinity for membranes and signalling capacities [41],[42]. In fact, it has been recently described that a lipid-unmodified form of Hh unable to signal does not decorate filopodia-like structures in the wing imaginal disc epithelium, confirming the link between Hh transport along cytonemes and Hh signalling [40]. Thus, cytonemes may ensure specific targeting of the Hh ligand to the receiving germline cells in a context of intense signalling between niche cells and the GSCs. Interestingly, in both en− and smo− mosaic niches, the long processes projected towards the signalling-deficient area of the niche, which showed that competent CpCs sense the strength of Hh signalling activity in the microenvironment. While the nature of the signal perceived by the CpCs or the receptor(s) involved in the process are unknown, we postulate that Hh-decorated filopodial extensions represent the cellular synapsis required for signal transmission that is established between the Hh-producing cells (the CpCs) and the Hh-receiving cells (the ECs). In this scenario—and because Ptc, the Hh receptor, is a target of the pathway—the membranes of mutant ECs, in which the transduction of the pathway is compromised, contain lower Ptc levels. Thus, longer and perhaps more stable projections ought to be produced to allow proper signalling. In addition, the larger the number of en mutant cells (and hence the stronger the deficit in Hh ligand concentration or target gene regulation), the longer the cellular projections decorated with Hh (Tables S3 and S4), which indicates that the niche response is graded depending on the degree of signalling shortage.
Do the longer cytonemes found in mosaic germaria represent structures created de novo, or do they simply reflect a pre-existing meshwork of thin intercellular bridges that can regulate the amount of Hh protein in transit across them? Because we utilised an anti-Hh antibody to detect the cytonemes and all of our attempts to identify other markers for these structures have failed, we cannot presently discriminate between these two possibilities. In any case, since we did not detect increased Hh levels in wild-type CpCs that contained cytonemes relative to those that did not, it is clear that long filopodia do not arise solely by augmenting Hh production in the CpCs. Rather, if long cytonemes are not synthesised in response to a Hh signalling shortage and if they already existed in the niche, they ought to restrict Hh spreading independently of significant Hh production. Furthermore, because the strength of Hh signalling in the niche determines the distance of Hh spreading, either cytoneme growth or Hh transport (or both) are regulated by the ability of the CpCs to sense the Hh signalling output.
Our demonstration that a challenged GSC niche can respond to insufficient signalling by the cytoneme-mediated delivery of the stem cell survival factor Hh over long distances has wider implications. Niche cells have been shown to send cellular processes to their supporting stem cells in several other scenarios: the Drosophila ECs of the ovary and the lymph gland, the ovarian niche of earwigs, and the germline mitotic region in the hermaphrodite Caenorhabditis elegans [5],[6],[14],[43],[44]. Similarly, wing and eye disc cells project cytonemes to the signalling centre of the disc [34],[36]. However, definitive proof that the thin filopodia described in the lymph gland, the earwig ovary, or imaginal discs deliver signals from the producing to the effector cells is lacking. Our findings strongly suggest that cytonemes have a role in transmitting niche signals over distance, a feature that may underlie the characteristic response of more complex stem cell niches to challenging physiological conditions. Careful analysis of the architecture of sophisticated niches, such as the bone marrow trabecular zone for mouse haematopoietic stem cells, will be needed to further test this hypothesis and to determine whether it represents a conserved mechanism for stem cell niche signalling.
Flies were grown at 25°C on standard medium for Drosophila. The following genetically null alleles were used: enE, en54 [45], hh21, hhAC [46], dispSH21 [28], smoD16 [47], and smo3 [48]. ptcAT96 is a LacZ enhancer trap inserted in the gene [49]. enspt [50] is a temperature-sensitive allele. To express UASt-DsRed and UASt-flp we used the bab1-Gal4 line [51]. The expression of UAS transgenes in ECs was done utilising the ptc-Gal4 driver. In order to generate experimental enspt/enE adult females, flies were shifted from 25°C to 28°C for 7 or 14 d upon eclosion.
To obtain adult females overexpressing WaspMyr [38] or DiaCA [37], w; tub-Gal80ts/CyO; bab1-Gal4/TM2 flies were crossed to w; UAS-waspMyr or w; UAS-diaCA, respectively. To overexpress dpp RNAi (VDRC) or smo RNAi (Bloomington Stock Center) in ECs, w; ptc-Gal4, UAS-GFP; tub-Gal80ts/SM6∧TM6B flies were crossed to w; UAS-dpp RNAi, w; UAS-smo RNAi. The offspring were grown at 18°C, and upon eclosion adult F1 flies were shifted to 31°C for 5, 7, or 14 d.
Ovaries were dissected at room temperature in PBS containing 0.1% Tween-20 (PBT), fixed for 20 min with 4% PFA, blocked with PBT+10% BSA for 1 h, and washed in PBT before they were incubated for 15 h with primary antibodies diluted in PBT supplemented with 1% BSA. Primary antibodies were washed three times in PBT containing 1% BSA. Secondary antibodies were diluted in PBT containing 0.1% BSA. Primary antibodies were used at the following concentrations: mouse anti-Hts (1B1) (Developmental Studies Hybridoma Bank [DSHB], University of Iowa), 1∶50; rabbit anti-Vasa (a gift from R. Lehmann), 1∶1,000; mouse anti-En (4D9) (DSHB), 1∶50; rabbit anti-α-Spectrin (a gift from R. Dubreuil), 1∶400; rabbit anti-GFP (Molecular Probes), 1∶500; mouse anti-GFP (Molecular Probes), 1∶50; mouse anti–Lamin C (LC28.26) (DSHB), 1∶50; rabbit anti-Hh (a gift from S. Eaton [52]), 1∶500; rabbit anti-phospho-Mad 1/5/8 (a gift from E. Laufer), 1∶5,000; rabbit anti-β-galactosidase (Cappel), 1∶1,000; rabbit anti-cleaved Caspase 3 (BioLabs), 1∶50; and mouse anti-Orb (6H4+4H8) (DSHB), 1∶50. Secondary antibodies (Cy2- and Cy3-conjugated, Jackson ImmunoResearch) were used at 1∶100. DNA staining was performed using the DNA dye Hoechst (Sigma) at 1∶1,000. Images were captured with a Leica SPE confocal microscope and processed using ImageJ, Adobe Photoshop, and Adobe Illustrator.
Fluorescence intensity units and cytoneme length were measured using the Leica LAS-AF software. Images were captured with a Leica SPE confocal microscope and processed using ImageJ, Adobe Photoshop, and Adobe Illustrator.
To generate mitotic clones we induced the Flipase enzyme using either a heat shock promoter or the bab1-Gal4 driver to activate expression of a UAS-flp construct. en and smo mutant germline clones were induced by giving 3-d-old females two 1-h-long heat shocks at 37°C spaced by 10 h at 25°C. hh, disp, en, and smo mutant somatic clones were induced expressing UAS-flp with the bab1-Gal4 driver. Ovaries were processed 3 d (for somatic clones) or 7, 14, or 21 d (for germline clones) after treatment. To eliminate smo function in adult females, 3-d-old HS-flp1112/+; smoD16 FRT40A/ubi-nls:GFP FRT40A flies were subjected to three 1-h-long heat shocks at 37°C separated by 6-h periods at 25°C. The following chromosomes were used: HS-flp1112, FRT42D en54, FRT42D enE, FRT42D ubi-nls:GFP, smoD16 FRT40A, ubi-nls:GFP FRT40A, FRT42D ubi-nls:GFP, hhAC FRT82B, hh21 FRT82B, dispSH21 FRT82B, bab1-Gal4 FRT82B ubi-nls:GFP, UASt-flp, smo3 FRT40A ptcAT96, and bab1-Gal4 UASt-flp.
The relative amounts of hh, dpp, and gbb mRNAs were determined by real-time quantitative PCR using the comparative cycle threshold (CT) method [53], Fam-dye-labelled TaqMan MGB probes (Applied Biosystems), and an ABI-PRISM 7700 Sequence Detection System. RNA polymerase II (RpII140) was used to normalise mRNA levels. hh, dpp, or gbb mRNA relative amount was calculated from the determination of the difference between the CT of the given gene and that of RpII140. CT values used were the result of three different replicas from three independent experiments. Primers and TaqMan probes for the different cDNAs were obtained from the Assays-by-Design Service (Applied Biosystems) with the following sequences (5′–3′): RpII140, forward, ACTGAAATCATGATGTACGACAACGA, reverse, TGAGAGATCTCCTCGGCATTCT, probe, TCCTCGTACAGTTCTTCC; hh, forward, GCAGGCGCCACATCTACT, reverse, GCACGTGGGAACTGATCGA, probe, CCGTCAAGTCAGATTCG; dpp, forward, GCCAACACAGTGCGAAGTTTTA, reverse, TGGTGCGGAAATCGATCGT, probe, CACACAAAGATAGTAAAATC; gbb, forward, CGCTGTCCTCGGTGAACA, reverse, CGGTCACGTTGAGCTCCAA, probe, CCAGCCCACGTAGTCC.
cDNA was synthesised from ∼100–200 ovary pairs of the following characteristics: enspt/CyO (control) and enspt/enE (experimental) females were shifted from 25°C to 28°C for 7 d after eclosion prior to dissection. +; UAS-smo RNAi/SM6∧TM6B (control) and ptc-Gal4, UAS-GFP/+; UAS-smo RNAi/tub-Gal80ts (experimental) females were shifted from 25°C to 31°C for 7 d after eclosion prior to dissection.
A Student's t test was used to determine whether the following were significantly different between control and experimental samples: (i) the mean number of GSCs and differentiated cysts per germarium, (ii) the relative levels of hh, dpp, and gbb expression, and (iii) the length of cytonemes. To analyse whether the observed differences in the percentages of cytoneme-containing germaria between control ovaries and ovaries overexpressing WaspMyr or DiaCA were significant, we applied the Chi-square test. Differences were considered significant when the p-values were less than 0.01.
|
10.1371/journal.ppat.1005976 | Evolution of Fitness Cost-Neutral Mutant PfCRT Conferring P. falciparum 4-Aminoquinoline Drug Resistance Is Accompanied by Altered Parasite Metabolism and Digestive Vacuole Physiology | Southeast Asia is an epicenter of multidrug-resistant Plasmodium falciparum strains. Selective pressures on the subcontinent have recurrently produced several allelic variants of parasite drug resistance genes, including the P. falciparum chloroquine resistance transporter (pfcrt). Despite significant reductions in the deployment of the 4-aminoquinoline drug chloroquine (CQ), which selected for the mutant pfcrt alleles that halted CQ efficacy decades ago, the parasite pfcrt locus is continuously evolving. This is highlighted by the presence of a highly mutated allele, Cam734 pfcrt, which has acquired the singular ability to confer parasite CQ resistance without an associated fitness cost. Here, we used pfcrt-specific zinc-finger nucleases to genetically dissect this allele in the pathogenic setting of asexual blood-stage infection. Comparative analysis of drug resistance and growth profiles of recombinant parasites that express Cam734 or variants thereof, Dd2 (the most common Southeast Asian variant), or wild-type pfcrt, revealed previously unknown roles for PfCRT mutations in modulating parasite susceptibility to multiple antimalarial agents. These results were generated in the GC03 strain, used in multiple earlier pfcrt studies, and might differ in natural isolates harboring this allele. Results presented herein show that Cam734-mediated CQ resistance is dependent on the rare A144F mutation that has not been observed beyond Southeast Asia, and reveal distinct impacts of this and other Cam734-specific mutations on CQ resistance and parasite growth rates. Biochemical assays revealed a broad impact of mutant PfCRT isoforms on parasite metabolism, including nucleoside triphosphate levels, hemoglobin catabolism and disposition of heme, as well as digestive vacuole volume and pH. Results from our study provide new insights into the complex molecular basis and physiological impact of PfCRT-mediated antimalarial drug resistance, and inform ongoing efforts to characterize novel pfcrt alleles that can undermine the efficacy of first-line antimalarial drug regimens.
| Point mutations in the Plasmodium falciparum chloroquine resistance transporter (PfCRT) earlier thwarted the clinical efficacy of chloroquine, the former gold standard, and constitute a major determinant of parasite susceptibility to antimalarial drugs. Recently, we reported that the highly mutated Cambodian PfCRT isoform Cam734 is fitness-neutral in terms of parasite growth, unlike other less fit isoforms such as Dd2 that are outcompeted by wild-type parasites in the absence of CQ pressure. Using pfcrt-specific zinc-finger nucleases to genetically dissect the Cam734 allele, we report that its unique constituent mutations directly contribute to CQ resistance and collectively offset fitness costs associated with intermediate mutational steps. We also report that these mutations can contribute to resistance or increased sensitivity to multiple first-line partner drugs. Using isogenic parasite lines, we provide evidence of changes in parasite metabolism associated with the Cam734 allele compared to Dd2. We also observe a close correlation between CQ inhibition of hemozoin formation and parasite growth, and provide evidence that Cam734 PfCRT can modulate drug potency depending on its membrane electrochemical gradient. Our data highlight the capacity of PfCRT to evolve new states of antimalarial drug resistance and to offset associated fitness costs through its impact on parasite physiology and hemoglobin catabolism.
| Human malaria remains a leading global health scourge in part due to multidrug resistance mechanisms evolved by Plasmodium falciparum, the protozoan species responsible for the most severe forms of disease [1]. Artemisinin-based combination therapies (ACTs) are the current first-line means of controlling pathogenic asexual blood-stage infections, including ones dominated with drug-resistant strains that arose during previous selective sweeps resulting from the global use of chloroquine (CQ) and sulfadoxine-pyrimethamine [2–4]. The 4-aminoquinoline compound CQ was especially pivotal earlier in reducing mortality rates [5]. However, the multi-focal emergence and spread of CQ resistance (CQR) contributed to stalled control measures and substantial increases in malaria-associated hospitalizations and deaths [6]. Nevertheless, owing to its safety, affordability, and established efficacy against non-resistant parasites, CQ continues to be deployed in regions that are free of CQR or that harbor CQ-sensitive P. vivax [7]. Interestingly, studies of infections with CQ-resistant P. falciparum strains in Guinea-Bissau recently revealed a ~5–fold increase in CQ efficacy upon doubling the standard dose in children aged <5 years, the age demographic at highest risk for malaria mortality [8]. These findings coincide with renewed efforts to delineate the molecular basis of resistance to antimalarials bearing the hallmark CQ-type quinoline moiety [9].
Genetic linkage and allelic replacement studies have previously identified pfcrt variants as the primary determinant of CQR [10,11]. These findings are supported by evidence of directional selection for mutant pfcrt alleles in P. falciparum parasite populations subjected to extensive CQ pressure [12]. A secondary, strain-dependent contribution to CQR has also been noted for the P. falciparum multidrug resistance 1 (pfmdr1) gene [13–15]. Among CQ-resistant field isolates, PfCRT isoforms are comprised of geographically distinct clusters of single-nucleotide polymorphisms (SNPs), namely K76T and 3 to 8 additional point mutations. PfCRT K76T is a critical, albeit insufficient, determinant of parasite in vitro CQR [16]. This mutation also predicts in vivo CQ treatment failure with high sensitivity but lower specificity [17]. At the cellular level, PfCRT is a multi-pass transporter embedded in the intra-erythrocytic parasite’s digestive vacuole (DV) membrane, with enigmatic functions that may include transport of ions and/or peptides [18–21]. In the absence of PfCRT structural information, mutational approaches have guided studies into the effect of specific PfCRT mutations on drug transport and parasite growth [16,22–24].
Point mutations in PfCRT have also been associated with altered parasite susceptibility to ACT component drugs, namely artemisinins and their partner drugs (including amodiaquine, lumefantrine, and piperaquine) [25–29]. This is of particular relevance given reports of emerging clinical resistance to these first-line agents [30,31]. To various degrees, these compounds interfere with or are otherwise impacted by parasite-mediated catabolism of host hemoglobin (Hb), which supplies parasites with amino acids and helps maintain intracellular osmolarity [32–35]. This catabolic process produces ferriprotoporphyrin IX heme, which in its reactive free form can exert lethal oxidative damage to the parasite [36]. For quinoline-based antimalarials, drug-heme interactions in the DV cause toxicity by preventing incorporation of ferriprotoporphyrin IX heme dimers (β-hematin) into the non-reactive hemozoin (Hz) crystals that account for >95% of total heme [37,38]. Consistent with this inhibition of β-hematin mineralization and detoxification, CQ treatment of drug-sensitive D10 parasites was recently observed by transmission electron microscopy to disrupt the highly ordered fringe pattern of Hz crystals [32].
Cell fractionation methods in P. falciparum D10 parasites have further demonstrated that, upon CQ treatment, the proportion of total heme present as Hz significantly diminishes, whereas the proportion corresponding to free heme increases [32]. These responses are dose-dependent and inversely proportional to parasite survival [32]. As a weak base and lipophilic drug, CQ traverses multiple lipid bilayers and accumulates as CQ2H+ up to a thousand-fold in the acidic DV, where it binds hematin, hemozoin, or both [39]. CQR-promoting PfCRT isoforms appear to efficiently transport CQ out of the DV, consequently restricting CQ-heme contacts and allowing Hz formation to proceed [7,21]. Of note, the mutational status of pfcrt can also impact DV volume and pH, both of which influence Hz formation kinetics [40].
Interestingly, recent metabolomic analyses of CQ-resistant versus CQ-sensitive P. falciparum strains detected a link between mutant pfcrt-mediated CQR and the elevated accumulation of peptides derived from Hb digestion [41]. Given the reduced growth of CQ-resistant parasites (expressing the Dd2 or 7G8 mutant pfcrt alleles) relative to recombinant isogenic parasites encoding wild-type pfcrt, defective Hb degradation was postulated as a cellular basis for the reduced fitness associated with mutant pfcrt [41]. Reduced fitness of these mutant alleles was confirmed in in vitro cell culture studies [41,42] and was observed at a population level in Africa, where the removal of CQ pressure led to the attrition of mutant pfcrt-expressing parasites in favor of wild-type, CQ-sensitive strains [43,44].
A pathogen’s fitness refers to its capacity to support infection and generate new progeny. For P. falciparum parasites, fitness is influenced in part by the rate of growth of pathogenic asexual blood-stage parasites, selective forces exerted by drug pressure, mosquito-human transmission, and selection within the mosquito vector [41,44,45]. In general, these factors are impaired in parasites expressing mutant, CQR-associated PfCRT isoforms [7,46]. To date, over 50 distinct PfCRT haplotypes have been reported [22]. Of these, the Asian haplotype Dd2 (M74I/N75E/K76T/A220S/Q271E/N326S/I356T/R371I) and the South American/Western Pacific haplotype 7G8 (C72S/K76T/A220S/N326D/I356L) account for a large proportion of global mutant types, with additional isoforms harboring four or more SNPs and resembling the PfCRT haplotypes Dd2 or 7G8 (notably the six-SNP African variant GB4 or the four-SNP South American variant Ecu1110 respectively) [7]. Recent modeling in P. falciparum suggests that pfcrt evolution occurred via punctuated periods of mutation that were too brief to allow fixation of partially mutated alleles (i.e. bearing 1 to 3 SNPs), shedding light on physiologic constraints that explain the rarity of mutant pfcrt emergence in the field [16]. Intriguingly, studies from Cambodia, an epicenter of multidrug resistance in P. falciparum, also revealed a highly polymorphic CQR-conferring pfcrt allele, Cam734 [47]. This allele encodes nine mutations (see Table 1), five of which (N75D, A144F, L148I, I194T, T333S) are not found in the predominant Southeast Asian CQ-resistant pfcrt allele, Dd2 [22]. After Dd2, Cam734 pfcrt represents the second most prevalent allele in Southeast Asia [27]. Unlike other CQR-associated isoforms such as Dd2, the Cam734 allele has been found to be fitness-neutral in that it supports parasite growth comparable to recombinant pfcrt-edited parasites (engineered on the same strain, and referred to as isogenic) that encode the CQ-sensitive, wild-type pfcrt allele [42].
This unique Cam734 PfCRT isoform presents an opportunity to explore P. falciparum genetic determinants that concurrently confer drug resistance and fully neutralize fitness costs, a unique feature not associated with other mutant PfCRT variants. Herein, we leveraged drug resistance versus growth profiling of isogenic, pfcrt-modified asexual blood-stage parasites. These studies were combined with biochemical approaches—including metabolomic, heme fractionation, and heterologous expression studies—in order to address the following questions: (1) To what extent do the mutations unique to Cam734 PfCRT directly impact parasite resistance to clinically employed antimalarials? (2) Which mutations are compensatory and thus serve to preserve PfCRT function? and (3) Mechanistically, how does mutant Cam734 PfCRT confer CQR without an accompanying fitness cost? Results provided herein broaden our present understanding of the mechanistic basis of CQR and inform field efforts that evaluate pfcrt genotypes as a tool to predict the drug susceptibility status of clinical isolates.
To dissect the contributions of the rare mutations comprising Cam734 PfCRT to parasite drug resistance and fitness, we utilized a recently established [48] gene-editing approach (S1 Fig) with pfcrt-specific zinc-finger nucleases (ZFNs). Starting with the GC03 strain, a CQ-sensitive progeny of the HB3×Dd2 genetic cross [49], we engineered isogenic parasites encoding full-length Cam734 pfcrt (GC03Cam734; the PfCRT haplotype of recombinant lines is listed in superscript) as well as partial Cam734-like isoforms containing “back-to-wild-type” mutations at PfCRT residues 75, 144, 148, 194, and 333 (see Table 1). Our parasite panel also included the GC03Dd2 and GC03GC03 lines, which encode the mutant Dd2 (CQ-resistant) and wild-type GC03 (CQ-sensitive) haplotypes, respectively and which were similarly engineered using ZFN-based editing (see Table 1). This ZFN approach enables the expression of only full-length pfcrt, and is thus a significant improvement over the prior allelic exchange method [11] that in addition to expressing full-length pfcrt also generated truncated fragments that created the possibility of internal recombination events. We note that the host GC03 strain has been used in multiple prior pfcrt allelic exchange and gene editing studies [11,16,42,48,50], providing extensive background data on parasite drug susceptibilities, transport properties, transcriptional changes, drug-heme interactions and metabolomics [41,51–59]. For each recombinant line, two independent clones were selected, and pfcrt sequence integrity was verified using sequencing primers listed in S1 Table. Recombinant parasite design and validation, including PCR-based confirmation of the recombinant locus, cDNA sequencing that showed error-free editing, and Western blot analysis confirming equivalent expression, is further detailed in S1 Fig and Supplementary Materials and Methods.
We examined the roles of Cam734 PfCRT-constituent mutations in mediating CQR by assessing the responses of recombinant, pfcrt-modified parasites to CQ and its clinically relevant metabolite, monodesethyl-CQ (md-CQ). Md-CQ was included in our analysis as it shows a greater distinction between CQ-sensitive and CQ-resistant lines and may have been the primary evolutionary selective agent [11,16]. For all drug assays, genetically unmodified Dd2 and GC03 parasites were included as reference lines (see Table 1). Using 72 h flow cytometry-based drug susceptibility assays, we determined antimalarial drug concentrations that result in 50% (IC50) and 90% (IC90) inhibition of parasite proliferation (S2 Table). Both values have earlier proven informative in defining CQ susceptibility phenotypes, particularly in cases of low-level resistance or tolerance [50]. Statistical comparisons were performed against GC03Cam734 parasites, which express the full-length Cam734 pfcrt allele. In our analysis, recombinant GC03 parasites encoding the major Southeast Asian pfcrt variants Cam734 and Dd2 conferred moderate (~5-fold and ~14-fold) and high-level (~26-fold and ~43-fold) increases in CQ and md-CQ IC50 values, respectively, when compared to CQ-sensitive GC03GC03 parasites (S2 Table). This is consistent with earlier P. falciparum drug susceptibility studies [42].
CQ (Fig 1A; S2 Table) and md-CQ (Fig 1B; S2 Table) susceptibility profiles revealed significant roles for multiple Cam734 PfCRT-defining mutations in conferring CQR. Among these, PfCRT A144F was indispensable for CQ and md-CQ resistance (compare GC03Cam734 with the GC03Cam734 F144A line). Indeed, removal of this mutation restored complete to near-complete sensitivity to CQ and md-CQ respectively, at IC50 and IC90 levels (S2 Table). The Cam734 F144A PfCRT haplotype (see Table 1) is equivalent to Cam738 PfCRT, which, like Cam734, was initially documented in Cambodia but, in contrast, did not achieve wide regional spread [47]. Notably, although the Cam734 F144A haplotype harbors K76T and seven additional mutations, its mutational configuration is nevertheless insufficient for CQR (see Fig 1A and S2 Table). This underscores the notion of K76T as an insufficient predictor of CQR status and supports the fact that CQ resistance or susceptibility can depend on additional PfCRT substitutions (e.g. C101F, L272F, C350R, or in this case A144F), even when K76T is present [22,60–62].
Our analysis identified additional contributory roles for PfCRT mutations N75D, L148I, and T333S in conferring parasite resistance to CQ (Fig 1A; S2 Table) and md-CQ (Fig 1B; S2 Table), as parasite lines encoding back-mutations at each of the corresponding PfCRT residues demonstrated significant reductions in resistance (range of 1.5 to 3.7-fold reductions in CQ or md-CQ IC50 values for GC03Cam734 D75N, GC03Cam734 I148L, and GC03Cam734 S333T as compared to GC03Cam734 parasites). Thus, to various degrees, N75D, A144F, L148I, and T333S directly contribute to CQR and are not merely compensatory in nature in terms of restoring function or fitness to mutant PfCRT isoforms. Only the I194T mutation was found to not significantly contribute to CQR.
A defining molecular feature of P. falciparum CQR is resistance reversal by the calcium channel blocker verapamil (VP) [52]. The extent to which VP modifies CQR, referred to as the CQ response modification index (RMI), is PfCRT isoform-specific and is calculated by dividing the IC50 value for CQ in the presence of 0.8 μM VP by the IC50 value for CQ alone [63]. Our CQR reversibility results are depicted in Fig 1A and S3 Table. Consistent with previous studies [42], isogenic parasites expressing PfCRT variants Cam734 (GC03Cam734) and Dd2 (GC03Dd2) exhibited moderate and high-level CQR reversibility (2.4-fold and 8.1-fold reductions in the CQ RMI, respectively, versus the GC03GC03 line that showed no CQR reversibility). Among the recombinant lines encoding partial, back-to-wild-type Cam734 PfCRT haplotypes, GC03Cam734 D75N and GC03Cam734 I148L parasites exhibited statistically significant increases in CQ RMI values as compared to GC03Cam734 parasites, highlighting critical roles for mutations N75D and L148I in the VP reversibility effect. These findings uncover a novel role for PfCRT residue 148 in mediating VP reversal of CQR and align with previous studies that implicate an important role for mutations at residue 75 in mediating this reversal phenotype [42,64].
PfCRT variants can modulate parasite susceptibility to a host of antimalarials beyond CQ [7]. We consequently assessed the effects of Cam734 PfCRT-constituent mutations on parasite responses to various clinically employed antimalarials (Fig 1 and S2 Fig). Our drug panel consisted of the following: (1) ACT partner drugs, including monodesethyl-amodiaquine (md-AQ, the active metabolite of AQ), lumefantrine (LUM), piperaquine (PPQ), and pyronaridine (PND); (2) the ACT artemisinin derivative artesunate (AS); and (3) quinine (QN), a second-line agent used to treat severe malaria. In keeping with known cross-resistance relationships between CQ and AQ [7], we observed significant reductions in md-AQ resistance (Fig 1B; S2 Table) among parasites harboring reversions of Cam734 PfCRT mutations N75D, A144F, L148I, and T333S (range of 1.4 to 4.0-fold reductions in md-AQ IC50 values for GC03Cam734 D75N, GC03Cam734 F144A, GC03Cam734 I148L, and GC03Cam734 S333T as compared to GC03Cam734 parasites). Reversion of the PfCRT mutation A144F back to wild-type (compare GC03Cam734 with GC03Cam734 F144A) was likewise associated with a significant (~1.6-fold) reduction in QN IC50 values (Fig 1B; S2 Table), emphasizing A144F as a critical determinant of parasite resistance to multiple quinoline-type antimalarials. We further detected a modest (~1.5-fold), although statistically significant, increase in PPQ IC50 for GC03Cam734 I148L parasites, as compared to GC03Cam734 parasites. This highlights the capacity of PfCRT mutations to impact parasite PPQ resistance, a rising problem in Southeast Asia with a presently unclear genetic basis [30,65,66]. As compared to wild-type (GC03) pfcrt, full-length Cam734 sensitized parasites to the antimalarial compounds LUM, AS, and PND, and this phenotype was not modulated by any of the Cam734-constituent mutations studied herein (S2 Fig; S2 Table).
Previous efforts to disrupt the pfcrt gene demonstrated that it is essential for survival of asexual blood-stage P. falciparum parasites [67]. Furthermore, PfCRT mutations can be deleterious to parasite growth [16,42], a measurable phenotype that serves as a proxy for fitness and reflects parasite functional requirements [68]. To evaluate the contributions of Cam734-unique PfCRT mutations to parasite growth, we used previously established co-culture methods to derive relative growth estimates [16,69]. Co-cultures consisted of equal proportions of a pfcrt-modified GFP-negative (GFP−) test line and a wild-type pfcrt-expressing GFP-positive (GFP+) reporter line (see Materials and Methods). To assess the impact of a sub-therapeutic dose of CQ on parasite growth, experiments were performed in the absence or presence of 7.5 nM CQ (~0.5× CQ IC50 of CQ-sensitive GC03GC03 parasites; see S2 Table). These co-cultures were monitored for 10 generations, with the GFP− proportion of the co-culture determined every 48 h generation by flow cytometry (S3 Fig). As detailed in Materials and Methods and Supplementary Materials and Methods, these data were used to derive the per-generation selection coefficient (s) of each test line (Fig 2; S4 Table). This coefficient serves as a proxy for the degree of fitness of a given parasite line, as compared to GC03Cam734 parasites, which encode the full-length Cam734 pfcrt allele (s = 0, s>0, and s<0 indicate fitness levels equal, greater than, or less than that of GC03Cam734 parasites).
As determined in our in vitro growth assays (Fig 2; S4 Table), the growth rate of parasites encoding Cam734 pfcrt was markedly increased relative to isogenic parasites encoding mutant Dd2 pfcrt (s = -0.25 for GC03Dd2 versus GC03Cam734) and was more comparable to growth of parasites encoding wild-type GC03 pfcrt (s = 0.06 for GC03GC03 versus GC03Cam734). These findings agree with earlier studies of isogenic, pfcrt-modified lines constructed via an independent allelic exchange strategy [42]. Remarkably, growth of GC03Cam734 parasites was significantly enhanced (~1.2–fold; Fig 2; S4 Table) by the presence of a sub-lethal dose of CQ (7.5 nM), surpassing the growth rates of all other parasite lines, including those expressing GC03 pfcrt. Our results further demonstrate that the Cam734 PfCRT-constituent mutations N75D, A144F, L148I, I194T, and T333S distinctly contribute to the growth of asexual blood-stage parasites expressing full-length Cam734 PfCRT, as the corresponding partial, back-mutant haplotypes conferred significantly decreased relative growth rates (with s values ranging from -0.07 to -0.30 for GC03Cam734 D75N, GC03Cam734 F144A, GC03Cam734 I148L, GC03Cam734 T194I, and GC03Cam734 S333T; see S4 Table).
Among the back-mutant parasites, the most deleterious growth was observed for parasites expressing Cam734 I148L PfCRT, which was reminiscent of the growth phenotype of the Dd2 PfCRT isoform that is known to substantially impair parasite fitness (compare GC03Cam734 I148L and GC03Dd2; Fig 2; S4 Table). Considering our dissection of the roles of Cam734 PfCRT mutations in parasite CQ susceptibility and growth, our results suggest that PfCRT mutations N75D, A144F, I148L, and S333T play dual roles, directly contributing to CQR and compensating for associated fitness costs, whereas I194T has no impact on CQR and only helps improve growth.
The catabolism of host-derived Hb is an essential parasite process that liberates two major products: (1) free heme that is subsequently incorporated into crystalline Hz; and (2) free Hb-derived peptides that can contribute to the parasite’s nutrient pool [70]. Recent studies have shown that peptides derived from either the α or β chains of Hb can accumulate up to 32–fold within isogenic parasites expressing CQ-resistant Dd2 or 7G8 pfcrt alleles as compared to CQ-sensitive (wild-type) pfcrt-expressing parasites [41]. The accumulation of Hb-derived peptides in parasites expressing PfCRT Dd2 or 7G8 variants was linked to impaired Hb catabolism and was proposed to be a causal determinant of their reduced fitness, as measured using in vitro growth rates [41]. Given the unique capacity of mutant Cam734 PfCRT to neutralize fitness costs that are typically associated with CQ-resistant PfCRT isoforms, we examined whether Cam734 PfCRT mitigated the accumulation of Hb-derived peptides.
To test this, we measured endogenous metabolite levels in isogenic lines encoding the CQ-resistant pfcrt alleles Cam734 and Dd2 (GC03Cam734 and GC03Dd2). Briefly, red blood cells (RBCs) harboring late-stage (~36–42 h) trophozoites were magnetically purified, metabolites were extracted, and extracts were analyzed using established mass spectrometry-based metabolomic methods [41]. Results are depicted in Fig 3 for compound classes and S4 Fig for individual metabolites. Metabolite signal intensities and z-scores are reported in S5 and S6 Tables, respectively. Our results show that the CQ-resistant pfcrt alleles Cam734 and Dd2 accumulated comparable levels of Hb-derived peptides (P = 0.33; see Fig 3 and S6 Table). Moreover, the GC03Cam734 and GC03Dd2 peptide levels were significantly elevated relative to genetically matched GC03GC03 parasites expressing the wild-type pfcrt [41]. Consequently, peptide accumulation in these lines was not correlated with the observed differences in fitness between Cam734 and Dd2.
To better understand the impact of the Cam734 allele, we conducted a more comprehensive metabolic analysis of GC03Cam734 and GC03Dd2 parasites. This analysis revealed several distinguishing metabolic phenotypes. Most significantly, ATP to ADP and ATP to AMP ratios were significantly higher in GC03Cam734 compared to GC03Dd2 parasites (P = 0.0002 and P = 0.0001, respectively; see Fig 3 and S6 Table). Low ATP to AMP ratios are a classic marker of lower cellular energy and metabolic stress [71]; the relatively higher ratios seen in GC03Cam734 parasites are consistent with their increased in vitro growth rate as compared with GC03Dd2 (see Fig 2). GC03Cam734 parasites also exhibited significantly increased levels of glycolytic and tricarboxylic acid (TCA) cycle-associated metabolites (P = 0.003 and P = 0.01, respectively; see Fig 3 and S6 Table). These elevated central carbon metabolites may suggest that GC03Cam734 achieves its altered energy state via a metabolic compensatory mechanism. Collectively, these findings indicate that parasites encoding Cam734 and Dd2 PfCRT both suffer from impaired parasite Hb catabolism, but that the Cam734 PfCRT isoform compensates for this defect via a mechanism that may involve alterations in central carbon metabolism.
CQ treatment of P. falciparum parasites affects their disposition of the different forms (“species”) of heme, namely free heme, Hz, and Hb. In a dose-dependent manner, CQ causes an increase in toxic free heme, a decrease in the formation of chemically inert Hz crystals, and an accompanying reduction in parasite survival [32]. To date, the profiles of heme fractions have only been explored in CQ-sensitive (D10, NF54 and D6) parasites [72,73]. Given the central role of pfcrt in dictating parasite responses to CQ, we examined the composition of heme species in CQ-treated and untreated isogenic parasites encoding either the CQ-sensitive (wild-type) pfcrt allele GC03 or the CQ-resistant (mutant) pfcrt alleles Cam734 or Dd2. Our heme fractionation assay entailed treating synchronized early ring-stage parasites with CQ across a range (0 to 3×) of its IC50 values for the different lines. After 32 h, trophozoite-stage parasites were subjected to a series of cellular fractionation steps. The abundance of free heme, Hz and Hb was subsequently determined spectrophotometrically and reported both as a percentage of total heme (Fig 4) and as an amount of heme iron (Fe) per cell (S5 Fig).
Our results demonstrate that parasite exposure to CQ caused a dose-dependent increase in free heme for all pfcrt-modified parasite lines. Considering all CQ treatments, the highest accumulation of free heme was observed in the CQ-sensitive GC03GC03 parasites, with maximal mean free heme Fe concentrations of 19.3, 12.6, and 10.5 femtograms (fg) per cell observed in GC03GC03, GC03Dd2, and GC03Cam734 parasites respectively (Fig 4A and S5A Fig). The difference between GC03GC03 and GC03Cam734 parasites achieved statistical significance (P = 0.02 by unpaired t test with Welch’s correction). Inversely correlating with free heme profiles, Hz amounts showed comparable, CQ dose-dependent decreases (Fig 4B and S5B Fig).
We note that GC03GC03 parasites showed a higher, statistically significant level of free heme at baseline (i.e. in the absence of CQ) as compared to GC03Dd2 and GC03Cam734 parasites (P = 0.004 and P = 0.0012, respectively, by unpaired t tests with Welch’s correction), although the lowest level of Hz achieved in all three CQ-treated strains was comparable (Fig 4B). Maximal free heme amounts in the presence of CQ even at very high concentrations were also markedly lower in GC03Dd2 and GC03Cam734 parasites compared to the CQ-sensitive GC03GC03 parasites (S5 Fig).
Several notable differences were also observed among the Hb profiles of pfcrt-modified lines (Fig 4C and S5C Fig). First, GC03 parasites encoding wild-type (GC03) pfcrt exhibited lower concentrations of Hb at baseline as compared to isogenic parasites expressing the mutant pfcrt alleles Dd2 or Cam734 (mean amounts of 2.3, 3.7 and 6.3 Hb fg per cell respectively in untreated samples; S5C Fig), with the difference between the GC03GC03 and GC03Cam734 lines achieving statistical significance (P = 0.006 by unpaired t test with Welch’s correction). Consistent with previous findings [32], CQ-sensitive GC03GC03 parasites (Fig 4C and S5C Fig) showed a significant elevation in Hb species that did not occur until 2.5× CQ IC50. A comparable accumulation in Hb starting at 2.5× CQ IC50 was observed for GC03Cam734 parasites, contrasting with the profile of GC03Dd2 parasites, which showed elevations in Hb amounts at a lower (1×) CQ IC50 fold (Fig 4C and S5C Fig). The increase in Hb observed for GC03Dd2 parasites coincided with a statistically significant increase in free heme at 1× CQ IC50 (see Fig 4A). This is consistent with previous studies of CQ-treated parasites, in which increases in undigested Hb were found to follow significant increases in free heme [32].
For each pfcrt-modified line, we also compared the CQ dose dependence of free heme fractions versus parasite growth (Fig 5). Interestingly, for each line, the free heme concentration curve crossed the parasite growth curve at approximately the same mid-point (IC50 value), indicating that the inverse relationship between CQ action on free heme levels and growth inhibition that was previously observed for parasite lines encoding wild-type pfcrt [73] is preserved among lines encoding CQ-resistant pfcrt alleles. These data provide compelling evidence that for both resistant and sensitive parasites, CQ-mediated growth inhibition results primarily from this drug’s inhibition of Hz formation, which the parasite uses to detoxify reactive free heme.
A key feature of CQR in P. falciparum parasites is the ability of mutant PfCRT to efflux CQ from the parasite DV, in turn reducing CQ access to heme. Measurement of drug transport is experimentally challenging in Plasmodium parasites due to the presence of multiple membrane-bound intracellular compartments. Accordingly, we evaluated CQ transport mediated by Cam734 PfCRT using a recently optimized Saccharomyces cerevisiae galactose-inducible PfCRT expression system [22]. Yeast-expressed PfCRT isoforms localize largely to the cell membrane and mediate transport of CQ from the external medium (low pH, positive membrane potential [Ψ]) to the yeast cytosol (high pH, negative Ψ), recapitulating the electrochemical gradient-driven transport of CQ from the parasite DV (low pH, positive Ψ) to the parasite cytosol (high pH, negative Ψ) [74]. Of note, at baseline, the yeast cell membrane possesses a high ΔpH and a low Δψ. By increasing the pH of the external medium (pHexternal), the cell membrane ΔpH can be experimentally lowered, yielding a compensatory potassium channel-dependent increase in Δψ [74]. In this system, at higher Δψ, CQ transport by CQ-resistant PfCRT isoforms is more pronounced as compared to the basal level of transport mediated by the CQ-sensitive wild-type GC03 PfCRT isoform [22,75]. Earlier studies have validated that growth rates of PfCRT-expressing yeast serve as a useful proxy for CQ transport [75].
Using quantitative growth rate analyses, we examined the effect of varying external pH (and hence the Δψ) in yeast strains expressing PfCRT isoforms GC03, Cam734, the back-mutant Cam734 F144A (chosen because this back mutation was the most effective at ablating CQ and md-CQ resistance; see Fig 1), or Dd2 (see Table 1 for haplotypes). Growth was assessed in the presence of 5 mM CQ, a concentration required for this drug to exert differential growth inhibitory activity on yeast strains expressing various PfCRT isoforms [22]. As a negative control, we also included yeast harboring no PfCRT (vector control). To examine the effect of Δψ on transport, we assessed growth over a range of pHexternal values (range of 7.20 [low Δψ] to 7.45 [high Δψ]; S6A Fig). Intriguingly, in low Δψ conditions, growth of yeast expressing Cam734 PfCRT was comparable to that of yeast expressing the CQ-sensitive GC03 PfCRT isoform (Fig 6). However, when the Δψ was clamped to higher values, Cam734 PfCRT conferred a CQR-associated delayed growth phenotype that was intermediate to that of GC03 (wild-type) and Dd2 PfCRT (Fig 6). Of note, the growth phenotype associated with the Cam734 F144A isoform was intermediate to that of empty vector and wild-type PfCRT. This provides evidence that the A144F mutation is critical for drug transport mediated by the Cam734 isoform and is consistent with our drug assay data showing a CQR phenotype for parasites expressing Cam734 pfcrt but not the F144A back-mutant (see Fig 1A). These PfCRT-specific phenotypes were not attributable to differences in protein expression, as comparable protein expression of PfCRT variants was observed upon galactose induction of yeast (S6B Fig).
Investigations utilizing the entrapment of a dextran-conjugated NERF to probe the pH and volume of the parasite DV have previously documented the ability of PfCRT mutations to alter DV physiology [40,76,77]. Using similar methods (see Supplementary Materials and Methods), we determined the DV pH and volume in isogenic GC03 parasites encoding GC03, Dd2, Cam734 or Cam734 F144A PfCRT, in the presence or absence of CQ concentrations corresponding to twice the CQ 50% lethal dose (LD50; see Supplementary Materials and Methods). From smallest to largest DV volume, as well as from most alkaline to most acidic DV pH, we observed the order of parasite lines to be: GC03GC03, GC03Cam734 F144A, GC03Cam734, and GC03Dd2 (S7 Table). This order was preserved upon a brief (30 min) addition of CQ, which consistently increased DV volume (by 13% to 33%).
To ensure successful progression through their life cycle, drug-resistant P. falciparum parasites must balance the acquisition of resistance properties with the maintenance of required and often interrelated physiological processes. Focusing on the pathogenic intraerythrocytic stages of parasite growth, we explored herein how novel mutations comprising the unusually polymorphic Cam734 PfCRT variant contribute to this complex relationship. Our analysis of isogenic, pfcrt-modified lines reveals that multiple PfCRT mutations possess dual roles, contributing to both quinoline resistance and parasite proliferation. This was most notable for the A144F mutation that is unique to Cam734 PfCRT, which in addition to affecting growth rates proved to be indispensable for parasite resistance to multiple quinoline-type compounds, including CQ, QN, and the first-line ACT partner drug AQ. While these drug IC50 shifts are often relatively small, studies have shown that these translate into clear patterns of selection in field parasite populations [7,12]. The pleotropic requirement for the A144F mutation in Cam734 PfCRT-mediated drug resistance is reminiscent of earlier work, in which back-mutation of K76T ablated CQR and nearly halved the degree of parasite resistance to QN [52]. GC03Cam734 F144A parasites appeared CQ-sensitive, but nonetheless showed a 3–fold higher IC90 value for md-CQ, believed to be the major driver of selection for mutant pfcrt [16], relative to the fully sensitive GC03GC03 line (S2 Table). By comparison, GC03Cam734 parasites showed a 27–fold md-CQ IC90 increase. Similar findings were earlier observed with a PfCRT variant of 7G8 that carries the C350R mutation (the H209 isolate found in French Guiana) [50,60]. This variant was shown to mediate a phenotype of CQ tolerance, which manifested as low CQ IC50 values but elevated md-CQ IC90 values as well as parasite recrudescence after exposure to CQ concentrations lethal to CQ-sensitive parasites expressing wild-type pfcrt [50,60]. We note that parasites encoding Cam734 F144A PfCRT retained K76T as well as 7 other mutations [7]. The clear importance of mutations other than K76T in contributing to CQR can help explain, in areas where novel PfCRT variants have arisen, why the K76T mutation predicts clinical CQR with good sensitivity but only moderate specificity [17,62]. Another important factor driving the reduced specificity of the K76T marker is patient immunity, which in higher-transmission settings of Africa is known to help resolve CQ-resistant infections in CQ-treated patients [78].
Prior studies with asexual blood stage parasites have shown that CQ affects heme disposition (increasing free heme and reducing Hz) and that CQ access to its heme target in the DV is significantly reduced by CQ-resistant mutant forms of PfCRT, which are thought to efflux CQ away from the DV [32,38,52]. We built on these observations by comparing wild-type and variant PfCRT isoforms expressed on the same genetic background. Our results, shown in Figs 4 and 5, provide compelling evidence that the degree of CQ-mediated inhibition of parasite growth is closely correlated with the level of inhibition of Hz formation. Reduced Hz formation was accompanied by the accumulation of reactive free heme, which at high concentrations is presumably the major trigger of parasite death, either alone or conjugated to CQ [7]. Recombinant GC03 parasites expressing the mutant PfCRT Dd2 and Cam734 isoforms (i.e. GC03Cam734 and GC03Dd2) differed notably from the isogenic clone expressing the wild-type GC03 isoform (GC03GC03) in that accumulation of free heme occurred at higher CQ concentrations in the former. Increased free heme was accompanied by lower levels of Hz, consistent with mutant PfCRT being able to efflux drug away from its heme target. Both mutant pfcrt-expressing lines also showed reduced levels of heme at low concentrations of CQ or at baseline (no CQ), suggesting a more efficient process of Hz formation under those conditions. The reason for this difference in baseline free heme is not yet known. Evidence suggests that free heme in untreated parasites is sequestered, possibly through association with neutral lipids in the DV [79]. The baseline difference between wild-type and mutant pfcrt-expressing parasites may thus be attributed to the larger DV of GC03Dd2 and GC03Cam734 lines, as compared to GC03GC03 parasites (see below and S7 Table), resulting in a lower lipid to aqueous volume ratio. Indeed, a fixed lipid-aqueous portioning coefficient and fixed ratio of lipid to aqueous heme concentration would yield an increased quantity of aqueous free heme (volume × concentration), which in turn would be mostly incorporated into Hz.
With all three lines, levels of undegraded Hb also rose at relatively high CQ fold IC50 concentrations, with the highest levels recorded in the GC03Cam734 and GC03Dd2 lines (see Fig 4C), potentially reflecting increased CQ amounts in the cytosol of these parasites because of higher rates of CQ efflux from the DV. These results expand on the previous observation, in CQ-treated drug-sensitive parasites, that increases in undigested Hb follow significant increases in free heme [32]. This suggests a secondary mode of CQ action, whereby Hb proteolysis is inhibited at higher CQ concentrations, possibly through a physiologic effect of elevated CQ concentrations on Hb endocytosis or the activity of DV-resident hemoglobinases [80,81]. Alternatively, elevated concentrations of free heme in aqueous environments have been shown to form heme aggregates capable of disrupting lipid bilayers and triggering membrane disorder, which in turn could disrupt Hb import and catabolism [82].
Our metabolomics analysis (see Fig 3) and total Hb quantification (see Fig 4C and S5C Fig) reveal major changes in the Hb digestion pathway in the GC03Cam734 line when compared with the isogenic lines GC03Dd2 and GC03Dd2 lines. Given that these changes have been associated with impaired fitness in mutant PfCRT parasites expressing the CQ resistance-conferring Dd2 or 7G8 haplotypes [41], our results raise an obvious question: does Cam734 PfCRT impart a metabolic compensatory mechanism that allows these parasites to circumvent the normally deleterious effects of altered Hb digestion? Previous heterologous expression studies using Xenopus laevis oocytes have suggested that mutant PfCRT isoforms might selectively confer transport of the tripeptide glutathione [57], which was earlier proposed to facilitate the degradation of reactive heme and reduce heme-mediated toxicity [36,83]. However, we saw no significant differences in glutathione or any other redox-associated metabolites (see S4 Fig and S6 Table) between isogenic lines encoding Cam734 or Dd2 PfCRT, suggesting that major redox-related metabolic changes are unlikely to account for the improved fitness associated with the Cam734 pfcrt allele. In contrast, we observed significant differences in ATP/AMP ratios and central carbon metabolism between lines encoding Cam734 and Dd2 PfCRT (see Fig 3 and S6 Table), implicating changes in energy metabolism as a potential physiologically compensatory mechanism.
Our heterologous yeast expression studies also found that Cam734 is significantly affected by Δψ, resembling a CQ-sensitive PfCRT isoform at low Δψ and a mutant, CQ-resistant PfCRT isoform at high Δψ. This unique plasticity in mediating drug transport may underlie the improved asexual blood-stage fitness associated with the Cam734 PfCRT isoform, as compared with Dd2 (see Fig 2), whereby Cam734-defining mutations confer drug transport only in certain Δψ DV conditions. We also note that CQ transport, as assessed in heterologous expression systems, may only partially account for in vitro parasite CQR. This is highlighted by the earlier observation that the CQ-resistant Ecu1110 PfCRT variant (K76T/A220S/N326D/I356L) confers lower parasite CQR, but higher CQ transport, than the related 7G8 PfCRT variant (that in addition carries the C72S mutation) [16,23,77]. The Cam734 isoform might therefore facilitate CQR in part by alleviating CQ-mediated inhibition of an endogenous PfCRT function. Continued elucidation of the elusive function of PfCRT will assist in clarifying these distinctions.
Our physiological studies of isogenic pfcrt-modified lines revealed that the DV pH and volume parameters in GC03Cam734 parasites were intermediate to GC03GC03 and GC03Dd2 parasites, consistent with the Cam734 allele producing CQ IC50 values between the sensitive wild-type and highly-resistant Dd2 isoforms. In all three isogenic lines, brief exposure to CQ caused DV swelling, with the more CQ-resistant parasites showing the greatest increase in DV volume (S7 Table). As the composition of the DV environment governs the degree of heme-to-Hz conversion [38], we propose that, compared to Dd2 PfCRT, the reduced DV size and more wild-type (GC03) PfCRT-like DV pH associated with the Cam734 isoform might play a role in neutralizing the fitness costs typically associated with mutant PfCRT variants.
Recent evolutionary genetic studies of the adaptive landscapes (i.e. mutational paths and their accompanying fitness costs) associated with drug resistance-conferring mutations in pfcrt and P. falciparum dihydrofolate reductase (dhfr) highlight key considerations when hypothesizing how Cam734 pfcrt might have evolved: (1) forward pfcrt evolution is a physiologically constrained process that is consistent with the rarity of pfcrt alleles bearing three or fewer polymorphisms; (2) forward and reverse processes of gene evolution are associated with distinct adaptive landscapes; and (3) adaptive landscapes can be substantially modified by their drug environment [16,84–86]. The spread of CQR in Asia and Africa has long been attributed to a single (Dd2 or Dd2-like) pfcrt allele [4]. Of note, Cam734 shares four of the eight mutations comprising both the eight-amino acid Dd2 variant (see Table 1) and the related 6-amino acid variant GB4 (equivalent to Dd2 S326N T356I) found in Africa and also seen in Southeast Asia. Our recent analysis of close to 900 Asian P. falciparum genomes recently sequenced by the Pf3K consortium [87] estimates the prevalence of Cam734, GB4 and Dd2 pfcrt alleles at 15%, 13% and 58% respectively, with the remainder comprising the wild-type allele (3%) and several minor variants. For Cam734, the highest abundance was observed in Cambodia (105 of 570 genomes), Laos (29 of 85) and Vietnam (32 of 97), with a far lower prevalence in Thailand (1 of 148). We posit that, faced with high CQ pressure, parasites underwent mutational bursts (as previously suggested [16]) that led to the evolution of Dd2 pfcrt. With reduced CQ pressure, a “reverse” evolutionary process might have led to the loss of some mutations and eventual acquisition of novel ones, as in the case of Cam734 pfcrt.
In their report documenting Cam734 pfcrt in Cambodia, Durrand et al. also reported the related allele Cam738 (akin to Cam734 without the A144F mutation) [47]. We posit that Cam738 served as a mutational precursor of the more evolutionarily successful Cam734 allele. This is supported by the inferior growth of isogenic parasites expressing Cam738 pfcrt as compared with Cam734 pfcrt, in the absence or presence of CQ or other quinoline drugs (see Figs 1B and 2). Our evidence of reduced growth rates of parasites harboring the Cam738 allele compared with Cam734 is consistent with the absence of Cam738 haplotypes in the recent Pf3K genome data set (https://www.malariagen.net/projects/pf3k). We also note that selective forces favoring mutation of PfCRT residue 144, found herein to be a key mediator of CQR, are apparent in Asia, in some cases requiring two nucleotide substitutions. For example, in the Philippines or in China, PfCRT haplotypes have been detected that, respectively, harbor the mutations A144T or A144Y [88,89]. Interestingly, addition of A144Y to the CQ-resistant Dd2 PfCRT isoform was previously found to abrogate CQ transport in S. cerevisiae [22].
With sustained exposure to drug selective forces, parasites may evolve intragenic and/or intergenic compensatory changes that allow them to persist even in the absence of drug pressure [90]. The Cam734-defining compensatory mutations identified in our analysis reveal an intragenic basis for the enhanced fitness of this CQ-resistant allele, which could explain its continued presence in Southeast Asia as a minor allele despite the lack of CQ use for several decades to treat P. falciparum malaria. We note that CQ resistance-conferring mutant pfcrt alleles (including Cam734 and Dd2) might also persist in Southeast Asia because of local conditions of decreased genetic diversity and complexity of infections, resulting in less competition among parasite lines, as compared with high-endemicity settings in sub-Saharan Africa, where mutant pfcrt alleles are known to rapidly decrease in prevalence in areas without CQ pressure [7].
The degree to which secondary genetic factors also play a role in maintaining mutant pfcrt in Southeast Asia is presently unclear. We note that our pfcrt-modified lines were generated in GC03 parasites, a clone of the HB3 (Central America) × Dd2 (Asia) genetic cross [91]. These parasites encode the HB3 PfMDR1 haplotype, which differs from Dd2 PfMDR1 at three distinct residues (86, 184, 1042) [92]. A recent study has shown that the PfMDR1 N86Y mutation present in Dd2 augments the degree of CQR imparted by the mutant Dd2 PfCRT isoform [15]. We have observed in prior transfection-based studies that the parasite genetic background dictates the level to which mutant pfcrt alleles can mediate CQR [50]. Of note, results obtained herein in GC03 parasites might potentially differ from ones that would be produced with culture-adapted field isolates that naturally harbor the pfcrt Cam734 allele. However, to the best of our knowledge, no such isolate has been culture-adapted and reported in the literature. Furthermore, GC03 has been the primary strain used in multiple prior pfcrt transfection studies, using either the ZFN method or the earlier approach that used single-site crossovers, thus providing a benchmark against which to assess the current data set [11,16,42,48,50].
The notion that additional genetic changes are required to produce high-level CQR in parasites encoding Cam734 PfCRT evokes a previous finding that mutant PfCRT-encoding parasites can exhibit increased expression of proteins involved in pH regulation, including a V-type H+ pyrophosphatase [93]. Our observation that, compared with Dd2, Cam734 PfCRT required a higher Δψ to manifest increased growth in the presence of CQ (consistent with elevated drug transport; see Fig 6) suggests that high-level Cam734 PfCRT-mediated drug resistance may be potentiated by parasite proteins that govern the Δψ across the DV membrane. We speculate that this plasticity in drug transport might be a reflection of the balance that Cam734 PfCRT has achieved in mediating resistance while also avoiding fitness costs. Future genetic dissections of pfcrt alleles, as well as candidate secondary genetic modulators (e.g. pfmdr1), are possible with the recent advent of efficient parasite ZFN or CRISPR/Cas9–based genome-editing tools [15,94]. Leveraging these approaches with analysis of parasite whole-genome sequences will aid in deciphering the genetic complexities that underlie new and emerging multidrug resistance phenotypes.
P. falciparum asexual blood-stage parasites were cultured in human RBCs (Interstate Blood Bank) at 2–4% hematocrit in RPMI-1640-based malaria cell culture medium supplemented with 0.5% Albumax II (Invitrogen) [95]. Cultures were incubated at 37°C in 5% O2 / 5% CO2 / 90% N2. Genetic modification of the parasite pfcrt locus is detailed in Supplementary Materials and Methods and S1 Fig.
Drug inhibitory concentrations that result in 50% (IC50) or 90% (IC90) growth inhibition were determined for a panel of drugs (CQ ± 0.8 μM VP, md-CQ, md-AQ, QN, PPQ, LUM, AS, and PND), as described [50]. After 72 h exposure to drug, parasite growth was quantified by staining with SYBR Green I and MitoTracker Deep Red and measuring parasitemia on an Accuri C6 flow cytometer. Reversibility of CQR by 0.8 μM VP was expressed as the CQ RMI, equivalent to the quotient of the CQ+VP IC50 divided by the CQ IC50 [63]. Statistical significance was determined via non-parametric Mann-Whitney U tests using GraphPad Prism 6 software.
As a proxy for in vitro fitness, growth of parasite lines was assessed in 1:1 co-culture assays with the fluorescent reporter line NF54eGFP, using previously described methods [16]. Briefly, 1:1 co-cultures consisting of the reporter line (GFP+) and individual pfcrt-modified test lines (GFP−) were propagated for 10 generations, and parasitemias were maintained between 0.3% and 8%. The proportion of GFP− parasites was regularly determined by flow cytometric detection of the far-red fluorescent dye SYTO61, which labels the nuclei of infected RBCs (iRBCs), and GFP. Derivation of per-generation selection coefficients (s) of test strains is detailed in Supplementary Materials and Methods. Statistical significance was assessed via two-way analysis of variance (ANOVA) with Sidak’s post-hoc test using GraphPad Prism 6 software.
All parasite culturing, metabolite extraction, mass spectrometry data acquisition, and data analyses were conducted using previously established methods [41]. Briefly, after double synchronization with 5% sorbitol, late-stage (~36–42 h) P. falciparum trophozoites were magnetically purified using a SuperMACS magnetic separator (Miltneyi Biotec) and CS columns. Eluted iRBCs were resuspended at 0.4% hematocrit and allowed to recover for 2 h at 37°C in a tissue culture incubator. Cells were then rapidly cooled to 4°C and pelleted by centrifugation at 2,000×g for 5 min. Media was then aspirated away from the iRBC pellets and metabolites were extracted by resuspending cells in cold (4°C) 90% methanol. Samples were homogenized by vortexing and centrifuged at 10,000×g for 5 min at 4°C. The supernatant metabolite extracts were harvested and stored at -80°C until mass spectrometry analysis. Just prior to mass spectrometry, samples were dried under a stream of N2 and were resuspended in HPLC-grade water at a 4:1 dilution (relative to the original iRBC pellet volume). High-resolution mass spectrometry data were acquired on a Thermo Fisher Exactive Mass spectrometer in negative mode using 25 min reverse phase gradients and ion-pairing chromatography [41]. Metabolites were identified using the known chromatographic retention times of standards, and metabolite signals were quantified using MAVEN [41]. To allow for more direct metabolite-to-metabolite comparison of phenotypes, raw mass spectrometry signals were expressed as z-scores. Briefly, for each metabolite, the mean expected signal (x¯) was defined as the mean intensity observed in the control line (GC03Dd2). Likewise, the standard deviation (s) for each metabolite signal was calculated from signals observed in the GC03Dd2 line (deduced from 3 independent harvests with 4 replicates). The z-score (zi) for each observed signal (xi) in test lines was then computed as per the relationship zi = [xi−x¯]/s and plotted according to metabolite class. For summary statistics (Fig 3), z-scores were calculated from the signals observed for each class. These classes were comprised of metabolites that are directly associated with a metabolic pathway (e.g. TCA metabolism included TCA intermediates as well as the TCA-associated amino acid glutamate). P values were computed by one-way ANOVA. All data analyses and statistical tests were conducted using custom in-house software written in R. Metabolite signal intensities are summarized in S5 Table. Metabolite z-scores and associated P values are found in S6 Table.
The heme fraction profiles of pfcrt-modified GC03GC03, GC03Dd2, and GC03Cam734 parasites were determined following recently published and validated protocols [73]. First, parasite growth in response to CQ was determined using the lactate dehydrogenase assay [96]. Heme fractionation assays were then initiated by incubating sorbitol-synchronized, early ring-stage parasites in the absence or presence of CQ in multiples (0.5×, 1×, 2×, 2.5×, and 3×) of the biological CQ IC50. After 32 h, iRBCs were treated with 1% saponin to release mature trophozoites, followed by hypotonic lysis and centrifugation. Supernatants were treated with 2% SDS and 2.5% pyridine, yielding the Hb fraction. Pyridine was used as it coordinates to heme forming a monomeric low-spin complex with a distinctive spectrum and is easily detectable by UV-visible spectroscopy, thereby allowing heme species to be quantified. Pellets were treated with 2% SDS and 2.5% pyridine, sonicated, and centrifuged, and supernatants were removed to isolate the free heme fraction. The remaining pellets were solubilized in 2% SDS and 0.1 M NaOH, sonicated, neutralized with HCl, and treated with 2% SDS and 2.5% pyridine to generate the Hz fraction. For each fraction, the UV-visible spectrum of heme present as a heme-pyridine complex was measured with a multi-well plate reader (Spectramax 340 PC, Molecular Devices). The abundance of Hb, free heme, and Hz species was reported as a percent and as an absolute amount of heme Fe per cell. Parasites were quantified using flow cytometry, as previously described [73]. Statistical significance was assessed via unpaired t tests with Welch’s correction using GraphPad Prism 6 software.
Cultivation, transfection, and quantitative growth rate analysis of S. cerevisiae yeast strains employed previously detailed protocols [22,75]. Briefly, CH1305 yeast strains were transfected with either pYES2 (blank vector) or pYES2-derived plasmids encoding the galactose/raffinose-inducible, codon-optimized PfCRT isoforms GC03 (HB3; wild-type), Cam734, Cam734 F144A (also known as Cam738), or Dd2. Quantitative assessment of yeast growth, a validated proxy for CQ transport [74], was performed in PfCRT-inducing (galactose/raffinose) or PfCRT-noninducing (glucose) conditions, with a starting cell density (OD600) of 0.1. Yeast growth ± 5 mM CQ was measured in triplicate with a Tecan GENios microplate reader following established parameters [22]. PfCRT protein expression of yeast lines was evaluated using Western blot analysis, demonstrating comparable protein levels across all lines (see Supplementary Materials and Methods and S6B Fig).
|
10.1371/journal.pntd.0004590 | Assessing Lymphatic Filariasis Data Quality in Endemic Communities in Ghana, Using the Neglected Tropical Diseases Data Quality Assessment Tool for Preventive Chemotherapy | The activities of the Global Programme for the Elimination of Lymphatic Filariasis have been in operation since the year 2000, with Mass Drug Administration (MDA) undertaken yearly in disease endemic communities. Information collected during MDA–such as population demographics, age, sex, drugs used and remaining, and therapeutic and geographic coverage–can be used to assess the quality of the data reported. To assist country programmes in evaluating the information reported, the WHO, in collaboration with NTD partners, including ENVISION/RTI, developed an NTD Data Quality Assessment (DQA) tool, for use by programmes. This study was undertaken to evaluate the tool and assess the quality of data reported in some endemic communities in Ghana.
A cross sectional study, involving review of data registers and interview of drug distributors, disease control officers, and health information officers using the NTD DQA tool, was carried out in selected communities in three LF endemic Districts in Ghana. Data registers for service delivery points were obtained from District health office for assessment. The assessment verified reported results in comparison with recounted values for five indicators: number of tablets received, number of tablets used, number of tablets remaining, MDA coverage, and population treated. Furthermore, drug distributors, disease control officers, and health information officers (at the first data aggregation level), were interviewed, using the DQA tool, to determine the performance of the functional areas of the data management system.
The results showed that over 60% of the data reported were inaccurate, and exposed the challenges and limitations of the data management system. The DQA tool is a very useful monitoring and evaluation (M&E) tool that can be used to elucidate and address data quality issues in various NTD control programmes.
| The Global Programme for the Elimination of Lymphatic Filariasis has been conducting yearly treatment of entire communities in endemic countries since the year 2000. During the treatments various information is collected on the populations, number of medicine tablets distributed and remaining, the number of people treated, etc. that can be used to evaluate the performance of the lymphatic filariasis control programme. For example, information on the number of people treated in a District gives an indication of the success of the programme. In line with this, the World Health Organization in collaboration with other agencies developed a tool for Neglected Tropical Diseases (NTD) to help national control programmes assemble and analyse their data. This study was undertaken to evaluate this tool and the information collected from some endemic communities in Ghana. Community registers were reviewed and personnel involved in drug distribution in the communities were interviewed to collect the necessary information. The results showed that more than half of the data reported in the endemic communities surveyed were inaccurate. It also revealed some weaknesses in the data management and reporting system. The tool, however, is good for identifying and quantifying the magnitude of the challenges encountered in the information management for NTD programmes, especially at peripheral levels.
| The Global Programme to Eliminate Lymphatic Filariasis (GPELF) started its activities in the year 2000, with the aims of eliminating lymphatic filariasis (LF) as a public health problem by the year 2020, through mass drug administration (MDA) in endemic implementation units (IU) [1]. In many countries significant progress has been made in controlling the disease; however, many programmatic challenges continue to affect the performance of National LF Control Programmes. Notable among these is the effective implementation of the preventive chemotherapy strategy in endemic communities [2, 3].
The quality of data reported in healthcare systems is important for evaluating programmes, as such high quality health information is crucial in addressing global health challenges and building strong public health systems [4]. Data Quality Assessment (DQA) is a scientific and statistical evaluation of data to determine if they meet the objectives, and are of the right type, quality, and quantity to support their intended use [5]. At present yearly MDA has been undertaken in 53 of the 73 LF endemic countries [6]. During the MDA various data are collected at various levels to help in the planning and improvement of activities. As such high quality becomes the prerequisite for better information, better decision-making and better population health [7].
In all LF endemic Districts in Ghana, various information are collected during MDAs, including number of treatments given, number of people treated, number of tablets used, reasons for non-treatment, place of treatment, individual identification (name and address), name of drug used, age and sex of drug recipients, etc. Public health data can be useful for decision-making, effective service delivery, and evaluating prevailing programmes in order to maintain high quality of healthcare. Poor data quality not only contributes to poor decisions and loss of confidence in the systems, but also affects the validity of impact evaluation studies [8]. Furthermore, variability in data quality from health management information systems in sub-Saharan Africa threatens utility of these data as a tool to improve health systems [9]. Thus, collecting accurate data will aid appropriate intervention for elimination.
The WHO, in collaboration with NTD partners, including ENVISION/RTI, developed a DQA tool to identify and characterize challenges with NTD data quality–including incomplete and inaccurate data or data not timely reported–following a recommendation from the WHO Working Group on Monitoring and Evaluation of Preventive Chemotherapy. The DQA tool focuses exclusively on verifying the quality of reported data quantitatively and assessing the underlying data management and reporting systems qualitatively for standard programme-level output indicators. Data quality dimensions include accuracy, reliability, completeness, timeliness, precision, integrity and confidentiality [10]. In 2014, training was undertaken to introduce the tool to various stakeholders, with field testing to follow [10]. This study was undertaken to evaluate the quality of data reported in selected communities or service delivery points (SDP), and the data management functions and capabilities in three LF endemic Districts in Ghana.
Approval for this study was obtained from the Ethical Review Board of the Noguchi Memorial Institute for Medical Research (IRB 077/13-14). The District Health Office was informed of the study and permission sought to assess data from the registers. Written informed consent was obtained from all individuals interviewed during the study.
This study was undertaken in the Ahanta West, Nzema East and Agona East Districts of Ghana (Fig 1). The Ahanta West and Nzema East Districts, both located in the Western Region of Ghana, started MDA in the years 2000 and 2002, respectively. Both districts represent areas with persistent transmission, with MDA ongoing at the time of this study in 2014. The Agona East District is located in the Central Region of Ghana and started MDA activities in 2002. By 2010, LF infection rate in Agona East District was considered to be below the 2% antigenemia and 1% microfilaremia thresholds required to stop MDA [11]. As such, treatment in Agona East ceased in 2010. Thus, for the purpose of comparing data between sites, the 2010 data registers were analysed for all the sites surveyed. While the DQA tool advocates the selection of sites based on probability proportionate to size (PPS), the survey communities in this study were randomly selected because the population estimates of the communities could not be obtained beforehand. Eight sites were surveyed in the Ahanta West District and 6 sites from the Nzema East and Agona East Districts respectively.
A cross sectional study involving the review of data registers and interview of drug distributors, disease control officers and health information officers was done. Data registers at the SDPs capture data during MDA for compilation by health workers. Information contained in these registers includes age, sex, height, number of households, population, drug used, number of tablets used, number of tablets received, etc. While the tool is capable of being used at the SDPs and all intermediate data aggregation levels (IALs), in this study only the SDPs were evaluated for data quality since they represent the first data collection and handling locations. In assessing the data management systems and functions, the intermediate data aggregation level 1 (IAL-1) represents the last point for information dissemination into the SDPs, and the first point of data collection from the SDPs. The interview tool used is a standard questionnaire, developed as part of the DQA tool, with scoring guidelines coded 3 for “Yes, completely”, 2 for “Partly” and 1 for “No, not at all”. These scores take into consideration the response from all the interviewees. The DQA tool and further information can be obtained from the WHO Department of Control of Neglected Tropical Diseases.
To evaluate the quality of data reported in the study areas, data registers for SDPs for 2010 were obtained from IAL-1, for assessment. The assessment verified reported results (from IAL-1) in comparison with recounted values (from SDPs) for five indicators, i.e. number of tablets received, number of tablets used, number of tablets remaining, MDA coverage, and population treated. Sources of data for the five indicators were examined to determine the percentage of reports that were available, on-time, completed, collected and measured consistently, protected from deliberate bias, and maintained according to national or international standard. Further, drug distributors, disease control officers and health information officers available at the time of the study (at the first data aggregation level) and who were willing, were interviewed using the DQA tool, to determine the M&E structure, functions and capabilities, indicator definitions, links with national reporting system, data management processes and data collection and reporting forms and tools.
For each of the five indicators assessed a verification factor (VF) was calculated as the ratio of recounted value (from the data register) of the indicator to the reported value, expressed as a percentage. A value of 100% indicates a high level of accuracy. Values above 100% indicate under reporting, whiles values below 100 suggest over reporting. In interpreting the results, indicator values between 95–105% across multiple sites were considered as high quality reporting. Indicator values less than 90% and greater than 110% were considered poor quality reporting. Verification factors above 300 were excluded from the analysis. The values were compiled and graphs generated in GraphPad Prism version 6.05. Statistical significance was set at p-values less than 0.05.
Scores for the functional areas in the Data Management Assessment were automatically computed by the DQA tool, which also generates a spider chart. At individual sites, functional areas with scores >2.5 indicate high quality, whiles scores<2.0 reflect low quality. However, when comparing scores across sites, scores> = 2.8 indicate good performance and scores < = 1.5 indicate that a functional area needs to be improved.
In each District, 3–4 days were spent in reviewing and evaluating the data, and conducting interviews. The SDPs were visited by the study team to review the selected indicators from the community registers. It is worth mentioning that except for being interviewed by the study team, individuals working at any level with the LF control programme were not directly involved in the use of the tool, and the study was undertaken independently by personnel from the Noguchi Memorial Institute for Medical Research (NMIMR)–University of Ghana. Further, the evaluation of the tool took advantage of ongoing parasitological and entomological surveys in the Districts.
The results showed that 40% (40/100) of all data examined were over reported while 22% (22/100) were under reported. The only consistent indicator that was accurately reported across sites was the number of tablets received. For the five indicators assessed, the VF were plotted for comparison between Districts (Fig 2). Between Districts, there was no significant difference in the indicators assessed, except for the population treated in Agona East.
Results of the data management assessment in the Districts are shown in Fig 3. In Agona Nkwanta, indicator definitions and reporting guidelines were the strongest functional areas followed by data collection and reporting forms and tools. Data management processes was the weakest functional area, followed by M&E structure, function and capabilities. In Axim District Health Directorate, the strongest functional area was indicator definitions and reporting guidelines, followed by M&E structure, functions and capabilities then data-collection and reporting forms and tools. In Konyarko health post, the strongest functional area was the data collection and reporting forms and tools.
In terms of the data quality dimensions, the observed values were more or less consistent between sites (Fig 4). In Ahanta West, the lowest reporting performance was confidentiality (75%), followed by timeliness (79%). However, the best reporting performance was reliability (88%), followed by availability (86%) and integrity (86%). In Nzema East District, the lowest reporting performance was confidentiality (77%), followed by completeness (79%). On the other hand, the best reporting performance was reliability (89%), followed by integrity (86%). In Agona East, the lowest reporting performance was availability (68%), followed by timeliness (70%), while the best reporting performance was reliability (90%), and followed by integrity (82%).
Data are vital to public health, since they signify and provide a documented account of public health practice. The extensive application of data, for the evaluation of public health responsibility and performance, highlights the importance of data quality and how to evaluate it. This study reflected the poor quality of data reported following MDA for LF, indicated as the over reporting or under reporting of the indicators assessed. In particular, is the overestimation of MDA coverage in many of the sites surveyed. Overestimation of MDA coverage in NTD programmes has been reported in many studies [12–14]. MDA coverage is the core indicator required for global reporting on preventive chemotherapy, thereby reflecting the performance of control programmes [14] and must therefore be strictly monitored. It is important to note that before MDA, drug distributors are gathered at the District level for training and orientation. In some of the communities, the drug distributors recalled having supervision support during the MDA, while this was not the case in other communities.
In terms of the functional areas of the LF control programme, even though study sites registered a score >2.0, which is considered passable, no site demonstrated a high quality data reporting system, given that no site scored >2.5 for all functional areas. However, in some of the surveyed areas, health workers reported that strict guidelines were received, defining the indicator to report on, where, when and to whom to send reports, and these guidelines have been vividly stated and followed. Additionally, this study found that community drug distributors always used a standard data capture tool, made available at the national level. t. Training plans, and trained data management staff were also available in some of the areas. On the other hand, data quality controls, and back-up procedures, confidentiality of personal data and feedback on data quality were not available. Moreover, trained data management staff and training plans were not sufficient, as well as non-availability of M&E organizational structure. Results from this study suggest that data management processes is the weakest functional area across sites. The reasons for this must further be investigated and addressed accordingly. Similar gaps, such as lack of M&E guidelines, poor feedback given to sub-national levels, poor data use and poor general data management capacity, lack of training programmes to build M&E skills, few standard practices related to confidentiality and document storage, have been reported following DQA in other countries [15, 16].
Overall, data confidentiality, completeness and timeliness require improvement in terms of reporting performance. In the survey sites, data was not managed according to protection and use standard and in most cases data were not reported on time. The completeness of data also needs to be improved as the data examined had missing reports. The data quality dimensions reported in this study are somewhat comparable to those reported elsewhere [17–19] and may indicate the general quality of data in health care systems. The validity of data reported also varied with the various indicators assessed. The most accurately reported indicator was number of tablets received. This is because the supply of drugs to communities was strictly supervised by the Districts.
With the renewed commitment from pharmaceutical companies and other NTD partners at the London Declaration on NTDs in 2012 [20], emphasis must be placed on value for money to ensure that the resources invested are worthwhile. The expanded drug donations and the programme goals presented in the NTD roadmap for implementation point to the importance of having a robust monitoring and reporting system, from the point of treatment by a drug distributor to the national and international levels [10, 21]. While some communities reported good quality data for some of the indicators assessed in this study, the majority of the indicators assessed were of poor quality, necessitating the need to get all communities up to standard. Thus, an important programmatic application of the DQA is to enable objective identification of context-specific issues that compromise data challenges and thereby trigger corrective action before the subsequent MDAs. For example, the use of the tool in LF may help inform how long to treat communities if measures are put in place to address challenges in order to consistently attain the required 65% MDA coverage rates, thus reducing the cost involved in undertaking further yearly treatment beyond the recommended 5–6 years, as per WHO guidelines [1, 22]. While Agona East appears to have the poorest data quality, it is the one District where MDA has been stopped. It is plausible that other factors such as baseline disease prevalence, vector competence and non-compliance to MDA may be at play [23–25], prompting the need for continued treatment in Districts with persistent transmission. As such, the link between data quality and programme success/failure needs to be further evaluated. Nonetheless, this tool may complement the Transmission Assessment Surveys (TAS) undertaken to inform on the need to stop MDA [26], by ensuring that the reported MDA coverage rates have truly been achieved in the evaluation units under assessment. Similarly, donated drugs must be properly accounted for, to ensure that they are put to their intended use, while monitoring the population treated may help in forecasting and budgeting for future MDAs.
While other assessment methods rely on the use of questionnaires and ability of the study respondents to recall past events [13, 14, 27], such as having taken the drug, this tool relies on examining and recounting data recorded in community registers. Though the former method may be subject to proper description of the specific MDA by the questionnaire administrator (considering the various treatment regimen for different NTDs) and the honesty of the study respondents, the DQA tool relies on data recorded for each individual in the register at the SDP. Thus, the tool may be considered as providing a more reliable estimate of indicators for assessing programme performance, with the ability to compare retrospective to current data. However, it is important to note that the tool is also limited to the raw data recorded, such that incorrect entries (especially individual records such as age and the number of tablets given to a particular person) in community registers cannot be detected using the tool.
In this study, PPS sampling wasn't used. Thus, the findings are only applicable to the areas under study, even though it is likely that the same findings occur nationwide. Nevertheless this would need to be confirmed by doing DQA with a representative sampling methodology. Further, in the Districts, data from 2010 was evaluated. While the evaluation of retrospective data can provide valuable information, the use of the tool for programme evaluation should consider the most up-to-date data in order for challenges to be resolved in real-time. In addition to these challenges in the study, the WHO protocol requires co-implementation with Ministry of Health (MoH), NTD programme and other NTD partners operating in the country. This study was undertaken as a research activity, with limited funding, taking advantage of on-going surveys in the study areas. As such, future implementations of the tool should involve the MoH, NTD programme and other partners, in order for the outcomes to be owned by the MoH and therefore more likely to be acted upon to improve programme performance.
In conclusion, this study revealed that the majority of data reported in LF control programme in the study areas was inaccurate, and highlighted some programmatic challenges that must be addressed. At the time of this study, only five indicators could be assessed at a time using the DQA tool and perhaps provision could be made for many more indicators to be evaluated. The DQA tool holds tremendous value in evaluating NTD control programmes, and its use in indicator assessment points to its usefulness in assisting programme managers to address the issues of inaccurate reporting and data quality, following MDA. Using the tool is quite simple and it is recommended that sub-District, District, regional and national management levels use the tool in assessing their NTD programme performance. However, further sensitization and training on this tool for NTD programme personnel or teams at sub-national levels is recommended to ensure its use in the M&E of NTD control programmes. While the results from this study are informative, a more complete assessment of the LF Control Programme (involving the MoH, NTD programme and other NTD partners in the country) must be undertaken at all levels, in order to establish appropriate programmatic responses. All programme activities need to be closely supervised in order to ensure accurate data.
|
10.1371/journal.pbio.2005712 | Within-host competition can delay evolution of drug resistance in malaria | In the malaria parasite P. falciparum, drug resistance generally evolves first in low-transmission settings, such as Southeast Asia and South America. Resistance takes noticeably longer to appear in the high-transmission settings of sub-Saharan Africa, although it may spread rapidly thereafter. Here, we test the hypothesis that competitive suppression of drug-resistant parasites by drug-sensitive parasites may inhibit evolution of resistance in high-transmission settings, where mixed-strain infections are common. We employ a cross-scale model, which simulates within-host (infection) dynamics and between-host (transmission) dynamics of sensitive and resistant parasites for a population of humans and mosquitoes. Using this model, we examine the effects of transmission intensity, selection pressure, fitness costs of resistance, and cross-reactivity between strains on the establishment and spread of resistant parasites. We find that resistant parasites, introduced into the population at a low frequency, are more likely to go extinct in high-transmission settings, where drug-sensitive competitors and high levels of acquired immunity reduce the absolute fitness of the resistant parasites. Under strong selection from antimalarial drug use, however, resistance spreads faster in high-transmission settings than low-transmission ones. These contrasting results highlight the distinction between establishment and spread of resistance and suggest that the former but not the latter may be inhibited in high-transmission settings. Our results suggest that within-host competition is a key factor shaping the evolution of drug resistance in P. falciparum.
| The malaria parasite Plasmodium falciparum has evolved resistance to most antimalarial drugs, greatly complicating treatment and control of the disease. Curiously, although sub-Saharan Africa accounts for the majority of the global burden of malaria, the evolution of drug resistance in Africa has been markedly delayed compared to Asia and the Americas. One reason might be that, in a population in which the prevalence of infection is high, a newly emerged drug-resistant strain faces a high risk of extinction due to competition from drug-sensitive parasites that already “occupy” most of the host population. Using a mathematical model, we confirm that drug-resistant parasites face a much greater risk of extinction in a “high-transmission” setting like sub-Saharan Africa than in a “low-transmission” setting like Southeast Asia. However, we also find that when drug-resistant parasites manage to avoid extinction, their subsequent spread may be more rapid in high-transmission settings than in low-transmission settings, especially when selection is strong. These results offer a novel explanation for global patterns of drug resistance evolution in malaria and suggest a new dimension to consider in resistance prevention and containment efforts: namely, the intrinsic favorability of low- and high-transmission settings for establishment and spread of drug resistance.
| Drug resistance is a recurring threat to effective treatment and control of P. falciparum malaria. Chloroquine—a highly effective synthetic antimalarial—was used to combat malaria on a massive scale in the 1950s; by the 1960s, resistance to chloroquine had developed in Southeast Asia and South America, and over the next few decades, chloroquine resistance spread to virtually every corner of the malaria-endemic world [1]. Sulfadoxine-pyrimethamine, the prevailing substitute for chloroquine, was also rapidly undermined by resistance [1]. Artemisinin-based combination therapies (ACTs), which are now the standard treatment for P. falciparum malaria in most countries, are in danger of widespread failure, with resistance to artemisinin and piperaquine entrenched and spreading in Southeast Asia despite focused containment efforts [2]. At present, ACTs are the last broadly effective antimalarial drugs left; the prospect of widespread resistance therefore represents a looming crisis in global health.
A substantial body of evidence indicates a strong tendency for resistance to evolve in low-transmission settings. Studies that reconstructed the evolutionary histories of resistance to chloroquine, sulfadoxine, and pyrimethamine [3–6] found that drug-resistant parasites from all over the world are frequently derived from just a handful of lineages. Furthermore, these lineages disproportionately originated in low-transmission settings, especially Southeast Asia and South America. These findings present a great puzzle: although the high-transmission settings of sub-Saharan Africa account for roughly 90% of the global burden of malaria, resistance has seldom evolved locally within Africa [7]. In at least two cases—those of resistance to chloroquine and pyrimethamine—drug-resistant genotypes were carried to Africa from Southeast Asia, after which the imported drug-resistant alleles swept across the African continent via gene flow [4,5]. Resistance mutations must occur frequently in the high-transmission settings of Africa but fail to be selected and spread through the population. This leads to the counterintuitive conclusion that low-transmission settings are more conducive to the sustained transmission of drug-resistant parasites (at least when resistance is rare)—a prerequisite for resistance to become established in the population.
There are at least three different (and nonmutually exclusive) mechanisms that could explain this counterintuitive pattern. The first is recombination during sexual reproduction in the mosquito vector, which occurs more frequently in high-transmission settings [8]. Recombination will tend to separate multiple mutations that contribute to a drug-resistant phenotype, slowing down the spread of resistance; the more loci involved, the stronger the effect will be [9]. Thus, recombination could explain the delayed appearance of multilocus resistance (such as for sulfadoxine-pyrimethamine and ACTs) in high-transmission settings [10,11]. However, recombination is unlikely to slow the spread of resistance encoded at a single locus, such as resistance to chloroquine [4,12]. The fact that both single- and multilocus resistance mechanisms show similar tendencies to evolve in low-transmission settings therefore suggests that recombination is not the primary mechanism delaying the evolution of resistance in high-transmission settings.
The second proposed mechanism is that higher levels of acquired immunity reduce selection for resistance in high-transmission settings. Clinical immunity to malaria, which reduces the likelihood of symptomatic infection, is thought to be acquired through years of frequent exposure, mainly in high-transmission settings [13]. As a result, it is believed that malaria infections are frequently asymptomatic in high-transmission settings (especially in older age groups) but generally symptomatic in low-transmission settings and that levels of antimalarial drug use will accordingly be higher in low-transmission settings, resulting in stronger selection for resistance [14, 15]. However, a 2013 meta-analysis [16] found that, although there is a positive correlation between transmission intensity and asymptomatic infections, a majority of infections are asymptomatic even in low-transmission settings, which suggests that the relationship between transmission intensity and selection for resistance is, at a minimum, weaker than previously believed. Furthermore, historical records indicate that chloroquine resistance spread rapidly across the African continent following its appearance in Kenya and Tanzania in 1978 [3], suggesting that the selection pressure was in fact adequate to drive rapid spread of resistance in Africa.
Here, we address the alternative hypothesis that within-host competition between drug-sensitive and drug-resistant parasites could inhibit the spread of resistance in high-transmission settings. Human malaria infections frequently consist of multiple parasite strains or genotypes, with higher multiplicity of infection observed in high-transmission settings [17]. Strains inhabiting the same host are limited by shared resources and strain-transcending immune responses, leading to ecological competition. Within-host competition between drug-sensitive and drug-resistant parasites has been unequivocally demonstrated in the rodent malaria parasite P. chabaudi; in this model system, drug-sensitive parasites strongly suppress the growth of drug-resistant parasites and reduce their transmission to the mosquito vector [18–22]. Empirical evidence strongly supports within-host competition in P. falciparum in humans as well [23,24]. In particular, a series of cross-sectional studies across the African continent showed that densities of chloroquine-resistant parasites were significantly reduced in children that were coinfected with chloroquine-sensitive parasites, suggesting competitive suppression [23].
To date, few mathematical models have been developed to explore the impact of within-host competition on the spread of drug resistance in malaria. Some of these have suggested that within-host competition could inhibit the spread of resistance [25,26]; however, their conclusions were likely driven by assumptions that drug resistance carried a fitness cost that was either contingent on or exacerbated by within-host competition. This essentially ensured that resistant parasites were intrinsically more affected by competition than their drug-sensitive counterparts, a finding that is not strongly supported by empirical data [23]. Thus, whether within-host competition can explain the delayed evolution of resistance in high-transmission settings on the basis of ecological competition alone, without invoking fitness costs of resistance, remains an unanswered question.
Other models have suggested that within-host competition might actually accelerate the spread of resistance in high-transmission settings because of a phenomenon known as “competitive release.” Experiments in P. chabaudi have shown that competition between sensitive and resistant parasites can be alleviated by antimalarial drug treatment, which removes the sensitive parasites from the host, allowing the resistant population to expand [18]. In P. chabaudi, competitive release increases not only the density of resistant parasites but transmission to mosquitoes as well [22,27]. Relatively simple models incorporating competitive release have shown that it can increase the rate at which resistance spreads through a population [9,28], suggesting that drug resistance should evolve more rapidly in settings with more mixed-strain infections—i.e., high-transmission settings. On the whole, empirical data contradict these predictions, but it is not immediately clear why. Competitive release has never been demonstrated in P. falciparum, so it is not known with certainty that the assumptions made in these models are valid. Alternatively, it may be that rates of antimalarial drug use are not high enough in most circumstances for the effects of competitive release to matter, especially if other factors tend to inhibit the spread of resistance in high-transmission settings. Curiously, although mathematical models have examined the effects of within-host competition and competitive release individually, no effort has been made to examine their combined effects—a significant gap, since the two almost certainly go hand in hand.
Here, we develop a fuller understanding of how within-host competition might affect the evolution of drug resistance in P. falciparum. We do so by considering the complex relationships between transmission intensity, within-host dynamics, and the frequency of drug-resistant parasites in the population (Fig 1). Transmission intensity and the frequency of resistant parasites affect within-host dynamics indirectly by determining the occurrence of mixed-strain infections as well as the acquisition of immunity; treatment with antimalarial drugs affects within-host dynamics as well. Within-host dynamics then determine the transmission potential for sensitive and resistant parasites and thereby feed directly back into the frequency of resistant parasites in the population. Thus, dynamics at the within-host and population levels are connected by “reciprocal feedbacks” that make this question amenable to using a nested model that explicitly describes dynamics on both levels [29].
In order to explore the effects of within-host competition on the establishment and spread of drug resistance, we use a fully nested, individual-based model consisting of a mechanistic model of within-host dynamics embedded into a “between-host model” that describes transmission between humans and mosquitoes. The model captures the complex network of feedbacks illustrated in Fig 1, and—unusually, even for a nested model—allows for within-host dynamics to vary as a result of ongoing exposures and acquisition of immunity. We use this model to explore key questions outlined above: whether within-host competition alone can inhibit the spread of resistance, whether competitive release can accelerate the spread of resistance, and how the net effect depends on the level of antimalarial drug use.
We developed a nested individual-based model consisting of a population of human hosts for which the within-host dynamics of parasites, red blood cells (RBCs), and immune responses are modeled using a set of ordinary differential equations (ODEs). The parasite population is assumed to comprise an effectively infinite variety of strains, which are phenotypically classified as either sensitive or resistant, and the dynamics of these two types are explicitly described in the model. (We effectively assume that there are multiple strains of each type in the population but only a single strain of each type within a given host.) Transmission between human hosts and a mosquito population is simulated using stochastic sampling algorithms; the frequency of human–mosquito contact is used as a proxy for transmission intensity.
We first give an overview of the general behavior of the within-host model and the full nested model. We illustrate the dynamics of the within-host model for a single strain and for two coinfecting strains; for the latter, we show how timing of infection, cross-reactivity between strains, and fitness differences between strains can affect within-host dynamics. For the full nested model, we show how infection prevalence, mixed-strain infections, and the gradual acquisition of immunity are affected by transmission intensity.
We then present the results of simulations in which resistant parasites are introduced at low frequency into populations in which drug-sensitive parasites are endemic. We examine the fate of the resistant parasites in low- and high-transmission settings with zero, low, or high rates of antimalarial drug treatment; we also consider how the outcomes are affected when resistant parasites suffer a fitness cost relative to sensitive parasites. In each instance, we present results for two different degrees of cross-reactivity between strains, because greater cross-reactivity tends to intensify immune-mediated competition at the within-host and population levels.
The within-host model accurately mimics the dynamics of parasites and immune responses observed in P. falciparum infections. Fig 2 shows the simulated within-host dynamics of infection with a single strain, in the absence of preexisting immunity or antimalarial drugs. The dynamics of infected RBCs (Fig 2A) are qualitatively similar to those observed in malaria-therapy patients (examples shown in [30,31]), which are still one of the best sources of data on malaria infection dynamics in naïve hosts. Gametocyte density (Fig 2B), innate immunity (Fig 2C), and acquired immunity (Fig 2D) all more or less track the dynamics of infected RBCs. Note that gametocytes persist several weeks longer than asexual parasites, consistent with observations from human malaria infections [32]. Acquired immunity fluctuates (a result of the way antigenic variation is built into the model) but eventually stabilizes at a high level that serves to reduce parasite growth in future infections.
In contrast to single-strain infections, empirical data on the dynamics of mixed-strain malaria infections in humans are extremely limited; however, two-strain dynamics produced by our within-host model are biologically reasonable and also consistent with observations from the P. chabaudi model system. Fig 3 shows within-host dynamics of two strains (sensitive and resistant), introduced either 20 or 100 days apart, with either high or low cross-reactivity between strains (cross-reactivity is controlled by antigenic overlap, the proportion of antigens or epitopes that are shared between strains). The resistant strain, which is introduced second, reaches higher densities and persists longer when cross-reactivity with the sensitive strain is small, demonstrating that cross-reactive immune responses can contribute to competition between strains. The effect of timing can be most clearly seen in the simulations with higher antigenic overlap. When the strains are introduced 20 days apart, the newly arrived resistant strain immediately goes extinct (probably due to high levels of innate immunity provoked by the drug-sensitive strain); however, when the strains are introduced 100 days apart, the resistant strain overtakes the sensitive strain because of strain-specific adaptive immune responses that disproportionately affect the sensitive parasites. This shows the sensitivity of mixed-strain infection dynamics to the timing of the introduction of different strains, a factor that has been explored in the P. chabaudi model system [19] but is frequently overlooked in mathematical models of superinfection.
Many studies have suggested that drug resistance imposes a fitness cost, with resistant parasites exhibiting impaired growth in vivo [23,33]. When incorporating even a modest cost of resistance, our model suggests that such costs may be strongly exacerbated by within-host competition. Fig 4A shows that, in single infections (without any competition between strains), the impact of a 10% fitness cost of resistance is small; the area under the curve for the resistant strain (total parasites produced over the entire infection) is 4.5% lower than for the sensitive strain. In a mixed infection, however, the area under the curve for the resistant strain is 91.4% lower than for the sensitive strain (Fig 4B).
The full nested model produces epidemiological patterns that are consistent with empirical observations. First, we show how the mosquito biting rate—the parameter that determines transmission intensity in our simulations—determines the entomological inoculation rate (EIR; the number of infective mosquito bites per person per year). EIR was estimated for simulations with four different values of b (mean number of mosquito bites per person per day). For simulations with b = 5, b = 10, b = 15, and b = 20, respectively, the estimated EIRs are 6.1, 25.3, 47.9, and 72.6 infective bites per person per year. These values are not fixed, since EIR depends on the total prevalence of infection in the human population, which can vary both within and between simulations (for instance, an intervention can reduce the prevalence of infection and consequently decrease EIR, even if the mosquito biting rate remains the same). The exact values are less important than the fact that for a two-fold change in biting rate, EIR more than doubles, reflecting the fact that mosquito biting rate affects not only the frequency of contact between humans and mosquitoes but also the likelihood that any particular mosquito is infective, since the prevalence of infection among humans increases with transmission intensity as well. This finding is consistent with a well-known feature of the Ross-MacDonald model, in which R0 is proportional to a2, where a is the rate of contact between humans and mosquitoes [34].
Our model also reproduces the fundamental relationship between transmission intensity and prevalence of infection. Fig 5A shows the relationship between transmission intensity (determined by the mosquito biting rate, b) and total prevalence of infection. Prevalence increases from approximately 70% in the lowest transmission setting (b = 5) to near 100% in the highest transmission setting (b = 20). These results are consistent with epidemiological observations of the relationship between transmission intensity and infection prevalence [35]. Actual estimates of malaria prevalence are significantly lower than those given by the model; this may be attributable to the rapid host turnover in the model, as well as the limits of detection of malaria screening tools (which generally range from 1 to 100 parasites/μL).
The model also reproduces one of the key relationships outlined in Fig 1: that the frequency of mixed-strain infections increases with transmission intensity. Fig 5B shows the relationship between transmission intensity and the frequency of mixed-strain infections (in this context, meaning infections with both drug-sensitive and drug-resistant strains). The median frequency of mixed infections increases from 33% for the lowest transmission intensity (b = 5) to 93% for the highest transmission intensity (b = 20). These results are consistent with observed correlations between transmission intensity and average multiplicity of infection (number of strains per host) [17].
Epidemiological studies have observed that immunity to malaria is acquired over time, with immunity developing more rapidly as transmission intensity increases; in high-transmission settings, clinical immunity to malaria is more or less fully developed by five years of age [13,36]. Again, our model accurately reproduces these patterns. Fig 6 shows the relationship between age and parasite density for four different levels of transmission intensity. In all settings, parasite density decreases with age, but this decrease is more rapid in higher transmission settings, consistent with empirical observations.
In order to understand how within-host competition affects the spread of resistant parasites at the population level, it is helpful to first examine what happens in the absence of selection. We ran simulations lasting 8,000 days (21.9 years), with sensitive parasites introduced at the start and resistant parasites introduced after 3,000 days (8.2 years). Resistant parasites were randomly introduced into 2% of human hosts, with no antimalarial drug use and no fitness cost of resistance. We ran sets of 10 independent simulations for four conditions: low transmission (b = 5) and high transmission (b = 15) with high cross-reactivity and low cross-reactivity between strains. As discussed above, greater cross-reactivity tends to intensify competition between strains.
As shown in Fig 7, the resistant parasites are able to persist longer and achieve higher prevalence in low-transmission settings, particularly with low cross-reactivity between strains. The simulations continue for 5,000 days (13.7 years) after introducing the resistant parasites; the frequency with which the resistant parasites go extinct in this time frame differs across the four conditions tested. The resistant parasites go extinct more often in high-transmission settings than in low-transmission settings (10 out 10 simulations for high transmission/high cross-reactivity compared to 8 out of 10 for low transmission/high cross-reactivity, and 7 out of 10 simulations for high transmission/low cross-reactivity compared to 2 out of 10 for low transmission/low cross-reactivity). The length of time that the resistant parasites persist prior to going extinct also varies: looking only at the cases in which the resistant parasites go extinct, the mean time to extinction is lower for high-transmission settings than for low-transmission settings (2.2 years for high transmission/high cross-reactivity compared to 5.8 years for low transmission/high cross-reactivity, and 2.3 years for high transmission/low cross-reactivity compared to 3.3 years for low transmission/low cross-reactivity).
The cumulative prevalence of the resistant parasites is also starkly different across the four conditions tested. The total number of infection days for resistant parasites is an order of magnitude greater in low-transmission settings compared to high-transmission settings (Table 1). These differences in cumulative prevalence—a measure that accounts for factors such as duration of infection, as well as force of infection and time to extinction—suggest that low-transmission settings are more conducive to persistence and transmission of newly introduced resistant parasites and may consequently offer more opportunities for the rare resistant parasites to be selected by antimalarial treatment. In contrast, persistence of the resistant parasites appears to be strongly inhibited in high-transmission settings because of within-host competition from sensitive parasites occupying the majority of hosts. Cumulative prevalence of the resistant strain is also an order of magnitude lower in the simulations with high cross-reactivity between strains, compared to those with low cross-reactivity. This suggests that immune-mediated competition, which is stronger when cross-reactivity between strains is greater, might play an important role in limiting the persistence of newly introduced resistant parasites.
The higher probability of extinction of resistant parasites in high-transmission settings appears to be driven by more frequent and more severe within-host competition between sensitive and resistant strains. Fig 8 shows cumulative densities of resistant parasites in hosts that are uninfected when the resistant parasites are introduced (single infections) and in hosts that already harbor sensitive parasites when the resistant parasites are added (mixed infections); the proportion of infections falling into each category (single or mixed) is shown above each box-and-whisker plot. Mixed infections occur much more frequently in high-transmission settings, meaning the resistant parasites encounter competitive suppression more often than in the low-transmission settings. Unsurprisingly, cumulative densities of the resistant parasites are reduced in mixed infections compared to single infections; interestingly, the reduction is more severe in high-transmission settings, suggesting that the higher average level of acquired immunity in the high-transmission conditions may contribute to competitive suppression of the newly introduced resistant parasites.
The above analyses demonstrate that within-host competition can inhibit the spread of resistance in the absence of selection from antimalarial drugs or a fitness cost of drug resistance. The next step is to consider how within-host competition might affect the spread of resistance under selection. Antimalarial drug treatment and fitness costs of resistance, of course, exert opposing selective pressures on resistance; the net outcome will depend on the relative strengths of these selective forces. Fig 9 shows the results of simulations with three different rates of antimalarial drug use combined with three different fitness costs (one of which is zero, in order to look at the effect of selection from antimalarial drugs alone). Results for a slightly expanded range of parameter values are presented in S1 Fig.
Panels A–C of Fig 9 show the results of simulations with no cost of resistance and varying levels of antimalarial drug use. With the lowest rate of treatment (Fig 9A), the results are fairly similar to those with no selection (Fig 7); in particular, the resistant parasites consistently go extinct in the high-transmission/high-cross-reactivity setting, despite increasing in frequency in the low-transmission/high-cross-reactivity setting. This suggests that within-host competition is able to prevent the establishment of resistance in high-transmission settings even in the presence of positive selection. With higher rates of treatment (Fig 9B and 9C), however, resistance is able to spread in all settings, but the relationship between transmission intensity and spread of resistance flips, such that the rate of spread of resistance is higher in high-transmission settings. Taken together, these results suggest that in high-transmission settings, drug-resistant parasites are at higher risk of extinction when rare but may actually spread more rapidly if selection is sufficiently strong to avoid extinction in the early stages.
Panels D–I of Fig 9 show results of simulations that include fitness costs of drug resistance as well as antimalarial drug use. These yield three main observations. First, with a low rate of treatment and/or a high fitness cost of resistance, the establishment of resistant parasites is again inhibited in high-transmission settings (Fig 9D, 9E, 9G and 9H). Second, with a very high rate of treatment, resistance spreads to fixation in both low- and high-transmission settings, even in the presence of a fitness cost (S1D, S1H, S1L and S1Q Fig). Third, with treatment rate and fitness cost both at intermediate levels, resistance is able to spread in both low- and high-transmission settings but often fails to reach fixation (Fig 9F and 9I); this effect is particularly pronounced in high-transmission settings. The stable coexistence of sensitive and resistant parasites may result, in part, from exacerbation of the fitness cost of resistance in mixed-strain infections (as shown in Fig 4), which will make the average fitness of the resistant parasites depend on the frequency of mixed-strain infections, as well as the rate of antimalarial drug use.
A key result from these simulations is the fact that drug resistance often spreads (increases in frequency) more quickly in high-transmission settings, despite being at higher risk of extinction when rare. This could be partly attributable to the fact that the prevalence of infection is greater in high-transmission settings; thus, if equal proportions of infections are treated in low- and high-transmission settings, a higher proportion of hosts will be treated (and therefore effectively unavailable to drug-sensitive parasites) in high-transmission settings. However, the spread of resistance in high-transmission settings may also be accelerated by “competitive release,” which refers to enhanced growth of the resistant parasite population in a host following removal of drug-sensitive competitors by drug treatment.
Evidence for competitive release at the within-host level can be seen by comparing the posttreatment density of resistant parasites to the density that would have been observed if the host had gone untreated. We examined the within-host dynamics underlying spread of resistance in low- and high-transmission settings, using model output for one simulation from Fig 9B(i) and one from Fig 9B(ii). Fig 10 shows the distributions of resistant “gains” through competitive release in treated mixed infections from each of these simulations. In most cases, the net gains for resistant parasites are small: 77% of treated mixed infections in low-transmission settings and 97% in high-transmission settings show net gains of less than 1 parasite/μL. However, a minority of treated mixed infections (approximately 9% in the low-transmission setting and 1% in the high-transmission setting) show net gains of over 100 parasites/μL.
In high-transmission settings, the gains achieved by resistant parasites through competitive release might be somewhat reduced due to higher levels of acquired immunity (which acts in a more strain-specific manner than resources or innate immunity and is also slower to respond to changes in parasite density). However, this decreased competitive release is likely offset somewhat by the greater frequency of mixed-strain infections (roughly 28% of all infections in the high-transmission settings—counting only over the period during which sensitive and resistant coexist in the population—versus 8% in the low-transmission setting). In addition, the nonlinear relationship between gametocyte density and transmission to mosquitoes means that the increase in fitness obtained from competitive release is not necessarily proportional to the change in parasite density.
One of the most compelling pieces of evidence that within-host competition could inhibit the evolution of resistance comes from the results of simulations with no selection at all, in which the only forces at work are ecological ones. Simulations with no antimalarial drug use and no intrinsic fitness differences between sensitive and resistant parasites show that resistant parasites introduced at low frequency persist longer in a low-transmission setting than in a high-transmission setting with otherwise identical parameters. Analysis of the within-host dynamics in these simulations shows that, in a high-transmission setting, the newly introduced resistant parasites almost always finds themselves in mixed-strain infections, in which they suffer from competition with drug-sensitive parasites. As the “newcomers” in these mixed infections, the resistant parasites suffer disproportionately from within-host competition, even though they are not intrinsically less fit; this effect of prior residency has been observed in the P. chabaudi model system [19]. In contrast, in low-transmission settings, the newly introduced resistant parasites are more likely to find “unoccupied territory” in hosts without drug-sensitive competitors. As a result, the resistant parasites suffer less from within-host competition and are better able to persist in the population. Since the majority of P. falciparum infections are probably untreated at any given time [16], being able to persist in the population long enough to experience episodes of positive selection by antimalarial drugs is likely to be a key determinant of the evolution of resistance.
These results suggest that within-host competition may impede the spread of resistance even in the absence of a fitness cost of resistance. Other theoretical studies have concluded that within-host competition would slow the evolution of resistance in high-transmission settings, but these results are contingent on a fitness cost of resistance being exacerbated in mixed-strain infections [25,26]. Results from simulations with a fitness cost of resistance are consistent with the idea that costs may be exacerbated in mixed infections, and this may well contribute to the impaired spread of resistance in high-transmission settings; however, fitness costs do not appear to be intrinsic to the effects of within-host competition on the spread of resistance. However, if the magnification of fitness costs in mixed-strain infections serves to prevent the fixation of resistance in the parasite population, the resulting preservation of drug-sensitive parasites could play a key role in the recovery of drug susceptibility when an antimalarial drug is retired—a phenomenon that has been documented in numerous areas [23,37–39] and that has raised hopes for a viable “drug cycling” strategy to manage resistance in P. falciparum.
Across all simulations, the resistant parasites persist longer and/or spread more rapidly with lower cross-reactivity between strains. Cross-reactivity, in this context, determines the extent to which resistant parasites are affected by acquired immunity to sensitive parasites, and vice versa; greater cross-reactivity therefore translates to stronger immune-mediated “apparent competition.” The results showing that lower cross-reactivity is favorable for newly introduced resistant parasites are consistent with the idea that within-host competition may inhibit the spread of resistance. Furthermore, cross-reactivity is likely to be greatest for sensitive and resistant strains drawn from the same local parasite population; a resistant strain imported from a distant population may have markedly lower cross-reactivity with local drug-sensitive strains, giving it a competitive advantage over local parasites [40]. This may favor the establishment of imported resistant parasites in the population and facilitate their spread; this is an interesting hypothesis to explain the spread of drug-resistant genotypes from Southeast Asia across much of sub-Saharan Africa [4,5].
Simulations with antimalarial drug use also show resistance emerging more readily in low-transmission settings—more readily, but not necessarily more quickly. In cases in which the resistant parasites are able to avoid extinction, resistance actually spreads more rapidly in high-transmission settings than in low-transmission settings with otherwise identical parameters. There are two mechanisms that may contribute to this rapid spread of resistance in high-transmission settings. The first is simply that the higher prevalence of infection in high-transmission settings translates into a higher overall rate of use of antimalarial drugs, effectively increasing the strength of selection for resistance. The second is that treatment of mixed-strain infections results in “competitive release” of drug-resistant parasites; this occurs in both low- and high-transmission settings but more frequently in the latter because of higher frequencies of mixed-strain infections. In contrast to prior experimental and modeling studies [9,18,20], we find that the gains attained through competitive release are generally modest; with weaker drug-resistant phenotypes (such as delayed clearance in response to artemisinin-based therapies), the gains are likely to be even smaller. However, with drugs that target gametocytes as well as asexual parasites, treatment of mixed infections may enhance transmission of resistant parasites even without large gains from competitive release, since rapid clearance of drug-sensitive gametocytes will reduce competition for transmission to mosquitoes. Thus, the impact of competitive release is likely to depend on the mode of action and degree of resistance to the drug(s) being used.
Overall, the results presented here suggest that the relationship between transmission intensity and evolution of resistance may be more nuanced than previously appreciated. This nuance arises in the distinction between establishment of resistance and spread of resistance [41,42]. High-transmission settings may be less conducive to the establishment of resistance, since the probability of extinction for newly introduced resistant parasites is higher [43]. However, the rate of spread of resistance can nevertheless be greater in high-transmission settings, as a result of competitive release, a higher overall frequency of antimalarial drug use, or both. These opposing findings help to reconcile previous models predicting rapid evolution of resistance in high-transmission settings [9,28] with the observed tendency for resistance to evolve in low-transmission settings. The distinction between probability of establishment and rate of spread may be a fruitful one to explore further in models of drug resistance evolution. For instance, although models have previously been used to explore the effects of recombination on the spread of multilocus drug resistance, the impact of recombination on establishment of multiple-mutant genotypes in the population, particularly in the presence of epistatic interactions, may be an interesting puzzle to explore in future work.
Finally, we emphasize that this model cannot rule out the possibility that recombination, unequal rates of antimalarial drug use, and/or fitness costs of resistance may be important drivers of the observed relationship between transmission intensity and evolution of drug resistance. However, our results show that the observed relationship can be qualitatively explained without invoking any of the three. A key area in which to expand this work is consideration of how these different factors might interact—for example, how the effects of within-host competition might change when mutations at multiple loci contribute to resistance. A great deal of theoretical and empirical work is still needed to quantify the relative contributions of recombination, selection pressure, within-host competition, and other mechanisms to variation in the rate of drug resistance evolution in P. falciparum.
Understanding the relationship between transmission intensity and evolution of resistance is likely to lead to improved strategies for controlling malaria and managing drug resistance [43]. For instance, the results presented here suggest that the fate of drug-resistant parasites may be more sensitive to the rate of antimalarial drug use in high-transmission settings, in the sense that strong selection might be a prerequisite for drug-resistant parasites to spread at all in these settings. Thus, interventions such as mass drug administration (MDA) may present different risks in low- versus high-transmission settings; although increased use of antimalarial drugs will always increase selection for resistance, in low-transmission settings, this seems likely to be a matter of degree, whereas in high-transmission settings, increased selection has the potential to “tip the scales” in favor of resistance, allowing resistant parasites to establish and spread where they otherwise would not. (It should be noted, however, that MDA is primarily recommended for use in low-transmission settings in pursuit of malaria elimination.) More broadly, reductions in malaria transmission could—paradoxically—increase opportunities for resistance to evolve, necessitating increased vigilance and/or compensatory measures to ensure that resistance does not gain a foothold in newly “unoccupied territory” [44,45]. Knowledge of the factors that limit—or drive—the spread of resistance will aid in optimizing control strategies for a wide range of different endemic settings [46,47].
We use a nested model of parasite population dynamics: a model of within-host dynamics embedded into another model that simulates parasite transmission between humans and mosquitoes. Essentially, the model tracks a population of humans, a population of mosquitoes, and the parasites that circulate among them. Infections in mosquitoes are tracked over time, while the dynamics of infection in each human host are modeled using ODEs. The model simulates within-host dynamics in 24-hour increments interspersed with daily transmission of parasites between humans and mosquitoes; any new parasites that are introduced to a human host are incorporated into the population of parasites tracked by the within-host model. The general structure of the full model is illustrated in Fig 11.
The model is designed to explore questions related to the evolution of drug resistance; therefore, it actually describes the dynamics of two “types” of parasite: drug sensitive and drug resistant. We assume the parasite population is comprised of a variety of strains, which are phenotypically classified as either sensitive or resistant (in this model, resistance is absolute; we do not consider multiple resistance mutations or different degrees of resistance—such as delayed-clearance phenotypes as seen with artemisinin resistance—but doing so will be an important direction for future work). Different strains are assumed to have partial overlap in their antigens; for computational tractability, we assume that any two strains have the same degree of overlap, and the number of strains in the population is infinite. Within an individual host, all of the parasites of one type are considered to constitute a single strain; thus, an infection with only one type is a “single-strain infection,” while one with both types is a “mixed-strain infection”.
We now give an overview of the within-host model, the between-host (transmission) model, and a few other important aspects. A detailed description of the model is provided in S1 Text.
The within-host model consists of a system of ODEs that describe the dynamics of infection for two parasite “types” (sensitive and resistant, denoted with subscripts 1 and 2, respectively). The dynamics of the following components are described:
Below are the ODEs for the within-host model. These equations, other than those governing acquired immunity, are similar to previous within-host models of malaria [48]. Subscripts i and j are used to indicate type-specific variables and parameters (if i = 1, then j = 2, and vice versa). The definitions of all parameters in Eqs 1–7 are given in Table 2 (parameter values in S1 Text), and Fig 12 shows a compartment-style schematic of the within-host model.
The dynamics described by the within-host model boil down to two essential processes: parasite replication and parasite–immune system interactions. Parasite replication is straightforward: merozoites invade RBCs, turning them into infected RBCs; infected RBCs either produce more merozoites or differentiate into gametocytes. RBCs are lost during this process but are continually replaced by new ones. Interactions between parasites and the immune system are more complicated. The model includes two types of immunity: innate and acquired (sometimes called adaptive). Innate immunity acts in a strain-transcending manner and is important for controlling parasite growth during the acute phase of the infection; acquired immunity is at least partly type specific (the degree of specificity is determined by the model parameters) and required for eventual clearance of the infection. Both innate and adaptive immune responses are triggered by parasites, and both decay in the absence of continued stimulation; however, the “on” and “off” rates for innate immunity are higher, resulting in a fast, self-limiting response. Acquired immunity takes longer to develop but decays very slowly and helps to limit parasite growth in subsequent infections. However, the dynamics of acquired immunity are complicated by antigenic variation, which is discussed in more detail below.
Antigenic variation is an evolved strategy for evasion of acquired immunity. The parasite has several dozen “variants” of an immunodominant surface protein but only expresses one variant at a time; when the immune system learns to recognize the current variant, the parasite switches to a different one [49,50]. Antigenic variation therefore interferes with recognition and killing by the adaptive immune system and helps to prolong the infection, which may not be cleared for several months (presumably when the variant repertoire has been exhausted). Thus, antigenic variation plays a key role in the dynamics of P. falciparum infections; however, explicitly modeling the process is extremely computationally demanding. Instead, our model implicitly incorporates antigenic variation by describing its effects on the dynamics of infection. Each switch to a novel variant interferes with recognition by the adaptive immune system, which makes the immune response less effective; this loss of effectiveness is mathematically indistinguishable from a loss of immune effectors. We therefore incorporate antigenic variation as a second decay term in the equations for adaptive immunity, which diminishes with the progress of the infection, since the variant repertoire is finite and eventually runs out.
Unlike the within-host model, which is governed by deterministic equations, the between-host model is stochastic; as a result, multiple simulations with the same parameters and starting conditions will tend to yield similar but not identical results—unless the conditions favor highly divergent trajectories, such as scenarios that tip either toward extinction or toward epidemic spread.
The between-host model describes human-to-mosquito and mosquito-to-human transmission of parasites. Each human host is assigned to be bitten by a randomly chosen set of mosquitoes each day (the number of mosquitoes is based on the transmission intensity); these mosquitoes can infect and/or become infected by the human host when they feed. The probability of a mosquito becoming infected is determined by the total gametocyte density in the human host, using a gametocytemia-infectivity function originally described by Churcher and colleagues [51]. If the mosquito is determined to be infected, the number of gametocytes of each type picked up is determined by the individual gametocyte densities of drug-sensitive and drug-resistant parasites. If parasites are acquired, there is a latent period (10 days in the model) before the infection reaches the salivary glands and the mosquito becomes infective. Mosquitoes are allowed to acquire parasites from multiple hosts (though only one host per day); parasites acquired from different blood meals are tracked separately through the latent period and up to the point of transmission to a human host.
An infective mosquito has a constant probability of introducing parasites to the human host it feeds on. If this does occur, a fixed number of sporozoites is introduced; how many of these are drug-sensitive and drug-resistant depends on the ratio of the two types in the pool of gametocytes the mosquito originally acquired. Once sporozoites are transmitted to the human host, the infection goes through a latent period (i.e., the liver stage), which ends with merozoites being released into the bloodstream, at which point the parasites become subject to the within-host model.
An aspect of the model with particularly broad ramifications is the diversity of the parasite population, particularly as it pertains to recognition by the adaptive immune system. In the simulations presented here, we assume a complete lack of population structure: all strains in the population, whether sensitive or resistant, are equally “related.” (However, we note that the design of the within-host model results in competition between sensitive and resistant strains being slightly weaker than competition between strains of the same type, particularly in hosts that have had only a few infections. This permits stable coexistence between the strains in settings in which transmission and cross-reactivity between different strains are both low—which is not unreasonable, since some degree of linkage between resistance mutations and antigenic loci is to be expected, particularly in low-transmission settings in which recombination rates are low) [52].
Relatedness of different strains is governed by the parameters λ and μ, which specify the fractions of fixed (nonvariant) and variant antigens shared by any two strains (for high cross-reactivity, we have λ = 0.7 and μ = 0.3; for low cross-reactivity, we use λ = 0.35 and μ = 0.15). These parameters determine cross-reactivity, or how much protection against one strain is conferred by previous exposure to a different strain [53]. This affects how quickly immunity is acquired: if cross-protection is minimal, then it may take many exposures to build up effective immunity [54, 55]. Cross-reactivity also determines the severity of immune-mediated competition between sensitive and resistant parasites, which affects the ability of resistant parasites to survive and spread in the face of competition from drug-sensitive parasites.
We note that in P. falciparum, genetic diversity is known to be greater in high-transmission settings [8], so strains in high-transmission areas might be expected to have lower cross-reactivity than strains in low-transmission regions. Thus, although it may be more straightforward to compare simulations that differ only in transmission intensity or only in cross-reactivity between strains, it may be important to compare high-transmission, low-cross-reactivity settings with low-transmission, high-cross-reactivity ones. Therefore, we present simulation outputs in sets of four—high and low transmission with high and low cross-reactivity—to make relevant comparisons while disentangling the effects of transmission intensity and cross-reactivity between strains.
In the simulations presented here, antimalarial drug treatment is conditional only on the host being infected with parasites (above the extinction threshold, a total infected RBC density greater than 10−4). If a host is infected (and not already being treated), there is a fixed daily probability of beginning antimalarial treatment. If started, drug treatment is maintained for a fixed duration of 14 days, regardless of whether parasite clearance is achieved.
Simulated populations consisted of 400 human hosts and 12,000 mosquitoes; transmission intensity was determined by the rate of contact between humans and mosquitoes rather than the ratio of mosquitoes to humans. Simulations were run for a total duration of 8,000 days. Age was uniformly distributed with a maximum of 3,000 days; when hosts reached this limit, they were removed and replaced with naïve hosts of age zero. Drug-sensitive parasites were introduced at an initial prevalence of 10%, and the simulation was run for 3,000 days to allow the system to reach equilibrium before introducing drug-resistant parasites at a prevalence of 2%. Use of antimalarial drugs, when included, was initiated at the start of each simulation.
Certain parameters (human population size, starting prevalence of resistant parasites, host lifespan) were constrained by computational demands. A larger population size would considerably increase simulation run times and would not significantly change the results if the number of hosts initially infected with resistant parasites were held constant (this is because transmission is frequency dependent rather than density dependent). A lower starting prevalence of resistance, or stochastic introduction of resistance, would better simulate de novo emergence of resistance mutations; however, this would have significantly increased the number of simulations required, due to higher rates of extinction. Instead, we introduced resistant parasites into a fixed, slightly higher number of hosts in each simulation, representing hosts infected by a recently emerged drug-resistant mutant. Host lifespan was limited by the time required for the simulation to reach equilibrium before introducing resistant parasites. A longer lifespan would be expected to alter the distribution of immune states in a population, which would affect within-host dynamics (including competitive suppression and competitive release in mixed-strain infections) as well as between-host dynamics (including infection prevalence and the frequency of mixed infections). However, the qualitative differences between low- and high-transmission settings that underlie the results of our model would be expected to hold.
|
10.1371/journal.pmed.1002431 | Validity of a minimally invasive autopsy for cause of death determination in maternal deaths in Mozambique: An observational study | Despite global health efforts to reduce maternal mortality, rates continue to be unacceptably high in large parts of the world. Feasible, acceptable, and accurate postmortem sampling methods could provide the necessary evidence to improve the understanding of the real causes of maternal mortality, guiding the design of interventions to reduce this burden.
The validity of a minimally invasive autopsy (MIA) method in determining the cause of death was assessed in an observational study in 57 maternal deaths by comparing the results of the MIA with those of the gold standard (complete diagnostic autopsy [CDA], which includes any available clinical information). Concordance between the MIA and the gold standard diagnostic categories was assessed by the kappa statistic, and the sensitivity, specificity, positive and negative predictive values and their 95% confidence intervals (95% CI) to identify the categories of diagnoses were estimated. The main limitation of the study is that both the MIA and the CDA include some degree of subjective interpretation in the attribution of cause of death.
A cause of death was identified in the CDA in 98% (56/57) of cases, with indirect obstetric conditions accounting for 32 (56%) deaths and direct obstetric complications for 24 (42%) deaths. Nonobstetric infectious diseases (22/32, 69%) and obstetric hemorrhage (13/24, 54%) were the most common causes of death among indirect and direct obstetric conditions, respectively. Thirty-six (63%) women were HIV positive, and HIV-related conditions accounted for 16 (28%) of all deaths. Cerebral malaria caused 4 (7%) deaths. The MIA identified a cause of death in 86% of women. The overall concordance of the MIA with the CDA was moderate (kappa = 0.48, 95% CI: 0.31–0.66). Both methods agreed in 68% of the diagnostic categories and the agreement was higher for indirect (91%) than for direct obstetric causes (38%). All HIV infections and cerebral malaria cases were identified in the MIA. The main limitation of the technique is its relatively low performance for identifying obstetric causes of death in the absence of clinical information.
The MIA procedure could be a valuable tool to determine the causes of maternal death, especially for indirect obstetric conditions, most of which are infectious diseases.
The information provided by the MIA could help to prioritize interventions to reduce maternal mortality and to monitor progress towards achieving global health targets.
| Since 1990, the maternal mortality ratio (MMR) has dropped by 43%, but despite this progress, hundreds of women still die every day in large parts of the world due to pregnancy or childbirth complications.
A reliable knowledge of the causes of maternal death is a necessary condition to reduce this burden through adequate health planning.
Current methods, such as verbal autopsies or the review of clinical data, have shown a high degree of misclassification in this specific group of deaths.
We aimed to validate a minimally invasive autopsy (MIA) approach as a possible complement to verbal autopsies by comparing its performance against the complete diagnostic autopsy, the gold standard for cause of death investigation.
We performed paired MIA and CDA in 57 maternal deaths that occurred at the Maputo Central Hospital, Mozambique, and assessed the concordance between both methods.
Indirect obstetric diseases caused 56% of all deaths and direct obstetric complications accounted for 42% of deaths.
Infectious diseases and obstetric hemorrhage were the most common causes of death among indirect and direct obstetric conditions, respectively.
The overall concordance of the MIA with the CDA was moderate and both methods agreed in 68% of the diagnostic categories.
The etiological microorganisms in infectious diseases causes of death were identified in 67% of the MIAs.
The MIA procedure could be a valuable tool to determine the causes of maternal death, especially in indirect obstetric conditions, most of which are infectious diseases.
This information may be helpful for decision-making on health planning and prioritization of interventions to reduce maternal mortality and for monitoring progress towards achieving global health targets.
| As the Millennium Development Goals came to a close in 2015, the maternal mortality ratio (MMR) had dropped by 43% since 1990 [1]. Although this progress is certainly encouraging, it is lower than the 75% target initially planned, and hundreds of women still die every day due to complications of pregnancy or childbirth. In 2015, the number of estimated maternal deaths occurring worldwide was 303,000, most of which were preventable and disproportionately took place in low- and middle-income countries. The MMR in low-income countries in 2015 was 239 per 100,000 live births versus 12 per 100,000 live births in high-income countries, according to WHO [2]. In Mozambique, in 2015, the MMR was 489 per 100,000 live births, with an annual rate of reduction of 4.4% from 2005–2015 [3]. The aim of Sustainable Development Goal 3.1 is to reduce the global MMR to less than 70 per 100,000 live births by 2030 [4]. The accomplishment of this objective requires robust data sources to develop accurate estimates and, importantly, a thorough understanding of the causes of these deaths. Monitoring progress towards the 5th Millennium Development Goal, focusing on maternal health, revealed the lack of high-quality data in most countries, especially those with the highest MMR [1].
Recently, a minimally invasive autopsy (MIA) protocol has been adapted for cause of death determination in middle- and low-income settings [5]. This postmortem procedure, which consists of the sampling of key organs and fluids for histological and microbiological analysis, could add value to the currently used methods that rely on verbal autopsies and clinical records, which have been shown to have a high level of imprecision, especially for maternal and perinatal deaths [6,7].
The complete diagnostic autopsy (CDA) is the gold standard for cause of death determination but it is not free of limitations. The CDA is an invasive procedure, often not accepted by the relatives, and requires trained pathologists to perform it. The MIA, as opposed to the CDA, has been shown to be simple and more acceptable [8] and can be performed by less qualified personnel. The MIA has been recently validated in a series of in-hospital stillbirth, neonate, pediatric, and adult deaths in Mozambique [5]. These validation studies have shown that the MIA may reliably identify the cause of death with high concordance when compared with the gold standard CDA diagnosis, particularly for infectious diseases.
In this study, we analyzed the validity of the MIA to determine the cause of death in a series of in-hospital maternal deaths from Mozambique.
This observational study received the approval of the Clinical Research Ethics Committee of the Hospital Clinic of Barcelona (Spain; approved, File 2013/8677) and the National Bioethics Committee of Mozambique (Mozambique; approved, Ref. 342/CNBS/13).
The study was conducted at the Departments of Gynecology and Obstetrics and Pathology of the Maputo Central Hospital, a 1,500-bed government-funded institution that serves as the referral center for other hospitals in southern Mozambique. Recruitment of maternal deaths was conducted from November 2013 to March 2015 as part of a comprehensive validation study of the validity of the MIA in different age groups [9].
We included in the study all deceased women fulfilling the following inclusion criteria: (1) death during pregnancy or within 42 days of termination of pregnancy, irrespective of its cause (maternal deaths, as defined by WHO) [10,11]; (2) a CDA requested by the clinician as part of the medical evaluation of the patient; and (3) oral informed consent to perform the autopsy given by the relatives. Accidental or incidental deaths were excluded.
The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement and the prospective analysis plan are included as supplementary information (S1 STROBE Checklist and S1 Text, respectively).
The overall study plan, indicating how postmortem procedures were performed, the laboratory investigations involved, and site and timing of each procedure, has been reported elsewhere [5].
Detailed MIA pathological and microbiological methods have been reported elsewhere [12,13]. The procedure, tailored for maternal deaths, included an initial disinfection of the surface of the body followed by the collection of blood and cerebrospinal fluid (CSF), aiming to collect about 20 mL. The procedure also included the puncture of solid organs (liver, lungs, and central nervous system [CNS]) for microbiological and pathological analysis using biopsy needles (14G–16G). In addition, the heart, spleen, and kidneys were targeted for pathology examination. In this particular group, biopsy samples of the uterus were attempted with a 14G biopsy needle, which were analyzed histologically and microbiologically. In addition, a sample of amniotic fluid, if available, was obtained.
Immediately after the MIA, the CDA procedure was conducted by a second pathologist not involved in the MIA and following a standardized protocol for maternal autopsies [5]. Histological and microbiological analyses were conducted in samples from the same viscera collected in the MIA and from any grossly identified lesions. The microbiological results of the blood and CSF were also included in the CDA evaluation.
A team of 2 pathologists (J.O., P.Ca.) and 2 microbiologists (M.J.M., J.H.) reviewed and analyzed the samples from the MIA blindly to any clinical data except the information on the maternal status at death (pregnant, puerperal, previous abortion) that allowed classifying the case as a maternal death. All samples collected for histology were routinely stained with hematoxylin and eosin. Additional histochemical (e.g., Masson trichrome) and/or immunohistochemical stains and in situ hybridization techniques (e.g., Epstein–Barr virus-encoded small RNAs) were used, whenever needed, to reach a diagnosis. The microbiological methods have been reported in detail elsewhere [13]. In all patients, screening for context-epidemiologically common pathogens was conducted, including Plasmodium falciparum by real-time PCR detection of antibodies against human immunodeficiency virus (HIV)-1/2, hepatitis B virus surface antigen, and antibodies against hepatitis C virus. In all HIV-infected patients, the viral load was determined routinely. Additionally, Toxoplasma gondii, Mycobacterium tuberculosis, and Cryptococcus spp. in CSF and CNS samples and Pneumocystis jirovecii, Cryptococcus spp., and M. tuberculosis in the lungs were tested by real-time PCR. Any other microorganisms were only investigated depending on the pathological findings observed in the MIA-obtained tissues.
After a washout period of 3–6 months, the same team of experts analyzed the samples of the CDA following the same approach used for the analysis of the MIA samples.
The methodology for cause of death determination has been detailed elsewhere [5]. Briefly, once all the analysis of the MIA samples had been completed, a panel composed of a pathologist, a microbiologist, an obstetrician, and an epidemiologist evaluated all the data of the MIA and assigned the MIA diagnosis, i.e., the disease or condition putatively leading to death. The participants in the MIA cause of death attribution panel were aware of the external macroscopic examination of the body (but not of the organs) and the histological and microbiological results obtained in the MIA sampling but were blind for the information on the clinical records, except for the pregnancy status and gestational age.
After a washout period (minimum time 3 months, range 3–6 months), the same panel evaluated the data from the CDA and assigned the CDA diagnosis of cause of death, which was considered the gold standard. As per routine practice in CDA procedures, investigators involved in the CDA diagnosis were aware of the clinical data, the macroscopic (external and internal) findings, and the histological and microbiological results. Using a combination of the strength of the evidence of the histological and the microbiological findings, a category was assigned to the certainty of the cause of death attribution of the MIA diagnosis and the CDA diagnosis (details published elsewhere) [5]. In this particular group, the obstetric history of severe intrapartum hemorrhage increased by 2 levels the strength of the pathological findings of the CDA.
All morbid conditions were codified following the International Classification of Diseases, 10th revision for maternal mortality (ICD-10 MM) [10,14]. Up to 4 diagnoses were established following the most probable sequence of events leading to death, starting from the immediate cause and going back to the earliest event, i.e., the underlying condition. Finally, other conditions or concomitant infections that may have existed prior to development of the underlying cause of death or developed during the chain of events leading to death and that, by their nature, contributed to the death, were considered as contributory conditions (e.g., multiple pregnancy) [10]. As indicated by WHO, all deaths were classified either as (1) direct obstetric deaths (i.e., those resulting from obstetric complications of the pregnancy, labor, or puerperium) or (2) indirect obstetric deaths (i.e., those resulting from previous existing diseases or diseases developed during pregnancy but not due to direct obstetric causes) [10].
Additionally, all deaths were aggregated in 8 categories according to the underlying causes of death as stated in the ICD-10 MM [10], which included the following: (1) pregnancies with abortive outcome; (2) hypertensive disorders in pregnancy, childbirth, and puerperium; (3) obstetric hemorrhage; (4) pregnancy-related infections; (5) other obstetric complications; (6) unanticipated complications of management; (7) nonobstetric complications; and (8) unexplained deaths. Nonobstetric complications include the following diseases: cardiovascular (including preexisting hypertension), endocrine, gastrointestinal tract, CNS, respiratory, genitourinary, autoimmune, skeletal, psychiatric diseases, neoplasm, and infections that are not a direct result of pregnancy. Categories 1 to 6 were considered direct obstetric deaths, and deaths classified as category 7 were considered indirect obstetric deaths.
The concordance between the MIA and the CDA diagnosis (gold standard) was established based on the concordance in the underlying cause of death categories, was assessed by the kappa statistic, and was interpreted as suggested by Landis and Koch [15]. The diagnostic performance of the MIA to identify the categories established by the gold standard diagnosis was evaluated in terms of sensitivity, specificity, positive and negative predictive values, and total percentage of cases correctly classified.
The level of coincidence between the MIA and the CDA diagnoses was analyzed by comparing the ICD-10 MM codes in order to identify the conditions that were concordant in the disease category but were not coincident in the main diagnosis (e.g., a case categorized as an indirect obstetric death in both methods but as pneumonia in the CDA and as sepsis in the MIA). As the ICD-10 system classifies diagnoses into nested classes of different hierarchical levels in which diseases or conditions are organized in chapters, blocks, and 3-character categories [14], a coincidence was classified as (i) perfect when the ICD-10 codes were identical in chapter, block, and 3-character categories; (ii) moderate when the codes were within the same chapter and block but there was a discrepancy in the 3-character category; or (iii) low when the codes were within the same chapter but not in the same block and 3-character categories. When the MIA and the CDA diagnoses were in different chapters, the coincidence was classified as “none.” The statistical analyses were performed using Stata version 14.1 (Stata, College Station, TX, USA). The analytical plan was determined once all the histological and microbiological results as well as the ICD10-MM codes were available.
Coupled MIA and CDA procedures were performed in 57 women. Median age was 27 years (range 15–39 years). Thirty-one women (54%) died in the puerperal period, 9 (16%) died after an abortion, and 16 (28%) were pregnant at the time of death. In all 16 cases, the fetus and the placenta were in situ. Nevertheless, no autopsy was performed on these in situ fetuses because it was not considered as relevant to determine the cause of death. The placenta was analyzed as part of the CDA in 8 of these cases but was not contributory to the final diagnosis. Forty-four lived in the Maputo urban area, whereas 13 were transferred from a rural district.
In 40 cases (70%), the interval between death and the beginning of postmortem procedures was ≤24 hours, whereas in 17/57 (30%), it ranged between 25 and 64 hours. The age of the patients, time from death to procedure, HIV status, MIA and CDA diagnoses, ICD-10 coding, level of certainty, and concordance between both diagnoses are shown for all patients in S1 Table.
Blood was obtained in all cases, but in 3 women, less than 10 mL were obtained (in 1 of these cases, the MIA was done 34 hours after death). In 56/57 cases (98%), 10 mL of CSF were obtained. In the MIA procedure, liver, CNS, and bone marrow samples were obtained in 100% of the cases, whereas lung was sampled in 93%, heart in 86%, spleen in 79%, and kidneys in 61% of the cases. The uterus was successfully obtained in 53/57 (93%) of the cases. Amniotic fluid was obtained in 7 (12%) cases. The placenta was available for analysis in 8/16 (50%) women that were pregnant at the time of death.
S1 Table summarizes the cause of death for each patient according to each of the 2 methods used.
A cause of death was identified in the CDA in 98% (56/57) of cases. The level of certainty of the CDA diagnosis was considered high or very high in 51/57 (89%) cases and moderate in 5 cases (9%).
Direct obstetric conditions accounted for 24 out of 57 deaths (42%). This included obstetric hemorrhages (13/24; 54%), complications of abortive pregnancies (6/24; 25%), pregnancy-related infections (4/24; 17%), and hypertensive disorders of the pregnancy (1/24; 4%) (Table 1, S1 Table). No deaths were classified in the categories “other obstetric complications” and “unanticipated complications of management”.
Indirect obstetric conditions accounted for 32/57 (56%) deaths. Among them, infectious diseases accounted for 22/32 deaths (69%), other diseases not related to pregnancy for 7/32 (22%), and malignant tumors for 3/32 (9%) of the deaths (Table 1, S1 Table). Five women died of tuberculosis (M. tuberculosis) and 4 of cryptococcal infection. Cerebral malaria with histological evidence of sequestration of parasitized erythrocytes in the capillaries of the CNS and P. falciparum detected by PCR was diagnosed in 4 women. Of these 4 patients, 3 (out of 44; 6.8%) came from the Maputo urban area, whereas 1 (out of 13; 7.7%) was transferred from a rural district.
HIV was detected in 36 women (63%). All HIV positive cases showed detectable viral load for HIV-1. HIV infection was identified in 13/24 (54%) women who died from direct obstetric causes and in 22/32 (69%) women who died from nonobstetric complications. HIV was also detected in the single case with nonconclusive diagnosis. Nearly 3 quarters (16/22; 73%) of the HIV-positive nonobstetric maternal deaths were considered to be AIDS related, which included all 5 tuberculosis cases, the 4 cryptococcal disseminated infections, 3 pneumonias, 1 streptococcal sepsis, 1 pyelonephritis, 1 meningoencephalitis, and 1 Burkitt lymphoma. The other 6 HIV-positive cases died from cerebral malaria (3 cases), liver failure due to massive liver necrosis (2 cases), and dilated cardiomyopathy (1 case). Hepatitis B virus infection was identified in 3 HIV-positive cases.
A cause of death was identified in the MIA in 48 out of the 57 cases (84%). In 9 cases (16%), the MIA diagnosis was nonconclusive. The certainty of the MIA diagnosis was high or very high in 22/48 (46%) cases, moderate in 19/48 (40%), and low in 7/48 (14%).
Table 1 shows the diagnostic concordance between the MIA and the CDA (gold standard) for the categories of causes of death. The 2 procedures agreed in the diagnostic categories in 39/57 cases (68%) and the concordance was moderate according to the kappa statistics (kappa = 0.48, 95% CI: 0.31–0.66).
A perfect or almost perfect coincidence in the cause of death established by each method was observed in 36 of the 39 cases that were concordant in the diagnostic category. One case showed a low level of coincidence (diagnosed as chronic hypertensive disease in pregnancy in the CDA and as suggestive of cardiovascular disease in the MIA). The 2 remaining cases were noncoincident (a case of pneumonia and a case of liver necrosis according to the CDA, diagnosed as sepsis due to Escherichia coli and pneumonia, respectively, in the MIA).
Table 2 shows the sensitivity, specificity, and the positive and negative predictive values of the MIA diagnosis for the major diagnostic categories, as well as the percentage of false-positive and false-negative diagnoses, and the cases correctly classified (accuracy) in the MIA. The sensitivity of the MIA was very high for indirect causes of death (91%) and low (<10%) for obstetric hemorrhage. Overall, the accuracy of the MIA was at least 79% for all diagnostic categories.
Overall, infectious diseases accounted for 30/57 (57%) of maternal deaths. Infectious diseases included 8 obstetric infections (4 puerperal sepsis and 4 septic abortions) and 22 nonobstetric infections. An etiologic agent was identified in the CDA in 22 out of the 30 (73%) infectious diseases (6 obstetric and 16 nonobstetric). The same microorganism was identified in both the CDA and the MIA in 20 out of the 22 cases (91%). A microorganism was identified in the MIA in 13 additional cases and considered the MIA cause of death. In 5 cases, both the CDA and the MIA diagnosed an infectious disease, but whereas the CDA did not identify any microorganism, the MIA identified a causative agent. In 6 cases, the CDA (gold standard) cause of death was hemorrhagic shock and, consequently, the microorganism identified in the MIA was considered as not contributing to the death. Finally, in 2 cases, the CDA diagnosed an infectious disease, probably of bacterial origin, with no agent identifiable (a meningoencephalitis and a puerperal sepsis, respectively), whereas in the MIA the cause of death was considered to be a disseminated cytomegalovirus infection. In both cases, the cytomegalovirus was also identified in the CDA but it was not considered as the cause of death. Table 3 shows the etiological agents identified in the CDA and in the MIA.
This validation study shows, for the first time to our knowledge, that the MIA, a simplified postmortem procedure, can provide an acceptable correlation with the gold standard CDA diagnosis in maternal deaths. In this group, the agreement of the technique was 68% (kappa statistic: 0.48, 95% CI 0.31–0.66), a percentage of agreement similar to those observed in other age groups [16,17]. These findings suggest that this new method could provide reliable and relevant data regarding the causes of mortality associated with pregnancy and childbirth, particularly for indirect obstetric deaths, and thus contribute to reducing maternal mortality in the settings where this burden is highest.
This study was not designed to describe the causes of maternal death. However, our findings are similar to previous reports indicating that obstetric hemorrhage and indirect obstetric conditions are the leading causes of maternal mortality in low-income countries [18,19]. Interestingly, and unlike what it is usually reported [20,21], the proportion of indirect obstetric deaths (56%) observed in our study is larger than that of direct obstetric deaths (42%). These results are similar to a previous study carried out over 10 years ago in the same setting [19] and suggest that nonobstetric complications may be missed using the current methodologies and sources of data collection utilized for maternal cause of death estimation. Our study also confirmed the significant contribution of malaria and HIV to maternal mortality in endemic countries such as Mozambique [1,19,22,23]. In this series, malaria accounted for 7% (4/57) and HIV-related conditions for 28% (16/57) of all maternal deaths. Cerebral malaria could be detected in the MIA in all 4 cases [24], a finding of particular relevance, taking into account that the study was conducted in an urban setting where malaria transmission is known to be minimal. Interestingly, 3 out of the 4 women with cerebral malaria were of urban origin, although we cannot exclude a visit to a rural area with higher malaria transmission during pregnancy. This finding is very relevant for improving the knowledge on the impact of malaria on maternal mortality in endemic areas, which is usually underestimated [25]. As observed in previous reports [19,26,27], HIV and tuberculosis coinfection were found to be important causes of indirect obstetric deaths (9% of deaths). Tuberculosis was identified in the MIA in 4 out of 5 cases in which the infection was determined to be the cause of death in the CDA. This is of relevance given the low sensitivity of the clinical diagnosis of tuberculosis as a cause of maternal mortality [28]. Importantly, disseminated cryptococcal infection contributed significantly to death among HIV-positive women (7%, 4/57), and all the cases were also identified in the MIA. Cryptococcus spp. is the leading cause of adult meningitis in sub-Saharan Africa, where it is estimated to cause 15%–20% of all AIDS-related deaths [28–30]. Nevertheless, data in pregnant women are scarce, with only a few published studies having reported cryptococcal infection among pregnant women. This study highlights the importance of cryptococcosis as a cause of maternal death and thus the need to improve the diagnosis and management of this fungal infection during pregnancy [31–33].
Interestingly, the sensitivity of the MIA procedure was very high (91%) for indirect causes of death. In contrast, the sensitivity was much lower (36%) for direct obstetric causes. The sensitivity of the MIA procedure was particularly poor (less than 10%) for obstetric hemorrhage. Only 1 case was captured in the MIA because a retained placenta was identified in a woman during puerperium. Although only 1 death was attributed to eclampsia, the disease was identified in the MIA. This accuracy was probably influenced by our strict criteria to diagnose eclampsia, which required the presence of the typical pathological lesions in the liver and the absence of any other lethal lesion, in order to avoid overassignment of this condition as a cause of death [28]. This is relevant given the high number of false-positive clinical diagnoses of eclampsia that has been reported and the likely overestimation of this condition in clinical reports and verbal autopsies [34,35]. Indeed, it is essential to establish thoroughly the contribution of eclampsia to maternal mortality to guide maternal health programs that are based on preeclampsia prevention.
This study was designed to determine the validity of the MIA procedure by itself, without using any additional clinical data. The only exception to this purist approach was the knowledge of whether the death of the woman occurred while pregnant or within 42 days of termination of pregnancy (irrespective of the duration and the site of the pregnancy), as this information was essential to know when a case fulfilled the definition of WHO and, consequently, had to be included as a maternal death. Our findings indicate that the MIA without any clinical data has a relatively limited performance for direct obstetric causes of death. Some degree of obstetric information from the clinical record or the verbal autopsy could significantly improve the results of the MIA. Importantly, obstetric hemorrhage was missed in all but 1 case in the MIA; however, this condition is easily identified in the clinical records and is likely not to be difficult to retrieve from a verbal autopsy. On the contrary, most indirect obstetric deaths that are frequently misclassified in verbal autopsies were identified in the MIA. Future research should focus on the performance of the MIA with the inclusion of clinical data.
The extensive microbiological sampling and analysis proposed in our MIA protocol result in the adequate identification of the etiological agent in a high number of infectious diseases. Moreover, in 5 cases considered as infectious by the gold standard, the etiological agent was successfully identified in the MIA but not in the CDA. It is possible that the less invasive sampling procedure performed in the MIA might improve the performance of the microbiological analysis by reducing the probability of microbiological contamination [36,37]. On the other hand, the MIA may result, in the absence of any clinical information, in an overestimation of the microbiological results, as it may be the cases of the 5 women who died of obstetric hemorrhage that were considered as infectious deaths in the MIA.
A limitation of this study is the relatively low sample size, which has resulted in a poor representation of some causes of death. Nevertheless, the study was not designed to describe all the causes of maternal death and validate the method for each specific cause or category of disease but rather to assess the validity of the method for the group of maternal death as a whole. A second limitation of this study is that its diagnostic accuracy of the MIA could have been influenced by the dissemination of many diseases. More than half of the patients were HIV-infected adults with highly disseminated infections, and the performance of the procedure could be significantly reduced in focal lesions and in limited infections in immunocompetent hosts. Finally, both the MIA and the CDA include some degree of subjective interpretation in the attribution of cause of death.
In conclusion, establishing reliably the causes of maternal deaths is crucial for health planning and prioritization, which in turn are essential elements to reduce maternal mortality. The lack of quality data on these causes of death in high-burden countries has been recognized as a major limitation to achieving the 5th Millennium Developmental Goal [38], especially in sub-Saharan Africa, where the MMR only dropped by 26% since 1990 [1]. In these settings, a tool such as the MIA might provide quality information mainly for those conditions that are more difficult to identify through routinely used methods for cause of death determination, while improving clinical management and verbal autopsies with refined algorithms. This would lead to improving the most underachieving global health goal, which relates to maternal health.
|
10.1371/journal.ppat.1006187 | Multiple UBXN family members inhibit retrovirus and lentivirus production and canonical NFκΒ signaling by stabilizing IκBα | UBXN proteins likely participate in the global regulation of protein turnover, and we have shown that UBXN1 interferes with RIG-I-like receptor (RLR) signaling by interacting with MAVS and impeding its downstream effector functions. Here we demonstrate that over-expression of multiple UBXN family members decreased lentivirus and retrovirus production by several orders-of-magnitude in single cycle assays, at the level of long terminal repeat-driven transcription, and three family members, UBXN1, N9, and N11 blocked the canonical NFκB pathway by binding to Cullin1 (Cul1), inhibiting IκBα degradation. Multiple regions of UBXN1, including its UBA domain, were critical for its activity. Elimination of UBXN1 resulted in early murine embryonic lethality. shRNA-mediated knockdown of UBXN1 enhanced human immunodeficiency virus type 1 (HIV) production up to 10-fold in single cycle assays. In primary human fibroblasts, knockdown of UBXN1 caused prolonged degradation of IκBα and enhanced NFκB signaling, which was also observed after CRISPR-mediated knockout of UBXN1 in mouse embryo fibroblasts. Knockout of UBXN1 significantly up- and down-regulated hundreds of genes, notably those of several cell adhesion and immune signaling pathways. Reduction in UBXN1 gene expression in Jurkat T cells latently infected with HIV resulted in enhanced HIV gene expression, consistent with the role of UBXN1 in modulating the NFκB pathway. Based upon co-immunoprecipitation studies with host factors known to bind Cul1, models are presented as to how UBXN1 could be inhibiting Cul1 activity. The ability of UBXN1 and other family members to negatively regulate the NFκB pathway may be important for dampening the host immune response in disease processes and also re-activating quiescent HIV from latent viral reservoirs in chronically infected individuals.
| A family of human genes termed UBXN is thought to control many cellular processes, including protein destruction. While working with these proteins, we noticed several profoundly blocked the production of human immunodeficiency virus (HIV) and other, similar viruses. We delved into the how this occurs, and it appears that at least three of the proteins affect a central pathway in man’s immunologic response to viral and other pathogens, termed NFκB, by a mechanism not previously described. When UBXN1 protein levels were reduced, HIV synthesis was enhanced in certain cells. We also tested what happens when UBXN1 is removed or blocked in human and mouse cells, and saw consistent effects on NFκB. Removing UBXN1 from mouse cells changed the expression hundreds of genes (about 5% of all mouse genes), notably those involved in cell stickiness and also the immune response. Blocking UBXN1 in T cells harboring a silent HIV caused HIV to be resynthesized. We believe that the UBXN gene family members that negatively impact NFκB may be important for dampening immune responses and also re-animating silent HIV, important for curing or eliminating HIV in man.
| The UBX family member proteins are thought to regulate diverse cellular processes, including protein stability and degradation. Members of the gene family include UBXN2a (also termed UBXD4), UBXN2b (p37), UBXN2c (UBXD10, p47, or UBX1), UBXN3a (FAF1 or UBXD12), UBXN3b (FAF2 or UBXD8), UBXN4 (UBXD2 or UBXDC1), UBXN6 (UBXD1 or UBXDC2), UBXN7 (UBXD7), UBXN8 (REP8 or UBXD6), UBXN9 (ASPCR1, RCC17, TUG, or UBXD9), UBXN10 (UBXD3), and UBXN11 (UBXD5, COA-1, or SOC). These proteins may be grouped into five evolutionary conserved families that are represented by UBXN1, UBXN2c, UBXN3a, UBXN6, and UBXN8. Perhaps the best studied family member is UBXN2c, known to play a crucial role in homotypic membrane fusion processes as an adaptor or co-factor for the AAA ATPase p97/valosin-containing protein (VCP) [1–6]. p97 is thought to control multiple aspects of cellular homeostasis, and recently dominant mutations in p97 that cause rare multisystem degenerative diseases with varied phenotypes have been linked to altered UBXN2c co-factor regulation [7]. The UBX domain of p47 interacts directly with p97/VCP [8], imitating ubiquitinated substrates of this chaperone [9].
Other UBXN family members have been implicated as co-factors that cooperate with p97 [10–14]. In C. elegans, UBXN1, UBXN2, and UBXN3 appear to redundantly control spermatogenesis via degradation of TRA-1A [12]. UBXN9, by modulating the activity of the p97-Ufd1-Npl4 complex, has been shown to be critical for the degradation of polyubiquitinated proteins via endoplasmic reticulum-associated protein misfolding pathway [15]. However, for most of the other UBXN family members the physiological or cellular function is poorly if at all understood. Of note, members of the UBXN1, UBXN2c, and UBXN3a families also possess an N-terminal UBA domain, thought to be involved in (but not restricted to) binding ubiquitin monomers and higher order forms [16].
Previously we had reported that UBXN1 negatively regulated RIG-I-like receptor (RLR) signaling by binding to MAVS and sterically blocking the interaction of MAVS with several downstream effectors [17]. The overall outcome of RLR pathway inhibition by UBXN1 was enhanced replication of several RNA viruses, including vesicular stomatitis, Sendai, West Nile, and Dengue. Subsequently, a separate group reported that UBXN1 inhibited TNFα-stimulated NF-κB signaling by cIAP recruitment, independent of VCP/p97 [18]. This action of UBXN1 blocked cIAP1 recruitment to TNFR1, inhibiting RIP1 polyubiquitination in response to TNFα.
NFκB signaling is central to the innate immune response in higher organisms, inducing expression of multiple genes via interferon signaling that inhibit pathogen replication, including human immunodeficiency virus type 1 (HIV) and several other viruses [19–22]. At the same time, HIV is dependent upon NFκB activation in order to promote viral RNA transcription—the viral long terminal repeat or LTR has two NFκB binding sites, and mutations or deletions in these binding sites modulate transcript initiation [23–25]. Stimulation of the canonical NFκB pathway requires phosphorylation and degradation of cytosolic IkBα, the latter by the 26S proteasome, and nuclear translocation of NFκB, comprised of a p50 and p65 heterodimer, to directly activate transcription of responsive genes [26–29]. Depending upon the model in vitro T cell system used, quiescent, genomically integrated HIV, which is transcriptionally silent and a barrier to viral eradication, can be stimulated out of latency by activation of the NFκB pathway [30–32]. Such transcriptional activation, coupled with other treatment modalities and immune recognition of HIV-infected cells, is a possible path towards HIV eradication and cure [33–38].
Over the years a number of viral and cellular negative regulators of NFκB signaling have been identified. One of the best known is A20, which functions as a ubiquitin editing gene, both adding and removing different polyubiquitin chains from NFκB signaling proteins [39–41]. Related to this mechanism of inhibition are the A20-Binding Inhibitors of NFκB (ABINs), which were originally identified as A20-binding proteins, and these are also thought to be involved in the negative feedback regulation of NFκB activation [42–43].
In the cytosol, NFκB heterodimer is held in an inactive state by IκBα, which after cell stimulation is phosphorylated by the activated IKK complex composed of NEMO and IKKα/β. Phosphorylated IκBα is recognized by the Cullin (Cul)1 E3 ubiquitin ligase scaffolding complex [44–46], which includes Skp1, Skp2 F-box (β–TrCP), and Rbx1, and IκBα undergoes polyubiquitination and subsequent 26S proteosomal degradation, allowing NFκB heterodimer nuclear translocation. Cul1 activity is itself regulated by the COP9 signalosome and neddylation [47–48]. Although it has been reported that rotavirus NSP1 induces proteosomal degradation of β–TrCP [49] and ORF2 of hepatitis E virus directly binds β–TrCP and stabilizes IκBα [50], we are unaware of any cellular factors known to obstruct or interfere with Cul1 function. Here we report that UBXN1 and other UBXN family members block the canonical NFκB signaling pathway and inhibit retroviral and lentiviral production via interaction with Cul1.
To further investigate the biology of UBXN1, we constructed an HIV-based, third generation lentiviral vector encoding the full-length UBXN1 cDNA (297 aa) and attempted to produce VSV G-pseudotyped lentiviral vector particles in 293T cells by co-transfection of HIV packaging and VSV G expression plasmids, but resultant HIV vector titers were reduced by two orders-of-magnitude compared to control, empty vector. To verify this reduction in titer was not a result of an intrinsic defect to the vector perhaps due to inhibitory cis-acting sequences present within the UBXN1 open reading frame, we co-transfected UBXN1 expression plasmid along with an HIV transfer vector and VSV G envelope plasmid again into 293T cells and observed >100-fold reduction in resultant titer of the HIV vector on HOS targets, compared in parallel to the use of empty cDNA expression plasmid (Fig 1B). To exclude the possibility that the reduction in titer was a result of less functional VSV G being expressed in the producers, 293T cells transfected with VSV G were acid –shocked at pH 5.2 for two minutes and cell syncytia were enumerated an hour later, and there was no observable difference in the number of multinucleate 293T cells in the presence or absence of UBXN1.
To determine whether this reduction in titer was generalizable to other retroviral vectors, we tested other third generation, replication-defective retroviral vectors, including murine leukemia virus (MLV), feline immunodeficiency virus (FIV), simian immunodeficiency virus (SIV), and equine infectious anemia virus (EIAV), all in the presence or absence of co-transfected UBXN1 expression plasmid (Fig 1F). Each retroviral vector was produced in 293T cells using a three-component plasmid system comprised of VSV G, retroviral packaging vector driven by CMV immediate-early enhancer/promoter (IE), and retroviral transfer vector encoding either eGFP or eYFP. In each case, there was a significant decrease in retroviral vector titer in the presence of co-transfected UBXN1, as assessed by flow cytometry readout on target HOS cells 72 h post-transduction (Fig 1F).
To see whether other UBXN family members had a similar effect, we obtained or cloned ourselves CMV IE-driven, FLAG epitope tagged versions of p47, UBXD4, UBXN3a, UBXD8, UBXN4, UBXN6, UBXN7, UBXN8, UBXN9, UBXN10, and UBXN11 (for UBXN11 both rat and human cDNAs), and verified appropriate expression of each after transient transfection of 293T cells, although there was some variability in expression levels between each family member (S1a Fig). Each of these UBXN constructs was co-transfected with HIV or FIV transfer and packaging vector components as described above, along with VSV G, and reproducibly we observed significant, marked inhibition of lentiviral vector production when UBXN1, UBXD8, UBXN6, UBXN11 and UBXN9 were co-transfected and expressed (S1b and S1c Fig).
In the HIV vector used in the above experiments, the autofluorescent reporter is driven off the viral long terminal repeat (LTR), and reporter gene expression was also significantly reduced in a dose-dependent manner by UBXN1 in 293T producer cells, which paralleled that of an HIV LTR-FFLUC construct (Fig 1D). This marked reduction in LTR activity was also observed for the SIV and FIV LTRs after plasmid transfection of 293T cells (Fig 1E). Because of this result, we tested whether UBXN1 and the four other family members directly affected transcriptional initiation of HIV and other retroviruses. After transient transfection into 293T cells of replication-defective HIV, SIV, and MLV vectors that all have LTRs driving eGFP, all five UBXN family members reduced LTR activity, with UBXD8 having the least amount of inhibitory activity (S1d Fig). Additionally, using firefly luciferase reporter transient transfection assays, full-length UBXN1 and other family members, including the four mentioned above, inhibited NFκB and HIV LTR activity but with rare exception had no or little effect on the AP-1 promoter (Fig 2A and 2B, S1e Fig).
NFκB activity is controlled by nuclear translocation, which can only occur if IkBα is degraded via the 26S proteasome, mediated in part by the E3 ubiquitin ligase Cullin1 (Cul1) [51–52, 45]. Previously, UBXN1 had been shown to interact with several Cullins, including Cul1 (http://www.ebi.ac.uk/intact/pages/interactions/interactions.xhtml?query=Q04323*). When epitope-tagged versions of both UBXN1 and Cul1 were both over-expressed in 293 cells, they interacted with each other by co-immunoprecipitation (IP)-immunoblotting (Fig 3A). This co-IP-IB interaction was also observed between endogenous Cul1 and exogenous, epitope-tagged versions of UBXN1, UBXN9, and UBXN11 expressed in 293 cells (S2g Fig).
Cul1 is considered a scaffolding protein that binds to the adaptor protein Skp1 near its amino terminus and ring box protein Rbx1 near its carboxy terminus; the receptor protein Skp2fbox binds to Skp1 and also to substrate protein. To determine which regions of Cul1 interacted with UBXN1, a series of Flag epitope-tagged derivatives of Cul1 were constructed and expression in 293 cells assessed by immunoblotting (Fig 3B, left and right). All Cul1 proteins expressed at detectable levels. By co-IP and immunoblotting in 293 cells, UBXN1 interacted with both the N and C-terminal thirds of Cul1, but not the middle third of the protein (3B, right). These results suggest that there are multiple points of contact between UBXN1 and Cul1.
We next examined the ability of the endogenous forms of both Cul1 and UBXN1 to interact with each other, before and after TNFα stimulation of primary, non-transformed cells. In the absence of TNFα, the two proteins co-IP in human foreskin fibroblasts (HFFs), but after TNFα stimulation there was visibly less interaction as seen by co-IP, which temporally coincided almost precisely with IkBα degradation (Fig 3C, quantified on the right). Once IkBα levels returned to baseline, so did the observed interaction between Cul1 and UBXN1. Total levels of Cul1 and UBXN1 were unaffected by TNFα treatment of HFF cells, nor were there any obvious changes in the mobility of either protein by SDS-PAGE (Fig 3C).
Degradation of IkBα is central to the activation of NFκB, which is mediated by Cul1 scaffolding complex of Skp1 adapter, Skp2fbox receptor, and Rbx1, which binds to IkBα substrate and results in its ubiquitination and subsequent targeting to the 26S proteasome for degradation. Over-expression of UBXN1 completely blocked IkBα degradation (Fig 3D, quantified on the right), which mirrored the inhibition of NFκB activity (Fig 3E). This was also true for UBXN9 and UBXN11, but not UBXN6 (S2c and S2d Fig, quantified on the right). These results suggest that UBXN1, N9, and N11 but not UBXN6 block IkBα degradation, thus inhibiting NFκB activity.
In order to map the functional domains of UBXN1, previously we had made a series of epitope-tagged UBXN1 deletion constructs (Fig 1A) [17]. These constructs, which consisted of both N and C terminal truncations, and internal deletions of the UBA, UBX, and coiled-coiled domains to allow functional assessment of these three regions of UBXN1, were also tested in the above assays. Of note, reproducibly the shorter constructs were not well-expressed, either because they were unstable or due to poor transfer to the filter membrane (Fig 1C). The ability of UBXN1 to inhibit HIV-based vector production (Fig 1B), NFκB activity (Fig 2A), HIV LTR activity (Fig 2B), degradation of IκBα (S2a and S2b Fig), along with interaction with Cul1 (S2e and S2f Fig) mapped to its UBA domain (amino terminus), with a second important determinant between amino acids 240 and 253. One puzzling result is that the 1–240 aa construct was well-expressed and variably interacted with Cul1 but was less inhibitory in these assays compared to 1–230 aa and the 1–253 aa constructs, suggesting that it was not completely folded correctly or lacked necessary post-translational modifications. Neither the C terminus (amino acids 253–297) nor the coiled-coil domain (amino acids 87 to 172) of UBXN1, the latter of which mediates interaction between UBXN1 monomers [17], was necessary for these activities. A triple alanine (AAA) substitution mutation in the UBA domain thought to bind to ubiquitin multimers (amino acid positions 13–15) markedly reduced the inhibitory activity of UBXN1. This was true for HIV vector production (Fig 1B and S1b Fig), HIV LTR transcriptional activity (Figs 2B and 1D), production of other retroviral vectors with the possible exception of MLV (Fig 1F and S1b Fig), and NFkB transcriptional activity (Fig 2A). This AAA UBXN1 mutant is quite informative since it was very well expressed but clearly did not interact with Cul1 by co-IP (S2e and S2f Fig). This is consistent with the inhibitory activity of UBXN1 being mediated by interaction with Cul1.
In order to examine whether there is a genetic interaction between UBXN1 and Cul1, increasing amounts of UBXN1 and a dominant negative (DN), COOH-truncated form of Cul1 were co-transfected into 293T cells in the presence of an HIV LTR-FFLUC reporter. As expected, in the absence of the DN Cul1 increasing amounts of UBXN1 were inhibitory, in a dose-dependent fashion (S3 Fig). Additionally, increasing amounts of DN Cul1 in the absence of UBXN1 were also inhibitory, presumably because DN Cul1 interferes with the function of endogenous Cul1 in targeting IκBα for proteasomal degradation. Intermediate amounts of DN Cul1, however, were stimulatory in the presence of UBXN1, likely because the exogenous DN Cul1 interacts with UBXN1, blocking its inhibitory function against endogenous Cul1. Moreover, at the highest amounts of both UBXN1 and DN Cul1 there was no further inhibition of HIV LTR activity, suggesting that both are acting similarly, on the same pathway.
Both Skp1 and Rbx1 are known to bind to the N and C termini of Cul1, respectively, in order to target substrate proteins for proteosomal degradation. Because UBXN1 appeared to bind both the N and C termini of Cul1 as well, we investigated whether UBXN1 interacted with Skp1 or Rbx1 and whether UBXN1 interfered with either Skp1 or Rbx1 binding to Cul1. We constructed epitope-tagged versions of both Skp1 and Rbx1 cDNAs driven by the CMV promoter and co-transfected 293 cells with those along with UBXN1 and Cul1 expression plasmids. As anticipated, by co-immunoprecipitation Skp1 bound to full-length Cul1 (Fig 4A). Furthermore, Skp1 bound to Cul1 at the N-terminus, and increasing amounts of UBXN1 bound to Cul1 and yet did not displace Skp1 bound to Cul1 (Fig 4B). In addition, UBXN1 did not interact with Skp1.
As expected, Rbx1 bound Cul1 and also to various forms of UBXN1, but not the coiled-coil domain by itself (Fig 4A). Skp1 did not bind the latter half of Cul1 whereas UBXN1 did (Fig 4B), however the binding of Skp1 to full-length Cul1 was not displaced by increasing amounts of UBXN1 (Fig 4B). Also as expected Rbx1 bound to Cul1 at the C-terminus, and increasing amounts of UBXN1 did not displace Rbx1 bound to Cul1 (Fig 4C). UBXN1, however, did bind to Rbx1 by co-immunoprecipitation (Fig 4C).
We also examined whether there is a genetic interaction between Skp1 and UBXN1. Similar to the experiment with DN Cul1, increasing amounts of UBXN1 and full-length Myc-tagged Skp1 were co-transfected into 293T cells but in the presence of an HIV LTR-FFLUC reporter. As expected, in the absence of Myc-Skp1 increasing amounts of UBXN1 were inhibitory, in a dose-dependent fashion (S4 Fig). Increasing amounts of Myc-Skp1 were also inhibitory, likely because Myc-Skp1 is titrating away Skp2fbox from the Cul1 scaffolding complex. In this case, however, at intermediate levels of Myc-Skp1 and increasing amounts of UBXN1 there was no stimulation in HIV LTR activity. This may be explained by the fact that both UBXN1 and Myc-Skp1 bind to Cul1 (Fig 4A) and at high amounts are inhibitory to Cul1 function. At the highest amounts of combined UBXN1 and Myc-Skp1 there was no further inhibition of HIV LTR activity, suggesting that both are indeed acting on the same pathway.
In order to examine the role of UBXN1 in a more physiological experimental setting, we first investigated the effect of knocking down UBXN1 using both shRNA and siRNA approaches in 293T cells. Validated shRNA against UBXN1 was ligated into an HIV-based vector, and when co-transfected into 293T cells with HIV packaging vector and VSV G expression plasmid the presence of the shRNA resulted in a 3 to 10-fold increase in viral titer of both the vector it was introduced into and also a separate HIV vector (encoding either eGFP or eYFP) used in parallel, compared to empty vector control (Fig 5A). This increase in titer was observed for all three of the separate HIV vectors that were co-transfected. The boost in titer of the anti-UBXN1 shRNA-encoding HIV vector is even more surprising considering that any shRNA-containing vector typically reduces titer by 10 to 30-fold, due to activation of RNA interference in the 293T cells [53]. siRNA and shRNA-encoded lentivirus vector-mediated knockdown of UBXN1 in 293 cells also resulted in significantly enhanced NFκB signaling (Fig 5B). The degree of knockdown using the siRNA against UBXN1 was ~80% whereas for the shRNA it was ~50% (Fig 5C).
We attempted to produce a murine conditional knockout animal, placing the LoxP-selection cassette in one of the small, 5' introns of murine UBXN1. Although we were able to obtain several heterozygous male and female mice, we were unable to obtain homozygous targeted animals, despite multiple matings and pup screenings. We also harvested fetuses from heterozygous matings, prepared mouse embryo fibroblasts (MEFs) at embryonic days 7, 11, 18, but only observed heterozygous (UBXN1 +/-) or wild-type homozygous (UBXN1 +/+) MEFs, as assessed by PCR using primers that spanned an inserted LoxP site (S2 Table). These results suggest that UBXN1 is essential for fetal development and eliminating the gene (or reducing its function) resulted in early embryonic lethality, to the point that homozygous null MEFs are not recoverable, even early on.
Instead, we decided to stably knockout or knockdown UBXN1 in hTERT-immortalized MEFs and HFFs, respectively. To reduce UBXN1 expression in HFFs, we used the shRNA described above, encoded within the HIV-based vector, to produce stable HFF UBXN1 knockdown cells. After verifying ~50% knockdown (Fig 6A, quantified on right), HFFs were stimulated with TNFα. There was a subtle but quantifiable enhanced recovery of IκBα levels in control versus knockdown HFFs, which was evident at both early and late time points (Fig 6A). There were no differences in Cul1 levels between control and knockdown HFFs (Fig 6A). We also assessed NFκB signaling and HIV LTR activity in the UBXN1 KD HFFs, which were both increased, compared to HFFs transduced with empty vector control (Fig 6B). Of note, UBXN1 KD in HFFs or KO in MEFs did not modulate IkBα phosphorylation levels, most evident with the use of the proteosomal inhibitor Bortezomib, which suggests that UBXN1 does not inhibit or modulate IKK activity (Fig 6A and 6C).
To knockout UBXN1 in the immortalized MEFs, we used Cas9/CRISPR and guide (g)RNAs targeting both a 5’ and 3’ exon of the gene simultaneously, along with a gRNA targeting a coding exon of murine HPRT. The latter allowed us to select knockout MEFs using 6-thioguanine (6TG) and enriched for UBXN1 knockout clones several hundred fold (see Methods). Control MEFs had HPRT alone knocked out. Knockout of UBXN1 in MEFs was verified by immunoblotting (Fig 6C), PCR using flanking primers, and confocal immunofluorescence microscopy (Fig 6D). We were unable to obtain KO HFFs using a similar strategy that involved gRNA co-targeting of human UBXN1 along with HPRT. Although small (<100 cells) colonies resistant to 6TG arose in culture we were unable to propagate these HFFs; this may be related to altered cell cycle progression we observed in the KO MEFs (see below).
Knockout (both UBXN1-/- and HPRT-/-) MEFs appeared visibly larger and proliferated more slowly than HPRT -/- (control) MEFs (Fig 6D and S5a and S5b Fig). UBXN1 KO MEFs were uninucleate but the nuclei were larger (Fig 6D). Cell cycle analysis of mid-logarithmic growth UBXN1 KO MEFs indicated most were in G2/M, in sharp contrast to the HPRT KO MEFs, which had a normal cell cycle distribution (S5a and S5b Fig). Despite this, UBXN1 KO cells were able to be passaged repeatedly, without evidence of senescence or progressive changes in morphology. After TNFα stimulation of UBXN1 knockout MEFs, there was delayed recovery of IκBα protein levels, most obvious at the 10–40 min time points (Fig 6C, top left, quantified on right). In the presence of the proteasomal inhibitor Bortezomib after TNFα treatment there was no discernible degradation of IkBα, irrespective of the presence or absence of UBXN1 (Fig 6C, bottom left, quantified on right).
NFκB is well-known to induce IκBα in order to extinguish its own signaling, as an autoregulatory negative feedback mechanism. NFκB had significantly enhanced activity in the UBXN1 KO MEFs, as did the HIV LTR reporter, compared to control cells (Fig 6E). Cul1 levels were unchanged in the UBXN1 knockout MEFs, compared to control MEFs (Fig 6C). To test the reversibility of these effects, UBXN1 was added back to the UBXN1 KO MEFs via lentiviral vector stable transduction. After doing so, there was an increase and stabilization of IκBα protein levels after stimulation of the cells with TNFα (S6a Fig, quantified on right). We also observed significant inhibition of both NFκB and HIV LTR activity after UBXN1 was added back to the UBXN1 KO MEFs (S6b Fig).
In order to quantify the transcriptome after UBXN1 KO, RNA-Seq was performed on both control and UBXN1 KO MEFs. Nearly 800 genes were significantly up- and down-regulated in UBXN1 KO compared to control MEFs (S3 Table). Using KEGG analysis, several gene pathways were significantly over-represented in terms of modulated genes, including focal adhesion, cytokine-cytokine receptor interaction, ECM-receptor interaction, chemokine signaling, and protein digestion and absorption (S4 Table and S7a and S7b Fig). Whether UBXN1 KO directly or indirectly modulated the levels of these genes in MEFs is difficult to ascertain at this time.
We next examined the effects of reducing UBXN1 expression on lentiviral and retroviral replication. Despite higher NFκB activity, compared to control MEFs UBXN1 KO MEFs were significantly less susceptible in single round infectivity assays to VSV G-pseudotyped, replication-defective HIV, SIV, FIV, EIAV, and MLV VSV G-pseudotyped particles (S8a Fig). All of these vectors encoded an autofluorescent gene such that transduction was measured by flow cytometry. This was also true for replication-defective adenoviral vectors (S8b Fig), but the UBXN1 KO MEFs were equally or more susceptible to infection by adeno-associated virus vector (S8c Fig), suggesting that the observed phenotype is not simply due to a general unhealthy or toxic cellular state. Similar results were observed with the KD MEFs (S8d Fig).
To examine the effects of reduced UBXN1 expression in a more physiological system related to HIV replication, we stably knocked down UBXN1 in C8166 T cells using the shRNA-encoding HIV vector described above, with ~50–80% KD confirmed by immunoblotting and RNA Seq, respectively (S9a Fig and S5 Table). The C8166 knockdown T cells had enhanced NFκB signaling and HIV LTR activity compared to control C8166 T cells (S9b Fig). Using replication-defective, single cycle HIV-based vectors pseudotyped with either VSV G or CXCR4-tropic HIV envelopes, we observed that KD C8166 T cells were equally susceptible to HIV infection, compared to control C8166 T cells (S9c Fig). This is likely because multiple transcription factors, not just NFκB, regulate transcription off the HIV LTR.
HIV latency in the memory T cell population is thought to be a major barrier to virus eradication in man, with multiple layers of regulatory control at the level of transcriptional initiation, including reduced activity of NFκB. In order to test the effects of UBXN1 on HIV latency, control or shRNA against UBXN1 was introduced by lentiviral vector transduction into JLAT 10.6 T cells, a Jurkat T cell-based line with an integrated but replication-defective HIV encoding eGFP that is a model of HIV latency. Degree of knockdown of UBXN1 was assessed to be ~80% by immunoblotting (S9e Fig). In control JLAT 10.6 T cells, percentage eGFP+ cells was ~3%, which increased to ~75% after 48 h of TNFα treatment, whereas shRNA UBXN1 JLAT 10.6 T cells at baseline were ~25% eGFP+, which increased to 90% after 48 h of TNFα stimulation (Fig 6F). shRNA UBXN1 JLAT 10.6 T cells did not appear to be more sensitive to TNFα treatment (S9f Fig), nor did they apoptose after TNFα stimulation. These results suggest that reduction in UBXN1 levels resulted in activation of NFκB, leading to enhanced HIV LTR-based transcription in this JLAT Jurkat cell line.
Although UBXN1 was previously thought to be involved in protein degradation and turnover [54–57], we have now shown that UBXN1 inhibits two key antiviral pathways, both RLR and NFκB signaling, by binding to cellular components and interfering with downstream effector functions. While other cellular, viral, and bacterial gene products are known to inhibit one or the other pathway, to our knowledge UBXN1 is unique in this regard and appears to tonically inhibit NFκB signaling by somehow interfering with Cul1 function in unstimulated cells.
Of interest, several other UBXN family members, notably N3b (D8), N6, N9, and N11, consistently and quite markedly inhibit the production of replication-defective HIV and other retroviruses from 293T cells. Although all four of those family members and UBXN1 have inhibitory activity directed against both NFκB and the HIV LTR, only UBXN1, UBXN9 and UBXN11 stabilized IkBα after TNFα stimulation. Those three family members also interacted with Cul1, at least by co-immunoprecipitation. That in UBXN1 KO MEFs after TNFα stimulation there is enhanced degradation of IkBα but levels return to baseline suggests that there is some redundancy in the system. The fact that the four other UBXN family members were expressed in MEFs, with RPKM values ranging from ~2 to 30 (lowest for UBXN11), supports this contention. Of note, UBXN1 RPKM values in MEFs were much higher, at approximately 100. Precisely how UBXN family members N6 and N3b (D8) act to inhibit NFκB signaling and retrovirus production and whether different UBX gene family members can compensate for each other remains to be explored.
At this juncture we cannot exclude the possibility that UBXN1 has other, yet unknown inhibitory activities. For example, UBXN1 and the four other UBXN family members inhibited MLV LTR activity, and yet the MLV LTR does not have any NFκB sites. The mechanism of NFκB inhibition we describe here regarding IkBα stabilization is completely unrelated to and distinct from that recently reported by an independent group of investigators [18]. It is not entirely clear how UBXN1 becomes inactivated after TNFα cellular stimulation and then is no longer able to interact with Cul1 (or interacts only weakly with Cul1). Precisely how UBXN1 inhibits Cul1 function is not known. Based upon the fact that UBXN1 independently interacts with both the N and C-termini of Cul1, does not interact with Skp1 but does with Rbx1, and does not inhibit the binding of Skp1 or Rbx1 to Cul1, we present schematic models of how UBXN1 may be inhibiting the canonical NFκB pathway (Fig 7A). In one model, UBXN1 binds to Cul1 and multimerizes to somehow inhibit Cul1 function without much allosteric effect (Fig 7B); in the other, the presence of UBXN1 results in a conformational change in Cul1 between the N and C terminal domains, which are thought to be connected by a flexible linker region (Fig 7C). In both of these models Skp1 binds to Cul1 independently of UBXN1 whereas UBXN1 interacts with Rbx1 and Rbx1 also binds to Cul1. Whether UBXN1 interferes with the binding of other proteins or enzymes (i.e., Skp2 or E2 ubiquitin conjugating enzyme) to Skp1, Rbx1, or Cul1 remains to be established. Interaction between UBXN1 and Cul1 may be indirect, mediated by another protein(s) (e.g., ubiquitin multimers) or by a small molecule; experiments to examine direct binding of purified UBXN1 to Cul1 and Rbx1 are in progress. Whether UBXN1 also disrupts the function of other Cul family members is also under investigation. It is important to note that have we not yet observed any post-translational modifications of UBXN1 after cell stimulation that could potentially regulate its activity.
UBXN1 is not a viral restriction factor since it interacts with cellular proteins and not any viral gene products and in fact can enhance RNA virus replication [17]. Here it inhibits HIV and other lentivirus production in single cycle assays likely because virus production in 293T cells is very dependent upon NFκB binding to the viral LTR enhancer as part of the enhanceosome and stimulating viral mRNA biosynthesis. This underscores the delicate and intricate balance the NFkB pathway plays during HIV infection since it is absolutely required for high level virus production and yet the downstream effectors or interferon stimulated genes can be detrimental to virus replication. Thus, although much more HIV can be produced from activated CD4+ T cells, at later stages those same cells can block viral replication. With regards to quiescent memory T cells, thought to be the major reservoir of latent HIV and the chief barrier to cure [58–61], the data presented here knocking down UBXN1 gene expression in JLAT cells suggest that interfering with the function of UBXN1 and perhaps other family members may facilitate the reactivation of HIV gene expression by enhancing NFκB activity, as part of the ‘shock and kill’ strategy of eliminating latently HIV-infected T cells.
That UBXN1 blocks both the RLR and NFκB pathways suggests that it is involved in innate immunity. Thus far, examination of several thousand whole exome sequences from both the Yale and NHLBI collections have revealed no homozygous single nucleotide variants or indels within the UBXN1 coding sequence, consistent with its essential function in mouse embryogenesis. Perusal of the ExAC dbase demonstrates only a handful of homozygous missense mutations within the coding sequence of UBXN1 in man (http://exac.broadinstitute.org/gene/ENSG00000162191); several of these are conservative substitutions. It will be of interest to test the functionality of the other missense mutations and correlate with disease processes. It is possible that enhancement of UBXN1 function could ameliorate inflammatory disorders, certain forms of autoimmunity, and malignancy in which either the RLR or NFκB pathway has been activated [62]. Additionally, other UBXN family members could play heretofore unappreciated roles in innate and adaptive immunity.
pNFκB-Luc, pRL-TK, and AP1-Luc reporters, and all UBXN1 expression plasmids were as previously described [17]. HIV LTR-Luc and Tat expression plasmids were kind gifts of Dr. Andrew Rice (Baylor College of Medicine). pcDNA3-MYC3-Cul1 and pCDN3-DN-hCul1-Flag were obtained from Addgene; to prepare a FLAG-tagged version the Cul1 insert was excised using flanking BamH1 sites, Klenowed, and inserted into FLAG-pcDNA3.1/Zeo (Invitrogen) at the EcoR5 site, and expression confirmed by transfection and immunoblotting using anti-FLAG antibody as described above. Cul1 truncation mutants were made by PCR amplification using various combinations of primer pairs (S1 Table) and PrimeStar HS DNA Polymerase (Takara), TOPO-cloned into pCR-Blunt II-TOPO vector (Invitrogen), and the insert sequenced by the dideoxy method. pSil-mHPRT-Tomato was a gift of Richard Flavell (Yale), and is based upon pSilencer-Tomato, with the gRNA sequence 5’-GATCCGAAAAAGTGTTTATTCCTCAGTTTAAGAGCTATGCTGGAAACAGCATAGCAAGTTTAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTA-3’ inserted between the BamH1 and HindIII sites (underlined sequence targets second exon of murine HPRT). pSil-mUBXN1-Tomato-1 and pSil-mUBXN1-Tomato-2 were similarly constructed, with insert sequences being 5’-GATCCGCATCGAGGCTGCGATGGATGTTTAAGAGCTATGCTGGAAACAGCATAGCAAGTTTAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTA-3’ and 5’-GATCCGCCTTCTGCTGTCCTCATTGGTTTAAGAGCTATGCTGGAAACAGCATAGCAAGTTTAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTA-3’ (underlined sequences target third and last exon of murine UBXN1, respectively). HIV-CIY has been described [63]; NL4-3-Renilla is based upon HIV provirus pNL4-3, with a frameshift in Vpr at the unique EcoR1 site and the 1.0 kb coding sequence of Renilla luciferase inserted at the unique Xho1 site in the same 5’-3’ orientation as the other viral genes. FG12-SV40 is based upon the third generation HIV vector FG12 with the UbiC promoter replaced by a 0.35 kb SV40 origin-promoter. Both full-length Skp1 and Rbx1 were PCR-amplified from a HeLa cDNA library and cloned into three different expression plasmids such that the epitope tags (HA, FLAG, and Myc) were at the amino terminus; DNA sequence and protein expression of all six constructs were respectively confirmed by Sanger sequencing and immunoblotting, the latter after transient transfection of 293T cells.
Adherent or suspension cells were cultured in DMEM or RPMI (ThermoFisher Scientific), respectively, supplemented with 10–15% fetal calf serum (FCS, Life Technologies), ultraglutamine as needed, and antibiotics. Mouse embryo and human foreskin fibroblasts were immortalized using an HIV-based vector encoding hTERT, the catalytic subunit of human telomerase, or SV40 virus large T antigen, coupled by an internal ribosome entry site to bsd, and maintained in DMEM supplemented with 10 μg/ml blasticidin (Invivogen).
For cell cycle analysis, mid-logarithmic cells were lifted in PBS plus 2 mM EDTA, pelleted, resuspended in complete medium supplemented with 10 μg/mL Hoechst 33342, and incubated for 45 min at 37°C prior to data acquisition on an LSRII flow cytometer (B-D). Cell cycle analysis was performed using FlowJo software. Because MTT assay results were misleading due to the increased size of the UBXN1 KO cells, for proliferation studies cells were plated at low density in replicates in 6-well format and enumerated manually every 48 h after trypsinization, using a hemocytometer and trypan blue exclusion dye staining, passaging as necessary.
Rabbit anti-UBXN1 (HPA012669), mouse anti-tubulin (T5168), and mouse anti-flag (F1804) antibodies were from Sigma-Aldrich; mouse anti-Cullin1 (32–2400) was from Invitrogen; mouse anti-Cullin1 (sc-17775) was from Santa Cruz Biotechnology. Rabbit anti-Cul1 monoclonal antibody was from Novus Biologicals (NBP1-40523). Rabbit anti-Myc (71D10), mouse anti-IκBα (L35A5), mouse anti-NFκB p65 (L8F6), rabbit anti-COX IV (3E11), rabbit anti-β-tubulin (9F3), rabbit anti-phospho-IκBα (Ser32) were all from Cell Signaling Technology. Secondary antibodies used from Cell Signaling Technology were anti-mouse IgG-HRP (#7076) and anti-rabbit IgG-HRP (#7074). Rabbit anti-p65 RelA (10745-1-AP) were from Proteintech; Phosphatase Inhibitor Cocktail (5870) and human Tumor Necrosis Factor-α (8902) were from Cell Signaling Technology; Protein A/G Agarose (20423) was from Thermo Scientific, Protease Inhibitor Cocktail Tablets (Complete EDTA-free) were from Roche Life Sciences. Alexa Fluor 488 Goat anti-Rabbit IgG (H+L) Secondary Antibody (Cat #: A-11008), Alexa Fluor 546 Goat anti-Mouse IgG (H+L) Secondary Antibody (Cat #: A-11003), TO-PRO-3 Iodide (642/661) (Cat #: T3605), and ProLong Gold Antifade Mountant with DAPI (Cat #: P-36931) were from Life Technologies. For enhanced chemiluminescence (ECL) HyGLO Quick Spray Chemiluminescent HRP Antibody Detection Reagent (Cat #: E2400) and HyBlot CL Autoradiography Film (Cat #: E3012; both from Denville Scientific Inc) were used.
Approximately 106 HEK293 cells (from the American Type Culture Collection, Manassas, VA) were reverse transfected in 6 well plates with various expression plasmids using Lipofectamine 2000 (Life Technologies). Whole-cell extracts were prepared from transfected cells using lysis buffer (50mM Tris-HCl pH 7.4, 150mM NaCl, 2mM EDTA, 1% Triton X-100, and 0.1% SDS) and incubated overnight at 4°C with 1:100 dilution of mouse monoclonal anti-FLAG antibody, then bound to protein A/G PLUS-Agarose beads for 2 hr at 4°C. Beads were washed three times and proteins eluted by boiling for 10 min in SDS sample lysis buffer, electrophoresed on pre-made SDS-PAGE gradient gels, transferred to nitrocellulose membranes (Bio-Rad), and probed with anti-Cul1 rabbit monoclonal antibody (Novus Biologicals) as primary and anti-rabbit IgG-HRP as secondary to detect Cul1-UBXN1 interactions by ECL and autoradiography. For input protein expression, 10% of whole cell extracts were gel electrophoresed in parallel and probed with anti-FLAG, anti-Cul1, and anti-β-Tubulin antibodies.
For co-immunoprecipitation of endogenous proteins, ~2 × 106 of immortalized human foreskin fibroblasts were stimulated with TNFα at a concentration of 10 ng/mL. Whole cell extracts (10 μg) were prepared at different time points using lysis buffer as described above, incubated overnight at 4°C with 1:100 dilution of rabbit anti-UBXN1 antibody (Sigma-Aldrich), bound to protein A/G Agarose beads, and further processed as described above.
For confocal immunofluorescence microscopy of UBXN1 and Cul1 in MEFs, cells were fixed in one of 8-chamber culture slides (Falcon, #354108) with 4% formaldehyde-PBS for 15 min at room temperature, rinsed in PBS, then immunostained according to Immunofluorescence General Protocol (http://www.cellsignal.com/contents/resources-protocols/immunofluorescence-general-protocol/if). Primary rabbit anti-UBXN1 (Sigma-Aldrich) was diluted by 1:400, mouse anti-Cul1 (Santa Cruz) was diluted by 1:100. UBXN1 and Cul1 were stained using 1:1000 diluted secondary antibodies conjugated to Alexa Fluor 546 and 488 (Life Technologies), respectively, and nuclear DNA stained using TO-PRO-3. Images were acquired using a Zeiss LSM 510 Meta (objective 633).
Four pmol of siRNA directed against human UBXN1 (#SASI_Hs01_00134629, Sigma-Aldrich) or Trilencer-27 Universal scrambled negative control siRNA (SR30004, Origene) were reverse-transfected into HEK 293 cells using Lipofectamine 2000 in 12-well plates, together with luciferase reporter plasmids (80 ng NFkB-luc per well, for example, and 16 ng pRL-TK per well). Ninety-six h post transfection, cells were washed with PBS and lysed for Dual Glow Luciferase assay, following the manufacturer’s instructions (Promega). FFLUC RLU was normalized to that of Renilla.
In order to generate stable UBXN1 knockdown cell lines, UBXN1-specific shRNA duplex oligonucleotides were designed targeting the same sequence as the siRNA above (#SASI_Hs01_00134629, Sigma-Aldrich), and inserted into pLKO.1 cloning vector between Age1 and EcoR1 sites, and DNA sequence confirmed.
For shRNA vector production, 293T cells were transfected using the calcium phosphate method with 10 μg shRNA plasmid (or pLKO.1 empty cloning vector), 10 μg HIV-PV, and 10 μg VSV-G expression plasmid. Replication-defective particles were harvested 72 h later, filtered, and then used for transducing various cell lines. Stable knockdown cells were maintained in 10 μg/ml puromycin (Sigma-Aldrich), with other supplements as needed.
For knocking out UBXN1 in murine cells, 5 x 106 immortalized MEFs were nucleofected using MEF reagent (Lonza) and A-024 Amaxa program with 5 μg of codon-optimized Cas9-eGFP (gift of Dr. Richard Flavell of Yale), 5 μg of pSil-mHPRT-Tomato, 5 μg of pSil- mUBXN1-Tomato-1 and 5 μg of pSil-gadRNA-mUBXN1-Tomato-2 or just 5 μg of pSil-mHPRT-Tomato. After 96 hr, cells were selected in 100 μM 6-thioguanine (Sigma-Aldrich). Surviving cells were diluted and seeded into 48-well plates to select for mouse knockout cell clones, expanded, and screened via immunoblot using anti-UBXN1 antibody. Genomic DNA was extracted from candidate knockout cell clones using DNeasy (Qiagen), and an ~4.0 kb region of murine UBXN1 PCR amplified using primers 5'-TGGAGAGCCTCATCGAGATGGGCTTT-3' and 5'-TGCCCTTCTCAGAAAGGCAG TTCTGG-3' with PrimeStar HS DNA Polymerase (Takara), TOPO-cloned into pCR-Blunt II-TOPO vector (Invitrogen), and both ends of the insert sequenced by the dideoxy method to confirm deletion or rearrangement of UBXN1.
VSV G-pseudotyped, replication-defective HIV particles were produced from 293T cells using calcium-phosphate transfection protocol as described [65], as were FIV-eGFP(VSV G) particles. In addition to VSV G expression plasmid, for production of replication-defective, single-cycle EIAV packaging and transfer vector plasmids were pCEV53B and SIN6.1CeGFPW, respectively (kind gifts of Dr. John Olsen, UNC-Chapel Hill), for MLV pHIT60 and pBabe-IRES-eYFP, for SIV pSIV-PV and pSIV-NIG (gifts of Dr. Hung Fan, UC Irvine).
For deep sequencing of mRNA, total RNA from MEF UBXN1 knockout clones was extracted using RNeasy Mini Kit (Qiagen) according to the manufacture’s instructions. RNA-Seq was performed by Yale’s Stem Cell Center’s Genome core by preparing a cDNA library depleted of rRNA, using Illumina HiSeq2000 platform (50 nucleotide paired-end reads). Reads were mapped to the hg19 human reference genome using TopHat2 aligner [64], and results analyzed using CuffDiff pipeline to identify differentially expressed genes [65].
List of genes that were determined to be significantly differentially expressed by RNA Seq analysis between total RNA samples was uploaded to the WebGestalt web server {http://www.webgestalt.org} and analyzed for enriched KEGG pathways. Statistical significance was calculated by WebGestalt. Pathway graphics were generated using the KEGG website and performing pathway enrichment analysis.
|
10.1371/journal.pgen.1001220 | Systematic Dissection and Trajectory-Scanning Mutagenesis of the Molecular Interface That Ensures Specificity of Two-Component Signaling Pathways | Two-component signal transduction systems enable bacteria to sense and respond to a wide range of environmental stimuli. Sensor histidine kinases transmit signals to their cognate response regulators via phosphorylation. The faithful transmission of information through two-component pathways and the avoidance of unwanted cross-talk require exquisite specificity of histidine kinase-response regulator interactions to ensure that cells mount the appropriate response to external signals. To identify putative specificity-determining residues, we have analyzed amino acid coevolution in two-component proteins and identified a set of residues that can be used to rationally rewire a model signaling pathway, EnvZ-OmpR. To explore how a relatively small set of residues can dictate partner selectivity, we combined alanine-scanning mutagenesis with an approach we call trajectory-scanning mutagenesis, in which all mutational intermediates between the specificity residues of EnvZ and another kinase, RstB, were systematically examined for phosphotransfer specificity. The same approach was used for the response regulators OmpR and RstA. Collectively, the results begin to reveal the molecular mechanism by which a small set of amino acids enables an individual kinase to discriminate amongst a large set of highly-related response regulators and vice versa. Our results also suggest that the mutational trajectories taken by two-component signaling proteins following gene or pathway duplication may be constrained and subject to differential selective pressures. Only some trajectories allow both the maintenance of phosphotransfer and the avoidance of unwanted cross-talk.
| Maintaining the specificity of signal transduction pathways is critical to the ability of cells to process information, make decisions, and regulate their behavior. Preventing cross-talk often relies predominantly on molecular recognition and a set of specificity-determining residues in cognate proteins. Identifying these residues and understanding how they dictate specificity is still a major challenge. Additionally, we have a rudimentary understanding of how specificity evolves, particularly after gene duplication events. We tackled these questions using two-component signaling proteins, the largest family of bacterial signaling proteins. Using analyses of amino acid coevolution, we pinpointed a set of specificity residues in histidine kinases and their cognate substrates. Then, using systematic mutagenesis, we characterized the complete set of intermediates between two different signaling systems, EnvZ/OmpR and RstA/RstB. The results demonstrate that specificity residues contribute unequally and, importantly, that some residues depend substantially on the identity of neighboring residues. We also demonstrate how the specificity of EnvZ/OmpR can be reprogrammed to match that of RstB/RstA through a series of individual substitutions without disrupting the kinase/regulator interaction. Notably, this property is not shared by all trajectories from EnvZ/OmpR to RstA/RstB, suggesting that the duplication/divergence process that likely produced these two pathways may have been fundamentally constrained.
| Protein-protein interactions are crucial to virtually every cellular process. Within the crowded confines of the cell, proteins must distinguish between their cognate partners and non-cognate partners, in order to avoid unproductive and potentially deleterious interactions. The problem of interaction specificity is particularly acute for paralogous protein families where proteins with diverse cellular functions share significant structural and sequence similarity. Cells have evolved many mechanisms to cope with potential cross-talk and to ensure the specificity of protein-protein interactions [1]–[2]. In multicellular organisms, spatial mechanisms that prevent related, but distinct, proteins from coming in contact with one another are often used to create specificity. For example, scaffold proteins, the localization of proteins to different subcellular compartments, and tissue-specific expression can all insulate distinct pathways. Temporal mechanisms, such as the differential timing of expression, are also used to insulate pathways. Although cells employ each of these strategies, in many cases the primary means of preventing unwanted interactions is molecular recognition. However, our understanding of precisely how proteins discriminate between cognate and non-cognate partners at the molecular level is surprisingly rudimentary. Identifying the amino acids responsible, elucidating the precise roles played by each residue, and understanding their complex interdependencies remain major challenges for most protein-protein interactions.
Two component signal transduction pathways provide a tractable system for addressing these questions. These signaling pathways, which are the dominant form of signaling in bacteria, typically consist of a sensor histidine kinase (HK) and a cognate response regulator (RR) [3]. Upon activation of the pathway, a histidine kinase dimer will autophosphorylate on a conserved histidine that then serves as the phosphodonor for a cognate response regulator. Phosphorylation of the response regulator typically activates an output domain which can effect changes in cellular physiology, often by modulating gene expression [4]. Many histidine kinases are bifunctional and when not active for autophosphorylation, will drive the dephosphorylation of their cognate response regulators.
Two-component signaling systems are used for sensing and adapting to a wide range of environmental and intracellular stimuli [3] and most bacterial species encode dozens, if not hundreds of kinase-regulator pairs. Most histidine kinases have only one or two cognate response regulators, and there is minimal cross-talk between different pathways at the level of phosphotransfer [5], [6]. The specificity of phosphotransfer is dictated, on a system-wide level, at the level of molecular recognition [6]. That is, histidine kinases exhibit a large kinetic preference in vitro for their in vivo cognate regulator(s) relative to all other response regulators [6]–[8]. Hence, cellular context is not essential and the basis of in vivo phosphotransfer specificity can be dissected in vitro.
To identify the amino acids that govern the specificity of phosphotransfer in two-component pathways, several groups have examined patterns of amino acid coevolution in cognate pairs of histidine kinases and response regulators [9]–[12]. The rationale behind this approach is that if a residue critical to molecular recognition mutates, it must either revert or be compensated for by a mutation in the cognate protein. Many of the residues identified in these computational approaches are at the molecular interface formed in a co-crystal structure of a histidine kinase-response regulator complex [13]. However, residues in direct contact do not necessarily dictate specificity [9] and computational approaches alone cannot reveal how a histidine kinase discriminates between cognate and non-cognate substrates.
Using the E. coli histidine kinase EnvZ as a model, we mapped a subset of coevolving residues that are critical to the specificity of phosphotransfer [9]. Mutating as few as three residues within the DHp (Dimerization and Histidine phosphotransfer) domain of EnvZ was sufficient to reprogram its phosphotransfer specificity from OmpR to the non-cognate substrate RstA. Although a set of residues that could switch the phosphotransfer specificity of EnvZ was identified, several fundamental questions remain unanswered. Can phosphotransfer specificity also be rewired by making mutations in a response regulator? Do individual specificity residues function as positive elements to promote cognate interactions, as negative elements to prevent non-cognate interactions, or both? Do individual residues contribute equally and independently or are there “hot spots” and dependencies at the amino acid level?
Here, we couple analysis of amino acid coevolution with alanine-scanning mutagenesis and an approach we call trajectory-scanning mutagenesis to systematically dissect the basis of phosphotransfer specificity in two-component signaling pathways. The results provide new insights into how histidine kinases use a set of amino acids to “choose” their cognate substrates, and vice versa. The results have important implications for understanding the evolution of two-component signaling pathways and the mechanisms that cells can use to insulate pathways following gene duplication.
To identify the amino acids responsible for determining the specificity of phosphotransfer in two-component signaling pathways, we searched for residues that covary in cognate HK-RR pairs. Histidine kinases and response regulators that are encoded in the same operon typically form exclusive one-to-one pairings, exhibiting a highly specific interaction both in vivo and in vitro. We identified ∼4500 operonic pairs of histidine kinases and response regulators from a phylogenetically diverse set of 400 sequenced bacterial genomes. To identify coevolving residues, we concatenated cognate HK-RR pairs, performed a large multiple sequence alignment, and then measured mutual information between columns of the sequence alignment. We noted that some columns tended to have high mutual information scores with many other columns in the alignment, an observation also made in other analyses of mutual information [14]. For example, positions 8 and 270 have relatively broad score distributions with long tails, while positions 18 and 202 have narrower distributions centered closer to the origin (Figure S1A and S1B). Consequently, the pairs 8–270 and 18–202, which possess identical mutual information scores of 0.35, cannot be treated identically. We used a relatively simple correction in which raw MI scores were normalized by each column's average raw MI score with all 310 positions in the sequence alignment (Figure S1C).
At an adjusted score threshold of 3.5, we found 12 coevolving pairs, comprising 9 residues in the histidine kinases and 7 in the response regulators (Figure 1A–1C). These residues form a single, densely-interconnected cluster of coevolving residues. The residues are all solvent-exposed in the individual molecules, but buried within the molecular interface formed in a co-crystal structure of T. maritima HK853 and RR468 (Figure 1D) [13]. The residues identified here overlap substantially with, but are not identical to, those we identified previously [9]. Of the coevolving residues in the kinase, all are in the DHp domain, consistent with this domain being the primary site of interaction with the response regulator. Within the DHp domain, the coevolving residues are found on both alpha helices and are located below the histidine phosphorylation site (Figure 1D). The covarying residues in the response regulator are spatially near the conserved aspartic acid phosphorylation site (Figure 1D), predominantly on a single face of alpha helix-1 in the receiver domain with one additional residue within the β5-α5 loop. At lower score thresholds, an additional cluster of coevolving residues are found (Figure S2), but we focus here on the set of 16 residues identified at a threshold of 3.5.
Our previous studies demonstrated that many of the coevolving residues in the kinase (Figure 1) are critical to the phosphotransfer specificity of EnvZ and when mutated can reprogram its substrate selectivity [9]. To test whether we could also rewire the specificity of a response regulator, we again coupled our analyses of coevolution with site-directed mutagenesis. We aimed to mutate the response regulator OmpR such that it was no longer phosphorylated by its cognate kinase EnvZ and instead was phosphorylated by the non-cognate kinase CpxA or RstB. Each kinase was autophosphorylated, purified away from unincorporated nucleotide, and tested for phosphotransfer. In our reaction conditions at a 1 minute time point, EnvZ phosphotransfers exclusively to OmpR, whereas CpxA and RstB phosphotransfer exclusively to CpxR and RstA, respectively (Figure 2).
We first substituted residues in OmpR at the positions within alpha helix-1 identified by mutual information analysis with the corresponding residues from CpxR and RstA to create OmpR(MI-CpxR) and OmpR(MI-RstA); in each case three amino acid substitutions were made in OmpR. The mutant OmpR(MI-RstA) was not phosphorylated to a significant extent by RstB and was still a robust target of EnvZ (Figure 2A). The mutant OmpR(MI-CpxR) showed diminished phosphotransfer from EnvZ and was now phosphorylated by CpxA, although less efficiently than wild type CpxR (Figure 2B). The residues in alpha helix-1 are thus important for phosphotransfer specificity, but other residues must contribute.
We hypothesized that residues within the β5-α5 loop may also affect specificity of the regulator. One of these residues covaried strongly with residues in the histidine kinase (Figure 1) and other loop residues covaried at a slightly lower score threshold of 2.8. We thus swapped the residues in the OmpR loop with those from CpxR and RstA to create OmpR(MI+loop-RstA) and OmpR(MI+loop-CpxR), respectively, and examined phosphotransfer to each of these constructs; the former required three amino acid substitutions and the latter just one. Both constructs exhibited a nearly complete switch in phosphotransfer specificity. EnvZ was unable to phosphotransfer to either OmpR(MI+loop-RstA) or OmpR(MI+loop-CpxR), whereas phosphotransfer from RstB or CpxA to the respective rewired OmpR mutants was efficient and at near wild-type rates (Figure 2). Thus, the top coevolving residues appear sufficient, when mutated along with the β5-α5 loop, to rewire the phosphotransfer specificity of OmpR.
We note that the residues mutated to change the specificity of OmpR constitute a subset of the molecular interface formed by a cognate kinase and regulator (Figure 1D). For instance, the residues in the β4-α4 loop of the response regulator contact the histidine kinase, are in close proximity to the top coevolving residues, and coevolve with sites in the kinase at lower score thresholds (Figure S2), but mutating them was not required to change phosphotransfer specificity (Figure 2). We conclude that the strongest coevolving residues are necessary and sufficient to change the phosphotransfer partnering specificity of OmpR. Other residues may fine-tune the interaction, but do not make major contributions.
Our results indicate that kinase-substrate interaction specificity in two-component pathways is determined by a relatively small set of residues. But does each residue contribute equally to specificity or are there “hotspots” that contribute disproportionately? Do individual residues help bind the cognate substrate or help prevent interaction with non-cognate substrates? To address these questions, we performed alanine-scanning mutagenesis on the DHp domain of EnvZ. Surprisingly, despite being one of the best-characterized histidine kinases, EnvZ has never been explored through alanine-scanning mutagenesis. One study described a series of cysteine mutants [15], but the set of residues examined was limited and the interpretation of cysteine mutations can be ambiguous. We created a series of 33 EnvZ mutants to probe the role of most of the solvent-exposed residues in the DHp domain, generating alanine mutations for all residues except for A255, which was substituted with a threonine (Figure 3A).
We first examined the autophosphorylation activity of each EnvZ mutant (Figure 3B, Figure S3A). As expected, mutating the conserved phosphorylation site H243 (data not shown), or the highly conserved aspartate that follows, D244, completely abolished autophosphorylation. Other residues strongly affecting autophosphorylation flank H243, including L236, G240, R246, T247, P248, L249, R251, and I252. Many of these residues are highly conserved among all histidine kinases suggesting they are critical for catalyzing phosphoryl transfer from ATP to histidine. Alternatively, they may impact folding or stability of the kinase; however, these residues are mostly solvent-exposed and none of the mutants significantly affected purification of soluble protein (data not shown). Of the top coevolving residues (Figure 1), only R251A showed substantially lower autophosphorylation than wild type, suggesting that residues required for docking to a response regulator are distinct from those required for docking to the kinase's CA (catalytic ATP-binding) domain.
For each EnvZ mutant that was able to autophosphorylate to reasonably high levels after an extended incubation, we tested phosphotransfer to OmpR, CpxR, and RstA (Figure 3C–3E, Figure S3B). For an assessment of significance, see Figure S3C and Materials and Methods. For wild-type EnvZ, phosphotransfer to OmpR manifests as a decrease in the EnvZ∼P band and a weak or absent OmpR∼P band, resulting from high rates of phosphotransfer and subsequent dephosphorylation of OmpR∼P by EnvZ. Several alanine mutants did not show the same decrease in EnvZ∼P as the wild-type protein. However, for most of these mutants, such as R246A, T247A, and P248A, a more intense OmpR∼P band was also seen, suggesting that phosphotransfer had occurred but that the mutant could no longer dephosphorylate OmpR∼P. We confirmed the loss of phosphatase activity by measuring the dephosphorylation of purified OmpR∼P by each EnvZ mutant (Figure 3D, Figure S4). Only one mutant, I252A, showed a significant defect in phosphotransfer with no effect on phosphatase activity. Strikingly, mutating most of the coevolving specificity residues, including T250, R251, A255, E257, M258, S269, K272, and D273 had no major effect on phosphotransfer to OmpR. This finding suggests that there is no single “hot spot” and, instead, that specificity and molecular recognition are distributed over a number of residues. There may also be non-additive or synergistic effects between residues such that single point mutations do not significantly affect phosphotransfer in isolation, a possibility probed in more detail below.
Finally, we examined the EnvZ alanine mutants for phosphotransfer to the non-cognate regulators RstA and CpxR (Figure 3D, Figure S3B). For these reactions, in contrast to those shown in Figure 2, EnvZ constructs were autophosphorylated and tested for phosphotransfer without purifying them away from ATP. Under these conditions, EnvZ phosphotransfers weakly to RstA, permitting us to assess whether the alanine mutations affected this non-cognate interaction. Most mutants phosphorylated RstA at a level equivalent to or less than the wild type EnvZ. However, four mutants, P248A, A255T, E257A, and D273A, each showed increases in RstA phosphorylation; E257A also showed detectable phosphorylation of CpxR. Notably, three of the four residues were identified as specificity residues (Figure 1) in our coevolution analysis. The increase in cross-talk seen with these mutants suggests that these residues function, at least in part, as negative elements that prevent phosphotransfer to non-cognate substrates without significantly affecting transfer to the cognate substrate.
Although alanine-scanning provides some insight into specificity, an alanine substitution does not necessarily result in a simple loss of functionality, especially considering that EnvZ has a specificity residue that is already an alanine. In addition, as noted, there may be non-additive interdependencies between residues such that individual substitutions have minimal effect. We therefore sought to characterize the role of specificity-determining residues by examining the complete set of mutational intermediates between two histidine kinases with different specificities. For this analysis we focused on the paralogous systems EnvZ/OmpR and RstB/RstA, and term the approach trajectory-scanning. We constructed each possible specificity intermediate between EnvZ and RstB. This was feasible as the conversion of EnvZ phosphotransfer specificity to match that of RstB required only three substitutions, T250V, L254Y, and A255R [9]; the other major specificity residues identified by coevolution analysis are identical between EnvZ and RstB. In addition, we were able to rewire the specificity of RstB to match that of EnvZ by mutating the same three sites (Figure 4). The triple mutant RstB(V228T, Y232L, and R233A) no longer phosphorylated RstA and, instead, efficiently phosphorylated OmpR. These three residues thus play the dominant roles in dictating the specificity of both EnvZ and RstB. Other residues may make minor contributions.
We constructed each possible single and double mutant intermediate between EnvZ and RstB, in the context of each protein for a total of 12 mutants. To simplify nomenclature we have named mutants based on the protein mutated and the identity of the three specificity residues being considered. For example, wild-type EnvZ is EnvZ(TLA) and the single point mutant EnvZ(T250V) is EnvZ(VLA). Each mutant was tested for phosphotransfer to the regulators OmpR, RstA, and CpxR (Figure 4). Under the conditions used, the wild type EnvZ and RstB are specific for, and only phosphorylate, their cognate substrates, OmpR and RstA, respectively.
In the context of EnvZ, each single mutant continued to phosphorylate OmpR (Figure 4A). The single mutants EnvZ(TYA) and EnvZ(TLR) also showed weak phosphorylation of RstA. Of the double mutants, EnvZ(VYA) and EnvZ(TYR) both preferentially phosphorylated RstA, with the former not detectably phosphorylating OmpR and the latter only weakly phosphorylating OmpR. The other double mutant, EnvZ(VLR) appeared to have an approximately equal preference for phosphotransfer to RstA and OmpR. In the context of RstB, none of the three single mutants had a major effect on specificity and each continued to phosphotransfer only to RstA (Figure 4B). By contrast, the double mutants each behaved differently; the mutant RstB(TYA) phosphorylated only RstA, the mutant RstB(TLR) was promiscuous and phosphorylated RstA, OmpR, and CpxR, while the mutant RstB(VLA) did not phosphorylate any of the response regulators under these reaction conditions.
The systematic mapping of the mutational trajectories from EnvZ to RstB and vice versa led to several interesting observations (Figure 4). First, the behaviors of intermediates along individual trajectories are often quite different. The most dramatic example is the double mutants of RstB, with RstB(TLR) phosphorylating all three substrates examined, RstB(TYA) phosphorylating only RstA, and RstB(VLA) not phosphorylating any of the substrates. Second, we found that the individual specificity residues strongly influence each other. For example, the substitution V228T in the wild type RstB had very little effect on substrate preference, while the same substitution into RstB(VLA) converted a kinase that phosphorylated none of the regulators into a kinase that specifically phosphorylates OmpR (Figure 4B). The effect of the V228T substitution thus depends critically on the identity of other residues. As another example, the substitution Y230L in wild type RstA had little effect on specificity, but when introduced into RstA already harboring the V228T substitution produced a kinase that phosphorylated OmpR, RstA, and CpxR (Figure 4B). Similar observations were made for each of the other residues. Collectively, these data indicate that each specificity residue does not contribute independently or additively to the overall substrate specificity of a kinase. Rather, their contributions are frequently epistatic to one another and display context-dependence.
The mutational trajectory scanning done for both EnvZ and RstB was extended to the response regulator OmpR. Converting OmpR to have the phosphotransfer specificity of RstA required 3 mutations in alpha helix-1 and 3 mutations in the β5-α5 loop (Figure 2A). We treated the loop as a single entity and made the 15 possible OmpR-RstA intermediates: 4 single, 6 double, 4 triple, and 1 quadruple mutant. We then examined phosphotransfer from each of the 7 EnvZ-RstB mutants (Figure 4A), as well as wild type EnvZ, RstB, and CpxA, to each of the 15 OmpR mutants and to wild-type OmpR, RstA, and CpxR, for a total of 180 pairwise combinations. The complete data are shown in Figure 5 and Figure 6. All phosphotransfer reactions were run for 10 seconds, except for RstB and CpxA, which were run for 10 seconds and for 1 minute. To evaluate phosphotransfer, we quantified the relative intensity of each response regulator band for a given histidine kinase, yielding a profile of phosphotransfer activity for each kinase. From the comprehensive profiles, several observations and trends emerged (Figure 5 and Figure 6).
First, the triple mutant EnvZ(VYR) robustly phosphorylated wild type RstA as well as the quadruple mutant of OmpR in which all major specificity residues have been mutated to match those found in RstA. EnvZ(VYR) no longer phosphorylated OmpR, consistent with a complete change in specificity. However, it still phosphorylated two other OmpR mutational intermediates that the wild type RstB kinase did not, at least at the time point examined. This comparison supports the notion that the three residues we mutated in EnvZ are the dominant determinants of partner specificity, but that other residues play minor, fine-tuning roles, particularly in preventing non-cognate interactions.
Second, the data demonstrated that EnvZ and OmpR can tolerate some mutations in the specificity residues of their partner and still retain the ability to readily phosphotransfer. Wild-type EnvZ phosphorylated each of the single mutants of OmpR and three of the six double mutants nearly as well as it phosphorylated wild-type OmpR; however, it did not significantly phosphorylate the triple mutants or the quadruple mutant. Wild-type OmpR was efficiently phosphorylated by each of the EnvZ single mutants and one of the double mutants, but not by the triple mutant.
Third, these profiles reveal mutational paths from the specificity of the EnvZ/OmpR pair to that of RstB/RstA in which phosphotransfer is maintained. In other words, there is an ordered series of single mutations that can be made in EnvZ and OmpR that convert them to the specificity of RstB and RstA, respectively, without disrupting their ability to phosphotransfer to one another along the way. For example, wild-type EnvZ phosphorylates OmpR and the single mutant OmpR(RLAPFN) to similar levels, and conversely the single mutant EnvZ(TLA) phosphorylates both OmpR and OmpR(RLAPFN). In Figure 7 we extend this example to show how EnvZ and OmpR could, in principle, change its specificity to that of the RstB/RstA system by a series of alternating mutations in the two molecules without ever severely disrupting their interaction. There are several such paths, although each path is not necessarily equivalent because CpxA phosphorylates some mutational intermediates of OmpR and some EnvZ mutants phosphorylate CpxR. For instance, EnvZ(TLR) phosphorylated CpxR, and OmpR(ELRPFN) was phosphorylated by CpxA (Figure 5, also see Figure 4). The avoidance of cross-talk may limit the possible evolutionary pathways between EnvZ/OmpR and RstA/RstB, or at least favor some relative to others (Figure 7).
We also quantified the phosphotransfer profiles for each EnvZ mutant and the wild type kinases (Figure 5) and performed hierarchical clustering in two dimensions, i.e. both the kinase and regulator dimensions (Figure 6). As expected, clustering the kinases places RstB close to the EnvZ(VYR) while CpxA is separated from EnvZ, the EnvZ mutants, and RstB. Similarly, clustering the regulators placed RstA close to the quadruple mutant OmpR(EVATTP) while CpxR formed a clear outgroup on its own.
The hierarchical clustering analysis provides insight into the relative importance of individual specificity residues. The profiles were clustered based on phosphorylation levels, but show a clear correspondence to sequence features. For instance, the two primary clusters of OmpR mutants (labeled A and B in Figure 6) differ in the identity of their β5-α5 loops; that is, each OmpR mutant in cluster A has the residues ‘PFN’ whereas each mutant in cluster B has the residues ‘TTP’. The branch lengths separating these clusters are long relative to the total length of the tree, indicating that the identity of the loop strongly splits the phosphotransfer profiles of the regulators. Within both cluster A and B, the next split in the tree correlates with the identity of position 1; that is, each OmpR mutant in cluster C (or cluster E) has an arginine at position 1 while each OmpR mutant in cluster D (or cluster F) has a glutamate at position 1. Again, the branch lengths are relatively long indicating a clear correlation between phosphotransfer behavior and sequence. The next split is based on identity at the second position, either a leucine or valine. The final split is based on the identity at the third position. In each case, this final split has extremely short branch lengths, reflecting the near identity of each profile pair that follows the split. In sum, the clustering analysis suggests a hierarchy to the contribution made by individual specificity residues within the regulators. The loop, which includes three residues, made the strongest contribution, followed by, in order, positions 1>2>3. A similar analysis was applied to the EnvZ mutants revealing that position 2 (Y or L) drives the initial clustering of EnvZ mutants, followed by position 3 (R or A), and finally position 1 (V or T).
Maintaining specificity and preventing unwanted cross-talk between highly similar proteins is a fundamental challenge for cells, and one that remains poorly understood. In many cases molecular recognition plays a critical role, but the ability to pinpoint the amino acids responsible and to determine the contributions of each residue to specificity has been elusive. Here, we tackled this problem in the context of bacterial two-component signal transduction systems where specificity is dictated by molecular recognition [6]. We note, however, that two-component signaling pathways are not insulated at all levels – for instance, distinct signaling pathways sometimes converge transcriptionally by regulating overlapping sets of genes [5]. However, the focus here is on the specificity of phosphotransfer for which there is little evidence of significant, physiologically-relevant cross-talk [5].
To identify the amino acids that enforce the specificity of phosphotransfer, we examined patterns of amino acid coevolution in cognate kinase-regulator pairs. However, computational approaches alone do not unequivocally establish which residues are critical for specificity or reveal how each contributes to substrate selection. We therefore focused on experimentally rewiring the specificity of the model two-component proteins, EnvZ and OmpR. Previously we reported that EnvZ could be rewired to exhibit the substrate specificity of RstB by mutating as few as three of the coevolving residues [9]. Here we extended these results by rewiring OmpR to partner specifically with the histidine kinase RstB instead of EnvZ.
The residues mutated to rewire the partnering specificity of EnvZ and OmpR are predicted to be in close physical proximity during phosphotransfer. While no structure of EnvZ bound to OmpR exists, a co-crystal structure of a histidine kinase from Thermotoga maritima in complex with its cognate response regulator was recently solved [13] and can be used to infer physically proximal residues for EnvZ and OmpR. However, the spatial proximity of residues does not reveal how they govern specificity and whether individual residues promote the binding of a cognate protein or prevent interactions with non-cognate proteins. Moreover, the relative contribution made by each residue is difficult to discern from structural or spatial considerations alone.
To better dissect the role played by individual residues, we used alanine-scanning mutagenesis of EnvZ. However, of the nine major specificity residues in EnvZ (Figure 1), only one disrupted phosphotransfer to OmpR when mutated to alanine. These data suggest that no major hot spot exists for the EnvZ-OmpR interaction and that specificity is distributed across the interface. However, single alanine mutants do not always reveal the role of a particular residue. For example, EnvZ(L254A) showed very little change in substrate specificity, whereas EnvZ(L254Y) (Figure 4A) showed a significant level of cross-talk to RstA. Alanine-scanning mutagenesis also ignores any potential interdependencies that may exist between residues. Such relationships and non-additive effects on specificity were revealed in our comprehensive characterization of the mutational intermediates separating EnvZ and RstB. In several cases, the effect of a given substitution on phosphotransfer specificity depended significantly on what other substitutions had already been made; for example the mutation A255R in EnvZ had very little effect in the context of EnvZ(VYA) but led to significant promiscuity in the context of EnvZ(TLA). These sorts of contextual and epistatic effects have been seen in other studies of molecular interaction specificity including corticosteroid receptor-ligand interaction [16] and transcription factor-DNA binding [17]. In principle, the context dependence of amino acids could lead to ‘negative’ epistasis in which one mutation on its own is detrimental until a second mutation is introduced. For example, the protein β-lactamase has evolved resistance to cefotaxime by accumulating five different mutations [18]. While each mutation contributes to resistance, certain mutations actually decrease resistance unless, or until, one of the other mutations also occurs. We did not see any obvious case of negative epistasis when converting EnvZ to RstB or converting OmpR to RstA, as each mutation either increased interaction with the target molecule or had no effect. However, negative epistasis could exist when converting the specificity of other two-component signaling proteins.
Our trajectory-scanning analysis provides a glimpse into the possible evolutionary history of two-component signaling proteins. The EnvZ/OmpR and RstB/RstA systems are relatively closely related and likely evolved by duplication of a common progenitor followed by sequence divergence, including at specificity sites. Mutations in specificity residues following duplication presumably required corresponding changes in their cognate regulators in order to maintain operation of each pathway as they diverged from one another to avoid pathway cross-talk. Our results demonstrate that an ordered series of mutations could occur in EnvZ and OmpR such that the two proteins would maintain significant levels of phosphotransfer while transiting through sequence space to the specificity residues of RstB/RstA (Figure 7), or vice versa. In addition, this series of mutations can occur without ever entering the sequence space occupied by another closely related (in sequence) pair, CpxA/CpxR thereby preventing cross-talk. Interestingly though, not all mutational trajectories have these characteristics of maintaining phosphotransfer and avoiding cross-talk, raising the possibility that sequence evolution following duplication is constrained or that natural selection may have favored certain trajectories over others. Analysis of other proteins, including β-lactamase, lambdoid phage integrases, hormone receptors, and the metabolic enzyme isopropylmalate dehydrogenase [18]-[21], have led to similar suggestions about the constraints on protein evolution.
Our trajectory scanning approach is related to other systematic studies of protein-protein interaction specificity, including homolog-scanning [22] and site-saturation mutagenesis [23]. In many cases, however, such approaches involve single substitutions rather than an exploration of the entire mutational landscape separating two different proteins. Because the major specificity-determining residues of two-component signaling proteins have been previously mapped and are relatively limited in number, we were able to systematically generate all intermediates between EnvZ/OmpR and RstB/RstA. We note, however, that for the three major specificity residues in EnvZ, T250, L254, and A255, conversion to the corresponding residue in RstB requires two nucleotide substitutions. There are thus a great number of additional mutational intermediates that will be important to characterize in the future when considering the evolutionary history of EnvZ and RstB.
Intriguingly, our clustering analysis of the trajectory-scanning data also reveals an underlying hierarchy of the specificity-determining residues in EnvZ and OmpR. The clusters mapped based on phosphotransfer relationships were strongly correlated with the sequence of specificity residues. For example, the first branch point in the histidine kinase clusters separated those with a leucine at position 254 in EnvZ from those with a tyrosine at that position. These observations demonstrate that different residues contribute unequally to specificity. So although our alanine-scanning mutagenesis did not reveal any major hot spots and suggested that specificity is distributed, the trajectory-scanning study indicates that certain residues play more important roles than others. It will be interesting to see whether the hierarchies revealed here have influenced or constrained evolutionary trajectories of two-component signaling proteins, and if the relative importance of positions is similar in other two-component pairs.
The rational rewiring of two-component signaling proteins represents a stringent test of how well specificity is understood. Additionally, it opens the door to improved construction of synthetic signaling pathways in bacteria. Here, we used analyses of amino acid coevolution to guide the rational rewiring of the response regulator OmpR, a prototypical DNA-binding response regulator. With only a handful of mutations, the phosphotransfer specificity of OmpR was rewired to match that of RstA or CpxR. A recent study of Rhodobacter used structural data to guide the rewiring of chemotaxis response regulators to partner with the non-cognate kinase CheA3 [24]. The residues mutated in that study were in alpha helix 1 of the response regulator and most were identified here as coevolving residues. A genetic screen for altered partnering specificity of the regulator PhoB also identified residues in alpha helix 1 [25]. The successful rewiring of CheY and PhoB along with EnvZ and OmpR suggests that two-component proteins will be generally amenable to synthetic biology. However, it is not yet clear whether any histidine kinase (or response regulator) can be reprogrammed to behave like any other histidine kinase (or response regulator). For example, response regulators have been categorized into eight subfamilies, with the majority falling into just three [26]. OmpR, RstA, and CpxR all fall within one subfamily perhaps facilitating the interconversion of their specificities. Another important challenge for the future is to create novel kinase-regulator pairs with specificity residues that are orthogonal to those used in naturally occurring pairs. The functional hierarchies and interdependencies identified here will be important guides in engineering new, specific interactions. Similarly, these functional relationships should help in designing better algorithms for predicting kinase-regulator pairs in genomes of interest.
The life of a cell depends critically on the specificity of protein-protein interactions. Yet we still have a relatively primitive understanding of how such specificity is encoded within proteins and how a set of amino acids can allow binding of a cognate partner while excluding all other non-cognate partners. Two-component signal transduction systems represent an ideal model for addressing these fundamental issues as specificity is determined predominantly by a small set of residues. The consequent reduction in scope and scale enabled the systematic and comprehensive analyses presented here. More generally, the approaches used, including analyses of amino acid coevolution and trajectory-scanning mutagenesis, will be widely applicable to the study of specificity and molecular recognition in many other protein-protein interactions.
The software HMMER (http://hmmer.org) was used, with an E-value cutoff of 0.01, to identify and align histidine kinase and response regulator sequences from fully sequenced bacterial genomes in GenBank. For histidine kinases, the models HisKA, HisKA_2, HisKA_3, and HWE_HK from the PFAM database were used. For response regulators, the model Response_reg was used. Histidine kinases and response regulators with GenBank genome identifier numbers differing by one, indicating adjacent genes, were identified, concatenated, and treated as cognate pairs. Sequences were filtered to ensure that no two sequences were more than 90% identical. The final set contained 4375 concatenated pairs of histidine kinase and response regulators. Columns in the multiple sequence alignment (MSA) containing greater than 10% gaps were eliminated.
Mutual information (MI) between columns was measured as described previously [9]. MI scores were adjusted to account for differences in the average MI of each column. For columns i and j in a multiple sequence alignment, we defined MI(i,j)adj = MI(i,j)raw/(MI(i)avg+MI(j)avg)/2 where MI(i)avg and MI(j)avg are the average MI scores for column i and j paired with every other column in the alignment.
Phosphorylation profiles in Figure 6 were constructed by quantifying response regulator bands in each profile (Figure 5) using ImageQuant (GE Healthcare) and then normalizing such that each regulator's value was represented as a percentage of the maximally phosphorylated regulator for a given kinase. Profiles were then subjected to hierarchical clustering in two dimensions, with response regulators clustered using uncentered correlation and histidine kinases using Euclidean distance. Profiles were clustered using Cluster 3.0 [27] and visualized using Java Treeview [28].
All cloning and site-directed mutagenesis was done with Gateway pENTR vectors (Invitrogen) following procedures described previously [9]. Mutagenesis primers are listed in Table S1. Clones in pENTR vectors were mobilized into destination vectors for expression and purification using Gateway LR reactions according to the manufacturer's protocol (Invitrogen). Histidine kinases were moved into pDEST-His6-MBP and response regulators into pDEST-TRX-His6. Expression and purification was carried out exactly as described previously [6].
For autophosphorylation analysis of alanine mutants, histidine kinases were at a final concentration of 5 µM in HKEDG buffer (10 mM HEPES-KOH pH 8.0, 50 mM KCl, 10% glycerol, 0.1 mM EDTA, 2 mM DTT) supplemented with 5 mM MgCl2, 500 µM ATP, and 0.5 µCi [γ32P]-ATP from a stock at ∼6000 Ci/mmol (Perkin Elmer). Reactions were incubated at room temperature for 1 minute, stopped by the addition of 4X loading buffer (500 mM Tris-HCl pH 6.8, 8% SDS, 40% glycerol, 400 mM β-mercaptoethanol), and analyzed by SDS-PAGE and phosphorimaging.
For phosphotransfer analysis, histidine kinases were autophosphorylated as above, but were incubated for 60 minutes at 30°C. Phosphotransfer was assessed by incubating autophosphorylated kinases with response regulators, each at a final concentration of 2.5 µM, at room temperature for the indicated time (either 10 seconds or 1 minute). Reactions were stopped by the addition of loading buffer, and analyzed by SDS-PAGE and phosphorimaging. For the experiments in Figure 2, Figure 4, and Figure 5, autophosphorylated kinases were purified away from unincorporated nucleotides by diluting them 1∶10 in HKEDG and then washing eight times in Nanosep 30K Omega columns (Pall Life Sciences) to minimize the effect of any phosphatase activity. The final eluate was diluted back to the original volume and MgCl2 added to 5 mM before assessing phosphotransfer.
For alanine-scanning mutagenesis, to gauge reproducibility and assess significance in the changes observed, we repeated the phosphotransfer reactions for wild type EnvZ six times and a subset of the mutants three times. Standard deviations in each case were ∼5–10% of the mean.
|
10.1371/journal.ppat.1000764 | Six RNA Viruses and Forty-One Hosts: Viral Small RNAs and Modulation of Small RNA Repertoires in Vertebrate and Invertebrate Systems | We have used multiplexed high-throughput sequencing to characterize changes in small RNA populations that occur during viral infection in animal cells. Small RNA-based mechanisms such as RNA interference (RNAi) have been shown in plant and invertebrate systems to play a key role in host responses to viral infection. Although homologs of the key RNAi effector pathways are present in mammalian cells, and can launch an RNAi-mediated degradation of experimentally targeted mRNAs, any role for such responses in mammalian host-virus interactions remains to be characterized. Six different viruses were examined in 41 experimentally susceptible and resistant host systems. We identified virus-derived small RNAs (vsRNAs) from all six viruses, with total abundance varying from “vanishingly rare” (less than 0.1% of cellular small RNA) to highly abundant (comparable to abundant micro-RNAs “miRNAs”). In addition to the appearance of vsRNAs during infection, we saw a number of specific changes in host miRNA profiles. For several infection models investigated in more detail, the RNAi and Interferon pathways modulated the abundance of vsRNAs. We also found evidence for populations of vsRNAs that exist as duplexed siRNAs with zero to three nucleotide 3′ overhangs. Using populations of cells carrying a Hepatitis C replicon, we observed strand-selective loading of siRNAs onto Argonaute complexes. These experiments define vsRNAs as one possible component of the interplay between animal viruses and their hosts.
| Short RNAs derived from invading viruses with RNA genomes are important components of antiviral immunity in plants, worms and flies. The regulated generation of these short RNAs, and their engagement by the immune apparatus, is essential for inhibiting viral growth in these organisms. Mammals have the necessary protein components to generate these viral-derived short RNAs (“vsRNAs”), raising the question of whether vsRNAs in mammals are a general feature of infections with RNA viruses. Our work with Hepatitis C, Polio, Dengue, Vesicular Stomatitis, and West Nile viruses in a broad host repertoire demonstrates the generality of RNA virus-derived vsRNA production, and the ability of the cellular short RNA apparatus to engage these vsRNAs in mammalian cells. Detailed analyses of vsRNA and host-derived short RNA populations demonstrate both common and virus-specific features of the interplay between viral infection and short RNA populations. The vsRNA populations described in this work represent a novel dimension in both viral pathogenesis and host response.
| Biological systems are protected by innate immune mechanisms initiated by host sensors called pattern recognition receptors (‘PRRs’) that recognize specific “foreign” features of invading pathogens to initiate multiple downstream anti-pathogen cascades. PRRs that detect nucleic acid structures characteristic of viral infection (such as single- or double-stranded RNA or DNA) are among the innate responders that protect diverse cell types from viral pathogenesis (for review, see [1],[2]). How the cell handles viral double-stranded RNA (dsRNA) is of special interest because dsRNA is a necessary intermediate in the replication of RNA viruses. In addition to dsRNA that forms during replication of the virus genome, RNA duplexes can form due to self-complementarity in the virus genome, and in some instances, from sense-antisense transcription of overlapping genes.
Four of the most studied families of PRRs for dsRNA are: (a) cytoplasmic RNA helicases like Retinoic acid-inducible gene I & Melanoma differentiation-associated gene-5 (“RIG-I” & “Mda-5,” which trigger mitochondrial-localized antiviral pathways); (b) Protein Kinase R (“PKR,” which induces a translational arrest state in cells after sensing dsRNA); (c) 2′–5′ oligoadenylate synthetase (“OAS,” which stimulates the ssRNase activity of RNase L in response to dsRNA); and (d) Toll-like receptors (“TLRs,” which bind various forms of RNA or DNA). All of these PRRs trigger the Interferon (IFN) responses, and activate IFN-stimulated genes (ISGs) that establish an antiviral state in the infected cell (for review, see [3]). The IFN signaling pathway is central to the detection of, and response to, virus infections in cells. Type I IFNs (IFN-α and IFN-β) make up one of the first lines of defense in the innate immune response to viruses by inducing antiviral ISGs, modulating the levels of specific host-encoded miRNAs [4], and in a feedback loop, that of PKR and OAS. Many viruses are also susceptible to treatment with Type I IFNs, and conversely, cells that have higher basal activity of ISGs seem to mount a more successful antiviral response, and are not targeted by viruses [5].
Dicer is another PRR that recognizes dsRNA, chopping it into smaller duplexes called siRNAs that are 19–27 nucleotides (nt) long [6],[7]. These siRNAs have a terminal 5′ mono-phosphate and a terminal 3′ hydroxyl on both strands, generally have 2 nt 3′ overhangs, and are fed into an RNA-induced silencing complex “RISC” (for review on Dicer and Argonautes, see [8],[9]). siRNA duplexes are unwound, and only one strand remains associated with RISC (the mechanism of unwinding and choice of strand is poorly understood; for review, see [10]). One of the key components of RISC is a protein called Argonaute-2 (Ago-2), which belongs to the Argonaute family of proteins. Ago-2 is the only member of the family that has cleavage activity, and is the designated ‘slicer’ protein in RISC that mediates cleavage of mRNA in a sequence-directed manner by a process termed RNA interference, or ‘RNAi’ [11],[12],[13],[14].
There is strong evidence for an antiviral role for RNAi in plant and invertebrate systems (for review, see [15],[16],[17]). Viruses replicate most effectively in these systems in the absence of key elements of the RNAi pathway: either in cells lacking components of the RNAi machinery, or in the presence of virus-encoded suppressors of the silencing pathway (for review, see [18],[19]). As expected, virus-derived siRNAs (vsRNAs) can be detected in some plant and invertebrate systems that are capable of mounting a successful/partially successful RNAi response [15],[16],[17]. A population of vsRNAs would be an expected component of any viral defense pathway that acted through an RNAi mechanism.
In mammalian cells, short duplex RNAs can effectively enter the RNAi pathway and function in sequence-specific silencing, while duplexes longer than 30 nt generally produce a more complex response including the induction of multiple non-specific pathways including the IFN response (for review, see [20],[21]). Indeed, RNA and DNA viruses have evolved a host of defense mechanisms to counteract the nonspecific signaling effects of dsRNA. For example, Adenovirus VA RNA sequesters PKR [22], while proteins from Vaccinia virus (E3L), Porcine Rotaviruses (NSP3), and Influenza A virus (NS1) sequester dsRNA and prevent stimulation of the IFN response [23],[24],[25],[26]. Viral proteins can also inhibit signaling downstream of dsRNA binding, as in the case of the HCV protease NS3/4A, which cleaves IPS-1 (the RIG-I/MDA-5 signaling partner) to consequently disrupt induction of IFN responses [27]. Several of these dsRNA-binding proteins may also facilitate viral evasion of host immune responses by inhibiting RNAi [28]. Additionally, some viruses make their genomes inaccessible to PRRs of various types including IFN effectors and the siRNA-programmed RISC complex (e.g. [29]).
Viruses may also perturb another class of effectors involved in RNAi called micro-RNAs (miRNAs), which are a class of cellular small RNAs generated by Dicer from hairpin structures. Cellular miRNA profiles are frequently modulated upon infection by viruses, and this may contribute in some cases to infectivity and pathogenesis [30]. Conversely, some viruses usurp the host miRNA machinery for processing miRNA-like structures encoded in the viral genome, potentially using these molecules for regulation of virus/host gene expression [31].
With so much potential for RNA-mediated cross talk between the IFN response, the RNAi pathway, and the virus itself, it has been difficult to demonstrate a precise role for the RNAi pathway in vertebrate antiviral defense. The difficulties in segregating IFN and RNAi functions have given rise to speculations that the antiviral role of RNAi may have been lost during evolution, or alternatively, that RNAi-based defense may only be harnessed by triggers such as short hairpins and siRNAs that do not stimulate the IFN pathway. There has been some attempt at demonstrating recognition of viral RNA by the RNAi machinery. For instance, in Vero cells (which lack IFNα/β), inhibition of RNAi by Dicer knockdown increases replication of an RNA virus, the Influenza A Virus [32]. Additionally, there are cases where short virus-derived RNAs can be detected in vertebrate systems (e.g. from HDV [33] by high-throughput sequencing, and the HCV replicon [34], by bulk analysis methods). However, it is still not clear how general the presence of such RNAs is, and whether these RNAs can participate in host defense mechanisms. To complicate this issue, many of the classically-studied virus-host systems have been chosen based on the ability of the virus to rapidly replicate and kill host cells; these experimental infection systems may artificially under-represent the capacity of vertebrate cells to protect themselves, hence biasing against systems where RNAi might have a significant role in host-virus interactions.
Here, we sought a broader survey of potential RNA-derived defenses in viral infection systems. Given no knowledge of which virus type might engage the RNAi machinery, and which cell types might efficiently use this machinery in defense, we cast a wide net in terms of both virus families and host cells. In this study, we describe small RNA populations from six different RNA viral pathogens, each in a variety of animal cell infection systems (including both immune-competent and immune-compromised hosts). Upon examining small RNA populations from ∼150 samples with sample-specific DNA barcodes, we found viral-derived small RNAs (vsRNAs) from each virus, with vsRNA populations sensitive to both viral and host characteristics. A more detailed analysis of vsRNAs in two viral infection models (Hepatitis C Virus and Poliovirus) in various host types revealed that multiple distinct pools of vsRNAs may co-exist during infection: as single strands, as part of duplexes, and in complexes that may contain Argonautes. We also observed specific changes in cell-derived miRNA populations, providing a clear indication of host perturbation by the virus. The characterization of small RNA populations during RNA virus infections provides both an experimental entry point, and an indication of the complexity that will need to be addressed in understanding roles for small RNAs in host and viral processes.
In the following sections, we will describe small RNA populations present during infection of animal cells with six different viruses. In each case, we have taken infected cells, extracted small RNA populations, and characterized these populations using high-throughput sequencing methods. Two high-throughput sequencing platforms were used: Roche/454 pyrosequencing (http://www.454.com/), to obtain several hundred thousand sequences from pools of appropriately linkered amplicon templates; and Solexa/Illumina technology, which yields larger datasets of shorter reads (http://www.illumina.com/). Due to the large number of samples to be analyzed, we used DNA barcodes to ‘tag’ RNA samples from individual experiments, which facilitated sequencing in parallel from multiple samples. This allowed us to work with samples from different viral systems and diverse experimental conditions in a cost-effective manner, with a small number of instrument runs. Viral-derived sequences were identified in sequence datasets through pattern matching using standard software (BLAT [35] and BLAST [36]). We use the term ‘vsRNA’ to refer to small RNA segments whose sequences show perfect complementarity to the infecting viral genome at every base position (reference genomes listed in Table S1). vsRNAs are distinct from host-derived miRNAs that may show partial complementarity to sites in the viral genome (e.g. [37],[38],[39], Fig. S1). We detected 77,609 vsRNAs out of 19,425,777 sequences from 151 datasets (Fig. 1, Figure S2: length distributions). The most abundant vsRNAs from each virus are listed in Table S2.
For a small number of vsRNAs (0.033%) we observed a perfect match to both the host and viral genomes (Table S3, Table S4). The fractions of vsRNAs that matched host genomes were approximately as expected by random sequence coincidence (for example: the human genome, with a unique genome complexity of 2×109 bp, would match approximatly 1 in 4000 arbitrary 22-mer sequences). The perfect nature of the homology makes it difficult to determine whether this minor class of sRNAs was derived from the host or from the virus.
Furthermore, to validate the specificity of the barcoding and sequencing assays, we carried out sequence comparisons to the full set of viruses for each experimental sample. We identified 13 vsRNAs that were ‘rogue’ hits i.e. mapped to one of the other 5 viruses not used in that particular experiment. In no sample were the ‘rogue’ matches present at more than 0.008% of all parsed sequences (Table S3).
For certain purposes, it will be of interest to compare vsRNA incidences in different samples. Such comparisons require some normalization for total depth of RNA sequencing. In Table S3, we provide two distinct normalizations for each sample: normalization to total small RNAs recovered and sequenced (v/sRNA), and normalization to the population of cellular miRNAs that are expected to represent a large proportion of bona-fide small RNA effectors (v/miR; miRNAs are defined as documented in miRBase ver9.2 [40],[41],[42]). There is a substantial challenge in choosing and interpreting appropriate normalization schemes: any change in sample character that results in increased levels of non-specific degradation of RNA will increase the levels of non-specific decay products (which may include decay products of both cellular and viral long RNAs), and impact both v/miR and v/sRNA ratios. Another important consideration is whether the difference in v/miR (or v/sRNA) ratios between two samples is above the variance in ratios observed between technical replicates. For all relevant technical replicates in our analysis, the variances in v/miR and v/sRNA ratios were <3-fold and <1.5-fold, respectively. These notes provide caution in interpreting small differences in normalization values between samples.
In describing the results of this work, we have taken care to avoid any a-priori assumption that small RNAs identified by sequencing play a functional role in gene silencing, viral pathogenesis, or host response. In the Discussion section, we will summarize arguments pertaining to this question.
Components of the worm RNAi machinery such as the argonaute, rde-1 [43], the dsRNA binding protein, rde-4 [44],[45], and the RNA-dependent RNA Polymerase or RdRP, rrf-1 [46] are essential for protection against Vesicular Stomatitis Virus ‘VSV’ [47], and Flock House Virus ‘FHV’ replication [48]. To characterize small RNA populations in an animal system known to utilize the RNAi machinery in antiviral defense, we used C. elegans experimentally infected with FHV RNA1ΔB2 (FHV RNA1 that expresses a mutant version of the RNAi suppressor protein, B2; [48]).
Two different vsRNA capture and library production schemes were used to enrich for Dicer products or for RdRP products, both of which have structures distinct from those of RNA fragments generated by alkali-induced degradation. The first (5′-phosphate-dependent cloning) requires a single phosphate at the 5′ end of the RNA, and allows for the capture of Dicer products (which have a mono-Phosphate and a hydroxyl moiety at their 5′ and 3′ termini). The second (5′-phosphate-independent cloning; [49]) is designed to capture RNA populations with any number of 5′ phosphates (zero, mono, di, tri), including both RdRP products (which have a tri-Phosphate and a hydroxyl moiety at their 5′ and 3′ termini) and Dicer products. Both procedures require a 3′ end that can ligate to a pre-adenylated linker, and allow for the capture of 3′-OH and 2′-O-Methyl structures but not 3′ phosphate termini, thus minimizing the extent of capture of degradation products (many of which have 3′ mono-phosphate termini).
5′ mono-phosphorylated (5′-P) vsRNAs were present during abortive FHV RNA1ΔB2 replication in wild-type animals (v/miR = 0.007; Fig. 2B). vsRNAs were absent in two RNAi-defective mutants, rrf-1(pk1417)I and rde-4(ne299)III (Table S3), while as predicted, genomic viral RNA replicated to high levels in these mutants (Parameswaran P, unpublished). Similarly, vsRNAs were much reduced (19-fold; P-value = 2.3E-227) in rde-1(ne300)V mutants (Fig. 2C). We also observed a difference in strand ratios of vsRNAs (Positive∶Negative) between strains: 1∶2.4 in wild-type, versus 1∶1.1 in the mutant, rde-1 (P-value = 0.0016).
The population of RNAs captured with no requirement for a 5′-P terminus (i.e. 5′-xP RNAs) yielded a stronger signature for vsRNAs in wild-type worms with replicating RNA1ΔB2 (v/miR = 0.019; Fig. 2F). Fewer vsRNAs mapped to the positive strand of FHV than to the negative strand, with a Positive∶Negative vsRNA strand ratio of 1∶3.5 (P-value = 1.1E-48). rde-4−/− was the only RNAi-defective mutant that yielded a detectable signature for 5′-xP vsRNAs (v/miR = 0.0014), with a strand ratio (Positive∶Negative) of 1.3∶1 (Fig. 2G). Interestingly, in wild-type worms, both 5′-P and 5′-xP vsRNAs were distributed throughout the length of the genome, with increased frequencies of positive-strand vsRNAs detected in the 3′ region that also encodes the subgenomic RNA species RNA3 (Fig. 2B, 2F).
To identify virus-host systems in which RNAi might participate as an antiviral defense mechanism, we sequenced small RNAs from diverse populations of cells (of human or mouse origin) infected with one of five viruses: Vesicular Stomatitis Virus (VSV), Poliovirus, West Nile Virus (WNV), Dengue Virus, or Hepatitis C Virus (HCV). These viruses were purposefully chosen as token members of diverse families (Fig. 1), and are mostly positive-stranded (except for Vesicular Stomatitis Virus, which is negative-stranded). We identified vsRNAs from all six surveyed viruses (Fig. 1; Table S3), albeit in only a fraction of all infected samples investigated. From this initial survey, we made a choice of a single host-virus system in which to further investigate vsRNA biogenesis. The viral system chosen for this purpose was HCV infection of human Hepatoma cells. While the remainder of the Results section will focus primarily on HCV, we will briefly summarize our observations in the four other virus systems. For the Polio, VSV, West Nile and Dengue (Fig. S3) systems (Table S3), the abundance and molecular features of vsRNAs were dependent on the nature of the host and/or the virus, with some notable trends:
A more detailed description of small RNA profiles from West Nile Virus, Dengue, Vesicular Stomatitis Virus, and Poliovirus is provided in the supporting document (Text S1), and in Supplementary Tables & Figures.
HCV is an enveloped, positive-stranded RNA virus that is a member of the Flaviviridae family. Its genome is flanked by short stretches of structured RNA in the 5′ and 3′ UTRs, is uncapped, and lacks a 3′ poly-A tail. A previous study with HCV-1b-infected Huh7.5 cells failed to identify HCV-derived vsRNAs using standard sequencing protocols [57]. We expanded on this work by choosing two cell-culture-based systems that are used for studying HCV replication: an HCC cell line (Huh7) harboring a subgenomic replicon of genotype 1b [55], and an infectious virion system (Huh7.5 cells infected with tissue-culture-produced virions of genotype 2a [58]).
v/miR levels of 5′-P vsRNAs from replicon cells varied between 0.03 and 0.14 (Fig. S8, Table S3), with an estimated 7300 +/− 2200 vsRNA molecules per cell (based on the approximation that the most abundant miRNA, miR-122a, is present at 15,000 copies per hepatoma cell in culture [59]). In virus-infected Huh7.5 cells, we detected a very low incidence of vsRNAs at early time points, with an increase over time (Fig. S9, S10). vsRNAs were found starting at 1 day post-infection (dpi) in Huh7.5 cells (v/miR = 0.000025), and steadily increased (excluding a possible dip at 11dpi), reaching a v/miR value of 0.056 at 15 dpi.
vsRNAs from the sense (positive) and antisense (negative) viral strands were roughly equally abundant in both the replicon and the infectious virion systems (sense-to-antisense ratios of 1.07 to 1.9; Fig. 5, Table S3). This contrasts with the observed ratios of genome-length viral RNAs, where the sense strand is 5- to 10-fold more abundant in replicon-harboring cells and in infected hepatocytes [53],[54],[55]. The near-equivalent abundance of vsRNA strands is consistent with vsRNAs deriving from cleavage of a double-stranded replication intermediate, which has an equimolar ratio of positive and negative strands.
HCV vsRNAs from both the 1b and 2a genotypes were distributed throughout the length of the genome, with several ‘hotspots,’ where many vsRNAs were found clustered in specific regions of the genome (Fig. 5). Direct comparison of vsRNA distributions, and of individual vsRNA hotspot species between the replicates demonstrated that both were reproducible properties of HCV infection (Fig. S11A–D). Additionally, the ability of structured sequences such as those found in the HCV IRES and EMCV IRES to produce specific small RNA populations is of considerable interest. A comparison of vsRNA localization and published secondary structures of the HCV IRES and EMCV IRES is shown in Fig. S12 & S13.
We also found some evidence for nucleotide bias among HCV replicon (HCVrep)-derived positive-strand and negative-strand vsRNA populations, including a bias toward strings of Cs and Gs at the 5′ and 3′ termini respectively (Fig. S14). This potentially creates favorable conditions either for intramolecular base pairing within a vsRNA, or for base pairing between overlapping sense-antisense vsRNA pairs at their termini.
We further investigated the potential for duplexed structures of sense and antisense vsRNAs from HCVrep, by comparing sequence placement for sense and antisense vsRNAs within the viral genome. There are examples of independently captured sense and antisense HCVrep-derived vsRNAs that could derive from a dsRNA duplex with a 0–3 base 3′ overhang (Fig. 6A–C). These are similar to canonical overhangs in Dicer-generated siRNAs. If we first separated HCVrep-derived vsRNAs into different size ranges and then calculated the distribution of overhangs, duplexes formed by overlapping sets of 20–21 nt sense and antisense vsRNAs had a strong bias for one or two nt 3′ overhangs (Fig. 6A). On the other hand, duplexes formed by vsRNAs that are 24–26 nt long have a wider overhang range of zero to three nucleotides (Fig. 6B; False Discovery Rate is less than 0.01%).
To explore the possibility that vsRNAs may associate with core components of the RISC machinery (the Argonaute, or “Ago” proteins) despite our inability to detect a role for vsRNAs in silencing pathways, we used transient transfection to express FLAG/HA-tagged Ago-1, Ago-2, Ago-3 or Ago-4 [12] in HCVrep cell lines. We note a limitation of the Argonaute immunoprecipitation (IP) assays in that a large fraction of small RNAs from the cell may be capable of associating with Argonautes in a specific or non-specific manner; nonetheless, the expectation of such experiments is that immunoprecipitation will lead to enrichment for small RNAs that specifically associate with the tagged Argonaute. We compared RNA populations from each of the four Argonaute IPs to RNAs from the Mock-IP (i.e. IP with FLAG Ab, using lysates from mock-transfected cells), to give us an indication of the specificity of the IPs, and conversely, of the degree of non-specificity due to “stickiness” of the α-FLAG-M2 Antibody (Fig. 7A). Specifically, we compared the enrichment for vsRNAs in the Ago IPs (relative to Mock IPs), first to the enrichment for RNAs previously known to be Ago-associated (miRNAs, some miRNA*s) [12],[60],[61],[62], and second to the de-enrichment for RNAs which have less (or no) specific association with Argonaute (ribosomal RNAs). In the Ago IPs, we observed a several-fold enrichment for vsRNAs (similar to that observed for miRNAs), accompanied by a marked de-enrichment for rRNA fragments (Fig. 7A; for raw data, see Fig. S15B–C). This indicates that at least a subpopulation of vsRNAs associates specifically with all four Argonautes.
A striking feature of Ago association in general is the rapid reduction of the initial dsRNA duplex to a single-stranded guide RNA [63]. We compared the duplex properties of vsRNA populations in total cell lysates to those of vsRNAs that are specifically associated with the Argonautes (Fig. 7B). The IP datasets for Ago-2 and Ago-4 showed a notable feature: a striking de-enrichment for duplexes with 0–3 nt 3′ overhangs, compared to their respective total RNA samples (P-values of 0 and 2.1E-49 respectively). Ago-1 IP and Ago-3 IP showed a de-enrichment for such duplexes, but total RNA samples for Ago-1 and Ago-3 did not have sufficient sequence coverage to allow for a comparison. These data suggest that HCVrep cell lysates have populations of duplexed (and some single-stranded) vsRNAs, with only a single strand of each duplex reproducibly incorporated into an Ago complex.
We were interested in understanding the role played by the RNAi machinery in shaping the course of viral pathogenesis in vertebrate and invertebrate host systems. Our work builds on prior observations that effective replication by certain RNA viruses in plants, C. elegans and in D. melanogaster requires suppression of the antiviral RNAi response [15],[16],[64]. In each of these invertebrate systems, there is strong evidence for an antiviral mechanism that is directed by small RNAs derived from the virus genome (‘vsRNAs’). Similar questions of great interest in mammals remain unresolved. Using a sequencing approach to investigate the involvement of small RNA-based responses in viral infection, we detected vsRNAs from several mammalian host-viral systems.
We consider two possible sources for the vsRNA populations that were observed during infection: (i) the vsRNAs could be participants in a specific pathway (or pathways) in which small RNAs are generated from the viral genome for host or viral functions; and (ii) the vsRNAs could be products of non-specific degradation of longer (e.g. full-length or subgenomic) viral RNAs mediated by ssRNA nucleases, chemicals, pH, mechanical shear etc.
We note that the small RNA populations characterized by sequencing may be a mixture of (i) biologically relevant small RNAs, and (ii) degradation products with limited significance. In particular, any population of larger RNAs, on extraction and experimental manipulation, can yield a sub-population of RNAs in every size range, including the miRNA and siRNA size range of 19–30 nt. Since viral genomic RNAs and mRNAs are abundant in infected cells, we would certainly expect degraded derivatives to contribute to sequenced pools.
Despite the likely capture of some degradation products, there are several strong indications of vsRNA populations that are not simply the result of degradative mechanisms.
The above characteristics of vsRNA populations strongly argue that mammalian cells retain the ability (present in lower organisms; for review, see [16],[17]) to utilize vsRNAs as RNAi effectors. As for any host-virus interaction, the expectation would be that the utilization of small RNA-based mechanisms would be highly dependent on the biology of the virus and the host. This is evident from observed differences in the strandedness of vsRNAs in different systems infected with the same virus, and may be accounted for by several factors such as (a) substantial contributions from potential degradation mechanisms; (b) nuclease activity on ssRNA templates, especially if the secondary structures in one strand are targeted preferentially; (c) strand accessibility; (d) Dicer processivity and activity, and (e) selective RISC loading. Some infectious systems (e.g. Poliovirus or VSV in HeLa cells; Fig. S18B, S17C) yield a vsRNA profile with a skew towards positive strand vsRNAs. By contrast, productive infections with the same viruses in other host environments (e.g. Poliovirus in mouse muscle (Fig. 4C, S18H) or VSV in MEF/BHK cells (Fig. 3D–E, S17B) can show strong signatures from both strands, consistent with potential generation by dsRNA-related mechanisms.
Host-virus interactions may also determine bulk abundance of vsRNA populations. In the systems we surveyed, vsRNA abundance varied between 0 and 12.8 (v/miR; relative to miRNAs), or between 0 and 0.02 (v/sRNA; relative to all small RNAs; Table S3). Even within a single host, tissue specific factors seem to govern the efficacies of small RNA-related mechanisms. This was observed in IFNαβR−/− mice infected with West Nile Virus: even though levels of full-length viral RNA in the brain were comparable to levels in spleen & lymph nodes [5],[50], vsRNAs in the brain were undetectable (454-141, 454-144; Table S3).
Despite the complex nature of the factors that govern vsRNA abundance, a number of trends are suggested by comparison of vsRNA levels in paired “wild-type” and “mutant” hosts infected with various viruses. In each of the six sets of experiments where we compared vsRNA levels in parallel infections in IFNαβR(+/−), or ago-2(+/−) hosts, we observed an increase (ranging from 1.7-fold to >30-fold) in vsRNA levels in the infected “mutant” host. For five out of these six comparisons, the variation in vsRNA abundance between “wild-type” and “mutant” hosts is higher than the (maximum) 3-fold difference we observed in technical replicates.
Viruses can hijack the RNAi machinery at various levels: at the level of Dicer, Argonaute, RISC-mediated silencing, or a combination of the above (for review, see [19]). Downregulation of Dicer has an attenuating effect on Hepatitis C Virus replication [88]. Several studies have shown that HCV proteins inhibit Dicer and Argonaute [34],[89],[90]. In these experiments, inhibition is not complete (∼60–70%; [89]), concordant with our ability to detect vsRNA populations with structures consistent with synthesis by Dicer. We also see evidence for loading of vsRNAs onto Ago complexes, indicating that some downstream steps are not entirely impaired.
Key players in the piRNA pathway, Piwi and Aubergine, are required for protection against Drosophila X Virus infections [84]. The PIWI-piRNA pathway produces 26–31 nt RNAs in a Dicer-independent manner, and is mostly active in the germline [91], and in adjacent somatic tissues [92],[93]. We note that animal systems with PIWI proteins (vertebrates), we observe a wide size range for vsRNAs (Fig. S2). These diverse size classes, together with our observation that vsRNAs are still present in systems that lack Dicer (Poliovirus infections in dcr−/− MEFs), suggest that the Dicer pathway (which is thought to primarily produce RNAs shorter than 27 nt) may not be the only source for these vsRNAs, and that there may some contribution from baseline levels of PIWI proteins (or other novel proteins) in the various systems. We also identified a population of Polio-derived vsRNAs in dcr-1−/− and in eri-1+/+ MEFs that can form duplexes with 9 or 10 bp overhangs (Fig. S16I, S16L), which are hallmarks of piRNAs that are formed by a ping-pong mechanism [94],[95]. This observation brings up the question of how the piRNA pathway interfaces with the RNAi pathway during viral infections, and whether in the absence of the RNAi (and perhaps the IFN) machineries, we could uncover a role for the piRNA pathway in antiviral immunity.
Long dsRNAs, such as those found in the viral replication intermediate, primarily induce the non-specific interferon response in mammalian cells. In the absence of the robust IFN pathway, long dsRNA becomes a trigger for the sequence-directed RNAi pathway [96],[97],[98],[99]. Accordingly, we detected a more abundant signature for dsRNA-derived vsRNAs in IFNαβR−/− mice (Poliovirus and West Nile Virus; Table S3). Conversely, when we stimulated the Interferon pathway by providing an exogenous supply of IFN-α to a culture of HCVrep-harboring cells, we found that concurrent with reduction in full-length genomic RNA (as reported in: [55],[100]), HCV-derived vsRNAs also dropped several fold (Fig. S20). Thus vsRNA abundance seems to associate with the strength of the IFN response: the more robust the IFN response, the fewer the number of vsRNAs. The observed boost in vsRNA abundance in IFN knockout conditions could conceivably reflect a number of distinct effects including augmented viral replication in the absence of IFN, and/or specific interactions between IFN stimulation and the RNAi machineries.
We observed consistent effects of viral infection on miRNA profiles in distinct experimental systems (Fig. S7). For example:
Downregulation of the miR-17-92 cluster (i) in virus-infected cells may be a pro-apoptotic indicator, since an increase in expression of the miR-17-92 cluster is associated with an inhibition of apoptosis [101],[102],[103]. A decrease in miR-125b (ii) is observed post-LPS-stimulation of macrophages, and causes de-repression of TNF-α in a sequence-specific manner [104]. We speculate that miR-125b may be one of the regulators of the TNF-α response during viral infection, and that regulation of miR-125b may require an intact IFN response. Increased levels of the anti-apoptotic miR-21 (iii) were found in memory and effector T cells, compared to naïve T cells [105]. It is tempting to speculate that this upregulation of miR-21 may be indicative of proliferating immune cells post-recognition of viral antigens.
The identification of multiple vsRNAs, some of which are derived from ‘hotspot’ locations in diverse viral genomes may be useful for designing cocktails of siRNAs for therapeutic purposes, and for mapping areas of the viral genome that are more susceptible to RNAi machineries. Much work still remains to be done in designing siRNA duplexes such that the most accessible strand of the virus may be successfully targeted by siRNAs. A major concern is whether RNAi would be effective in combating viruses with fast replication kinetics. Poliovirus [106] and Semliki Forest Virus [107] replication in cultured cells have been effectively attenuated by an exogenous supply of siRNA triggers against the virus. Whether this holds true for clearing infections in whole organisms remains to be tested.
Mice used for experimental infections with West Nile Virus were genotyped and bred in the animal facilities of the Washington University School of Medicine, and experiments were performed with approval from, and according to the guidelines of, the Washington University Animal Studies Committee (which is IACUC approved). For infections with Poliovirus, mice that express the human Poliovirus Receptor gene were maintained in BSL-2 animal facilities at Stanford University. The methods for mouse use and care were approved by the Stanford University Administrative Panel on Laboratory Animal Care (APLAC), and are in accordance with the USDA Animal Welfare Act and the Public Health Service Policy on Human Care and Use of Laboratory Animals.
Worms and mouse tissues were flash frozen in liquid nitrogen, powdered and lysed. Cell lines were trypsinized and washed in PBS pre-lysis. The mirVana kit (Ambion) was used for isolation of RNAs shorter than 200 nt from all samples. Different protocols were used to prepare libraries of RNAs with mono-phosphorylated, or with modified 5′ termini. Briefly, in the 5′-P-Dep protocol, RNA was linkered at the 3′ terminus, size-selected, linkered at the 5′ terminus, reverse-transcribed, amplified using ten-nucleotide barcoded PCR primers [115], and sequenced on the Roche/454 GS-20 or the GS-FLX platforms. In one version of the 5′-P-IND protocol (used for all samples sequenced on the GS-20/FLX), the RNA was linkered on the 3′ end, size-selected and reverse-transcribed before addition of the second linker [49]. In the 5′-P-IND protocol for preparing Solexa libraries, RNA was linkered on the 3′ end, dephosphorylated and re-phosphorylated (to replace any multi-phosphate moieties with a monophosphate), linkered on the 5′ end, reverse-transcribed and amplified (method courtesy of Guoping Gu). All 5′-P-Dep and 5′-P-IND libraries for Solexa were prepared by introducing four-nucleotide barcodes as part of the 5′ linker, rather than during PCR amplification (Lui WO, Parameswaran P; unpublished). Shorter barcodes allowed for the use of non-barcoded primers for amplification, and most importantly, for greater allocation of sequence space to the sequence of interest. After preparation of barcoded libraries, the libraries were pooled together in molar ratios that were proportional to the sequencing depth required from each sample. The presence of a large number of libraries required multiple sequencing runs on the Roche/454 and the Illumina sequencers.
Sequences obtained from the Roche/454 platform were handled differently from those obtained on the Solexa platform due to different amplicon structures. Sequences from the Roche/454 platform were binned based on their barcodes using Barsort [115], trimmed using perl scripts (PP) and aligned using a local copy of the multiple alignment program Blast (word size = 11). Sequences from Illumina's platform were segregated into individual datasets based on perfect barcode match, trimmed to ensure removal of the flanking adapter sequences before analysis, and aligned using Blat (tile size = 11; step size = 5, run on Mac OS X). Alignments were performed to databases of species-specific miRNAs, and of the various viral genomes. The alignments were subsequently parsed to yield unique hits with the highest homology for each matching read. We filtered for matches of >16 nt for Roche/454 sequencing, and of >19 nt for Solexa sequencing.
For comparing incidence of miRNAs across samples, the frequency of miRNAs and standard error were computed as per the following formulae [116]:
The distribution, length and strandedness of vsRNAs (i.e. RNAs that mapped to the genome of the infecting virus) were plotted as a function of the length of the viral genome. Starts and ends of positive-strand and negative-strand vsRNAs were then compared to generate a matrix of the percent incidence of different types of overhangs. For generating the nucleotide bias, the vsRNAs were aligned at the 5′ termini (or at the 3′ termini), and the nucleotide frequencies were counted at each position ten nucleotides upstream and ten nucleotides from the 5′ end of the vsRNA population (or ten nucleotides downstream and ten nucleotides from the 3′ end of the vsRNA population). These numbers were then fed into Pictogram (Chris Burge, MIT; http://genes.mit.edu/pictogram.html), and were normalized to the total numbers of A, C, G and T in the population of cloned vsRNAs, thus circumventing any bias that arose from a skewed ratio of nucleotides inherent to either the cloning process, or to the genome of the virus.
Calculation of P-values: For establishing statistical significance, two-tailed P-values were calculated using either the Z-test for two proportions, or Fisher's Exact test. Fisher's Exact test was used in test cases with small sample or population sizes, while the Z-test was used if sample or population sizes were large.
Calculation of False Discovery Rate (FDR): Populations of 20,21-mers and 24,25,26-mer HCV vsRNAs were pooled, and positive strand and negative strand vsRNAs were randomly selected from the pool. 10,000 datasets that mirrored positive strand and negative strand vsRNA abundances from the original 20,21-mer population, and another 10,000 datasets that mirrored vsRNA abundances (of both polarities) from the original 24,25,26-mer population were generated. The percent of randomly generated datasets wherein 1–2 nt 3′overhang duplexes were as abundant as in the original 20,21-mer population, or as depleted as in the 24,25,26-mer population was calculated (for example, FDR of 0.01% indicates that we were unable to detect percent incidence of duplexes similar to what was observed, in at least 10,000 randomly-generated datasets).
Cultures of HCVrep cells were grown to ∼80% confluency in 10-cm. plates, and transfected with 20ug each of FLAG/HA-tagged Ago-1, Ago-2, Ago-3 or Ago-4 (codon-optimized), and harvested 24 hours post-transfection. Cells were washed twice with 4mL PBS, and 0.75mL of ISOB/NP40 (10mM Tris pH 7.9, 0.15M NaCl, 1.5mM MgCl2, 0.8% NP40, proteinase inhibitor at 1 tablet per 13mL solution) was added to each plate. Cell lysates were vortexed, incubated on ice for 20 minutes, spun at 4°C for 10 min at 13,000 rpm, and the supernatants were transferred to new tubes. 10 uL ‘Fake’ (uncoated) Sepharose 4B beads were washed twice with 1.5ml NET-1 buffer (1×TBS-0.2% Tween), and incubated with supernatants at 4°C for 2 hrs. The supernatants post-‘fake-bead’ IP were subsequently incubated with 10uL of washed Anti-Flag M2 beads [Sigma] at 4°C for 4 hrs. Finally, beads were washed 3× with 500uL NET-1 buffer, and RNA was extracted with lysis buffer (mirVana kit, Ambion). Transfections were performed in duplicate: 1 plate was used as input for IPs, while the other was used for cloning of total small RNA populations. For Mock-IPs, lysates from mock-transfected cells were used as described above.
The loading control was a 5.7 kb region of the pGEM plasmid carrying the Poliovirus 1 genome (pGEM-PV1), obtained by digestion of the plasmid with HindIII and AgeI. Equivalent amounts of RNA (as measured using a NanoDrop) from each sample were run on a 1% agarose-formaldehyde gel, and transferred under basic conditions to a Hybond+ membrane. The loading control was visualized by staining the blot with methylene blue. 60-mer DNA probes to specifically detect either the positive strand (AF-PP-350: GTCACCGCTTGTAGAATTGTCATTGCCCTGTTGATGTTCCTTTCTGTTTGAACCTGGCTG & AF-PP-351: TCATCTATGGTTTGCCGATACGTGGTGTTGCTAATCCATGGCACT ACCATAGTACATGAG), or the negative strand (AF-PP-84: TTCACGGGTACGTTC ACTCCTGACAACAACCAGACATCACCTGCCCGCAGGTTCTGCCCG, AF-PP-86: ATTC GGACACCAAAACAAAGCGGTGTACACTGCAGGTTACAAAATTTGCAACTACCACTT) were end-labeled with 32P, and were successively used on the same blot (the blot was stripped in between the two hybridizations). The blots were hybridized and washed at 50°C, and exposed using a phosphorimager screen.
|
10.1371/journal.pgen.1002940 | Transcriptional Repression of Hox Genes by C. elegans HP1/HPL and H1/HIS-24 | Elucidation of the biological role of linker histone (H1) and heterochromatin protein 1 (HP1) in mammals has been difficult owing to the existence of a least 11 distinct H1 and three HP1 subtypes in mice. Caenorhabditis elegans possesses two HP1 homologues (HPL-1 and HPL-2) and eight H1 variants. Remarkably, one of eight H1 variants, HIS-24, is important for C. elegans development. Therefore we decided to analyse in parallel the transcriptional profiles of HIS-24, HPL-1/-2 deficient animals, and their phenotype, since hpl-1, hpl-2, and his-24 deficient nematodes are viable. Global transcriptional analysis of the double and triple mutants revealed that HPL proteins and HIS-24 play gene-specific roles, rather than a general repressive function. We showed that HIS-24 acts synergistically with HPL to allow normal reproduction, somatic gonad development, and vulval cell fate decision. Furthermore, the hpl-2; his-24 double mutant animals displayed abnormal development of the male tail and ectopic expression of C. elegans HOM-C/Hox genes (egl-5 and mab-5), which are involved in the developmental patterning of male mating structures. We found that HPL-2 and the methylated form of HIS-24 specifically interact with the histone H3 K27 region in the trimethylated state, and HIS-24 associates with the egl-5 and mab-5 genes. Our results establish the interplay between HPL-1/-2 and HIS-24 proteins in the regulation of positional identity in C. elegans males.
| Linker histone (H1) and heterochromatin protein 1 (HP1) play central roles in the formation of higher-order chromatin structure and gene expression. Recent studies have shown a physical interaction between H1 and HP1; however, the biological role of histone H1 and HP1 is not well understood. Additionally, the function of HP1 and H1 isoform interactions in any organism has not been addressed, mostly due to the lack of knockout alleles. Here, we investigate the role of HP1 and H1 in development using the nematode C. elegans as a model system. We focus on the underlying molecular mechanisms of gene co-regulation by H1 and HP1. We show that the loss of both HP1 and H1 alters the expression of a small subset of genes. C. elegans HP1 and H1 have an overlapping function in the same or parallel pathways where they regulate a shared target, the Hox genes.
| Linker histone H1 and heterochromatin protein HP1 are involved in numerous processes ranging from stabilizing heterochromatin condensation to the regulation of gene expression [1]–[5]. As has been reported, a methylation mark on vertebrate histone H1 is specifically recognized by the chromodomain of HP1. However, the exact biological role of HP1 binding to linker histone has not been determined [6].
The functions of HP1 and H1 proteins are mainly dependent on the cell type in which particular variants are expressed. Although the number of H1 (11) and HP1 variants (3) presents difficulties in studying the effect of H1 and HP1 depletion in mice, some data has emerged [3], [7]–[10]. For example, loss of HP1β results in defective development of neuromuscular junctions and the cerebral cortex [10], whereas depletion of three of eleven H1 genes causes lethality connected with a very broad range of defects in mice [11]–[12]. In ES cells, the lack of three somatic H1 variants leads to changes in nucleosome spacing and local chromatin compaction, and this is correlated with decreased levels of H3K27 trimethylation [11]. Additionally, H1 is necessary to establish and maintain the DNA methylation pattern in a subset of genes including the reproductive homeobox (Rhox) gene cluster [13].
C. elegans possesses eight linker histone variants and two HP1 homologues, HPL-1 and HPL-2 [14]–[16]. Mutation of hpl-2 results in defective vulval and germline development at elevated temperatures [15]–[17]. hpl-1, in contrast to hpl-2, does not have visible effects on C. elegans development at different temperatures, however, hpl-1 acts redundantly with hpl-2 to control larval development, somatic gonad development and vulval cell fate determination [17]. Our previous study revealed that HPL-1 recognizes the linker histone variant HIS-24 when it is mono-methylated at lysine 14 (HIS-24K14me1), similar to the situation in vertebrates [16]. Additionally, we showed that HIS-24 interacts with H3K27me3 [18]. The H3K27me3 modification correlates with a repressive chromatin state that inhibits expression of many developmentally regulated genes. This is consistent with studies of Hox loci demonstrating that enrichment of H3K27me3 recruits the binding of Polycomb group proteins (PcG) [19].
The Hox genes encode conserved homeodomain-containing transcription factors that control the positional identities of cells along the anterior–posterior axis [20]–[21]. The expression pattern of Hox genes appears to be regulated by two evolutionarily conserved PcG complexes, the ESC/E(Z) complex and the PRC1 complex. Both have been identified in flies and mammals and are linked to modulation of repressive chromatin structures [21]. The C. elegans Hox cluster consisting of lin-39, ceh-13, mab-5 and egl-5 (orthologs of Drosophila Scr, labial, ftz and Abd-B, respectively) is quite degenerated in comparison to Hox clusters in other species [22] but, as in mammals, is also globally repressed by Polycomb group (PcG) proteins [20], [23]. Mutations in mes-2 and mes-6, which encode the C. elegans ESC/E(Z) complex, result in ectopic expression of Hox genes [23]. A similar phenotype has also been observed in the absence of sop-2 or sor-1 genes. SOP-2 and SOR-1 form another C. elegans PcG-like complex which shares many structural and functional properties with the Drosophila PRC1, and is involved in the global repression of Hox gene expression. Loss of sop-2 and sor-1 results in gross homeotic transformations [24]–[25].
To elucidate the function of H1 and HP1 related proteins in C. elegans, we decided to generate double and triple mutants, since hpl-1, hpl-2 and his-24 deficient nematodes are viable, and since HIS-24K14me1 is recognized by HPL-1 [16]–[17], [26]. We performed global transcriptional analyses of single, double and triple mutant animals, and we found that HPL-1/-2 and HIS-24 regulate a relatively small number of genes. We provide evidence that the methylated form of HIS-24 (HIS-24K14me1) and HPL-2 are involved in the regulation of mab-5 and egl-5 expression by binding to H3K27me3, although HIS-24K14me1 does not interact with HPL-2 [16]. Furthermore, we observed that HIS-24 and HPL-2 act in parallel pathway as MES (PcG) proteins, and loss of their activity causes defects of male tail structures. Overall, our data suggest a common and dual role for C. elegans H1 and HP1, functioning both as chromatin architectural proteins and at the same time as modifiers of a small subset of genes. Furthermore, we provide the first direct evidence for redundant functions of H1 and HP1 in metazoan development.
C. elegans contains two related HP1 proteins (HPL) and eight linker histone variants [14]–[15]. Only one of the eight linker histone variants, HIS-24 is important for germline development, with its absence resulting in reduced fertility and de-repression of extrachromosomal transgenic arrays in the germline [14]. As we previously reported, the absence of HIS-24 did not affect protein levels of the other histone variants, in contrast to the mammalian H1 subtypes which are sufficient to compensate for the loss of a single linker histone [7], [16]. Furthermore, we showed that C. elegans heterochromatin protein 1 variant, HPL-1 recognizes and binds the methylated form of HIS-24 [16]. Given the physical interaction of HPL-1 with HIS-24 mono-methylated at lysine 14 and their role in chromatin silencing and germline developmental processes [15]–[17], we decided to study HPL and HIS-24 function in transcriptional regulation in C. elegans. It was of great interest to determine how the HPL subtypes (HPL-1 and HPL-2) and HIS-24 affect gene expression. To determine the contribution of HIS-24 and HPL-1/-2 to the control of gene transcription, we compared the gene-expression profiles of single null mutations in the hpl-1, his-24 and hpl-2 as well as profiles of hpl-1his-24, and hpl-2; his-24 double, and hpl-2; hpl-1his-24 triple mutant animals in L4 larval stages grown at 21°C. We decided to use L4 larval stages because HIS-24 is the most abundant linker histone H1 variant at this stage according to mass spectrometry-based protein expression data (Figure 1).
By microarray we observed very few changes in the gene expression profiles of either single, double, or triple mutants when compared with wild type animals at L4 larval stages. Among the 16,278 target probe sets assayed, we identified only modest changes in expression of just a small number of genes (Figure 2A–2H, Table 1). The majority of genes exhibiting changes were upregulated (6.5%) in the absence of the three heterochromatin components HIS-24, HPL-1 and HPL-2, in contrast to 3.7% downregulated genes from a total of 16,278 genes analyzed (FDR<0.05) suggesting that HPL-1/-2 and HIS-24 are not global repressors of transcriptional activity (Table 1). The deletion of both hpl-1 and hpl-2 genes caused up-regulation of 4.5% genes and downregulation of 2.1% of a total 16,278 genes when compared to wild type (WT) animals.
As previously reported, HPL-2 binds to HIS-24K14me1 through its association with HPL-1, and the heterochromatin proteins HPL-1 and HPL-2 play redundant roles in C. elegans development [16]–[17]. Considering these observations we compared transcriptional profiles between hpl-2 (tm1489); hpl-1(tm1624) double mutants and hpl-2 (tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant animals. We found that 464 up-regulated (2.9% of 16,278) and 195 down-regulated (1.2% of 16,278) genes were commonly affected (FDR<0.05; p-value<0.000001 for all pair-wise comparisons by hypergeometric tests) (Figure 2, Table S1). Among the 464 up-regulated genes we identified some significantly enriched in GO terms associated with growth regulation (Fisher exact test (FET) P = 4×10−6), determination of adult life span (FET P = 2×10−6), locomotion (FET P = 0,003), protein phosphorylation (FET P = 0,04), reproduction (FET P = 0,05) and lipid storage (FET P = 0,05). The 195 genes that are down-regulated are enriched in GO terms associated with oxidation reduction (FET P = 0,003), embryonic development (FET P = 0,002) and metabolic process (FET P = 0,04). We identified common response proteins including heat shock proteins (HSP-12.3, -12.6, -16.2 and -17), enzymes (cytochromes) of the P450 family involved in protection against toxins (CYP-13A12, CYP-33C4, CYP-33D3, CYP-34A2, CYP-34A4 and CYP-34A9), metabolic enzymes such as the fatty acid-coenzyme A (CoA) synthetase ACS-1 and the fatty acid/retinol binding proteins FAR-5, -7 (Table S1). Furthermore, we observed the induction of oxidative stress proteins such as glutathione S-transferases (GST) and genes commonly associated with increased stress resistance – for example, the mitochondrial sod-3 superoxide dismutase gene (Table S1).
In conclusion, deletion of the different HPL variants and HIS-24 caused an alteration in the expression of a limited number of genes, different in each HPL variant and HIS-24. Most of the genes are affected by a single HPL variant and HIS-24, supporting the theory that HPL isoforms or HIS-24 play specific roles in gene expression. Nonetheless, a proportion of genes are altered by more than one HPL variant as well as HIS-24, suggesting redundant roles for HIS-24 and HPL variants, and for HPL-1/-2 may also exist.
In parallel to microarray analysis we investigated the biological role of HIS-24 and HPL proteins in C. elegans. For morphological defects we scored hpl-1(tm1624) his-24(ok1024), and hpl-2(tm1489); his-24(ok1024) double mutants as well as hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant animals. In particular, we focused on germline nuclei morphology, hermaphrodite vulval development and the somatic patterning of the male tail since these tissues are known to be affected by mutations in chromatin factors, and HPL-2 influences vulval cell fate specification in the synMuv (synthetic multivulva) pathway [14]–[15], [27]. We found that the deletion of hpl-2(tm1489) together with his-24(ok1024) results in synergistic non-lethal defects of vulval cell fate specification (everted vulva, multivulva) and sterility at 21°C, and at 25°C (Table 2). While the observed phenotypic effects at 21°C were minor in contrast to the situation at 25°C, it is tempting to speculate that the effects can be also modulated through unknown mechanisms, environmental cues (temperature), which in itself may also lead to significant side-effects. Additionally, decreased brood sizes were observed in hpl-2(tm1489); his-24(ok1024) double and hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant animals grown at 21°C (Figure 3). The brood size of the hpl-2(tm1489); his-24(ok1024) was strongly decreased by 35% of wild type worms, and was further decreased to about 50% in the hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant animals (Figure 3). These results were consistent with our microarray data analysis that revealed differential expression of genes involved in the embryonic development or reproduction (Table S1). Furthermore, we observed several defects in the morphology of the somatic gonad of hpl-1(tm1624) his-24(ok1024) double mutant animals grown at 21°C. In wild type, single mutant and hpl-2(tm1489); his-24(ok1024) double mutant the gonad arms form an U-shaped structure (Figure 4A–4D). In contrast, in the double mutant hpl-1(tm1624) his-24(ok1024) 25% of gonad arms (161 of 642) form a loop (Figure 4E). These results suggest that both proteins HIS-24 and HPL-1 are involved in the somatic gonad development whereas HIS-24 and HPL-2 influence vulva cell fate specification and reproduction (Table 2).
Since HPL-2 and HIS-24 are required for germline development and for the chromatin based germline-specific silencing mechanism [14]–[15], [26], we asked whether they influence the structure of nuclei. In-depth analysis revealed that the germline nuclei of hpl-2(tm1489); his-24(ok1024) double mutants differ in size and morphology when compared to single mutants or to wild type worms grown at 21°C (Figure 4F–4I, 4L). The observed chromatin of 86% of gonad arms (36 of 42) had a more open, relaxed structure suggesting that HIS-24 and HPL-2 play a function in chromatin condensation in the germline (Figure 4J, 4M). To assess the specific requirements for HIS-24 among the H1 isoforms, we also tested hpl-2(tm1489);hil-3(ok1556) double mutant strain to determine if the observed changes in the chromatin compaction is linker histone variant specific (Figure 4K). As shown, loss of hpl-2 and linker histone variant hil-3 did not cause defects in chromatin compaction in contrast to hpl-2; his-24 strain. In addition, we also did not observe involvement of HPL-2 and HIL-3 on brood size (Figure 3).
To determine if the loss of HIS-24 and HPL proteins also influence chromatin histone modifications as well as core histone H3 level, we performed western blot analysis of mutant animals. No gross changes were observed in the methylation and core histone H3 levels using antibodies directed against H3K9me3, H3K27me3, and H3 (Figure S1). In addition, we did not detect changes in chromatin modification marks on a cellular level by immunofluorescence (data not shown) indicating that the observed effects of chromatin compaction are not correlated with alterations of histone modifications in hpl-2(tm1489); his-24(ok1024) double mutant animals.
Loss of hpl-1, -2 and his-24 function results in changes of transcriptional regulation of genes encoding nuclear hormone receptor family genes (nhr-60, nhr-156), transcription factors (miz-1, zip-3, zip-8, madf-2), homeobox ceh-82 and homeodomain lim-7 genes (Table S1). Moreover, hpl-2 regulates lin-39 Hox gene expression in vulval precursor cells (VPCs) [27]. Therefore we tested whether hpl-1, -2 and his-24 genes are involved in the regulation of Hox gene expression during the somatic patterning of the male tail.
The wild type male tail possesses nine pairs of bilateral sensory rays that function in locating and mating with hermaphrodites. Normally, the posterior hypodermal blast cells V5 and V6 produce six pairs of rays (ray 1 to ray 6), while the blast cell T gives rise to the three rays (rays 7–9) [23]–[25]. We found that mutations in both his-24 and hpl-2 (37%, 51 of 73 males with defected rays) as well as in his-24, hpl-1 and hpl-2 (83%, 76 of 107 males with defected rays) cause abnormalities in patterning of blast cells V that result in fused and atypical (under-developed) rays, while the single and hpl-1; hpl-2 and hpl-1 his-24 double mutations have normal development of rays (Table S2, Figure 5A–5E, 5G). Although hpl-1 mutation alone or in combination with his-24 or hpl-2 had no visible effect on the male tail at 21°C (Figure 5C, 5G–5H), it appeared to be partially redundant in combination with hpl-2 and his-24 double mutations. As Figure 5J and Table S2 show, the number of under-developed rays is significantly increased (up to 42%, 39 of 107 males) in the hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant compared to the hpl-2(tm1489); his-24(ok1024) double mutant males (13%, 18 of 73 males) (Figure 5I). This synergism suggests that hpl-1 only in combination with his-24 and hpl-2 plays functions in the patterning of the male tail. We also tested hil-3; hpl-2 double mutant animals for the mail tail phenotype. We did not observe any defects in the patterning of the male tail of hil-3; hpl-2 double mutant animals in contrast to hpl-2; his-24 animals suggesting that HIS-24 (in combination with HPL-2) specifically affects the patterning of the mating structures in C. elegans (Figure 5F–5I).
In agreement with previous observations we analyzed the ability of his-24, hpl-1 and hpl-2 genes to regulate mab-5 and egl-5 expression [28]–[29]. Interestingly, these two Hox genes are required for V ray development [23] and mab-5 was slightly up-regulated in our microarray analysis of hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) mutant animals (data not shown).
We compared expression of egl-5::gfp and mab-5::gfp reporter genes in wild type animals and in combination with his-24(ok1024), hpl-1(tm1624) and hpl-2(tm1489) background mutations (Figure 6). We observed that mab-5::gfp reporter is ectopically expressed in approximately 30% of early L3 stage of hpl-2(tm1489); his-24(ok1024) double mutant males scored (n = 100) (Figure 6A, 6B). Altered expression of this reporter was also detected in adult males. Similarly, about 80% of L3 stage of hpl-2(tm1489); his-24(ok1024) double mutant males (n = 100) displayed ectopic expression of EGL-5::GFP protein in two daughters of ray precursors anterior to R4, R5 and R6 sublineages (Figure 6C–6E). We did not observe any significant enhancement of the ectopic expression of mab-5 and egl-5 Hox genes in hpl-1 depleted hpl-2(tm1489); his-24(ok1024) double mutant animals. For the crossing with mab-5::gfp transgenic strain we did not use triple mutant animals due to hpl-1 his-24; hpl-2 phenotype (sterile worms, worms with everted vulva or multivulva; Table 2).
To verify HPL-1 depletion directly and to examine the extent of HPL-1 knockdown we tested the hpl-1 depleted hpl-2; his-24 mutant animals for presence of HPL-1 on the western blot. We found that hpl-1RNAi strongly reduces HPL-1 level compared to the controls (Figure S2).
Since mutations in hpl-2 and his-24 affect transgene expression in C. elegans [14]–[15] we assessed the expression level of the endogenous EGL-5 in hpl-2; his-24 double mutant males. Western blot of hpl-2; his-24 double mutant males probed with EGL-5 antibody revealed an increased level of endogenous EGL-5 protein of predicted size (26 kDa) compared to EGL-5 level of wild type C. elegans and egl-5::gfp transgenic line (Figure 6F) [29]. Altogether, these results suggest that HIS-24 and HPL-2 silence the Hox gene cluster, either by general repression of the transcriptional activity, or through a specific biochemical and structural function in Hox gene silencing.
Since HIS-24 and HPL-2 are required for inhibiting the ectopic expression of mab-5 and egl-5 Hox genes, we tested if HIS-24 and HPL-2 bind directly to their promoters in vivo and therefore regulate egl-5 and mab-5 transcription. The primer sets used for quantitative ChIP-PCR (qChIP-PCR) analysis were directed to the promoters, introns and 3′UTR regions of mab-5 and egl-5 genes. Remarkably, mab-5 and egl-5 are tightly clustered on chromosome III, suggesting that chromatin structure coordinately regulates the expression of these genes (Figure 7A). qChIP-PCR analysis revealed that HIS-24 is indeed associated with the promoters and introns of mab-5 and egl-5 genes (Figure 7B). In contrast, we did not see any HIS-24 binding to 3′UTR regions (Figure 7B). However they are occupied by H3 (Figure 7D). As shown, the anti-HIS-24 antibody binds with higher affinity to egl-5 and mab-5 genes than the anti-HIL-4 antibody, which is cross-reactive to C. elegans linker histone variants [14] (Figure 7C). Next, to verify the specificity of the HIS-24 binding to Hox genes, we tested the HIS-24 binding to mab-5 gene ectopically expressed in sor-1 background mutation. As previously reported, SOR-1 (together with SOP-2) shares many structural and functional properties with the PRC1 complex, and is involved in the global repression of egl-5 or mab-5 Hox gene expression [25]. As shown, we detected a significantly decreased level of HIS-24 at this region compared to the situation in wild type animals, implicating that HIS-24 enables mab-5 transcriptional repression, thereby influencing its expression (Figure 7D). Additionally, we observed lower levels of histone H3 occupancy at the mab-5 promoter in sor-1 background mutation than in wild type animals, suggesting that the difference in H3 levels could be due to the nucleosome free region that forms at high levels of expression (Figure 7D). In addition, mab-5 promoter and intron regions in the his-24 mutant animals showed decreased enrichment of the histone H3 than in wild type animals, suggesting that binding of H3 and HIS-24 can be positively correlated at regulatory regions. In comparison, the H3 changes at 3′UTR region of mab-5 in sor-1 and his-24 background mutation were relatively mild than in wild type animals (Figure 7D).
Unfortunately, we have failed so far to detect HPL-2 at this region using direct ChIP approach.
Hox genes are transcriptionally repressed by the evolutionally conserved Polycomb group (PcG) proteins through the H3K27me3 mark in a lineage specific fashion [30]–[31]. In Drosophila, a member of the Polycomb group (PcG), the H3K27 histone methyltransferase E(Z) has been identified as a stable repressor of Hox genes [32]. In C. elegans, orthologs of the PcG chromatin repressors E(Z) and ESC, namely MES-2 and MES-6 influence expression of Hox genes and male tail development [23]. Since Polycomb group (PcG) proteins (MES-2/3/6 complex) are involved in the repression of Hox genes, we performed genetic epistasis analysis of mes-2- and mes-3-depleted triple mutant animals [23], [33]. Interestingly, hpl-2(tm1489); his-24(ok1024) double as well as hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant males on mes-2 or mes-3 feeding plates showed an increased number of ectopic rays (∼2-fold) and defective rays in comparison to mes-3- or mes-2 - depleted double mutant males (Figure 8A, 8C; Table S2). As shown, loss of HPL and HIS-24 together with depletion of mes-2 or mes-3 resulted in additive defects implying that HPL and HIS-24 act in parallel pathway as MES-2 or MES-3.
We also phenotyped hpl-2(tm1489); his-24(ok1024) double and hpl-2(tm1489); hpl-1(tm1624) his-24(ok1024) triple mutant males on sop-2 feeding plates (Figure 8B, 8C, Table S2). As previously reported, SOP-2 forms a novel PcG-like complex that may function analogously to PRC1 in C. elegans and regulates expression of Hox genes [25]. Homologs of SOR-1 and SOP-2 are not found in other organisms, including even the very closely related C. briggsae suggesting a C. elegans specific mechanism on an essential global gene regulatory system [25]. Remarkably, we did not observe any influence of SOP-2 depletion in the hpl-2; his-24 double and hpl-2; hpl-1 his-24 triple mutant background suggesting that sop-2 appears to be epistatic to hpl-2; his-24 deletion.
Recently, we have reported that HIS-24 specifically interacts with H3K27 trimethylated and H3K27 unmodified peptides [18]. While HPL-1 and HPL-2 were able to pull down native HIS-24K14me1, and HPL-2 failed to bind either modified or unmodified HIS-24 peptides in vitro, we asked whether HPL-2 and HIS-24K14me1 repress the transcription of egl-5 and mab-5 genes by binding to H3K27me3 [16], [18].
By peptide pull down assay (PD) we observed that HIS-24K14me1 interacts preferentially with H3K27me3 peptide when compared to the unmodified, mono- or di-methylated H3K27 peptides, and conversely, native H3K27me3 binds only the methylated form of HIS-24 peptide (Figure 9A). Furthermore, we found strong preference of HPL-2 for the trimethylated form of H3K27, as well as for H3K27me2 and H3K9me2/3 as previously reported (Figure 9A) [16]. No interaction was observed between H3K9me0/1 or H3K27me0/1. We confirmed the results obtained from peptide pull down (PD) by an immunoprecipitation (IP) approach using antibodies raised against different chromatin modification marks and lysates of wild type animals (Figure 9B). Additionally, we were able to pull down native H3K27me3 using a GFP-antibody directed against GFP-tagged HPL-2 and HIS-24 (Figure 9C). As a control we used GFP expressed protein under the his-24 promoter to demonstrate the specificity of HPL-2 and HIS-24 binding to H3K27me3 (Figure 9C). To confirm that HPL-2 and HIS-24 indeed display H3K27me3 binding, we expressed HPL-2 and HIS-24 in E. coli. We did not detect the interaction of HPL-2 with H3K27me3 in contrast to HIS-24, suggesting that additional factors (transcription factors, RNAi machinery, post-translational modifications of HPL-2) are involved in the mediation of HPL-2 binding to H3K27me3 (Figure 9D, 9F). In the case of HIS-24 we detected strong preference for H3K27me peptides apart from H3K27me1 (Figure 9D). The differences in the binding to H3K27me3 between bacterially expressed HIS-24 and native HIS-24 can be explained by the fact that bacterially expressed proteins are not methylated and only the methylated form of HIS-24 binds specifically the H3K27me3. Finally, to exclude that the binding of HPL-2 to H3K27me3 takes place via interaction with the C. elegans HIS-24, we repeated the pull downs using extracts obtained from his-24(ok1024) mutant animals (Figure 9E). We detected a preference of HPL-2 for H3K27me3 independently of HIS-24 however this binding was reduced compared to binding of HPL-2 to H3K27me3 in the presence of HIS-24 (Figure 9B, 9E).
To assess whether the methylated form of HIS-24 has a causal role in the observed changes of the male tail morphology, we generated his-24::gfp and his-24K14A::gfp transgenic worms in the hpl-2(tm1489); his-24(ok1024) mutant background. We observed that the restoration of HIS-24 levels by expression of HIS-24::GFP rescued the male phenotype and the fused/missing rays were down nearly to zero in the transgenic line (Figure 10). Importantly, the nonmethylatable HIS-24K14A::GFP mutant failed to rescue the wild type rays development in hpl-2; his-24 animals, suggesting that HIS-24 methylation at lysine 14 is necessary to regulate male tail development (Figure 10B, 10D, 10E). These results also imply that, at 21°C, hpl-2 and his-24 play a redundant role in the regulation of positional identity in the C. elegans males. Importantly, the analysis of transgene expression at the cellular level by immunostaining and immunoblotting of the rescued hpl-2(tm1489); his-24(ok1024) animals verified that the exogenous HIS-24K14A::GFP mutated form was expressed at a level comparable to that in animals carrying HIS-24::GFP wild type form (Figure 10C).
HP1 and H1 are heterochromatin components that are believed to be associated with global repression of transcriptional activity 4–5. Surprisingly, our microarray analysis showed that H1 and HP1 play more dynamic and gene-specific roles in the roundworm C. elegans. They grossly affect only a few genes and can have an overlapping function in the same or parallel pathways where they regulate common target genes.
In particular, we found that HIS-24 and HPL-2 can regulate a shared target, the Hox genes. Although, the C. elegans homeobox genes (egl-5, mab-5) are silenced by mechanisms involving H3K27 trimethylation, we showed that the methylated form of HIS-24 and HPL-2 can also serve as essential protein components in establishing and/or maintaining the repressive chromatin state at the selected Hox genes, presumably through their binding to H3K27me3 (Figure 9F).
Our microarray analyses support a role of H1 and HP1 in specific gene regulation, rather than a general repressive function [34]–[36]. Despite global changes in chromatin compaction and synergism of HIS-24 and HPL in aspects of many developmental processes we observed very few and slight changes in gene expression profile of mutants when compared with wild type animals. We detected a set of shared up- and down-regulated genes by HIS-24 and HPL suggesting that redundant roles for HIS-24 and HPL also exist. The relatively small number of regulated genes in observed triple mutant animals may indicate that HPL proteins and HIS-24 serve to fine-tune the regulation of key genes during development or differentiation. This model can be explained by the fact that the sequential arrangement of the linker histone HIS-24 and HPL-2 on the chromatin fibre might influence higher-order chromatin structure and effect nucleosome positioning, and stability [36]. It is possible that the different HPL subtypes and HIS-24 confer subtle differences in the properties of the chromatin fiber which allow for quantitative modulation of gene expression [34], [35]. Although the changes in gene transcription are subtle, we think that even 1.5-fold differences in expression can contribute to the marked phenotypic consequences we observed.
We demonstrated that HIS-24K14me1, together with HPL-2, has a causal role in transcriptional silencing of egl-5 and mab-5. We propose that HPL-2 and HIS-24K14me1 may serve as essential protein components in establishing and/or maintaining the repressive chromatin state at the selected Hox genes through their interactions with H3K27me3. While we did not observe any phenotypic effects on male tail development either in hpl-2; hpl-1 nor in hpl-1 his-24 background, we speculate that HPL-2 acts redundantly with HIS-24K14me1 to regulate the positional identity in the C. elegans males. Loss of the two heterochromatin components, HIS-24K14me1 and HPL-2, causes significant changes in chromatin structure affecting Hox gene expression in C. elegans. However, since no interaction of HPL-2 and HIS-24K14me1 was observed in immunoprecipitation experiments, it is possible that HPL-2 together with HIS-24K14me1 might be a part of the same protein group involved in the regulation of Hox gene expression. The high degree of redundancy between his-24 and hpl-2 in Hox gene regulation might indicate that these two proteins are the only readers acting in parallel to perform the same role in translating the effects of histone H3K27 trimethylation. However, since we have failed so far to detect HPL-2 at the Hox gene region using direct ChIP approach, it is possible that the mechanisms by which HPL-2 regulates mab-5 and egl-5 might be indirect, involve intermediate factors (RNAi machinery, transcription factors) and depend on an architectural level in the cellular context.
In mammals, H1 regulates Hox gene activation by promoting DNA demethylation [13]. Although C. elegans does not possess methylated DNA, we speculate that H1 can still influence Hox gene regulation and, together with HPL-2, regulate Hox gene expression as a part of the PcG silencing complex. The interaction of HPL-2 and HIS-24K14me1 with H3K27me3 can regulate the Hox gene in parallel pathway as MES-2 or MES-3, and can be directed to specific parts of the genome. Notably C. elegans HP1/HPL-2 does not follow the H3K9me2/me3 code [37]–[41] but it is sufficient to recognize, and to bind H3K27me2/me3. Remarkably, HIS-24 is required for optimal HPL-2 binding to H3K27me3 in vivo.
Interestingly, some PcG proteins containing a chromodomain similar to that found in C. elegans HPL-2 and mammalian HP1s have been shown to bind H3K27me3 [30], [42].
Overall, these and our previous results implicate that HPL and HIS-24 share some common functions even though there are differences among these proteins [16]–[17], [26]. We conclude that a combination of the H3K27me3 methylation mark, HPL-2 and HIS-24K14me1 could be a major factor in the establishment of stable patterns of selected homeotic gene expression.
Nematodes strains were cultured and genetically manipulated as previously described [43]. The Bristol strain (N2) was used as wild type. The following strains, obtained from the Caenorhabditis Genetics Center (CGC), were used in this study: his-24(ok1024)X, hil-3(ok1556)X (both strains outcrossed 1×), hpl-1(tm1624)X (outcrossed 4×), hpl-2(tm1489)III (outcrossed 4×). Transgenic strain (transcriptional reporter) expressing GFP under the control of the his-24 promoter was kindly provided by BC C. elegans Gene Expression Consortium, Canada. The double mutants hpl-1(tm1624)X his-24(ok1024)X, hpl-2(tm1489)III; his-24(ok1024)X, hpl-2(tm1489)III; hil-3(ok1556)X and the triple mutant strain hpl-2(tm1489)III; hpl-1(tm1624)X his-24(ok1024)X were generated by crossing. his-24::gfp (stable integrated EC602 strain [26]) and his-24K14A::gfp transgenic strains were crossed with the hpl-2(tm1489)III; his-24(ok1024)X. The generation of his-24K14A::gfp transgenic strain was previously described [16].
For the reporter gene analysis following transgenic strains: EM599 [egl-5::gfp; him-5(e1490)V; lin-15B(n765)X; bxIs13], OP27 [unc-119(ed3)III; wgIs27], OP54 [unc-119(ed3)III; wgIs54] and HZ111 [mab-5::gfp; muIs16 II; sor-1(bp1)/qC1 dpy-19(e1259) glp-1(q339)III; him-5(e1490)V], kindly provided by CCG, were used.
The brood size was scored as previously described [14].
All C. elegans strains were maintained at 15°C or at 21°C, unless otherwise specified.
C. elegans H1 extraction was performed as previously described [16].
Worms from wild type strain and the mutant worms were fixed and stained as previously described [26]. Gonads of worms were stained with fluorescent dye 4′,6′-diamidino-2-phenylindole (DAPI) diluted 1∶1000. The slides were mounted with Vectashield Mounting Medium and analyzed by using Leica DMI 6000 microscope.
Microarray analysis from two biological replicates was performed as previously described [16], [44]. In brief, 80 to 100 animals in L4 larval stage raised at 21°C were used. The gene expression fold change was calculated from the duplicate microarray data. The fold change cut-off was 1.5 from 2 biological replicates.
Abnormalities of rays were identified in single, double and triple mutant males in comparison to the wild type worms. Animals were transferred on agar pads (2% agarose) and examined with differential interference contrast (DIC), using Leica DMI 6000 microscope. The number of rays, their position in relation to the anterior-posterior body axis and their shape served as basics of the analysis. Rays which were found outside of their normal formation region were defined as ectopic.
RNAi feeding experiment was performed in 50 mm NGM feeding plates (NGM plates with 100 µg/ml ampicillin, 1 mM IPTG). him-14 (RNAi) (CGC, USA), hpl-1 (RNAi), mes-2 (RNAi) and mes-3 (RNAi) bacterial clones (Sanger Institute, UK) were grown overnight at 37°C in LB medium with 100 mg/ml ampicillin and were spotted onto 50 mm NGM plates. Mixed stage L3 and L4 mutant larval worms were transferred onto feeding plates and incubated at 21°C through several generations. Males were examined on the agar pads using Leica DMI 6000 microscope. Male progeny were scored for the presence of ectopic, under-developed rays and/or ray fusions.
L3 stage and adult animals from each line were mounted on the agar pads and examined under Leica DMI 6000 microscope. Males were scored for the presence of ectopic EGL-5::GFP or MAB-5::GFP expression.
ChIP was performed as previously described [45] with several modifications. Mixed stage L4 and adult worms were homogenized in ice-cold FA lysis buffer (50 mM HEPES/KOH pH 7.5, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate; 150 mM NaCl) with complete protease inhibitor cocktail (Roche) and 1% Triton X-100 using liquid nitrogen. Worm lysate was sonicated with a Branson Digital Sonifier using following settings: 30% amplitude for 3 min total. Protein concentration of the extract was determined by the Coomassie Plus (Bradford) Protein Assay. Worm extract was incubated with the following antibodies: anti-H3 (Abcam 1791), anti-H3K27me2 (Upstate 07-322), anti-H3K27me3 (kindly provided by T. Jenuwein), anti-H3K9me2 (Abcam 1220), anti-H3K4me3 (Abcam 1012), anti-GFP (Roche) and anti-HIS-24. Proteins were immunoprecipitated using G-agarose beads (Pierce). mab-5 and egl-5 genes were detected by qPCR using iCycler iQ™ Multi- Color real time PCR detection system (Bio-Rad). Primer sequences are available on request.
Peptide pull downs were performed as previously described [46]. 10 µg of each biotinylated peptide was coupled to streptavidin- agarose beads (Pierce). For peptide binding experiments following peptides were used: H3 mono-, di- or trimethylated at K9, H3 mono-, di- or trimethylated at K27, H3 unmethylated at K27, HIS-24 monomethylated at K14 and HIS-24 unmethylated at K14. Peptides were generated by Squarix (Germany). Worm extracts were incubated for 2 h at 4°C with the beads (constant rotation). Beads were washed six times with PD 150 buffer (20 mM Hepes pH 7.9, 150 mM KCl, 0.2% Triton-X 100, complete protease inhibitor cocktail, 20% glycerol). Bounded proteins were separated on gradient NuPAGE SDS gel (4–12%).
C. elegans lysates were prepared and analyzed by western blot as previously described [16], [18].
Mixed populations of L4 worms carrying the hpl-2::gfp transgene or wild type worms were homogenized [47]. About 1.5 mg of total precleared protein was incubated with following antibodies GFP-TrapR –A beads (Chromotek, Germany), anti- H3 (Abcam 1791), anti-H3K27me2 (Upstate 07-322), anti-H3K27me3 (kindly provided by T. Jenuwein), anti-H3K9me2 (Abcam 1220), anti-H3K4me3 (Abcam 1012), anti-H3K9me3 (Abcam 8898) or anti-H4K20me3 (Abcam 9053) at 4°C overnight. Next, the complexes were washed six times with PD150 buffer for 5 minutes at 4°C (20 mM Hepes, pH 7.9; 150 mM KCl, 0.2% Triton X-100, 1× Protease Inhibitor (Roche), 20% glycerol). Finally, the immunoprecipitated proteins were resolved on NuPAGE SDS gradient gel (4–12%) and western blotted with antibodies against H3K27me3 (1∶20 000 dilution), HPL-2 (1∶2000 dilution; kindly provided by F. Palladino), HIS-24K14me1 (1∶10000 dilution) and GFP (Roche; 1∶20000 dilution).
The pGEX HPL-2a plasmid (kindly provided by F. Palladino) and HIS-24 pet3a plasmid were expressed in E. coli BL21(DE3) and the recombinant proteins were used for the peptide pull down assay.
The microarray data can be found in the Gene Expression Omnibus (GEO) of NCBI through accession number GSE33339.
|
10.1371/journal.ppat.0030155 | RNAi Screen in Drosophila Cells Reveals the Involvement of the Tom Complex in Chlamydia Infection | Chlamydia spp. are intracellular obligate bacterial pathogens that infect a wide range of host cells. Here, we show that C. caviae enters, replicates, and performs a complete developmental cycle in Drosophila SL2 cells. Using this model system, we have performed a genome-wide RNA interference screen and identified 54 factors that, when depleted, inhibit C. caviae infection. By testing the effect of each candidate's knock down on L. monocytogenes infection, we have identified 31 candidates presumably specific of C. caviae infection. We found factors expected to have an effect on Chlamydia infection, such as heparansulfate glycosaminoglycans and actin and microtubule remodeling factors. We also identified factors that were not previously described as involved in Chlamydia infection. For instance, we identified members of the Tim-Tom complex, a multiprotein complex involved in the recognition and import of nuclear-encoded proteins to the mitochondria, as required for C. caviae infection of Drosophila cells. Finally, we confirmed that depletion of either Tom40 or Tom22 also reduced C. caviae infection in mammalian cells. However, C. trachomatis infection was not affected, suggesting that the mechanism involved is C. caviae specific.
| Chlamydia spp. are intracellular bacterial pathogens that infect a wide range of hosts and cause various diseases, including preventable blindness in developing countries, sexually transmitted disease, and pneumonia. Chlamydia spp. are able to establish their replication niche inside the host cell, residing in a membrane-bound compartment that serves as a protector shield against immune surveillance and antimicrobial agents but also acts as a “filter” to exchange factors with the host cell. Despite the primary importance of Chlamydia for human health, little is known about the mechanisms underlying the infection process. The study of Chlamydia pathogenesis is challenging because Chlamydia spp. are not amenable to genetic manipulation and it is difficult to conduct extensive genetic approaches in the mammalian host. To circumvent these difficulties, we have used Drosophila cells to model Chlamydia infection. We conducted a genome-wide RNA interference screen and identified host factors that, when depleted, reduce Chlamydia infection. Validating our approach, we further showed that the identified factors were also required for infection in mammalian cells. This work will help us better understand the complex interaction between Chlamydia and its host and potentially identify novel targets for therapeutic treatment.
| Chlamydia spp. are Gram-negative, obligate, intracellular bacterial pathogens that infect a wide range of hosts and cause various diseases. Three species infect humans. C. trachomatis is the leading cause of preventable blindness in developing countries [1] and the most common cause of bacterial sexually transmitted disease in developed countries [2]. Infection with C. pneumoniae leads to pneumonia, and in the past 10 years, C. pneumoniae has been implicated in atherosclerosis [3] and Alzheimer disease [4], although the direct links between the bacteria and these diseases is still unclear. C. psittaci infects various animals and is responsible for pneumonia in humans [5]. Many Chlamydia species are recognized as animal pathogens [6]. C. muridarum infects mice and hamsters. C. suis, C. abortus, and C. felis infect swine, ruminants, and house cats, respectively. Finally, infection with C. caviae in guinea pig resembles ocular and genital infections caused by C. trachomatis in humans.
Chlamydia are characterized by a biphasic developmental cycle that occurs exclusively in the host cell. The bacteria alternate between an infectious, metabolically inactive form called elementary body (EB) that is characterized by a condensed nucleoid, and an intracellular, metabolically active form named reticulate body (RB). Once internalized, Chlamydia resides in a membrane-bound compartment, named the inclusion. Shortly after uptake, an uncharacterized switch occurs, leading to the differentiation of EBs into RBs. The RBs then start to replicate until the inclusion occupies a large part of the cytosol of the host cells. At the end of the cycle, which lasts 2 to 3 d depending upon the species, the RBs differentiate back into EBs. The host cell is lysed, leading to the release of EBs and the infection of neighboring cells [7,8].
Both bacterial and host factors contribute to the biogenesis of the inclusion, but little is known about the mechanisms involved. Chlamydia spp. possess a type III secretion system (TTSS) responsible for the secretion of effector proteins in the cytoplasm of the host cell. An example of such effectors is the family of highly hydrophobic Inc proteins. Some of them are present on the surface of the inclusion membrane and are thought, in combination with other bacterial effector proteins, to modify the host cell environment and allow bacterial replication [9–13]. During the cycle, Chlamydia targets various host cell functions in order to establish its replication niche and disseminate from cell to cell [14]. The bacteria acquire amino acids, nucleotides, and other precursors from the host cell. The mechanism of chlamydial entry is not well understood, but among others, heparan sulfate proteoglycans, tyrosine phosphorylation of the bacterial effector Tarp, and activation of small GTPases and signaling pathways leading to actin remodeling are involved in this process [15]. Once internalized, Chlamydia directs the trafficking of the nascent inclusion to a perinuclear localization via a mechanism involving microfilaments, microtubules, and the motor protein dynein [16]. The inclusion does not interact with the endocytic pathway [14,17]. However, it intercepts exocytic vesicles and lipids from the Golgi [18]. Some Rab GTPases are recruited to the inclusion membrane [19], and a recent study suggests that Chlamydia targets host lipid droplets to enhance its intracellular survival and replication [20]. Finally, Chlamydia has the ability to modulate the programmed cell death pathway of infected cells [21,22]. During the early stage of infection, the infected cells are resistant to apoptosis signals but, by the end of Chlamydia developmental cycle, the programmed cell death pathway is induced, presumably to facilitate the release of the bacteria and the initiation of the next round of infection.
In the past few years, Drosophila has been established as a useful model to dissect microbial pathogenesis [23]. Among others, Pseudomonas aeroginosa [24], Mycobacterium marinum [25], Salmonella [26], and Listeria monocytogenes [27] successfully infect Drosophila adult flies. Host–pathogen interaction can also be analyzed in Drosophila S2 cells, which resemble embryonic hemocytes/macrophages. For example, the intracellular replication of L. monocytogenes [27,28] or Legionella pneumophila [29] in Drosophila cell lines is similar to the one observed in mammalian cells, and the first steps, but not the latest (RB to EB differentiation), of C. trachomatis developmental cycle can be observed in Drosophila cells [30].
An important discovery was made by Clemens et al., who reported that the simple addition of dsRNA to Drosophila cells in culture reduces or eliminates the expression of target genes by RNA interference (RNAi), thus efficiently phenocopying loss-of-function mutations [31]. Combined with the sequence of the Drosophila genome, it has opened a new area of research, allowing scientists to test the involvement of any Drosophila gene in a given cellular process [32,33]. Several screens have already shed light on various cellular processes such as cell viability [33], cytokinesis [34], wnt signaling [35], JAK/STAT signaling [36], and mechanisms of host–pathogen interaction, including Listeria and Mycobacterium pathogenesis [37–39], Candida albicans phagocytosis [40], and L. pneumophila exploitation of the early secretory pathway [29].
We have investigated the possibility of using Drosophila Schneider's Line 2 (SL2) cells [41] as a model system to dissect Chlamydia pathogenesis. We have shown that C. caviae enters, replicates, and performs a complete developmental cycle in Drosophila SL2 cells. We performed a genome-wide RNAi screen and identified 54 factors that, when depleted, inhibit C. caviae infection in Drosophila cells. We identified factors expected to have an effect on Chlamydia infection, but most importantly we also identified uncovered host factors, including components of the Tim-Tom complex. Clearly validating our approach, we showed that depletion of either Tom40 or Tom22 also reduced C. caviae infection in mammalian cells. We discuss how further investigation of the identified candidates may shed light on the molecular mechanisms involved in Chlamydia pathogenesis.
Drosophila SL2 cells [41] were cultured at 25 °C in Schneider media (Invitrogen) supplemented with 10% heat inactivated FBS (JRH).
HeLa 229 cells were cultured at 37 °C with 5% CO2 in DMEM high glucose (Invitrogen) supplemented with 10% heat inactivated FBS (Invitrogen).
C. caviae, the guinea pig model of genital and ocular infection of C. trachomatis, were obtained from R. Rank (University of Arkansas). C. trachomatis Lymphogranuloma venerum, Type II, were obtained from ATCC (VR-902B).
SL2 cell infection with GFP-expressing L. monocytogenes was conducted as previously described [37].
For propagation, HeLa 229 were incubated with C. caviae or C. trachomatis for 48 h in the presence of 2 μg/ml cycloheximide (Sigma). The infected cells were centrifuged (10 min, 1,000 rpm) and the cell pellet was resuspended in SPG buffer (218 mM sucrose, 3.76 mM KH2PO4, 7.1 mM KH2PO4, 4,9 mM glutamate [pH 7.4]). The cells were broken by passing them through a 261/2 gauge needle and the unbroken cells and nuclei were pelleted by centrifugation (10 min, 1,000 rpm). The supernatant was centrifuged (30 min, 12,000 rpm), and the bacterial pellets were resuspended in SPG buffer and stored at −70 °C.
For Drosophila SL2 cell infection, C. caviae were diluted in Schneider media supplemented with 10% heat inactivated FBS and incubated with the cells at 30 °C for the indicated time. For HeLa 229 cell infection, C. caviae or C. trachomatis were diluted in DMEM high glucose supplemented with 10% heat inactivated FBS and incubated with the cells at 37 °C in the presence of 5% CO2. One hour post infection, the bacteria were washed away and the cells were incubated with fresh media for the indicated length of time at 37 °C in the presence of 5% CO2.
The following primary antibodies were used: (FITC)-conjugated C5+C8 monoclonal antibodies directed against Chlamydia MOMP and LPS (1:300, Argene), rabbit polyclonal anti IncA (1:200, [42]), guinea pig polyclonal antibody directed against C. caviae EBs (Kind gift of R. Rank, University of Arkansas), rabbit polyclonal antibody anti-hTom40 (1:500, Kind gift of M. Ryan, La Trobe University, Australia [43]), mouse monoclonal anti-Tom22 (1:2000, Sigma, clone 1C9–2), and rabbit polyclonal anti-actin (1:10,000, Sigma A2066).
The following secondary antibodies were used: goat anti-rabbit AlexaFluor 594 antibody (1:1,000, Molecular Probes), fluorescein (FITC)-conjugated AffiniPure donkey anti-guinea pig IgG (1:500, Jackson ImmunoResearch), peroxidase-conjugated goat anti-rabbit IgG (1:10,000, Jackson ImmunoResearch), and peroxidase-conjugated goat anti-mouse IgG (1:10,000, Jackson ImmunoResearch).
At the indicated time, the cells were fixed for 30 min in PBS containing 4% paraformaldehyde. Immunostainings were performed at room temperature. Antibodies were diluted in PBS containing 0.16 μg/ml Hoechst (Molecular Probes), 0.1% BSA, and 0.05% saponin. Samples were washed with PBS containing 0.05% saponin, and a final PBS wash was performed before examination under an epifluorescence microscope.
Drosophila SL2 cells (108) were incubated at 30 °C with C. caviae (MOI ∼ 5), fixed 45 h post infection by addition of 0.125% glutaraldehyde / 2% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4), postfixed with osmium tetroxide, dehydrated in ethanol, embedded in epoxy resin, sectioned, stained with 1% uranyl acetate, and examined by electron microscopy [44].
HeLa 229 cells cultured on coverslips were fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate (pH 7.4) for 1 h at room temperature, postfixed in 1% osmium tetroxide in the same buffer for 1 h at room temperature, stained in 2% uranyl acetate in 50 mM sodium maleate (pH 5.2) for 1 h at room temperature, dehydrated in ethanol, and embedded in Embed 812 epoxy resin (all reagents from Electron Microscopy Sciences). Ultra-thin sections (60 nm) were obtained on a Reichert ultra microtome, transferred onto formvar- and carbon-coated hexagonal nickel grids, stained with 1% lead citrate and 2% uranyl acetate, and examined in a Tecnai 12 Biotwin electron microscope (FEI Company). Random images of vacuoles were recorded at a magnification of 11,500 using a Morada CCD camera (Olympus Soft Imaging Solutions). For quantitation of the percentage of vacuolar membrane or nuclear envelope covered by mitochondria, a grid with a distance of 560 nm between lines was superposed on top of the images, and the number of intersections of vertical and horizontal lines with membranes counted. The number of intersections of these lines with mitochondria was also counted, but mitochondria were counted as being associated with the vacuolar or nuclear membrane only if the distance between the point of intersection of the grid with the mitochondrial outer membrane and the closest vacuolar or nuclear membrane was 50 nm or less. The ratio of the number of intersection with mitochondria divided by the number of intersections with the vacuolar or nuclear membrane gives an estimate of the percentage of these membranes covered by mitochondria.
Drosophila SL2 cells (108) were incubated at 30 °C with C. caviae. At the indicated time, the infected cells were processed as described above for Chlamydia propagation. The bacterial pellets were resuspended in 100 μl of SPG. To test for the presence of infectious C. caviae in the preparation, 300 μl of a 1:100 dilution were incubated with 6.104 HeLa cells seeded onto coverslips at 37 °C in the presence of 5% CO2. After 1 h, the bacterial suspension was replaced by 500 μl of fresh medium. The cells were fixed 24 h post infection, stained, and the percentage of cells containing a large inclusion was determined by visual inspection using an epifluorescence microscope.
The infection was performed in 384-well format such that 75% of the cells were infected. At the indicated time, the infected cells were collected and transferred to an eppendorf tube containing 100 μl of glass beads (Sigma, G8772) and 300 μl of DMEM high glucose supplemented with 10% FBS. The cells were broken by vortexing for 1 min, and 40 μl of dilutions of the lysat were added to 4.103 HeLa 229 cells seeded in 384-well plate. After 1 h at 37 °C in the presence of 5% CO2, the lysat was washed away and 40 μl of fresh media was added to each well. The cells were fixed and stained 24 h post infection and the percentage of infected cells was determined.
Two sets of 42 384-well plates containing 0.25 μg of dsRNA per well were provided by the Drosophila RNAi Screening Center (Harvard Medical School, Boston, Massachusetts, http://www.flyrnai.org). Drosophila SL2 cells (2.104), resuspended in 20 μl of serum-free Schneider media, were seeded in each well and incubated 1 h at 25 °C before the addition of 20 μl of Schneider media containing serum. After 3.5 d, the cells were infected by addition of 10 μl of Schneider media containing C. caviae. The cells were centrifuged for 1 min at 1,000 rpm and incubated at 30 °C for 48 h. The cells were processed for immunofluorescence by using the DNA dye Hoeschst and FITC-conjugated C5+C8 monoclonal antibodies. An automated microscope was used to automatically track, focus, and capture fluorescent images of the cells within each well across an entire plate. One set of images was captured in the blue channel to detect the cells' nuclei and one set in the green channel to detect Chlamydia. The qualitative analysis of the image data was done by visual inspection.
dsRNA used for validation and secondary assays were synthesized using a MEGAscript High Yield transcription kit (Ambion) according to the recommendation of the manufacturer.
The protocol used for siRNA transfection was adapted from Dharmacon's HeLa cells transfection protocol. One volume of siRNA buffer containing 200 nM of siRNA was incubated with 1 volume of serum-free DMEM high glucose containing 5 μl/ml DharmaFECT-1 transfection reagent for 20 min at room temperature. Two volumes of DMEM high glucose supplemented with 20% FBS containing 5.104/ml HeLa 229 cells were added to each well and the cells were incubated at 37 °C with 5% CO2 for 3 d. The total volume was 40 μl in 384-well and 400 μl in 24-well. In 24-well format the transfection mix was replaced by 500 μl of fresh media 24 h post transfection.
The knock down of Tom40 or Tom22 was performed as described above in 24-well plate. Three days post transfection, the cells were harvested in 100 μl of protein sample buffer and 20 μl of cell lysates were run on SDS-PAGE gels and analyzed by western blot using HPR-conjugated secondary antibodies and Amersham ECL western blotting detection reagents.
Images were acquired using the Metamorph software (Molecular Devices). The integrated morphometry analysis module was used to quantify the size of C. caviae inclusions.
In an attempt to use Drosophila as a model system to study Chlamydia pathogenesis, we investigated C. caviae replication in Drosophila SL2 cells. For this purpose, 80% confluent Drosophila SL2 cells cultured in 96-well dish were incubated with C. caviae. At various times post infection, the cells were transferred to Concanavalin A–coated coverslips (Sigma, 2 mg/ml) in Schneider media for 2 h. The samples were then fixed and stained with the DNA dye Hoescht and a FITC-conjugated antibody directed against Chlamydia to stain the inclusion.
As shown in Figure 1, C. caviae is able to infect and replicate in Drosophila SL2 cells. Although most of the cells contained at least one bacterium 1 h post infection, only 20% to 30% of the cells had an inclusion 48 h post infection (not shown), suggesting that some bacteria were actually cleared in the phagocytic SL2 cells. However, when the bacteria were successful in establishing their niche, the infected cells displayed a perinuclear inclusion whose size increased between 24 and 72 h post infection. At 96 h post infection, the size of the inclusions was more heterogeneous (not shown) and some cells displayed disrupted inclusions, suggesting that the developmental cycle was completed and that reinfection was occurring between 72 and 96 h post infection.
We next determined whether C. caviae were undergoing a full developmental cycle in Drosophila SL2 cells. To this end, we determined whether the different developmental forms of C. caviae were present in the inclusion by electron microscopy. As shown in Figure 2A, 45 h post infection the bacteria were found in a membrane-bound compartment that occupies most of the cytosolic space. The inclusions mainly contained RBs and intermediate bodies (IBs) in the process of differentiating to EBs and are characterized by their DNA condensation stage, but they also contained some bacteria with an EB morphology (Figure 2B), suggesting that in Drosophila SL2 cells, RBs start to differentiate back to EBs 45 h post infection.
In order to demonstrate that infectious progeny was produced, C. caviae harvested from Drosophila SL2 cells at different times post infection were used to infect HeLa cells (Materials and Methods; Figure 2C). When C. caviae were harvested 3 h post SL2 infection, 10% of the HeLa cells displayed an inclusion. This number decreased to less than 5% when the bacteria were isolated 24 or 48 h post infection, suggesting that a substantial amount of bacteria were either cleared or had differentiated into non-infectious RBs. In contrast, 12.5% and 19% of the HeLa cells contained a large inclusion when the bacteria were harvested 72 and 96 h post SL2 infection, respectively. After 96 h, the number of infected HeLa cells remained constant. These results indicate that infectious forms of C. caviae are produced in Drosophila SL2 cells. Moreover, they are in agreement with the immunofluorescence (Figure 1) and electron microscopy (Figure 2A and 2B) data and confirm that 48 h post infection of Drosophila SL2 cells, the inclusion mainly contains RBs and IBs, whereas EBs are produced in the next 24 h. Taken together, these data show that C. caviae undergo a full developmental cycle in Drosophila SL2 cells and suggest that the cycle lasts 72 to 96 h.
The TTSS of C. caviae was functional in Drosophila SL2 cells as shown by determining the presence of the Inc family protein, IncA, on the C. caviae inclusion membrane (Figure 3). Drosophila SL2 cells were fixed 48 h post infection with C. caviae, stained with the DNA dye Hoescht to visualize the nuclei (N) and the inclusions (Inc), and antibodies directed against IncA. A ring-like signal (IncA, red) that surrounded the inclusion (Inc, blue) was observed, indicating that, in Drosophila SL2 cells, the TTSS of C. caviae is functional and that TTS substrates such as IncA, are delivered to the inclusion membrane.
Sixteen thousand Drosophila genes were individually knocked down by RNAi and screened for their ability to reduce C. caviae infection of Drosophila SL2 cells. The assay was performed as follows (Materials and Methods; Figure 4A). After 3.5 d of RNAi treatment, the Drosophila SL2 cells were incubated with C. caviae for 48 h. The infected cells were fixed and stained with a DNA dye and a Chlamydia-specific FITC-conjugated antibody. An automated microscope was used to capture fluorescence images that were subsequently analyzed by visual inspection. The primary screen was performed in duplicate. We identified 162 candidates that, when depleted, reduced C. caviae infection (Table S1). Figure 4B is representative of the phenotype observed: few cells displayed wild-type size inclusion (middle top panel) and the number of infected cells, as well as the size of the inclusion, was largely reduced (middle bottom panel). The candidates were grouped into 14 functional categories (Figure 5): miscellaneous (32), unknown (32), metabolism (18), transcription (14), vesicular trafficking (12), cytoskeleton (9), mitochondria (8), transporter (8), kinase and phosphatase (7), chromatin organization (5), endosome and lysosome (5), protein biosynthesis (5), RNA processing (4), and cell cycle (3).
The dsRNA targeting most of the candidates of the miscellaneous, metabolism, vesicular trafficking, cytoskeleton, mitochondria, transporter, kinase and phosphatase, and endosome and lysosome categories were resynthesized to confirm the phenotype observed in the primary screen (Table S1). Out of the 100 candidates retested, the phenotype was confirmed for 54 candidates in at least two out of three replicates (Table 1). The validation rate varied among the categories: miscellaneous (40%), metabolism (47%), vesicular trafficking (75%), cytoskeleton (75%), mitochondria (67%), transporter (37%), kinase and phosphatase (57%), and endosome and lysosome (100%).
In an attempt to assay for Chlamydia specificity, the knock down of the candidates was tested for the inhibition of L. monocytogenes infection (Table 1). The knock down of most vesicular trafficking (9/9), cytoskeleton (4/6), and endosome and lysosome (5/5) candidates inhibited both C. caviae and L. monocytogenes infection, and an equal number of kinase and phosphatase candidates inhibited C. caviae or L. monocytogenes infection. The knock down of most miscellaneous (11/13) and metabolism (6/8) candidates and of all mitochondria (6/6) and transporter (3/3) candidates inhibited C. caviae infection only. These results suggest that the latter categories are likely to represent candidates specifically involved in C. caviae infection.
The RNAi screen in Drosophila cells revealed that the silencing of six mitochondrial genes inhibited C. caviae, but not L. monocytogenes infection. Moreover, four out of the six candidates were members of the mitochondrial membrane translocase, a multiprotein complex involved in the recognition and import of nuclear-encoded mitochondrial proteins to the mitochondria [43,45]. Taken together, these results suggested a specific role of this machinery for optimal C. caviae infection in Drosophila cells. To address the relevance of these findings in Chlamydia pathogenesis, this observation was further investigated in mammalian cells.
Tom40 or Tom22 expression was knocked down in HeLa 229 cells using a mix of four siRNA duplexes directed against their respective mRNA (ThermoFisher). In addition, each siRNA was tested individually to rule out any potential off-target effects. The depletion of either Tom40 or Tom22 was assayed 3 d post transfection of the siRNAs by western blot analysis. As shown in Figure 6A, both Tom40 and Tom22 were efficiently depleted after incubation with the mix of four siRNAs or with individual siRNA duplexes.
The effect of Tom40 or Tom22 depletion on C. caviae infection was analyzed. HeLa 229 cells were incubated for 3 d with either CDH1 siRNA control directed against E-Cadherin, or Tom40 or Tom22 siRNAs pooled (mix), or individually (1, 2, 3, 4), infected with C. caviae for 24 h, and processed for immunofluorescence. The corresponding low and high magnification images are depicted in Figure 6B and 6C, respectively. The nuclei were labeled with the DNA dye Hoeschst (Figure 6B and 6C: left panel, DNA, blue) and the inclusions were stained with a guinea pig polyclonal antibody against C. caviae (Figure 6B and 6C: middle panels, C. caviae, green). Although the number of infected cells was similar, the inclusions appeared smaller upon Tom40 or Tom22 depletion (compare CDH1 middle panels to Tom40 or Tom22 middle panels).
A computer-assisted analysis of the images was used to quantify the size of the inclusions (Materials and Methods). In the control situation, we determined that each inclusion could be defined as a 10- to 150-μm2 object and 40% of the inclusions were larger than 30 μm2. We defined 10- to 30-μm2 and 30- to 150-μm2 objects as small and large inclusions, respectively. The impact of Tom40 or Tom22 knock down on C. caviae ability to form large inclusions was analyzed (Figure 6D). A 5- to 3-fold reduction in the percentage of large inclusions was observed upon depletion of either Tom40 or Tom22, confirming the overall reduction in the size of the inclusions and suggesting that upon Tom40 or Tom22 depletion, C. caviae intracellular growth is impaired.
Electron microscopy analysis of C. caviae inclusions in control or Tom40 depleted cells confirmed the immunofluorescence results. Although a mixed population of small and large inclusions was observed 24 h post infection, the overall size of Tom40 depleted cell inclusion was smaller (Figure 7). In addition, RBs had already started to differentiate back into EBs in control cells, and 85% of the inclusions contained more than 25% EBs. In contrast, although some IBs were present, very few EBs were visible in Tom40-depleted cells, and only 25% of the inclusions contained more than 25% EBs. This result suggested that, in addition to a reduction in intracellular growth, differentiation back into EBs is also lessened in Tom40-depleted cells.
The electron microscopy results suggested that RB differentiation into EBs was reduced upon Tom40 or Tom22 depletion. We therefore investigated the production of infectious progeny by Tom40- or Tom22-depleted cells. The cells were incubated with the siRNA in pool or individually for 3 d before incubation with C. caviae for 48 h to allow completion of the developmental cycle. The infected cells were collected, lysed with glass beads, and dilutions of the lysate were used to infect fresh HeLa 229 cells (see Materials and Methods). The cells were fixed 24 h post infection and the number of inclusion forming units (IFUs) was determined after assessment of the number of infected cells by immonulabeling (Figure 8A). We observed a 2- to 3-fold reduction in the production of infectious progeny upon Tom40 or Tom22 depletion. On the contrary, a similar number of infectious C. trachomatis were recovered from control or Tom40- or Tom22-depleted cells (Figure 8B).
These results demonstrate that, as suggested by the electron microscopy analysis, the reduction in the size of C. caviae inclusions is accompanied with a decrease in the number of infectious progeny produced. Altogether, our results indicate that depletion of members of the Tom complex in mammalian cells have a detrimental effect on C. caviae intracellular replication, which impairs bacterial replication and differentiation. Since Tom40 or Tom22 depletion had no effect on C. trachomatis infection, our results also indicate that the mechanism involved is C. caviae–specific.
Chlamydia infections represent an enormous burden to human health, and although some of the host cellular processes targeted by Chlamydia have been identified, most of the factors involved in the infection process remained to be identified. This paucity of knowledge is mainly due to the fact that Chlamydia is not genetically tractable and to the difficulty of conducting genetic approaches in the mammalian host. Drosophila has recently emerged as a powerful alternative model to dissect microbial pathogenesis, and we show here that Drosophila SL2 cells constitute a viable model to study Chlamydia infection and identify host factors involved in the infection process.
We demonstrated that, similar to the situation in mammalian cells [8], infectious forms (EB) of C. caviae enter Drosophila SL2 cells, differentiate into the replicative form (RB), replicate within a membrane-bound compartment, and differentiate back from RBs to EBs.
A previous report showed that different serovars of C. trachomatis, including C. trachomatis LGV serovar L2, could initiate their developmental cycle in Drosophila S2 cells [30]. However, the later stages of the developmental cycle were not achieved. Similarly, we found that when Drosophila SL2 cells were incubated with C. trachomatis LGV serovar L2 most cells were also infected 1 h post infection. However, the pattern of staining did not change over a 72-h period post infection and the cells never displayed large perinuclear inclusions (not shown), confirming that C. trachomatis developmental cycle is not complete in Drosophila SL2 cells.
We noticed a difference in the morphology of C. caviae inclusion in Drosophila cells compared to mammalian cells. The inclusions appear multilobed in mammalian cells [46], whereas they appeared as a single membrane-bound compartment in Drosophila SL2 cells. This morphology resembles that of C. trachomatis inclusions that are known to undergo homotypic fusion and are therefore monovacuolar. This observation suggests that C. caviae inclusions may also undergo homotypic fusion in Drosophila SL2 cells.
IncA, a type III secretion (TSS) substrate known to be in involved in the homotypic fusion of the C. trachomatis inclusions [47], was present on the surface of C. caviae inclusion in Drosophila SL2 cells. Since IncA from C. caviae can interact with itself [48], it is possible that in Drosophila SL2 cells it participates to the biogenesis of a single large inclusion. If it is the case, some Drosophila factors probably interact with IncA and help promote the fusion. However, one cannot exclude that in Drosophila cells the homotypic fusion of C. caviae inclusions is IncA independent.
Using the Drosophila cell / C. caviae model system, we have performed an RNAi screen and identified 54 host factors that, when depleted, reduced C. caviae infection. By testing the effect of the candidates' knock down on L. monocytogenes infection, we have identified candidates presumably specific of C. caviae infection. In the following section, we discuss their potential relevance in Chlamydia pathogenesis.
The attachment of most Chlamydia species to the host cell is dependent on host cell heparan sulfate glycosaminoglycans (GAGs) [15]. C. caviae is no exception, because its adhesion is GAG dependent and can be blocked by heparin [49]. Drosophila contains two main glypicans: Dally (Division abnormally delayed) [50] and Dlp (Dally-like protein) [51,52]. They are composed of cell-surface heparan sulfate proteoglycans linked to the plasma membrane by a glycosyl phosphatidylinositol linker. Our screen showed that the knock down of Dlp reduced C. caviae infection, suggesting that Dlp may promote the attachment of C. caviae to the cell surface.
Activation of Rho family of GTPases and actin remodeling has also been implicated in Chlamydia entry [15]. For example, Cdc42 and actin polymerization are involved in C. caviae entry of mammalian cells [53], and we show here that their depletion also reduced infection of Drosophila cells. Rac1, which is also involved in C. caviae entry in mammalian cells [53], was not identified in our screen. The Drosophila genome contains two rac genes, and it is possible that the single knock down of one or the other was not sufficient to block C. caviae entry. In addition, we also identified Ssh, a phosphatase that controls actin reorganization through the dephosphorylation of cofilin [54]. Ssh was not previously reported to play a role in Chlamydia pathogenesis, but our data suggest that it may be implicated in regulating actin dynamics upon entry of C. caviae.
After their internalization, C. trachomatis EBs direct the nascent inclusion to a peri nuclear area. This movement is dependent of microtubules and the motor dynein [16].We have identified two candidates that are linked to motors and microtubules. The first candidate, pav, encodes a kinesin-like protein [55]. Although kinesin is not involved in the trafficking of C. trachomatis inclusion to the peri nuclear area [16], the microinjection of antibodies against kinesin prevents the recruitment of mitochondria to C. psittaci inclusions and delays the developmental cycle [56]. A defect in mitochondria recruitment to the inclusion may therefore explain the phenotype observed upon pav knock down in Drosophila cells. The potential importance of mitochondria in Chlamydia infection will be further discussed in the following section. The second candidate related to microtubule is katanin-60. In mammalian cells, katanin concentrates at the centrosome of the cell, where the p60 subunit exerts its microtubule severing activity and induces the release of microtubules from the centrosome [57]. C. trachomatis inclusions associate with centrosomes [58]. It is possible that Chlamydia interacts with the centrosome and induces a katanin-mediated local destabilization of the microtubule network, thus allowing the expansion of the inclusion.
A recent study revealed a dynamic interaction between multi-vesicular body–derived constituents and C. trachomatis inclusion [59]. We have identified two candidates involved in lysosomal transport (CG11814) and organization (CG5691), as well as two subunits of the v-ATPase (VhaAC39 and VhaPPA1–1). The identification of such factors suggests that, at some point during the developmental cycle, Chlamydia inclusions may interact with compartments of the endocytic pathway. Further analysis of theses candidates may shed light on the mechanism involved.
C. trachomatis inclusion also intercepts vesicles and lipids from the Golgi [18] and targets lipid droplets [20]. We have identified several enzymes involved in fatty acid synthesis, desaturation, elongation, and oxidation. The identification of such enzymes reinforces the idea that the acquisition of lipids is an important aspect of Chlamydia intracellular replication, and further investigation may shed light on the host metabolism pathways targeted by Chlamydia.
Our screen revealed that the knock down of members of the Tim-Tom complex, the multiprotein complex involved in the recognition and import of nuclear-encoded mitochondrial proteins to the mitochondria [45,60], inhibited C. caviae infection in Drosophila cells. Importantly, we have shown that the knock down of two major components of the outer membrane complex of mitochondria, Tom40 and Tom22, also inhibited C. caviae infection in mammalian cells. In the following section we discuss potential mechanisms that may explain the phenotype observed.
The National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/) accession numbers for the mammalian genes are CDH1 (NM_004360), Tom40 (NM_006114), and Tom22 (NM_020243). |
10.1371/journal.ppat.1000699 | T Cell Detection of a B-Cell Tropic Virus Infection: Newly-Synthesised versus Mature Viral Proteins as Antigen Sources for CD4 and CD8 Epitope Display | Viruses that naturally infect cells expressing both MHC I and MHC II molecules render themselves potentially visible to both CD8+ and CD4+ T cells through the de novo expression of viral antigens. Here we use one such pathogen, the B-lymphotropic Epstein-Barr virus (EBV), to examine the kinetics of these processes in the virally-infected cell, comparing newly synthesised polypeptides versus the mature protein pool as viral antigen sources for MHC I- and MHC II-restricted presentation. EBV-transformed B cell lines were established in which the expression of two cognate EBV antigens, EBNA1 and EBNA3B, could be induced and then completely suppressed by doxycycline-regulation. These cells were used as targets for CD8+ and CD4+ T cell clones to a range of EBNA1 and EBNA3B epitopes. For both antigens, when synthesis was induced, CD8 epitope display rose quickly to near maximum within 24 h, well before steady state levels of mature protein had been reached, whereas CD4 epitope presentation was delayed by 36–48 h and rose only slowly thereafter. When antigen expression was suppressed, despite the persistence of mature protein, CD8 epitope display fell rapidly at rates similar to that seen for the MHC I/epitope half-life in peptide pulse-chase experiments. By contrast, CD4 epitope display persisted for many days and, following peptide stripping, recovered well on cells in the absence of new antigen synthesis. We infer that, in virally-infected MHC I/II-positive cells, newly-synthesised polypeptides are the dominant source of antigen feeding the MHC I pathway, whereas the MHC II pathway is fed by the mature protein pool. Hence, newly-infected cells are rapidly visible only to the CD8 response; by contrast, latent infections, in which viral gene expression has been extinguished yet viral proteins persist, will remain visible to CD4+ T cells.
| Many viruses infect cells in which both the MHC I and MHC II pathways of antigen presentation are active, and so viral proteins expressed in those cells may be presented as MHC I-peptide complexes to CD8+ T cells and as MHC II-peptide complexes to CD4+ T cells. Here we study these processes in a model system involving Epstein-Barr virus-infected human B lymphocytes (MHC I/II-positive) where viral antigen expression can be induced or suppressed at will, and antigen presentation tracked with specific CD8+ and CD4+ T cell clones. In this system, we find that the MHC I pathway is entirely fed by newly-synthesised polypeptides, whereas the MHC II pathway depends upon antigen supplied from the mature protein pool. Hence, while only CD8+ T cells can rapidly recognise new infections, only CD4+ T cells will recognise latent infections in which viral gene expression is extinguished yet a pool of viral antigens remains.
| Many intracellular pathogens, particularly viruses, naturally infect cells of the haemopoietic system that express both MHC I and MHC II molecules. Such infected cells may be rendered visible to the host T cell response through the intracellular processing of virally-encoded proteins, leading to cell surface display of MHC I- and MHC II- peptide complexes recognised by CD8+ and CD4+ T cells respectively. With regard to MHC I-restricted presentation, the speed with which virus-infected cells become recognisable by CD8+ T cells [1] and the involvement of the proteasome in that process [2] led to the idea that a proportion of all newly-synthesised viral polypeptides were marked for immediate degradation, generating peptides that were fed into the MHC I pathway [3]. While the concept has evidential support [4],[5],[6],[7], questions remain about the proportion of translation products sacrificed in this way [8],[9], the mechanism that underpins their selection [10],[11] and most importantly the degree to which, in latently-infected cells where viral antigen synthesis has been extinguished, cells may still be visible to the virus-specific CD8 response through MHC I-restricted processing of antigen from the mature protein pool. Only two studies have attempted to address this latter issue by specifically regulating antigen expression rather than resorting to general inhibitors of translation [12],[13]. Though both studies supported the dominance of newly-synthesised protein as an antigen source, in each case the evidence came from a single epitope studied at a very limited number of time points leaving the generality of the results, with respect to such variables as antigen dose, epitope location and target cell identity, unresolved.
Less is known about the rules governing MHC II-restricted presentation of endogenously expressed viral antigens, though it is clear that under some circumstances this can occur [14],[15]. To date there are examples of endogenous antigen accessing the MHC II pathway either through location in the endoplasmic reticulum itself [16], through delivery to endosomes/lysosomes by macro- [17],[18] or chaperone-mediated [19] autophagy, or through release and re-uptake by neighbouring cells [20]. However there is little information on two important issues: firstly the kinetics with which MHC II-restricted epitopes are presented following antigen expression, which determines when a newly-infected cell becomes visible to the CD4+ T cell response, and secondly the relative importance of newly-synthesised polypeptides and the mature protein pool as antigen sources.
Here we address these issues using Epstein-Barr virus (EBV), a human gamma-herpesvirus that transforms B cells in vitro into MHC I/II-positive lymphoblastoid cell lines (LCLs) expressing eight viral proteins, the nuclear antigens EBNAs 1, 2 3A, 3B, 3C and -LP, and the latent membrane proteins LMPs 1 and 2 [21]. Such LCLs resemble the virus-transformed B cells that arise during EBV infection in vivo and elicit the MHC I- and MHC II-restricted T cell responses that control the infection [22]. Many of these responses have been mapped to individual peptide epitopes and epitope-specific CD4+ and CD8+ T cell clones shown to recognise MHC-matched LCL targets [22]. Here we sought to use such clones to follow the presentation of EBV antigens via the MHC I and MHC II pathways in an LCL background which lacked base-line epitope display and where expression of the cognate antigen could be temporally controlled. For this purpose we chose two indicator antigens, EBNA3B and EBNA1. EBNA3B is non-essential for transformation in vitro and therefore one can establish LCLs with an EBNA3B gene-deleted virus [23],[24]; EBNA1, the virus genome maintenance protein, is required for transformation but shows sequence variation between virus isolates, allowing one to establish LCLs using a virus that lacks many of the relevant T cell epitopes [25],[26]. In both cases we then introduced the cognate antigen-coding sequence under the control of a doxycycline-regulated promoter and monitored CD4 and CD8 epitope display after inducing or suppressing new antigen synthesis.
Figure 1A shows the vector used to achieve dox-dependent antigen expression [27]. Rat CD2 expression from the vector backbone allows initial enrichment of transfected cells, while the EBV ori-p sequence promotes episomal maintenance in LCLs. Antigen-coding sequences lie under the control of a dox-regulated promoter. We first introduced an EBNA3B-carrying vector (pEBNA3B-tet, Figure 1A) into LCLs made using a recombinant EBNA3B-KO virus. Figure 1B illustrates the pattern of results consistently observed with stable pEBNA3B-tet transfectants on three different LCL backgrounds. EBNA3B protein expression, undetectable by immunoblotting in non-induced cells, showed a clear dose-dependent response to 7 day treatment with dox, reaching a level equivalent to that seen in wild-type EBV-transformed LCLs at 25 ng/ml dox and increasing to supra-physiologic levels at higher dox concentrations.
We then assayed these same cells, dox-induced for 7 days, as targets for T cell recognition. CD8+ T cell clones were generated against five well-defined epitopes in EBNA3B (HRC/B*2705, RRA/B*2702; AVF/A*1101, IVT/A*1101, VEI/B*4402; positions shown in Figure 1A, see Table S1 for details). Because EBNA3B had not been studied before as a CD4 target, we first screened EBV-immune donors for CD4+ T cell reactivity to an EBNA3B peptide panel in IFNγ Elispot assays, generated CD4+ T cell clones against three of the epitopes thus defined and determined their MHC II restriction using standard approaches [28],[29]. These epitopes (FIE/DRB1*1501, ILR/DRB4*01 and QAP/DRB3*0201; see Table S1) are located on the EBNA3B sequence in Figure 1A. Figure 1C shows representative results from such experiments, here using a pEBNA3B-tet LCL (A*1101/DRB3*0201-positive) as a target for CD8+ clones against the IVT/A*1101 epitope and for CD4+ clones against the QAP/DRB3*0201 epitope. All such experiments included, as a positive control target, a wild-type EBV-transformed LCL from the same individual expressing EBNA3B from the resident EBV genome. Target cell recognition is assayed by IFNγ release after 18 h of co-culture. There was no response to the non-induced pEBNA3B-tet LCL by either CD8+ or CD4+ effectors, whereas dox-induced cells were recognised at levels which increased in a dose-dependent manner. For both effector populations, the recognition of target cells exposed to 25 ng/ml dox (i.e. the dose inducing physiologic levels of EBNA3B) was similar to that seen for the wild-type LCL, whereas higher levels of induction increased recognition accordingly. Assays with different pEBNA3B-tet LCLs, using effector cells against the other four CD8 and two CD4 epitopes in EBNA3B, gave very similar results (data not shown). All subsequent studies were therefore conducted on cells induced to express indicator antigens at physiologic (25 ng/ml dox) and at supra-physiologic (100 ng/ml dox) levels, with similar patterns of results obtained.
We first asked how quickly target cells became susceptible to CD8+ and CD4+ T cell recognition following dox-induction. Figure 2 shows one such experiment inducing the above pEBNA3B-tet LCL at the two dox concentrations. In both cases, expression of EBNA3B protein was detectable by immunoblotting within 6 h of dox addition, and by 72 h had increased to reach a stable steady-state level that was again higher (relative to a wild-type LCL) at the higher inducing dose (Figure 2A). Aliquots of the same cells were used as targets in T cell assays, each time alongside cells from the appropriate non-induced and long-term-induced cultures. To examine epitope display at the precise time of harvest, all target cells were fixed in 1% PFA before addition to the assay. As shown in Figure 2B, while absolute levels of IFNγ release were always higher with targets given 100 ng/ml dox, the same pattern of results was obtained following antigen induction at either dose. Thus, recognition by CD8+ T cells specific for the IVT/A*1101 epitope was detectable within 6 h of dox induction and by 36 h had increased to plateau at the same level as seen against long-term dox-induced targets. In contrast, recognition by CD4+ T cells specific for the ILR/DRB4*01 epitope was not detectable until 36–48 h and increased quite slowly thereafter, only reaching the long-term dox plateau level on targets induced for 168 h. In further experiments with this and other pEBNA3B-tet LCLs, these temporal differences between CD8 and CD4 epitope display held true for all eight EBNA3B epitopes tested (data not shown).
The existence of EBNA1 sequence variation between geographically distinct EBV isolates [26] allowed us to generate LCLs using a Chinese virus strain (CKL) with epitope mutations that, for the T cell clones used in these experiments, abrogated CD8 recognition and reduced CD4 recognition to a very low base-line. Into these LCLs, we then introduced an epitope-positive EBNA1 allele under dox-regulated control. As shown in Figure 3, we used both a full length EBNA1 sequence and a sequence (E1dGA) from which the internal glycine-alanine repeat (GAr) domain had been deleted. Note that this GAr domain reportedly offers the wild-type protein some level of protection from CD8+ T cell recognition through reducing the rate of its translation from mRNA [30] and/or though stabilising the protein from proteasomal digestion [31]. Figures 3A and B show immunoblots of EBNA1 expression induced in the pEBNA1-tet and pE1dGA-tet LCLs following 100 ng/ml dox induction. As with inducible EBNA3B, the two forms of EBNA1 accumulated to reach their steady state levels by 72–96 h post-induction, though E1dGA was detectable slightly earlier than full length EBNA1 (6 versus 12 h post-induction), and accumulated to slightly higher steady-state levels, a finding consistent with published data [30],[32].
We examined the kinetics of EBNA1 and E1dGA presentation using clones against two CD8+ (HPV/B*3501 and IPQ/B*07) and two CD4+ (GLR/DQB1*06 and VYG/DRB1*11) T cell epitopes (see Table S1). Results from one such set of assays are shown in Figures 3C and D, using HPV- and VYG-specific effectors and target LCLs established from a B*3501, DRB1*11-positive donor. Focusing first on the CD8+ T cell data, we found that both EBNA1 and E1dGA were rapidly recognised by CD8+ T cells and reached their plateau values (shown by long-term induced cells) within 48 h. Note that these plateau values were always some 20–30% higher with target cells expressing the E1dGA construct. Given the reported effect of the GAr domain on MHC I processing, we looked in greater detail at early time points in the above experiment, repeating the CD8 assays hourly over the first 12 hr post-induction. As illustrated in Figure S1 for assays conducted with 25 and 100 ng/ml dox inductions, we found that CD8 epitope display from E1dGA was indeed slightly accelerated at early times, typically reaching 35% of its plateau value by 12 hr compared to 25–30% for full length EBNA1. Turning now to the CD4+ T cell data in Figures 3C and D, antigen presentation by the MHC II pathway was again profoundly delayed. Thus there was no CD4+ T cell recognition of dox-induced target LCLs (other than very weak base-line recognition of the CKL virus-coded EBNA1) until 48 h post-induction, followed by a slow rise that did not reach the long-term plateau value even by 168 h. Both the EBNA1 and E1dGA proteins gave similar results in this respect, although here the plateau level of CD4+ T cell recognition was always slightly higher with cells expressing the full length protein. Experiments conducted on a different pair of pEBNA1-tet and pE1dGA-tet LCLs using T cell clones specific for the IPQ/B*07 and GLR/DQB1*06 epitopes gave the same pattern of results (data not shown).
The temporal differences between CD8 and CD4 epitope display therefore held true for all epitopes studied both in EBNA3B and in EBNA1. However we reasoned that the delayed presentation of CD4 epitopes might simply reflect their processing by an indirect route if, as previously shown for EBNA3A and 3C, the source antigens access the MHC II pathway through antigen release and uptake by neighbouring cells in the LCL culture [20]. We first investigated this for EBNA3B by co-cultivating “antigen donor” cells (a pEBNA3B-tet LCL lacking relevant MHC restriction alleles but dox-induced to express cognate antigen) with “antigen-recipient” cells (an antigen-negative EBNA3B-KO LCL with the relevant MHC alleles) for 7 days, then used this mixture as a target for EBNA3B-specific CD4+ and CD8+ T cell clones. As shown in Figure S2A, we found that co-culture could indeed sensitise recipient cells to recognition by CD4+ T cell clones specific for the EBNA3B ILR epitope, although not by the corresponding CD8+ IVT clones. However, in parallel experiments where we co-cultured dox-induced pEBNA1-tet “antigen donor” cells with a CKL virus-transformed “antigen recipient” LCL, there was never any recognition of the co-culture by EBNA1-specific CD4+ T cells (Figure S2B). Furthermore a second sensitive method of detecting inter-cellular antigen transfer, where recipient cells are fed with 25x-concentrated culture supernatant from donor LCLs [20], again never sensitised recipient cells to EBNA1-specific effectors (Figure S2C). This clearly shows that inter-cellular antigen transfer likely contributes to EBNA3B's presentation via the MHC II pathway in LCL cells; however, as others have also observed [17], endogenously expressed EBNA1 is presented by an intracellular route. Yet, irrespective of these differences, both antigens show delayed presentation via the MHC II pathway following the induction of antigen synthesis.
We therefore sought reassurance that this slow presentation did not simply reflect an intrinsic feature of MHC class II maturation and epitope display in our LCL cells. To do so, we used the inducible vector system to express E1dGA fused with an invariant chain (Ii) tag that delivers the protein directly into endosomes and the MHC II processing compartment [33]. As shown in Figure 4A, expression of the E1dGA-Ii protein is detectable by immunoblotting 24, 48 and 72 h after 100 ng/ml dox-induction but at very low levels compared to non-tagged EBNA1 and E1dGA. This reflects on-going degradation of the endosomally-targeted E1dGA-Ii protein, since adding chloroquine, an inhibitor of endosomal proteolysis, 24 h prior to harvest increased the level of protein detectable. Figure 4B shows the corresponding T cell assay data following dox-induction. The Ii-tagged protein was rapidly presented not just to CD8+ T cells, where it was processed as quickly as the non-targeted constructs, but also to CD4+ T cells. In this latter case, recognition appeared within 12 h and became almost maximal by 48 h, much quicker than with the non-tagged proteins. Thus our LCLs can rapidly process and present endogenously expressed antigen, once that antigen gains access to the MHC II presentation pathway.
We then examined antigen presentation in long-term 100 ng/ml dox-induced cells after switching off new antigen synthesis by dox-withdrawal. As illustrated in Figure 5A, using a Q-RT-PCR assay for vector-encoded EBNA3B mRNA transcripts, we first showed that >80% of transcripts are lost within 6 h and none are detectable by 24 h. This implies that new antigen synthesis must terminate quite rapidly after dox withdrawal. However, as shown in Figure 5B, the EBNA3B protein is clearly very stable since it remained easily detectable in immunoblots for several days post-withdrawal. Indeed, as the immunoblots were loaded with equal number of cells each time, the falling EBNA3B levels reflect both slow natural turnover of the protein and also dilution from cell doubling (in cultures with a doubling time of 48–72 hr).
Aliquots of LCL cells from the same experiment (HLA B*2702, DRB3*0201-positive) were used in parallel as targets for EBNA3B-specific T cells. As shown in Figure 5C, target cell recognition by an RRA epitope-specific CD8+ T cell clone fell progressively after dox withdrawal, down to half of the original level by 48 h, to <10% by 96 h and approaching zero thereafter. By contrast, recognition by a CD4+ T cell clone against the QAP epitope fell much more slowly, being still >50% of the original level after 96 h and >20% even after 192 h. Indeed the rate of fall in CD4 epitope display closely paralleled the level of EBNA3B protein detectable in these target cells by immunoblotting (cf. Figures 5B and 5C). Such experiments were conducted on all three pEBNA3B-tet LCL backgrounds, whether first induced at 25 ng/ml or 100 ng/ml dox, and included clones against five CD8 epitopes and three CD4 epitopes. In each case CD8+ T cell recognition had fallen to <10% of its original value by 96 h after dox withdrawal, whereas CD4+ T cell recognition was still at 35–50% of its original value at the much later time of 168 h (data not shown).
Results from a corresponding experiment involving pEBNA1-tet and pE1dGA-tet LCLs are shown in Figure 6. Q-RT-PCR assays using primer/probe combinations specific for vector-encoded EBNA1 and E1dGA mRNAs showed mRNA levels fell rapidly after dox-withdrawal and were undetectable beyond 12 h (Figure 6A). Again, therefore, new antigen synthesis must rapidly terminate following dox withdrawal yet, as shown by the immunoblots in Figures 6B and 6D, both the EBNA1 and E1dGA proteins are relatively stable, levels per cell falling slowly over time and being still detectable at 168 h. When these same dox-withdrawn cells (HLA B*3501, DRB1*11-positive) were used as targets in T cell assays, recognition by HPV-specific CD8+ T cells fell to <50% of the original level by 48 h and was undetectable by 120 h, whereas recognition by a VYG-specific CD4+ T cells fell much more slowly, being still 30–40% of the original value as late as 168 h. Again, parallel experiments using a different LCL background and T cell clones against the other CD8 and CD4 epitopes in EBNA1 produced a very similar pattern of results.
While Figures 5 and 6 showed that CD8 and CD4 epitope display fell at different rates after switching off new antigen synthesis, in both cases target cells remained susceptible to T cell recognition for some time. We therefore asked how the observed rates of fall compared to the half-lives of pre-existing MHC I-peptide and MHC II-peptide complexes on the LCL surface. Thus pEBNA3B-tet and pEBNA1-tet LCLs of the appropriate MHC type maintained in the absence of dox were briefly exposed to a non- saturating dose of epitope peptide, washed well (time 0 h) and the subsequent fall in epitope display tracked over time by T cell assay. For comparison, all experiments included long-term-induced cultures of the same LCLs, from which dox was either withdrawn at time 0 h or maintained throughout. Figure 7 shows representative data obtained for pairs of epitopes from EBNA3B and from EBNA1. Both CD8 epitopes had half-lives on the LCL surface of 36–48 h; indeed the rate with which exogenously loaded CD8 peptides disappeared from the surface was only slightly faster than the rate at which CD8 epitope display fell following cessation of new antigen synthesis. However, both CD4 epitopes also had half-lives in the same range, the levels of display on peptide-pulsed cells therefore falling much quicker than seen on pEBNA3B-tet and pEBNA1-tet LCL cells after cessation of antigen synthesis. A similar pattern of results was observed for all CD8+ and CD4+ T cell epitopes tested (see for example Figure S3). Such results strongly suggest that, after the cessation of antigen synthesis, new CD4 epitope complexes continue to reach the cell surface whereas the supply of new CD8 epitope complexes is rapidly curtailed.
To test this further, we used a protocol (briefly exposing cells to citrate/phosphate buffer at pH 3.1) that efficiently strips pre-existing EBV epitope/MHC I and/MHC II complexes from the LCL surface without affecting cell viability. Having switched off antigen synthesis in pEBNA3B-tet and pEBNA1-tet LCLs by dox withdrawal, we followed the recovery of epitope peptide display by T cell recognition, stripping pre-existing epitopes off the cell surface either at the time of dox withdrawal (time 0 h) or 48 h later. The results from such assays are illustrated in Figure 8, again comparing CD8/CD4 epitope pairs from EBNA3B and from EBNA1. In each case, new epitope supply after stripping at time 0 h (blue line) or 48 h (red line) is shown against the level of surface epitope display seen on the same target cells that had been similarly dox-depleted at time 0 h but not stripped (black line). The CD8 epitopes showed significant recovery of cell surface display 24 h after stripping at time 0 h but then levels fell away rapidly, down to the same low values remaining on dox-depleted, non-stripped cells. When stripping was delayed until 48 h after dox-withdrawal, there was only a small recovery of CD8 epitope display, recapitulating the low residual values on non-stripped cells. By contrast, the CD4 epitopes showed a substantial recovery whether the cells were stripped at 0 h or 48 h following dox-withdrawal. Furthermore the recovery was sustained for up to 192 h, with stripped cells regaining the same persistent levels of CD4 epitope display as shown by non-stripped cells. Figure S4 shows the results of a similar experiment involving different target LCLs, here initially induced at 25 ng/ml dox, and effectors against different epitopes. This emphasises the point that consistent results were obtained for all CD8/CD4 epitope pairs tested, whether antigen was initially expressed at physiological or supra-physiological levels.
Here we address a generic question regarding pathogens, particularly viruses, that naturally infect target cells in which both the MHC I and MHC II pathways of antigen presentation are active. Antigens endogenously expressed within such an infected cell could potentially be presented by both pathways, rendering the cell visible to CD4+ as well as CD8+ T cells. However, the relative timing of those events and their degrees of dependence upon new antigen synthesis have never been rigorously examined in parallel. Our experimental system, based on EBV-infected B cell lines and the regulatable expression of EBV antigens, allows one to study these processes in a physiologically relevant cell context, select appropriate levels of antigen expression and track the presentation of CD8 and CD4 epitopes from the same source antigen after inducing or suppressing antigen synthesis. We studied five CD8 and three newly-defined CD4 epitopes from EBNA3B and two CD8 and two CD4 epitopes from EBNA1, in each case probing epitope display with at least two independent clones per epitope. To cover the wide range of MHC restricting alleles involved, assays were conducted on three different pEBNA3B-tet LCLs and two different pairs of pEBNA1-tet and pE1dGA-tet LCLs. The contrasting patterns of CD8 versus CD4 epitope display were remarkably consistent across the whole range of epitopes and antigens studied, and were reproducible whether the antigen was being expressed at physiologic (LCL-like) or supra-physiologic levels. We infer that these differences are not chance consequences of particular epitope selection but reflect fundamental differences in the way that endogenously expressed viral antigens are handled by the MHC I and MHC II presentation pathways in human B cells. At the same time, we would emphasise that both EBNA3B and EBNA1 are native nuclear proteins; there could possibly be differences in detail were one to study the processing of viral antigens normally resident in the cytoplasm or marked for export, but we would nevertheless expect the basic pattern of results to remain the same.
With antigen induction, we found that EBNA3B and both forms of EBNA1 were rapidly recognised by CD8+ T cells. Recognition was first apparent soon after dox addition and rose to almost maximal levels within 24 h, well before steady-state levels of these proteins, as detected by immunoblotting, were reached. The results with EBNA1 were particularly interesting given the history of work on this protein as a target for CD8+ T cells. Thus early studies found that the GAr domain was able to protect EBNA1 from presentation via the MHC I pathway [31] and that this was associated with resistance to proteasomal degradation [31],[34]. However, more recent results have shown that this protection from CD8+ T cell recognition is only partial [35],[36],[37],[38] and may reflect a GAr-mediated reduction of the rate of protein translation rather than of sensitivity to the proteasome per se [30],[32],[39],[40]. Importantly, many of these studies involved chimaeric antigen constructs, often with indicator epitope insertions, tested in in vitro translation or transient transfection assays, leaving the effects of the GAr domain in its physiologic setting open to question. In the present work we found that, after inducing antigen synthesis, E1dGA was presented to CD8+ T cells slightly quicker than the wild-type protein, though the magnitude of the effect was not as great as noted in other less physiologic experimental settings. We believe that our system is robust in this regard since we also found that CD8+ T cell recognition of cells induced to express E1dGA long-term was consistently 20–30% greater than seen with cells induced to express EBNA1. This exactly mirrors levels of EBNA1 epitope display seen earlier in LCLs transformed with EBV expressing a GAr-deleted EBNA1 protein versus LCLs transformed with wild-type virus [35].
Overall the results of the antigen induction experiments were consistent with MHC I presentation of newly synthesised polypeptides. However, in the same experiments, the MHC II-restricted presentation of EBNA3B and EBNA1 was grossly delayed; CD4 epitope display only became detectable after 36–48 h and took some 7 days to reach the long-term steady state level. This delay is not an intrinsic feature of MHC II processing in LCL cells since an invariant chain-targeted E1dGA protein expressed in the same dox-inducible system was detected by CD4+ T cells within 12 h and maximum recognition was reached within 48 h. This reinforces a large amount of earlier evidence testifying to the efficiency of MHC II antigen processing in LCL cells [41]. Our findings therefore imply that endogenously expressed antigens such as EBNA3B and EBNA1 are delivered very slowly into the MHC II processing pathway, even though they may access that pathway by different routes. Thus co-cultivation experiments showed that EBNA3B (like EBNA3A and 3C, [20]) is processed, at least in part, via the inter-cellular transfer of antigen between LCL cells. The precise form of antigen being transferred in LCL cultures is not known, except that it clearly requires active processing and, by analogy with our earlier work using donor cells transformed with a replication-defective EBV strain [20], does not derive from cells dying as a result of lytic virus replication. By contrast, the same experimental approaches never detected any evidence for inter-cellular transfer of EBNA1. Thus the CD4 epitopes recognised by our EBNA1-specific T cell clones must derive from antigen processed by an intracellular route. In that regard, others have also observed that endogenously expressed EBNA1 is processed intracellularly in LCL cells, and have suggested the involvement of autophagy in that process [17].
Dox-withdrawal from pre-induced LCLs allowed us to ask whether, in the absence of new antigen synthesis, the pre-formed intracellular pool of mature protein can feed the MHC I and MHC II pathways. We first verified that gene transcription from the dox-inducible promoter terminated rapidly after dox-withdrawal, with EBNA3B and EBNA1 transcript levels falling by >80% within 6 h and becoming undetectable by 12-24 h. New antigen synthesis must therefore be curtailed at least at the same rate yet, as is clear from the immunoblots, the pre-formed EBNA3B, EBNA1 and E1dGA proteins remain detectable for days thereafter. In this regard the natural turnover of EBNA3B has not been investigated previously, while ours is the first attempt to compare turnover of the EBNA1 and E1dGA proteins having switched off their synthesis specifically, rather than non-specifically with general protein synthesis inhibitors. Previous studies of the latter kind, where EBNA1 is first expressed by transient transfection or from recombinant viral vectors, all indicate that the wild-type protein has a long half-life, but differ in the degree to which this is shortened by GAr deletion [31],[32],[36],[42]. Our finding, that in the natural setting of the LCL cell both EBNA1 and E1dGA are stable proteins, accords with the most recent findings from transiently transfected cells with protein synthesis inhibitors [42]. For our present purpose, however, the essential point is that both EBNA3B and the two forms of EBNA1 are sufficiently stable that a large pool of mature protein persists in the cells for several days after the cessation of new antigen synthesis, providing a source of antigen that is potentially available to both MHC I and MHC II presentation pathways.
It is therefore significant that, upon dox-withdrawal, T cell assays showed a marked fall in cell surface display of all seven CD8 epitopes tested, typically to <50% of the initial level by 48 h and to <10% by 96 h. Indeed the rate of fall was in each case close to that seen when the corresponding epitope-negative LCL cells were loaded with epitope peptide at non-saturating levels and tracked over time to follow the natural half-life of the MHC I-peptide complex on the cell surface. These half-life measurements accord with earlier work, for example the RRA/B*2702 epitope from EBNA3B was estimated to have a half-life of 40 h in the present T cell assays and of 37 h in earlier antibody-based assays [43]. While rates of fall were similar on dox-depleted and peptide-pulsed cells, there was often a slight delay in the timing of that fall on dox-depleted cells. At least part of this lag must reflect the fact that, for a short time after dox-withdrawal, new MHC-peptide complexes either already in the export pathway or generated from residual mRNA translation will be delivered to the cell surface. Overall, the results strongly suggest that continued CD8 epitope display depends upon continued antigen synthesis. By contrast, T cell recognition of CD4 epitopes consistently fell much more slowly after dox-withdrawal, typically being still >50% of the initial level at 96 h and still easily detectable as late as 192 h. Recognition persisted despite the fact that in peptide pulsing experiments the relevant MHC II/CD4 epitope complexes have half-lives similar to their MHC I/CD8 epitope counterparts, strongly implying that the MHC II presentation pathway was being fed from the mature protein pool.
These conclusions were further supported by experiments in which cells were stripped of cell surface peptides after dox-withdrawal, and then assayed for the recovery of epitope display over time. Interestingly, cells stripped immediately after dox withdrawal showed a significant recovery of detection by CD8+ T cells 24 h later; however this effect, which could be quite marked for some epitopes, was transient with recognition falling away at later times. We attribute this transient recovery to the continued supply of newly-formed complexes to the cell surface occurring immediately after dox-withdrawal (as above) and possibly also to the reappearance of pre-existing mature complexes that were recycling from the surface at the time of stripping [44]. Importantly, cells stripped 48 h after dox removal, by which time surface epitope display was declining rapidly, showed minimal recovery of CD8 recognition. This strongly suggests that the mature protein pool, which is still substantial in cells 48 h after dox-withdrawal, makes little if any contribution to the MHC I presentation pathway. By contrast, CD4 epitope display was extensive and prolonged, whether cells were stripped immediately after dox-withdrawal or 48 h later. Such sustained presentation of CD4 epitopes by cells in which de novo synthesis of EBNA3B, EBNA1 and E1dGA was terminated 48 h earlier must reflect processing of antigen derived from the mature protein pool.
In summary, we find that in virally-infected human B cells newly-synthesised viral polypeptides, by inference rapidly degraded translation products, are the dominant source of antigen feeding the MHC class I pathway. This does not discount the possibility that the mature protein pool may, in other circumstances or in other cell types, contribute to such a role. Indeed, prompted by a report that irradiation could increase MHC I processing activity in cycloheximide-treated cells [45], we irradiated pEBNA3B-tet and pEBNA1-tet LCLs several days after dox-withdrawal and showed a small, transient recovery of CD8 epitope display that, in the absence of antigen synthesis, must have come from mature protein (L.K. Mackay, unpublished observations). However we find no evidence of any major contribution to the MHC I pathway from this source in a naturally proliferating LCL cell. By contrast, in these same cells endogenous antigen presentation via the MHC II pathway is dependent upon the mature protein pool and shows no immediate connection with the presence or absence of de novo translation products. These fundamental differences have important implications for virus-specific CD8+ and CD4+ T cells as direct effectors against infections of MHC I/II-positive target cells. In such circumstances, only CD8+ T cells have the capacity to recognise newly infected cells as soon as de novo antigen synthesis begins; CD4+ T cell recognition will be delayed until the intracellular antigen pool has increased sufficiently to feed the MHC II presentation pathway. Interestingly however, our results imply that for viruses establishing latent infections in MHC I/II-positive cells where viral gene expression is extinguished but where viral proteins persist, a situation that could for example pertain to gamma-herpesviruses and their genome maintenance proteins, the latently-infected cell reservoir may remain visible to CD4+ T cells.
All experiments were approved by the South Birmingham Local Research Ethics Committee (07/Q2702/24). All patients provided written informed consent for the collection of blood samples and subsequent analysis.
LCLs were established using the reference EBV strain B95.8, a B95.8-based recombinant lacking the EBNA3B gene (EBNA3B-KO) [23], or the Chinese CKL strain (called NPC 15, [46]) with a variant EBNA1 sequence. All lines were maintained in RPMI 1640 medium supplemented with 2 mM glutamine and 10% fetal calf serum (standard medium). A derivative of the dox-dependent expression vector pRTS-1 [27] and the EBNA3B, EBNA1 and E1dGA constructs were kindly provided by Dr J Mautner, Munich; in cases where EBNA1 was expressed under dox control, a derivative of pRTS-1 lacking constitutively expressed EBNA1 was used. Ii-tagged E1dGA and FLAG-tagged EBNA1 and E1dGA constructs were constructed by PCR, verified by DNA sequencing, then introduced into the vector by standard DNA cloning procedures. To introduce these into LCLs, DNA (15 µg) was transfected into 107 cells by electroporation in 300 µl Optimem (Invitrogen) at 230 V and 960 µF using a Biorad electroporation apparatus. Immediately after electroporation, cells were resuspended in RPMI 10% FCS and were incubated at 37°C and 5% CO2. After 24 h in culture, cells were then stained with rat CD2-specific antibody OX34 and were positively selected by magnetic cell sorting with anti-mouse IgG2a/b Microbeads and LS columns (Miltenyi Biotech) according to the manufacturer's guidelines. Cells were then expanded and maintained in culture in the absence of dox, before testing for dox- inducibility of antigen expression.
Total RNA was extracted from 5×106 cells using a Nucleospin RNA extraction kit (Macherery Nagel) according to the manufacturer's instructions. 400 ng RNA was reverse transcribed into cDNA using a pool of primers specific for EBNA3B, EBNA1/E1dGA and (as an internal control) cellular GAPDH transcripts. In subsequent quantitative PCR (Q-PCR) assays, primer/probe combinations were used to amplify (i) the 3′ end of the major EBNA3B exon, or (ii) the unique 5′ end of EBNA1/E1dGA transcripts initiated from the dox-regulatable promoter in plasmid pEBNA-tet. After normalising to GAPDH expression, levels of EBNA3B or EBNA1/E1dGA transcription in test cells are expressed relative to that of a fully induced cell line.
Cells were sonicated in UTB buffer (8 M urea, 150 mM β-mercaptoethanol, 50 mM Tris/HCl pH 7.5) and cellular debris removed by centrifugation. Protein concentration was determined by using the BioRad Bradford Protein determination reagent. Solubilized proteins were separated by 8% SDS-PAGE and transferred to nitrocellulose membranes (Thermo Scientific Pierce). Cellular and viral proteins were detected by incubating the membranes with specific Abs followed by HRP-conjugated secondary Abs (Sigma). Bound HRP was visualized using the ECL-plus detection kit (Amersham Biosciences). Antibodies used include: anti-EBNA3B (ExAlpha), anti-EBNA1 (IH4, [47]), and anti-actin (Sigma).
CD4 epitope peptides within EBNA3B were identified by screening immune donor lymphocytes in IFNγ Elispot assays on peptide panels (20-mers overlapping by 15) covering the primary sequence of B95.8 strain EBNA3B. All peptides were synthesized using 9-fluorenylmethoxycarbonyl chemistry (Alta Bioscience; University of Birmingham, Birmingham, U.K.), dissolved in DMSO, and concentrations were determined by biuret assay. CD4+ and CD8+ T cell clones specific for these and for other defined epitopes within EBNA 1 or EBNA3B were generated as described [29]. All epitope sequences are shown, with their MHC restricting alleles, in Table S1.
Immediately before all T cell assays, target LCL cells were fixed in 1% paraformadehyde for 10 min followed by quenching with 0.2 M glycine for 10 min, and then washed with PBS before resuspension in standard medium. Assays therefore measured the level of epitope display at a defined time point, with no further changes occurring during the 18 h assay period itself. Unless otherwise stated, fixed target cells were seeded at 105 cells per triplicate assay well, to which 2000 T cells were added; after 18 h incubation, supernatant medium was harvested and assayed for IFNγ release by ELISA (Endogen). Assays routinely included the following control targets: the wild-type B95.8 virus-transformed LCL from the same donor as the pEBNA-tet-transfected LCLs, the relevant pEBNA-tet-transfected LCL both without dox induction and long-term dox-induced, and the pEBNA-tet-transfected LCL without dox-induction but exogenously loaded with 10−7 M concentration of the relevant epitope peptide. In all assays, at least two different T cell clones were tested for each epitope specificity.
In assays measuring the half-life of peptide/MHC complexes at the cell surface, LCLs with relevant HLA types but transformed with EBNA3B-KO or CKL (variant EBNA1) virus strains were exposed for 1 h to epitope peptide at concentrations mediating half-maximal recognition, then washed several times and either fixed immediately for T cell assay, or cultured in standard medium then harvested and fixed for assay at later times. For assays measuring the continued supply of complexes to the surface from intracellular sources, we used an acid-stripping protocol that preliminary work confirmed would completely remove both MHC I- and MHC II-bound epitope peptides without affecting cell viability ([35] and L. Mackay, unpublished observations). Cells were washed with PBS and pellets were gently resuspended in citrate/phosphate buffer (0.131 M citric acid, 0.066 M Na2HPO4), pH 3.1, for 20 min on ice before neutralization by addition of excess standard medium. Stripped target cells were then washed several times and an aliquot of cells fixed immediately for T cell assay, while the remaining cells were re-cultured in standard medium, then harvested and fixed for assay at later times.
|
10.1371/journal.pntd.0000535 | The Production of Antibody by Invading B Cells Is Required for the Clearance of Rabies Virus from the Central Nervous System | The pathogenesis of rabies is associated with the inability to deliver immune effectors across the blood-brain barrier and to clear virulent rabies virus from CNS tissues. However, the mechanisms that facilitate immune effector entry into CNS tissues are induced by infection with attenuated rabies virus.
Infection of normal mice with attenuated rabies virus but not immunization with killed virus can promote the clearance of pathogenic rabies virus from the CNS. T cell activity in B cell–deficient mice can control the replication of attenuated virus in the CNS, but viral mRNA persists. Low levels of passively administered rabies virus–neutralizing antibody reach infected cells in the cerebellum of B cell–deficient mice but are not sufficient to mediate virus clearance. Production of rabies virus-specific antibody by B cells invading CNS tissues is required for this process, and a substantial proportion of the B cells that accumulate in the CNS of mice infected with attenuated rabies virus produce virus-specific antibodies.
The mechanisms required for immune effectors to enter rabies virus-infected tissues are induced by infection with attenuated rabies virus but not by infection with pathogenic rabies viruses or immunization with killed virus. T cell activities can inhibit rabies virus replication, but the production of rabies virus–specific antibodies by infiltrating B cells, as opposed to the leakage of circulating antibody across the BBB, is critical to elimination of the virus. These findings suggest that a pathogenic rabies virus infection may be treatable after the virus has reached the CNS tissues, providing that the appropriate immune effectors can be targeted to the infected tissues.
| Every year over 50,000 people die from rabies worldwide, primarily due to the poor availability of rabies vaccine in developing countries. However, even when vaccines are available, human deaths from rabies occur if exposure to the causative virus is not recognized and vaccination is not sought in time. This is because rabies virus immunity induced by the natural infection or current vaccines is generally not effective at removing disease-causing rabies virus from brain tissues. Our studies provide insight into why this is the case and how vaccination can be changed so that the immune response can clear the virus from brain tissues. We show that the type of immune response induced by a live-attenuated rabies virus vaccine may be the key. In animal models, live-attenuated rabies virus vaccines are effective at delivering the immune cells capable of clearing the virus into CNS tissues and promote recovery from a rabies virus infection that has spread to the brain while conventional vaccines based on killed rabies virus do not. The production of rabies-specific antibody by B cells that invade the CNS tissues is important for complete elimination of the virus. We hypothesize that similar mechanisms may promote rabies virus clearance from individuals who are diagnosed after the virus has reached, but not extensively spread, through the CNS.
| Rabies viruses spread from peripheral sites of entry to the central nervous system (CNS) tissues via axonal transport thereby bypassing the specialized features of the neurovasculature known as the blood-brain barrier (BBB). Once the virus reaches CNS tissues three alternative outcomes are likely: (1) the BBB remains intact and the infection is lethal due to the absence of an antiviral CNS immune response (2) immune effectors cross the BBB and mediate a CNS antiviral immune response with extensive immunopathology that contributes to the disease, or (3) immune effectors cross the BBB and clear the virus from the CNS without significant pathological consequences. It is well known that in humans naturally infected with rabies virus the latter outcome is exceedingly rare. In addition, CNS inflammation is generally limited in individuals who succumb to rabies [1]. Consequently, it is probable that the BBB remains intact through much of the course of rabies infection in humans. In the absence of a mechanism to compromise the barrier function of the neurovasculature, circulating rabies virus-specific immune effectors, whether raised by the infection or by active or passive immunization, would be unable to mediate an antiviral response in CNS tissues. This may explain why conventional post-exposure treatment of human rabies, consisting of active and passive immunization, is unsuccessful if begun after the appearance of signs of the disease [2]–[4]. At this stage of the infection the virus has likely begun to replicate in the CNS. Thus, the primary function of current post-exposure regimens may be limited to preventing the virus from reaching CNS tissues.
Unlike humans where rabies viruses may take weeks to reach the CNS from the site of exposure [5], the spread of most rabies virus strains to the CNS in mice is rapid with virus generally being detectable in CNS tissues within 48 hours of infection [6]. Nevertheless, normal mice survive infection with laboratory-attenuated strains of rabies virus [7]. While certain of these viruses may be deficient in the capacity to spread from the periphery to the CNS, most of the attenuated rabies virus variants that we have tested spread to and replicate in the CNS but are cleared by immune effectors that cross the BBB and infiltrate neural tissues [7]. In contrast, BBB integrity is maintained and immune effectors do not accumulate in the CNS tissues during infection of mice with common pathogenic rabies virus strains, despite the development of virus-specific immunity in peripheral lymphoid organs and innate immunity in the infected CNS tissues [7],[8]. These observations have led us to speculate that the lethal outcome of infection with wildlife and pathological strains of rabies virus is at least in part due to the evasion of immune clearance as a consequence of the maintenance of BBB integrity [8]. Perhaps the best evidence that this may be the case is that disruption of the BBB in mice infected with a highly lethal silver-haired bat-associated rabies virus (SHBRV), by triggering autoimmune CNS inflammation, promotes the clearance of the virus from the CNS tissues and survival [9]. Due to the associated pathology, the approach of using an autoimmune response to induce elevated BBB permeability and permit rabies virus-specific immune effectors to infiltrate CNS tissues is clearly inappropriate for use in human rabies. On the other hand, the neuroimmune response induced by infection with attenuated rabies virus, which also has the appropriate specificity, is not associated with significant pathology. In this study, we show that the functional changes in the BBB required to deliver immune effectors to the CNS tissues can be induced in mice infected with a lethal rabies virus strain by immunization with a live-attenuated virus vaccine strain but not by administration of killed virus vaccine. Furthermore, our data suggests that clearance of rabies virus from CNS tissues is dependent upon the production of virus-specific antibodies by infiltrating B cells.
Eight to 12-week old wild-type control 129/SvEv and C57BL6 mice and JHD−/− mice on a C57BL6 background were obtained from the in-house breeding colony at Thomas Jefferson University. RAG-2−/− mice on a 129/SvEv background were obtained from Taconic (Germantown, NY). Mice were infected or immunized intranasally (i.n.) with 105 focus forming units of CVS-F3, CVS-N2c or UV-inactivated CVS-F3 in PBS as previously described [10]. In some experiments, mice were infected and immunized with a combination of the viruses. Where indicated, CVS-F3-infected JHD−/− mice were treated intraperitoneally with 1 mg of the monoclonal, rabies virus glycoprotein-specific, virus-neutralizing antibody 1112 in 500 µl of saline or with the vehicle alone at the time points noted in the figure legends. All procedures were carried out according to the protocols approved by the Institutional Animal Care and Use Committee of Thomas Jefferson University.
BBB integrity was assessed by quantifying the leakage of a low molecular weight fluorescent marker (Na-fluorescein, 376 kDa) from the circulation into CNS tissues as previously described [10]. Briefly, 100 µl of 10% solution of Na-fluorescein was injected intraperitoneally and after 10 minutes mice were anesthetized and cardiac blood was collected followed by transcardial perfusion. Serum samples as well as supernatants of homogenized and centrifuged tissues were clarified by precipitating proteins with 15% TCA and the level of fluorescence measured with a CytoFluor™II fluorimeter. The amount of Na-fluorescein in the CNS tissue is normalized to its level in serum by (µg of Na-fluorescein in CNS tissue/mg of tissue)/(µg of Na-fluorescein in serum/µl of serum) and is expressed as a fold increase in fluorescence uptake by comparison with the results obtained from naïve controls.
For immunohistochemical analysis, brains from perfused mice were snap frozen in Tissue-Tek O.C.T. Compound (Sakura Finetex, Torrance, CA), sectioned using a Thermo Shandon cryostat (Pittsburgh, PA), and fixed in either 80% acetone or 95% ethanol. Immunoglobulin (Ig) was detected using either biotinylated monoclonal rat anti-mouse kappa light chain (1 hour at 1∶50) (BD Pharmingen, San Jose, CA) followed by Alexa Fluor 568 streptavidin (1 hour at 1∶1000) (Invitrogen, Eugene, OR) or the VECTASTAIN ABC-AP KIT with polyclonal rabbit anti-mouse biotinylated IgG (1∶200) developed using the peroxidase antiperoxidase method and 3′3-diaminobenzidine as substrate (Vector Laboratories, Burlingame, CA) according to the manufacturer's protocol. For the additional staining shown in one of the figure panels (#3D), a 1 hour incubation with 1 mg/ml 1112 was performed prior to detection of Ig. To assess virus infection sections were stained for 1 hour with FITC-conjugated anti-rabies virus nucleoprotein monoclonal antibody (1∶50) (Centocor, Malvern, PA). Photographs were taken with a Nikon digital camera on an Olympus BX-60 microscope.
Total RNA was isolated from CNS tissue samples and mRNA expression levels of rabies virus nucleoprotein, CD4, CD8, IFN-γ and L13 in CNS tissues were measured by quantitative reverse-transcriptase (RT)-PCR as previously described [10]. Real-time quantitative RT-PCR was carried out on cDNA using specific primer and probe sets and a Bio-Rad iCycler iQ Real Time Detection System (Hercules, CA). The number of copies of specific mRNAs in each sample was determined as previously described [10] and normalized to the mRNA copy number of the housekeeping gene L13 in that sample. Data are expressed as the number of copies of mRNA for a particular gene in a sample per copy of mRNA for the housekeeping gene L13 in that sample.
Mononuclear cells were prepared from peripheral blood collected by retro-orbital bleeding in heparinized capillary tubes by centrifugation at 300 g for 20 minutes. The white cell layer was washed in PBS twice before analysis. Mononuclear cells were isolated from CNS tissues as described elsewhere by isolation at the interface of a 30/70 Percoll (Sigma) step gradient centrifuged at 800 g for 25 minutes [11]. For flow cytometry, mononuclear cells were suspended in staining buffer (PBS with 2% FBS and 0.1% NaN2) and incubated with anti-CD16/32 (1 ug/106 cells) (2.4G2 BD Pharmingen, San Jose, CA) antibody to prevent non-specific binding. Cells were washed in PBS and incubated with anti-mouse CD19 (1∶1000) (1D3, BD Pharmingen, San Jose, CA) and MHC class-II (1∶1000) (120.1, BD Pharmingen, San Jose, CA) antibodies. Phenotypic characterization of antibody-labeled cells was performed on a BD-FacsCaliber Flowcytometer. CD19-MHC class II double-positive cells were defined as B cells. Numbers of rabies virus-antigen specific antibody secreting B cells were assessed using Millipore Multiscreen HA® ELISPOT plates coated with 5 ug/mL of UV-inactivated whole rabies virus. Peripheral blood or brain derived mononuclear cells were suspended in RPMI media supplemented with 25 mM HEPES and 10% FBS and 200,000 cells were incubated in each well for 18 hours. Plates were washed and bound rabies virus-specific antibodies were detected by addition of alkaline-phosphatase conjugated anti-mouse IgG antibody (1∶500) (Sigma, St. Louis, MO) followed by BCIP/NBT substrate. Spots were counted using a dissecting microscope.
Results are expressed as the mean±standard error mean (S.E.M.). Statistical significance of the differences between groups was tested using the Mann-Whitney test and the symbol * indicates a p value<0.05.
Mouse models are not considered to be particularly suited to studies of post-exposure prophylaxis (PEP) with rabies due to the rapid spread of the viruses to the CNS. However, our prior studies suggest that the lethal outcome of rabies in mice is more a consequence of the inability to deliver immune effectors into CNS tissues than its spread [8]. In our view, the key feature is that BBB integrity is maintained during infection with lethal rabies viruses, while infection with attenuated rabies virus variants causes enhanced BBB permeability and a virus-clearing CNS immune response [7],[10]. The reason for this difference could be that infection with highly pathological rabies virus strains causes the inhibition of immune mechanisms that mediate the changes in BBB function necessary for rabies-specific immune effectors to cross. Alternatively, these may not be triggered due to a subtle difference in rabies virus immune mechanisms induced by pathogenic and attenuated viruses. To distinguish between these two possibilities, mice were infected with the attenuated CVS-F3 variant or immunized with killed CVS-F3 and then 3 or 5 days later were super-infected with the pathogenic CVS-N2c rabies virus. CVS-F3 was administered first because the virus spreads to, and replicates in the CNS more slowly than CVS-N2c (data not shown). The delays were limited to 3 and 5 days so that the CVS-N2c infection would have the 48 hours required to spread to the CNS before the appearance of serum rabies virus-specific antibodies approximately 8 days following CVS-F3 infection [10]. To control for unanticipated effects caused by the administration of either immunogen, groups of mice were also given both live and inactivated CVS-F3. As shown in Table 1, administration of an inactivated CVS-F3 vaccine preparation that is effective when given several weeks before a CVS-N2c challenge does not protect when given 3 or 5 days prior to challenge. On the other hand, the majority of mice infected with CVS-F3 as recently as 3 days previously survive CVS-N2c infection regardless of whether or not inactivated virus is also administered. These results suggest that the processes required to clear pathogenic rabies virus from CNS tissues are induced by infection but not immunization with CVS-F3.
JHD−/− mice lack B cells but have functional T cells and, unlike RAG-2−/− mice, which lack both T and B cells, are capable of elevating fluid-phase BBB permeability in response to the infection (Fig. 1A). These mice are therefore suitable for analyzing the effects of antibody administration on CVS-F3 infection. As a preface to such studies we compared the course of CVS-F3 infection in JHD−/− and RAG-2−/− mice. As is the case for wild-type mice infected with CVS-F3 [10], both JHD−/− and RAG-2−/− mice lose weight as the infection progresses (Fig. 1A). However, while RAG-2−/− mice continue to lose weight and die approximately 20 days following infection, up to 70% of JHD−/− mice survive past this time-point, most showing a modest weight gain (Fig. 1B). At the same time virus replication, which continues to increase in RAG-2−/− mice, becomes reduced in the JHD−/− mice (Fig. 1C). These JHD−/− mice exhibit signs of rabies infection including ataxia and partial paralysis but survive the infection for extended periods of time ( >40 days). This raises the possibility that T cell activities may be able to partly control the virus infection independently of antibody. To gain insight into the contributing T cell subsets, we compared the levels of CD4, CD8 and IFN-γ mRNAs in CNS tissues from wild-type conventional mice 24 days after CVS-F3 infection when there is little virus replication remaining (see below) and JHD−/− mice 40 days after infection. As can be seen in Fig. 2, the levels of CD4 and CD8 mRNA are somewhat lower in the JHD−/− mice but IFN-γ mRNA levels have remained relatively high and may therefore be contributing to the control of virus replication.
The inability of JHD−/− mice to clear CVS-F3 from the CNS reaffirms the importance of rabies virus-specific antibodies in this process. However, little is known with respect to how these antibodies may be delivered to infected CNS tissues. Our studies of mice clearing CVS-F3 suggest that the leakage of naturally developing antibodies from the circulation into the CNS tissues may be minimal since elevated BBB permeability occurs before serum antibody titers peak [10]. In addition, over the short term, extensive fluid phase exchange across the BBB is seen but little accumulation of markers of the molecular mass of antibody is detectable [12]. Nevertheless, it may be expected that some antibody would cross the BBB in conjunction with infiltrating immune cells and, over time, sufficient levels may accumulate to impact virus replication. To examine this possibility, JHD−/− mice were treated with 1 mg of the mouse IgG1 monoclonal rabies virus-neutralizing, glycoprotein-specific antibody 1112, which is highly effective in post-exposure treatment models [13], on each of days 7 and 9 post-infection when BBB permeability is at a peak. Several hours later, CNS tissues were obtained and stained with antibodies specific for rabies nucleoprotein and for mouse IgG to determine if there was any antibody associated with infected cells. While extensive infection of Purkinje cells can be readily detected with nucleoprotein-specific antibodies in sections from the cerebellum of JHD−/− mice (Fig. 3A), as expected, there is no evidence of IgG in sections from animals that had not received antibody (Fig. 3B). IgG-specific staining of Purkinje cells in sections from mice receiving 1112 antibody could be detected (Fig. 3C) but the treatment of these sections with additional 1112 antibody in vitro prior to IgG detection resulted in more extensive staining (Fig. 3D). When cells stained for both nucleoprotein (Fig. 3E,G green) and antibody (Fig. 3F,G red) were examined more closely, distinct inclusions of nucleoprotein and antibody/glycoprotein can be seen. These findings suggest that low levels of 1112 antibody can leak from the circulation to interact with rabies virus-infected cells in the CNS tissues provided that their application coincides with elevated BBB permeability.
To determine whether 1112 antibody administration to CVS-F3-infected JHD−/− mice leads to the clearance of the virus from CNS tissues, we administered saline or 1 mg of the antibody 5 times at two day intervals between days 7 and 15 post-infection. This antibody dose regimen achieved a half-maximal serum rabies-specific antibody titer of approximately 1/240, which is roughly equivalent to the serum titer found in normal mice 8 days post-infection with CVS-F3, during the period of time when BBB permeability is maximal. Viral nucleoprotein mRNA levels in the CNS tissues of surviving animals, both saline and antibody treated, were substantial several weeks later (Fig. 4) at a time when they are virtually undetectable in wild-type mice [10]. Moreover, no impact on the health or survival of the mice was noted.
CVS-F3 clearance from the CNS tissues of wild-type mice occurs prior to the development of high titers of circulating virus-neutralizing antibodies (VNA) and after BBB permeability has peaked [10]. However, B cells that have infiltrated the CNS tissues express high levels of κ-light chain mRNA during this time period indicating that there is likely to be substantial antibody production in the CNS tissues [10]. To assess this possibility more directly, we used antibodies specific for mouse IgG to stain CNS tissues from wild-type mice infected 12 days previously with CVS-F3. Extensive foci of antibody are seen throughout the cerebellum (Fig. 5). At higher magnification the antibodies appear to be diffusing in stellate patterns from the foci (Fig. 5). To determine if B cells may be the source of these antibodies and whether or not they are likely to be rabies virus-specific, we assessed rabies virus-specific antibody production by B cells from the peripheral blood and CNS tissues of CVS-F3-infected mice. While the proportion of CD19+ B cells in mononuclear cells recovered from the CNS tissues of CVS-F3-infected mice is lower than in peripheral blood from the same animals, the fraction of the cells that produce rabies virus-specific antibodies is considerably higher (Fig. 6). This suggests that B cells producing rabies virus-specific antibodies either selectively invade or expand in the CNS tissues in response to CVS-F3 infection.
Prompt administration of PEP is the recommended course for an individual who has come in contact with a rabid animal. Since this prevents the development of clinical rabies it is impossible to be certain how many of the tens of thousands of people who receive PEP on an annual basis have actually been infected with the virus. It is also impossible to know how far the rabies virus may have spread before being cleared by the immune effectors provided or induced by PEP and the infection. The commonly held view that pathogenic wildlife rabies virus that has spread to the CNS cannot be cleared by immune mechanisms is supported by the absence of significant immune cell infiltration into the CNS tissues of individuals who die from rabies [1] and the failure of PEP in individuals that have developed signs of rabies [2],[4]. Our studies in animal models of rabies suggest that this is a consequence of the inability of virus-specific immune effectors to cross the BBB and enter CNS tissues infected with pathogenic rabies viruses [7],[8]. The rabies virus-specific immune effectors that are raised in lethally infected mice are able to clear rabies virus from the CNS if provided access across the BBB. For instance, when the BBB is compromised by the induction of autoimmune CNS inflammation, rabies-specific immune effectors infiltrate CNS tissues and can clear the highly pathogenic SHRBV [9]. In addition, the adoptive transfer of immune effectors recovered from mice lethally infected with SHBRV results in clearance of the attenuated CVS-F3 virus from the CNS tissues of mice lacking T and B lymphocytes [8]. In contrast, the transfer of cells from mice clearing CVS-F3 has no impact on the outcome of SHBRV infection [8]. Regardless of the infecting virus strain, elements of the innate immune response that are important for the early control of virus replication and for attracting immune cells into infected tissues are induced [7],[8],[10]. These findings led us to speculate that functional changes at the BBB required to provide immune effectors access to the CNS tissues are induced during infection with attenuated rabies virus strains but not during pathological rabies virus infection [7]–[9]. A key issue examined in this study is whether or not this is due to an inhibitory process triggered by infection with pathogenic rabies virus. If so, it may be expected that BBB integrity would be maintained during infection with both pathogenic and attenuated rabies viruses and the outcome would be lethal, but it is not. Infection with an attenuated rabies virus induces BBB integrity changes and immune effector entry into CNS tissues regardless of whether or not there is also an ongoing infection with pathogenic rabies virus. However, protection is not provided by immunization with killed virus. We therefore conclude that the generation of a rabies virus-specific immune response in the periphery is not sufficient to clear pathogenic rabies viruses from the CNS tissues. A mechanism selectively induced by infection with attenuated rabies virus, likely manifested at the BBB, is necessary to provide immune effectors access to CNS tissues.
To gain further insight into the mechanism of rabies virus clearance from the CNS tissues, we have used gene-deleted mice to study the role(s) of different antiviral immune effectors in the CNS tissues of mice clearing the attenuated rabies virus CVS-F3. Mice lacking T and B cells cannot clear this virus and die from the infection [8],[14]. CD8 T cells contribute to, but are not required for the clearance of CVS-F3 as clearance is merely delayed in mice without this cell population [14],[15]. On the other hand, JHD−/− mice, which lack B cells but have functional CD4 and CD8 T cells, often survive CVS-F3 infection over extended periods despite being unable to clear virus from CNS tissues and exhibiting neurological symptoms. This leads us to conclude that elements of the T cell response, likely including IFN-γ production by CD4 and CD8 T cells, can control certain features of the infection that make significant contributions to its lethality but that antibody is required for virus clearance. To examine the contribution of circulating antibody to virus clearance from CNS tissues, we administered high levels of the rabies virus neutralizing mouse monoclonal 1112 antibody to CVS-F3-infected JHD−/− mice during the stage of infection when BBB permeability is maximal. While leakage of a 150 kDa molecular weight marker from the circulation into the CNS tissues of CVS-F3-infected mice is minimal over a 4-hour period [12], antibodies present in the circulation over a more extensive period of time can evidently leak into the CNS tissues of the infected mice. 1112 antibody was found associated with the Purkinje cells in the cerebellum that express high levels of rabies virus antigen. The antibody was primarily localized in inclusion bodies which is consistent with previous in vitro studies showing that 1112 antibody is rapidly internalized by rabies virus-infected neuroblastoma cells where it accumulates in intracellular vesicles [13]. Of note in our studies is that the intracellular inclusions of glycoprotein-specific 1112 are generally distinct from inclusions of the virus nucleoprotein. The amounts of antibody reaching rabies virus-infected cells in vivo appears to be relatively low as considerably greater amounts of the antibody can bind to the cells when applied to tissue sections in vitro. While it is possible that even low levels of virus-neutralizing antibody may impact the replication and spread of the virus while BBB permeability is enhanced, treatment of CVS-F3-infected JHD−/− mice with 1112 antibody failed to clear the virus.
It should be noted with respect to the origin of the antibodies that participate in rabies virus clearance that serum rabies virus-specific antibody titers peak some time after BBB integrity has been restored [10]. The presence of cells expressing the B cell phenotypic marker CD19 and mRNAs specific for κ- light chain in the CNS tissues of mice clearing CVS-F3 [8],[10] led us to examine the possibility that rabies virus-specific antibodies are produced by infiltrating B cells. The current findings indicate that this is the case. Focal concentrations of antibodies can be readily detected in the CNS tissues of mice clearing CVS-F3 and a high proportion of B cells recovered from the tissues produce rabies virus-specific antibodies in vitro. This leads us to conclude that the high levels of antibodies required for rabies virus clearance from the CNS tissues are produced at the site of infection rather than diffusing in from the circulation. In this case, passively administered antibody during PEP would primarily impact virus in the periphery and an active immune response leading to elevated BBB permeability and immune effector delivery to the CNS tissues would likely be required to clear virus from the CNS.
As certain of the aspects of BBB function required for immune cell infiltration are unchanged by CNS infection with pathogenic rabies viruses [8], once the virus has reached the CNS a PEP protocol capable of altering the BBB, so that virus-specific immune effectors can reach the infected tissues, is required. Inactivated CVS-F3 can induce rabies virus-specific T and B cells, but fails to promote recovery from CVS-N2c infection over a time frame during which the administration of live CVS-F3 is therapeutic. We consider that this is a consequence of the inability of the inactivated virus to induce the functional changes in the BBB that are required for antiviral immune effectors to enter CNS tissues. In our view, administration of a live-attenuated rabies virus vaccine is the most reasonable, currently available, approach to providing the appropriate immune effectors access to the CNS tissues. The results of our experiments with a new, highly attenuated recombinant rabies virus vaccine which expresses three copies of a mutated glycoprotein gene, strongly support this hypothesis [16]. In these studies, the triple G vaccine was shown to promote immune effector delivery into CNS tissues and normal mice were found to survive the intracranial injection of a mixture of the vaccine virus and a highly pathogenic dog strain which was nearly 100% lethal when administered alone [16]. The triple G vaccine also proved effective in the post-exposure treatment of mice infected with a highly pathogenic dog rabies virus several days previously [16]. However, when UV-inactivated and given peripherally to mimic conventional post-exposure vaccination, there was little protective effect [16]. In addition to boosting the antiviral response, attenuated rabies virus vaccines spread to the CNS where they trigger the mechanisms required for T cells and B cells to enter the tissues and clear, not only the attenuated, but also pathogenic rabies viruses. It is clear from the commonly lethal outcome of rabies that these mechanisms are not induced in a timely fashion in the context of the spread of a wildlife rabies virus to the human CNS.
|
10.1371/journal.ppat.1002754 | African Swine Fever Virus Uses Macropinocytosis to Enter Host Cells | African swine fever (ASF) is caused by a large and highly pathogenic DNA virus, African swine fever virus (ASFV), which provokes severe economic losses and expansion threats. Presently, no specific protection or vaccine against ASF is available, despite the high hazard that the continued occurrence of the disease in sub-Saharan Africa, the recent outbreak in the Caucasus in 2007, and the potential dissemination to neighboring countries, represents. Although virus entry is a remarkable target for the development of protection tools, knowledge of the ASFV entry mechanism is still very limited. Whereas early studies have proposed that the virus enters cells through receptor-mediated endocytosis, the specific mechanism used by ASFV remains uncertain. Here we used the ASFV virulent isolate Ba71, adapted to grow in Vero cells (Ba71V), and the virulent strain E70 to demonstrate that entry and internalization of ASFV includes most of the features of macropinocytosis. By a combination of optical and electron microscopy, we show that the virus causes cytoplasm membrane perturbation, blebbing and ruffles. We have also found that internalization of the virions depends on actin reorganization, activity of Na+/H+ exchangers, and signaling events typical of the macropinocytic mechanism of endocytosis. The entry of virus into cells appears to directly stimulate dextran uptake, actin polarization and EGFR, PI3K-Akt, Pak1 and Rac1 activation. Inhibition of these key regulators of macropinocytosis, as well as treatment with the drug EIPA, results in a considerable decrease in ASFV entry and infection. In conclusion, this study identifies for the first time the whole pathway for ASFV entry, including the key cellular factors required for the uptake of the virus and the cell signaling involved.
| ASFV is a highly pathogenic zoonotic virus, which can cause severe economic losses and bioterrorism threats. No vaccine against ASFV is available so far. A strong hazard of ASFV dissemination through EU countries from Caucasian areas has recently emerged, thus making urgent to acquire knowledge and tools for protection against this virus. Despite that, our understanding of how ASFV enters host cells is very limited. A thorough understanding of this process would enable to design targeted antiviral therapies and vaccine development. The present study clearly defines key steps of ASFV cellular uptake, as well as the host factors responsible for permitting virus entry into cells. Our results indicate that the primary mechanism of ASFV uptake is a macropinocytosis-like process, that involves cellular membrane perturbation, actin polarization, activity of Na+/H+ membrane channels, and signaling proceedings typical of the macropinocytic mechanism of endocytosis, such as Rac1-Pak1 pathways, PI3K and tyrosine-kinases activation. These findings help understanding how ASFV infects cells and suggest that disturbance of macropinocytosis may be useful in the impairment of infection and vaccine development.
| ASFV is a 200 nm large DNA virus that infects different species of swine, causing acute and often fatal disease [1]–[3]. Infection by ASFV is characterized by the absence of a neutralizing immune response, which has so far hampered the development of a conventional vaccine. A strong hazard of ASFV dissemination through EU countries from Caucasian areas has recently emerged, thus making progress of knowledge and tools for protection against this virus urgent.
Analysis of the complete DNA sequence of the 170-kb genome of the Ba71V isolate, adapted to grow in Vero cells, has revealed the existence of 151 genes, a number of enzymes with functions related to DNA replication, gene transcription and protein modifications, as well as several genes able to modulate virus-host interaction [4]–[12].
ASFV replicates within the host cell cytosol, although a nuclear step has been reported [13], [14]. Discrete cytoplasmic areas are reorganized into viral replication sites, known as factories, during the productive virus cycle. Regarding this, we have recently described ASFV replication as fully dependent on the cellular translational machinery since it is used by the virus to synthesize viral proteins. Thus, during infection, factors belonging to the eukaryotic translational initiation complex eIF4F are phosphorylated, and then redistributed to the periphery of the ASFV factory. Furthermore, ASFV late mRNAs, ribosomes and mitochondrial network were also located in these areas [15]. Such phosphorylation events and redistribution movements suggest, first, a reorganization of the actin skeleton induced by ASFV, and second, virus-dependent kinases activation mechanisms. Several other critical steps of the infection, probably including virus entry and trafficking, might be also regulated by phosphorylation of key molecules targeted by the virus.
As the first step of replication, entry into the host cell is a prominent target for impairing ASFV infection and for potential vaccine development. Endocytosis is a major pathway of pathogen uptake into eukaryotic cells [16]. Clathrin-mediated endocytosis is one of the best studied receptor-dependent pathways, characterized by the formation of clathrin coated pits of 85–110 nm in diameter that bud into the cytoplasm to form clathrin-coated vesicles. Relatively low size viruses, as Vesicular stomatitis virus, Influenza virus, and Semliki forest virus all enter their host cells using this mechanism [17]–[19]. On the other hand, the caveolae-mediated pathway is dependent on small vesicles termed caveolae (50–80 nm) enriched in caveolin, cholesterol, and sphingolipid. It has been implicated in the entry of other small viruses such as Simian virus 40 [20].
Macropinocytosis is another important type of endocytic route used by several viruses to enter host cells. It is defined as an actin-dependent endocytic process associated with a vigorous plasma membrane activity in the form of ruffles or blebs induced by activation of kinases and Rho GTPases. This pathway involves receptor- independent internalization of fluid or solutes into large uncoated vesicles sized between 0.5–10 µm called macropinosomes [21], [22]. In recent years, it has been reported that macropinocytosis is responsible for virus entry of Vaccinia virus (VV) [23], [24], Coxsackievirus [25], Adenovirus-3 [26], Herpes simplex virus [27]–[29], and is required for other viruses to promote viral internalization after entry by some different endocytic mechanism [30]–[32].
Regarding ASFV entry, preliminary studies were reported many years ago by our lab describing this process as temperature, energy, cholesterol and low pH-dependent, and also showing that ASFV strain Ba71V enters Vero cells by receptor-mediated endocytosis [33]–[37]. However, the cellular molecules involved and the precise mechanisms for ASFV entry remain largely unknown.
A recent paper [38] reported that ASFV uses dynamin and clathrin-dependent endocytosis to infect cells. However, it is noteworthy that this work employed the expression of ASFV early proteins as readout of virus entry, which is not equivalent to virus uptake, since several post-entry events could be involved in virus early protein expression. Hence, explanation of several controversial points, such as the larger size of ASFV (200 nm) compared to the smaller size (50–80 nm) of clathrin coated pits, or the existence of several other possible roles for dynamin in addition to virus entry [39], are not discussed in that work.
In the present work we have characterized the mechanisms of entry of ASFV-Ba71V and ASFV-E70 strains either in Vero or swine macrophages, as representative models for ASFV infection. By means of a combination of pharmacological inhibitors, specific dominant-negatives and confocal and electron microscopy, we show that ASFV is taken up predominantly by macropinocytosis. Therefore, we provide evidence, for the first time, that the ASFV entry requires sodium/proton exchanger (Na+/H+), activation of EGFR and PI3K, phosphorylation of Pak1 kinases together with activation of Rho-GTPase Rac1 and relies on actin-dependent blebbing/ruffling formation, all events fully linked with macropinocytosis activation.
Vero (African green monkey kidney) cells were obtained from the American Type Culture Collection (ATCC) and grown in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 5% fetal bovine serum (Invitrogen Life Technologies). IPAM cells (porcine macrophage-derived cell lines) were kindly provided by Dr. Parkhouse (Fundaçao Calouste Gulbenkian - Instituto Gulbenkian de Ciência, Oeiras, Portugal) and grown in RPMI 1640 medium supplemented with 10% fetal bovine serum. Cells were grown at 37°C under a 7% CO2 atmosphere saturated with water vapour in a culture medium supplemented with 2 mM L-glutamine, 100 U/ml gentamicin and nonessential amino acids. The Vero-adapted ASFV strain Ba71V and isolate E70 were propagated and titrated by plaque assay on Vero cells, as described previously [40], [41]. In brief, subconfluent Vero cells were cultivated in roller bottles and infected with ASFV at a multiplicity of infection (MOI) of 0.5 in DMEM 2% fetal bovine serum. After 72 h post infection the cells were recovered and centrifuged at 3000 rpm for 15 min and the cellular pellet was discarded. The supernatant containing viruses was clarified at 14000 rpm for 6 h at 4°C and the purified infectious virus was resuspended in medium and stored at −80°C. Vero cells were infected with Ba71V isolate and IPAM cells with E70 or Ba71V as indicated. The MOI used ranged from 1 to 3000 pfu/cell, as explained.
Viral adsorption to cells was performed at 4°C (synchronic infection) or at 37°C (asynchronic infection) during 90 min (or 60 min when indicated), followed by one wash with cold PBS, and a shift to 37°C to allow the infection until indicated times.
Pharmacological inhibitors were prepared either in water or DMSO following the manufacturer's recommendation and used at the indicated concentration. 5- ethylisopropyl amiloride (EIPA), Cytochalasin D (Cyto D), Genistein, IPA-3, Chlorpromazine (CPZ), Dynasore (Dyn) and Nocodazole were purchased from Sigma. ±Blebbistatin, EGFR inhibitor (324674) and Rac1 inhibitor (Rac1 Inh, NSC23766) was purchased from Calbiochem, and LY294002 (LY) from Echelon.
Specific antibodies against Akt, phospho-Akt (Thr308), phospho-Akt (Ser473) and PI3K p85 were purchased from Cell Signaling Technology; anti-Pak1, anti- phospho-Pak1 (Thr423), anti-Rock1 and anti-β-actin from Santa Cruz Biotechnology, Inc. Rac1 was detected with a monoclonal antibody from Millipore, kindly provided by Dr. C. Murga (CBMSO, Madrid, Spain). Monoclonal anti-p72 (17LD3) [42] was a kind gift from Ingenasa and polyclonal antibodies risen against p72, p32 and most of the ASFV structural proteins (anti-ASFV) were generated in our laboratory. Alexa Fluor 594-WGA, TRITC- phalloidin, Alexa Fluor 488-phalloidin, Topro3, anti-mouse Alexa Fluor-488, anti-goat Alexa Fluor-555 and anti-mouse Alexa Fluor-555 were purchased from Invitrogen, and anti-rabbit, anti-mouse and anti-goat immunoglobulin G coupled to peroxidase from Amersham Biosciences.
GFP-tagged versions of wild type forms of actin (pEGFP-actin) and Rac1 (pEGFP-Rac1) were kindly provided by Dr. J. Mercer (ETH Zurich, Institute of Biochemistry, Zurich, Switzerland) and Rac1 mutant form (pGFP-Rac1-N17) was a generous gift from Dr. R. Madrid (CBMSO, Madrid, Spain). GFP-tagged versions of WT, AID, and T423E of Pak1 constructs were a gentle gift from Dr. J. Chernoff (Fox Chase Cancer Center, Philadelphia, PA, USA) and pEGFP-C2 was purchased from Invitrogen.
To analyze ASFV uptake, Vero cells were pretreated with the pharmacological inhibitors listed above at 37°C for 60 min in serum free medium. Ba71V synchronic infection was carried out at a MOI of 10 pfu/cell in the presence of the drugs. After binding, cells were washed once with cold PBS, followed by the addition of containing drug medium, and infection was allowed to proceed for 60 min at 37°C. After infection, cells were fixed and prepared either for Fluorescence Activating Cell Sorting (FACS) or Confocal Laser Scanning Microscopy (CLSM) analysis.
The specific effect of the drugs on virus entry and post entry steps was analyzed by incubation of the cells either 60 min before virus addition or 60 min after virus addition, and viral infection was allowed in the presence of the drugs at 37°C, in each case. Ba71V or E70 asynchronic infection was carried out for 16 or 48 h at a MOI of 1 pfu/cell or at a MOI of 5 pfu/cell to analyze viral proteins by Western blot or number of infected cells by CLSM, respectively.
To analyze Akt phosphorylation upon ASFV infection, Vero cells were infected at a MOI of 10 pfu/cell and viral adsorption was allowed for 60 min at 37°C. Actin distribution analysis was performed at different times post infection since virus addition at 37°C at MOI 50. Rac1 distribution and Pak1 phosphorylation was measured after synchronic infection at a MOI of 10 pfu/cell. At the indicated times, cells were prepared for Western blot or CLSM analysis.
Vero cells were pretreated with DMSO or pharmacological inhibitors for 60 min at 37°C. The asynchronic infection was carried out at a MOI of 1 pfu/cell for 48 h in the presence of the inhibitors and the supernatant was recovered. The number of productive viral particles was titrated by plaque assays on Vero cells as described in [41].
Cells were grown on glass coverslips, serum starved for 24 h, infected synchronously (MOI 50) and at the indicated times post infection, fixed in 2.5% glutaraldehyde and 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) for 3 h at 4°C. They were washed three times in phosphate buffer, postfixed in 2% OsO4/water at RT for 60 min, washed in water, dehydrated in acetone, critical point dried for 2 h and coated with graphite-gold in a sputter coater. The samples were analyzed with a JSM-6335-F (JEOL) Field Emission SEM (Electron Microscopy National Center, UCM; Madrid, Spain).
Vero cells were serum starved 24 h and virus binding was allowed for 90 min at 4°C with Ba71V (MOI 3000). Cells were fixed with 2% glutaraldehyde and 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) for 3 h at 4°C. Sections of infected cells were prepared as described [43] and analyzed in a JEOL 100B electron microscope.
In order to study real-time live imaging of ruffles formation induced by ASFV infection, Vero cells were serum starved for 24 h and virus binding was allowed for 90 min at 4°C at MOI 100. After binding, cells were washed with cold PBS and images were collected for 30 min with an Orca R2 digital camera (Hamamatsu) on a wide-field microscope (LeicaDMI6000B, Leica Microsystems) with controlled environmental chamber (temperature 37°C and 5% CO2 humidified atmosphere). Images were captured with LAS AF version 2.6.0 software (Leica Microsystems) at a resolution of 1344×1024 pixels using a 20×, 0.40 NA objective with a 1.6× magnification-changer, and analyzed with Image J software.
To analyze blebs formation, IPAM cells were infected synchronously (MOI 50) and at different times post infection, fixed with paraformaldehyde 4% for 20 min. Images were taken with a ccd monochrome camera (Hamamatsu) on a invert microscope (Axiovert200, Zeiss) using a 63× objective and analyzed with Image J program.
Mock-infected or infected cells in the presence of pharmacological inhibitors were detached with trypsin-EDTA after 60 min post infection (mpi), fixed with 2% paraformaldehyde for 30 min at 4°C and then permeabilized with PBS-Staining buffer (PBS 1×, 0.01% sodium azide, 0.5% BSA) 0.2% saponin for 15 min at RT. Detection of infected cells was performed by incubation with an anti-p72 monoclonal antibody (17LD3) (diluted 1∶100 in PBS-Staining buffer 0.2% saponin) for 20 min at 4°C, followed by incubation with an anti-mouse Alexa Fluor-488 (diluted 1∶500 in PBS-Staining buffer, 0.2% saponin) in the same conditions. Finally, 2×104 cells were analyzed in a FACSCalibur flow cytometer (BD Science) to determine the percentage of infected cells. All FACS analyses were performed at least in triplicate and displayed as the average percentage of infected cells relative to control infection in the absence of a pharmacological inhibitor. Error bars represent the standard deviation between experiments.
Cells were grown on glass coverslips and, at indicated times post infection, were fixed with 4% paraformaldehyde for 20 min and permeabilized with PBS-0.2% Triton X-100 for 30 min at RT. Viral particles or infected cells were stained with an anti-p72 monoclonal antibody (17LD3) (diluted 1∶250 in PBS-5% BSA) for 60 min at RT, followed by incubation with an anti-mouse Alexa Fluor-488 or an anti-mouse Alexa Fluor-555 for the same time. Alexa Fluor-488 phalloidin (dilution 1∶100) or TRICT- phalloidin and Topro3 (dilution 1∶500) were used to stain actin filaments and nuclei of cell, respectively. Goat anti-Rock1 was used at a dilution 1∶50.
To analyze the virus binding to the cellular membrane, the viral adsorption was allowed for 90 min at 4°C (MOI 10) and after 60 min from virus addition cells were incubated with Alexa Fluor 594-WGA for 30 min. Cells were washed twice with cold PBS-0.1% BSA Buffer and incubated with anti-p72 monoclonal antibody (17LD3) and Alexa Fluor-488 for 60 min at 4°C. Finally, cells were fixed with 4% paraformaldehyde at RT for 20 min.
Samples were analyzed by CLSM (Zeiss LSM510) with a 63× oil immersion objective. To investigate ASFV uptake as well as actin, Rock1 and Rac1 distribution, Z-slices per image were collected and displayed as maximum z-projection of vertical slices (x–z plane) and/or maximum z-projection of horizontal slices (x–y plane). For presentation of images in the manuscript, LSM images were imported into Image J software for brightness and contrast enhancements. In all instances one image is representative of three independent experiments.
ASFV uptake in the presence of inhibitors was analyzed automatically by a Macro algorithm from Image J program (developed by CBMSO Confocal Microscopy Service, Spain) in which Intermode threshold was used to count the number of virus inside the cells.
Vero cells were serum starved for 24 h and pretreated with DMSO or EIPA. After 60 min at 37°C the cells were synchronously infected (MOI 10) or treated with PMA (200 nM) at 37°C for 30 min. Fifteen min prior to harvesting or fixation, cells were incubated with 0.5 mg/ml 10 KDa 647-dextran or 3 KDa Texas Red-dextran (Invitrogen) at 37°C. Dextran uptake was stopped by placing the cells on ice and washing three times with cold PBS and once with low pH buffer (0.1 M sodium acetate, 0.05 M NaCl, pH 5.5) for 10 min. Then, the cells were prepared for FACS or CLSM analysis. In FACS experiments dextran uptake was displayed as fluorescence mean of three independent experiments. Error bars represent the standard deviation between experiments. Cells without wash acid buffer were added as an experiment control.
At indicated times post infection, cells were washed with PBS and lysed in RIPA modified buffer (50 mM Tris-HCl pH 7.5, 1% NP40, 0.25% Na-deoxycolate, 150 mM NaCl, 1 mM EDTA) supplemented with protease and phosphatase inhibitor cocktail tablets (Roche). The protein concentration was determined by a Pierce BCA Protein Assay kit based on the bicinchoninic acid spectrophotometric method (Thermo Scientific). Cell lysates (15–50 µg of protein) were fractionated by SDS-PAGE and electrophoretically transferred to an Immobilon extra membrane (Amersham) and the separated proteins reacted with specific primary antibodies. The antibodies used were the following: polyclonal anti-p72 (dilution 1∶2000), anti-p32 (dilution 1∶1000), anti-ASFV (dilution 1∶3000); anti-Akt, anti-phospho Akt (Thr308), anti-phospho Akt (Ser473), anti-Pak1 and anti-phospho Pak1 (Thr423) (dilution 1∶500); anti-β-actin and anti-Rac1 (dilution 1∶1000). Membranes were exposed to horseradish peroxidase-conjugated secondary antibodies (dilution 1∶5000) followed by chemiluminescence (ECL, Amersham Biosciences) detection by autoradiography. In all instances the figures are representative of three independent experiments.
Vero cells were serum starved for 24 hours and treated with DMSO or LY294002 for 60 min at 37°C in serum free medium. Asynchronic infection (viral adsorption for 60 min) was carried out at a MOI of 10 pfu/cell in the presence of the drug at 37°C until indicated times. PI3K subunit p85 was immunoprecipitated from lysed cells and PI3-kinase activity was measured as PI(3,4,5)P3 production by ELISA activation kit, following the manufacturer's recommendations (Kit#1001s Echelon).
Vero cells were serum starved for 24 hours before synchronic infection at a MOI of 10 pfu/cell. The cells were washed once with cold PBS, shifted to 37°C and harvested at the indicated times post infection. Rac1 activation was measured with a G-LISA activation kit (Kit #BK128 Cytoskeleton, Inc.) and by immunoblotting after a Pak1-PBD-Agarose Beads (Upstate) pull down step as described following the manufacturer's recommendations. Bound Rac1-GTP was detected by incubation with an anti-Rac1 specific antibody followed by a secondary antibody conjugated to HRP and a detection reagent. The signal was read by measuring absorbance at 490 nm using a microplate reader and by autoradiography.
To check if the EIPA inhibitor was specifically blocking virus entry and not a down-stream process such as early gene expression, we induced the fusion of the viral membrane with the plasma membrane (PM) by lowering the pH of the medium [23]. The cells were pretreated with EIPA for 60 min at 37°C in serum free medium. Viral adsorption was allowed at MOI 1 for 90 min at 37°C in neutral (7.4) or acid (5.0) pH. Cells were washed once with cold PBS and infection was allowed to proceed for 16 h at 37°C in the presence of the inhibitor in neutral pH. Samples were prepared for Western blot analysis.
Vero cells were transfected with 1 µg of specific expression plasmids per 106 cells using the LipofectAMINE Plus Reagent (Invitrogen) according to the manufacturer's instructions and mixing in Opti-MEM (Invitrogen) in a 6-well plate. Cells were incubated at 37°C for 4 h in serum free medium, washed and incubated at 37°C. After 16 or 24 h post transfection the cells were infected at indicated MOI and either lysated and analyzed by Western blot, or fixed and prepared for CLSM analysis.
In order to analyze the localization of p72 in the viral particle, we carried out an experimental procedure as described in [44]. In brief, purified virus was treated with different buffers (Buffer 1 and 2) for 30 min at RT and the separate samples were centrifugated over sucrose (20% in PBS) cushion in a Beckman Airfuge at 24 p.s.i. for 15 min. The supernatant (SP) and pellet (P) were analyzed by Western blot by incubation with anti-p72 monoclonal antibody (17LD3). Buffer 1: 10 mM Tris pH 8, 0.65 M NaCl, 0.5% octil β- D- Glucopyranoside (Sigma); Buffer 2: 10 mM Tris pH 8, 0.65 M NaCl, 0.2% octil β- D- Glucopyranoside (Sigma) and 0.1% Dithiothreitol (DTT).
To check cell viability after treatment with inhibitors cells were dyed with Trypan Blue and dead cells were counted in hemocytometer as blue cells.
After Western Blot analysis, bands developed by ECL chemiluminescence were digitalized by scanning and quantified with Fujifilm Multi Gauge V3.0 software. Data were normalized after subtracting background values and calculated as factors by their ratio against the highest or lowest positive value obtained. All quantifications represent the mean of three independent experiments.
ASFV proteins in Swiss Prot database: p72: MCP_ASFB7; p32: P30_ASFB7; p17: P17_ASFB7; p12: P12_ASFB7.
Cellular proteins in ENSEMBL database: Pak1: ENSMMUG00000001387; Rac1: ENSFM00250000002337; β-actin: ENSMMUG00000012054; Akt: ENSMMUG00000001041; Rock1: ENSFM00540000717933.
Macropinocytosis mainly differs from other endocytic processes in the requirement of extensive actin cytoskeleton restructuring and formation of blebs or ruffling in the cellular surface, through which the specific cargo enters the cell [22]. These rearrangements are coupled to an external-induced formation of plasma membrane extensions. Several viruses have been described to use macropinocytosis for entry, including Vaccinia virus [23], [45], [46], Ebola virus [47] and Kaposi's sarcoma-associated herpesvirus [29], [48].
Receptor-mediated endocytosis has been postulated in classic studies as the most likely mechanism for ASFV entry into Vero cells [33]–[35]. Yet the specific characteristics to further depict the viral entry procedure have not been elucidated. To analyze the possible perturbation of the cellular membrane induced by ASFV, the virus strain Ba71V was used to synchronously infect Vero cells at MOI 50. To achieve this, we have analyzed by Field Emission SEM analysis (FESEM) the induction of ruffling and bubbles-like perturbations at 10, 60 and 90 min after ASFV uptake. The results are shown in Figure 1A, where a maximum level of membrane perturbation similar to ruffles appears in ASFV-infected Vero cells between 10 and 60 mpi, decreasing after 90 mpi, indicating that ASFV-induced macropinocytosis is a transient event.
On the other hand, Figure 1B shows that ASF virions internalize in Vero cells adjacent to retracting ruffles, thus indicating that the macropinocytic uptake of viral particles seems to occur as part of the macropinocytic process.
Finally, we have analyzed in vivo in real-time the membrane protrusions observed during Ba71V infection in Vero cells. Figure 1C shows the sequence of images during the first minutes of the infection (Video S2), illustrating the ASFV-induced ruffling, and in concordance with the data shown in Figure 1A. For comparison to Mock-infected Vero cells, see Video S1.
To assess whether the ASFV entry also induces membrane perturbation in swine macrophages, the natural target cell of ASFV infection in vivo, the virulent strain E70 was used to synchronously infect IPAM cells at MOI 50. As early as 10 mpi, strong membrane protrusions were observed by FESEM analysis (Figure 2A). To better characterize these membrane rearrangements, IPAM cells were synchronously infected with E70 strain at MOI 50, during 30, 45 and 60 mpi. Next, IPAM cells were fixed and analyzed by optic microscopy. Figure 2B shows images compatible with blebs induced by ASFV infection in swine macrophages from 30 mpi. To prove this point, we achieved an additional experiment showing the inhibition of virus entry with different doses of blebbistatin, an inhibitor of blebbing and macropinocytosis [23], [29], [49]–[51]. Western blot analyses have shown that blebbistatin impairs the entry of the virus in IPAM cells, as the drug inhibits the expression of ASFV proteins when preincubated before virus addition. Hence, when blebbistatin was incubated 60 min after virus addition, a much lower inhibition of viral proteins was observed, thus indicating the role of blebbistatin on early steps of virus entry. Results are presented in Figure 2C.
Last, by using a specific anti-Rock1 antibody as a marker of blebs [52], we have shown that Rock1 colocalizes with virus particles on blebs in IPAM cells from 30 min after ASFV uptake (Figure 2D), revealing the close relation between bleb and viral particle.
Taken together, these data strongly indicate that ASFV induces a vigorous plasma membrane activity during the first steps of the infection, both in Vero and IPAM cells, well-matching with macropinocytosis-mediated entry.
With the membrane perturbation pattern shown above, it was likely that ASFV was using macropinocytosis to enter cells. Macropinocytosis is dependent on the Na+/H+ exchanger [21], and thus amiloride and its analogue 5-(N-ethhyl-n-isopropil)-amiloride (EIPA) are frequently used as the main diagnostic test to identify macropinocytosis because this drug has been shown to be specific to this endocytic pathway without affecting others [53]–[55]. Consequently, to further assess the involvement of macropinocytosis in ASFV entry, the effect of EIPA was investigated. When tested on Vero cells, EIPA had no significant cytotoxic effect as assessed by cell monolayer integrity and trypan blue cell viability assessment (Table S1).
It has been previously described that after 60 mpi more than 90% of the ASF viral particles are located in the cell [34]. Furthermore, the viral uncoating does not completely occur before 2 hours post infection (hpi) [34]. According to these data, we measured viral uptake by using the specific antibody 17LD3 against p72, the major protein of ASFV capsid [42], [56] (see Materials and Methods and Figure S1A, B and C). Interestingly, amounts of EIPA from 40 µM to 60 µM caused a significant reduction (60%) in the uptake of ASFV infective particles after 60 mpi (Figure 3A), suggesting that ASFV entry depends on Na+/H+ exchanger activity/function.
To further visualize the effect of EIPA on virus uptake, Ba71V strain was added to Vero cells, previously treated with DMSO or 60 µM EIPA. Sixty min after infection, the cells were incubated with anti-p72 antibody 17LD3 to stain the virus. A confocal microscopy analysis revealed that there was a noticeable drop in virus particles incorporated into the cells incubated with EIPA, as compared to those incorporated into DMSO-incubated cells (Figure 3B, bottom panels). Images were taken as a maximum z-projection (x–y plane). For clarification, individual channels are shown in Figure S2A. Moreover, we also analyzed images of a maximum z-projection of vertical slices to determine whether viral particles could be imbibed into the membrane in the presence of the inhibitor. As shown in the Figure 3B upper panels, a different distribution of viral particles in the cells infected in the presence of EIPA, compared to that found in cells infected in the absence of the drug, was observed. This last data strongly suggests that in EIPA-treated cells the virus can bind to the membrane but is not able to internalize. This could be the explanation for the percentage of cells that were positive for 17LD3 antibody detected in Figure 3A. The total number of virus obtained in the confocal images was automatically quantified using a macro algorithm in the Image J program (Figure S3).
In regard to this, it is also remarkable that, although a small amount of viral particles can still be detected inside the cells in the presence of EIPA, neither early, p32, nor late ASFV proteins, p17, p24, p12 and p72 [57]–[60] could be detected by Western blot in the presence of the drug (Figure 3C). Hence, it is likely that EIPA is mainly affecting virus uptake since when drug is added 60 min after virus uptake, it does not affect the viral protein synthesis (Figure S4A). As expected, no viral factories detected by using anti-p72 antibody (green) and Topro3 (blue) for viral and cellular DNA, were found after EIPA treatment by confocal microscopy (Figure 3D). Separate channels are shown in Figure S2B and a morphological detail of an ASFV factory is shown in Figure S1C. Consequently, viral production was also strongly inhibited by the drug (Figure 3E). Finally, and to fully ascertain if EIPA was specifically blocking ASFV entry and not a downstream step, we performed the infection by using the acid-mediated fusion of plasma membrane. Briefly, in the presence of acid pH, endocytosis is subverted and virions fused with the plasma membrane and then directly carried into the cytosol. When an inhibitor blocks virus endocytosis, inhibition of viral protein synthesis in the presence of drug can be bypassed through fusion. If membrane fusion could not rescue viral gene expression, the blocking would most probably occur at a post-entry step [23]. By using this method, we find that when the viral adsorption is performed in the presence of EIPA in acidic pH, p72 viral synthesis is clearly recovered in relation to the infection developed at neutral pH (Figure 3F).
Next, we investigated the dextran uptake during ASFV infection, since it has been described that macropinocytosis activation induces a transient increase of this fluid phase marker [61], [62]. To achieve this, Vero cells were treated with EIPA for 60 min and then infected synchronously with Ba71V for 30 min, or stimulated with PMA as a positive control. Fifteen minutes before stopping the infection, cells were pulsed with dextran and prepared for FACS analysis. As indicated in Figure 3G, ASFV infection induces dextran uptake during the virus entry and this action is inhibited by EIPA. Moreover, to reinforce the hypothesis that ASFV entry occurs mainly by macropinocytosis, we developed an experiment to assess the colocalization between the virus particles and the macropinocytosis marker dextran. These results are included in Figure 3H.
All together, these data strongly indicate that ASFV induces activation of macropinocytosis to enter cells.
Macropinocytosis is a very specific actin-dependent endocytic process since it depends on acting rearrangements to induce membrane ruffling formation, and inhibitors of actin microfilaments, such as Cytochalasin D (Cyto D) [63], [64], Latrunculin A [65] and Jasplakinolide [66], are commonly used to inhibit this process.
To demonstrate whether ASFV depends on actin to enter cells, we used Cyto D, which binds to the positive end of F-actin impairing further addition of G-actin, thus preventing growth of the microfilament [67]. Vero cells were pretreated with Cyto D at a concentration of 8 µM and ASFV uptake (MOI 10) at 60 mpi was next analyzed by FACS. As shown in Figure 4A, the disruption of actin dynamics by the inhibitor reduced ASFV entry in a percentage of about 50%. To assess whether the drug impairs the synthesis of viral proteins, Vero cells were untreated or treated with Cyto D (4 µM) and then infected with Ba71V, MOI 1. After 16 hpi, we used a specific antiserum against both early and late ASFV proteins (generated in our lab), to analyze viral protein expression. As expected, Cyto D treatment importantly reduced both the synthesis of p32, one of the main ASFV early proteins, and the synthesis of p12, p17 and p72, three typical late proteins in the ASFV cycle (Figure 4B). In agreement with this, both virus production and viral factories clearly diminished as shown in Figure 4C and 4D, respectively. However, it is noteworthy that even in the presence of Cyto D, a number of virions seem able to enter the cell and induce a productive infection, thus suggesting that the actin cytoskeleton is involved in ASFV entry and also in successive post-entry steps, as shown in Figure S4B.
To further assess the importance of actin microfilaments in the first steps of ASVF entry, we examined whether ASFV infection causes rearrangements of actin cytoskeleton in Vero cells, by using phalloidin in confocal microscopy experiments. Data are presented in Figure 4E, showing the change of actin pattern after 10 and 30 min after virus uptake at MOI 50. Furthermore, and to reinforce these data, Vero cells were transfected with pEGFP-actin plasmid (kindly gifted by Dr. J. Mercer), and infected with Ba71V, MOI 50. Figure 4F shows the redistribution in aggregates of GFP-actin in transfected Vero cells, which are similar to those observed when endogenous actin was analyzed. Not only that, but also, viral particles (red) are found together with the actin aggregates both in endogenous and ectopically expressed actin.
Since it has been described that blebs and ruffles contain actin, Rac1 and cortactin [23], [68], it is likely that these actin spots correspond to membrane active places where ASFV-induced ruffling should occur, thus suggesting that actin dynamics is a very important factor to ASFV in the host cell to mediate cell-wide plasma membrane ruffling.
Another component of the cytoskeleton that has been reported to be involved in several virus entry processes is the microtubules system, although the importance of microtubules specifically regarding the macropinocytosis pathway is controversial [69]. In respect to ASFV infection, whereas it has been reported that nocodazole (a specific inhibitor of microtubules system [70]) does not affect viral DNA replication [71], a report from Health et al. [72] describes that nocodazole produces a decrease in the expression of p72 and p12 late proteins, but not in the early proteins of ASFV. To investigate whether the microtubule system has a role in ASFV entry, Vero cells were treated with different concentrations of nocodazole and then infected with ASFV at MOI 1. Microtubule disruption had no effect on early viral protein synthesis and barely on late proteins synthesis such as p12 and p72 (Figure S5). Therefore, we conclude that the microtubules system is not likely significant for ASFV entry.
Macropinocytosis is typically started by external stimulation. This stimulation is usually associated with growth factors that trigger activation of receptor tyrosine kinases (RTKs). These molecules then activate signaling pathways that induce changes in the dynamics of actin cytoskeleton and disturb plasma membrane [21]. Among them, epidermal growth factor receptor (EGFR) has been connected with actin rearrangement and activation of Rho family GTPases, and its activation is known to trigger macropinocytosis [45], [73].
Besides the membrane perturbations and actin remodeling observed following ASFV uptake, we have found that EGFR activation was essential for ASFV infection, since 324674, the specific inhibitor of this receptor tyrosine kinase [74], efficiently inhibited ASFV uptake in a dose-dependent manner as assessed by FACS experiments in Vero cells. Accordingly, ASFV entry relies on tyrosine kinases activity, as preincubation of the cells with genistein (tyrosine kinase inhibitor [75]) also inhibited ASFV infection (Figure 5A).
The PI3K/PDK1/Akt/mTORC1 pathway regulates vital cellular processes that are important for viral replication and propagation, including cell growth, proliferation, and protein translation [76]. Concerning macropinocytosis, it has been described that PI3K and its effectors induce the formation of lipid structures in ruffles and macropinocytic cups involved in cytoskeleton modulation [77]–[79]. In recent years, it has been reported that several viruses use the PI3K-Akt pathway to support entry into cells and early events of the infection [23], [80].
In order to investigate the importance of this pathway on ASFV entry, we have developed, after different times of ASFV uptake, an ELISA test that directly measures the activity of PI3K by analyzing phosphorylation of its specific substrate PI(4,5)P2. The results (Figure 5B), show the increase of substrate phosphorylation from 5 min after virus uptake, reaching a maximum after 30 min of infection. Importantly, the presence of the PI3K inhibitor LY294002 (LY) [81] strongly impaired the kinase activation by the virus.
It has been reported that Akt is the major downstream effector of the PI3K pathway and is commonly used as readout of PI3K activation [82], since Akt phosphorylation has been considered to be a direct consequence of PI3K activation pathway [83]–[85]. To analyze the effect of virus uptake on Akt phosphorylation, Vero cells were serum starved for 4 h and then infected with Ba71V (MOI 10) from 5 to 90 min. Figure 5C shows that Akt is phosphorylated from 5 min after virus uptake, reaching a maximum at 30 min. It has been established that Akt phosphorylation of Thr308 is a direct consequence of PI3K activation pathway [83] while phosphorylation of Ser473 depends on mTORC2 [84], [85]. Since phosphorylation in both residues of Akt is required for its complete activation, we measured the ASFV-induced Akt phosphorylation with two different anti-phospho antibodies. Figure S6 shows that Akt is phosphorylated both in Thr308 and in Ser473 early after ASFV infection, suggesting that ASFV entry fully activates this pathway in the infected cell.
To further investigate whether the PI3K activation observed early during ASFV infection involves mainly upstream steps, we pretreated Vero cells with LY at a concentration of 60 µM. Cells were then infected with Ba71V MOI 10, and the virus uptake was analyzed by FACS at 60 mpi. Figure 5D shows that virus uptake decreased to about 45% in treated Vero cells in respect to DMSO-treated cells, indicating that PI3K activation is involved in the virus entry. Not only that, but we also found that the activation of this kinase has a key role in the consecution of infection since, as shown in Figure 5E, the presence of 20 µM LY severely impairs the synthesis of both ASFV early and late virus proteins. Recently, our group has described that ASFV regulates the cellular machinery of protein synthesis to guarantee the expression of its own proteins [15]. Since it has been reported that one of the main roles of PI3K is regulating the translational machinery through the PI3K-Akt-mTOR pathway [86], the strong effect observed of LY on the ASFV protein synthesis is not surprising (Figure S4C). Finally, and to confirm the role of PI3K on ASFV infection, we performed experiments to analyze the number of cells presenting viral factories in the presence of LY. As shown in Figure 5F, a dramatic decrease of infected cells was observed after 16 hpi (MOI 5) when the infection was performed in the presence of the inhibitor. Similarly, virus production was diminished about 3 logs units by the effect of LY after 48 hpi (Figure 5G).
Since activation of Rac1-GTPase has been involved in the regulation of macropinocytosis by triggering membrane ruffling in the cell [87], we investigated the activation status of Rac1 during the first steps of ASFV entry in Vero cells. Ba71V was used to synchronously infect cells (MOI 10), and Rac1 activation was measured with the G-LISA activation kit following the manufacturer's instructions. The results showed that Rac1 activation is a very fast and strong event during ASFV entry, reaching a maximum (2.5 fold) at 10 mpi compared to mock-infected cells (Figure 6A). It has been shown that Rac1 controls macropinocytosis by interacting with its specific effectors, the p21-activated kinases (Paks), thus modulating actin cytoskeleton dynamics [88], [89]. It is also known that Rac1 binds and activates Pak1 only under its Rac1-GTP active form. To confirm the results obtained by G-LISA, we further analyzed the Rac1 activation during ASFV entry by performing a pull down assay using Pak1-PBD-Agarose Beads, which carried the PBD-Pak1 ready to bind Rac1-GTP. As shown in Figure 6B, Rac1-GTP was found together with the pulled Pak1-PBD-Agarose Beads after 10 min post ASFV infection, slightly diminishing 30 min after the infection. This result further corroborates that ASFV entry induces the formation of the Rac1 active conformation. Since it has been described that Rac1 is contained in blebs and ruffles [22], [23], [90] and, as shown above, ASFV induces these type of the structures when it infects cells, we next analyzed the localization of Rac1 during the process of ASFV entry. To achieve this, Vero cells were first transfected for 24 h with pEGFP-Rac1 (kindly given by Dr. J. Mercer) and then infected with Ba71V, MOI 10. As shown in Figure 6C, we found clusters of the GTPase as early as 10 min after infection. Accordingly with the experiments shown above, this effect was clearly perceptible at 30 mpi, demonstrating, first, that ASFV infection induces accumulation of active Rac1 in ruffling areas, and second, that this is an event that takes place mainly during ASFV entry.
The effect of Rac1 inhibition on virus uptake was next investigated. Cells were pretreated with 200 µM Rac1 inhibitor [91] and the virus uptake was measured after 60 mpi by FACS analysis, using the specific antibody against the ASFV capsid protein p72, as described in Materials and Methods. Figure 6D shows the dramatic decrease of virus uptake when the infection is performed in the presence of the pharmacologic inhibitor of Rac1. Furthermore, we analyzed the effect on the ASFV uptake in the presence of the inhibitor by CLSM experiments, using the same conditions as above. The images were taken as a maximum z-projection of horizontal and vertical slices. As Figure 6E (bottom panels) indicates, a strong inhibition of virus uptake could be observed in the presence of the Rac1 inhibitor, since the number of ASFV particles in the cell (green) is visibly lower in the presence of the drug. Moreover, and as shown in the upper panels of Figure 6E, virus (green) colocalized (yellow), with cortical actin (red), indicating that the drug immobilizes the virions imbibed into the plasma membrane and impairs their entry into the cell. Separated channels are also shown in Figure S2D.
Alternatively, and to reinforce the role of Rac1 on ASFV infection, we studied the level of ASFV protein synthesis in Vero cells previously transfected with the mutant pGFP-Rac1-N17 (a kind gift from Dr. R. Madrid). The expression of the inactive form of Rac1 strongly inhibited the expression of the ASFV early p32 protein (Figure 6F). As expected, the synthesis of viral late proteins was also affected by treatment with the inhibitor (Figure S7). Not only that, but also, when Rac1 inhibitor was added 60 min after virus addition, the level of viral protein synthesis observed was completely recovered, thus reinforcing the role of Rac1 in virus entry (Figure S4D).
Hence, the role of Rac1 on ASFV morphogenesis and virus production was investigated. To achieve this, Vero cells were treated with the Rac1 inhibitor and then infected during 16 h, MOI 5. Cells were fixed and stained with anti-p72 to visualize the viral factories by CLSM and the percentage of infected cells in the presence or absence of the inhibitor was represented in the graph (Figure 6G). As observed, the number of cells containing ASFV factories decreased about 65% in the presence of Rac1 inhibitor compared to the untreated controls (separate channels are shown in Figure S2E). In line with these results, the viral production at 48 hpi decreased strongly when the activity of Rac1 GTPase was inhibited (Figure 6H).
Finally, since Rac1 has been reported to be an important component of ruffles [22], [23], [90], we have used the Rac1 inhibitor to assess its involvement in the inhibition of these membrane perturbations and therefore, indirectly, the role of ruffles in ASFV uptake. To achieve this, we have performed FESEM assays in Vero cells treated with 200 µM Rac1 inhibitor during 60 min prior to virus addition. As shown in Figure 6I, Rac1 inhibitor strongly decreases the ASFV-induced ruffles, in accordance with the decrease in virus uptake (Figure 6D), viral infection (6G) and virus production (6H) previously observed.
Taken together, these results demonstrate the significant role of Rac1 on ASFV entry.
The p21-activated kinase 1 (Pak1), a serine/threonine kinase activated by Rac1 or Cdc42 [89] is one of the most relevant kinases related to several virus entry processes since it is involved in the regulation of cytoskeleton dynamics and is needed during all the stages of macropinocytosis [88], [92], [93]. Among the different residues to be phosphorylated in Pak1 activation, the Thr423 plays a central role because its phosphorylation is necessary for full activation of the kinase [94].
To determine whether Pak1 was activated during ASFV entry, we first analyzed the phosphorylation on Thr423 in Vero cells synchronously infected (MOI 5) with Ba71V. At different times post infection, samples were collected and analyzed by immunoblotting using an anti-phospho-Pak1 Thr423 antibody. As early as 30 mpi, phosphorylation of Pak1 could be detected, increasing until 120 mpi (Figure 7A).
IPA-3 has been identified as a direct, noncompetitive and highly selective Pak1 inhibitor. In the presence of IPA-3, Thr423 phosphorylation is inhibited since the Pak1 autoregulatory domain is targeted by the inhibitor [95]. To assess the role of Pak1 activation in ASFV uptake, we measured by FACS analysis the p72 levels detected into the Ba71V-infected Vero cells (MOI 10) after 60 mpi. As shown in Figure 7B, the p72 levels incorporated into the cells in the presence of 30 µM IPA-3 were significantly lower (70%) than those obtained in the absence of the inhibitor. These results indicate that Pak1 activation is involved in the first stages of ASFV entry, since phosphorylation of the kinase occurs at very early times after virus addition, and even more importantly, the uptake of the virus into the host cells is strongly dependent of Pak1 activity.
Apart from the role played by Pak1 in viral entry, the sensitivity of ASFV infection to IPA-3 was investigated in Ba71V-infected Vero cells by Western blot. Using specific antibodies against both early and late ASFV proteins, the effect of the inhibitor from 1 to 10 µM on viral protein synthesis was evaluated. Figure 7C shows the strong dose-dependent IPA-3 inhibition over the most important early (p32) and late proteins (p72, p24, p17 and p12). To reinforce the role of Pak1 in ASFV entry, a similar experiment performed by incubation with IPA-3 during 60 min after virus addition is shown in Figure S4E. These data indicate that the drug is mainly affecting virus entry as it does not induce important inhibition on viral protein synthesis when incubated after virus uptake. Moreover, virus title was reduced 1.5 log units in cells pretreated with 5 µM IPA-3 and then infected with Ba71V (MOI 1) in the presence of the inhibitor during 48 h (Figure 7D).
To corroborate the significant role of Pak1 during ASFV infection, we used different Pak1 constructs affecting Pak1 activation (see Materials and Methods). Vero cells were transfected for 24 h with pEGFP, pEGFP-Pak1-WT, pEGFP-Pak-AID and pEGFP-Pak1-T423E (all of them kindly gifted by Dr. J. Chernoff) and infected for 16 h with ASFV at a MOI of 1 pfu/cell. As shown in Figure 7E, the constructs containing the Pak1 autoinhibitory domain (AID) inhibited p12 and p32 viral protein expression, whereas cells transfected with wild type (WT) form showed the same protein levels than infected control cells. It is noteworthy that constitutively active Pak1 construction T423E (even although it was only shortly expressed in the transient transfection process) induced a remarkable enhancement on the expression of the ASFV early protein p32, indicating that increasing Pak1 activity intensifies the early protein synthesis, probably due to its effect on virus entry. Numeric values of these data are shown in Figure 7F.
These data, together with those of Rac1 activation explained above, strongly supports our hypothesis of ASFV triggering the Rac1-Pak1 pathway during the virus entry.
Dynamin is a cellular essential GTPase which plays an important role in cellular membrane fission during vesicle formation [96]. It is likely involved in Rac1 localization and function, since it has been shown that Rac1-dependent macropinocytosis is blocked by the dynamin-2 (DynK44A) dominant-negative [39].
Since, as we demonstrated above, Rac1 is important to ASFV entry, we have analyzed whether dynamin-2 pathway plays a role either in ASFV entry or infection. To achieve this, we first investigated the effect of Dynasore (Dyn), a reversible inhibitor of GTPases activity [97], over ASFV uptake. After 60 min of pretreatment with 100 µM Dyn, Vero cells were infected with Ba71V at MOI 10 and virus uptake was measured by FACS using the specific antibody against the capsid viral protein p72. The result showed that treatment with Dyn partially inhibited virus uptake (35%) (Figure 8A). A higher effect of the inhibitor on ASFV entry could not be found by using different experimental conditions (data not shown), further indicating the partial involvement of dynamine in virus uptake. Moreover, the role of clathrin-mediated endocytosis was examined in parallel using Chlorpromazine (CPZ), which inhibits the assembly of coated pits at the plasma membrane and is considered a specific inhibitor of clathrin-mediated endocytosis [98]. Using parallel experimental conditions, and in contrast with the data obtained after treatment with Dyn, we observed that the virus uptake was not likely affected in the presence of 20 µM CPZ (Figure 8A). These data indicate that whereas dynamine is to some extent involved in ASFV entry in accordance with its role in macropinocytosis [39], clathrin is not related to ASFV uptake in Vero cells.
In order to investigate whether other steps downstream ASFV entry were affected by Dyn and CPZ, Vero cells were separately pretreated with the inhibitors, and then infected with ASFV (MOI 1). At the indicated times after infection, the synthesis of both early and late ASFV proteins was analyzed by Western blot. The treatment with 100 µM Dyn strongly inhibited p72 and p32 expression from early times post infection (Figure 8B), consequently indicating that dynamine is required for ASFV both early and late infection course. As Figure 8C shows, CPZ had a similar effect to Dyn both on ASFV early and late protein synthesis, in concordance with the data from Hernaez et al. [38], in which the expression of the viral protein p32 depends on clathrin function. Higher amounts of CPZ could result in an inhibition of p72, but this effect is likely due to the cytotoxic effect of the drug, as reported in Table S1. Taken together, our data showed that whereas the effect of Dyn on viral protein synthesis is probably due to dynamine participation on ASFV entry events, the clathrin inhibition does not involve virus uptake, but only viral protein synthesis, thus indicating a role for clathrin function merely in post entry events. Future experiments are planned to more specifically study which are the ASFV post entry events regulated by clathrin.
Finally, and as expected, both inhibitors had an important effect on viral production measured after 48 hpi (MOI 1) in Vero cells (Figure 8D).
Endocytosis constitutes an efficient way for viruses to cross the physical barrier represented by the plasma membrane and to pass through the underlying cortical matrix. Knowledge of the specific pathway of virus entry and of the precise mechanisms regulating is key to understand viral pathogenesis, since virus entry into host cell is the first major step in infection. Whereas there is ample evidence showing that ASFV enters cells through endocytosis in a pH-dependent manner and that saturable binding-sites on the plasma membrane mediate the productive entry of the virus into Vero cells and swine macrophages [33], [34], the specific endocytic and signaling pathways used by the virus are largely unknown.
In this report, by combining different and independent approaches, we have achieved an exhaustive analysis of the ASFV endocytic pathway. We have obtained a precise picture of how ASFV enters the cell and have identified the main cellular proteins required. Careful assessment of specificity and functionality of each pathway was performed and correlated with infection and virus uptake.
Many recent reports have shown that viruses can directly use macropinocytosis as an endocytic way for productive infection [21], [23]–[29], and also to promote the penetration of viral particles that enter by other endocytic mechanisms [31], [32]. Macropinocytosis activation is related to significant cell-wide membrane ruffling mediated by activation of actin filaments. These structures may have different shapes: lamellipodia, circular-shaped membrane extrusions (ruffles) or large membrane extrusions in form of blebs.
Here we have illustrated by FESEM that ASFV strain Ba71V induced prominent membrane protrusions compatible with ruffles after 10 mpi. Transmission electron microscopy images further support this result by showing that ASF virions internalize adjacent to retracting ruffles, likely indicating uptake of viral particles occurs as part of the macropinocytic process. Not only that, but also, we found that inhibition of Rac1, an important component of ruffles, importantly impaired the ASFV uptake, thus involving the formation of these membrane perturbations in virus entry.
Moreover, and in parallel to the data obtained in Vero cells, we found that the E70 virulent strain induced a type of membrane protrusion similar to blebs a few minutes after the infection of the swine macrophage line IPAM. This last result is important, since macrophages are probably the natural target cell of the infection in vivo and suggests that different macropinocytic programs can be used by different ASFV strains, as has been published for other virus as Vaccinia [45]. Because of this, we have carefully characterized these structures. First, we showed the inhibition of virus entry with different doses of blebbistatin, and second we demonstrated that Rock1 (a marker of blebs [52]) colocalized with virus particles on blebs in IPAM cells from 30 min after virus uptake.
Apart from characteristic membrane perturbations, macropinocytosis is also distinguished from other entry pathways by features that include actin-dependent structural changes in the plasma membrane, regulation by PI3K, PKC, Rho family GTPases, Na+/H+ exchangers, Pak1, as well as ligand-induced upregulation of fluid phase uptake. In this regard, our work demonstrates that EIPA, a potent and specific inhibitor of the Na+/H+ exchanger [23], [53], [54], [99], severely impairs ASFV infection and entry. By using FACS analysis we found that EIPA treatment caused a significant dose-dependent manner reduction (more than 60%) in the uptake of ASFV infective particles. Confocal microscopy analysis also revealed that there was an evident drop in virus particles incorporated into the cells incubated with EIPA. It is important to note that macropinocytosis is the only endocytic pathway susceptible to the inhibition of the Na+/H+ exchangers. Thus, these results strongly indicate the involvement of macropinocytosis in ASFV virus entry.
Actin plays a central role in formation and trafficking of macropinosomes. Cyto D, which binds to the positive end of F-actin (impairing further addition of G-actin and preventing the growth of the microfilament [67]), reduced ASFV entry by approximately 50% and inhibited the synthesis of both early and late viral proteins, together with viral morphogenesis. However, it is remarkable that virions that escape from the action of Cyto D induce a productive infection, thus suggesting that actin cytoskeleton is mainly involved in ASFV entry, although it could have a role in successive post-entry steps.
Corroborating this hypothesis, we have observed that ASFV infection causes rearrangements of endogenous actin cytoskeleton in Vero cells as early as 10 min post infection. These data were reinforced by overexpression of GFP-actin that was concentrated in aggregates in virus-infected cells. Together, these data provide evidence for a role of actin in ASFV entry and suggest that the virus can actively promote localized actin remodeling to facilitate its uptake through macropinocytosis or a similar mechanism.
The first reports describing the endocytic entry of viruses into their host cells presumed that incoming viruses took advantage of ongoing cellular endocytosis processes [16]. However, it is now clear that several viruses are not only passive cargo but activate their own endocytic uptake by eliciting cellular signaling pathways. The activation of these pathways significantly depends on the interaction of the virus with cellular receptors specific to the type and activation status of the host cell [100], [101]. ASFV, as Vaccinia virus [21], [45], seems to belong to the viruses that actively trigger their endocytic internalization. In this respect, we have found that entry of ASFV is dependent on signaling through tyrosine kinases as EGFR, and activation of PI3K together with Rho-GTPases as Rac1, which have been all described to be important regulators of macropinocytosis [69].
Concerning the function of the PI3K pathway, activation of this kinase early after virus uptake was confirmed by analyzing the phosphorylation of its specific substrate PI(4,5)P2. Also, phosphorylation of both residues Thr308 and Ser473 of Akt was observed early after ASFV infection. Besides, pretreatment of Vero cells with the specific PI3K pharmacological inhibitor LY strongly inhibited virus uptake at 60 mpi. Not only that, but we also found that the activation of this kinase has an important role in the infection, since the presence of LY severely impairs the synthesis of both ASFV early and late virus proteins. In this regard, our group has recently described [15] that ASFV uses the cellular machinery of protein synthesis to express its own proteins. Since it has been reported that one of the main roles of PI3K is to regulate the translational machinery through the AKT-mTOR pathway [86], the strong effect observed of LY on ASFV protein synthesis is very much expected.
We have also demonstrated that Rac1, a regulatory guanosine triphosphatase of Pak1, was activated during ASFV entry. Rac1 protein belongs to the Rho family of small guanosine triphosphatases, a subgroup of the Ras superfamily of GTPases [102]. In the last years, several viruses have been described to target Rho-GTPases activation to enter the host cells, such as Vaccinia virus [23], [45], Ebola virus [80], Echovirus [92] or Adenoviruses type 2 [103], among others. Through interaction with its specific effector Pak1, Rac1 modulates actin cytoskeleton dynamics and controls macropinocytosis [88], [89]. Consistent with the data reported by Mercer and Helenius, 2008 [23], showing that active Rac1 is contained in virus-induced membrane perturations, our results show that ASFV induces clusters of this GTPase as early as 10 min after infection. Hence, Rac1 accumulates in ruffling areas very early during the process of ASFV entry, suggesting that ASFV targets Rac1 to entry in host cells. In agreement with this hypothesis, a strong inhibition of virus uptake, in parallel with ruffle formation, was observed in the presence of the Rac1 inhibitor. Moreover, by performing CLSM experiments, we showed that the drug immobilized the virus particles imbibed into the plasma membrane, thus impairing their entry into the cell. Taken together, these results demonstrate the significant role of Rac1 on ASFV entry. Our data strongly contrasts with a recent study [104], which reported that, although Rac1 is activated by ASFV infection, it is not involved in either ASFV entry or viral protein synthesis. In that study by Quetglas et al. [104], Rac1 would be responsible of a downstream process that only affected viral production. The discrepancies about the role of Rac1 in ASFV entry and infection might be explained by the fact that the Rac1 inhibitor concentration used does not match with the amounts usually employed to analyze the role of Rac1 in virus uptake [80], and it is likely too low to disturb ASFV entry or viral protein synthesis. Moreover, confocal microscopy images to measure ASFV uptake were taken as mid z-section, in contrast to our procedure that includes several z-sections that allow us to count the total virus particles inside the cells. Finally, important information regarding the effect of the dominant-negative Rac1-N17 on viral protein synthesis were not shown in that study, in contrast to our results described in Figure 6F. Therefore, the limitations of that work [104] make it difficult to reach any conclusions about the function of Rac1 on ASFV entry and infection. Furthermore, in support of our data, we should note that we have found an important role for Pak1 in Ba71V entry in Vero cells. Pak1 is a serine/threonine kinase activated by Rac1 or Cdc42 involved in the regulation of cytoskeleton dynamics and needed during all stages of macropinocytosis [88], [93], [105]. Our results indicate that Pak1 activation is involved in the first steps of ASFV entry, since phosphorylation of the kinase occurs at very early times after virus addition, and even more importantly, the uptake of the virus into the host cells is strongly dependent of Pak1 activity. However, our preliminary studies using the E70 strain did not show a clear effect of the Pak1-specific inhibitor IPA-3 on the synthesis of ASFV proteins (data not shown), either in IPAM or in alveolar swine macrophages. These data suggest that ASFV may activate other different pathways in macrophages or that IPA-3 cannot be efficient enough to inhibit Pak1 if this kinase is constitutively activated in these cells [106], [107]. Nevertheless, the synthesis of viral proteins was strongly inhibited in macrophages after EIPA and LY treatments, indicating that Na+/H+ exchangers and the PI3K pathway are involved in macropinocytosis-mediated ASFV entry into these cells (Figure S8).
In conclusion, the involvement of the EGFR and PI3K, the nature of the signaling pathway, the involvement of Rac1, Pak1 and Na+/H+ exchangers, and the actin-cytoskeleton rearrangements, all support a macropinocytosis-driven endocytic process for ASFV entry. In addition, ASFV caused significant induction of dextran uptake (a specific fluid phase marker of macropinocytosis), and colocalization of the internalized ASF virus particles with dextran was also observed.
The ASFV genome encodes several glycoproteins [108], whose role in host-cell binding and entry has not yet been described. However, it has been shown that glycoproteins and lipids are required for several virus binding and entry steps to the host cells [23], [109], [110]. It has been also reported that cellular partners that bind to specific regions of viral glycoproteins translocate from intracellular compartments to regulate the susceptibility of different cells to the infection [111]. These kinds of mechanisms could explain the differences found among ASFV viral isolates and their ability to infect different host cells. Future experiments are planned to study the role of both ASFV glycolipids and the putative host partners involved in the mechanisms of ASFV entry and infection of different cell populations.
Dynamin is a large GTPase that is involved in scission of newly-formed endocytic vesicles at the plasma membrane [112]–[114]. Although we have shown that dynasore partially inhibits virus entry, we have found no evidence for a role of clathrin in ASFV entry despite the use of multiple approaches. The fact that in our hands dynamin was only partially involved in ASFV entry further ruled out roles for clathrin or caveolae-mediated pathways, as both require dynamin activity. Therefore, our data contrast with a recent study concluding that clathrin-mediated endocytosis is the major entry pathway for ASFV [38]. The key concern about the conclusion of this work is that virus entry is merely measured by the synthesis of ASFV early proteins in the presence of chlorpromazine, and not by specific analysis of virus uptake. Moreover, it is important to note that whereas chlorpromazine disrupts clathrin-coated pits, it may also interfere with biogenesis of large intracellular vesicles such as phagosomes and macropinosomes [115].
Here, by combining different and separate strategies we have carried out a precise analysis of each key endocytic pathway concerned, obtaining, for the first time, a relatively complete description of the mechanism by which ASFV enters into a cell, including identification of several cellular molecules and routes. We have carefully evaluated the specificity and functionality of each pathway and correlated them with virus uptake and infection. Two different strains of ASFV, the virulent E70 and the virulent Ba71V, adapted to growth in Vero cells, have been used to study the virus entry mechanism either in swine macrophages or Vero, respectively.
Several drugs were used to inhibit pathways, but specificity was evaluated by testing the function of the main pathways after treatment. Furthermore, highly specific dominant-negative mutants were used to confirm the data obtained by pharmacological inhibitors. More importantly, all throughout this work either a FACS-based or a confocal sensitive virus entry assays were used in discriminating blockage in virus entry versus blockage in downstream steps of the infection cycle. This is particularly relevant when using drugs that frequently affect multiple cellular functions in addition to specific entry.
Overall, our data provide strong evidence that ASFV entry takes place by a process closely related to macropinocytosis, adding new and valuable information regarding endocytosis mechanisms in the context of ASFV entry (plotted in Table 1).
The evidence presented demonstrates for the first time, that ASFV utilizes a macropinocytosis-like pathway as the primary means of entry into IPAM and Vero cells. However, we cannot state that virus entry occurs exclusively by this pathway, especially in swine macrophages. But our data clearly show that its disruption blocks the greater part of infection and particle uptake. Our work also indicates that clathrin-mediated endocytosis plays at most a minor role in ASFV entry. However, and in accordance with the data of Hernaez et al. [38], we found that CPZ diminishes both ASFV early and late protein synthesis, together with viral production. Thus, our data demonstrate a role for clathrin function merely in post entry events.
A strong hazard of ASFV dissemination from Sardinia and Caucasian areas to EU countries has recently appeared, thus making the progress of knowledge and tools for protection against this virus urgent. Infection by ASFV is characterized by the absence of a neutralizing immune response, which has so far hampered the development of a conventional vaccine. Therefore, our findings are relevant as they not only provide a detailed understanding of ASFV entry mechanism, but also identify novel cellular factors that may provide new potential targets for therapies against this virus. In parallel, further studies are planned to characterize viral factors that may interact with components of the macropinocytosis pathway, probably useful for vaccine development.
|
10.1371/journal.pcbi.1005713 | A conceptual and computational framework for modelling and understanding the non-equilibrium gene regulatory networks of mouse embryonic stem cells | The capacity of pluripotent embryonic stem cells to differentiate into any cell type in the body makes them invaluable in the field of regenerative medicine. However, because of the complexity of both the core pluripotency network and the process of cell fate computation it is not yet possible to control the fate of stem cells. We present a theoretical model of stem cell fate computation that is based on Halley and Winkler’s Branching Process Theory (BPT) and on Greaves et al.’s agent-based computer simulation derived from that theoretical model. BPT abstracts the complex production and action of a Transcription Factor (TF) into a single critical branching process that may dissipate, maintain, or become supercritical. Here we take the single TF model and extend it to multiple interacting TFs, and build an agent-based simulation of multiple TFs to investigate the dynamics of such coupled systems. We have developed the simulation and the theoretical model together, in an iterative manner, with the aim of obtaining a deeper understanding of stem cell fate computation, in order to influence experimental efforts, which may in turn influence the outcome of cellular differentiation. The model used is an example of self-organization and could be more widely applicable to the modelling of other complex systems. The simulation based on this model, though currently limited in scope in terms of the biology it represents, supports the utility of the Halley and Winkler branching process model in describing the behaviour of stem cell gene regulatory networks. Our simulation demonstrates three key features: (i) the existence of a critical value of the branching process parameter, dependent on the details of the cistrome in question; (ii) the ability of an active cistrome to “ignite” an otherwise fully dissipated cistrome, and drive it to criticality; (iii) how coupling cistromes together can reduce their critical branching parameter values needed to drive them to criticality.
| Pluripotent stem cells possess the capacity both to renew themselves indefinitely and to differentiate to any cell type in the body. Thus the ability to direct stem cell differentiation would have immense potential in regenerative medicine. There is a massive amount of biological data relevant to stem cells; here we exploit data relating to stem cell differentiation to help understand cell behaviour and complexity. These cells contain a dynamic, non-equilibrium network of genes regulated in part by transcription factors expressed by the network itself. Here we take an existing theoretical framework, Transcription Factor Branching Processes, which explains how these genetic networks can have critical behaviour, and can tip between low and full expression. We use this theory as the basis for the design and implementation of a computational simulation platform, which we then use to run a variety of simulation experiments, to gain a better understanding how these various transcription factors can combine, interact, and influence each other. The simulation parameters are derived from experimental data relating to the core factors in pluripotent stem cell differentiation. The simulation results determine the critical values of branching process parameters, and how these are modulated by the various interacting transcription factors.
| Stem cells present an important instance of the complexity of cellular function and gene regulation. Pluripotent stem cells possess the capacity both to renew themselves indefinitely and to differentiate to any cell type in the body. This second capacity means that the ability to direct stem cell differentiation would have immense potential in regenerative medicine. High throughput datasets are available to facilitate understanding of stem cells, but such data provide only snapshots of biological complexity, with no dynamics. Here we consider these data within a particular theoretical framework to explore their abstract dynamics, with the longer term aim to help understand cell behaviour and complexity.
Here, we suggest that in order to understand how stem-cell gene regulatory networks drive cell fate, we need to appreciate the non-equilibrium nature of regulatory networks in living cells. How can a regulatory network self-organise from one stable state of pluripotency, to another state of differentiation? Ultimately, such better understanding could aid attempts to direct such state changes. First, we need an understanding of the dynamics of such networks.
We expand on our previous theoretical framework in which self-organization and natural selection are intimate partners in non-equilibrium dynamic systems generally [1–5], and investigate an abstract model of four transcription factors (TFs) central to the self-organization of the core pluripotent network of mouse embryonic stem cells (ESCs) [6, 7]. The underlying theory of the model presented here is the Transcription Factor Branching Process (TFBP) concept explained in Halley et al. [3].
Our principal aim is to capture and integrate the results of multiple high-throughput experiments using a logical and transparent computational framework. This would allow us to model protein expression, particularly TF expression, across multiple layers of stem cell regulation. However, this first requires a sound theoretical framework to understand and predict how regulatory layers self-organize and interact, both within individual cells and between multiple cell types within larger cell assemblies.
Here, we begin to address the problem by describing a novel computational concept derived from the hypothesis of exploratory behaviour described (see Exploratory hypothesis of stem cell fate computation, above) and from branching process theory (see Branching Processes, above).
The utility of such a computational approach relies on the scale-invariant nature of the reproducing units. By its very nature, critical dynamics describes multiple parallel processes that propagate in some way, but because these processes occur within a bounded system, not all can realize their full potential to propagate. Unlike self-organized critical systems, which typically incorporate local stability thresholds, the propagation of a branching process in a critical-like system depends on other processes propagating at that time (although of course there may also be local stability thresholds involved).
The key point is that the critical-like state is achieved or computed via direct interaction among branching processes. In contrast, in self-organized critical systems all of the branching processes are temporally separated such that they cannot interact directly, only indirectly via local stability thresholds.
In our model, branching processes are used conceptually to define boundaries of information flow. ChIP-Seq data are used to capture the entire genome of a pluripotent stem cell, where the expression of each pluripotency TF is defined in terms of a branching process that propagates through time, interacting and competing with others. In this view, fluctuation in the expression of pluripotency genes when mouse ESC are withdrawn from self-renewal conditions (2i) is not trivial: it is central to model dynamics. They are expected to determine differentiation trajectories.
When we consider the expression of a single TF as a distinct branching process, the population of TF molecules can be thought of as the backbone of the branching process as it propagates through time. This can be likened to propagation of a family name through the male offspring; female offspring traditionally fail to propagate the family surname, similar to bursts of TF expression that activate target genes that do not feed back into the transcriptional network. In the context of TF expression, the branching process effectively defines how the population of TF molecules reproduces via the entire regulatory network.
In this sense, the trajectories of individual ESCs are intrinsically knowable and able to be calculated from patterns of competition and interference among cascades of gene expression bounded by the cistrome of each TF. Genome location data, which describe interactions between TFs and other genes at genome-wide scales, can be used to simulate these branching processes and estimate patterns of interference that give rise to individual cell trajectories.
From the simulation of the TFBP, we can demonstrate the existence of mcrit values, values of the branching parameter below which the simulation rapidly dissipates and above which supercritical branching can take place; see Greaves et al [30] for details.
The model includes the following parameters and values:
The determination of mcrit for Nanog, Sox2 and Oct4 is found in Greaves et al [30]. (The experiments were repeated with the new multi-cistrome code used in this paper, run with a single cistrome; the same results were found.) The determination of mcrit for Nanog is illustrated in Fig 1. The value of mcrit for the Nanog cistrome (8) is found to be somewhat higher than that for the other core pluripotency cistromes of Oct4 and Sox2 (6 and 7 respectively). The observed mcrit results are summarised in the final table below.
An infinite limit model discussed in [30] is used to calculate an estimate of mcrit; in this limit the system tips from fully dissipated to supercritical immediately. Fig 1 is not such a step function. The finite sized, noisy, system tips sharply from dissipated to ignited at the experimentally observed value mcrit, but then requires a somewhat higher value of m to become fully saturated. The observed mcrit is slightly higher than the infinite limit value (Table 1), due to noise. Importantly, even though the system is noisy, it still maintains supercriticality in the face of fluctuations in all runs with mcrit ≤ m.
We have also verified that that mcrit is not affected by whether the initial value s0 = r or s0 = 0.5r and that results are scalable i.e. we can alter the values of p and r in a cistrome, providing that the ratio p / r remains constant and obtain the same results [30].
In addition to these previously existing findings, we have run similar simulations for the cMyc cistrome to determine its value of mcrit (see Fig 2 and Table 1). This cistrome has a relatively low value of mcrit (of 4) due to its relatively high number of active sites compared to the other cistromes investigated.
The multi-cistrome simulation shows that one active cistrome can “ignite” a fully dissipated cistrome, and drive it to criticality. This is illustrated in two cases. The critically branching Oct4 cistrome can drive fully dissipated Nanog cistrome to criticality (Fig 3). The figure shows the first of 200 simulation runs carried out with these initial conditions (the others runs are qualitatively similar). The inset shows the first 25 timesteps. Similarly, the critically branching Nanog cistrome can drive a fully dissipated Oct4 cistrome to criticality (Fig 4). These show in each that, although the cistrome begins the simulation fully dissipated, it is swiftly ignited to sustainable branching by the critically branching cistrome to which it is coupled.
This is a non-trivial result because: (i) the driving cistrome is working just at the critical value of m needed to sustain itself; (ii) the driving cistrome incurs a “cost” to do so, because some of the TPs it produces are transferred to the other initially dissipated cistrome, rather than being used to maintain its own activity.
In addition to igniting a dissipated cistrome, an effect of coupling cistromes is the reduction of mcrit.
In [30] we use an infinite limit model to calculate an estimate of mcrit in a single cistrome. We here use a similar approach to calculate an estimate in the reduction of mcrit in coupled cistromes. This infinite limit case is essentially noiseless, with each TF site being ignited the average number of times.
Consider the case of two cistromes, X and Y. At time t let there be stX sites active in cistrome X. In the TFBP model, each of these active sites emits mX TFs, so a total of stX × mX TFs are emitted. Let each of these TFs be absorbed by a separate site with uniform probability. There are three cases (three kinds of sites):
Similar arguments, mutatis mutandis, hold for TFs emitted by cistrome Y. So at timestep t+1, the number of active sites in cistrome X is those activated from cistrome Y plus those activated from cistrome X
st+1X=stXmX(rX−σ)pX+stYmYσpY
A similar equation holds for st+1Y. At the critical tipping point, st+1 = st. We take these two equations, eliminate sX / sY, then solve for mX, to get
mX=pX(pY−mYrY+mYσ)(pY−mYrY)(rX−σ)+mYrXσ
This gives the infinite limit predicted value of mX in the case of a given mY. If we substitute mY = pY/rY, the infinite limit single cistrome critical value for Y, we get mX = pX/rX. That is, in the infinite limit, the critical values are unchanged. Alternatively, if we substitute σ = 0 (isolated cistromes), we also recover the original predicted value of mX. However, if we substitute the observed critical value in the finite sized noisy case for mY, and the experimentally derived value of σ (shown in Table 2) we get a different prediction for mX, as shown in Table 3.
Table 3 shows the prediction that critical value for cistrome X should fall when coupled with cistrome Y run at its observed critical value. Under the assumptions used to generate Table 3, cistrome Y is being run at a higher rate than it needs in the infinite limit, and the excess TF production can be used to lower cistrome X’s required rate. The question is: does this reduction carry over in the finite case, or does the presence of noise, requiring a higher than predicted rate to sustain, negate any such change? Our simulation results, for several of these cases, are presented below.
Figs 5–10 show the behaviours of various combinations of coupled cistromes. Each plot shows the first 250 of 1000 timesteps performed, and shows the first run from a set of 200 runs performed; the others runs are similar. In each case, the critically branching cistrome(s) can drive the branching process in the other cistrome, and the sub-critically branching cistrome dissipates the branching process in the critically branching cistrome(s). Coupling the Nanog cistrome to that of Oct4 (with Oct4 having its m value fixed at its observed value mcrit = 6) reduces mcrit for the Nanog cistrome by one (Fig 5). Similar results hold for coupling the Nanog cistrome to Sox2. Coupling the Oct4 cistrome to that of Nanog (with Nanog having its m value fixed at its mcrit = 8) reduces mcrit for the Oct4 cistrome by one (Fig 6). Coupling the Nanog cistrome to both the Oct4 and the Sox2 cistromes reduces mcrit for the Nanog cistrome by 2 (Fig 7). The c-Myc cistrome is extensively overlapped with the core pluripotency cistromes. Coupling the Nanog cistrome to the c-Myc cistrome alone reduces mcrit for the Nanog cistrome by 2 (Fig 8). Hence the c-Myc cistrome has twice the effect on Nanog mcrit as either Oct4 or Sox2 alone. Similar results hold for Oct4. Coupling the Oct4 cistrome to that of cMyc cistrome reduces the value of mcrit for the Oct4 cistrome from 6 to 4 (Fig 9). Coupling the cMyc cistrome to that of Oct4 cistrome leaves the value of mcrit for the cMyc cistrome unchanged at 4 (Fig 10).
The various reductions in mcrit caused by coupling cistromes are summarised in Table 4. Similar to the single cistrome case, the observed values of mcrit are a little higher than the infinite case predictions. In all cases but one the observed coupled value is nevertheless lower than the observed uncoupled value, demonstrating that the reduction can be maintained in the presence of noise. In the case of Oct4 driving cMyc, there is no reduction in the observed value of mcrit, possibly because the value is so low in the first place, and so any reduction would be proportionally smaller.
Our simulation model supports the argument that critically branching processes in cistromes can drive sub-critically branching processes in other cistromes to criticality, as described by the branching process model proposed by Halley and Winkler [5].
Our simulation of the theoretical TFBP occurring within the various TF cistromes of the genome clearly illustrates that there is a critical value of the branching ratio below which branching effectively ceases to occur. Critical or super-critical branching, however, can lead to a sustained branching process in a given cistrome. This is roughly equivalent to saying that a certain minimum transcription of genes for a given TF cascade is required for its sustained production throughout time.
Our simulation model also illustrates the interplay between branching processes in cistromes, in that critically branching processes in one cistrome can drive the processes in other cistromes to which they are coupled, and can also lower the critical value of the branching parameter in those cistromes by one (Oct4 and Nanog, Sox2 and Nanog, Oct4 and Sox2) or by two (cMyc and Nanog, cMyc and Oct4, cMyc and Sox2, Sox2 with Oct4 acting on Nanog).
The simulation model as it stands provides evidence that the Transcription Factor Branching Process (TFBP) of Halley and Winkler [5] has utility in describing the regulation of TF gene regulatory networks. This is a first step towards the understanding of the non-equilibrium dynamics of the core pluripotent network of ESCs. The simulation additionally provides some low level description of the interaction between TF cistromes, demonstrating that igniting transcription of one TF can further prompt the ignition of transcription of other TFs. To gain a more realistic insight to the extent of this effect we need to, ideally, extend the model to incorporate fully interacting branching processes and also to allow for inhibitory binding of TF to sites in the cistrome of TFs whose transcription they repress.
We have taken a first step towards the creation of a multi-layered model of the stem cell regulatory network and in our opinion, these results are interesting and begin to tease out the utility of the Branching Process Theory in understanding Stem Cell fate computation. We have demonstrated that a regulatory network may self-organise from one equilibrium state to another, through ignition of a coupled dissipated cistrome. Understanding the complex dynamics of such self-organising changes of state possible with communicating cistromes may ultimately give insight into how a pluripotent state can tip to a differentiated state, and may help in understanding how to control and even reverse such tipping.
However, our model does not, as yet, capture all the biologically-relevant dynamics. For example, in our current model, a TF can interact with genes in another cistrome only in a stimulatory manner. We need to include the possibility of inhibitory TF binding of TFs to their binding sites on the various gene segments in the cistromes modelled. We therefore need to develop a Domain Specific Language to allow a standardised way of describing the TF circuitry (the pattern of excitatory and inhibitory relationships between individual TFs) to be modelled in the simulation.
The current model is essentially ‘blind’ to the identity of any TFs produced via transcription of the gene segments in any given cistrome, and also ignores the potential requirement for multiple TF to bind to a gene segment in order to activate it (or inhibit it). There are also other considerations that we need to take on board in order to capture further biologically-relevant dynamics. We also need to accommodate combinatorial binding of TFs to gene segment promoter sequences, as a gene may require binding of specific combinations of TFs to their binding sites in order to promote transcription of the gene segment. Such combinatorial TF binding to enhancer sites can impart transcriptional synergy in a future multicellular model.
We need to include expression data to factor in the strength of activation: currently each gene segment is assumed to produce the same amount of TF each timestep. We also need to include some concept of TF half-life into the model as currently all TFs are assumed to survive and remain bound to their binding sites for a single simulation time-step only. As epigenetic histone modifications may help to shape the circuitry of self-organization, it would also be useful to be able to take into account epigenetic considerations and the effect of enhancer sites within the cistromes modelled.
Our model is currently aspatial, in that it lacks any detail of three-dimensional genomic architecture, which affects how TFs access their binding sites. Inclusion of histone data to incorporate spatial effects is a future goal.
Additionally, we remain uncertain of how the degree of cistrome overlap (number of shared segments) determines the extent of the effect of cistrome coupling on reduction of mcrit. More experimental data on this overlap, particularly in the case of multiple cistromes, is needed to investigate this.
The model presented represents a novel example of self-organization that may apply to other complex systems of interest from a theoretical point of view because it helps to demonstrate how distributed interactions among units result in higher order emergent behaviours. Such complexity could provide dynamic templates of organization upon which natural selection builds additional elaborations [5].
However, even this extremely pared down implementation of the TFBP demonstrates the ability of coupled TFBPs (equivalent to transcription patterns) to influence and modulate each other’s branching behaviour via constructive interference as suggested by the theoretical model proposed by Halley and Winkler [5].
The code for the simulation, batch scripts for running the simulation on an SGE enabled compute cluster, Python scripts for generating real or synthetic cistromes, and example R scripts for processing simulation results into graphical form, are all available on GitHUB at: github.com/CellBranch/CellBranch
In Greaves et al. [30] we detail the development of a simulation of a single cistrome branching process using the iterative CoSMoS simulation design protocol [40], taking the Transcription Factor Branching Process (TFBP) Model [2, 3, 5] as our initial domain model. That simulation was written as object-oriented code in Java, using the MASON simulation development environment. The reader is referred to Greaves et al. [30] for the relevant design details, and assumptions made about the domain. The context of the simulation development remains the same as in our previous work, i.e. the investigation of the TFBP approach to modelling stem cell fate computation. Fig 11 shows how the ChIP-Seq data is used to produce a model of a single cistrome. Fig 12 shows how this single cistrome is used to underpin the TFBP model.
In Greaves et al. [30] we have reasoned that if the activities of single TFs can be adequately described as a critical-like Branching Processes, as suggested by our results in that publication, then their interplay should define a critical-like genome-wide interference pattern. This pattern would then in some way, capture the nature of the entire pluripotency TF regulatory network [3].
We now wish to gain a deeper understanding of the behaviour of constructive interference between interfering branching processes. In particular, we aim to characterise TF branching processes and how they might propagate in the presence of cross-cistrome communication. So we now discuss the extension of the earlier, simple model Greaves et al. [30], to a model of two or more interacting, branching cistromes, to enhance the biological relevance of the simulation. We also need to allow for segments that are shared by multiple cistromes and to specify TF sharing algorithms. These refinements of the simulation require us to make further assumptions about system behaviour and to revise exiting ones. Most obviously, this includes assumptions about how cistromes communicate throughout the simulation. By communication, we mean the transfer of TF branching behaviour products from one cistrome to the appropriate binding sites on another. We assume that the inclusion of constructive interference of TFBPs is a sensible increment in the model of the system, but acknowledge that it does not yet yield a simulation of full biological relevance.
In Greaves et al. [30] we use the language of the most abstract of our models of the system, our ‘sparking posts’ model, but here we use the language of the more biological abstraction, i.e. cistromes and segments and ‘transcription’. In our new multi-cistrome model, the TF binding sites in a cistrome can be occupied by TFs transcribed from genes in the same cistrome or those from another cistrome. We have found it useful to implement the model so that we can distinguish the separate origins of incoming TFs–same cistrome or different cistrome, with those TF incoming from a different cistrome being bound to a binding site that we nicknamed the ‘spark bucket’ in our abstract sparking posts model, but which is merely the equivalent of a second TFBS for the appropriate incoming TF in our abstract biological model. We start our augmentation of the simulation by examining the case of cooperation between two TFs at a genome-wide scale. This is still far from a biologically realistic system as very many TFs will intercommunicate to regulate cellular systems i.e. multiple TFBPs will interfere constructively and destructively such that we would expect to ultimately be able to generate the interference patterns predicted to underpin cell circuitry self-organization.
Strictly speaking, in biology, any segment shared by two or more cistromes will have a TF binding site for each of the TFs produced by the cistromes which share the gene segment (remembering that in our simplified model cistrome X branching gives rise to TF X at all times if a TF product binds to a binding site in cistrome X).
However, since the model allows a segment in cistrome X to produce TF products when we have either a TF derived from cistrome X or a TF derived from any other coupled cistrome, then it is not necessary for us to include the possibility of TF transfer from more than one cistrome to the segment in Cistrome X. Again, we believe that the distinction between this description and the one we actually employ will be subtle, but acknowledge that it should be fully tested before we progress the model further.
We also assume that TFs can remain bound to their binding sites for one simulation time-step only. At each time-step a segment in a particular cistrome, let us call this cistrome X for sake of argument, can receive a TF molecule resulting from branching within Cistrome X. It can also receive a TF molecule generated by branching within another cistrome, say Y, which shares this particular segment with Cistrome X. We have further assumed that in any single time-step, only one TF from another cistrome can be transferred to a corresponding binding site in Cistrome X.
Our implementation model is illustrated in Fig 13, and is an implementation of this description of TF transfer between cistromes.
In our model, each gene segment has two TF binding sites: one for TF derived from the same cistrome and a secondary one for incoming TF from another cistrome. If either of these two sites is occupied at the end of a simulation step, then that gene segment will be activated and a TF product will be released to propagate the appropriate TF branching process. Fig 11 outlines four special cases of inter-cistrome communication in our model:
Alternatively, we could have assumed that if a segment in Cistrome X has a TF molecule resulting from branching within Cistrome X bound to it, then no TFs resulting from branching in other cistromes can be accepted by the segment in Cistrome X. However, we have decided to reject this description of the system as in the biological system, communication between cistromes will occur via shared gene segments and these segments will therefore have binding sites for both the TFs concerned.
This model has limitations from the point of view of studying biologically relevant systems, because we have deferred the inclusion of destructive interference between branching processes to a later increment in order to permit the acquisition of a fuller understanding of this simplified representation of the system prior to adding another layer of complexity to the model. We also, at this point, acknowledge that another equally valid assumption about inter-cistrome communication would be that a TF transfer cannot occur between cistromes if the destination cistrome already has any TF bound to a site in the segment under consideration. i.e. in Fig 11C, Cistrome X would not be able to transfer its bound TF to Cistrome Y’s second TFBS. We believe that this alternative model will not substantially alter the qualitative nature of the results obtained from the simulation and we mean to undertake this verification prior to any further expansion of our computational model.
c-Myc is a TF that is connected with approximately 30,000 target areas throughout the genome. Thus, c-Myc represents an evolutionarily ancient undercurrent that underpins circuitry self-organization and cell behaviour in many different ways [41, 42]. c-Myc overlaps core pluripotent TF cistromes to roughly the same extent as they overlap each other (refer to Tables 1 and 2), but with mcrit (4) being half that for the Nanog cistrome (8) and lower than that for both Oct4 and Sox2 (6 and 7 respectively).
We tested the extended simulation by first using it to replicate the results obtained from the single cistrome simulation presented in Greaves et al. [30]. We then ran simulations in which two or more cistromes were coupled under a variety of starting conditions, e.g. one or more saturated cistromes coupled to an initially dissipated cistrome.
Visualisation of results was carried out using R scripts to present plots of the proportion of ‘red’ segments activated at a given timestep or the proportion of ‘red’ segments activated at the end of the simulation (t = 1,000) against the value of the branching parameter, m.
|
10.1371/journal.ppat.1000250 | The Staphylococcus aureus Protein Sbi Acts as a Complement Inhibitor and Forms a Tripartite Complex with Host Complement Factor H and C3b | The Gram-positive bacterium Staphylococcus aureus, similar to other pathogens, binds human complement regulators Factor H and Factor H related protein 1 (FHR-1) from human serum. Here we identify the secreted protein Sbi (Staphylococcus aureus binder of IgG) as a ligand that interacts with Factor H by a—to our knowledge—new type of interaction. Factor H binds to Sbi in combination with C3b or C3d, and forms tripartite Sbi∶C3∶Factor H complexes. Apparently, the type of C3 influences the stability of the complex; surface plasmon resonance studies revealed a higher stability of C3d complexed to Sbi, as compared to C3b or C3. As part of this tripartite complex, Factor H is functionally active and displays complement regulatory activity. Sbi, by recruiting Factor H and C3b, acts as a potent complement inhibitor, and inhibits alternative pathway-mediated lyses of rabbit erythrocytes by human serum and sera of other species. Thus, Sbi is a multifunctional bacterial protein, which binds host complement components Factor H and C3 as well as IgG and β2-glycoprotein I and interferes with innate immune recognition.
| Staphylococcus aureus is a Gram-positive bacterium that can live as a commensal but can also cause severe life threatening infections in humans. Upon infection the bacterium is attacked by the host immune system, and in particular by the complement system which forms the immediate, first defence line of innate immunity. In order to survive, S. aureus has developed multiple evasion strategies and uses several virulence factors to evade and inactivate the host complement attack. Here we show that this pathogen binds the host complement regulators Factor H from human serum with the secreted and surface exposed Sbi protein, by a—to our knowledge—new type of interaction. Factor H binds to Sbi in combination with another host complement protein C3, C3b or C3d, and forms tripartite Sbi∶C3∶Factor H complexes. As part of this tripartite complex, Factor H is functionally active and inhibits further complement activation. Sbi, by recruiting Factor H and C3b, acts as a potent complement inhibitor, and inhibits alternative pathway-mediated lyses of rabbit erythrocytes by human serum and sera of different species. Thus, Sbi is a multifunctional bacterial protein, which binds host complement components Factor H and C3 as well as IgG and β2-glycoprotein I and interferes with innate immune recognition.
| In order to establish an infection pathogens have developed multiple mechanisms to avoid immune recognition and to escape host immune attack [1],[2]. Complement, which mediates a powerful immediate innate immune defense of vertebrate hosts, is activated, within seconds upon entry of a foreign invader [1],[2]. Activation of the complement system occurs through three pathways, the alternative, the classical, or the lectin binding pathway. The activated system cleaves the central complement protein C3 into the fragments C3a and C3b, and deposits C3b onto the surface of a microbe, which normally results in opsonization and elimination of the microbe by phagocytosis. This surface deposited C3b initiates further activation of the complement cascade and results ultimately in the formation of the membrane attack complex (MAC), which forms a pore in the membrane and destroys the microbe by complement-mediated lyses. However for Gram positive bacteria MAC mediated lyses seems of minor significance. The cleavage products C3a and C5a serve as potent anaphylatoxins, which attract immune effector cells to the site of infection. Non-pathogenic microbes are effectively killed and eliminated by the complement system [3].
In order to restrict complement activation to the surface of an invading microbe host cells are protected from complement attack by membrane bound and soluble regulators. Factor H is the major fluid-phase complement regulator that controls alternative pathway activation at the level of C3. The 150-kDa Factor H protein is exclusively composed of 20 structural repetitive protein domains, termed short consensus repeats (SCR) [4]. Factor H is a member of a protein family, that includes the Factor H like protein 1 (FHL-1), encoded by an alternatively spliced transcript of the Factor H gene, and five Factor H related proteins (FHRs) that are encoded by separate genes [5]. Factor H controls complement activation by acting as a cofactor for the serine protease Factor I, which cleaves surface-bound C3b into iC3b. In addition, by competing with Factor B for C3b binding Factor H accelerates the decay of the alternative pathway C3 convertase. Thus, Factor H blocks C3b deposition and amplification of the complement cascade on the cell surface [5],[6].
In order to survive and to establish an infection, pathogens need to inhibit the host complement attack and apparently utilize diverse escape mechanisms. Several pathogens acquire host fluid-phase complement regulators, like Factor H, FHL-1, FHR-1 and C4b-binding protein (C4BP) from host plasma and body fluids. Bound to the surface of a pathogen, these host regulators retain complement regulatory functions, and inhibit complement activation. Therefore, acquisition of host regulators masks the pathogenic surface, which results in survival of the pathogen [7],[8].
This common strategy of complement evasion has been identified for multiple pathogens, including Gram-positive and Gram-negative bacteria, human pathogenic fungi, parasites and viruses and several of the corresponding surface proteins were identified [1]. The vast majority of these pathogenic surface proteins bind additional host plasma proteins and display multiple functions. The M protein of Streptococcus pyogenes binds the complement regulators Factor H, FHL-1 and C4BP as well as other plasma proteins, i.e. plasminogen, fibronectin, thrombin, fibrinogen, IgA, IgG and kininogen [1], [9]–[14]. The Candida albicans surface protein Glyceratphosphat-Mutase 1 (Gpm1) binds Factor H, FHL-1 and plasminogen [15]. In addition, Complement Regulator Acquiring Surface Protein 1 (CRASP-1) of Borrelia hermsii and Tuf of P. aeruginosa, bind Factor H, FHR-1 and plasminogen [16],[17]. The additional Factor H binding pathogenic surface proteins e.g. CRASP1 of Borrelia burgdorferei, PspC of S. pneumoniae and porin protein 1A of Neisseria gonorrhoeae are candidates for combined Factor H and plasminogen binding [18]–[20]. These pathogenic surface proteins display multiple functions and interfere with the complement regulation and coagulation. Thus, multiple or potentially all pathogens acquire soluble host factors and utilize these proteins for immune evasion [1].
S. aureus is a major human pathogen responsible for hospital- and community-acquired infection. The Gram-positive bacterium permanently colonizes the human skin and mucous membranes of approximately 20% of the population [21]. Once the pathogen has crossed host immune barriers S. aureus can cause superficial skin infection, toxin-mediated diseases or serious invasive infections depending on the interaction of the pathogen's virulence factors and the defense mechanisms of the host [22]. The pathogen utilizes complex strategies to survive and disseminate within the host and expresses several virulence factors to block both innate and adaptive immune response [23].
S. aureus utilizes several proteins to control and evade the host complement attack. The cell wall-anchored protein A (SpA) binds the Fc region of IgG [24]. S. aureus expresses the zymogen staphylokinase, that cleaves human plasminogen into active plasmin, which in turn cleaves IgG. In both cases recognition of the pathogen by C1q, the initial component of the classical complement activation pathway, is inhibited [25],[26]. Sbi is an additional staphylococcal IgG-binding protein that similar to SpA interacts with the Fc part of IgG [27]. Furthermore, Sbi binds β2-glycoprotein I, which is also termed apolipoprotein H [28]. Recently, additional effector molecules of S. aureus are identified, that directly interfere with complement activation at the level of C3. The extracellular fibrinogen-binding protein (Efb), the Efb homologous protein (Ehp), and the extracellular complement-binding protein (Ecb), bind C3 and C3d, prevent further activation of C3b and consequently block the activity of C3b-containing convertases [29],[30],[31],[32],[33]. The staphylococcal complement inhibitor (SCIN) acts on surface-bound C3 convertases, C3bBb and C4b2a, by stabilizing these complexes, thereby reducing the enzymatic activity [34],[35].
Here we show binding of Factor H and FHR-1 to the surface of intact S. aureus and in addition identify the secreted staphylococcal Sbi protein as a Factor H binding protein. Native Factor H from human serum binds to Sbi, and this binding is mediated by a second serum factor, which was identified as C3. Factor H binding is increased in the presence of C3b or C3d suggesting formation of a tripartite complex. This complex blocks activation of the alternative complement pathway. The Factor H binding site of Sbi which was located to domains III and IV is distinct from the IgG binding sites which are contained in the N-terminal domains I and II [28]. Here, we demonstrate a novel mechanism for Factor H binding by Sbi. Sbi forms a tripartite complex with Factor H and C3b or C3d and this complex interferes with complement activation.
In order to analyze binding of host complement regulators to S. aureus, strain H591 was incubated in human serum. After extensive washing bound proteins were eluted, separated by SDS-PAGE, transferred to a membrane and analyzed by Western blotting. This approach identified three bands of 150, 43 and 37 kDa, which represent Factor H, FHR-1β and FHR-1α, respectively (Figure 1A, lane 2). These proteins were absent in the final wash fraction, thus suggesting specific binding (Figure 1A, lane 1 and lane 2). The same proteins were also identified in human serum (Figure 1A, lane 3). When bacteria were incubated with purified Factor H binding of the purified protein was also detected in the eluted fraction (Figure 1B, lane 2).
In order to characterize the bacterial ligand mediating this interaction we hypothesized that the staphylococcal Sbi protein might represent the binding protein. The N-terminal region of Sbi (i.e. Sbi-E) is composed of four domains and includes the IgG binding domains I and II, whereas domains III and IV lack antibody binding properties (Figure 2A and C) ([28], Burman et al. JBC in press). IgG binding of Sbi-E and Sbi-I was confirmed for one polyclonal antiserum and two monoclonal antibodies (mABs), which are directed to Factor H (Figure 2C, columns 1 and 2). Antibody binding was rather strong and exceeded the reactivity for the specific ligand Factor H (Figure 2, compare columns 5 and 1).
Sbi is an IgG binding protein, therefore Sbi-E and Sbi-I interaction with additional ligands cannot be studied by standard ELISA. Consequently we used the previously described combined ELISA and Western blot approach (CEWA) to study binding of human serum proteins to Sbi [36]. CEWA, which allows the identification of Sbi bound serum proteins by size and by reactivity with specific antisera, revealed that Factor H as well as both FHR-1α and FHR-1β bind to Sbi-E, comprising domains I–IV (Figure 3A, lane 1). Both Factor H and FHR-1α/FHR-1β bound to the deletion constructs Sbi-III/IV and with lower intensity to Sbi-IV (Figure 3A, lanes 3 and 4). The IgG binding domain Sbi-I did not bind the host complement regulators (Figure 3A, lane 2). As described previously Factor H bound to borrelial CRASP-1 and CRASP-5 and FHR-1α/FHR-1β to CRASP-5 (Figure 3A, lanes 5 and 6) [36].
Having demonstrated binding of Factor H, FHR-1α and FHR-1β from human serum to Sbi via domains III and IV, we wanted to confirm this interaction with purified proteins. However purified Factor H did not bind to Sbi, but did bind to CRASP-1 and CRASP-5 (Figure 3B). These results suggest that binding of Factor H to Sbi is mediated by an additional serum factor.
In order to identify the additional serum factor that mediates binding of the host complement regulatory, we hypothesized that the central complement component C3, which binds to the staphylococcal inhibitors Efb, Ehp and Ecb [29],[30],[32] might be such a mediator. Consequently binding of purified Factor H in the presence of the complement proteins C3b and C3d was analyzed by CEWA. When coincubated with either C3b or C3d Factor H bound to Sbi-E, Sbi-III/IV and Sbi-IV, but not to Sbi-I (Figure 4A). This binding suggests that Sbi forms a tripartite complex with Factor H and C3.
Factor H binds to domains III and IV of Sbi, but not to the IgG binding domain I. The interaction to the non-IgG binding domains was confirmed by standard ELISA. Purified Factor H together with C3b or C3d bound to Sbi-III/IV (Figure 4B, columns 1). Binding of Factor H together with C3b or C3d to Sbi-IV was rather low. In this assay the binding of Factor H together with C3b/C3d to Sbi-III/IV was more pronounced as compared to borrelial CRASP-1 (Figure 4B, compare columns 1 and 3). In addition the C3 fragment responsible for complex formation with Factor H was assayed by CEWA and ELISA (Figure 4D and 4E). The C3d-containing fragments C3, C3b and C3d mediate complex formation of Factor H with Sbi, but not C3a, C3c nor to iC3b. This result reveals a novel mechanism of capturing host immune regulators, as Sbi binds Factor H in combination with a second host ligand, namley C3.
Having identified staphylococcal Sbi as a protein that binds the host complement components Factor H together with C3b or C3d, we analyzed C3 binding and tripartite complex formation in more detail. First binding of the various forms of C3 was analyzed to immobilized Sbi-E in real time using surface plasmon resonance. C3 showed a strong association and a relative fast dissociation (Figure 5A: C3). C3b, used at the same molar ratio showed slower association, but the Sbi∶C3b complex was rather stable (Figure 5A: C3b). In addition C3d, the degradation product of C3, showed a more pronounced association and also a slow rate of dissociation (Figure 5A: C3d). This slow dissociation profile of both C3d and C3b suggests a high stability of the Sbi∶C3b and Sbi∶C3d complexes.
Based on the apparent stronger association of C3d to Sbi-E, this interaction was analyzed in more detail. Sbi-E showed a dose-dependent binding to immobilized C3d when used at a range of 200, 400 and 800 nM (Figure 5B). The same dose-dependent binding was observed in a reverse setting with immobilized Sbi-E (data not shown). These results demonstrate that C3, C3b and C3d bind directly to the staphylococcal Sbi.
In order to further analyze the interaction and complex formation Sbi-E representing domains I-IV were immobilized and complex formation was followed in real time. In this setting purified Factor H bound rather weakly to immobilized Sbi-E, while C3b binding was stronger (Figure 5C). An increase was observed in the presence of both Factor H and C3b confirming formation of a tripartite complex (Figure 5C). Formation of the tripartite complex was also analyzed with Factor H and C3d (Figure 5D). In this setting binding of C3d was similar to that of C3b (compare Figure 5D and Figure 5C) and based on the RLUs the tripartite Sbi∶C3d∶Factor H complex showed more pronounced interaction.
Binding and tripartite complex formation was analyzed to immobilized Sbi-constructs, i.e. Sbi-E, Sbi-I and Sbi-III/IV, to localize the C3 binding domains in Sbi. C3d did not bind to the IgG binding domain Sbi-I, but to Sbi-E and also to the construct Sbi-III/IV (Figure 5E). C3d interaction to Sbi-E and Sbi-III/IV was comparable, thus confirming the role of domains III and IV for the contact. Based on the strong interaction of the Sbi∶C3d∶Factor H complex to Sbi-E and to Sbi-III/IV (Figure 5E) it is concluded that the C3/Factor H interaction region of Sbi is located exclusively in Sbi domains III and IV.
To characterize the formation of Sbi∶C3d∶Factor H complex Sbi-E was coupled to an NTA-chip and complex formation was followed upon sequential addition of C3d and Factor H. Immobilization of Sbi-E was observed (Figure 5F, phase I) and upon addition of C3d formation of the Sbi∶C3d complex was followed in real time (Figure 5F, phase II). Upon addition of Factor H, a further association was detected by the increase in the surface plasmon resonance signals. These results demonstrate that Factor H binds directly to the Sbi∶C3d complex and that Factor H does not compete with C3b for Sbi-E binding (Figure 5F, phase III). The observed mass increase at the surface of the sensor chip during association of Factor H to the Sbi∶C3d complex was higher than that of Factor H to immobilized C3d (data not shown).
To further characterize this novel type of Factor H acquisition with C3, we decided to identify the Factor H domains that are involved in this interaction. Factor H deletion constructs were immobilized and used in an ELISA experiment. In the presence of C3b, Sbi-E and Sbi-III/IV, but not to Sbi-I bound to immobilized Factor H SCRs 19–20 and SCRs 15–20 (Figure 6, columns 5 and 6). In addition Sbi-I did not bind to any Factor H deletion construct. Thus the Sbi binding site was localized within the C-terminal surface binding region of Factor H, within SCRs 19–20 and is restricted to Sbi domains three and four.
Factor H bound to pathogenic ligands maintains complement regulatory activity which relates to complement evasion [1]. It was therefore of importance to assay if Factor H fixed in this tripartite complex is functionally active and has complement regulatory activity. Factor H and C3b were incubated simultaneously with immobilized Sbi-E or the deletion fragments Sbi-I, Sbi-III/IV and Sbi-IV. Subsequently, Factor I was added and the mixture was incubated further. Following this treatment the proteins were eluted, separated by SDS-PAGE and after transfer to a membrane the C3b degradation products were identified by Western blotting. Factor H bound to Sbi-E in the presence of C3b displayed cofactor activity as indicated by the disappearance of the α' band and the appearance of the α'68- and α'43 bands (Figure 7, lane 1). The same degradation profile of C3b was observed when Factor H was bound to Sbi-III/IV (Figure 7, lane 3) or to borrelial CRASP-1 (Figure 7, lane 5). In the absence of Factor H no degradation of C3b was observed (Figure 7, lanes 7 and 8). These results show that Factor H attached to Sbi in a tripartite complex maintains complement regulatory activity.
Tripartite Sbi∶C3d∶Factor H complexes represent –to our knowledge- a novel mechanism for Factor H attachment. Factor H has a C3b/C3d binding region within the C-terminal recognition region, which also forms the major contact with Sbi. Therefore we asked whether the tripartite complex is based on a sandwich type interaction, by which Sbi binds first intact C3, C3b or C3d and then Factor H. Alternatively a tripartite complex may be formed, in which Factor H directly contacts Sbi and C3. Inhibition experiments were performed to test this hypothesis and to characterize this interaction in more detail. First Factor H and C3b were incubated in the presence of Factor H antiserum and Factor H binding to immobilized Sbi was studied. Preincubation of Factor H with the specific antiserum decreased binding to Sbi-E and blocked binding to the fragments Sbi-III/IV and Sbi-IV (Figure 8A, lower panel). The weak binding of antiserum treated Factor H to intact Sbi-E and to Sbi-I is explained by binding of the Factor H∶IgG complex via the IgG binding site of Sbi located within domain I.
First binding of Factor H to immobilized Sbi-III/IV in the presence of increasing amounts of C3d was studied. Already 1 ng of C3d, resulting in a molar Factor H∶C3d ratio of 25∶1 enhanced Factor H∶Sbi interaction (Figure 8B).
Secondly, Sbi-III/IV was immobilized, C3d was added and Sbi-III/IV bound C3d was blocked with increasing amounts of specific C3d antiserum. Subsequently, the binding of Factor H was analyzed. Factor H binding was not impaired with antisera titers up to 1∶1000, and was reduced but not completely blocked at the highest titers (1∶100 and 1∶10) (Figure 8C). This result shows direct binding of Factor H to Sbi and indicates that the presence of C3d, Sbi enhances formation of the tripartite complex.
Similarly, Sbi-III/IV was immobilized and a saturating amount of C3d was bound. In order to block C3d binding sites on the Factor H protein, Factor H was preincubated with increasing concentrations of C3d prior to binding. The preincubated Factor H∶C3d complexes were added to the immobilized Sbi∶C3d complexes and after incubation Factor H binding was analyzed. Again tripartite Sbi∶C3d∶Factor H complexes were detected and complex formation was independent of the amount of C3d used for preincubation (Figure 8D). This result is in agreement with a direct Factor H∶Sbi contact. In summary the inhibition and blocking experiments reveal that Factor H binds directly to Sbi and that binding is assisted by C3d.
Staphylococcal Sbi forms a tripartite complex with host complement proteins Factor H and C3. Consequently the complement inhibitory activity of Sbi was assayed in a standard hemolysis assay, using human serum and rabbit erythrocytes. In this assay Sbi-E and also Sbi-III/IV inhibited complement-mediated lyses of rabbit erythrocytes in a dose-dependent manner. Complete inhibition was observed at a concentration of 600 ng of either Sbi-E or Sbi-III/IV (Figure 9A). In contrast, Sbi-I had no effect (data nor shown) indicating that C3b and Factor H binding is relevant for complement inhibitory activity. These results demonstrate that Sbi acts as a potent complement inhibitor. Hemolysis of rabbit erythrocytes in human serum was dose-dependent over a range from 5 to 15% and Sbi blocked hemolysis efficiently at all serum concentrations (Figure 9B).
To analyze the species range of Sbi-E the inhibitory effect of Sbi-E was tested using sera of different species. Complement mediated inhibition was observed in human, mouse and guinea pig sera, and no effect was detected in dog, goat and sheep sera (Figure 9C). Thus Sbi acts in human serum but also displays a broader species range.
Sbi is a potent complement inhibitor. Therefore we investigated the inhibitory effect of Sbi-E in all three complement pathways. Sbi-E clearly inhibited alternative pathway activity (Figure 9D, column 2 and 3). When all pathways were activated hemolysis was reduced in a dose-dependent manner, indicating that the alternative pathway, which was blocked by Sbi-E, is involved and that the classical and lectin pathway are unaffected (Figure 9D, columns 7 and 8). This effect was confirmed upon analyzing the impact on the classical and the lectin pathway. Sbi-E did not inhibit hemolysis of rabbit erythrocytes when complement was activated via the classical and the lectin pathway (Figure 9D, columns 12 and 13).
As Sbi inhibits complement we asked whether Sbi protects S. aureus from phagocytosis mediated killing. S. aureus was incubated with complement active human serum in presence or absence of Sbi-E. Subsequently bacteria were harvested and incubated together with activated phagocytic THP-1 macrophages. At the indicated times points bacteria were recovered and the survival rate was quantitated. The presence of Sbi increased bacterial survival (Figure 9E), thus indicating that the inhibitor Sbi protects bacteria from opsonisation and phagocytosis.
These results demonstrate that Sbi-E efficiently inhibits the alternative, complement pathway and aids in bacterial resistance against complement mediated phagocytosis.
The Gram-positive bacterium S. aureus, similar to other human pathogens binds the complement regulators Factor H and FHR-1 from human serum. We identify the staphylococcal Sbi protein as a ligand for the two host complement regulators. Apparently Sbi binds Factor H by a new mechanism, as this human regulator binds to Sbi together with C3, which likely results in formation of a tripartite Sbi∶C3∶Factor H complex. Arranged in this tripartite complex Factor H is functionally active and displays complement regulatory activity. Thus Sbi is a potent complement inhibitor, and inhibits the hemolytic activity of human and rodent serum on rabbit erythrocytes via the alternative pathway. Thus the multifunctional bacterial Sbi protein interferes with innate immune recognition, by acquisition of multiple host proteins in form of the complement components Factor H, C3 as well as IgG and β2-glycoprotein I.
Purified Factor H bound to intact bacteria, but dependent on the assay showed weak or even no binding to Sbi (compare Figure 1, Figure 3B and Figure 5C,D). This difference in binding suggests that intact S. aureus bacteria express an additional Factor H binding surface protein. The identification of this protein is subject to further studies. The staphylococcal Sbi protein was identified as a ligand for Factor H. However Factor H binding is enhanced in the presence of an additional complement protein C3. A tripartite Sbi∶C3b∶Factor H complex is formed (Figure 4 and Figure 5).
The Factor H contact region for Sbi is located within SCRs 19–20 (Figure 6). Very similar contact domains were identified for other microbial Factor H binding proteins, e.g. CRASP-1 and CRASP-2 of B. burgdorferi, Tuf of P. aeruginosa and Gpm1 of C. albicans [12],[14],[16],[37]. Inhibition experiments showed that polyclonal Factor H antiserum blocks Factor H binding to Sbi (Figure 8A). In the proposed tripartite complex the regulatory region of Factor H (i.e. SCRs 1–4) is freely accessible as demonstrated by the Factor I mediated cleavage of C3b (Figure 7).
The staphylococcal Sbi protein is composed of four globular N-terminal domains connected to a tyrosine-rich C-terminal domain via a prolin-rich region (Figure 2A) [38]. A recombinant fragment with domains I–IV (Sbi-E), as well as constructs Sbi-III/IV and Sbi-IV, but not Sbi-I bound Factor H in combination with C3b or C3d (Figure 4 and Figure 5), thus localizing the Factor H binding region to Sbi domains III and IV. As the Factor H/C3b binding sites in domain III and IV and the IgG binding sites in domain I and II are separated, the Sbi protein may simultaneously bind several host proteins.
The binding properties of Sbi are unique, as –to our knowledge– Sbi is the first bacterial protein identified that forms such a tripartite complex with Factor H and C3, C3b or C3d. It will be of interest to demonstrate whether other proteins of pathogen origin or virulence factors form similar tripartite complexes. The Sbi∶C3 interaction appears rather complex, as intact C3 and the two processed forms C3b and C3d display different binding profiles resulting in different stabilities (Figure 5A). C3d complexed to Sbi showed the highest binding intensity of binding, and C3b or C3 a lower interaction. In addition the rate constants of C3d and Sbi-III/IV when assayed by surface plasmon resonance did not fit a 1∶1 langmuir model of interaction, but rather fit a bivalent analyte model (Figure S2A and S2B, Figure S3).The proposed bivalent analyte interaction together with the different binding profiles for the three C3 forms suggest that C3 undergoes a conformational change upon binding to Sbi and exposes additional binding epitopes, which affect Sbi interaction, or that these C3b/C3d binding region(s) is/are differently accessible to the bacterial Sbi protein.
During complement activation C3 is cleaved, the C3 cleavage products bind to Sbi, increase Factor H binding and enhance the stability of the tripartite complex. Such a feed back regulation may increase the amount of inhibitory host regulators like Factor H at the site of infection and result in protection of the pathogen from complement attack and thus improves bacterial survival (Figure 9E).
This inhibition of the alternative pathway by Sbi indicates that Factor H bound to Sbi affects the C3 convertase. Within the tripartite complex Factor H displays complement regulatory activity (Figure 7) and seems responsible for hemolytic activity (Figure 9A and data not shown). This explains why Sbi domains III and IV display an inhibitory effect. Compared to the other staphylococcal complement regulators Efb and Ecb, Sbi does not interfere with the activity of the classical pathway and did not affect hemolysis mediated by the classical pathway (Figure 9D) [29]. Thus Sbi forms a tripartite complex with the two human complement proteins Factor H and C3, revealing- to our knowledge- a novel mechanism for complement inhibition.
The inhibitory activity of Sbi is not restricted to human complement as the protein also blocks complement of other species i.e. mouse and guinea pig. Demonstrating that Sbi is a staphylococcal virulence factor with a broader species range as compared to the human specific inhibitor SCIN, which acts specifically in the human system [29],[34].
Sbi is a potent complement inhibitor, which interferes with the hemolytic activity of human serum. In hemolytic assays with rabbit erythrocytes Sbi used at 2 µg/ml ( = 0.3 µg) exclusively blocked the alternative pathway whereas the classical and the lectin pathways were unaffected (Figure 9D). However when used at higher concentrations of 1000 µg/ml in an ELISA approach the Sbi-III/IV fragment blocks all three activation pathways of human complement but the Sbi-IV fragment is a specific inhibitor for the alternative pathway (Burman et al. JBC in press). This activity differs from that of SCIN and its homologues SCIN-B and SCIN-C, which affects all three complement pathways [29],[34].
The staphylococcal Sbi protein is a multifunctional protein which binds the complement effectors Factor H, FHR-1 and C3 and also the processed forms C3b and C3d, as well as IgG and β2-glycoprotein I. Thus Sbi mediates innate and adaptive immune escape (i) by acquiring host complement inhibitors, which correlates with the activation state of complement, (ii) by inhibiting complement activation at the level of alternative pathway C3 convertase, (iii) by binding and inactivation of IgG to avoid recognition by phagocytes, and (iv) most likely by blocking C3dg binding to complement receptor 2 (CR2) (Burman et al. JBC in press).
S. aureus strain H591 (MSSA clinical isolate, UK) was grown at 37°C in tryptic soy broth (TSB, Sigma). The strain was characterized for the presence of Sbi on DNA and protein level (Figure S1A, S1B and S1C).
Overnight cultures of S. aureus were diluted to OD600 = 0.2 in TSB and incubated for about 1.5 h at 37°C to OD600 = 1.0 (approximately 1.2×109 cfu). Cells (2×109 cfu) were harvested by centrifugation (6000 g, 8 min at room temperature), resuspended in veronal buffered saline (GVB2+, Sigma) supplemented with 10 mM EDTA and incubated with either normal human serum (NHS, diluted 1∶10) or Factor H (100 µg/ml, Aventis Behring) for 1 h at 37°C with agitation. Subsequently, the cells were washed four times with EDTA-GVB2+ and bound proteins were eluted with SDS buffer (60 mM Tris-HCl, pH 6.8, 2% SDS, 25% glycerine) for 5 min at 98°C. Wash and elute fractions were separated by SDS-PAGE, transferred to a membrane and analyzed by Western blotting using a polyclonal goat Factor H antiserum (Merck) and horseradish peroxidase (hrp) coupled rabbit anti goat antiserum (DAKO) for detection.
Recombinant fragments of the N-terminal region of Sbi (adjacent to the poly-proline region) were engineered, expressed and purified as described previously by (Burman et al. JBC in press). The following Sbi constructs were used in this study: Sbi-E (amino acids 28–266) containing IgG-binding domains I and II and C3 interacting domains III and IV; Sbi-I (amino acids 42–94); Sbi-III-IV (amino acids 150–266) and Sbi-IV (amino acids 197–266).
The Factor H deletion mutants SCRs 1–7, SCRs 8–11, SCRs 11–15, SCRs 15–18 and SCRs 19–20 were expressed in insect cells infected with recombinant baculovirus as described earlier [39]. Briefly, Spodoptera frugiperda cells (Sf9) were grown at 28°C in monolayer cultures in protein-free expression medium for insect cells (BioWhittaker). Adherent Sf9 cells were infected with recombinant virus using a multiplicity of infection of five. The culture supernatant was harvested after 9 days and recombinant Factor H constructs were purified by affinity chromatography using Ni-NTA-Agarose (Qiagen).
The complete extra cellular region, Sbi-E, and the extra cellular deletion mutants Sbi-I, Sbi-III/IV, Sbi-IV, BSA (2 µg/ml each) and Factor H (1 µg/ml) were immobilized onto a microtiter plate for 2 h at room temperature. Unspecific binding sites were blocked with 0.2% gelatine in DPBS (Lonza) over night at 4°C. After extensive washing with PBSI (3.3 mM NaH2PO4×H2O, 6.7 mM Na2HPO4, 145 mM NaCl, pH 7.2) supplemented with 0.05% Tween 20 a polyclonal rabbit SCR1–4 antiserum and the two mABs B22 and C18 (all specific for Factor H) were added for 2 h at room temperature. Protein-antibody complexes were detected using secondary horseradish peroxidase (HRP)-coupled antiserum (e.g. rabbit anti goat-hrp (DAKO) rabbit anti mouse-hrp (DAKO)) Respectively. All antibodies and antisera were used at 1∶1000 dilutions. The reaction was developed with 1,2-phenylenediamine dihydrochloride (OPD, Dako) and the absorbency was measured at 490 nm.
A combined ELISA and Western blot approach (CEWA) was used to assay Factor H binding to Sbi-E and the deletion constructs Sbi-I, Sbi-III/IV and Sbi-IV [36]. The proteins (10 µg/ml) were immobilized onto a microtiter plate over night at 4°C. After blocking with 0.2% gelatine in DPBS (Lonza) for 6 h at 4°C, NHS (diluted 1∶10), Factor H (5 µg/ml), a combination of Factor H (5 µg/ml) and C3b (10 µg/ml, Merck), or Factor H (5 µg/ml) and C3d (2,6 µg/ml, Merck) were added. For the C3-CEWA a mixture of Factor H (5 µg/ml) and C3b or C3, iC3b, C3d (each 10 µg/ml, Merck), C3c, C3a (each 10 µg/ml, Comptech) were added. Samples were incubated over night at 4°C. After extensive washing protein complexes were removed with SDS buffer, separated by SDS-PAGE and analyzed by Western blotting using a polyclonal anti C3 antibody (Calbiochem) and anti-goat – hrp (DAKO) was used for the detection of C3 and its degradation products. For Factor H detection the mAB C18, which is specific for SCR 20 of Factor H and rabbit anti mouse-hrp (DAKO) as secondary antibody was used. As positive controls the borrelial Factor H binding protein CRASP-1, and also the Factor H/FHR1 binding protein CRASP-5 (kindly provided by Dr. Peter Kraiczy (University of Frankfurt a. M.) and by Prof. Dr. Reinhard Wallich (University of Heidelberg)) and as negative control BSA were used.
The Factor H deletion constructs SCRs 1–7, SCRs 8–11, SCRs 11–15, SCRs 15–18, SCRs 19–20 and SCRs 15–20 were immobilized equimolar onto a microtiter plate over night at 4°C. After blocking with Blocking Buffer I (AppliChem) for 2 h at 37°C, a combination of C3b (5 µg/ml) and the Sbi deletion mutants Sbi-E, Sbi-I and Sbi-III/IV used at equimolar amounts were added and incubated for 1 h at room temperature. After excessive washing bound Sbi deletion mutants were detected with polyclonal Sbi antiserum (1∶1000) and a secondary horseradish peroxidase-coupled anti rabbit antiserum (1∶1000, DAKO). To analyze the complex formation, Sbi-III/IV (10 µg/ml) was coated and a combination of Factor H (15 µg/ml) and C3, C3b, C3d, C3c (Calbiochem) or C3a (15 µg/ml; Comptech) was added. The complex was detected by polyclonal goat anti Factor H (1∶1000) and rabbit anti goat-hrp (1∶1000, Dako).The reaction was developed with 1,2-phenylenediamine dihydrochloride (OPD, Dako) and the absorbency was measured at 490 nm.
Sbi-E and the extra cellular deletion mutants Sbi-I, Sbi-III/IV, Sbi-IV or CRASP-1, and BSA (10 µg/ml) were immobilized and unspecific binding sites were blocked as described. Factor H (5 µg/ml), polyclonal Factor H antiserum (diluted 1∶100) and C3b (10 µg/ml) were preincubated for 2 h at 4°C. Subsequently the mixture was added to the immobilized proteins and incubated over night at 4°C. After extensive washing protein complexes were removed from the well with SDS buffer, separated by SDS-PAGE and analyzed by Western blotting with the polyclonal rabbit SCR1–4 antiserum and swine anti rabbit-hrp (DAKO) as secondary antibody.
For determining the regulatory activity of Sbi-bound Factor H, the regulator (3 µg/ml) together with C3b (6 µg/ml) or C3b (6 µg/ml) alone were added to immobilized Sbi-E, Sbi-I, Sbi-III/IV, Sbi-IV, CRASP-1, or BSA (10 µg/ml) incubated over night at 4°C. After extensive washing Factor I (0.8 µg) was added and the mixture was incubated for 30 min at 37°C. C3b conversion to inactive C3b (iC3b) was detected after separating the protein solution by SDS-PAGE with Western blot analysis using a polyclonal goat C3 antiserum (1∶1000, Merck) and rabbit anti goat-hrp (DAKO) as secondary antibody.
Samples were separated by SDS-PAGE using 10% and 12% gels. After the transfer of the proteins onto nitrocellulose membranes by semi-dry blotting [40], the membranes were blocked with 5% (w/v) dried milk in PBSI for 30 min at room temperature and incubated with the indicated primary antibodies over night at 4°C. Antibodies were diluted in 2.5% (w/v) dried milk in PBSI. The proteins were detected by ECL using appropriate secondary antisera that was coupled with horseradish peroxidase.
Protein-protein interactions were analyzed by the surface plasmon resonance technique using a Biacore 3000 instrument (Biacore AB) as described [41]. Briefly, the staphylococcal proteins Sbi-E, Sbi-I, Sbi-III/IV or human C3d were coupled to the surface of the flow cells of the sensor chip via a standard amine-coupling procedure (carboxylated dextran chip CM5, Biacore AB) until about 2000 resonance units were reached. A control cell was prepared under identical conditions that lacked a protein. Sbi-E, Factor H, C3, C3b or C3d were diluted in DPBS (Lonza), adjusted to equal molarities and injected with a flow rate of 5 µl/min at 25°C. Alternatively, Ni2+ and Sbi-E was loaded to a NTA-chip, and C3d followed by Factor H were injected at equimolar amounts. Each interaction was analyzed at least three times.
In order to analyze the complement regulatory effect of Sbi, hemolytic assays were performed using rabbit erythrocytes (rE, Rockland). Rabbit erythrocytes represent activator surfaces for human serum and lyse due to MAC formation. Thus the complement activity correlates directly with the erythrocyte lysis as monitored by the increase in absorbance. Following preincubation of NHS with Sbi-E or Sbi-III/IV for 30 min at 37°C, 5×106 rE were added (150 µl total volume) and further incubated for 30 min at 37°C. After centrifugation (2 min, 5000 rpm) the absorbency of the supernatant was measured at 414 nm. NHS, Sbi-E and Sbi-III/IV were used at the indicated concentrations. Samples were diluted in HEPES buffer (20 mM HEPES, 144 mM NaCl, 7 mM MgCl2, 10 mM EGTA, 1% BSA, pH 7.4). The effect of Sbi on different animal sera (Innovative Research) was assayed using 30% animal serum and 2 µg (13 µg/ml) Sbi-E.
In order to analyze and distinguish between the alternative and the classical/lectin pathway complement activation was pursued in different buffers. Alternative pathway activity was measured in EGTA-HEPES buffer. Activation of all three pathways was assayed in Ca2+-HEPES buffer (20 mM HEPES, 144 mM NaCl, 5 mM CaCl2, 2,5 mM MgCl2, pH 7.4). The effect of the classical and the lectin pathway was assayed in Factor B deficient serum (Complement Technology Inc.) and the Ca2+-HEPES buffer. All three approaches (AP, AP+CP/LP and CP/LP) were analyzed in the presence of none, 0.3 µg (2 µg/ml) and 1.0 µg (6,7 µg/ml) Sbi-E.
Bacteria S. aureus strain H591 (6×104) were incubated in 40% NHS supplemented with HEPES EGTA in presence or absence of Sbi-E for 15 min at 37°C. Samples were added to 8×105 PMA primed THP-1 macrophages in antibiotic free RPMI-1640 resulting in a final Sbi-E concentration of 2 µg/ml. THP-1 cells incubated without S. aureus were used as negative control. After shaking 20 µl sample were plated hourly. Plates were incubated overnight and colonies were counted.
National Centre for Biotechnology Information (www.ncbi.nlm.nih.gov): Homo sapiens complement factor H (CFH), gi|62739185|ref|NM_000186.2|[62739185]; Homo sapiens complement factor H-related 1 (CFHR1), NM_002113.2 GI:118442838; Homo sapiens complement component 3 (C3), NM_000064.2 GI:115298677; immunoglobulin G-binding protein Sbi [Staphylococcus aureus subsp. aureus str. Newman], YP_001333351.1 GI:151222529.
|
10.1371/journal.pcbi.1006601 | Determinants of drug-target interactions at the single cell level | The physiochemical determinants of drug-target interactions in the microenvironment of the cell are complex and generally not defined by simple diffusion and intrinsic chemical reactivity. Non-specific interactions of drugs and macromolecules in cells are rarely considered formally in assessing pharmacodynamics. Here, we demonstrate that non-specific interactions lead to very slow incorporation kinetics of DNA binding drugs. We observe a rate of drug incorporation in cell nuclei three orders of magnitude slower than in vitro due to anomalous drug diffusion within cells. This slow diffusion, however, has an advantageous consequence: it leads to virtually irreversible binding of the drug to specific DNA targets in cells. We show that non-specific interactions drive slow drug diffusion manifesting as slow reaction front propagation. We study the effect of non-specific interactions in different cellular compartments by permeabilization of plasma and nuclear membranes in order to pinpoint differential compartment effects on variability in intracellular drug kinetics. These results provide the basis for a comprehensive model of the determinants of intracellular diffusion of small-molecule drugs, their target-seeking trajectories, and the consequences of these processes on the apparent kinetics of drug-target interactions.
| Small-molecule drug design assumes target binding of high affinity. Most small molecules can interact with other macromolecules in the cell ‘nonspecifically,’ i.e., with significantly lower affinity. The extent to which these nonspecific interactions influence the availability and action of the drug for its specific target depends upon the relative concentrations of drug, the specific target, and nonspecific targets. The structure of the cell is quite crowded with a highly non-uniform distribution of macromolecules that can interact with the drug of interest both specifically and nonspecifically. Thus, some compartments or micro-domains within the cell may have a comparatively high concentration of nonspecific targets, sufficient to trap the drug and retard its diffusion toward the specific target. Here, using small-molecule binding to DNA and single cell monitoring, we demonstrate that this effect results in apparently anomalous small molecule-DNA binding kinetics in cells at rates that are 1000-fold slower than in a homogeneous, dilute, aqueous environment. This slow intracellular diffusion, however, has an advantageous consequence: it leads to virtually irreversible binding of the small molecule (drug) to specific DNA targets in cells. We study and quantify the effect of nonspecific interactions between small DNA-binding molecules, including known DNA-binding drugs, in different cellular compartments in order to identify factors that account for the variability in binding kinetics among individual cells.
| Drug efficacy in vivo is notoriously difficult to predict owing, in part, to the complexity of the underlying biochemical processes that govern drug–target interactions in situ. Semi-empiric pharmacokinetic/pharmacodynamic (PK/PD) models typically describe accumulation of the drug in tissue(s) and, hence, do not address the question of variability in efficacy for individual cells, which is determined by the drug’s access to and interaction with its target(s) within a cell. Variability in drug efficacy may, therefore, be a key factor driving resistance, selection, and toxicity.
Here, we investigate factors affecting drug–target interactions at the single cell level. Our model system is a monolayer cell culture that allows continuous monitoring of drug binding to its target in individual cells. While this model system is, of course, far from ideal, provided that the free drug concentration in a given tissue is fairly uniform, cell culture experiments can meaningfully address the question of heterogeneity of response in a cell population.
We monitor the kinetics of 2’-[4-ethoxyphenyl]-5-[4-methyl-1-piperazinyl]-2,5’-bi-1H-benzimidazole trihydrochloride trihydrate (Hoechst 33342 dye) incorporation in individual cell nuclei by measuring the dye’s fluorescence signal intensity. Hoechst dye becomes significantly more fluorescent upon binding to the minor groove of DNA and, therefore, fluorescence intensity corresponds to the amount of bound target (DNA) in the nucleus.
Fluorescence microscopy permits resolution of both the temporal and spatial dependence of dye incorporation. It is instructive to investigate the incorporation process on two different spatial scales. By integrating out spatial degrees of freedom, we can assess overall dye incorporation kinetics with measurement of fluorescence intensity over time, Itot(t), for individual cells. At a sub-nuclear scale, we can analyze the time dependence of individual pixel intensities, I ( x → , t ), that typically correspond to a spatial resolution two orders of magnitude smaller than the whole nucleus in our system. Individual pixel intensities are noisy, for which reason we developed a method based on moments of distribution to characterize drug diffusion and signal ‘homogenization’ within the nucleus.
We introduce a physical multi-compartment model of drug diffusion and binding/dissociation that can explain our experimental findings. Within this model, we also incorporate the effects of membrane permeability and partitioning (as recently addressed [1]). We further extend this reaction scheme to include diffusion [2–4] and account for non-specific interactions (high capacity, low affinity) between drug and macromolecules other than intended targets.
Non-specific interactions are often driven by chemical reactions requiring close proximity of interacting species. In a crowded intracellular environment with high local concentrations of non-specific binders, this proximity can be achieved. We, therefore, incorporated non-specific binding and dissociation processes into our reaction-diffusion model.
With this detailed model, we show computationally that owing to their spatial localization in the intracellular environment, non-specific binders act as a trap, reducing extracellular drug concentration and retarding specific drug-target kinetics. The implications of these findings for drug-target interactions and pharmacological efficacy are discussed.
We used the MFC10A cell line with the NLS-Venus (nuclear) reporter for microscopy. Human epitheloid cervical carcinoma cells (HeLa cell line) were used for spectrofluorimetric measurements.
We used a spectroflurorimetric plate reader (SpectraMax Gemini) to monitor binding kinetics on a cell population-average level. To this end, HeLa cells were fixed with 4% formalin and resuspended in Dulbecco’s phosphate-buffered saline (dPBS, Sigma-D5652) at various cell densities. Next, cells were incubated with Hoechst 33342 dye (Invitrogen-H1399), and fluorescence changes over time were monitored using the microplate reader (excitation 350 nm, emission 461 nm). In order to measure free dye concentration in solution, cells were centrifuged and the collected supernatant was incubated with calf thymus DNA. Using a DNA standard (calf thymus, Sigma-D1501), we estimated the free dye concentration in the supernatant as a function of concentration and time.
Fluorescent images were taken with the Operetta High Content Imaging System (Perkin Elmer). The 20x objective was used throughout the experiments unless otherwise noted. Image processing and analysis were performed using customized imageJ and Matlab scripts (S1 Text). In brief, the cherry-NLS signals were binarized and segmented in order to generate nuclear masks, which were applied to the Hoechst channel to obtain pixel intensities. Single-cell tracking for time-lapse experiments was archived with Python/Perl/Matlab scripts.
Doxorubicin efficacy at the single cell level can be measured in terms of DNA damage biomarker(s), such as histone γ-H2Ax. In order to combine kinetic measurements in live cells with antibody staining for the γ-H2Ax marker, we performed immunofluorescence microscopy experiments as follows: Live cells were incubated with both Hoechst dye and doxorubicin at different concentrations and imaged for relatively short periods (typically three hours) that were sufficient to detect dynamic patterns in fluorescence staining. Immediately thereafter, cells were fixed with paraformaldehyde, stained with an anti-γ-H2Ax antibody, and again imaged (see Movie C in S1 File). This protocol allowed us to combine both the dynamic measurement of dye incorporation and the resulting phenotype (extent of DNA damage) for individual cells.
Fluorescence images were analyzed using custom-designed in-house programs. Briefly, the image background was subtracted using ImageJ; and nuclear segmentation, tracking, and data analysis were performed using custom MATLAB code.
Wolfram Mathematica was used to simulate reaction–diffusion model(s).
Since MFC10A cells are fairly symmetric and ellipsoidal in shape, we can identify principal axes and positions of the ‘center of mass’ using the nuclear localization sequence marker (NLS-mCherry) as a reference (N.B., NLS fluorescence intensity is stable and unchanging over the time course of these experiments).
We introduced the distance r of any given pixel from the center of mass in the xy plane. The corresponding time dependent pixel intensity is I r ( θ , t ) = I ( x → , t ) and depends, of course, on the orientation θ of the pixel, as well. If the target (DNA) distribution were symmetric in the nucleus and the shape of the nucleus were spherical, one would expect that all pixels positioned the same distance r away from the center of the nucleus would have identical dye incorporation kinetics. Similarly, for a symmetric nuclear ellipse, pixels in the xy plane satisfy the condition:
x 2 a 2 + y 2 b 2 = c o n s t = r 2 (1)
and would be expected to have identical intensities at any given time (here a, b are principal axes of the nucleus).
In reality, owing to a non-homogeneous target distribution and other factors affecting dye mobility and dye transport, pixel intensities are not identical and are noisy. Averaging over all pixels that satisfy the geometric condition of Eq (1) yields a much more robust time-dependent observable variable Ir(t) = 〈Ir(θ, t)〉 where averaging is performed over orientation angle θ.
We note that the actual measured quantities correspond to the integrated intensity in the z-dimension within the depth of the confocal plane. We take this fact into account while matching experimental and theoretical results (see S1 Text for more details).
Finally, we defined moments of the pixel intensity distribution as follows:
M n ( t ) = ∑ j I ( j , t ) r j n ∑ j I ( j , t ) (2)
where I(j, t) and rj are, respectively, time-dependent fluorescence intensity and distance from the center of mass for pixel j. This representation of the front is robust and can be defined for any nuclear geometry. This method is often used in image processing and usually referred to as the image moment method. The main advantage of this method in our case is its invariance with respect to translation, scale, and rotation [5, 6] due to movements of the cell and microscope stage.
Time traces of overall dye intensity (incorporation), Itot(t), for a typical experiment in live cells are depicted in Fig 1a. There are two striking features of these traces: (i) the characteristic time scale of drug incorporation kinetics, and (ii) the broad population distribution in individual cell kinetics. The dynamics of Hoechst dye incorporation for a typical cell (population average) is depicted in Fig 1b for various dye concentrations.
The time scale of 103 sec for micromolar dye concentrations is rather unexpected based on first principles, which we next address. The simplest way to describe dye incorporation is to assume that the kinetics is driven by second order binding and first order dissociation reactions:
d d t v ( t ) = - k ˜ on u ( t ) v ( t ) + k ˜ off [ c - v ( t ) ] (3)
where v and u are free target and drug concentrations, respectively, and c is the concentration of available binding sites (capacity). The parameters k ˜ on and k ˜ off correspond to effective association and dissociation rates, respectively. These parameters depend not only on the intrinsic reaction rates, but also on the spatial disposition of the target molecules, potential competing binding targets, obstructive barriers to free diffusion, cell membrane properties, and active transport processes in the cell. It is a straightforward exercise to demonstrate that experimentally observed values of k ˜ on and k ˜ off are very different from the corresponding intrinsic values kon and koff. Let us assume that the extracellular dye concentration is constant over time, u(t) = u0 (we will see below that this is not always the case). Under this condition, one readily derives from Eq (3) the following equation:
v ( t ) = v s t + ( c - v s t ) e - β t (4) β = k ˜ on u 0 + k ˜ off = k ˜ off ( 1 + u 0 / K d ) (5) v s t = k ˜ off β c (6)
with the steady-state dye concentration vst, dissociation constant Kd, and relaxation rate β. The intrinsic dissociation rate and dissociation constant for dye-DNA complexes in vitro (in cell free systems) have been measured by several groups [7, 8]:
k off > 10 - 1 s e c - 1 (7) K d < 10 - 8 M (8)
Based on these intrinsic parameters, one would, therefore, expect a relaxation rate β faster than 10−1 sec−1 for any dye concentration u0. For a dye concentration in the micromolar range, u0 ∼ 1 μM, the relaxation rate is dominated by the binding reaction and would be expected to be 10 sec−1. Experimentally, however, we observed a much slower relaxation rate, of the order of 10−3 sec−1 (Fig 1a and 1b). We note that replacing the intrinsic association rate kon with a conventional diffusion-driven association rate does not explain the slowness of the observed relaxation rate. First, the exponent β is a sum of two terms [see Eq (5)]. Second, a typical value for a diffusion-driven association rate for a small molecule the size of the dye interacting with DNA (in water) is 109 M−1 sec−1, an order of magnitude faster than the intrinsic observed association rate, kon. In order to eliminate factors related to evolving cell phenotype in culture (i.e., cell fate), we also fixed cells with paraformaldehyde and measured fluorescence over an extended period of time. Of note, we observed no significant effect of fixation on the dynamics of the population average by comparing the fluorescence of live and fixed cells for time periods of less than 3 hours.
The time traces of dye incorporation are shown in Figures Aa and Ab in S1 Text for dye concentrations of 8 μg/ml. Here, we used digitonin (Fig. Aa) selectively or in combination with Triton X-100 (Fig. Ab) to permeabilize either the plasma membrane alone or all cell membranes, respectively [9]. The results (Fig. Aa) show that mild digitonin treatment at moderate dye concentrations does not affect incorporation kinetics. Digitonin at high concentration (50 ug/ml or Triton X-100 (0.1%) treatment), however, has a major impact on incorporation kinetics compared to the presence of an intact nuclear membrane (Fig. Ab). We observed acceleration in the initial phase of the incorporation rate by 2.5 − 3.5 -fold with a high concentration of digitonin or with Triton X-100 treatment of fixed cells. Nevertheless, even under these conditions, the incorporation kinetics is very slow compared to in vitro behavior. Since it has been reported [9] that even 5 μg/ml digitonin is sufficient to permeabilize the plasma membrane in HeLa cells, we hypothesized that the reason for accelerated kinetics in the presence of higher concentrations of detergents might not only be a consequence of dissolution of limiting membrane structures, but also dissolution of other membrane structures in the cell under these conditions.
We next assessed the effective dissociation rate of dye from cellular DNA by means of ‘cold chase’ experiments. After overnight incubation with dye, cells were centrifuged and the supernatant containing unbound dye aspirated and replaced with dPBS, after which fluorescence intensity was monitored over time. The resulting decay in fluorescence is depicted in Figure B in S1 Text. Here we compare the fluorescence intensity of cells that were chased with dye-free PBS (dPBS) (Fig. Ba) to cells that were maintained in dye-containing solution (Fig. Bb). Note that fluorescence decay was essentially unaffected by the presence of free dye in solution. The slight and near equivalent fluorescence decay in both conditions is most likely due to dye degradation at room temperature and not dissociation from DNA.
We observed that effectively irreversible tight binding of dye, resulting in fluorescence, occurs only in intact nuclei (Figure C in S1 Text). Here, lysed cells were incubated with dye, and after achieving steady-state fluorescence, chased with dye-free buffer as described above. Unlike intact cells, the fluorescence intensity of the cell lysate decreases instantaneously (on the time scale of our typical experiments) after the chase and quickly equilibrates at a new steady-state level.
In order to tease out factors contributing to the slow kinetics of dye incorporation, we studied the spatial distribution of bound dye as a function of time. Surprisingly, we observed a reaction front propagation in live cells that lasted several minutes (cf. Fig 2a, and Movies (A, B) in S1 File).
The dependence of Ir(t) as a function of time is shown in Figure D in S1 Text for a typical spheroidal nucleus with principal axes of the nucleus a ≈ b. It is clear from the results of Figure D that dye incorporation dynamics is non-uniform (at least during the initial several minutes of monitoring). The observed front is a result of faster incorporation of the dye at the periphery of the nucleus compared with the center. While this is rather expected behavior, what is surprising, once again, is the kinetics of front propagation. Free dye diffusion in water is characterized by an estimated diffusion constant of 500 μM2 sec−1 [10] and, hence, the expected homogenization time in a nucleus of radius 20 μM is 1 sec, two to three orders of magnitude faster than what we observed experimentally.
Note that the results in Figure D suggest that after an initial period of homogenization (i.e., completed front propagation), the kinetics becomes uniform across the entire nucleus. To make this observation more apparent, we compared fluorescence intensities of the whole nucleus and sub-regions of the nucleus at different time points in Figure E in S1 Text (see S1 Text for the computational details). Here, a sub-region corresponds to 10% of all pixels situated around the geometric center of each individual nucleus (sub-regions were defined by “shrinking” the nucleus’s shape in each dimension proportionately and, hence, preserving nuclear geometry).
Comparison of sub-regional to total fluorescence intensity, indeed, demonstrates slow reaction front propagation dynamics that varies among cells. However, quantification of the dynamics based on this representation relies heavily on a uniform distribution of target density and symmetry of the nuclei.
A more direct and rigorous way to quantify and characterize front propagation is to calculate moments of the pixel intensity distribution Mn, a parameter that is not dependent on symmetries in geometry and target distribution. Typical time traces of the second moment M2 are depicted in Fig 2b for individual nuclei ([dye] = 2 μM) and for the population average (Fig 2c, different dye concentrations). Front propagation initially drives a large second moment (only a thin shell of the nucleus incorporates dye) towards a steady-state that depends on the DNA distribution. While the typical time scale of homogenization is significantly faster than the relaxation time for overall dye intensity, it is still much slower than the 1 sec time scale discussed above. Note that time traces of M2 depicted in Fig 2b display variability in both relaxation kinetics and the steady–state achieved, similar to total dye incorporation Itot. Furthermore, we observed excellent correlation between the relaxation rates of M2 and Itot time traces (cf. Fig. F in S1 Text).
The observed pattern of front propagation and incorporation suggests that the slow kinetics is driven by slow mixing of the dye in the nucleus. To confirm this hypothesis, we introduce a reaction–diffusion model that takes into account the interaction between dye and DNA, and diffusion of free dye. Assuming that DNA binding sites are largely stationary compared to dye molecules, we derive:
R ( u , v ) = k on u v - k off ( c - v ) (9) ∂ t u ( x , t ) = D ∇ x 2 u - R ( u , v ) (10) ∂ t v ( x , t ) = - R ( u , v ) (11)
Here, R is a local reaction rate and D corresponds to the diffusion coefficient of free dye in the nucleus. Note that the model implicit in Eqs (9)–(11) corresponds to a mean field description and, therefore, is not suitable for the study of variability in incorporation dynamics across the nucleus. Eqs (9)–(11) also need to be supplemented by the appropriate boundary condition:
D ∇ x u ( x , t ) = h m [ u e x t - u ( x , t ) ] , x ∈ Ω (12)
where Ω corresponds to the position of the nuclear membrane, uext is the external dye concentration, and hm is an effective mass transfer coefficient through the boundary Ω. Unlike other parameters that appear in Eqs (9)–(11), the value of hm is difficult to estimate since it depends on multiple electrostatic and other chemical properties of the cytosol and cell membranes, such as macromolecular obstructions to diffusion, partition coefficient, dielectric properties, and specific transporter kinetics. Instead, we can attempt to determine the value of the coefficient hm by fitting experimental data to Eqs (9)–(12).
Note that under the assumption that Eqs (9)–(12) correctly describe dye incorporation kinetics, the variability among individual cells is driven by the effective mass transfer coefficient hm and nuclear radius R. Indeed, all cells are exposed to an identical dye concentration in cell culture (even if that concentration is itself time-dependent), and all cells (in the same cell cycle phase) have a similar number of available binding sites.
Upon further consideration, one realizes that the model described by Eqs (9)–(11) is inadequate. Dynamics and steady-state prediction based on Eqs (9)–(11) cannot adequately explain the experimental data (cf. S1 Text). Briefly, in the steady-state, the free extracellular dye concentration will be the same as the intracellular concentration, uext = ust. Therefore, the bound dye concentration in the steady-state is completely insensitive to uext in the range of concentrations higher than Kd ∼ 0.01 μM; however, we observed a sensitivity to dye concentration in cell culture in the concentration range of 0.1 μM − 10 μM. We note that with the introduction of continuous extracellular dye depletion through the boundary condition, Eq (12) does not remedy the inadequacy of the model described by Eqs (9)–(11) (cf. S1 Text for details). Other model modifications are, therefore, required to account for the observed experimental data.
A local dye concentration in excess of the Kd is a principal reason for the failure of the simple passive diffusion model of Eqs (9)–(11) to recapitulate the observed experimental data. The introduction of a barrier (such as a limiting membrane compartment) results in slower kinetics, as we have seen for small values of the Biot number (dimensionless transfer coefficient, B i = h m R n D) (cf. S1 Text), but by itself does not lead to a reduction in the local free dye concentration at later time points. This reduction in free dye concentration could be achieved by active transport of the dye molecules through the cell membrane boundary; however, we observed that cell fixation with formaldehyde does not qualitatively change the fluorescence kinetics. Thus, we turned to other possible explanations for the experimental observations, chief among which is non-specific binding leading to apparent anomalous diffusion.
Another possible explanation for the reduction in free dye concentration is ‘buffering’ by non-specific (i.e., weaker) binding to other macromolecules in the cytoplasm and nucleus. One obvious suspect in this regard is DNA itself, since dye binding to different base pair sequences occurs and results in much lower or undetectable fluorescence.
If such nonspecific binding (low affinity, high capacity) is a correct explanation for the significant reduction in free dye concentration inside the nucleus, one would expect much higher uptake of the dye by the cells during the course of the experiment than expected from specific binding alone. Indeed, if the dye binds only to the specific high affinity sites that constitute a small fraction (∼ 1%) of total DNA, the effect of dye binding to these specific sites on total dye concentration is expected to be small. In our experimental setting, the number of cells per well is ∼ 5 ⋅ 104, and, therefore, the number of total base pairs per well bptot is ∼ 1.5 ⋅ 1014. This number can serve as the basis for a rough estimate of the number of non-specific binding sites. For a typical dye concentration of 1 μM in a cell culture well of 150 μl volume, the number of available dye molecules dyetot is ∼ 9 ⋅ 1013, which is comparable to bptot. Provided that only ∼ 1% of total DNA binds dye molecules specifically [11], depletion of the total dye pool should be negligible with exclusive specific binding. We, however, observed a significant depletion of dye not only at [dye] = 1 μM, but also at higher dye concentrations (vide infra). This finding is consistent with lower affinity binding of high capacity.
The amount of non-specific binding sites that act as a dye buffer is proportional to cell density. In order to quantitate this relationship accurately, we used suspended fixed cells, which allows one to quantitate this relationship accurately and also to monitor the remaining free dye concentration in cell culture over time. The latter measurement was obtained by cell centrifugation and subsequent analysis of dye in the cell-free supernatant. Subtracting residual dye concentration from the initial concentration, we can estimate the amount of dye taken up by the cells and compare it to the amount of DNA in the cells.
In order to assess different fluorescence conditions, we incubated different combinations of dye concentration and cell density. The resulting fluorescence intensity at late time point (18 hours of dye incubation) is shown in Figures Ga and Gb in S1 Text, where we compare the fluorescence intensity from intact fixed HeLa cells (Fig. Ga) and the extrapolated signal from the calf thymus DNA (CT) titration data set (Fig. Gb). Namely, we extrapolated a CT signal assuming 6 pg/cell DNA concentration using CT/dye titration data shown in Figure H in S1 Text. Note that high dye concentration leads to quenching of the fluorescence signal (see also [11]) in CT, for which reason we restricted our analysis to dye concentrations < 8 μg/ml.
The results of Figures Ga and Gb suggest there may exist dye binding molecules in addition to the specific binding sites in the minor groove of DNA (e.g., other DNA binding sies, RNA, and/or proteins) that would account for higher fluorescence intensity in cells compared to cell-free CT standards. The existence of buffering molecules would also explain less tight binding that manifests in significantly more gradual titration curves for cells compared to cell-free CT samples.
We estimated residual (free) dye concentration in cell suspension samples using the standard CT method (N.B., we could not measure free dye by simple light absorption owing to sensitivity limits). Cells were centrifuged at 8000 g, and the collected supernatant was incubated with a fixed concentration of CT (≈ 100 μg/ml). The CT standard was obtained by titrating various dye concentrations in the presence of the same concentration of CT as above (cf. Figure I in S1 Text). Using this approach, one can estimate the residual free dye concentration in the cell suspensions. The results are shown in Figures Gc and Gd for the corresponding raw data (Fig. Gc) and extrapolated values of free dye in supernatant samples (Fig. Gd). Owing to the second incubation step that is necessary in this approach, the original free dye was diluted two-fold, which was taken into account in the results in Fig. Gd.
We note that due to limited sensitivity of the assay, the free dye concentration could not be accurately assessed for values < 1 μg/ml. For this reason, we did not apply extrapolation to samples with initial dye concentrations less than 4 μg/ml. For high initial dye concentrations, we observed dye uptake that cannot be explained by specific DNA binding alone. Indeed, for a cell density of 2.5 ⋅ 105 cells/ml, there is approximately 1.5 μg of DNA per ml volume in solution. The dye uptake by the cells shown in Figure Gd is at least 3 times greater than the total DNA concentration, 4.5 μg/ml for [dye] = 8 μg/ml. Taking into account that only a fraction of DNA is available for specific binding (cf. CT titration data, Fig. H in S1 Text), there must exist (macro)molecules with low binding affinity and much higher concentration (capacity) than specific DNA sites to account for the magnitude of dye uptake we observed.
Before we turn to a numerical simulation of the model that takes into account non-specific binding interactions, let us demonstrate the resulting behavior in a single cell. It is intuitively clear that any binding and dissociation reactions, whether specific or non-specific, can lead to anomalous diffusion of molecule(s) in the cell by impairing the theoretical unimpeded diffusion of the molecule in the cytosol. (Anomalous diffusion has been studied in some limiting cases of these interactions under the rubrics of ‘excluded volume’ and ‘fractal structure of the cell’; for review see e.g., [12, 13]).
We demonstrate anomalous diffusion behavior for a ‘toy’ system: diffusion of a single particle (drug molecule) in bulk. In what follows, we assume that the particle undergoes a random walk on a d-dimensional lattice and can interact with particles uniformly embedded in nodes of the lattice. A diffusing particle can be in one of two possible probabilistic states, p and q (i.e., free or bound, respectively). We introduce transition rates k+ and k− between these two states (which are the microscopic analogues to kon and koff, respectively). The major advantage of the model compared to a general case is linearity and, hence, the existence of an exact solution. Indeed, the continuous version of this model yields:
R 1 ( p , q ) = k + p ( x , t ) - k - q ( x , t ) (13) ∂ t p ( x , t ) = D ∇ x 2 p - R 1 ( p , q ) (14) ∂ t q ( x , t ) = R 1 ( p , q ) (15)
subject to boundary and initial conditions:
p ( Ω , t ) = 0 (16) p ( x , 0 ) = δ ( x ) (17) q ( x , 0 ) = 0 (18)
where the boundary Ω is assumed to be very far from the origin, x = 0.
In this setting, we wish to calculate the mean square displacement 〈x2〉 of the particle from its origin:
⟨ x 2 ⟩ = ∫ d x x 2 ( p + q ) (19)
On very short time scales k+ t ≪ 1, the diffusion is normal and is described by the usual rate law 〈x2〉 = 2 dDt where d is the lattice dimensionality. In the long time regime (k+ + k−)t ≫ 1, one expects the following asymptotic behavior:
⟨ x 2 ⟩ ≃ 2 d D * t (20) D * = D k - k - + k + (21)
The asymptotic behavior Eqs (20) and (21) is due to translational symmetry, namely, diffusion and reaction processes do not depend on the position of the particle on the lattice. Indeed, since the particle can move only while in a free state, the late time asymptotic diffusion rate is proportional to the steady-state probability that the particle is free at any given time. In order to derive an exact solution to the mean square displacement in the case of the toy model Eqs (13)–(18), we first derive the relaxation dynamics of the free particle state p0(t) in the case of d = 0 (single site lattice, corresponds to D = 0 in Eqs (13)–(15)):
p 0 ( t ) = k - k - + k + + k + k - + k + e - ( k - + k + ) t (22)
The exact solution of Eqs (13)–(15) is, therefore, given by:
⟨ x 2 ⟩ = 2 d D ∫ 0 t d τ p 0 ( τ ) (23) ⟨ x 2 ⟩ = δ [ 1 - e - ( k - + k + ) t ] + 2 d D * t (24) δ = 2 d D k + ( k - + k + ) 2 (25)
Here, the integral over time in Eq (23) corresponds to the total time the diffusing particle remains in the free state during observation time t. The numerical simulation of the mean square displacement for Eqs (13)–(15) in 1 d is presented in Fig 3a along with the exact solution, Eqs (23)–(25).
The mean square displacement in the presence of association and dissociation reactions for our model exhibits anomalous diffusion in the transient time regime, Eqs (23)–(25). One expects that if interacting particles are embedded on the lattice in a non-uniform fashion, this pattern will persist much longer since under these conditions a diffusing particle will explore different regions of space, and microscopic reaction rates k+ and k− will become position-dependent.
We next considered a locally non-uniform distribution of interacting particles (which is the case for DNA binding sites) and compared the time dependence of 〈x2〉 to the case of a uniform distribution (with identical average k+ values in both cases). One may expect that after sufficient space exploration time (late time limit), 〈x2〉 would exhibit similar asymptotic behavior for both uniform and non-uniform local distributions of interacting particles. In order to demonstrate this fact, we introduced a local perturbation to the binding rate, k + = k + 0 + Δ ( x ). For periodic local perturbation, Δ ∝ sin(ωx + θ), simulated mean square displacements 〈x2〉 are shown in Fig 3b.
Even for uniformly distributed interacting particles, the diffusion is anomalous if more than a single particle performs a random walk on the lattice. This anomalous diffusion occurs because binding and dissociation rates become time-dependent. Indeed, for a given walker, the state of interacting particles at any site on the lattice depends on whether other walkers are engaged at that site.
Disregarding spatial fluctuations, we can formulate a mean-field approximation for the multiple walkers problem by substituting k+ ∼ kon ρf(t), where ρf(t) describes the time-dependent concentration of available (i.e., not bound) reactive species interacting with the walker. For low dye (walker) concentrations, we can estimate the effective diffusion rate using Eqs (20) and (21):
D * = D k o f f k o f f + k o n ρ (26) D * = D K d K d + ρ (27)
where it is assumed that the free reactive species concentration does not change significantly, ρf(t) ≈ ρ. For the sub-micromolar dissociation constant Kd of non-specific binding reported in [11] and high intracellular concentrations with lower affinity binding sites ρ ≥ 100 μM, one may expect a 102 − 103 times slower effective diffusion rate D* compared to the diffusion of dye in water, D. This slow effective diffusion constant is consistent with the time scale we observed in our experiments. We note here that the mechanism of retardation of dye transport through membrane(s) most likely is also driven by non-specific interaction between dye and lipid molecules or dye and (membrane) protein molecules present in high local concentration.
Using Eq (27) we can approximate the time-dependent changes in effective diffusion constant in the bulk phase by assuming
D e f f ( t ) = D K d K d + ρ ( t ) (28)
where ρ(t) is a time-dependent spatial average concentration of available binding sites. This is, of course, a crude approximation that completely ignores spatial fluctuations in interacting particle density. In order to estimate the time dependence of ρ(t), we (i) assume that all cells are identical, and (ii) once again ignore spatial fluctuations in the distribution of interacting particles. Under these assumptions, we derive the autonomous evolution equation for ρ(t):
d d t ρ = - k o n ( u 0 + ρ - ρ t o t ) ρ + k o f f ( ρ t o t - ρ ) (29) ρ ( 0 ) = ρ t o t (30)
where we define ρtot as a total amount (capacity) of DNA and u0 is an initial amount of dye available for each cell. The solution of the nonlinear equation, Eqs (29) and (30), is:
ρ ( t ) = ρ 1 + ρ 2 tanh ( 1 2 β t + ρ 3 ) (31) β = [ k o f f + k o n u 0 + k o n ρ 0 ] 2 - 4 [ k o n u 0 ] [ k o n ρ 0 ] (32)
where all parameters, ρ1, ρ2, ρ3, and β, depend on reaction rates and initial conditions.
In order to estimate the value of rate β, we consider a case wherein u0 ≈ ρ0 ∼ 1 μM. (Note that here ρ0 corresponds to the average concentration of DNA in culture media, not in the individual cell). In this case one derives:
β ∼ 2 k o f f ρ 0 / K d (33)
Experimentally, we observed a very slow effective dissociation rate koff ≲ 10−5 sec−1, (see Fig. B in S1 Text). Hence, the dye depletion rate can be approximated from the above as β ≲ 10−4 sec−1 for sub-millimolar non-specific dissociation constant Kd.
The derivation of the solution Eqs (31) and (32) and its generalization to the case of multiple binding species can be found in S1 Text; also note Figure Ja for a comparison of analytical and numerical solutions for this case (cf. S1 Text section, Mean-field Solution to Autonomous Binary Reaction Model).
We also used a numerical simulations scheme that allows us to trace a single “molecule” (walker) displacement during a stochastic reaction-diffusion process implemented in 3d space. The typical time traces of 〈x2〉 for mobile species in the absence and presence of interactions with stationary interacting species are shown in Figs. Jc-Je in S1 Text. A conceptual diagram of the time dependence of D ˜ is shown in Fig 3c.
In order to incorporate nonspecific binding in the model defined by Eqs (9)–(11), we introduced an additional term that corresponds to an average (lower affinity, relatively) non-specific binding site. We further assume that this non-specific binding site is immobile compared to free dye in the time course of the experiment:
R ( u , v ) = k on u ( c - v ) - k off v (34) R n ( u , v n ) = k on n u ( c n - v n ) - k off n v n (35) ∂ t u ( x , t ) = D ∇ x 2 u - R ( u , v ) - R n ( u , v n ) (36) ∂ t v ( x , t ) = R ( u , v ) (37) ∂ t v n ( x , t ) = R n ( u , v n ) (38)
Here, the superscript ns refers to a (generic) non-specific binding site. We include an estimate of two additional parameters in the model in Eqs (34)–(38), k on n and k off n, from reference [7] and assume that the concentration of non-specific binding sites is ∼ 100-fold greater than specific sites, i.e., cn ∼ 100c. The results of numerical simulation of the model described by Eqs (34)–(38) are shown in Fig 4a–4d.
The corresponding experimental results are shown in Fig 5a and 5b. Numerical simulations of the full non-specific interaction model support the prediction of the qualitative estimate above that both dye incorporation and front propagation are consistent with a slow diffusion process. The front dynamics is not described by a simple exponent as expected in the case of normal diffusion but, rather, consistent with the anomalous behavior discussed above.
We turn next to the heterogeneity in incorporation kinetics. To begin, we examine the steady-state levels of dye incorporation. A typical histogram of steady-state fluorescence intensity is presented in Figure L in S1 Text. This multi-modal distribution is likely to be driven by the cell cycle, with the two largest peaks corresponding to G1 and G2 phases. The coefficient of variation (CV) for cells in G1 and G2 states is similar and has typical values of 0.1, excluding outliers (e.g., segmentation errors in quantifying nuclear fluorescence). We observed that low dye concentrations result in very slow kinetics (cf. Fig 1b) for dynamics of the population average. This slow kinetics is difficult to study experimentally, especially with live cells (tracking individual cells becomes difficult with cell motion over long times). In order to explore a possible link between the variability in kinetics and DNA target state (such as cell cycle phase), we performed a timed double (sequential) addition experiment, viz., dye was added to cell culture in two sequential steps. If, at the first step of the experiment, the dye concentration is low and it is experimentally impossible to achieve the steady-state, adding high dye concentrations to the cells in the second step allows us to achieve steady-state equilibration. Even though conditions at final equilibrium are different from conditions after the first addition of dye, the DNA binding capacity can be resolved using this method.
Using these sequential addition experimental data, it is straightforward to confirm the existence of the buffering molecules discussed above. If the dye were not depleted from cell culture, one would expect that adding less dye in the second phase of the experiment would result in a decrease or, in the best case, no change in final fluorescence intensity. This is not the case, as shown in Figures Ma and Mb in S1 Text. As an example, consider changes in the average fluorescence intensity for the experimental conditions [dye1] = 0.25 μM, [dye2] = 0.12 μM shown in Figure Ma (brown curve). Despite adding a lower concentration of dye, the average fluorescence intensity increases. This behavior persists for higher dye concentrations. For example, the experimental conditions [dye1] = [dye2] = 1 μM also result in an increase of fluorescence intensity (Fig. Mb, red line).
The steady-state dependence of mean and CV of individual nuclear intensities on dye concentration are shown in Figure N in S1 Text. We classified cells into two cycle phases based on the final fluorescence intensity observed in the sequential addition experiment. While steady-state intensities display significant variation for different fluorescence conditions (Figures Na and Nb), their degree of variability (CV) remains roughly constant for a broad range of dye concentrations (Figures Nc and Nd).
In contrast to the narrow distribution of dye incorporation in the steady-state, relaxation kinetics toward equilibrium exhibit a much greater variance. To demonstrate this point graphically, we introduce the normalized time-dependent variable I t o t * defined as:
I t o t * = I t o t ( t ) I t o t ( T ) , (39)
where T is a final dye incubation time point. Time traces of live cells’ raw intensity Itot and normalized intensity I t o t * are shown in Fig 6a and 6b, respectively. The estimated half-life of relaxation ranges from the fastest relaxation rate, τ1/2 ≈ 10 min, to the slowest relaxation rate, τ1/2 > 60 (min), for [dye] = 1 μM, a 6-fold difference.
It is clear from Fig 6b that the relaxation rate of incorporation correlates with the cell cycle, namely, cells in G1 phase achieve equilibrium faster than those in G2 phase. Therefore, the variability in relaxation rates is actually smaller if one takes into account cell cycle state. Even allowing for this distinction, the variability in relaxation rate is still several fold higher (CV ∼ 0.6) than the variability in the steady-state fluorescence intensity (CV ∼ 0.15) (Figs. Nc, Nd, Oa and Ob in S1 Text).
In order to determine the factors controlling the variability in dye kinetics, we performed experiments on fixed cells with permeabilized membranes (using Triton X-100). The resulting kinetics is depicted in Fig 6c and 6d. Cells permeabilized with Triton X-100 display fluorescence dynamics that is initially significantly faster (about 3-fold on average) than intact cells, consistent with the microplate reader data discussed above for the HeLa cell line, (cf. Fig 5a and 5c). The variability in intensity of permeabilized cells appears significantly lower compared to that of intact cells. As a result, late time behavior becomes almost uniform for permeabilized cells.
In addition and importantly, the data from the sequential addition experiments show that variability in kinetics among cells persists after the first addition using non-permeabilized fixed cells (Fig 6e and 6f). Therefore, the factor(s) that cause variability do not “saturate” during the incubation phase. Since the interaction of the dye with membrane(s) is most likely driven by non-specific association/dissociation reactions, one would expect that saturation of binding sites would result in more uniform dynamics during the second addition phase of the experiment. This result suggests that there may exist factor(s) other than transport through the cell membrane that control(s) variability in incorporation kinetics.
Some of the fixed cells’ time traces exhibit other interesting behaviors. Namely, a few traces reach peak fluorescence intensity during the incubation period after which their fluorescence intensity decreases with time. This biphasic behavior is especially apparent for low dye concentrations (Fig 6e and 6f). We examined images of cells that exhibit this behavior and discovered that the nuclei of these cells have region(s) that incorporate dye very quickly compared to the rest of the nucleus. The very same region is responsible for a decrease in fluorescence signal after it peaks. We hypothesized that the regions with fast reaction kinetics correspond to micro-damaged areas of the nucleus (i.e, exposed/accessible DNA binding sites) owing to fixation. The effective diffusion and, hence, mixing, of the dye is, therefore, enhanced. Under this assumption, the peak fluorescence intensity is caused by a decrease in the extracellular dye concentration during the time course of the experiment (due to depletion of the free dye by cells discussed above). This observation supports the hypothesis that the reason for accelerated kinetics in the presence of detergents might not only be a consequence of membrane dissolution, but also of the presence of other binding species and compartments within the cell.
We next investigated whether anomalous and slow diffusion in cells is unique to Hoechst dye. To this end, we studied the incorporation dynamics of another DNA binding drug, doxorubicin, a potent cancer chemotherapeutic agent. In order to characterize doxorubicin incorporation, we employed an indirect method based on doxorubicin-DNA intercalation competition with Hoechst dye 33342 [14, 15]. Since the total pool of DNA sites specific for binding to doxorubicin and Hoechst dye is limited, one may expect that dye fluorescence in cells would depend on the local concentration of doxorubicin.
We, indeed, observed this antagonistic (competitive) effect at the single cell level. If doxorubicin is delivered at the same time or later than dye to cultured cells, we observed a peak pattern in time traces shown in Fig 7a and 7b.
The peak position corresponds to the point at which doxorubicin concentration in the nucleus becomes high enough to compete effectively with bound dye for specific DNA binding sites. The timing of the peak fluorescence depends on relative dye and doxorubicin concentrations in cell culture, as can be seen in the case of high or low dye concentrations shown in Fig 7a and 7b, respectively. (A similar pattern is observed in fixed cells, Figs. Qa and Qb in S1 Text).
If cells are pre-treated with doxorubicin several hours prior to dye addition, however, traces exhibit simple plateau saturation (which is [Dox]-dependent). This observation leads to the conclusion that it takes a fairly long period of time for doxorubicin to achieve sufficient intracellular concentrations to compete effectively with Hoechst dye. As in the dye case, this time period is [Dox]-dependent (see timing of peaks in Fig 7a). Thus, slow incorporation is most likely a common feature of DNA binding drugs for exactly the same reasons as for Hoechst dye: (i) high local DNA concentrations, and (ii) non-specific interactions with other macromolecules in cells. Since these factors affect both dye and doxorubicin molecules similarly, one may expect that the kinetics of dye incorporation can be used as a proxy for doxorubicin kinetics.
Surprisingly, dye homogenization in cells does not seem to be affected by co-incubation with doxorubicin. This conclusion is supported by the time traces of either moment of inertia M2 introduced above or another proxy for homogenization, the coefficient of variation in individual nuclear pixel intensities CVp. The observed dynamics of CVp is shown in Fig 7c and 7d, and unlike total intensity of incorporation (Fig 7a and 7b), is largely [Dox]-independent. (A similar pattern is seen in fixed cells, Figs. Qc and Qd in S1 Text). The most likely explanation for this behavior is the very similar effective diffusion properties of dye and doxorubicin, since one would otherwise expect non-uniform displacement of bound dye molecules throughout the nucleus.
Doxorubicin is, of course, a clinically used chemotherapeutic agent and, hence, one can quantify drug efficacy in individual cells by assessing the time course of DNA damage after incubation. We used γ-H2Ax antibody intensity as a proxy for DNA damage in cells. To simplify phenotype characterization, we dichotomized DNA damage by introducing an assay threshold. The threshold was set based on a comparison of γ-H2Ax antibody intensity in doxorubicin-treated and untreated conditions. First, we observed that dye acts as a buffer at high dye concentration by competing for binding with doxorubicin in the DNA minor groove (Fig 8a and 8b).
For high dye concentration (16 μM), the extent of DNA damage is below the threshold (corresponding to an intensity of 100 arbitrary units of γ-H2Ax antibody) for most cells. By contrast, incubation with low dye concentration (0.5 μM) leads to extensive DNA damage for a large fraction of cells. This result is consistent with the peak pattern for dye and doxorubicin co-incubation discussed above, which is also driven by competition for DNA binding. In addition, slow dynamics of drug incorporation leads to a higher extent of DNA damage, which is a non-trivial effect. To demonstrate this phenomenon, we plotted time traces of dye fluorescence intensity in individual cells treated with doxorubicin, as depicted in Fig 8c and 8d. Most of the cells that undergo DNA damage are in G2 phase, which is typically characterized by slower incorporation kinetics compared to cells in G1 phase; however, cells that exhibit a lesser degree of DNA damage in G2 phase typically achieve peak dye fluorescence intensity faster. The temporal position of the peak is related to the rate of intracellular doxorubicin accumulation. Hence, counterintuitively, cells are more likely to escape DNA damage if doxorubicin incorporation dynamics is rapid.
We observed several striking features of binding kinetics in our model system: First, both binding and dissociation of dye are much slower (by three orders of magnitude) in cells than in cell-free systems. In fact, the effective dissociation rate is so slow that binding is essentially irreversible. We show that this dye “trapping” in the nucleus is due to (i) high local DNA concentrations; (ii) higher capacity, lower affinity interactions with other macromolecules; and (iii) lipid membrane(s) partitioning and permeability characteristics. Second, we observed reaction front propagation by monitoring the spatial distribution of the dye in the nucleus over time. Temporal dynamics of front propagation is also slow compared to the dye diffusion rate in water, and is most likely controlled by the same factors as mentioned above. Third, slow drug intake/extrusion is not unique to the dye. We demonstrate that a clinically used drug (doxorubicin) that has a binding mechanism similar to the Hoechst dye also exhibits slow binding kinetics. Finally, we demonstrate that drug incorporation dynamics varies significantly among individual cells. On the characteristic time scales of our experiments (minutes to hours), some of the heterogeneity is due to the effects of the cell membrane compartments in the cell and their kinetic effects on dye entry into the cytosol and nucleus. We observed a correlation between the dynamics of drug incorporation and its efficacy in causing DNA damage using doxorubicin as a drug and dye dynamics as a proxy for the kinetic properties of individual cells.
Effectively irreversible binding has a very interesting implication in terms of distribution of incorporated drug between cells. For sub- or even micromolar drug concentrations, one expects that cells with fast incorporation kinetics would effectively serve as a sink reducing drug availability to cells with slower kinetics. This behavior might be interpreted as “passive” drug resistance in subpopulations of cells. There might be nothing biologically unique about this cell subpopulation; however, the existence of cells that can take up drug rapidly is a driving factor for the drug-resistant subpopulation. The possible clinical solution in this case might be completely counterintuitive. Instead of improving targeting of passively resistant cells, the drug-sensitive subpopulation of resistant cells needs to be treated with reagents that decrease their drug incorporation rate. A similar notion of the effective sink might be applicable on a larger spatial scale to cells in solid tumors. Some cells (e.g., those in outer layers) may act as a shield, taking up the drug, which, in turn, may facilitate drug resistance of the inner layers of cells in the tumor.
Non-specific interactions are often short range, driven by chemical reaction requiring close proximity of interacting species. Owing to a crowded intracellular environment, these interactions can effectively trap drug molecules in subcellular regions with high local concentrations of non-specific binders. Hence, non-specific interactions between drug and macromolecules present in the cell may result in slow and anomalous intracellular diffusion of drug molecules. Since the spatial organization of the intracellular micro-environments depends on cell cycle phase, one may expect that drug incorporation kinetics will also be cell cycle-dependent.
The heterogeneity of drug incorporation is not driven exclusively by cell cycle state. We observed a high degree of variability in kinetics for both G1 and G2 subpopulations of cells. While active transport has been shown to be an important factor contributing to drug incorporation efficacy on long time scales, we have not detected significant changes in short-term kinetics between live and fixed cells at the population average level (at least not in HeLa and MFC10A cell lines). Hence, other factors, such as relative spatial organization of drug targets and non-specific interacting molecules, likely drive variability in incorporation kinetics and account for anomalous diffusion characterization of the drug.
Slow drug transport through the plasma membrane is often empirically taken into account during drug design and optimization stages. We observed, however, that a slow diffusion process occurs within a cell, as well, at least for cationic DNA-binding small molecules, such as Hoechst dye and doxorubicin. The immediate consequence of this slow diffusion is a dramatic mismatch between kinetic reaction rates in vivo and in vitro, which we observed experimentally. Hence, we believe that non-specific interactions have to be taken into account in order to describe drug kinetics adequately. By so doing, it is likely that different strategies will be needed to optimize drug efficacy and minimize drug resistance.
|
10.1371/journal.pgen.1006690 | Cis-eQTL-based trans-ethnic meta-analysis reveals novel genes associated with breast cancer risk | Breast cancer is the most common solid organ malignancy and the most frequent cause of cancer death among women worldwide. Previous research has yielded insights into its genetic etiology, but there remains a gap in the understanding of genetic factors that contribute to risk, and particularly in the biological mechanisms by which genetic variation modulates risk. The National Cancer Institute’s “Up for a Challenge” (U4C) competition provided an opportunity to further elucidate the genetic basis of the disease. Our group leveraged the seven datasets made available by the U4C organizers and data from the publicly available UK Biobank cohort to examine associations between imputed gene expression and breast cancer risk. In particular, we used reference datasets describing the breast tissue and whole blood transcriptomes to impute expression levels in breast cancer cases and controls. In trans-ethnic meta-analyses of U4C and UK Biobank data, we found significant associations between breast cancer risk and the expression of RCCD1 (joint p-value: 3.6x10-06) and DHODH (p-value: 7.1x10-06) in breast tissue, as well as a suggestive association for ANKLE1 (p-value: 9.3x10-05). Expression of RCCD1 in whole blood was also suggestively associated with disease risk (p-value: 1.2x10-05), as were expression of ACAP1 (p-value: 1.9x10-05) and LRRC25 (p-value: 5.2x10-05). While genome-wide association studies (GWAS) have implicated RCCD1 and ANKLE1 in breast cancer risk, they have not identified the remaining three genes. Among the genetic variants that contributed to the predicted expression of the five genes, we found 23 nominally (p-value < 0.05) associated with breast cancer risk, among which 15 are not in high linkage disequilibrium with risk variants previously identified by GWAS. In summary, we used a transcriptome-based approach to investigate the genetic underpinnings of breast carcinogenesis. This approach provided an avenue for deciphering the functional relevance of genes and genetic variants involved in breast cancer.
| There is a clear genetic basis of breast cancer, and previous work has identified numerous genetic variants that increase risk of this common disease. However, much of the underlying genetic variation in breast cancer remains unexplained. To address this void, as part of the National Cancer Institute’s “Up for a Challenge” (U4C) competition, we undertook a large-scale study of genetically regulated gene expression and breast cancer. Specifically, we estimated gene expression levels based on germline genetics for subjects in the seven breast cancer studies provided by U4C and for subjects in the UK Biobank. We then evaluated associations between gene expression and breast cancer and detected three novel and two known breast cancer genes. These genes exhibit potential biological mechanisms for impacting breast carcinogenesis. Our work highlights the value of leveraging different sources of data to more thoroughly study the genetic basis of complex diseases.
| Breast cancer is the most common solid organ malignancy and the most frequent cause of cancer death among women worldwide [1]. Family history is among the strongest known risk factors for breast cancer. Individuals with a first-degree relative affected by the disease have a roughly two-fold increased risk of developing breast cancer themselves, and a more extensive family history or having relatives diagnosed at an earlier age confers yet greater risk [2–4]. A recent twin study estimated the heritability of breast cancer to be 31% [5], but the combination of rare variants (e.g., in BRCA1, BRCA2) identified from linkage studies (summarized in [6]) and common single nucleotide polymorphisms (SNPs) at roughly 100 loci identified from genome-wide association studies (GWAS; summarized in [7]) explain only one-third of the excess familial risk of disease [8]. Thus, a substantial gap remains in the understanding of the genetic factors that contribute to breast cancer risk.
The National Cancer Institute’s Up for a Challenge (U4C) competition offered a unique opportunity to further elucidate the genetic basis of breast cancer. Seven ethnically diverse GWAS datasets were made available in dbGaP and participants were challenged to use innovative approaches to identify novel loci, genes, and/or genomic features involved in breast cancer susceptibility. Our group leveraged the U4C genotype data along with gene expression datasets to search for evidence of additional genes involved in breast cancer susceptibility.
Recently, methods have been developed to leverage genotypic data toward imputing gene expression that can then be evaluated in association studies [9]. These methods are based on strong evidence that expression quantitative trait loci (eQTLs), which contribute to regulating gene expression levels, account for a substantial portion of the risk of various disease phenotypes [10–12]. A reference dataset with genotype and gene expression data is used to derive a set of SNPs that optimally predicts the expression of each gene. These SNPs can then be used to impute genetically regulated gene expression in datasets without measured expression data, and these imputed values can then be tested for associations with a phenotype of interest. Evaluating gene expression with respect to breast cancer risk has the potential to offer insights distinct from those available from traditional GWAS. First, associations with gene expression have clear functional interpretations. In contrast, the functional relevance of SNPs discovered by GWAS usually remains unclear. Second, association testing for genes substantially reduces the multiple testing burden relative to single variant approaches. Third, association testing for gene expression allows for rational combination of multiple SNPs, which may help to enhance weak signals.
We conducted a transcriptome-wide association study of gene expression and breast cancer risk by applying an innovative method called PrediXcan [9] that uses eQTL reference transcriptome datasets to impute genetically regulated expression. We used reference expression data from breast tissue and whole blood to identify the SNPs that predict gene expression. We then used the U4C datasets combined with data from the UK Biobank to search for genes for which predicted expression is associated with susceptibility to breast cancer. The approach provided an avenue for deciphering the functional relevance of both genes and SNPs involved in breast cancer development.
After splitting the GWAS of Breast Cancer in the African Diaspora (African Diaspora), Breast and Prostate Cancer Cohort Consortium GWAS (BPC3), and Multiethnic Cohort GWAS in African Americans, Latinos, and Japanese (MEC) datasets into sub-populations, and excluding the Nurses’ Health Study (NHS2) sub-population from the BPC3 (because it was already included in the Cancer Genetic Markers of Susceptibility Breast Cancer GWAS [CGEMS] dataset), we imputed gene expression into 14 separate discovery studies with a total of 12,079 breast cancer cases and 11,442 controls. In addition, we used 3,370 cases and 19,717 controls from the publicly available UK Biobank cohort study as a replication population [13]. Additional details of the study populations, genotyping, and quality control (QC) process are provided in Table 1 and the Materials and Methods section.
The developers of PrediXcan previously determined the cis-eQTL SNPs relevant to the estimation of gene expression in 44 distinct tissue types. The weights that should be applied to each SNP to impute transcript levels in other datasets are maintained in the publicly available database PredictDB. For our study, we elected to use the weights developed based on gene expression in breast tissue and, separately, in whole blood. We used the former for its direct relevance to breast cancer (developed based on n = 173 samples) and the latter because the weights were developed based on the largest number of samples among all tissues (n = 922).
Weights for the prediction of breast tissue expression were available for 4,473 genes based on 102,762 unique SNPs. The mean expected correlation between imputed transcript levels and true gene expression across all transcripts was 0.097. Regarding the prediction of whole blood expression, weights were available for 9,791 genes based on 249,696 unique SNPs. The mean expected correlation between imputed transcript levels and true gene expression across all transcripts was 0.145.
A meta-analysis of the U4C discovery datasets yielded 280 transcripts with imputed breast tissue levels nominally (p-value < 0.05) associated with breast cancer risk (S1A Table). We evaluated all of these genes for associations in the UK Biobank data. Our genomic inflation factor was 1.07 (λ1000 = 1.01). All genes with a p-value < 0.10 in this replication cohort and effect estimates in the same direction as the discovery effect were included in a combined meta-analysis of discovery and replication. Table 2 describes the three genes for which the combined meta-analysis showed evidence of an association with breast cancer. Decreased expression levels of RCCD1 (p-value: 3.6x10-06) and DHODH (p-value: 7.1x10-06) showed significant associations with breast cancer risk based on a Bonferroni-corrected significance threshold (0.05 / 4,473 = 1.1x10-05), and higher expression levels of ANKLE1 demonstrated a suggestive association (p-value: 9.3x10-05). The DHODH association was largely driven by the discovery dataset (p-value: 2.4x10-05) with little contribution from replication (p-value: 0.056). Estimates from each of the discovery datasets and the replication dataset are presented in S1 Fig for each of the three genes. While RCCD1 and ANKLE1 have been implicated by GWAS of breast cancer risk, DHODH has not been previously identified.
The imputed expression of genes based on whole blood yielded no statistically significant associations with breast cancer risk after multiple testing correction (Bonferroni significance threshold = 0.05 / 9,791 = 5.1x10-06) (S1B Table). Our genomic inflation factor was 1.06 (λ1000 = 1.01). However, Table 2 shows results for three genes that showed suggestive evidence of an association (p-value < 1.0x10-04). Notably, decreased expression levels of RCCD1 in whole blood (as in breast tissue) were suggestively associated with breast cancer risk (p-value: 1.2x10-05). Furthermore, we found that higher expression levels of ACAP1 (p-value: 1.9x10-05) and LRRC25 (p-value: 5.2x10-05) were suggestively associated with an increased risk of breast cancer. Estimates from each of the discovery datasets and the replication dataset are presented in S2 Fig for each of the three genes. Neither ACAP1 nor LRRC25 have previously been implicated by GWAS of breast cancer risk.
The volcano plots in S3 Fig depict the U4C and UK Biobank meta-analysis summary statistics for 4,469 breast tissue transcripts and 9,768 whole blood transcripts. Outliers with beta estimates outside three standard deviations from the mean were excluded from the plots–four for breast tissue and 23 for whole blood. The x-axis gives the beta effect sizes reflecting the fold change in gene expression between cases and controls, and the y-axis plots the corresponding -log10(p-value). S3 Fig is thus illustrative of the differential expression between cases and controls for genes across the transcriptome. For breast tissue expression (S3A Fig), we saw few genes beyond those noted above showing any evidence of association. In contrast, the distribution of p-values for whole blood expression (S3B Fig) was slightly broader, albeit with a more stringent threshold for statistical significance. However, among those genes significantly or suggestively associated with breast cancer risk, the magnitudes of the effect sizes were larger for breast tissue expression (|Beta| ≥ 0.15) than for whole blood expression (|Beta| ≤ 0.11; Table 2). For the 2,840 genes that overlapped, the correlation of the betas for the breast tissue and whole blood analyses was significant (r2 = 0.32; p-value: 2.2x10-16).
We tested for heterogeneity of the associations across studies in the meta-analysis of the U4C datasets alone, and in the meta-analysis combined with the UK data. These analyses did not indicate any statistically significant heterogeneity (p-values > 0.10). Furthermore, we did not detect heterogeneity within ancestry groups (p-values > 0.15), except for ANKLE1 in the European only meta-analysis (p-value: 0.022). Upon restricting the analysis to women with ER negative breast cancer, however, we no longer found significant heterogeneity (p-value: 0.32).
Table 2 indicates the number of SNPs identified by PredictDB for optimal prediction of the genetically regulated expression of each of the genes showing suggestive associations with breast cancer risk. PrediXcan uses an elastic net method to determine the best set of SNPs for predicting gene expression. Because elastic net allows for highly correlated variables in prediction models, some of the SNPs are in high linkage disequilibrium (LD). We evaluated associations between each of the SNPs and breast cancer risk (S2 Table); those achieving nominal (p-value < 0.05) significance in a meta-analysis of the U4C and UK Biobank data are displayed in Table 3. The tables also indicate the proportion of total weight attributed to each SNP in the gene prediction models. The sum of the relative weights for all SNPs contributing to the prediction of any given gene always equals to one, and the SNP ranking remains static. Raw weights used for gene expression prediction can be found within the GTEx and DGN PredictDB databases.
Fig 1 displays the location of eQTL SNPs for the genes for which breast tissue expression levels were associated with breast cancer risk. The y-axis indicates the strength of association between the SNPs and breast cancer risk and each point is sized based on the relative contribution of the variant to gene expression. Among the 24 SNPs predicting expression of RCCD1, rs3826033 showed the strongest association with breast cancer risk (joint p-value: 9.5x10-06). It contributed 13% of the weight for predicting RCCD1 expression, third only to rs2290202 (24%) and rs17821347 (16%). rs2290202 was also strongly associated with breast cancer risk (p-value: 1.7x10-05). It should be noted that rs3826033 and rs2290202 are in high LD (r2 = 0.97 in 1000 Genomes Phase 3 European populations), and both SNPs are within close proximity of RCCD1 relative to the other eQTL SNPs. In contrast, rs17821347 is furthest away from RCCD1 among SNPs predicting RCCD1 expression and showed no evidence of an association with breast cancer risk (p-value: 0.89). Among the remaining RCCD1 eQTLs, only rs4347602 showed a nominal association (p-value: 2.4x10-03); it has not previously been identified by GWAS.
All three nominal associations that we identified for SNPs predicting DHODH expression in breast tissue have not been implicated by GWAS. rs3213422 showed the strongest signal (p-value: 4.5x10-06) and also contributed the majority of the weight (56%) among the seven SNPs predicting of DHODH expression. Both rs2240243 and rs12708928 (r2 = 1.0) are in moderate LD with rs3213422 (r2 = 0.50 for both variants) and also showed evidence of associations with breast cancer risk (p-values: 1.0x10-03 and 1.3x10-03 respectively). After rs3213422, the second most weight was contributed by rs7190257 (16%), which showed no evidence of association (p-value: 0.77).
We identified two SNPs out of six total eQTL SNPs predicting ANKLE1 expression in breast tissue that were associated with breast cancer; both have been previously associated with breast cancer risk [14–19]. The SNPs, rs34084277 (p-value: 4.7x10-05) and rs8170 (p-value: 6.3x10-05), are in perfect LD (r2 = 1.0) and both contributed substantial weight to the prediction of ANKLE1 expression (23% and 26% respectively). Notably, rs3745162 also contributed substantial weight (24%), but showed no evidence of an association with breast cancer risk (p-value: 0.32).
Fig 2 depicts the genes for which whole blood expression levels were associated with breast cancer risk. Among the 20 RCCD1 eQTL SNPs, rs3826033 (p-value: 4.1x10-03) and rs2290202 (p-value: 5.3x10-03) contributed the most weight to prediction (33% and 29% respectively) and were the most strongly associated with breast cancer risk. The other SNPs showing evidence of an association were rs7180016 (p-value: 7.3x10-03), rs11073961 (p-value: 9.9x10-03), rs11207 (p-value: 0.016), rs2285937 (p-value: 0.023), and rs3809583 (p-value: 0.035). rs3826033, rs2290202, and rs11207 were included in the both the breast tissue and the whole blood prediction models for RCCD1 expression. Only rs11073961 and rs3809583 have not been previously implicated in breast cancer GWAS.
Among the 19 ACAP1 whole blood eQTL SNPs, five were nominally associated with breast cancer risk. Most noteworthy was rs35776863, which not only had the strongest association with breast cancer risk (p-value: 1.4x10-04), but also contributed nearly half of the weight for predicting ACAP1 expression (49%). The other SNPs showing evidence of an association were rs9892383 (p-value: 3.6x10-03), rs5412 (p-value: 8.0x10-03), rs4791423 (p-value: 0.018), and rs35721044 (p-value: 0.019). None of these SNPs have been previously implicated in breast cancer GWAS.
Out of 33 LRRC25 whole blood eQTL SNPs, five showed evidence of an association with breast cancer risk. Again, the SNP that contributed the most weight (25%), rs11668719, also showed the strongest association signal with disease risk (p-value: 1.2x10-05). The next two strongest signals were for SNPs in moderate LD with rs11668719, namely rs7257932 (r2 = 0.39; p-value: 2.5x10-04), which is the only SNP predicting LRRC25 expression previously implicated in breast cancer GWAS, and rs13344313 (r2 = 0.43; p-value: 3.2x10-03). Also suggestively associated with breast cancer risk, albeit contributing less than 0.1% of the weight for predicting LRRC25 expression, was rs3795026 (p-value: 0.013). The last SNP nominally associated with breast cancer risk was rs7251067 (p-value: 0.041).
In this transcriptome-wide association study, we identified five genes for which genetically regulated expression levels may be associated with breast cancer risk. We also found 23 unique SNPs contributing to the expression levels of these five genes that were associated with disease. Out of the 23 SNPs, seven in breast cancer genes identified by GWAS and one in a breast cancer gene previously unidentified by GWAS have been previously implicated in breast cancer or are in high LD (r2 > 0.50 in 1000 Genomes Phase 3 populations) with known risk variants. The remaining SNPs have not been previously associated with breast cancer risk.
We found that lower predicted expression of RCCD1 (i.e., RCC1 domain containing 1) in both breast tissue and whole blood was associated with increased breast cancer risk. This finding supports limited existing evidence for the role of RCCD1 in breast cancer. A 2014 GWAS of East Asian women reported a genome-wide significant association for rs2290203, which is 5,712 bp downstream of RCCD1 on 15q26.1 [20]. The authors then replicated the association in a European population. They also showed a correlation between rs2290203 and expression of RCCD1 [20], which supported a previous eQTL analysis of human monocytes that indicated that rs2290203 is a cis-eQTL for RCCD1 [21]. A more recent study identified an association between rs8037137, another 15q26.1 SNP in moderate LD with rs2290203 (r2 = 0.59 in 1000 Genomes Phase 3 European populations), and both breast and ovarian cancer [7]. The effect alleles of both rs2290203 and rs8037137 decrease RCCD1 expression [7,20], aligning with our finding that lower RCCD1 expression is associated with increased breast cancer risk. Neither rs2290203 nor rs8037137 was among the SNPs included in PredictDB for the prediction of RCCD1 expression. However, these SNPs are in LD with RCCD1 eQTL SNPs that were included in the prediction models, namely rs2290202 (r2 = 0.59 for rs2290203, r2 = 0.99 for rs8037137) and rs3826033 (r2 = 0.57, r2 = 0.96). The PrediXcan breast tissue model explains approximately 30% of the variance in RCCD1 expression, and rs2290202 and rs3826033 account for approximately 37% of that variation. The histone demethylase complex formed by RCCD1 protein with KDM8 is important for chromosomal stability and fidelity during mitosis division [22]. It is thus plausible that lower expression of RCCD1 could lead to errors in cell division that could potentially increase the risk of breast cancer. Future studies should evaluate the specific mechanisms whereby reduced RCCD1 expression could be associated with breast cancer risk.
ANKLE1 (i.e., ankyrin repeat and LEM domain containing 1) has been previously implicated in breast cancer. Both cis-eQTLs for ANKLE1, rs8170 and rs34084277, among several other SNPs in the 19p13.11 region, have been identified as breast cancer risk variants in several GWAS[8,14–19,23–25]. Little experimental evidence exists regarding associations between over- or under-expression of ANKLE1 and cancer risk. In our study, we found that higher expression levels of ANKLE1 were associated with an increased risk of breast cancer. Variants in the two SNPs positively associated with ANKLE1 expression in our study were also positively associated with breast cancer risk in previous work by Antoniou et al. [14]. With regard to the genotypic association with breast cancer risk, the effect estimates corresponding to the same risk allele were similar. Specifically, for rs8170, the A allele was positively associated with breast cancer in the previous study (OR = 1.28 among BRCA1 carriers) and our study (OR = 1.08). Although the direction of effect was not previously reported for rs34084277, this variant is in almost perfect LD with rs8170 and shares the same direction of effect in our study (OR = 1.09). ANKLE1 is an endonuclease involved in DNA damage repair pathways [26]. Its overexpression could therefore perturb the delicate balance required for DNA damage repair. That SNPs in the 19p13.11 locus have also been implicated in ovarian cancer [27,28] implies that ANKLE1 may also be involved in hormonally-mediated carcinogenic pathways.
To the best of our knowledge, DHODH, ACAP1, and LRRC25 have not been implicated in GWAS of breast cancer risk. Even though the imputation quality of DHODH (i.e., dihydroorotate dehydrogenase [quinone]), was lowest among the genes of interest in our study, we still identified a statistically significant association between decreased expression levels of DHODH in breast tissue and breast cancer risk. The existing literature regarding the directionality of association for DHODH and breast cancer is potentially inconsistent; deletion of the 16q22.2 locus has been associated with both better prognosis [29] and increased risk of metastasis [30]. Still, DHODH inhibition has been leveraged in the treatment of breast cancer. In particular, a DHODH inhibitor called brequinar has been shown to have modest activity in patients with advanced breast cancer [31]. It is thus difficult to reconcile our findings regarding disease risk with those of existing studies of disease progression.
ACAP1 (i.e., ArfGAP with coiled-coil, ankyrin repeat and PH domains 1) has not been implicated in breast cancer risk, but it has been shown to potentially play a role in disease progression. Its protein product activates the Arf6 protein [32], the expression of which has been shown to be higher in highly invasive breast cancer than in weakly invasive or noninvasive breast cancer and normal mammary epithelial cells [33]. ACAP1 also interacts with the third cytoplasmic loop of SLC2A4/GLUT4. SLC2A4 encodes a protein that functions as an insulin-regulated facilitative glucose transporter; inhibition of this gene affects cell proliferation and cell viability, suggesting a potential biological hypothesis for how ACAP1 may be involved with breast cancer [34].
LRRC25 (i.e., leucine rich repeat containing 25) is more than one megabase away from ANKLE1 at 19p13.11. It is located in a leukocyte-receptor cluster and may be involved in the activation of hematopoietic cells, which play a critical role in innate and acquired immunity [35]. If LRRC25 overexpression results in an elevated inflammatory response, then it could also increase the risk of breast cancer. In a study of the cis-eQTL activity of known cancer loci, the 19p13.11 breast cancer risk SNP rs4808801 was most significantly associated with the expression of LRRC25 (p-value: 3.2 x 10-03) [36]. rs4808801 is in high LD (r2 = 0.88 in 1000 Genomes Phase 3 European populations) with the eQTL rs7257932 that we used to impute LRRC25.
It is our understanding that ours is the first study to use PrediXcan to impute eQTLs transcriptome-wide toward evaluating associations with cancer. It is important, however, that it be interpreted in the context of some limitations. The weights housed in PredictDB were largely developed based on Caucasian samples. However, no SNPs that were monomorphic in any of the 14 U4C ancestral populations were included in our analysis. Still, whether or not the weights are valid for application in non-Caucasian populations is unclear and requires further study. Furthermore, true gene expression was unmeasured. Rather, our study evaluated estimated genetically regulated gene expression, sometimes with low imputation quality. The mean expected correlation of imputed genetically regulated gene expression and true gene expression is 0.097 for breast tissue and 0.145 for whole blood. For most genes, we would not expect the correlation to approach one given that gene expression is regulated by factors other than germline genetics, but because PrediXcan was only recently developed, an appropriate threshold for usable imputation quality is not yet definitive. In the release of PredictDB used here (dated 8/18/16), the authors only included genes that had a false discovery rate ≤ 5% based on the elastic net models used to generate the SNP weights. With respect to our results, imputation quality seemed related to the number of SNPs included in the gene expression prediction model. It is interesting, however, that we were still able to detect signal for the genes in our study for which expression was predicted by the smallest number of SNPs (ANKLE1 and DHODH). The imputation quality and included genes will likely change as updated versions of PrediXcan and PredictDB become available. How sensitive findings are to PrediXcan updates is an important consideration given that prediction is dependent on the reference panel.
In summary, by employing a transcriptome-wide approach, we identified novel associations for gene expression with breast cancer risk that have not surfaced from traditional GWAS designs. The approach also allowed for the development of new hypotheses regarding biological mechanisms at play in breast carcinogenesis. Future research focusing on the downstream effects of imputed gene expression, such as gene-gene interactions and gene co-expression networks, may further advance the characterization of breast cancer etiology.
Discovery analyses used all seven dbGaP datasets provided for the purposes of U4C: African American Breast Cancer GWAS (AABC); African Diaspora; CGEMS [37,38]; BPC3 [19,39]; San Francisco Bay Area Latina Breast Cancer Study (Latina Admixture); MEC; and Shanghai Breast Cancer Genetics Study (Shanghai). All of the U4C datasets provided case-control status, age, and principal components of race/ethnicity. Genotyping platforms varied by study as outlined in Table 1. Imputed genotypic data were also made available for U4C, but we elected to impute each dataset to the same reference panel as described later on.
We used the publicly available UK Biobank as a replication population. The UK Biobank is a cohort of 500,000 persons aged 40 to 69 recruited from across the United Kingdom between 2006 and 2010. Its protocol has been previously described [13]. In brief, every participant was evaluated at baseline in-person visits during which assessment center staff introduced a touch-screen questionnaire, conducted a brief interview, gathered physical measurements, and collected both blood and urine samples.
In an interim data release, UK Biobank has made typed genotypic data available for 152,736 individuals whose blood samples passed QC. Affymetrix genotyped 102,754 of these individuals' samples with the UK Biobank Axiom array [40] and 49,982 with the UK BiLEVE array [41]. The former array is an updated version of the latter; it includes additional novel markers that replace a small fraction of the markers used for genome-wide coverage. In all, the two arrays share over 95% of their marker content, and 806,466 SNPs that passed QC in at least one batch [41].
In addition to the typed data, UK Biobank has released imputed data for 152,249 samples that were not identified as outliers. Imputation was conducted based on a consolidation of the UK10K haplotype and the 1000 Genomes Phase 3 reference panels [42]. It resulted in a dataset of 73,355,667 SNPs, short indels, and large structural variants.
From among the individuals in the UK Biobank with imputed data available, we identified 3,370 European ancestry women diagnosed with breast cancer according to ICD-9 (174) and ICD-10 (C50) codes. Because non-breast cancers are unlikely to metastasize to breast tissue [43], we assumed that all first diagnoses of cancers in the breast were primary malignancies and included women with prior non-breast cancer diagnoses. Of the 3,370 breast cancers included in the analysis, 171 (5.1%) had a previous diagnosis of a separate cancer-related condition. A majority of these were nonmelanoma skin cancers (n = 43) or in situ conditions (n = 50); the number of cases with other malignancies was very low (n = 78, 2.3% of total cases), and including them was thus unlikely to materially alter our findings. We defined European ancestry individuals as those classified as British, Irish, or any other European background according to the baseline questionnaire.
We randomly selected 19,717 controls frequency-matched to cases by five-year age groups from among European ancestry females in the UK Biobank cohort without an ICD9 or ICD10 code for any primary or secondary diagnosis of cancer and with imputed genotypic data.
We excluded from controls any women with a previous cancer to limit the potential for bias arising from a shared genetic basis underlying different cancers. Age at the time of initial assessment was calculated by subtracting year of birth from year of assessment; month and day of birth were unavailable.
The Institutional Review Boards of each project that made the data used here publicly available approved the research. Since these are non-identifiable data, we are exempt from Institutional Review Board approval at our home institution.
For each of the seven U4C datasets and the UK Biobank case-control sub-study, we used the KING toolset to calculate pairwise kinship coefficients and remove subjects with up to second degree familial relationships. We found that all participants of the NHS1 were included in both the CGEMS and BPC3 U4C datasets. We thus excluded the NHS1 from the latter dataset. For related individuals, we retained one individual from the relationship pair for potential inclusion in our analyses.
As a first QC step for the U4C datasets, we merged all dbGaP consent groups within each of the seven studies and then checked self-reported sex against genotypic data (i.e., the X chromosome). We excluded all individuals with sex discrepancies as well as any individuals with overall call rates < 0.95. Next, we evaluated the rate of heterozygosity for all subjects. Of the seven U4C datasets, some included data from multiple sub-populations or cohorts (i.e., BPC3, MEC, and African Diaspora). As a result, we split BPC3, having already excluded the NHS1, into six datasets (Cancer Prevention Study II [CPSII], European Prospective Investigation into Cancer and Nutrition [EPIC], MEC—European, Nurses' Health Study 2 (NHS2), Polish Breast Cancer Study [PBCS], and Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial [PLCO]), MEC into two datasets (MEC—Japanese and MEC—Latina), and African Diaspora into two datasets (African and African American / Barbadian). Within the four datasets that we did not split, and in each of the ten newly created split datasets (14 datasets total), we excluded individuals with a heterozygosity rate greater than three standard deviations from the mean rate. Regarding SNP QC, we excluded those with an array genotyping rate < 0.98 in each study, as well as those with a minor allele frequency < 0.02.
Our next step was to ensure that all 14 datasets mapped to the same human reference genome (hg19). We used liftOver to lift datasets mapped to hg18 over to hg19 as necessary. We then ran SHAPEIT for haplotype phasing of each dataset. Finally, we imputed all datasets to the Haplotype Reference Consortium using Minimac3 [44].
Before being made available, UK Biobank data had already undergone extensive individual- and SNP-level QC procedures as previously described [13]. We thus used the data as provided except as outlined in the section below. We also used the imputed data provided by UK Biobank as described in the Study Populations and Genotyping section above.
We implemented principal component analysis to assess genetic ancestry in each of the 14 U4C datasets and in the UK Biobank case-control sub-study of unrelated individuals. To do so, we first LD pruned typed SNPs with r2 > 0.2 in PLINK. Then we excluded SNPs with > 0.2% missingness in the U4C datasets and > 1% missingness in the UK Biobank dataset. With the remaining data, we determined the principal components (PC) using EIGENSTRAT within smartpca [45].
Based on the PCs for the U4C datasets, we excluded any individuals outside six standard deviations along any one of the top ten principal components (S3 Table). For the UK Biobank dataset, we first focused on the top two PCs to identify any clusters of individuals that may have comprised separate sub-populations. Upon identifying one such cluster, we excluded outliers with a PC eigenvector value greater than seven standard deviations from the mean; doing so excluded individuals in the identified cluster (S3 Table).
Details of the PrediXcan method have been previously described [9]. In brief, PrediXcan uses reference datasets in which both genomic variation and gene expression levels have been measured to train additive models of gene expression. The models are constrained using an elastic net method that allows for the inclusion of highly correlated variables. Estimates from the best fit models are stored in the publicly available database PredictDB. The application of PrediXcan to GWAS datasets entails imputing gene expression across the transcriptome using the weights stored in PredictDB and correlating transcript levels with the phenotype of interest.
For these analyses, we accessed the sets of imputation weights referencing the breast tissue transcriptome from the GTEx Project and the set of weights referencing the whole blood transcriptome from the Depression Genes Network(DGN) [46,47]. The versions of PrediXcan and PredictDB used here were dated 6/29/16 and 8/18/16, respectively. We used each set of weights to impute the transcriptome in each of our 14 discovery datasets and in our replication dataset based on the subset of SNPs with imputation quality ≥ 0.3. In each dataset, we performed logistic regression to estimate the associations between imputed transcript levels and breast cancer risk, adjusted for the top ten PCs and age. Finally, we combined the results from the 14 discovery datasets and then included the replication dataset using inverse-variance-weighted fixed-effects meta-analyses. We assessed heterogeneity in the meta-analyses of the discovery U4C datasets, and in the joint meta-analyses with the UK data using Cochran’s Q-test as implemented by METAL [48].
When a joint meta-analysis indicated a suggestive association between expression of a particular gene and breast cancer risk, we evaluated associations between its cis-eQTLs and breast cancer risk. Again, we performed logistic regression adjusted for the top ten PCs and age in each dataset and then combined estimates via meta-analysis.
|
10.1371/journal.ppat.1000235 | Structure of Reovirus σ1 in Complex with Its Receptor Junctional Adhesion Molecule-A | Viral attachment to specific host receptors is the first step in viral infection and serves an essential function in the selection of target cells. Mammalian reoviruses are highly useful experimental models for studies of viral pathogenesis and show promise as vectors for oncolytics and vaccines. Reoviruses engage cells by binding to carbohydrates and the immunoglobulin superfamily member, junctional adhesion molecule-A (JAM-A). JAM-A exists at the cell surface as a homodimer formed by extensive contacts between its N-terminal immunoglobulin-like domains. We report the crystal structure of reovirus attachment protein σ1 in complex with a soluble form of JAM-A. The σ1 protein disrupts the JAM-A dimer, engaging a single JAM-A molecule via virtually the same interface that is used for JAM-A homodimerization. Thus, reovirus takes advantage of the adhesive nature of an immunoglobulin-superfamily receptor by usurping the ligand-binding site of this molecule to attach to the cell surface. The dissociation constant (KD) of the interaction between σ1 and JAM-A is 1,000-fold lower than that of the homophilic interaction between JAM-A molecules, indicating that JAM-A strongly prefers σ1 as a ligand. Analysis of reovirus mutants engineered by plasmid-based reverse genetics revealed residues in σ1 required for binding to JAM-A and infectivity of cultured cells. These studies define biophysical mechanisms of reovirus cell attachment and provide a platform for manipulating reovirus tropism to enhance vector targeting.
| Mammalian orthoreoviruses (reoviruses) are useful models for studies of virus–receptor interactions and viral pathogenesis. They are closely related in structure to adenoviruses and share similar mechanisms of cell attachment and entry. The receptor for reovirus, junctional adhesion molecule-A (JAM-A), is a component of cellular junctions and also used as a receptor by feline calicivirus. To better understand how viruses engage cellular receptors, we determined the structure of reovirus attachment protein σ1 bound to JAM-A. The structure provides an understanding of the biological function of the interaction and yields information that may enable targeting of reovirus to alternate receptors. Since the repertoire of receptors bound by a virus contributes importantly to determining which types of cells are infected, such targeting plays an essential role in gene delivery for vaccine or therapeutic applications. New cancer therapy approaches include the use of viruses, including reovirus, to lyse tumor cells. New knowledge about reovirus attachment to cellular receptors at an atomic level will help to harness the therapeutic potential of this virus.
| Viruses have evolved a variety of strategies to engage cellular receptors, often taking advantage of the adhesive properties of these molecules. Immunoglobulin superfamily (IgSF) members mediate cellular adhesion functions including antigen recognition, stabilization of intercellular junctions, adhesion to extracellular matrices, and leukocyte extravasation [1]. These cell-surface proteins are also used as receptors by many viruses [2],[3]. Junctional adhesion molecule-A (JAM-A) is an IgSF member that mediates cell-cell contacts and serves as a receptor for mammalian orthoreovirus (reovirus) [4] and feline calicivirus [5]. Reovirus serves as a tractable experimental model for studies of virus-receptor interactions and viral pathogenesis. Virtually all mammals including humans serve as hosts for reovirus infection, but disease is restricted to the very young [6]. The recent development of plasmid-based reverse genetics for reovirus offers the opportunity to manipulate these viruses for oncolytic and vaccine applications [7].
Reoviruses form icosahedral particles approximately 850 Å in diameter [6]. At the virion fivefold symmetry axes, the trimeric attachment protein, σ1, extends from pentameric turrets formed by the λ2 protein [8],[9]. A similar arrangement of a trimeric attachment protein inserted into a pentameric base is also observed for the adenovirus attachment protein, fiber [10]. The σ1 molecule is about 480 Å in length and composed of a filamentous N-terminal tail and a globular C-terminal head [8],[9]. Discrete regions of the molecule mediate binding to cell-surface receptors. Sequences in the tail bind to carbohydrate [11], which is α-linked sialic acid for serotype 3 reoviruses [12]. The σ1 head binds to JAM-A [4],[13].
Structural analysis of the C-terminal region of strain type 3 Dearing (T3D) σ1, which includes the region that binds to JAM-A [4], has revealed details of its trimeric structure [13],[14]. Residues forming the head consist of two Greek-key motifs that fold into a compact β-barrel. The topology of this structure is identical to the β-sandwich that forms the receptor-binding knob of adenovirus fiber, pointing to a distant evolutionary relationship between the two proteins [14]. Loops connecting individual strands of the σ1 β-barrel are short with the exception of the D–E loop (connecting β-strands D and E), which contains a 310 helix. N-terminal residues in the crystallized fragment form a portion of the tail, which consists of three triple β-spiral repeats. To date, the triple β-spiral motif has been observed only in adenovirus fiber [15], bacteriophage PRD1 spike [16], and avian reovirus attachment protein σC [17].
JAM-A is an important component of tight junctions between endothelial and epithelial cells [18],[19]. It is also expressed on the surface of platelets and leukocytes [20]. JAM-A influences the migration of leukocytes across endothelial and epithelial barriers in response to inflammatory cues [21],[22]. The extracellular portion of JAM-A forms a homodimer in which the monomers are partially intertwined via interactions of the membrane-distal D1 domains [23],[24]. Interestingly, the only other example of structurally similar homodimeric interactions by an IgSF member is the coxsackievirus and adenovirus receptor, CAR [25].
Domain-swapping experiments indicate that the D1 domain of JAM-A is necessary for functional interactions with reovirus [26]. Thus, our efforts to identify σ1-binding regions in JAM-A have focused on D1. Biochemical studies have identified the dimer interface as the region of JAM-A bound by reovirus σ1, and individual residues in JAM-A that are required for efficient σ1 binding are located within this interface [24],[26],[27]. In addition, complexes formed between purified σ1 head domain and purified dimeric wild-type (wt) or monomeric point-mutant forms of JAM-A are indistinguishable by size-exclusion chromatography [27], suggesting that a monomeric form of JAM-A serves as the relevant binding partner for σ1.
To define the structural basis of σ1-JAM-A interactions, we crystallized a complex of the head domain of T3D σ1 and the D1 domain of human JAM-A (hJAM-A) and determined its structure at 3.4 Å resolution. Since σ1 binds to a monomeric form of JAM-A, we determined the dissociation constant (KD) of the homophilic JAM-A interaction by analytical ultracentrifugation to define the stability of the JAM-A dimer and the mechanism of σ1-JAM-A complex formation. Finally, we used plasmid-based reverse genetics to engineer reoviruses expressing mutant forms of σ1 to determine the contributions to binding and infectivity of specific residues that contact JAM-A. These studies reveal the biochemical basis of σ1-JAM-A interactions, provide clues about how σ1 successfully competes for the JAM-A dimer interface, and establish a platform for fine-tuning receptor recognition to enhance the targeting of reovirus vectors.
A T3D σ1 fragment comprising the head domain and one β-spiral of the tail (σ1H; residues 293–455) and the D1 domain of hJAM-A (D1; residues 28–129) were purified using glutathione S-transferase (GST)-affinity purification [13],[27]. The domain boundaries were chosen to eliminate regions of known flexibility [14],[24] and retain binding capacity [13],[24],[27]. Purified σ1H was mixed with an excess of D1 to ensure saturation of binding. Following incubation, σ1H-D1 complexes were separated from excess D1 by size-exclusion chromatography and crystallized.
The structure of the σ1H-D1 complex was determined by molecular replacement and refined to 3.4 Å resolution (Table 1). The crystallographic asymmetric unit consists of two σ1H trimers, each bound to three D1 monomers. The presence of six independent copies enabled us to carry out six-fold non-crystallographic averaging of the components and refinement using non-crystallographic symmetry restraints. These techniques helped to establish a reliable model in which the main chain and most of the side chains, including those at the contact interface, are defined by satisfactory electron density. Real-space correlation plots show that the structure is in good agreement with the electron density (Figure S1). The dataset was assembled from three individual crystals, which may explain the relatively high merging R-factor of 16.3% (Rmerge, Table 1). In contrast, the refinement R-factor is relatively low at 21.0% (Rwork, Table 1). Because of sixfold non-crystallographic symmetry in the crystals, our free set of reflections, used as a control for the R-factor during refinement, is most likely not totally “free.”
The crystallized complex consists of a σ1H trimer ligated by three D1 monomers. When viewed along the three-fold non-crystallographic symmetry axis, its overall structure resembles a three-bladed propeller, with σ1H forming the hub and D1 forming the blades (Figure 1A). Each D1 monomer interacts with one σ1H monomer, making extensive contacts that shield a combined area (the sum of contact areas on both proteins) of 1622 Å2 from solvent. Crystal packing results in additional contacts between the molecules. However, the interactions we describe are common to all σ1H-D1 pairs and likely represent the physiologic complex interface. D1 residues involved in contact formation are located at the most membrane-distal (top) part of the domain and on the face that mediates homodimer formation. These regions in D1 pack tightly into a recessed region of σ1H just below the β-barrel (Figure 1B and 1C). Residues at the D1 dimer interface form extensive contacts with the D–E loop and 310 helix of σ1H at the upper boundary of the recessed region, whereas the top of D1 contacts residues in the β-spiral of the σ1H tail at its lower boundary. In comparison to structures of isolated σ1 [13],[14] and hJAM-A [24], the architecture of both σ1H and D1 in the complex are largely preserved. Differences are observed primarily in side-chain orientations at the interfaces between σ1H and D1.
Four of the six σ1H-D1 pairs present in the asymmetric unit have similar structures and feature the same interactions. The analysis of the complex presented here is based on these pairs. The remaining two σ1H-D1 pairs exhibit larger intermolecular distances of up to 1.2 Å, resulting in fewer contacts and higher crystallographic temperature factors. The total buried surface area for these two interacting pairs is about 60 Å2 less. Crystal packing is very tight for a protein complex of this size, with only 50% solvent content [28]. The largest gaps in the packing occur directly beneath the D1 chains that exhibit larger intermolecular distances to σ1H. Flash-cooling of crystals prior to data collection may have partially dislodged D1 from its binding site at these locations [29].
Reovirus σ1H engages JAM-A D1 using two main contact areas: a larger region centered at the D–E loop and its 310 helix, just below the β-barrel, and a smaller region formed by the top of the β-spiral and the α-helix (Figure 1D). These two regions resemble “jaws” that grip the D1 domain at its interdomain interface and top (Figure 2A). Although exact placement of individual atoms is not possible at 3.4 Å resolution, there is unambiguous electron density in an omit map for all side chains in the interface (Figure 3), allowing for assignment of contacts.
The upper, larger σ1H jaw contacts the D1 interdomain interface. Contacts are largely polar, featuring numerous hydrogen bonds and two salt bridges. These interactions are centered at the σ1H 310 helix, in which residues Thr380, Gly381, and Asp382 interact with D1 residues Glu61, Asn76, and Arg59, respectively (Figure 2B). These contacts are augmented by interactions between σ1H D–E loop residues Val371 and Glu384 and D1 residues Asn76, Lys78, and Lys63, and by contacts between Asp423 in the F–G loop of σ1H and the main-chain nitrogen atom of Ala81 (Figure 2C). In addition to these polar interactions, D1 residues Leu72 and Tyr75 engage in hydrophobic contacts with D–E loop residues and the terminal part of β-strand F in σ1H (Figure 2C). Previous point mutagenesis studies indicate that D1 residues Arg59, Glu61, Lys63, Leu72, Tyr75, and Asn76 contribute to σ1 binding [27]. Interestingly, most of the D1 residues engaged in interactions with σ1H form contacts of a similar nature in the JAM-A homodimer. For example, D1 residue Arg59 forms a salt bridge with Asp382 in the complex and a salt bridge with D1 residue Glu61 in the JAM-A dimer. Similarly, Leu72 and Tyr75, which mediate hydrophobic contacts in the complex, also do so in the JAM-A dimer.
Contacts mediated by the smaller, lower jaw of σ1H lack hydrogen bonds and salt bridges. Instead, extensive hydrophobic interactions with substantial surface complementarity are found, indicating that this area also plays an important role in defining specificity and providing high affinity. In σ1H, interactions involve β-spiral residue Tyr298, a mostly hydrophobic surface of the α-helix connecting the β-spiral with the β-barrel, the non-polar portion of the Arg316 side chain, and Pro377 in the D–E loop (Figure 2D). These residues surround the D1 F–G loop, which contains several partially hydrophobic residues. The nearby B–C loop of D1 also faces towards the σ1H β-spiral, with its closest contact between the hydroxyl group of D1 residue Ser57 and the tip of the β-spiral in σ1H. Ser57 also contributes to σ1 binding [26].
The majority of interactions between σ1H and D1 involve hydrophilic residues, with a surprisingly large number of charged residues participating in contact formation. Three charged σ1H residues directly mediate polar interactions with D1, and two others do so indirectly. In D1, four direct contacts are formed with charged residues. As a result, the interacting surfaces of both σ1H and D1 display strong electrostatic potentials (Figure 4A). When comparing the two, the interacting surface of σ1H has a dominant electronegative potential in the upper jaw, whereas the lower jaw is electropositive. The interacting surface of D1 is complementary to σ1H, featuring an electropositive potential at the dimer interface and a more electronegative potential at the most membrane-distal part of the domain. The importance of charged residues in the interaction between σ1 and JAM-A is highlighted by the observation that the complex dissociates at pH values lower than 5 (Figure 4B).
The σ1H-D1 complex is readily produced in solution by mixing the two components. Although JAM-A dissociates under high salt or low pH conditions [30], we were not able to detect monomeric species of JAM-A in the neutral pH, low salt conditions used for complex formation (data not shown). Thus, we conclude that complex formation requires disruption of JAM-A homodimers by σ1. This process could be facilitated by a significantly higher affinity between σ1 and JAM-A D1 compared to that of the homophilic JAM-A interaction. The dissociation constant (KD) for the σ1-JAM-A complex is in the low nanomolar range [4],[27]. To determine a KD value for the JAM-A D1 homodimer, we performed analytical ultracentrifugation experiments at near-physiological conditions (Tris pH 7.5, 100 mM NaCl). Five JAM-A D1 samples at concentrations ranging from 0.06 to 1.31 mg/mL were used for the sedimentation velocity experiments. Sedimentation velocities showed little concentration dependence of the sedimentation coefficient (Figure S2A). The main component sediments at ∼2.35 S. This value corresponds to molar masses between 19 kg/mol and 22 kg/mol, close to that expected for dimeric JAM-A D1. We also detected significant but variable amounts of a second component, sedimenting at 3.8 S. This species is most likely tetrameric JAM-A D1. While tetramers of JAM-A in solution have been observed [30], our analytical ultracentrifugation experiments did not reveal a tendency of JAM-A to form tetramers in a concentration-dependent manner, suggesting that this species is not physiologic.
Sedimentation equilibrium experiments were conducted at four different concentrations (0.16 to 1.6 mg/mL) at three different speeds. The best fit (r.m.s.d. of 1.99×10−2 with 5109 degrees of freedom) for all available data sets was for a monomer-dimer model with variable amounts of tetramer (Figure S2B). The molar mass converged to a value of 10.94 kg/mol (10.90 to 11.16 kg/mol), which is very close to the expected molar mass for monomeric JAM-A D1 (11.5 kg/mol). The KD for this fit is 1.1×10−5 M (0.8 to 1.4×10−5 M). If the molar mass is constrained to the expected value, a poorer fit (r.m.s.d. error of 2.19×10−2, 5110 degrees of freedom) is obtained. The slight mismatch between the best-fit and the expected molar mass indicates an imprecision in the calculation of the partial specific volume or density of the buffer.
To identify contributions of individual residues in σ1 to JAM-A engagement, we employed plasmid-based reverse genetics [7] to engineer mutations into the σ1 protein of reovirus strain T3D. Mutant viruses were isolated following co-transfection of murine L929 (L) cells with nine RNA-encoding plasmids corresponding to wt T3D genes and a tenth plasmid corresponding to the σ1-encoding S1 gene incorporating site-specific mutations. Thus, each resultant virus is isogenic, with the exception of the S1 gene and its protein product, σ1. Guided by the structure of the σ1H-D1 complex, we engineered individual substitutions of Thr380, Gly381, and Glu384 in the D–E loop and Asp423 in the F–G loop of the JAM-A-binding region of σ1. In addition, we also mutated Asn369, which is located at the N-terminus of the D–E loop, but does not contact JAM-A. With the exception of Asp423, these residues are conserved in sequence alignments among prototype strains from all three reovirus serotypes [14]. All mutant viruses were recovered and produced sufficient titer to allow binding and infectivity studies.
To determine effects of substitutions in the JAM-A-binding region of σ1 on viral infectivity, we adsorbed HeLa cells with the parent or mutant viruses at a multiplicity of infection (MOI) of 50 plaque-forming units (PFU) per cell and quantified infected cells in confluent fields of view following 20 h of incubation. With the exception of E384A, each of the point-mutant viruses exhibited significantly diminished infectivity in comparison to the parent strain, with the G381A mutant infecting the fewest cells (Figure 5A). T3 reoviruses bind to sialic acid, an event mediated by sequences in the σ1 tail [11],[31], which enhances attachment and infectivity in HeLa cells [11],[31]. Therefore, the parent and σ1 point-mutant viruses should retain the capacity to bind sialic acid. To determine effects on viral infectivity of mutated residues in the JAM-A-binding surface of σ1 in the absence of sialic acid binding, we pre-treated HeLa cells with A. ureafaciens neuraminidase to remove sialic acid prior to viral adsorption (Figure 5A). As expected, neuraminidase-treatment resulted in decreased infectivity for all viruses, with ∼60% fewer infected cells for the parent virus. In comparison to the parent strain, the T380A, G381A, and D423A viruses exhibited a significant decrease in viral infectivity in the absence of sialic acid. The relative decrease in infectivity of N369A compared to the parent virus following neuraminidase treatment was less than that observed in untreated cells. The explanation for this result is not clear, but it may be due to some type of cooperative interaction between the σ1 receptor-binding domains unmasked by the N369A mutant. We conclude that targeted mutations in the JAM-A-binding surface of σ1 influence viral infectivity, presumably due to altered viral avidity for JAM-A.
To determine the JAM-A-binding capacity of the mutant viruses, we captured purified JAM-A, as an N-terminal fusion with GST, on a biosensor surface and employed surface plasmon resonance (SPR) to assess viral binding [27]. Upon injection of the parent virus at 6×1012, 8×1012, and 1×1013 particles/mL, we observed specific, concentration-dependent association with JAM-A over time (Figure 5B). In accord with the infectivity results, all mutant viruses except E384A exhibited diminished binding in comparison to the parent strain, suggesting that these residues contribute significantly to interactions with JAM-A. Interestingly, the E384A mutant exhibited higher overall binding responses than the parent virus, suggesting this virus has enhanced avidity for JAM-A. However, this enhanced avidity does not appear to translate into enhanced infectivity in HeLa cells (Figure 5A).
The interaction between reovirus σ1 and JAM-A is the first step in an infectious cycle that culminates in the death of the target cell. While some reovirus strains use additional co-receptors, all strains engage JAM-A [32]. JAM-A exists as a dimer in solution [30] and most likely at the cell surface, but monomers are bound by σ1 in our crystal structure. The binding studies we report here show that formation of the σ1-JAM-A complex is clearly preferred to the formation of JAM-A homodimers. The interaction between two JAM-A molecules has a KD of 1.1×10−5 M, whereas the KD for the σ1-JAM-A interaction is about 1,000-fold lower [27]. These differences in affinity are remarkable given that the surfaces buried in the two complexes are strikingly similar in shape, almost identical in size, and share many of the same residues (Figure 6A and 6B). Why might JAM-A have a higher affinity for σ1 than for JAM-A? The structure of the JAM-A dimer [24] reveals a cavity in the dimer interface of about 6.9 Å3 in size (Figure 6C) (calculated using VOIDOO [33]). In contrast, no cavities are found in the six copies of the σ1-JAM-A complex interfaces, which feature nearly perfect surface complementarity. Cavities in protein-protein interfaces usually contain water molecules that can significantly destabilize hydrogen bonds and salt bridges by lowering the dielectric constant of the medium. Indeed, two water molecules are visible in the cavity of the JAM-A dimer interface, and two more are adjacent to this surface [24]. The presence of water at the center of the JAM-A dimer interface could thus weaken the homophilic interaction. Concordantly, the JAM-A dimer interface is dynamic, which is thought to facilitate transitions between monomeric and dimeric forms [24]. The transitional nature of the homophilic JAM-A interaction may play a role in the regulation of tight junction permeability. A similar cavity is found in the crystal structure of murine JAM-A [23], which also can bind σ1 [4].
Our results indicate that several residues in the σ1 D–E loop are especially important for efficient JAM-A engagement. Mutation of Asn369, Thr380, Gly381, or Asp423 to alanine leads to drastically impaired JAM-A binding on a biosensor surface and reduced infectivity of HeLa cells (Figure 5). These results can now be rationalized by the structure of the complex. Mutation of Gly381 would adversely affect interactions with JAM-A, as any side chain at this position would lead to steric clashes with D1 residue Tyr75. The Thr380 side chain likely shields hydrophobic interactions from solvent (Figure 2C). Moreover, since Thr380 makes extensive contacts with other σ1 residues, truncation of its side chain would likely affect the structural integrity of the 310 helix and thus diminish JAM-A binding. Changes in local structure also might explain the reduced binding observed for the N369A mutant. Although Asn369 does not directly contact D1, its location at the N-terminus of the D–E loop may help to stabilize the 310 helix. Asp423 interacts with the main chain amide group of Ala81 in JAM-A and, like Thr380, shields hydrophobic interactions from solvent. Interestingly, the E384A mutant exhibits slightly enhanced binding to JAM-A. The Glu384 side chain interacts with nearby σ1 residues His388 and Trp421 and may stabilize this region, which probably includes several water molecules bound to surrounding side chains. These interactions are likely altered to allow σ1 to bind JAM-A. We think it possible that truncation of the Glu384 side chain would facilitate this process.
To visualize how σ1 interacts with JAM-A at the cell surface, we combined the structures of the σ1H-D1 complex, the JAM-A extracellular domain [24], and the C-terminus of σ1 [14] with a model of the N-terminus of σ1 [14],[34], as previously done to generate a model of adenovirus fiber binding to CAR [35] (Figure 7). The model was produced by superimposing JAM-A [24] and a full-length model of σ1 [14] onto the σ1H-D1 complex structure. Based on the positioning of σ1 and JAM-A in the model, JAM-A must reach beyond the approaching σ1 head to access residues in the C-terminal region of the σ1 tail. Residues in the predicted β-spiral repeat region of the σ1 tail, closer to the midpoint of the σ1 molecule, are required for engagement of carbohydrate [12]. Thus, the processes of JAM-A and carbohydrate engagement are likely facilitated by regions of flexibility within both the receptor and the viral attachment protein [8],[9],[24]. Since the binding sites for JAM-A are distinct from each other in the σ1 trimer, and since D1 projects from the cell surface, it is conceivable that each σ1 trimer simultaneously engages more than one JAM-A monomer. This scenario assumes that both monomers in the JAM-A dimer are located on the same cell. Binding of σ1 would lead to separation of JAM-A dimers into monomers, both of which likely remain in close proximity and could engage the same σ1 trimer. In this fashion, several molecules of JAM-A could form a clamp that engages σ1 and tightly adheres the virus to the cell, as depicted in our model.
Although the σ1 sequence is the most divergent among the reovirus proteins, prototype and field-isolate strains of the three most prevalent reovirus serotypes use JAM-A as a receptor [32]. Based on sequence alignment, the highest degree of conservation is observed among residues in the D–E loop, suggesting that this region forms part of the JAM-A-binding site [14],[32]. However, several T3D σ1 residues that interact with JAM-A are not conserved in prototype strains type 1 Lang (T1L) and type 2 Jones (T2J) σ1 [14]. For example, reovirus T2J possesses an alanine rather than an aspartate residue at position 423. We found that a mutant reovirus containing a D423A polymorphism exhibits reduced binding to JAM-A and diminished infectivity in HeLa cells in comparison to the parent virus (Figure 5A). These observations suggest that, while the binding sites may be similar, σ1-JAM-A interactions may differ at an atomic level among the reovirus serotypes. Serotype-specific differences such as the D423A polymorphism may in turn alter the affinity of σ1 proteins for JAM-A and thus influence reovirus tropism in vivo.
Structural analyses have revealed striking similarities between reovirus σ1 and adenovirus fiber and their respective receptors, JAM-A and CAR, pointing to an evolutionary relationship in the attachment strategies used by these viruses [36],[37]. A comparison of the σ1-JAM-A complex with that of the adenovirus type 12 (Ad12) fiber knob in complex with the D1 domain of human CAR [38] reveals conserved features, providing additional support for common ancestry among the two viruses. Both attachment proteins form trimers that bind three copies of the D1 domain of the receptor. Like JAM-A, CAR uses the dimer interface and the top (B–C and F–G loops) to engage its viral ligand. Also like JAM-A, fiber-contacting residues of CAR are mainly located in and adjacent to β-strands C, C′, C″, F, and G. Moreover, the thermodynamic properties of both interactions are remarkably similar. The KD for the fiber-CAR complex is in the nanomolar range (0.5 to 1.5×10−8 M for Ad5 fiber [39]), which also is about 1,000-fold lower than the KD of homodimeric CAR interactions (1.6×10−5 M [25]). However, unlike σ1, which uses sequences in the head and tail to bind JAM-A, the CAR-binding area in Ad12 fiber is located entirely in the knob and does not include residues in the shaft. In contrast to the σ1-JAM-A complex, in which one JAM-A D1 domain exclusively contacts one σ1 monomer, CAR also has some contacts with a second subunit in the fiber knob. Thus, the two virus-receptor complexes are similar in the contact areas formed by the receptors and the thermodynamic forces that contribute to complex formation, but the viral attachment proteins engage the receptors using different binding sites. Viruses in addition to adenovirus and reovirus engage CAR and JAM-A, respectively. Coxsackievirus binds CAR [40], and feline calicivirus binds fJAM-1, the feline homologue of JAM-A [5]. Both coxsackievirus and feline calicivirus, which are spherical nonenveloped viruses, require the D1 and D2 domains of their respective receptors for binding [41],[42]. The cryo-EM structure of feline calicivirus in complex with fJAM-1 shows that the virus binds both domains of fJAM-1 with more contacts located in the D1 domain [43]. Interestingly, the cryo-EM structure of coxsackievirus in complex with CAR shows that only the distal end of the D1 domain binds to the virus, but formation of complexes appears to require both CAR D1 and D2 [41].
The capacity to redirect viral vectors to specific target cells by modification of receptor-binding capacity provides a powerful approach for delivery of an engineered viral payload to an appropriate site. For example, retargeting adenovirus from cells expressing CAR to cells expressing JAM-A has been accomplished using a chimeric adenovirus that expresses reovirus σ1 in place of adenovirus fiber [44]. Development of plasmid-based reverse genetics for reovirus [7], coupled with the oncolytic potential of this virus [45]–[49], underscores the importance of a precise understanding of σ1 interactions with cellular receptors. Here, we provide proof-of-principle that reovirus mutants with structure-guided alterations in receptor-binding capacity can be engineered. This achievement represents a first step towards designing viruses containing modified σ1 proteins to target specific sites in the host based on receptor utilization.
The majority of known three-dimensional structures of viral proteins in complex with protein receptors involve molecules of the IgSF type. In addition to the complex presented here, such receptors are components of the HIV gp120-CD4 [50], rhinovirus-ICAM-1 [51], and adenovirus-CAR [38] complexes. In each case, the receptors exist as homodimers in solution [25],[52],[53] but are engaged as monomers by their viral ligands. For JAM-A and CAR, and possibly also for CD4 and ICAM-1, engagement by viruses is incompatible with the existence of a homodimer. Whether disruption of dimers alters cellular functions of these receptors is currently unclear. Although not an IgSF receptor, the recent crystal structure of ephrin-B2 bound to the Nipah virus G glycoprotein also shows that G engages an ephrin-B2 surface that normally interacts with the receptor Eph [54]. The σ1-JAM-A structure presented here may therefore reveal an ancient mechanism by which viruses usurp existing receptor interfaces and cleverly engage them in an energetically more favorable manner.
Sequences corresponding to residues 28–129 of hJAM-A D1 (UniProtKB/Swiss-Prot entry Q9Y624) were amplified from a plasmid encoding full-length JAM-A [24] and cloned as an N-terminal GST-fusion into pGEX-4T-3 (GE Healthcare) using BamHI-XhoI restriction sites. The D1 E121A mutant was engineered from this construct [27]. JAM-A D1 and the T3D σ1 head domain (σ1H; residues 293–455; UniProtKB/Swiss-Prot entry P03528) were purified as described [13],[24], with minor modifications. Expression of the GST-D1 fusion proteins was induced in 1 L Luria Broth (Sigma-Aldrich) with 0.2 mM IPTG in Escherichia coli strain BL21 (DE3) pLysS (Novagen) at 25°C for 16 h. Bacteria were harvested by centrifugation, resuspended in 50 mM Tris [pH 7.5], 50 mM NaCl, 3 mM EDTA, 1% Triton X-100, 2 mM β-mercaptoethanol, 1 mM phenylmethylsulfonyl fluoride, and 100 µg/mL lysozyme, sonicated with 50% duty-cycle using a Branson Digital Sonifier 250, and centrifuged at 15,000×g. The clarified supernatant was passed over a 5 mL GSTrapFF column (GE Healthcare), which was washed with buffer (50 mM Tris [pH 7.5], 3 mM EDTA), ATP-Mg2+-buffer (20 mM MgSO4 and 10 mM ATP in buffer), and high-salt buffer (1 M NaCl in buffer). D1 was cleaved from GST on-column by overnight incubation with 150 units of thrombin (GE Healthcare) in 20 mM Tris [pH 7.8], 2.5 mM CaCl2, 150 mM NaCl. Induction of the σ1H construct was achieved using 0.4 mM IPTG, and bacteria were lysed using a high-pressure homogenizer (Avestin EmulsiFlex). After removal of GST, the sequence of each protein was identical to the native sequence with the exception of two amino acids at the N-terminus: Gly291 and Ser292 for σ1H and Gly26 and Ser27 for D1. None of these amino acids contribute to complex formation. Purified σ1H and D1 were mixed at a ratio of 1∶1.2 and incubated at 4°C for 30 min. Complexes were separated from excess D1 by size-exclusion chromatography in 20 mM Tris [pH 7.5], 100 mM NaCl using a Superdex 75 column (GE Healthcare). Analytical-scale size-exclusion chromatography to assay complex stability was performed using a SMART system (GE Healthcare) with a Superdex 75 PC 3.2/30 column.
The σ1H-D1 complex was concentrated to 4 mg/mL according to direct measurement of A280 and A260 (c[mg/mL] = 1.55×A280−0.76×A260). Crystals were initially obtained by mixing equal volumes of protein and 0.1 M CHES [pH 9.5], 30% polyethylene glycol 3000 (Wizard I Screen, Emerald BioSystems) at 20°C. Larger crystals were grown upon replacement of polyethylene glycol 3000 with polyethylene glycol 3350 and with streak seeding using cat whiskers (collected after natural loss). Crystals were flash-frozen with 20% glycerol as cryoprotectant. Data were collected at the X06SA beamline of the Swiss Light Source (Villigen, Switzerland) at 100 K and a wavelength of 0.92 Å using a MarCCD detector. The crystals were extremely thin. They had to be exposed for 10 seconds to an unattenuated beam to yield any diffraction beyond 4.0 Å and suffered severe radiation damage after only brief exposure. A total of 286 images from several dozen crystals were collected, and 85 of those were used to assemble the final data set. Since the radiation damage led to dramatic decreases in spot intensity for many reflections at higher resolution, we evaluated all processed data files with an in-house program, calculating the signal-to-noise ratio (I/σI) according to resolution bins for each frame in order to apply individual resolution cut-offs. This procedure significantly improved the overall quality of the data set.
Data were integrated and reduced with HKL (HKL Research). Crystals belong to the orthorhombic space group P21212 (a = 105.9 Å, b = 124.3 Å, c = 130.6 Å). The asymmetric unit consists of two σ1H trimers, each complexed with three D1 monomers. Initial phases were obtained by molecular replacement with PHASER in CCP4 [55] using the trimeric T3D σ1H structure (PDB ID 2OJ5) [13] as a search model. Molecular replacement solutions for two σ1H trimers in the asymmetric unit were readily obtained and resulted in an overall R-factor of 40.1% (30–3.4 Å). Initial attempts to locate the D1 domains of hJAM-A (PDB ID 1NBQ) [24] by molecular replacement were not successful. However, 2Fobs-Fcalc and Fobs-Fcalc electron-density maps, calculated using phases obtained from the two σ1H trimers, which account for 61% of the protein atoms present in the crystal, clearly revealed the position and location of the six D1 domains. Adding the D1 domains to the structure reduced the overall R-factor to 34.7% (30–3.4 Å) before refinement.
The structure was refined using CNS [56] and Coot [57]. Refinement was performed using rigid body refinement, simulated annealing, restrained individual B-factor refinement, and conjugate gradient minimization. B-factors were refined individually because unrestrained group B-factor refinement was unstable. No sigma-cut-off was used. For the NCS restraints, we defined two groups of restrained coordinates. NCS group one contained all six copies of σ1, and NCS group two contained six copies of JAM-A D1. Thus, we did not restrain the complexes, but we did restrain the individual components, taking into account the partially dislodged D1 molecules (see results section). In all cases, loops that participate in crystal contacts and did not have the same structures in all copies were omitted from the restraining procedure. Electron-density maps were improved using non-crystallographic symmetry averaging [58] and data sharpening [59] by adding an overall B-factor of −70 Å2 to the observed structure factors with CAD [55]. Data sharpening improved some details in the electron density map and allowed us to resolve a number of side chains that had poor electron density prior to sharpening. However, the unsharpened map was traceable. Contact areas were calculated using AREAIMOL [55]. Coordinates and structure factors have been deposited with the Protein Data Bank with the accession code 3EOY. All structural figures were prepared using PyMOL [60].
The effect of pH on complex stability was investigated by concentrating purified σ1H, wt D1, monomeric D1 E121A [27], and the σ1H-D1 complex to 10% of the original volume using Millipore 5,000 MWCO filters. Samples were diluted in 20 mM citrate buffers [pH 4.0, 4.5, or 5.0] or 20 mM Hepes [pH 7.4] and re-concentrated. This procedure was repeated five times. Size-exclusion chromatography was performed using the respective buffer for each sample, containing 100 mM NaCl.
For analytical ultracentrifugation experiments, JAM-A D1 was subjected to size-exclusion chromatography using a Superdex 75 column in 20 mM Tris [pH 7.5], 100 mM NaCl. Sedimentation velocity and equilibrium experiments were performed at 25°C using a BeckmanCoulter (Krefeld, Germany) Xl-I analytical ultracentrifuge equipped with interference optics. The solvent density and partial specific volume of JAM-A D1 were calculated from composition using known density increments. Two-sector titanium centerpieces of 12 mm or 20 mm optical pathlengths (Nanolytics, Germany) were employed. A factor of 3.29 mg/mL/fringes was used to convert signal units into molar quantities. For sedimentation velocity experiments, 400 µL of protein solution at five concentrations between 0.06 and 1.31 mg/mL were centrifuged at 50 krpm. The concentration profiles were scanned every two minutes until all material had sedimented. Data were evaluated using the c(s)-function implemented in SedFit, version 9.4 [61]. For sedimentation equilibrium experiments, four initial concentrations between 1.6–0.16 mg/mL were prepared, and 150 µL of these solutions were centrifuged at three different velocities (17.5/25/35 krpm). Attainment of apparent sedimentation and chemical equilibrium was verified using MATCH. Equilibrium gradients were globally analyzed using NonLin (MATCH and NonLin are available at http://www.biotech.uconn.edu/auf/?i=aufftp). Suitable models to describe the experimental data were selected based on minimized variance and visual inspection of the residuals run pattern. Different initial starting values for the floated parameters were used to confirm that the parameters were well defined by the data.
HeLa cells were propagated as described [31]. Reovirus strain rsT3D-σ1T249I (parent) was engineered using plasmid-based reverse genetics [7]. Reoviruses were purified by cesium chloride-gradient centrifugation from infected L cells [9]. Particle concentrations were determined using the conversion factor 1 AU260 = 2.1×1012 particles. Titers of virus stocks were determined by plaque assay using L cells [62]. Attenuated vaccinia virus strain rDIs-T7pol expressing T7 RNA polymerase was propagated using chick embryo fibroblasts [63].
The parental S1 gene used for these studies encodes a σ1 molecule with a threonine to isoleucine substitution at position 249, which renders σ1 resistant to proteolytic cleavage [7]. Substitution mutations were engineered in pBacT7-S1T3D T249I [7] using QuickChange site-directed mutagenesis (Stratagene). Reoviruses were recovered from plasmids as described [7]. Mutations in the S1 gene were confirmed using the OneStep RT-PCR kit (Qiagen), gene-specific primer sets, and viral dsRNA extracted from infected L cells as template. Purified PCR products were directly subjected to sequence analysis.
HeLa cells (2×105/well) were plated in 12-well plates and incubated at 37°C overnight. Cells were treated with either phosphate-buffered saline (PBS) alone (mock) or 40 mU/ml of Arthrobacter ureafasciens neuraminidase (MP Biomedicals, LLC) diluted in PBS at 37°C for 1 h prior to adsorption with reovirus at an MOI of 50 PFU/cell. Following incubation at 25°C for 1 h, cells were washed with PBS and incubated at 37°C for 18–20 h. Infected cells were processed for indirect immunofluorescence as described [31]. Images were captured at 200× magnification using a Zeiss Axiovert 200 microscope. For each experiment, three fields of view were scored. Mean values from three independent experiments were compared using the unpaired student's t test as applied using Microsoft Excel. P values of less than 0.05 were considered statistically significant.
A BIAcore CM5 chip (GE Healthcare) was coated with mouse ascites containing monoclonal GST-specific antibody (Sigma) to ∼1800 resonance units by amine coupling. Purified GST or GST-JAM-A ectodomain fusion proteins at a concentration of 2 µM in HEPES-buffered saline [pH 7.0] were captured by injection across individual flow cells of an antibody-coated chip for 2.5 minutes at 20 µL/min using a BIAcore 2000 (GE Healthcare). Purified parent or mutant reovirus (6×1012, 8×1012, and 1013 particles/mL) was injected across the conjugated chip surface at 20 µL/min. Following reovirus binding, chip surfaces were regenerated with a 20 µL pulse of 10 mM glycine [pH 2.5]. Data analysis was performed using BIAevaluation 3.0 software (GE Healthcare).
|
10.1371/journal.pcbi.1003976 | Structure, Dynamics and Implied Gating Mechanism of a Human Cyclic Nucleotide-Gated Channel | Cyclic nucleotide-gated (CNG) ion channels are nonselective cation channels, essential for visual and olfactory sensory transduction. Although the channels include voltage-sensor domains (VSDs), their conductance is thought to be independent of the membrane potential, and their gating regulated by cytosolic cyclic nucleotide–binding domains. Mutations in these channels result in severe, degenerative retinal diseases, which remain untreatable. The lack of structural information on CNG channels has prevented mechanistic understanding of disease-causing mutations, precluded structure-based drug design, and hampered in silico investigation of the gating mechanism. To address this, we built a 3D model of the cone tetrameric CNG channel, based on homology to two distinct templates with known structures: the transmembrane (TM) domain of a bacterial channel, and the cyclic nucleotide-binding domain of the mouse HCN2 channel. Since the TM-domain template had low sequence-similarity to the TM domains of the CNG channels, and to reconcile conflicts between the two templates, we developed a novel, hybrid approach, combining homology modeling with evolutionary coupling constraints. Next, we used elastic network analysis of the model structure to investigate global motions of the channel and to elucidate its gating mechanism. We found the following: (i) In the main mode of motion, the TM and cytosolic domains counter-rotated around the membrane normal. We related this motion to gating, a proposition that is supported by previous experimental data, and by comparison to the known gating mechanism of the bacterial KirBac channel. (ii) The VSDs could facilitate gating (supplementing the pore gate), explaining their presence in such ‘voltage-insensitive’ channels. (iii) Our elastic network model analysis of the CNGA3 channel supports a modular model of allosteric gating, according to which protein domains are quasi-independent: they can move independently, but are coupled to each other allosterically.
| Cyclic nucleotide-gated (CNG) channels mediate the passage of cations through the cytoplasmic membrane. They are involved in sensory transduction and cellular development in the rod and cone photoreceptors, as well as in brain, kidney, heart and other cells, and are linked to achromatopsia and other rare genetic diseases. We used a hybrid modeling approach, combining comparative modeling and estimates of evolutionary conservation and couplings, to model the structure of a human cone CNG channel. The channel comprises a membrane domain that allows ion passage, and a regulatory cytosolic domain that binds cyclic nucleotides. The structure of each domain was modeled by homology on the basis of a suitable template. Our hybrid approach allowed us to evaluate the model structure, as well as to determine the conformations of regions where the templates overlapped and presented conflicting structural evidence. We then conducted normal mode analysis to reveal global motions of the channel. We suggest that the main mode of motion, counter-rotation of the membrane and cytosolic domains around the membrane normal, is associated with channel gating. Such rotational motion induces minimal perturbation to the lipid membrane, which could explain why the motion is observed in other types of channels.
| Cyclic nucleotide-gated (CNG) ion channels are nonselective cation channels, essential for visual and olfactory sensory transduction in vertebrates [1]–[4]. Like other members of the voltage-gated-like ion channel superfamily [2], the CNG channels are composed of four (identical or similar) monomers, each containing six transmembrane (TM) helices (referred to as S1–S6) [1], [3]. The first four TM helices in each monomer (S1–S4) form a voltage-sensor domain (VSD); the last two helices (S5 and S6, connected by the P-loop) of the four subunits assemble jointly to form the central pore. In spite of the presence of the VSD, CNG channels display very little voltage-dependent activity [1]–[4]. Rather, the channel is gated by cyclic nucleotide binding to a cytosolic cyclic nucleotide–binding domain (CNBD), connected by a so-called C-linker to the C-terminus of the S6 helix [1]–[4].
In vertebrates, the six known members of the CNG channel family, classified into A and B subtypes, can coassemble in several combinations to produce functional heterotetrameric channels [1]–[6]. In cone photoreceptors, functional tetrameric CNG channels are composed of two CNGA3 and two CNGB3 subunits, with alike subunits positioned next to each other [7] (see also a recent work by Ding and colleagues suggesting a composition of three CNGA3 subunits and one CNGB3 subunit in a cone channel [8]). Binding of cyclic guanosine monophosphate (cGMP) to CNBD controls the activity of the cone channels [1], [3], [6]. The cone system confers color vision; about 70 mutations in the CNGA3 and CNGB3 have been associated with achromatopsia, characterized by color blindness, photophobia, nystagmus and impaired visual acuity [3]. However, most of these mutations have been observed only in isolated cases, and even among the more commonly-observed mutations the precise mechanisms causing diseases are unknown (Tables S1 and S2) [3], [9], thus hindering drug-discovery efforts. These challenges are further complicated by the poor understanding of the gating mechanism of the CNG channel, despite the available structural information on the membrane domain of a bacterial homolog [10] and on the regulatory cytosolic domain of both CNG and closely related hyperpolarization-activated cyclic nucleotide-gated (HCN) channels from various organisms [11]–[20].
Several different models have been proposed to describe the gating process of CNG channels (reviewed in [5], [21]–[23]). The models differ from one another in the minimum number of bound nucleotides required for activation, the cooperativity among the binding sites, and the number of open and closed states. The simplest model, known as the sequential model, suggests that three ligand molecules bind to the closed channel, and that the binding of a fourth molecule causes the transition to the open state [24]. The sequential model does not account for the experimentally confirmed opening of unliganded or partially-liganded CNG channels and is therefore insufficient [25]. The classical Monod, Wyman, and Changeux (MWC) model [26], initially aimed at describing cooperative properties of bacterial regulatory enzymes and hemoglobin, postulates that the channel can open with any number of bound ligands, i.e., 0 to 4; the higher the number of bound ligands, the more favorable the transition from the closed to the open state [5]. Although the MWC model accurately describes some features of CNG channels, it does not account for the existence of sub-conductance states, allowing only one open state [22]. A third model describes the tetramer as a dimer of dimers; each dimer acts as an MWC unit, and the monomers within the dimer display cooperativity [27]. In order for the channel to open, both dimers must be activated independently (no cooperativity between the dimers). The last model, the modular gating model, defines modules within the channel: the VSD, the pore, the C-linker and the CNBD [28]. Each module can independently switch between two possible conformations: the VSD and the C-linker can be either resting or activated; the pore can be closed or open; and the CNBD can be ligand-free (apo) or –bound (holo). Yet, the modules are coupled to each other, so that the state of each module can affect the states of other modules.
In order to elucidate the channel gating mechanism, we built 3D models of the tetrameric CNG channel in its resting state, with the CNBDs in apo- and holo-conformations. We built the homology model from two distinct templates—one corresponding to the TM domain and the other corresponding to the cytosolic domain—and evaluated the model structure on the basis of its evolutionary conservation profile [29]. However, because the modeling process was complicated by the use of two distinct templates, as well as by low sequence similarity between the queries and the template for the TM domain, we also evaluated the model structure using EVcouplings, a recently-developed methodology that identifies evolutionarily coupled residues from sequence variation data [30], [31]. As evolutionarily coupled residues often interact physically in the 3D space of a protein [32], this and similar methods (reviewed in [33], [34]) have been successfully applied to predict structures of membrane and soluble proteins, as well as to detect residues participating in conformational changes and protein-protein interactions. Here we used evolutionary couplings (i) to evaluate our model structure of the cone channel and (ii) to select the best conformations in ambiguous regions of the model and in regions in which the two templates yielded overlapping, conflicting predictions.
To analyze the channel dynamics, we used coarse-grained elastic network models [35]–[37]. In the channel's slowest, dominant mode of motion, the TM and cytosolic domains rotate around the membrane normal in opposite directions. We related this mode of motion to gating, a proposition supported by experimental evidence [38], [39], as well as by comparison to the gating mechanism of the bacterial KirBac channel. KirBac channels resemble CNG channels in the architecture of the pore domain and in the function of the cytoplasmic domain. Additionally, investigation of the next-slowest modes of motion revealed that the TM and the cytosolic domains fluctuate alternatively in each mode, with cooperativity between the mobile and immobile domains. This observation is consistent with the modular gating model of the cone channel [28].
We modeled by homology the 3D structure of the cone CNG channel using two distinct templates: the TM region structure of the bacterial channel MlotiK1, and the cytosolic domain structure of the mouse HCN2 channel (Fig. 1). We projected ConSurf [40] conservation scores of the CNGA3 and CNGB3 subunits, composing the cone channel, onto the resultant model to evaluate it (Fig. 2). The evolutionary conservation profile has previously been demonstrated to be a valuable tool for assessing the quality of model structures; the assessment approach is based on the observation that the protein core is typically conserved, whereas residues that face the surroundings, either lipids or water, are variable [29]. Our model structure was compatible with the anticipated evolutionary profile (Fig. 2). The model structure of the TM domain was also compatible with the expected hydrophobicity profile: hydrophobic residues were exposed to the lipid, and charged and polar amino acids were buried in the protein core or located in the loop regions (Figure S1). The model correlates well with the experimental data available for the closely homologous bovine channel CNGA1 (Text S1; Figures S2 and S3). Furthermore, it suggests molecular interpretations for the effects of known clinical mutations (Text S1; Tables S1 and S2; Figure S4).
We further evaluated the predicted model structure of the CNGA3 monomer and tetramer using evolutionary couplings calculated separately for the TM domain and for the cytosolic domain. To this end, we used the EVcouplings algorithm [30], [31]. For each domain we carried out the comparison using the 2L/3 residues with the greatest coupling strength, where L is the sequence length; L = 240 in the TM region and L = 197 in the cytoplasmic region; see Methods. In both domains the overlay of the calculated evolutionary couplings and the contacts derived from the model structure was remarkable (Fig. 3): among the residue pairs used in the evaluations of the TM and cytosolic domains, 75% and 92%, respectively, were in contact. A control calculation using the templates revealed similar ratios of contacting evolutionary couplings in the two domains (Figure S5). A more detailed analysis of the couplings detected between residues that were not in contact according to the model structure (or the templates) suggested that most of these are related to flexibility in the loop regions (Text S1).
Overall, our model structure was compatible with the anticipated hydrophobicity (Figure S1) and evolutionary (Fig. 2) patterns, as well as with mappings of evolutionary couplings (Fig. 3). However, we considered alternative conformations for two specific regions in the model, i.e., helix S1 of the TM domain (Figure S6) and the S6-C-linker interface of the cytosolic domain (Fig. 1). First, secondary structure prediction methods (Figures S6A and S6B), in addition to the hydrophobicity profile of the homologs in the region (Figure S6D), pointed to two main alternatives for the boundaries of helix S1: 170–186, used in the original model, vs. 174–190 (Figure S6D). Second, for the region connecting helix S6 and the C-linker, either of the two templates could have been used (i.e., either the TM domain of bacterial MlotiK1 channel or the cytosolic domain of the mouse HCN2 channel), and we based our model structure on the mammalian template. However, we considered an alternative model structure, in which the conformation of this region was based on the bacterial template (Figs. 1A and 1B). Overall, we constructed two alternative model structures of the CNG channel (in addition to the original): one with different boundaries for the S1 helix, and another one with a different conformation of the S6-C-linker interface. We correlated the overlay of the contacts in these models with the evolutionary couplings and concluded that the original model agrees with the data better than either alternative (Fig. 3, the insets).
We used coarse-grained elastic network models to investigate global motions of the cone channel. We chose this methodology because it is insensitive to the atomic details of the (imprecise) model structure and is capable of exploring large-scale motions, related to channel gating and inactivation [37], [41]. For simplicity and to facilitate more convenient representation of the data, we focused on a symmetric tetrameric channel composed of four identical CNGA3 subunits. The results obtained for the structure, modeled based on CNBD templates in their apo- and holo-states, were, in essence, identical, and thus we only describe the results obtained for the apo-state.
A complete description of the results is provided in Text S1. Briefly, the channel dynamics are dominated by three types of motion. In motion I the channel is divided into two dynamic units: the TM domain and the cytosolic domain (Fig. 4A). The two domains rotate around the membrane normal in opposite directions (Fig. 4B). In motion II the VSDs are swinging, while the pore domain, as well as the C-linkers and the CNBDs, are essentially stationary (Fig. 4D and Figure S7B). The fluctuations of the VSDs are positively correlated with those of the C-linker and the CNBD of the same subunit, and with those of the pore region of the adjacent subunit (Fig. 4C). In motion III the pairs of diagonal CNBDs alternately move towards and away from each other (Fig. 4F). The fluctuations of the VSDs are positively correlated with those of the CNBD, C-linker and pore region of the adjacent subunits (Fig. 4E).
The analysis also shows that, in essence, all the motions, except motion I, can be categorized into modes in which only the TM domain is fluctuating (as in motion II) or modes in which only the cytosolic domain fluctuates (as in motion III). Thus, each domain is mobile individually. However, there is cooperativity between the domains, which suggests that they affect each other (Figs. 4C and 4E). This observation corroborates the modular gating model of the CNG channels, which postulates that the domains (modules) can undergo conformational changes individually, but that the state of each module affects the states of other modules, i.e., quasi-independent dynamic units [28].
In order to examine the effect of the VSDs on channel dynamics, we performed elastic networks analysis of a variant of the channel in which the VSDs were removed (Figure S8). The dynamics remained, in essence, the same, aside from the motions that directly involve the VSDs.
We presented a model structure of a human CNG channel, built using a unique computational protocol that includes homology modeling, as well as evolutionary data from conservation and couplings analysis. The model correlates well with mutagenesis and clinical data. Elastic network analyses of the model-structure enable us to provide concrete suggestions concerning the gating mechanism.
Motion I of the CNGA3 channel is a rotational, iris-like opening (Figs. 4A and 4B). This motion is unique in that it is associated with the only (non-degenerate) slow mode that manifests cooperativity among all subunits and allows symmetry-preserving conformational transition. Here we compare this motion with the gating motion in KirBac channels, prokaryotic homologs of mammalian inwardly rectifying potassium channels. KirBac channels share the architecture of the pore domain with other members of the voltage-gated-like ion channel superfamily, but they lack the VSD. Similarly to CNG channels, KirBac channels feature a cytoplasmic regulatory domain [42]. Recent crystal structures of the KirBac3.1 channel in the open and closed states (Fig. 5A) revealed its gating mechanism: upon activation, the TM and cytoplasmic domains of KirBac3.1 rotate in opposite directions around the membrane normal [43], [44]. For comparison, we performed elastic network analysis of the KirBac3.1 channel in its closed state.
The slowest mode of motion of the KirBac3.1 channel displayed rotational movement of the TM and cytoplasmic domains, which resembled the shift between the open and closed states captured in the crystal structures, i.e., the conformational change that occurs upon gating (Figs. 5A and 5B) [43]. Because of the analogy between this motion and motion I of the CNG channel, we associate the latter motion with gating as well (Fig. 5). Indeed, calculations using the edge conformations of the motion and the HOLE software [45] show that the counter-clockwise rotation of the TM domain (and the joint clockwise rotation of the cytoplasmic domain) leads to pore opening (Figure S9).
The idea of a rotational motion related to the gating of CNG channels is not new [38], [39]. Based on existing experimental evidence, Flynn and colleagues have associated a clockwise rotation of the C-linkers with channel gating (reviewed in [39]). The authors also indicated that a rotational gating motion requires definition of a ‘pivot point’, the residues on the two sides of which rotate in opposite directions. They proposed that in CNG channels the pivot point is located at the top of the S6 helices. Indeed, our elastic network analysis shows clockwise rotation of the cytosolic domain, which is coupled to counterclockwise rotation of the TM domain (Fig. 4A), an observation compatible with their proposition and with the gating mechanism of the KirBac channels [43]. However, our analysis indicates that the pivot point is located at the C-termini (rather than the top) of the S6 helices (Fig. 4A); it corresponds to the hinge in motion I around residues 401–407 ().
Although CNG and KirBac channels display similar rotational motion upon gating, there is an important difference between the two channels. KirBac channels contain two activation gates: one in the center of the S6 helix (near the P-loop) and one at the C-termini of the S6 helices (Fig. 5A) [43]. In contrast, CNG channels contain only a single gate, which is located in the pore region [21], [39]. We propose that in the absence of a second gate in the CNG channels, the VSD provides an additional means of gating regulation. Our conclusion derives from a close investigation of rotational motion I of the CNGA3 channel in the presence (Fig. 5C) and absence (Fig. 5D) of the VSDs. In the presence of the VSDs (Fig. 5C), the fluctuations of the N-termini of the S6 helices are smaller in magnitude than they are in the absence of these domains (Fig. 5D). In other words, the VSD stiffens the pore region, so that the fluctuations in the gate region are limited. This could explain why KirBac channels, lacking the VSD, have an extra gate at the C-termini of the S6 helices [43], whereas CNG channels, which do contain the VSD, have only one gate.
The idea that rotational motion can facilitate gating has been proposed for other ion channels from various families, both voltage-dependent and -independent [46]–[52]. This motion appears to have minimal effect on the channel-membrane interface, and causes minimal perturbation of the lipid membrane. Thus, it is possible that many more ion channels share a similar rotational gating mechanism.
The approach used to predict the structure of the cone channel has certain drawbacks. First, the TM region was derived from the structure of a distant homolog, with ∼10% sequence identity between query and template. This may have led to inaccuracies in the model structure due to errors in the alignment of the query and template sequences, as well as structural changes along the course of evolution. However, the compatibility of the model structure with the expected evolutionary and hydrophobicity patterns, evolutionary couplings and the available experimental data is encouraging (Figs. 2, 3 and Figures S1-S3). Second, our model structure of the cone channel was derived from two distinct templates. Therefore, the interface connecting these templates in the model, i.e., the S6-C-linker region, might be inaccurate. However, we have explored alternative conformations of the region, and the evolutionary couplings agreed best with our model structure (Fig. 3B). Third, the loop regions in any model structure are expected to be imprecise [53]. Reassuringly, the exact loop conformation that was chosen had little effect on the results of our elastic network analysis. The fact that the results are insensitive to minor structural changes suggests that other channels of the CNG family might share dynamics similar to those of CNGA3. Lastly, the fact that the cytosolic domain was modeled on the basis of the crystal structure of the corresponding domain of an HCN channel could also be problematic, as the resemblance of the CNG and the HCN channels in this region has previously been called into question, owing to conflicting experimental evidence [54]. That the evolutionary couplings in the cytosolic domain of the human CNGA3 channel and mouse HCN2 channel displayed similar patterns suggests that they do share the same conformation in this region (Fig. 3B and Figure S5B).
Given the weaknesses of the presented model structure, it is quite natural to study the channel dynamics using elastic network analysis, an approach that relies on a simplified representation of the channel structure, comprising α-carbons connected by Hookean springs of identical force constant [35], [55]–[57]. That is, the network models do not depend on the identity of the amino acids and the specificity of interactions between them. Moreover, the coarse-grained network models do not depend on the atomic details of the structure and can tolerate some variations in topology [37], [41]. In addition, elastic network models are capable of capturing large-scale conformational changes, including changes that are dependent on external stimuli, such as voltage or ligand binding. This is because the intrinsic modes of motion of a protein are determined by its architecture only, and the elastic network models can predict these modes solely from the structure, regardless of the environment (water or membrane) [37]. Whereas the environment can have an impact on the magnitude of the global motions and on local interactions, it does not usually determine their direction and nature [37]. Indeed, in previous studies, slow modes of motion have been shown to describe functional conformational changes in proteins, including membrane channels and transporters [35]–[37], [41]. Finally, the strongest support for the suitability of elastic network models as an approach for detecting functionally important motions in CNG channels comes from calculations we conducted with the KirBac3.1 channel: The dominant motions we detected for the CNG channels were very similar to motions calculated for the KirBac3.1 channel, inferred to correspond to gating motions, according to the known crystal structures of the channel in the open and closed states (Fig. 5).
Evolutionary couplings have been successfully used to predict the structures of membrane and soluble proteins (reviewed in [33], [34]), but our preliminary attempt to predict the structure of the CNG channel using evolutionary couplings alone was unsuccessful. This is perhaps because of the large number of inter-subunit contacts in the unique architecture of the tetrameric CNG channel; it is difficult to discriminate between couplings associated with the intra- and inter-subunit contacts. Instead, we used homology modeling to derive our model structure, and relied on evolutionary couplings to validate the model and to distinguish between possible conformations in regions of conflicting structural evidence. This is somewhat related to the approach used in reference [58].
In some cases, evolutionary couplings may reflect protein conformational changes, as demonstrated for several TM transporters [30]. However, our evolutionary couplings map provided no clues as to the channel's conformational changes. We suggest that the reason is that, in contrast to the major conformational changes observed in transporters, conformational changes in channels, including the rotational motion described above, are minor and have little effect on residue contacts.
While this paper was in review, a homology model of the TM domain of the canine CNGA3 channel was published [59]. The model, obtained based on a chimeric Kv1.2/2.1 structural template, corresponds to an open state of the channel, while our model structure represents a closed conformation. The models are based on different templates, and they also differ from each other in the boundaries of the TM helices (Figure S6A); most differences could perhaps be attributed to the conformational changes upon channel opening/closure.
We presented a 3D model structure of the heterotetrameric human cone channel, composed of CNGA3 and CNGB3 subunits, performed elastic network model analysis of the (equivalent homotetrameric CNGA3) channel, and obtained a mechanistic view of its gating. The following are three ‘take-home messages’:
A CNG channel subunit consists of a TM domain and a cytosolic domain comprising a CNBD and a C-linker. In the absence of a high-resolution structure of an intact subunit, we modeled the cone channel on the basis of existing structures corresponding to the individual domains. CNG and HCN channels are closely related in structure [5]; thus, HCN structures can serve as templates for modeling CNGs. A variety of high-resolution structures of mammalian cytosolic domains from HCN channels are available [15]–[20]. As a template for the apo-state, we used the mouse HCN2 channel (Protein Data Bank (PDB) entry 3FFQ [18]), the only available structure of a mammal cytosolic domain in an apo-state. As a template for the holo-state, we used the mouse HCN2 channel (PDB entry 1Q3E [15]), the only available structure of the cytosolic domain in complex with cGMP. Pairwise alignments between the queries and templates were needed for the modeling. The CNBD of the mouse HCN2 channel shares sequence identities of 33% and 29% with the cytosolic domains of CNGA3 and of CNGB3, respectively. When a query and a template display sequence identity of 30% or more, a model structure can be constructed on the basis of a simple pairwise alignment [53]. Still, extraction of the pairwise alignment from a multiple-sequence alignment can improve the model accuracy [53]. We searched for the homologs of the mouse HCN2 channel in the CleanUniProt database [61] and aligned their sequences using the MUSCLE program [62]. Both subunits, CNGA3 and CNGB3, were detected as homologs, and their alignments with the sequence of the mouse HCN2 channel were extracted from the multiple sequence alignment (Figure S10).
The crystal structure of the bacterial MlotiK1 channel in its closed state (PDB entry 3BEH [10]) served as a template for modeling the TM domains of CNGA3 and CNGB3. The sequence identities between the TM domain of MlotK1 and those of CNGA3 and CNGB3 were only 10% and 12.5%, respectively. Thus, aligning both queries to the template was challenging. In cases of such low query-template similarity it is recommended to use a variety of tools in order to produce a reliable pairwise alignment [53]. We exploited the secondary structure prediction algorithm PsiPred [63], several methods for the identification of TM segments, namely MEMSAT [64] and HMMTOP [65], a methodology for profile-to-profile alignment HHpred [66], and the FFAS03 server [67] for both profile-to-profile alignment and fold-recognition (Figure S6). Additionally, we created a multiple sequence alignment of 101 homologs that included the sequences of the queries and the template. To this end, we searched for MlotK1 homologs in the SwissProt database [68] using CS-BLAST [69]; 3 iterations were performed with maximal e-value of 10-5. The collected homologs were aligned using the MUSCLE program [62]. The final alignment between the queries and template took into account the outputs of all methodologies used (Figure S6). For the most part, in spite of the high sequence diversity, there was consensus in the hydrophobicity profiles of the homologs in the TM helices (e.g., Figure S6E), excluding S1. Because of the observed deviations in the predicted boundaries of the S1 segment, as well as the diversity of the hydrophobicity profiles of the homologs in the region, we considered also an alternative location of S1 in the sequence (Figure S6). Compared with this alternative, the final model was in better agreement with the evolutionary and hydrophobicity profiles, and with the evolutionary couplings. Nevertheless, our confidence regarding the boundaries of the S1 segment is lower compared with the other TM segments.
The templates for the TM (PDB entry 3BEH) and cytosolic domains (PDB entry 1Q3E) contain overlapping regions i.e., regions corresponding to the same amino acid segment in the queries: residues 217–223 of PDB entry 3BEH and residues 443–449 of PDB entry 1Q3E, corresponding to residues 407–413 in CNGA3 and residues 449–455 in CNGB3. This region corresponds to the interface between the S6 helix in the TM domain and the C-linker, and its orientation differs greatly between the templates (Fig. 1). As bacterial cytosolic domains do not include C-linkers [11], we chose to model the region of conflict according to the template of mammalian origin, i.e., PDB entry 1Q3E, covering the cytosolic domain. The initial 3D models of the CNGA3 and CNGB3 subunits were constructed using version v9.10 of the MODELLER software [70].
Long loops, i.e., the loops connecting S1–S2, S2–S3 and S5-P-loop, were refined with the Rosetta loop modeling application [71], [72]. Rosetta created 1,000 decoys for each loop. The decoys were then evaluated using the ConQuass method [29]. ConQuass is based on the anticipation that evolutionarily conserved amino acids are buried in the protein and that variable residues are exposed to the environment, and assigns scores to the decoys based on the degree to which they adhere to this pattern. In each of the three refined loops the decoy with the best ConQuass score was chosen for the final model. Evolutionary conservation profiles were calculated as described below.
For comparison, we also constructed a model structure of the CNG channel in which the S6-C-linker interface was modeled according to the bacterial template, i.e., PDB entry 3BEH, covering the TM domain. Overall, we constructed two alternative model structures of the CNG channel (in addition to the original): one with different boundaries of the S1 helix, and another one with a different conformation of the S6-C-linker interface.
We used CS-BLAST [69] to collect homologous sequences of CNGA3 and of CNGB3 from the CleanUniProt database [61]. Redundant (>99% sequence identity) and fragmented sequences were discarded. We aligned the sequences (183 and 223 amino acids for CNGA3 and CNGB3, respectively) using the MUSCLE program [62] and then calculated the conservation profiles using the ConSurf web-server [40].
A reliable calculation of evolutionary couplings requires a particularly large collection of homologous sequences [33]. We failed to find a sufficient number of CNGA3 homologs that included both the TM and CNB domains, and conducted evolutionary couplings calculations for each domain separately. Namely, we built a multiple sequence alignment for the TM domain, i.e., residues 161–410, and another alignment for the C-linker with the CNBD, i.e., residues 410–610. To this effect, we used the JackHMMer software [73] (5 iterations) to search for similar sequences against the Uniprot database [74] using the range 10−15–10−20 of e-values. These e-values yielded the largest numbers of aligned sequences while maintaining coverage of the input sequence. The final alignments covering the TM and cytosolic domains contained 6,439 and 7,203 sequences, respectively. Although proteins from families such as voltage-gated potassium channels can be aligned to the CNG channels at a high e-value threshold, these families were excluded, as they introduced many insertions and deletions. Importantly, this means that strong evolutionary couplings deduced from alignment of the VSD domain are not a consequence of inclusion of these known voltage-gated channels. Redundant (>90% sequence identity) and fragmented sequences were discarded. Evolutionary couplings were then calculated using the EVcouplings webserver (www.evfold.org) as described previously [31], [33]; covariation information was inferred using the plmDCA algorithm (pseudolikelihood maximization for Potts models with direct coupling analysis) [75]. All columns in the alignments containing gaps of less than 80% were considered informative for the calculation. The evolutionary couplings were ranked according to their coupling strength; for each domain, we took the top 2/3 L strongest evolutionary couplings (L = sequence length; L = 240 in the TM domain and L = 197 in the cytoplasmic domain) and correlated them to the contacts in the CNGA3 model structure.
For comparison, we also calculated evolutionary couplings for the template sequences of the bacterial MlotiK1 and mouse HCN2 using the same protocol. We compared the results to the contact maps deduced from the crystal structures using distance cutoffs between 10 and 15 Å; residue pairs with α-carbons below the cutoff were considered in contact. The results were qualitatively the same for all cutoff points selected. In our analysis of the model structure, we used the results based on a distance cutoff of 12 Å, reproducing over 95% of the detected evolutionary couplings.
We analyzed the model structure of the homotetrameric CNGA3 channel using two elastic network models, namely, the Gaussian network model (GNM) and the anisotropic network model (ANM). We focused on the homotetrameric channel for simplicity. The methodology of both elastic network models has been described previously [35], [55]–[57]; here we give a short summary.
GNM calculations use a simplified representation of the protein, in which the structure is reduced to α-carbon atoms and is treated as an elastic network of nodes connected by hookean springs of uniform force constant γ. Two nodes i and j are assumed to display Gaussian fluctuations around their equilibrium positions if the distance between them is below the (commonly used) cutoff of 10 Å. The inter-node contacts are then defined by an N×N Kirchhoff matrix Γ, where N is the number of amino acids in the protein. The correlation between the fluctuations of two nodes i and j, ΔRi and ΔRj, respectively, are calculated as follows:(1)where uk and λk are, respectively, the k-th eigenvector and k-th eigenvalue of Γ, kB is the Boltzmann constant, and T is the absolute temperature; kBT/γ was taken as 1 Å2. The summation is performed over all (k = 1 to N-1) non-zero eigenvalues. Overall, Eq. 1 predicts the mean-square displacement of each residue (node) when i = j, and when i ≠ j it predicts the correlations between the fluctuations of residues i and j as a superimposition of N-1 eigenmodes. λk is proportional to the k-th mode frequency, the inverse of which gives the relative contribution of this mode to the protein's overall structural motion. The minima in the obtained fluctuation profile for a given mode suggest possible hinge points that coordinate the cooperative motions between mobile structural elements in this mode.
In contrast to isotropic GNM, ANM determines the direction of fluctuations. Here Γ is replaced by the 3N×3N Hessian matrix H, the elements of which are the second derivatives of the inter-node potential described by Eq. 1, with a (commonly used) cutoff of 15 Å. The first 6 modes are zero eigenmodes, corresponding to the rigid-body rotations and translations of the protein [37], and the correlation between ΔRi and ΔRj is decomposed into 3N-6 modes and calculated as follows:(2)where tr[H-1]ij is the trace of the ij-th submatrix [H−1]ij of H−1. The summation is performed over all (k = 1 to 3N - 6) non-zero eigenvalues. The eigenvectors allow us to identify alternative conformations sampled by the individual modes, simply by adding/subtracting the eigenvectors to/from the equilibrium position in the respective modes. Thus, being an anisotropic model, ANM provides information on the directions of the motions in 3D, while GNM is more realistic with respect to the mean-square fluctuations and the correlations between fluctuations [37].
Several studies have demonstrated that the first few slowest GNM modes are implicated in protein function [36], [37]. Therefore, we focused on the six GNM modes identified as slowest on the basis of the distribution of the eigenvalues; these modes were responsible for approximately 40% of the overall motion of the channel (Figure S11). The superimposition of the residues' mean-square displacement predicted by GNM and ANM revealed the correspondence between the two elastic network models. Thus, using ANM, we were able to determine the direction of fluctuations characterized by GNM.
|
10.1371/journal.pbio.1000454 | Aging in a Long-Lived Clonal Tree | From bacteria to multicellular animals, most organisms exhibit declines in survivorship or reproductive performance with increasing age (“senescence”) [1],[2]. Evidence for senescence in clonal plants, however, is scant [3],[4]. During asexual growth, we expect that somatic mutations, which negatively impact sexual fitness, should accumulate and contribute to senescence, especially among long-lived clonal plants [5],[6]. We tested whether older clones of Populus tremuloides (trembling aspen) from natural stands in British Columbia exhibited significantly reduced reproductive performance. Coupling molecular-based estimates of clone age with male fertility data, we observed a significant decline in the average number of viable pollen grains per catkin per ramet with increasing clone age in trembling aspen. We found that mutations reduced relative male fertility in clonal aspen populations by about 5.8×10−5 to 1.6×10−3 per year, leading to an 8% reduction in the number of viable pollen grains, on average, among the clones studied. The probability that an aspen lineage ultimately goes extinct rises as its male sexual fitness declines, suggesting that even long-lived clonal organisms are vulnerable to senescence.
| Aging has been demonstrated in many animals and even in bacteria, but there is little empirical work showing that clonal plants age. Evidence for aging in long-lived perennials is scarce because it typically requires survivorship or fecundity schedules from long-term demographic data. Given the extreme lifespan of many long-lived perennials, it is difficult to follow cohorts of individual clones to collect late-life survivorship or fertility. Our work offers a novel approach for obtaining late-life demographic data on a clonal species by using genetic data to estimate the age of individual clones. We studied plant clones in a natural population of trembling aspen, which grows clonally via lateral root suckers. By coupling estimates of each clone's age with a measure of its male reproductive performance, we show that long-lived plant clones do senesce. Although clonal plants have the capacity for continued growth and reproduction even late in life, mutations that reduce fertility can accumulate because selection on sexual fitness is absent during clonal growth, potentially explaining senescence in this species.
| Many species of animals, and even bacteria, demonstrate a decline in survivorship or reproductive performance with increasing age (“senescence”) [1],[2]. Evidence for senescence in perennial plants, however, is scant [3],[4],[7]. One feature that distinguishes plants from many animals is indeterminate growth. Indeterminate growth is particularly extreme in clonal plants, where an individual (genet/clone) can continually produce new physiological and demographic units (ramets) without undergoing sex. Senescence is thought to be a by-product of natural selection, acting most effectively early in life, when many species have the greatest reproductive value [8]. Mutations that are deleterious to late-life survival and reproduction can spread because of early-life benefits and/or because selection against them is too weak when few individuals survive to old age [9]. Because many perennial plants and especially clonal plants continue to grow throughout life, their reproductive potential can rise over time [10]. This rising reproductive potential counters the decline in natural selection that accompanies aging, allowing selection to remain effective even in late life. It is this characteristic of indeterminate growth in perennial plants that has led some to speculate that these organisms defy aging [1],[3],[10]–[12].
Indeterminate growth is, however, a double-edged sword. Although it facilitates genet growth and renewal, it also results in more mitotic cell divisions, increasing the accumulation of somatic mutations [13]. Because somatic mutations arise in the cells of the plant body (roots and/or above-ground mass), they can be passed on to subsequent ramet generations. Furthermore, because plants do not sequester their germline, these mutations can be transmitted to reproductive organs and subsequently to sexual offspring [14]. During clonal growth, somatic mutations that negatively impact sexual fitness are free to accumulate as long as they have little or no deleterious effect on clonal growth [5],[6]. This led us to hypothesize that long-lived clonal organisms might suffer senescence. Typically, senescence results from an age-related decline in the intensity of natural selection, which allows late-acting mutations to accumulate within a population of individuals. Long-lived clones, however, might senesce because somatic mutations that reduce sexual fitness accumulate within a population of ramets (i.e., within a clone). Importantly, we expect such clonal senescence to occur even when the intensity of natural selection does not decline with age, because selection on sexual fitness is absent during clonal growth. This is not true of traits like root growth, ramet production, average photosynthetic rates, hormone sensitivity, or even a clone's susceptibility to stress from the abiotic environment, all of which are likely to remain under selection within the clone. To test if senescence of sexual fitness occurs at the level of the clone, we asked whether older clones of Populus tremuloides (trembling aspen) exhibit lower reproductive performance. Specifically, we examined pollen production and viability among male clones from natural stands in British Columbia, Canada. Coupling molecular-based estimates of clone age with pollen data, we observed a significant decline in male fertility with increasing clone age.
Populus tremuloides is a dioecious tree that forms clones consisting entirely of male or of female ramets. Each ramet within a clone is capable of both sexual and asexual reproduction. Sexual reproductive maturity is reached between 10–20 y of age while asexual maturity is reached at 1 y [15]. Individual reproductive shoots produce inflorescences (catkins) that often have between 80 and 100 flowers [16]. Projections based on the size of intermountain aspen clones, formed from lateral root suckers, suggest that clones vary in size from 1.5 to 43.6 hectares and that some of the oldest clones might even be as old as one million years [17],[18]. Because size-based age estimates might be biased if local ecological conditions constrain growth, we instead estimated clone age by measuring the amount of neutral genetic diversity that had accumulated within each clone at 14 nuclear microsatellite loci (Materials and Methods) [19]. Because aspen is dioecious, we assayed the fertility of male clones by sampling whole catkins and quantifying pollen viability and pollen count (Materials and Methods). Thus, our measure of male fertility for each genet/clone was one component of male sexual fitness, the average number of viable pollen grains per ramet per catkin. Although we recognize that male sexual fitness includes other components such as pollen germination, tube growth, number of anthers, and number of flowers, as shorthand we use the terms male sexual fitness and male fertility interchangeably.
In a previous study, we measured the accumulation of neutral somatic mutations within 20 clones from Riske Creek, British Columbia, by genotyping 719 ramets at 14 microsatellite loci [19]. Because variation within a clone is expected to accumulate over time since initiation from a seed, we used genetic diversity within the kth clone (πk) to estimate the age of the clone [19].
We found substantial variation among clones for male fertility, with older clones exhibiting significantly lower numbers of viable pollen than younger clones (Figure 1a,b). While our estimates of clone age are subject to error, we infer the same pattern in the raw data (Figure S6 and Figure S7): clones exhibiting more variation at microsatellite loci produce a significantly lower number of viable pollen grains per catkin per ramet. The observed variation in male sexual fitness among clones could not be explained by factors such as date of flower collection (F2,94 = 2.243, p = 0.11, n = 96) and inbreeding level (F1,18 = 1.142, p = 0.30, R2 = 0.06, n = 19) (Text S1; Figures S1, S2, and S4). Furthermore, there was no relationship between ramet age and male sexual fitness, suggesting that ramet age plays a minor role relative to genet age with respect to senescence via the accumulation of mutations deleterious to male fertility (F1,94 = 3.801, p = 0.054, R2 = 0.04, n = 95) (Figure S3). Empirical studies suggest that the presence of fungal pathogens and insect herbivory can exert a strong influence on reproductive success; thus we investigated the relationship between male fertility and these environmental factors. To quantify the effect of the variable of interest, clone age, on male sexual fitness, we performed a multiple linear regression accounting for the environmental factors that were substantially correlated with male sexual fitness. The best model consisted of three predictors: a composite measure reflecting the mechanical damage sustained by the average ramet in the clone (fourth principal component, PCD4), a composite measure reflecting the levels of moisture (second principal component, PCE2), and clone age (F3,16 = 7.312, p = 0.0026, adjusted R2 = 0.50, Akaike Weight = 0.55) (Table 1, Figure 1c). Holding these environmental effects constant, average number of viable pollen grains per catkin per ramet again declined significantly with clone age (Figure 1, Figure S6, Text S1).
If there were a trade-off between sexual and asexual function, selection could have facilitated the observed loss in male fertility with clone age because mutations deleterious to male sexual fitness would have increased vegetative success and risen in frequency during clonal growth. To investigate this possibility, we asked whether ramets with lower fertility exhibited higher asexual fitness, measured as the rate of increase in ramet volume per year. Specifically, we divided the volume of a ramet, V (in m3, obtained from the formula for a cylinder, V = πh(d/2)2, where d is the diameter at breast height and h is ramet height) by the age of the ramet from tree-ring data. We found no correlation between the male sexual fitness of male ramets and their growth rate (Figure 2a).
The accumulation of alleles that were beneficial to asexual growth might not have affected ramet growth but could have affected the rate of clonal expansion. Alternatively, the observed correlation between male sexual fitness and clone age could have been caused by genetic variation among clones (reflecting genetic variation among the seeds that established the clones), where some clones had higher asexual fitness and were more likely to survive for long periods of time but at a cost to male fertility. In either case, we would expect a negative relationship at the clone level between male fertility and asexual fitness. There was, however, no substantial correlation between male fertility and three potential measures of clone fitness: clone area (r = −0.33, t = −1.465, df = 18, p = 0.16), clone perimeter (r = −0.36, t = −1.662, df = 18, p = 0.11), and the maximum distance between any two ramets in the clone, Dmax (Figure 2c). Furthermore, an analysis of variance indicated that much of the variation in clone fertility can be explained by clone age (or, equivalently, genetic diversity, πk) with very little attributed to the size of a clone (ANOVA: clone age: F = 11.55, p = 0.0034; clonal spread (Dmax): F = 2.972, p = 0.10). We also considered whether the accumulation of mutations reducing male fertility might be associated with an increased density of ramets within a clone. Our previous work [20] determined that a patch is largely comprised of a single clone with smaller clones near the edge, so we used estimates of the density of ramets within a 10 m×10 m square at the centre of each patch as a proxy for the density of ramets within a clone. There was, however, no significant relationship between density of ramets in a patch and male sexual fertility (r = −0.124, t = −0.5031, df = 16, p = 0.62). We caution that all of the above measures provide only rough estimates of clone fitness. To measure clone fitness accurately and to determine any trade-offs with sexual fitness would require a long-term common garden study examining the rates of clonal spread from seed. Thus, while we find no evidence that trade-offs (negative pleiotropy) explain the reduction in male fertility with clone age, we do not exclude this possibility.
Evidence that perennial plants exhibit demographic senescence is scarce because obtaining data on survivorship or fecundity from late-life perennials typically requires long-term demographic data. This proves difficult even in short-lived perennials. For example, to demonstrate aging in Plantago lanceolata, one study followed 30,000 individuals over 7 y, finding that, during times of stress, older-aged cohorts had significantly higher mortality rates relative to younger-aged cohorts [7]. Unlike many herbaceous perennials, most woody tree species are large in stature and have an extended life cycle, rendering such experiments impractical. Although our data are not without their caveats and limitations, our work offers a novel approach for obtaining late-life demographic data on a variety of clonal species by using a molecular clock to age individual clones.
We observed a significant decline in male fertility with clone age (Figure 1), causing a reduction of 8% in the average number of viable pollen grains per catkin per ramet, on average, among the clones sampled. Given the maximum age of the oldest clone was ∼10,000 y based on glacial history in this region of British Columbia, we estimate that the rate of decline in average number of viable pollen grains was 5.8×10−5 per year (95% CI: 3.8×10−6 to 1.1×10−4 based on the multiple linear regression, Figure 1c). Given a minimum age of the youngest clones of 71 y based on tree ring data, the estimated decline was 1.6×10−3 per year (95% CI: 1.03×10−4 to 3.1×10−3). Assuming a constant linear decline, it would thus take between ∼500 and 20,000 y for a clone to lose sexual function with respect to pollen quantity and quality.
A plausible explanation for the observed decline in male sexual fitness with increasing clone age is that somatic mutations that negatively impact pollen fitness accumulate over time. As is the case with meiotic mutations, somatic changes that arise during mitosis can be neutral, deleterious, or beneficial. While somatic selection among cell lineages would act to eliminate deleterious somatic mutations, those mutations that have little to no effect on clonal growth but that reduce sexual fitness are free to accumulate. As mutations affecting fitness tend to be deleterious and partially recessive, at least some somatic mutations may be largely masked in the diploid clone phase but be deleterious in the haploid pollen stage, reducing pollen fitness among older clones. Although we observed a higher number of somatic mutations at microsatellite marker loci among the clones that exhibit reduced male fertility (Figure 1, Figure S6, Figure S7), we have no reason to expect these marker loci are directly responsible for the observed declines in sexual fitness. These marker loci only confirm that somatic mutations can and do accumulate. Two previous studies on clonal ferns showed a direct link between somatic mutations and reduced fitness, using segregation patterns of deleterious mutations among gametophytes obtained from fern clones [21],[22]. In long-lived plant taxa where higher per generation mutation rates are often found ([6],[23],[24] but see [25]), post-zygotic somatic mutations may contribute substantially to the total mutation rate and genetic load [25],[26].
An alternative explanation is that somatic mutations reducing sexual function have spread within these clones because they enhance clonal fitness (negative pleiotropy), for example, due to trade-offs in resource allocation. We looked for evidence of such trade-offs at two levels: ramet and clone. We found no evidence of a relationship between male fertility and volumetric growth per year (m3/y) of a ramet (Figure 2a). Additionally, a trade-off at the level of the clone might predict that larger-sized clones (regardless of clone age) should exhibit a reduced sexual fitness when compared to smaller-sized clones, due either to the accumulation of somatic mutations that enhance clonal spread and/or to genetic variation among the seeds that established the clones. We did not, however, find any significant correlations between sexual fitness and three potential measures of clone size/fitness (Figure 2c). Nor was there any evidence that clone size was related to clone age [19]. Although we did not detect evidence for negative pleiotropy, we cannot rule out the possibility that the loss of sex in aspen was driven by the spread of beneficial mutations that improve cell- or ramet-proliferation.
A final alternative explanation for why older clones exhibit lowered reproductive performance is that heritable epigenetic changes accumulate that reduce sexual traits. It has been shown previously that allopolyploidization, a change in reproductive mode, and nutritional stresses can lead to both genetic and epigenetic re-patterning [27]. Furthermore, there is growing evidence that epigenetic mechanisms like DNA methylation and siRNAs are responsible for natural population variation in traits like flower symmetry, self fertility, flower initiation, and number of reproductive organs [27]. Although epigenetic mechanisms like paramutation may be highly stable [28], it is unknown if such heritable epigenetic changes could persist over hundreds to thousands of years and over multiple ramet generations.
With current advances in sequencing technologies, it will become increasingly cost-effective to assess the age of clones using a molecular clock and to ask whether sexual fitness declines with clone age as we have found in trembling aspen. Furthermore, given that previous work has shown an increased transmission of deleterious mutations through the sperm than egg [29], it would be interesting to assess whether female versus male clones differ in the amount of senescence that they exhibit.
In long-lived perennials and clonal plants, substantial numbers of somatic mutations can accumulate over time [6],[13],[23],[24],[30]. This is because in plants there is no distinction between the soma and germline. Somatic cell lineages are not protected in a quiescent non-replicative state and can actively divide, eventually giving rise to gametes whenever reproductive tissues form. Although somatic mutations need not be immediately life-threatening, they can have a devastating impact on sexual function when they are unmasked in the haploid state. This suggests that, in the face of accumulating somatic mutations, plant clones may have a limited time span within which sexual function remains high. The aspen clones that we examined have lost, on average, 8% of their fertility, with less than a quarter of the pollen fertility remaining in the oldest clone (Figure 1). Without sex, clones of Populus tremuloides are unable to disperse beyond their immediate local environment. Our data provide evidence that male fertility declines with advancing age, demonstrating that aging is inevitable in aspen clones.
We collected foliage for microsatellite analysis from 871 trees of Populus tremuloides sampled from two populations in Canada: Riske Creek, British Columbia (Nclones = 20, Nramets = 719), and Red Rock, Waterton Lakes National Park, Alberta (Nclones = 29, Nramets = 152). Trees on the perimeter and along transects were physically mapped using both a measuring tape and a handheld Global Positioning System (GPS) unit. Details on the Red Rock population are not included here because this mountainous population was comprised of very small clones. The foliage from trees/ramets were sampled in two ways: on the perimeter of a patch and systematically along two or three 30–50 m transects within the patch. On average 30–50 individuals were sampled per patch. No tree less than 1.5 m in height was sampled, and only patches separated by at least 1 km of terrain lacking aspen trees were used. We physically mapped ramets, measured height and diameter at breast height on all ramets, and took an increment core from a sample of ramets belonging to each genotype.
Estimates of clone age in years are detailed in Ally et al. (2008) [19]. In short, if neutral mutations accumulate in a clock-like manner at such loci as microsatellites, then coupling the amount of molecular diversity within a clone (πk) with a mutation rate (μ) can provide a measure of a clone's age. We examined 14 microsatellite loci for somatic mutations across 719 ramets in 20 clones. We scored an allele as a somatic mutation if an individual ramet in a clone differed by one allele at one locus but shared the same alleles at all other loci as the most frequent genotype. Somatic mutations were counted only if we were able to confirm their presence with two subsequent PCRs on the same ramets. Because we found that mutations accumulated within a clone in a manner consistent with a star-like phylogeny [19], the probability that a mutation had accumulated at a locus in either of two ramet lineages is expected to equal 4 μTCCA. Here, TCCA represents the clone age or the time to the common cellular ancestor, the seed, and 2 μ is the mutation rate per diploid ramet per locus per year [31]. Clone age can thus be estimated from the pairwise genetic distance, πk, averaged over all pairs of ramets and all loci within the kth clone. This assumes that the ramets accumulate somatic mutations according to a star-like phylogeny, i.e., independently. In our study, sampled ramets were well spaced from one another, with an average distance between any two ramets of 38 m (s.e. = 3.31 m). Although somatic mutations were occasionally shared by neighboring ramets, this affected only a small number of pairwise comparisons within a clone. Furthermore, we have shown theoretically that the relationship between TCCA and πk is robust to small departures from a star-like coalescent history, allowing for the possibility that some ramets are more closely related [19].
We thus estimated the time since initiation of the clone, TCCA, as πk/(4 μ). Rather than estimating the mutation rate directly, we obtained upper and lower bounds based on the minimum and maximum possible ages of the clones. To estimate the youngest age any clone could be, we used tree ring data, reasoning that a clone had to be at least as old as the oldest cored ramet. This provided an upper bound on the mutation rate per year, μupper, by setting the average value of πk across clones to 4μ times the average age of the oldest cored ramets ( = 71 y), yielding . Here, we used all clones except the most divergent clone (which was likely much older than the oldest cored ramet). Given the neutral genetic diversity within the kth clone (πk), this upper bound on the mutation rate was used to estimate the absolute minimum age of each clone, . Similarly, the oldest that a clone could be is 10,000 y old. According to the glacial history of British Columbia, this is when the ice sheets retreated from the study area [32]. We thus obtained a lower bound estimate for the mutation rate, μlower, by setting the age of the clone with the most diversity (πmax = 0.0335) to 10,000 y and using to solve for μlower. Upper and lower bound estimates of the microsatellite mutation rate were thus μupper = 2.3×10−5 and μlower = 8.4×10−7 [19].
In Populus tremuloides, clones are either male or female; we chose to focus only on male clones, whose fitness components were more readily measured. In contrast, many plant evolutionary studies use monoecious plants with male and female organs on the same individual [7],[33]–[36]. In such cases, it is possible to measure all aspects of sexual function each generation, including pollen and ovule fitness. Aspen catkins produce between 50 and 100 flowers per inflorescence, with each male flower containing approximately 7–11 anthers [37]. Recognizing the limits of field-based measures, we thus treat the average number of viable pollen grains per catkin per ramet as a proxy for the potential of each male clone to produce further sexual offspring.
In the spring of 2003, we collected 5 whole catkins from each individual ramet, sampling 4–6 ramets per clone in Riske Creek (Nramet = 97, Nclone = 20). Every attempt was made to ensure that the catkins had flowers that were fully open and that functional anthers were in the two-lobed condition, indicative of a staminate flower just prior to the shedding of pollen [38],[39]. To determine if, at the time of collection, the degree of flower/catkin development affected our estimates of male fertility, we noted the state of the catkin (see Figure S2 for state descriptions). Attempts were made to collect replicate catkins from different parts of the ramet crown. Given time constraints and the small size of individual flowers, we did not separate out anthers and suspend them in a mixture of lactophenol-aniline blue as is typical of pollen viability studies. Instead, whole catkins were put immediately into a tube containing lactophenol-aniline blue [40], and a pestle was used to mechanically free the pollen grains from the anthers. If during the mechanical mixing of anthers (from a single catkin) not all anthers were physically opened, then we are likely to have detected fewer pollen grains per catkin. This is, however, a systematic sampling error that should be present across all samples. All sample tubes were assigned a randomly generated code. These were then brought back to the lab where pollen counts and estimates of the proportion of viable pollen were assayed by two “blind observers.” A pollen grain that was unstained, collapsed, and abnormally shaped was considered non-viable. Pollen count for each ramet was estimated from a sample using a Neubauer hemocytometer and a microscope (4× objective). Estimates of the proportion of viable pollen grains were made on a standard microscope slide (at 40×), making three sweeps lengthwise along the slide and counting both viable and non-viable pollen. The average number of pollen grains counted on a slide was 1,756.
Thus, mean male fertility was a composite measure that included pollen viability and pollen count. We acknowledge that our composite measure only captures some components of male sexual fitness. In controlled breeding trials where dormant floral branches are collected and then forced to flower in greenhouses, it may be relatively easy to measure additional fitness components like pollen tube growth and pollen germination. This was not possible in our field-based study.
Empirical studies suggest that plant sexual and asexual reproductive success is affected by the presence of fungal pathogens and insect herbivory [41]–[45]. Thus, for all sampled trees, we measured 11 morphological variables that have been shown previously to reflect disease status for P. tremuloides trees [46],[47]: diameter at breast height (cm), height (m), number of conks, number of cavities, percentage of dead branches in crown, presence/absence of sap, number of scars, average length of scar (cm), proportion of leaves scored as eaten, proportion of leaves with a gall, and proportion of leaves exhibiting a leaf minor.
A second aspect of the environment affecting plant reproductive success is site quality, which reflects available resources like soil moisture, nutrients, drainage level, light, and soil temperature. Environmental variables like moisture vary through time, and thus accurate and detailed assessments of site quality can be time consuming, expensive, and difficult to obtain. As a proxy for site quality, plant assemblages are often used because indicator plant species reflect differential resource availability and the form of competitive interactions. Previous work on aspen-dominated communities has indeed shown that understory vegetation was significantly correlated to site quality [48]. Thus, to obtain an assay of site quality, we measured the percent cover of trees, shrubs, herbaceous plants, moss, lichens, and bryophytes in the centre of each patch using a sampling plot with a radius of 5 m. Surveys were conducted during the months of May and July 2003. Where possible, all non-woody herbs and shrubs were identified to species level. In a few cases where habitats were similar and more than one species of a genus was found, we collapsed species into genus-level groups to reduce the number of variables in our analyses. In addition, we dug soil pits in each of the sampled patches, and from a combination of topographic and soil morphological properties, we obtained data on soil moisture regime, soil nutrient regime, and drainage class [48]. These soil characteristics were recoded into binary data for the patch.
We performed two separate principal component analyses (PCA) because our variables reflect different physical scales as well as different aspects of ecology. Specifically, our morphological variables were measured for every sampled tree and indicate susceptibility to disease and health of the individual ramet, while site variables reflect the environment found within a clone. Eleven different ramet health variables were reduced to four composite measures, while plant understory cover and abiotic site variables were reduced to eight environmental indices using principal component analysis (Supporting Information).
To assess the relevance of these abiotic and biotic variables to mean sexual fitness, we first examined the data using Pearson product moment correlations (r) and scatterplots. No correction was made for multiple comparisons because we were simply identifying potential predictors. From this we chose only those predictors that showed a sizable correlation with sexual fitness (r>|0.30|). The predictors that showed a correlation greater than |0.30| were PCD4: mechanical damage (r = −0.43), PCE2: moisture (r = 0.43), and the predictor of interest, clone age (r = −0.56). We calculated the Akaike Information Criterion (AIC) and then obtained the Akaike Weight (W) to determine the relative probability that a given model best fit the data (Table 1) [49]. These variables were then put into a stepwise multiple regression analysis. Subsequent model selection was based on AIC criterion, p values, and adjusted R2 values.
|
10.1371/journal.pgen.1003367 | A Novel Role for the RNA–Binding Protein FXR1P in Myoblasts Cell-Cycle Progression by Modulating p21/Cdkn1a/Cip1/Waf1 mRNA Stability | The Fragile X-Related 1 gene (FXR1) is a paralog of the Fragile X Mental Retardation 1 gene (FMR1), whose absence causes the Fragile X syndrome, the most common form of inherited intellectual disability. FXR1P plays an important role in normal muscle development, and its absence causes muscular abnormalities in mice, frog, and zebrafish. Seven alternatively spliced FXR1 transcripts have been identified and two of them are skeletal muscle-specific. A reduction of these isoforms is found in myoblasts from Facio-Scapulo Humeral Dystrophy (FSHD) patients. FXR1P is an RNA–binding protein involved in translational control; however, so far, no mRNA target of FXR1P has been linked to the drastic muscular phenotypes caused by its absence. In this study, gene expression profiling of C2C12 myoblasts reveals that transcripts involved in cell cycle and muscular development pathways are modulated by Fxr1-depletion. We observed an increase of p21—a regulator of cell-cycle progression—in Fxr1-knocked-down mouse C2C12 and FSHD human myoblasts. Rescue of this molecular phenotype is possible by re-expressing human FXR1P in Fxr1-depleted C2C12 cells. FXR1P muscle-specific isoforms bind p21 mRNA via direct interaction with a conserved G-quadruplex located in its 3′ untranslated region. The FXR1P/G-quadruplex complex reduces the half-life of p21 mRNA. In the absence of FXR1P, the upregulation of p21 mRNA determines the elevated level of its protein product that affects cell-cycle progression inducing a premature cell-cycle exit and generating a pool of cells blocked at G0. Our study describes a novel role of FXR1P that has crucial implications for the understanding of its role during myogenesis and muscle development, since we show here that in its absence a reduced number of myoblasts will be available for muscle formation/regeneration, shedding new light into the pathophysiology of FSHD.
| Muscle development is a complex process controlled by the timely expression of genes encoding crucial regulators of the muscle cell precursors called myoblasts. We know from previous studies that inactivation of the Fragile X related 1 (FXR1) gene in various animal models (mouse, frog, and zebrafish) causes muscular and cardiac abnormalities. Also, FXR1P is reduced in a human myopathy called Fascio-Scapulo Humeral Dystrophy (FSHD), suggesting its critical role in muscle that findings presented in this study contribute to elucidating. Cell-cycle arrest is a prerequisite to differentiation of myoblasts into mature myotubes, which will form the muscle. One key regulator is the p21/Cdkn1a/Cip1/Waf1 protein, which commands myoblasts to stop proliferating, and this action is particularly important during muscle regeneration. In this study, we have identified FXR1P as a novel regulator of p21 expression. We show that FXR1P absence in mouse myoblasts and FSHD-derived myopathic myoblasts increases abnormally the levels of p21, causing a premature cell cycle exit of myoblasts. Our study predicts that FXR1P absence leads to a reduced number of myoblasts available for muscle formation and regeneration. This explains the drastic effects of FXR1 inactivation on muscle and brings a better understanding of the molecular/cellular bases of FSHD.
| The Fragile X-Related 1 (FXR1) gene belongs to a small gene family that includes the Fragile X Mental Retardation 1 (FMR1) and Fragile X-Related 2 (FXR2) genes (reviewed in [1]). Human FMR1 is located on chromosome Xq27.3 [2] and inactivation of FMR1 expression leads to the Fragile X syndrome in human, the first cause of inherited mental retardation [5]. FXR1 and FXR2 are autosomal genes, respectively mapping at 3q28 and 17p13.1 [3], [4]. The FXR1 gene is highly expressed in muscle and its pre-mRNA is known to undergo extensive alternative splicing, which generates distinct FXR1 mRNA variants that produce FXR1P isoforms with divergent C-terminal regions [6], [7]. Four isoforms ranging from 70 to 80 KDa (Isoa, Isob, Isoc, Isod) are ubiquitously expressed, including in murine [7], [8] and human myoblasts [9]. Myoblasts also express long muscle-specific FXR1 mRNA variants, termed Isoe and Isof, which are massively induced upon muscular differentiation [7], [8], [9], [10]. Importantly, these muscle-specific mRNA variants of FXR1 are the only expressed in adult muscle [6], [7], [8], [9], [11]. Defects in FXR1 gene muscular pattern of expression have been observed in patients affected by Facio-Scapulo Humeral Distrophy (FSHD), the most prevalent muscular dystrophy affecting adults and children [9]. Similar defects were observed in a mouse model of myotonic dystrophy (DM1, [12]). As a result, the long isoforms FXR1P Isoe and Isof of 82–84 kDa are depleted in myopathic muscle. Consistent with these altered expression pattern of FXR1 in myopathic patients, Fxr1-knockout mouse die shortly after birth most likely due to an abnormal development of cardiac and respiratory muscles [13]. A mouse model with reduced levels of Fxr1 expression has also been generated, and displays reduced limb musculature and a shorter life span of about 18 weeks [13]. Moreover, during Xenopus embryogenesis, complete or partial inactivation of xFxr1 disrupts somitic myotomal cell rotation and segmentation, impeding normal myogenesis [14]. Finally, depletion of zFxr1p during early development of the zebrafish leads to cardiomyopathy and muscular distrophy [15]. All these data point out an evolutionarily conserved role for FXR1P in myogenesis.
FXR1P contains two KH domains and one RGG box that are characteristic motifs in RNA-binding proteins [4], [16]. In addition, FXR1P harbours nuclear localization and export signals (NLS and NES) enabling nucleocytoplasmic shuttling [4], [17]. In most cell types and tissues studied, FXR1P isoforms are associated to messenger ribonucleoparticles (mRNPs) present on polyribosomes, suggesting a consensus role in translation regulation for FXR1P [18]. However, it was reported that, in undifferentiated myoblasts, FXR1P long isoforms Isoe and Isof are not detected on polyribosomes, suggesting a role other than translation regulation for these isoforms at this stage [7], [8]. Very few specific target mRNAs for FXR1P have been identified so far, and even more scarcely in the context of myogenesis. First, two independent studies reported that the shortest isoform of FXR1P, Isoa, binds the AU-rich element (ARE) present in the 3′UTR of proinflammatory cytokine tumor necrosis factor (TNFα) mRNA [19], [20]. In this context, FXR1P associates with AGO2 on TNFα−ARE to modulate its translation [20]. Second, we have previously shown the ability of FXR1P Isoe, its long muscle-specific isoform, to interact specifically and with high affinity with the G-quadruplex RNA structure in vitro [21]. However, no mRNA target of FXR1P bearing a G-quadruplex has been identified yet in vivo. Finally, one study reports the presence of Desmoplakin and Talin2 mRNAs in FXR1P-mRNP complexes and subsequent disturbance of the expression of the encoded proteins in Fxr1-KO heart extracts [22]. However, neither the binding motif/sequence recognized by FXR1P on these mRNAs nor the exact functional significance of these interactions have been explored.
To gain further insights into the muscular roles of FXR1P and the pathways perturbed in its absence, we performed a large-scale microarray analysis of the C2C12 myoblastic cell line inactivated for Fxr1. This analysis revealed that Fxr1-depletion lead to premature cell cycle exit of myoblasts. We link this to a robust increase in the levels of the cyclin-dependant inhibitor p21/Cdkn1a/Cip1/Waf1, that is also observed in FSHD-derived myoblasts. In this study, we further explore the role played by the direct interaction of FXR1P with p21 mRNA in the post-transcriptional control of p21 levels.
To understand the functional role of FXR1P in myoblasts, we used as a cellular model the C2C12 myoblastic cell line. This murine cell line enables to reproduce myogenesis in vitro [23] and expresses all the myogenic factors as well as FXR1P [7], [8]. In this model, we inactivated the expression of all FXR1P isoforms by transient transfection of siRNAs targeting exon 14, a constitutive exon present in all Fxr1 mRNAs [6]. As shown in Figure 1A, quantitative RT-PCR performed on C2C12 cells transfected with siFxr1 siRNAs reveals a significant reduction in Fxr1 mRNA as compared to siControl-transfected cells (13.45%±3.4% residual expression, Figure 1A). Knockdown of all isoforms of FXR1P was obtained by siFxr1 transfection, as shown by western-blot analysis using the 3FX antibody (Figure 1B, [8]). Note that the levels of FXR2P, the close homologue of FXR1P, also recognized by 3FX antibody, remain unaffected, confirming the specificity of the knockdown strategy (asterisk, Figure 1B). In siFxr1-transfected myoblasts, the decrease in epifluorescence signal after FXR1P-immunolabeling as compared to siControl-transfected cells confirms the efficiency of the knockdown (Figure 1C). The knockdown appears to homogenously affect all the cells since the signal is uniformly decreased. Note that in C2C12 cells, FXR1P immunoreactivity is mainly cytoplasmic, however, signal is also detected in the nucleus (Figure 1C). Indeed, we confirmed the partial nuclear localization of FXR1P in myoblasts by confocal microscopy (Figure 1D), as described previously for the long isoforms of FXR1P in C2C12 myoblasts [7] and in human myoblasts [9].
To determine the impact of the inactivation of Fxr1 on gene expression in myoblasts, total RNA was extracted from siControl and siFxr1-transfected C2C12 myoblasts and simultaneously analysed using whole genome mouse microarrays. Among the genes showing measurable differential levels of expression, a significant change was observed for 105 transcripts (32 down- and 73 up-regulated) of which 79 were annotated in the RefSeq database (Figure 1E and Table S1). As expected, Fxr1 mRNA appears among the most significantly down-regulated in siFxr1-transfected cells (Figure 1E and Table S1). To confirm the observed dysregulation of a subset of mRNAs in Fxr1-knockdown C2C12 myoblasts, we performed quantitative RT-PCR analysis (Figure 1F). Interestingly, in Fxr1-depleted myoblasts, we were able to confirm by quantitative RT-PCR a significant upregulation of mRNAs encoding: Semaphorin 7a (Sema7a), the Ca2+-binding multiple C2 domains transmembrane protein 2 (Mctp2), asialoglycoprotein receptor 1 (Asgr1), the cyclin-dependant kinase inhibitor p21 (p21/Cdkn1a/Waf1/Cip1), Hepatocyte growth factor (Hgf), Dual specific phosphatase (Dusp6) and finally Limb-bud and heart protein (Lbh, Figure 1E). Conversely, we confirmed a significant down-regulation of Cdk15 mRNA encoding the cyclin-dependent kinase 15. Finally, the mRNAs encoding the myoregulatory factors MyoD and Myogenin for which no mRNA variations were detected by microarray analysis remained unaffected (Figure 1F). These analyses were further repeated on C2C12 cells inactivated for Fxr1 by transfection of a different siRNA (siFxr1#2) targeting Fxr1 exon 6, another constitutive exon of Fxr1 present in all its variants [6]. This second siRNA leads to a 37% residual expression of Fxr1 mRNA (Figure S1A) and reduces all FXR1P isoforms (Figure S1B) as compared to siControl. In addition, siFxr1#2-mediated knockdown of Fxr1 efficiently modulated the previously studied subset of mRNAs to induce variations similar to the one observed with the first siRNA against Fxr1 (Figure S1C). Importantly, this cross-analysis using two siRNAs targeting distinct regions of Fxr1 mRNA exclude the fact that the observed variations could derive from off-target effects of the siRNAs.
To gain insights into the pathways perturbed by Fxr1 depletion, we performed an analysis of the biological functions or processes selectively enriched among the altered transcripts, using the Ingenuity Pathway Analysis (IPA) software (Table S2). Interestingly, Fxr1 knockdown in C2C12 myoblasts significantly affected the functional categories ‘cell cycle’ (Table S2), ‘skeletal and muscular system development and function’ and ‘skeletal and muscular disorders’ (Table S2). Importantly enough, a subset of mRNAs perturbed in siFxr1-knockdown myoblasts compared to control repeatedly appeared determinant for the definition of the affected functional categories: the cyclin-dependent kinase (Cdk15), the cyclin-dependent kinase inhibitor (p21/Cdkn1a/Cip1/Waf1) and the Hepatocyte growth factor (Hgf).
One of the most recurrent terms in IPA analysis of dysregulated mRNA upon Fxr1 depletion were ‘cell cycle progression’, ‘arrest in G0/G1’, ‘proliferation’ and also ‘cell viability’ (Table S2). This prompted us to analyse myoblasts' viability and proliferation abilities upon Fxr1-depletion. Fluorescence-Activated Cell Sorting (FACS) analysis of the DNA intercalant Propidium Iodide (PI) incorporation on living cells allowed us to detect no changes in the overall viability of Fxr1-knockdown (92.5% viability) compared to control (90.53% viability) C2C12 cells (Figure 2A). To assess the proliferation ability of Fxr1-depleted myoblasts, we conducted tetrazole MTT proliferation assays. Interestingly, after 48 hours in culture, siFxr1-transfected C2C12 cells exhibit a significant 15% decrease in MTT reductase activity as compared to control (Figure 2B). This suggests that Fxr1 depletion may induce alterations of myoblasts cell cycle. We therefore further analysed the distribution in the various cell cycle phases of siFxr1- or siControl transfected myoblasts. The DNA content of the cells was assessed by FACS-measurement of the amount of PI incorporated in cells. Surprisingly, in a normal asynchronous cell population, we did not observe any significant change in the cell cycle phases distribution of the C2C12 cells transfected with siFxr1 or siControl, in normal growth conditions (Figure 2C).
To highlight specific defects in cell cycle, we synchronized siFxr1- and siControl-transfected myoblasts by treatment with the cell cycle blocker mimosine, that arrests cell cycle progression at the G1/S phase border [24]. Since the effects of this cell cycle blocker are fully reversible, we then allowed the synchronized cells to reenter cell cycle by incubating them in normal growth medium for 16 hrs before FACS analysis. In these conditions, we did observe a significant 27.6% increase in the number of cells in the G0/G1 phase in Fxr1-knockdown myoblasts, as compared to control. This increase in the G0/G1 population is accompanied by a 51.9% decrease in the number of cells in the G2/M phase. Importantly, no differences were observed in the proportion of cells in the Sub-G1 phase - corresponding to cellular debris with a lower DNA content liberated by apoptotic cells [25]- in asynchronous cells (Figure 2A) and after release from cell cycle blocker (Figure 2D). These data indicate that FXR1P depletion in myoblasts does not lead to cell viability defects but rather causes a blockade and accumulation of cells in the G0/G1 phase to the detriment of mitosis.
Thus, to determine whether the cells were blocked in G0 or G1, we performed immunolabeling of C2C12 cells in normal growth conditions and quantified the number of DAPI-positive nuclei and the amount of cells positive for the proliferation marker Ki67 (Figure 3). We observed that the number of nuclei in cultures of siFxr1-transfected myoblasts is decreased by 26%, suggesting that Fxr1 depletion limits the proliferating abilities of myoblasts (Figure 3B). Quantification of cells expressing Ki67 enabled us to detect that siRNA-meditated depletion in Fxr1 leads to a subtle, but significant 10% decrease in the number of Ki67-positive cells compared to control (Figure 3C). Since Ki67 is expressed during all active phases of the cell cycle (G1, S, G2, and mitosis), but absent from quiescent cells (G0) [26], the unlabeled cells most likely represent resting cells blocked in G0.
The premature cell cycle arrest we observed in Fxr1-depleted myoblasts prompted us to examine the subset of deregulated mRNAs identified by microarray analysis in order to identify candidates for regulation by FXR1P that could contribute to explain this phenotype. The most promising mRNA candidate appeared to encode the ubiquitous cyclin-dependent kinase inhibitor (CDKI) p21 –also known as Cdkn1a/Cip1/Waf1- that belongs to the Cip/Kip family of CDKI. In myoblasts, p21 is known to block cell cycle progression to trigger cell-cycle exit, a prerequisite to muscular differentiation [27], [28], [29].
In Fxr1-depleted myoblasts, we found that p21 mRNA level is significantly increased by microarray analysis (Figure 1E, Table S1) and confirmed a 1.76-fold upregulation of the transcript by quantitative-RT PCR in these Fxr1 loss-of-function experiments (cf Figure 1F). This upregulation of p21 mRNA level in Fxr1-depleted myobasts was further confirmed using a second siRNA targeting Fxr1 (Figure S1). We had previously shown that the muscle-specific long isoforms of FXR1P, notably Isoe, are depleted in myoblasts derived from Fascio-ScapuloHumeral Distrophy (FSHD) patients and had hypothesized that this could induce deregulation of mRNA targets specific to this isoform FXR1P Isoe [9]. To test this hypothesis on this new potential mRNA target of FXR1P, we assessed the status of human P21 in the same samples used in our previous study. Interestingly enough, P21 mRNA levels are significantly increased in FSHD patients by a 1.8 factor (Figure 4A).
We then sought to verify whether this increase in p21 at the mRNA level was translated at the protein level by western-blotting (WB) analysis. Quantification of WB of siFxr1-transfected C2C12 using the ImageJ software revealed a 1.92 fold increase in p21 protein levels (Figure 4B). Concomitantly, we observed by western-blotting that the levels of P21 protein are increased in FSHD myoblasts compared to control by a 1.66 factor (Figure 4C). These data indicate that depletion of FXR1P and particularly of its long muscle-specific isoforms increases p21 mRNA and correlatively increase the levels of p21 protein both in murine and human myoblasts.
To assess the specificity and the direct nature of the effects we observed on p21 mRNA levels by FXR1P loss of function experiments, we first used a gain-of-function approach. For these experiments, we used FXR1P long isoform Isoe since its depletion in FSHD myoblasts recapitulates the effects on p21 mRNA levels of a knockdown of all FXR1P isoforms in C2C12 cells (cf Figure 4). Interestingly, in contrast to Fxr1 loss-of-function in C2C12 myoblasts, over expression of FXR1P Isoe lead to a 19,1% significant decrease in endogenous p21 mRNA levels as compared to transfection with empty vector (Figure 5A). This ascertains the fact that the effects we observe on p21 mRNA levels are directly related to the levels of FXR1P present in the cell. Secondly, we performed rescue experiments using a pTL1 plasmid bearing FXR1 Isoe cDNA in which we generated by site-directed mutagenesis 4 mismatches to avoid recognition of the transgene by siFxr1 (Figure 5B). This strategy enabled to efficiently re-express FXR1P Isoe in Fxr1-knocked down myoblasts (Figure 5C). Rescue of the expression of FXR1P Isoe lead to a significant reduction in p21 mRNA levels as compared to unrescued myoblasts. The rescue with FXR1P Isoe is total since the levels of p21 mRNA in rescued cells are restored to control levels. Of notice, similar results were obtained using another mutant plasmid of pTL1.Isoe (data not shown), confirming the efficiency of the rescue strategy.
These data confirm the specificity of our approach and suggests that p21 mRNA may be a target of FXR1P in C2C12 murine myoblasts and in human myoblasts, either directly by RNA-protein physical interaction, or indirectly by modulating a pathway involved in p21 levels controls.
Murine p21 mRNA is 1910 nts long (GenBank Accession number: GI 161760647), with a very short 5′UTR of less than 100 nts, a 480 nts coding sequence and a 1329 nts long 3′UTR where lie most of the regulatory elements for the stability of this mRNA (Figure 6A). Notably, the ARE located at position 86–103 nts on the 3′UTR is bound by the RNA-binding protein HuR to regulate the stability of the mRNA during muscle differentiation [30]. Given the ability of FXR1P Isoa to bind ARE sequences [19], [20], we hypothesized that the ARE present in p21 mRNA could be the binding site of FXR1P.
To test the physical interaction between FXR1P and p21 mRNA and determine the portion of the mRNA involved in the interaction, we performed in vitro filter-binding assays [21] using recombinant FXR1P and radiolabeled fragments of p21 mRNA 3′UTR described in Figure 5A. We chose to use FXR1P Isoe, the longest muscle-specific isoform of FXR1P for binding experiments since i) it was described to have RNA-binding properties [21], ii) its depletion in FSHD myoblasts recapitulates the effect on p21 mRNA levels of a complete knockdown of all FXR1P isoforms in C2C12 cells (cf Figure 4) and iii) Isoe is able to restore p21 mRNA levels to normal in Fxr1-knockdown myoblasts (cf Figure 5). As controls for interaction, we used the N19 fragment of FMR1 mRNA containing a G-quadruplex RNA structure [31], known to be specifically bound by FXR1P Isoe, and its truncated version N19Δ35 unable to be bound by FXR1P [21]. As expected, FXR1P was able to recognize the G-quadruplex containing N19 fragment (Figure 6B). Surprisingly, the binding activity of FXR1P towards p21 3′UTR-α fragment (nts 1–345) that contains a well characterized ARE sequence was null, being equal to the binding activity of the negative control N19Δ35. Also, p21 3′UTR-β fragment (nts 324–868) was not recognized by FXR1P. Interestingly, the most distal portion of p21 3′UTR, termed γ fragment (nts 851–1321), was specifically bound by FXR1P. These data indicate that FXR1P Isoe does not recognize p21 mRNA via the ARE motif present in the proximal portion of the 3′UTR (α fragment), but most likely via an uncharacterized motif or sequence present in the distal portion of its 3′UTR-γ fragment.
Knowing that FXR1P interacted, at least in vitro, with p21 mRNA, we further sought to validate that this interaction occurs in vivo. To test this hypothesis, we isolated immunocomplexes containing FXR1P by performing UV-crosslinking and immunoprecipitation assays (CLIP, [32]). Immunoprecipitation of FXR1P mRNA complexes was carried out using the polyclonal antibody #830 against exon 16 of FXR1P present in all isoforms except the short ones [7], [8] on C2C12 cell extracts (Figure 6C). Control CLIP was performed using non-immune rabbit IgGs. As expected, using the #3FX monoclonal antibody [7] against the constitutive exon 14 present in all isoforms of FXR1P, all the isoforms of FXR1P were detected in both inputs (Figure 6C, lane 1 and 2). Medium and long isoforms of FXR1P were selectively enriched in #830 immunoprecipitates (Figure 6C, Lane 4) and concomitantly depleted in #830 post-immunoprecipitation supernatant (Figure 6C, lane 6). The low amount of FXR1P small isoforms detected in the #830 immunoprecipitates most likely corresponds to the fraction of small isoforms interacting with FXR1P medium and long isoforms, since FXR1P is known to homodimerize [4]. In contrast, FXR1P is not recovered in immunoprecipitates obtained with control rabbit IgGs (Figure 6C, lane 3) and still present in the corresponding post-immunoprecipitation supernatant (Figure 6C, lane 5), confirming the specificity of the CLIP assay performed with #830 antibodies.
RT-PCR analysis of mRNAs extracted from both inputs and immunoprecipitates was then carried out (Figure 6D). The mRNA encoding p21, β-tubulin and the myogenic factors Myogenin and MyoD are detected in the input fractions (Figure 6D, lanes 1 and 2). Interestingly, only p21 mRNA was found selectively enriched in #830 immunoprecipitates (Figure 6D, lane 4) as compared to control immunoprecipitates (Figure 6D, lane 3), while Myogenin, MyoD and β-tubulin mRNAs were undetectable. This confirms the specificity of the approach and suggests that, in the C2C12 myoblastic cell line, endogenous p21 mRNA is present in mRNA complexes containing FXR1P.
To elucidate the functional significance of FXR1P interaction with p21 3′UTR-γ fragment, we conducted luciferase assays on C2C12 cells expressing FXR1P normally (siControl-transfected) and inactivated for Fxr1 (siFxr1-transfected). The various portions of p21 3′UTR used for binding assays were cloned in the 3′ of Renilla luciferase cDNA in a reporter system (Figure 7A). The influence of the 3′ regulatory elements on Renilla mRNA and protein levels was then assessed, in the presence and in the absence of FXR1P, and compared to control vector without regulatory elements in the 3′UTR (Figure 7B, 7C). In the presence of FXR1P or when FXR1P is knocked-down, no significant difference to control is observed in the Renilla mRNA levels, when its cDNA is fused either to the proximal α or central β fragments of p21 mRNA 3′UTR. However, the distal γfragment bound by FXR1P significantly increases Renilla mRNA levels in the presence of FXR1P (1.33-fold, Figure 7B). Intriguingly, removal of FXR1P by siRNA-mediated knockdown potentiated the mRNA stabilizing effect of the p21 3′UTR-γ fragment (1.76-fold; Figure 7B) compared to control. To assess whether variations of Renilla mRNA correlated to protein variations, we performed classical luminescence luciferase assays (Figure 7C). Interestingly, Fxr1-depletion lead to a significant increase in Renilla luciferase activity when its cDNA was either fused to the central β or distal γ fragment of p21 3′UTR (Figure 7C). However, the amplitude of variation was, again, higher when considering the γ fragment in siControl conditions (2.2-fold) or Fxr1 knockdown conditions (3.4-fold), compared to control empty vector. These data support the hypothesis that FXR1P normally destabilizes p21 mRNA via binding to a motif present in the distal γ portion of its 3′UTR.
To test in vivo the hypothesis of FXR1P involvement in the control of endogenous p21 mRNA stability, we treated siControl- or siFxr1-transfected C2C12 cells with the transcription inhibitor actinomycin D (ActD), and measured the decay rate of p21 mRNA by quantitative RT-PCR. Interestingly, p21 mRNA appears to cycle rapidly in control myoblasts. Linear regression on semi-log values of p21 mRNA decay rate in siControl-transfected cells, provides an estimated half-life of 2.57±0.14 hrs (Figure 7D), with only 16% mRNA remaining after 8 hrs. Conversely, upon Fxr1-depletion, p21 mRNA decay rate is strongly affected and its half-life is significantly increased, reaching 5.98±0.42 hrs (p-val<0.05). As a consequence, even after 8 hrs of ActD treatment, 43% of p21 mRNA is still present (Figure 7D). The slowing down of p21 mRNA decay rate following Fxr1-knockdown was further confirmed using 5,6-Dichlorobenzimidazole riboside (DRB), an adenosine analogue inhibiting mRNA synthesis (Figure S2). These data suggest that Fxr1-depletion increases endogenous p21 mRNA stability.
The previous data support a negative role for FXR1P in the control of p21 mRNA stability via binding to the 561 nts long p21 3′UTR-γ portion. The next step was to determine the RNA motif responsible for FXR1P recognition. So far, two mRNA motifs have been described to be recognized by FXR1P: the ARE motif of TNFα mRNA [20] and the G-quadruplex present in FMR1 mRNA [21]. Our in vitro data clearly indicate that the ARE present in the 3′UTR of p21 mRNA does not mediate the binding of FXR1P Isoe to p21 mRNA, we therefore looked for the presence of putative G-quadruplex motifs in the γ fragment of p21 3′UTR. For this purpose, we used the QGRS webtool [33] that indicated three putative G-quadruplexes spread along the sequence of the γ fragment (Figure S3), and notably a high-score central G-quadruplex motif (nts 931–955). It is worth noticing that this high-score putative G-quadruplex is located within a 51 nts G-rich region (position 918–955, 54% of G). To confirm the predicted G-quadruplex, we used the property of G-quadruplex forming regions to be detected by comparing reverse transcriptase elongation on RNA templates in the presence of either K+, Li+ or Na+ [31]. Indeed, stabilization of G-quadruplex structures by K+, but not by Li+ or Na+, results in cation-dependent pauses detectable on a sequence gel. The experiments were performed on the full-length 3′UTR and on the γ fragment alone and allowed us to identify two strong (position 939 and 940) and two weak G-quadruplex pauses (position 955 and 969) in the 3′UTR of p21 mRNA (Figure 8A). Both the full-length and the γ fragment exhibited the same pauses, indicating that the γ fragment retains the ability to form the G-quadruplex structure in a comparable manner to the full-length native 3′UTR (Figure 8A). Alignment of sequences corresponding to G-rich regions of p21 distal 3′UTR in mouse and human indicate high evolutionary conservation of this portion of non-coding sequences (Figure 8B) and argues in favour of its functional importance.
To explore the functional role of the G-quadruplex present in the 3′UTR of p21 mRNA, we constructed γ fragments mutants with partial (γΔ9) or full (γΔ38) deletion of the G-rich region containing the G-quadruplex (Figure 8C) that were cloned downstream of Renilla luciferase mRNA. Then, the levels of Renilla mRNA of the resulting constructs were assessed for each mutant in C2C12 cells. As previously shown in Figure 6B, the presence of the γ fragment did increase significantly the levels of Renilla mRNA, but partial or full deletion of the G-quadruplex potentiated the increase in the cognate mRNA levels (Figure 8D), mimicking the effect of Fxr1 knockdown in C2C12 cells (cf Figure 7B). These data argue in favour of a role of the G-quadruplex in mRNA stabilization that is potentiated by deletion of the binding site of FXR1P.
Over the last decade, studies in Fxr1-knockout models have inferred that FXR1P plays a critical role in myogenesis [13], [14], [15]. However, even though FXR1P muscle-specific isoforms have unique RNA-binding properties [21], no specific mRNA targets and function have been identified so far for FXR1P in muscle. In this study, we have explored the functional consequences of the depletion of the FXR1P in myoblasts, with the purpose to understand its role in the early stage of myogenesis and in the cellular pathophysiology of FSHD, a human myopathy.
Microarray analysis of our myoblastic model inactivated for Fxr1 enabled to show that FXR1P depletion affects the expression of a wide range of mRNA species that control several cellular pathways. One of the most represented functional categories correspond to ‘skeletal and muscular system development’ and ‘skeletal and muscular disorders’, in line with the evoked role of FXR1P in myogenesis and its altered pattern of expression in two human myopathies: FSHD [9] and DM1 [12]. Interestingly, the functional category ‘cell cycle’ appears also overrepresented in the affected functions, in particular, terms corresponding to ‘arrested in G0/G1 phase transition’ (related to the genes p21/Cdkn1a, HGF, IGF, IL6) actually reflect what we observed at the physiological level for Fxr1 inactivated myoblasts which remain blocked in the G0 phase, without undergoing further differentiation. Apart from p21, several mRNAs with altered levels in the absence of FXR1P seem to influence the functional categories affected and appear iteratively in our Ingenuity pathway analysis. These candidates for interaction with FXR1P in the context of myogenesis now deserve further investigation. Notably, Hepatocyte growth factor (Hgf) mRNA is significantly upregulated in the absence of FXR1P (Table S1, Figure 1E and 1F, Figure S1) and is known to play an essential role in the migration and proliferation of myogenic cells [34]. Similarly, the Insulin-like growth factor 1 (Igf1) would be a relevant target of FXR1P in the muscle context, since Igf1 plays a key regulatory role in skeletal muscle development, as well as muscle fiber regeneration and hypertrophy [35]. Finally, Cyclin-dependent kinase 15 (Cdk15) mRNA which, contrary to p21 mRNA, is downregulated in Fxr1-deficient myoblasts (Table S1, Figure 1E and 1F, Figure S1) would be an interesting candidate for regulation of cell-cycle progression by FXR1P. In this case, FXR1P would stabilize Cdk15 mRNA via recognition of a yet unknown specific motif. Murine and human Cdk15 mRNA are not annotated in the AREsite database [36] and therefore do not seem to bear a canonical AU-rich element sequence in their 3′UTR. However, analysis of the 3672 nts long human Cdk15 mRNA using QGRS G-quadruplex mapping webtool reveals the presence of 8 putative G-quadruplex sequences (Table S3), with 2 putative G-quadruplex in the 3′UTR that represent binding sites for FXR1P. To ascertain the importance of FXR1P in the regulation of its putative mRNA targets newly identified in this study, it would be worth investigating the presence of ARE sequences, G-quadruplexes RNA structures in their 3′ untranslated region.
Adequate regulation of the balance between proliferation and cell cycle arrest of myoblasts is a crucial step during myogenesis. The decision to progress through a new division cycle appears primarily regulated before the G1 to S phase transition, with p21 upregulation playing an important role in this process by blocking the formation of proliferation-inducing Cyclin A/Cdk2-E2F complexes [37]. In this context, p21 gene undergoes extensive regulation, both at the transcriptional and posttranscriptional level. Our data do not support a transcriptional mechanism for the maintenance of elevated p21 mRNA levels in Fxr1-depleted muscle cells. Indeed, in myoblasts, p21 is under the sole transcriptional control of the myogenic transcription factor MyoD that activates its promoter [38]. Our microarray and quantitative RT-PCR analyses reveal that MyoD levels remain normal in Fxr1-deficient myoblasts (Figure 1E). Finally, in luciferase assays, Ren mRNA levels are increased when p21 mRNA G-quadruplex region is fused to its 3′UTR, even though this mRNA does not contain the endogenous promoter of p21/Cdkn1a gene (Figure 7B, Figure 8D). These evidences privilege an FXR1P-mediated posttranscriptional mechanism of regulation of p21 mRNA levels involving the binding of FXR1P.
In myoblasts, FXR1P long isoforms Isoe and Isof are most likely not playing a role in translational regulation, since they are detected in the nucleus and faintly in the cytoplasm but do not associate to polyribosomes [7], [8], [17], [39]. On the other hand, we cannot exclude a mechanism involving translational inhibition via binding of small or medium isoforms of FXR1P to p21 mRNA to another motif, which may be located in the central part of p21 mRNA 3′UTR (β fragment) that activates translation in the absence of FXR1P (Figure 7A, 7B). This would be consistent with the previously described role of FXR1P small isoform Isoa in translational control [20]. However, our data strongly support the fact that the FXR1P-dependant translational control of p21 mRNA is mainly regulated by FXR1P long isoforms, notably Isoe, via binding to a 3′UTR-located G-quadruplex motif (Figure 8).
To date, the G-quadruplex has been described to be a negative [31], [40] or positive [41] regulator of translation, and a zip-code for dendritic transport and synaptic localization [42] depending on its location on the mRNA (e.g. 5′UTR or 3′UTR) (for review see [43]). We report here an evolutionary conserved G-quadruplex motif as a novel RNA-binding motif present in a G-rich region of the distal portion of p21 mRNA 3′UTR. This motif, distinct from the classical ARE present in the proximal portion of the 3′UTR [30], appears nevertheless to control the stability of p21 mRNA. Indeed, when fused to the 3′UTR of Renilla luciferase, the G-quadruplex induces an increase in Renilla mRNA levels, (Figure 7B, Figure 8D) and this effect is potentiated by deletion of the G-quadruplex (Figure 8D). Collectively, these data argue that the G-quadruplex of p21 mRNA 3′UTR participates in the control of mRNA stability via a mechanism involving FXR1P. A few reports describe the involvement of 3′UTR-located G-rich stretches as downstream sequence elements (DSE) promoting polyadenylation and leading to increased stability of mRNA when located downstream the polyadenylation site [44], [45]. However, in the context of p21 mRNA, the G-quadruplex (position 918–955 nts) located upstream of p21 mRNA polyadenylation site (AAUAAA sequence in position 1309–1314 nts) could act as an upstream sequence elements (USE) promoting polyadenylation, as described for a U-rich sequence in Prothrombin mRNA 3′UTR [46]. An alternate mechanism would involve that FXR1P long isoforms drive degradation of p21 mRNA via recruitment of microRNAs and the RISC complex. RNA interference is well described to occur in the cytoplasm, but it was recently shown that small non-coding RNAs can associate with complementary pre-mRNA target both in the nucleus and in the cytoplasm, by binding to Ago2 [47]. The lattest is a key component of the RNA-Induced Silencing Complex (RISC) [47] and a well-known interactor of FXR1P in human cells [20], Xenopus oocytes [48], and in Drosophila [49], [50]. Interestingly, p21 mRNA 3′UTR contains an evolutionarily conserved binding site for miR-22 100 nts upstream of the G-quadruplex motif (Figure S3). This microRNA was recently shown to regulate p21 mRNA levels [47] and is bound in vivo by Ago2 [51]. In this context, Fxr1-depletion or p21 3′UTR G-quadruplex deletion could prevent recruitment of the RISC complex on p21 mRNA and contribute to increase its stability, ultimately leading to an accumulation of p21 mRNA and of the cognate protein.
In myoblasts, FXR1P is not the sole RNA-binding protein playing a key role in the regulation of p21 mRNA. Several reports demonstrate the importance of the proximal ARE of p21 mRNA 3′UTR- present in the α fragment- to control the stability of this mRNA. In myoblasts, the ARE-mediated stabilization of p21 mRNA is mediated by cooperative binding of HuR and hnRNPC1 [30], [52], while its decay is controled by KSRP [53]. Members of the hnRNPE family of proteins, PCBP1 and 2, control the central part of p21 3′UTR -the β fragment- [54]. Finally, another hnRNPE, PCBP4, binds and stabilizes the γ fragment [55], while we show in this study that binding of FXR1P to the G-quadruplex motif of p21 3′UTR-γ fragment destabilizes the mRNA. Here, we wish to propose a double system of regulation in which FXR1P and PCBP4 cooperate to regulate the levels of p21 using the distal 3′UTR while HuR, RNPC1 and KSRP use the ARE in the proximal part. These complex regulatory systems enable a fine-tuning of p21 mRNA levels, and our data indicate a prominent role for FXR1P as a modulator of p21 levels.
We report that, when FXR1P is depleted in the C2C12 cell line and in FSHD myoblasts, p21 levels increase (Figure 1, Figure 4). As a consequence, a subset of myoblasts becomes more permissive to cell cycle arrest, resulting in a reduced yield of myoblasts at each cycle of division (Figure 2, Figure 3). We also observed that the Cyclin-dependent kinase 15 (Cdk15) mRNA levels are decreased (Table S1; Figure 1E and 1F; Figure S1) it would be worth investigating whether its decreased levels also have an impact in this premature cell-cycle exit we observe in Fxr1-depleted myoblasts. Our data are in line with other studies in which overexpression of p21 in myoblasts is sufficient to trigger cell cycle exit, even in mitogenic medium [28], [56], [57]. In our study, p21 upregulation upon Fxr1-depletion causes cell cycle exit without onset of differentiation. Indeed, the levels of the myogenic factors MyoD and Myogenin remain normal, as assessed by microarray (Table S1) and quantitative RT-PCR (Figure 1F). Moreover, we did not observe spontaneous myoblasts fusion into myotubes in Fxr1-knockdown cultures in normal growth conditions, which would be indicative of premature differentiation (Davidovic & Bardoni, unpublished data). Nevertheless, it would be worth investigating in details the impact of Fxr1-knockdown on the differentiation of C2C12 myoblasts. Indeed, our data predict that premature cell cycle exit of myoblasts in the absence of FXR1P decreases the pool of myoblasts available for differentiation. This would directly contribute to explain the reduced musculature detected in Fxr1-KO mice [13] and in xfxr1-knockdown Xenopus [14] at early stages of embryogenesis and development.
The fact that p21 mRNA is an mRNA target for FXR1P Isoe has also crucial implications for the understanding of the pathophysiology of myopathies. Indeed, splicing defects of the FXR1 gene in FSHD myoblasts leads to reduced expression of the long FXR1P Isoe, the one that specifically binds p21 3′UTR. We and others have shown that FSHD myoblasts exhibited higher levels of p21 than controls, under normal growth conditions (this study and [58], [59]). It is now tempting to speculate that depletion in FXR1P Isoe directly participates to the physiopathology of FSHD, by causing p21-mediated premature arrest of the cell cycle in FSHD myoblasts. Ultimately, this may limit the pool of myoblasts available for regeneration of muscle fibers, inducing progressive muscle wasting in FSHD patients. This hypothesis is supported by a study which demonstrates that p21 is essential for normal myogenic progenitor cell function in regenerating skeletal muscle [60]. A similar scenario may be envisioned in the case of the mouse model of DM1 in which reduced expression of FXR1P Isoe was determined [12].
In conclusion, our study highlights for the first time the direct involvement of an RNA-binding protein, FXR1P, in a new pathway that regulates p21 levels to control myoblasts cell cycle exit. Perturbations of this pathway will have a strong impact in muscle development and implicates a new signal dependant on a 3′-UTR located G-quadruplex-RNA structure. In the future it will be important to explore the implication of FXR1P in pathophysiology of muscle disorders and the pleiotropic functions of FXR1P during myogenesis. Furthermore, our study opens new perspectives on the role of the other Fragile X related proteins in the control of cell cycle. Noteworthy, FMRP is known to recognize G-quadruplex mRNA structures and it would be tempting to speculate that FMRP could control p21-dependant cell cycle exit of neuronal progenitors during neurogenesis.
The C2C12 cell line, a subclone of the C2C4 murine myoblastic cell line [61], [62], was cultivated under confluence state in the conditions described by ATCC. C2C12 cells were transfected with siRNA targeting exon 14 or exon 6 of Fxr1 mRNA (see Table S1) and/or constructs using the Lipofectamine 2000 reagent (Invitrogen), according to the manufacturer's protocole. Control experiments were performed using commercially available control random siRNA of matching GC content (Invitrogen). Transfected cells were always analysed 48 hrs post transfection. mRNA decay experiments were performed by adding actinomycin D (Act D, 5 µg/mL) or 5,6-Dichlorobenzimidazole riboside (DRB, 50 µM) to culture medium for 0 to 8 hrs.
Human myoblasts derived from muscle biopsies of n = 3 FSHD patients and n = 3 controls of matching age and gender were described in [9]. The procedures to generate myoblasts derived from human muscle biopsies were agreed by the French Health Authorities (AFSSAPS). Myoblasts cultures were established as previously described [9].
Total RNA of C2C12 cells transfected with siFxr1 or siControl siRNAs was extracted using the RNeasy kit (Qiagen, Hilden, Germany). Integrity of RNA was assessed by using an Agilent BioAnalyser 2100 (Agilent Technologies) (RIN above 8). RNA samples were then labeled with Cy3 dye using the low RNA input QuickAmp kit (Agilent) as recommended by the supplier. 825 ng of labeled cRNA probe were hybridized on 8×60K high density SurePrint G3 gene expression mouse Agilent microarrays. Two biological replicates were performed for each experimental condition. The experimental data are deposited in the NCBI Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) under the series record number GSE40577. Normalization of microarray data was performed using the Limma package available from Bioconductor (http://www.bioconductor.org). Inter slide normalization was performed using the quantile methods. Means of ratios from all comparisons were calculated and B test analysis was performed. Differentially expressed genes were selected based on a B-value above 0. Data from expression microarrays were analyzed for enrichment in biological themes (Gene Ontology molecular function and canonical pathways) and build biological networks using Ingenuity Pathway Analysis software (http://www.ingenuity.com/) and Mediante (http://www.microarray.fr:8080/merge/index), an information system providing information about probes and data sets.
Total RNA was extracted from myoblasts using the RNeasy kit (Qiagen, Hilden, Germany) and a reverse transcription (RT) reaction was performed using the Superscript II RT-PCR system (Invitrogen, Carlsbad, California, USA) according to the manufacturers' protocol. RT products were subjected to polymerase chain reaction (PCR). All primers were designed using the Primer 3 software (Table S4). Standard RT-PCR was performed using the Promega PCR Master Kit (Promega, Madison, Wisconsin, USA). Real-time PCR reactions were carried out using the Syber Green I qPCR core Kit (Eurogentec, Liège, Belgium) in a LightCycler system (Roche, USA). The comparative threshold cycle (Ct) for the amplicons of each sample was determined by the LightCycler software and normalised to the corresponding Ct of TATA Box Binding Protein (TBP) mRNA for endogenous p21 mRNA levels, and to the Ct of Firefly luciferase in the case of Renilla luciferase mRNA assessment. Finally, the 2-ΔΔCt method [63] was used to analyse the relative changes in the various studied mRNAs between C2C12 myoblasts transfected with control siRNA (Invitrogen) or anti-Fxr1 siRNA (Invitrogen), or between FSHD myoblasts and controls (n = 3). Data were expressed as means ±SEM. Each assay was performed in triplicate with n = 3–4 independent replicates.
Cell extracts were analysed by western blotting as described previously [64], [65]. Previously described primary antibodies against FXR1P were polyclonal rabbit antibody #830 (1∶5,000) and monoclonal 3FX (1∶500), the latter also cross-reacting with FXR2P [7]. Anti-β-actin monoclonal antibody (Sigma) and anti-p21 polyclonal rabbit antibodies (Santa Cruz) were used respectively at 1∶10,000 and 1∶200. Digital acquisition of chemiluminescent signal was performed using the Las-3000 Imager system (Fujifilm). Quantitation of western-blot was performed using the ImageJ software and normalized to the β-actin signal.
Immunofluorescence was performed as described [9], using anti-FXR1P #830 polyclonal antibodies (1∶5,000; [8]) and anti-Ki67 monoclonal antibody (1∶100; Millipore). Secondary Alexa 594-coupled antibodies (Invitrogen, Carlsbad, California, USA) were used at 1∶250. After counterstaining with DAPI, coverslips were mounted on slides with anti-fading reagent and observed using a Zeiss Axioplan2 epifluorescence microscope equipped with a CoolSNAP HQ CCD cooled camera (Roper Scientific) or an Olympus FV10i confocal digital microscope. Micrographs were then analysed with ImageJ software.
For viability assessment 48 hrs post transfection with anti-Fxr1 and control siRNAs, both attached cells and culture supernatant were collected and then incubated in the presence of propidium iodide (PI, 50 µg/mL). The incorporation of PI in dead cells was then analysed with a FACScan instrument (Becton, Dickinson). MTT proliferation assay was used to determine the proliferation ability of the cells as recommended by the manufacturer (Sigma). For cell cycle distribution assessment, cells were fixed in 70% ethanol, treated with RNAseA (50 µg/mL), stained with PI (50 µg/mL) and their DNA content was assessed using FACS analysis. For synchronisation experiments, cells were treated with 500 nM of the cell blocker mimosine for 8 hrs. Release from cell cycle blockade was performed for 16 hrs in growth medium before FACS analysis.
Human FXR1P Isoe recombinant protein His-tagged in the C–terminus was produced in bacteria using the pET21a/FXR1 Isoe construct [21], as described [64]. The control RNA fragments used in this study: N19 (RNA sequence derived from FMR1 mRNA and containing a G-quadruplex forming structure) and N19Δ35 (N19 sequence in which the G-quadruplex is deleted) were cloned in pTL1 plasmid [31]. The various fragments from p21 cDNA were amplified by RT-PCR of C2C12 cDNAs and cloned in the pGemTEasy system (Promega) using the primers described in Table S1, as advised by the manufacturer. For filter binding assay, N19 or p21 constructs were in vitro transcribed using T7 RNA polymerase (Promega), the RNA products being labeled by cotranscriptional incorporation of [γ−32 P]-ATP. Labeled RNAs were purified on a 1% low-melting agarose gel (Ambion). Labeled RNAs (50,000 c.p.m., 4 fmol) were renatured for 10 min at 40°C in binding buffer (50 mM Tris–HCl (pH 7.4), 1 mM MgCl2, 1 mM EDTA, 150 mM KCl, 1 mM DTT). In the presence of 2 U/mL of RNase inhibitors (RNasin, Invitrogen), 0,1 mg/mL of Escherichia coli total tRNA and 0.01% BSA, radiolabeled RNA were incubated to increasing amounts of FXR1P protein. RNA–protein complexes were allowed to form for 10 min on ice, filtered through MF-membranes (0.45 HA, Millipore) and washed with 2 mL binding buffer. Filters were air-dried and Cerenkov counting was used to assess the levels of remaining radioactivity on filters. Data were plotted as percentage of total RNA bound versus the protein concentration and one-site binding curve was drawn using the Prism 4 software.
To isolate mRNAs associated with FXR1P in vivo, UV-crosslinking and immunoprecipitations (CLIP) were performed with extracts of C2C12 cells using a protocol adapted from [65] and the #830 polyclonal antibody directed against the C-terminus of FXR1P [8]. For each assay, 10 µg of polyclonal antiserum was used to immunoprecipitate 25×106 cells. An equivalent amount of unrelated rabbit IgGs (Sigma) were used as negative control. Approximately 1/20th of the homogenate and 1/4th of the immunoprecipitate were loaded on a 11% SDS–PAGE gel. Proteins transferred onto a 0.45 µm nitro-cellulose membrane were revealed using the 3FX antibody recognizing both FXR1P and FXR2P [8]. mRNAs were extracted from C2C12 input lysate and immunoprecipitates using Trizol reagent (Invitrogen) according to the manufacturer's protocole and subjected to reverse transcription (RT) using the SuperscriptScript III RT-PCR system (Invitrogen). RT products were subjected to polymerase chain reaction (PCR), using a PCR Master Kit (Promega) and primers detailed in Table S4 specific for p21, Myogenin, MyoD and β-Tubulin mouse cDNAs. The PCR program consisted in 10 min. of initial denaturation at 95°C followed by 35 cycles −30 s. at 95°C, 30 s. at 58°C, 30 s. at 72°C- and a final elongation step of 10 min at 72°C. PCR products were visualised on a 2% TAE agarose gel and amplicon size was verified using the 1 Kb+ DNA ladder (Invitrogen).
Luciferase assays were performed using the pSiCheck2 system (Promega) according to the manufacturer's protocole. Briefly, the various fragments from p21-3′UTR cDNA (α, β and γ) were excised from the pGemTEasy vectors using the NotI site and inserted downstream of the Renilla luciferase cDNA using the NotI site of the pSiCheck2 vector. C2C12 cells were co-transfected in 96-well plates with the siRNA control or against Fxr1 and pSiCheck2 constructs. Luciferase assays were performed 48 hrs post transfection using the DualGlow Luciferase Kit (Promega) according to the manufacturer's protocole.
pTL1/FXR1Isoe plasmid was cloned as described in [8]. The mutated version of this plasmid bearing 4 silent mutations in human FXR1 cDNA that impede recognition by siFxr1#1 was produced by site-directed mutagenesis using primers described in Table S4 and the QuickChange kit (Stratagene).
To compare numerical data, non-parametric Mann & Whitney test was used for small sample size (n<30) and a Student T-test was used when n>30. Wilcoxon non-parametric tests were used to assess significance of Renilla luciferase mRNA or activity levels variations between each fragment relative to the empty vector (arbitrarily set to 1). All statistical analysis and data graphs were performed with the Prism 4 software. Only significant differences are displayed on the graphs.
|
10.1371/journal.ppat.1006650 | Reduced accumulation of defective viral genomes contributes to severe outcome in influenza virus infected patients | Influenza A virus (IAV) infection can be severe or even lethal in toddlers, the elderly and patients with certain medical conditions. Infection of apparently healthy individuals nonetheless accounts for many severe disease cases and deaths, suggesting that viruses with increased pathogenicity co-circulate with pandemic or epidemic viruses. Looking for potential virulence factors, we have identified a polymerase PA D529N mutation detected in a fatal IAV case, whose introduction into two different recombinant virus backbones, led to reduced defective viral genomes (DVGs) production. This mutation conferred low induction of antiviral response in infected cells and increased pathogenesis in mice. To analyze the association between low DVGs production and pathogenesis in humans, we performed a genomic analysis of viruses isolated from a cohort of previously healthy individuals who suffered highly severe IAV infection requiring admission to Intensive Care Unit and patients with fatal outcome who additionally showed underlying medical conditions. These viruses were compared with those isolated from a cohort of mild IAV patients. Viruses with fewer DVGs accumulation were observed in patients with highly severe/fatal outcome than in those with mild disease, suggesting that low DVGs abundance constitutes a new virulence pathogenic marker in humans.
| Influenza A viruses are the causative agents of annual epidemics, sporadic zoonotic outbreaks and occasionally pandemics. Worldwide, acute respiratory infections caused by influenza A viruses continue to be one of the main causes of acute illness and death. The appearance in 2009 of a new H1N1 pandemic influenza strain reinforced the search to identify viral pathogenicity determinants for evaluation of the consequences of virus epidemics and potential pandemics for human health. Here we identify a new general virulence determinant found in a cohort of severe/fatal influenza virus-infected patients, a reduced accumulation of viral defective genomes. These molecules are incomplete viral genome segments that activate the innate immune response. This data will contribute to the prediction of influenza disease severity, to improved guidance of patient treatment and will enable the development of risk-based prevention strategies and policies.
| Acute respiratory infections are a main cause of severe illness and death worldwide. Influenza A virus (IAV) causes annual epidemics and occasional pandemics with potentially fatal outcome [1]; the global burden of seasonal influenza is >600 million cases, with 5 million cases of severe illness and up to 500,000 deaths each year. Annual influenza epidemics affect all age groups, although infants, the elderly, and individuals with underlying medical conditions are most severely affected. The existence of co-morbid conditions and the immune status may contribute to the patient outcome. Comorbid conditions for influenza include diabetes, chronic metabolic of lung, renal and cardiac diseases, immunosuppression, pregnancy and obesity [2–4]. Although comorbidities are found in many severe or even fatal cases, a considerable number of apparently healthy individuals nonetheless suffer severe infection, which suggests the coexistence of influenza strains with increased virulence among circulating viruses.
We previously tested this hypothesis by characterizing two IAV strains from the AH1N1 2009 pandemic (AH1N1pdm09), one isolated from a fatal case in a person with no known previously described comorbidities (F-IAV, fatal-case IAV) and the other from a patient with mild symptoms (M-IAV, mild-case IAV) [5]. F-IAV virulence was greater than that of M-IAV in cell culture, and showed higher pathogenicity in the in vivo murine model [5].
IAV virulence and pathogenesis are dependent on complex, multigenic mechanisms involving the viral genetic characteristics, the host conditions, the virus-host interactions, and the host response to the infection. Special effort has been previously made to identify virulence determinants of AH1N1pdm09 viruses and AH1N1 seasonal viruses. As a result, some residues distributed all over the genome have been associated to increased virulence of specific viral isolates [6, 7]. These determinants map mainly to the polymerase genes (PB1, PB2, PA), the hemagglutinin (HA), neuraminidase (NA), and non-structural protein 1 (NS1) (reviewed in [8]). Attenuating factors have also been described in cell culture [9, 10].
A proportion of influenza virus particles have defective genome RNAs (DVGs) due to internal deletions of viral segments [11–14]. The DVGs have the 3’ and 5’ ends of the parental RNA segments, and most have a single, large central deletion that generates viral RNAs of 180–1000 nucleotides [15–17]. DVGs have been found for all viral segments, but most derive from PB2, PB1 and PA RNAs [15, 17, 18]. The presence of DVGs potentiates the host response in cultured cells [19, 20] and in animal models and leads to attenuated infection [21], possibly through recognition of double-stranded RNA by receptors that activate antiviral signaling cascades [19] (reviewed in [22]).
Although our understanding of influenza pathogenesis is considerable, a potential general virulence determinant in humans remains to be identified. Here we used next-generation sequencing (NGS) to evaluate the role of defective genomes in the pathogenicity of influenza virus circulating in the human population. We found that the low amount of DVGs accumulated in tissue culture cells correlates with increased pathogenicity in mice both, in natural isolates or recombinant viruses. To corroborate these findings we performed a genomic analysis of viruses isolated from respiratory samples of a select cohort of IAV A(H1N1)pdm09-infected patients who suffered severe or fatal outcome, or from a cohort of infected patients with mild disease. The former viruses showed significantly less accumulation of DVGs than the latter. We suggest that low DVGs abundance has a major role in the severe outcome of IAV-infected patients.
We previously characterized two virus isolates, the F-IAV derived from a fatal case in a young person with no known previously described comorbidities and the M-IAV, from a young patient with mild symptoms.
Comparison of the F- and M-IAV consensus sequences showed nine amino acid changes in the F isolate [5] taken the A/California/04/2009 strain as reference. Changes in the viral polymerase PA (D529N) and PB2 (A221T) subunits and the surface glycoprotein HA (S127L) found in <1% of viruses circulating during 2009 influenza season were considered specific and potentially responsible for the difference in virulence [5]. Since DVGs are mainly produced by the viral polymerase, PA D529N and PB2 A 221T changes in the polymerase subunits were selected as putative responsible for low DVGs production and increased pathogenicity of F-IAV.
To further characterize the role of mutations in the F-IAV polymerase subunits as virulence determinants, we generated on the A/H1N1/California/04/09 virus backbone (CAL), recombinant influenza viruses bearing the combination of PA D529N and PB2 A221T mutations (PB2/PA mut; F-IAV-like polymerase), or viruses bearing single PA D529N (PA mut) or PB2 A221T (PB2 mut) mutations (Table 1). These viruses were grown in cell culture at a low moi (0.0001) to limit the production of DVGs and viruses obtained from this passage were used for the following assays.
The activity of the reconstituted polymerases of these three mutant viruses and the wild-type virus was first evaluated in a mini-replicon assay, which showed not significant differences (Fig 4A and 4B). Next, growth kinetics in cell culture showed that all recombinant viruses accumulates similar levels of viral proteins in a single cycle replication assay (Fig 4C) and replicated at a similar rate in a multiple cycle assay (Fig 4D).
To evaluate the pathogenesis induced by the different recombinant viruses carrying mutations present in F-IAV, we infected mice with various virus doses of CAL, PB2 mut, PA mut or PB2/PA mut viruses or with DMEM as control. Survival (Fig 7A) and body weight (S9 Fig) were monitored daily for two weeks and the lethal dose 50 (LD50) for each virus was determined. CAL, PB2 mut, PA mut and PB2/PA mut viruses showed an LD50 of 1x105, >106, 3 x 103 and 3.5 x 104, respectively (Fig 7B). These data confirmed that CAL virus is pathogenic in mice [1] and indicated that the PA D529N mutation greatly increased pathogenicity, suggesting a decisive effect of this polymerase change on disease outcome. PB2 mut, which accumulates high DVG levels, was greatly attenuated compared with the CAL virus. Reconstitution of the F-like polymerase-containing virus (PB2/PA mut) notably reduced DVGs accumulation (Fig 5A) and led to higher pathogenicity compared with PB2 mut virus (Fig 7A and 7B).
We additionally evaluated the pathogenicity of M mut and M-PA mut viruses in the same way indicated above and the results show that the introduction of the M1+M2 mutations on the CAL wt or PA mut backgrounds led to clear virus attenuation in mice, as the LD50 increased from 1 x 105 or 3 x 103, respectively, to >5 x 105 in both cases (Fig 7B). M-PA mut virus pathogenicity was greater than that of M mut virus, as indicated by body weight loss after sublethal infection (Fig 7C).
To further evaluate the pathogenicity differences among viruses bearing mutations present in F-IAV, mice were infected with a sub-lethal dose (103 pfu) of recombinant mutant viruses, or were mock-infected. Samples were recovered at several days post-infection (dpi) and viral titers determined in lung (Fig 8A). PA mut-infected mice showed the highest titers at 1, 2, and 4 dpi and the most rapid virus replication kinetics, whereas PB2 mut-infected mice showed the lowest titers at all times tested.
Next, we wanted to examine whether accumulation of DVGs play a role in the pathogenicity of influenza viruses in humans. Deep-sequencing of RNA from viruses isolated from respiratory samples of a select cohort of A(H1N1)pdm09-like virus-infected patients was performed. This cohort includes patients with highly severe outcome including severe pneumonia and acute respiratory distress syndrome (ARDS) requiring admission to the intensive care unit (ICU) with mechanical ventilation and endotracheal intubation for more than 96 hours (Fig 9A) from 2012–2013 influenza season in Spain. For more precise characterization of the intrinsic pathogenicity of these viruses, only those isolated from patients with no known comorbidities and aged under 65 and over 4 were included (Fig 9A). This cohort (n = 4) is a faithful representation (80–100%) of the total confirmed severe H1N1 influenza cases following these criteria in the 2012–2013 Spanish influenza season (n = 4–5) (S2 Table) [31]. Additionally, two viruses isolated from deceased patients who accomplished these criteria, but otherwise showed underlying medical conditions were evaluated; total severe/fatal cohort n = 6 (Fig 9A). These viruses were compared to those isolated from a cohort (n = 6) of mild IAV patients detected through the regular influenza surveillance system.
Influenza virus pathogenicity has been studied in depth for many years, and several amino acid changes have been identified as virulence determinants [1, 6–8, 32], however, a general pathogenicity determinant has not been characterized. Although DVGs have been described in natural animal infections [33, 34] and in pandemic AH1N1pdm09- and other respiratory virus-infected individuals [35, 36] their role in viral pathogenicity in patients has not been evaluated. The correlation between DVGs accumulation and severe disease observed in the severe/fatal- and mild-case viruses isolated from respiratory samples suggests that DVGs generation is a critical feature of severe influenza virus infection. We tested this hypothesis genetically by analyzing recombinant viruses bearing mutations identified in a fatal-outcome virus (F-IAV, mutations PB2 A221T and PA D529N) [5] or described elsewhere (mutations M1 S30N + M2 V86S) [20]. Whereas the non-pathogenic mutation PB2 A221T accumulates high levels of DVGs in cultured cells and is attenuated in mice (PB2 mut versus CAL), mutation PA D529N reduces DVGs accumulation alone or in combination with PB2 A221T change (Fig 5A) or with M1 S30N + M2 V86 S mutations (M-PA mut versus M mut) (Fig 5C). Moreover, recombinant viruses carrying PA D529N mutation displayed increased viral pathogenicity in the infected mice (PB2/PA mut and PA mut) (Fig 7).
Selection criteria used for determination of putative virulent markers found in 2009 in F-IAV has been updated, and the prevalence of these changes in H1N1 viruses circulating in humans was calculated using all sequences available in the NCBI Influenza Resource database from December 2009 to March 2016. This analysis showed that PA D529N change continues to be a rare mutation specific of F-IAV, but PB2 221 position admitted several changes including T. This data indicates that at position 221 of PB2 changes have been established in addition to the original A in the further circulating viruses after the influenza 2009 pandemic and this position might not be relevant for the increased pathogenicity of the F-IAV (Table 1). This data correlates with our findings, which indicate that PB2 A221T change is not responsible for the increased pathogenicity of the F-IAV.
Here we adopted a highly restrictive approach to evaluate the potential role of DVGs accumulation as a determinant of severe influenza disease in humans, although a contribution of the immune status of the patient to the infection outcome cannot be excluded. We analyzed influenza viruses from a select cohort of patients, under 65 years and over 4 years of age, who suffered severe or fatal influenza infection. These viruses were compared with those obtained from a cohort of mild infected patients detected through the regular surveillance system. Deep sequencing identified an inverse correlation between DVGs accumulation and virus pathogenicity in these cohorts (Fig 9). Those patients with no known comorbidities infected with low DVGs producer viruses developed a severe outcome, and those who showed comorbid conditions eventually died, indicating that both factors may contribute to the fatal outcome of the infection. The later data is in agreement with our previous study where we found that, besides being infected with the virulent F-IAV, which accumulates a reduced amount of DVGs, the infected patient presented a new confirmed genetic risk factor, a truncated form of Ccr5 gene [37, 38]. Thus, both the high virulence of the infecting virus, and the genetic risk factor, may have contributed to the fatal outcome of the patient.
None of the severe/fatal cohort viruses bore PA D529N change. This result suggest that low DVGs accumulation in severe/fatal-case viruses might be mediated by various changes other than D529N in PA polymerase subunit, or in distinct polymerase subunits (S3 Table), or in several viral proteins; this coincides with the complex, multigenic nature of the pathogenesis mechanisms [8]. Actually, changes in the polymerase [9] and in NS2 protein [10] modulate DVGs production in cell culture. Mutations in M1 and M2 [20] also modulate DVGs accumulation, probably by altering DVGs encapsidation in progeny virions [39, 40]. Therefore, reduced accumulation of DVGs constitutes a virulent factor itself, regardless the mutations responsible.
Regarding the possible mechanism of PA in low DVGs production, colleagues E. Fodor and G. Brownlee described some years ago that mutation A638R in the PA subunit was involved in the enormous accumulation of DGs in cell culture [9]. It was there described that this high DGs generation was due to an elongation defect by destabilization of RNA-PA subunit interaction, and that this phenomenon could be reverted by another mutation in the same PA polymerase subunit (C453R). The authors proposed a putative domain involved in elongation activity in the PA-C part of this polymerase subunit. We have localized PAD529N mutation in the viral polymerase structure (S14 Fig) [54], and they are spatially in the same PA-C domain nearby these previously described mutations. This data suggests that PA D529N may be involved in the same elongation process, although this activity would need to be further explored. In addition, Influenza A virus polymerase exists in different oligomerization state [55, 56], which allows different assembly of polymerase monomers during replication [57]. A new proposed mechanism suggests that the RNA depending RNA polymerase (vRdRP) dimer bound to viral RNA recruits another free polymerase dimer to form transient tetramer, which initiates replication of viral genome [58]. PA D529N mutation localizes on the interaction surface of this proposed dimerization model of the viral polymerase (S15 Fig). All these data suggest that this mutation (or other mutations on this same area) may alter the possible polymerase dimerization or, it may modify the stability of the RNA- polymerase complex, or additionally it may change the interaction with any cellular required factor, or any combination of them.
The role of DVGs in inducing the innate immune response has been demonstrated in cell culture [19] and animal models [21]. Studies in several animal models have led to proposals for the use of DG molecules or viruses modified to generate them in large numbers as protective elements against influenza virus infection [41]. It would be likely that the reduced activation of the antiviral state of cells infected with low DVGs producer viruses might induce an impaired immune response in infected animals or patients. The initially reduced antiviral state of the infected cells (Figs 3 and 6C) may allow the virus to grow uncontrolled for a short time and then this would induce an exacerbated immune response and inflammation which was described for severe IAV infected patients in the 2009 pandemic [42, 43]. In viral infections, neutrophils and alveolar macrophages play a key role in clearance and control of viral growth in infected lungs, thus their substantial migration to the site of inflammation in infected tissue contributes to overall viral pathogenicity. Depletion of alveolar macrophages leads to an uncontrolled viral proliferation and fatal outcome in infected mice [30, 44] while high influx of neutrophils in lungs and excessive inflammation has been associated with severe illness and high mortality rate in influenza infection [29]. The depletion of alveolar macrophages perfectly correlates with increased viral titers in lung tissue of PA mut and PB2/PA mut infected animals (Fig 8A and 8C), emphasizing again the crucial role of these cells in viral clearance and control of viral growth. In addition, an increased influx of neutrophils, which is associated with lethal influenza virus infection [29], has been observed in recombinant viruses carrying a mutation (PA mut and PB2/PA mut) (Fig 8B) present in a fatal-case virus, which produce reduced amount of DVGs (Fig 5A).
Although DVGs are produced in the polymerase segments (PB1, PB2 and PA), as previously described [13, 18], most viruses studied here also generate DVGs in other segments (Figs 5B and 10). The distribution of DVGs of the different cohorts is interestingly observed in some distinct viral segments, which is specially reduced (23-fold) for the polymerase subunits segments in the severe/fatal—associated case viruses compared to viruses isolated from mild cases (Fig 10 and S13 Fig). These findings suggest that the genomic combination of DVGs produced by each virus, and not only absolute numbers, may also contribute to pathogenicity.
In summary, we establish a significant association between low DVGs accumulation and an increase in severe or fatal outcome in human influenza virus infection (Fig 9); we provide genetic support for this association in infected cultured cells and in mice. In addition to the previous reports about the role of DVGs in natural animal infections here we present data indicating that a reduced accumulation of DVGs may be considered a new virulence marker for viral pathogenicity in humans. Evaluation of DVGs phenotype of circulating viruses might predict its potential to induce severe disease. Additional work is needed to define specific DVGs function and the mechanism by which they are produced in humans. These data could contribute substantially to the prediction of influenza disease severity and enable the development of risk-based prevention strategies and policies.
All procedures that required the use of animals complied with Spanish and European legislation concerning vivisection and the use of genetically modified organisms, and the protocols were approved by the National Center for Biotechnology Ethics Committees on Animal Experimentation and the Consejo Superior de Investigaciones Científicas (CSIC) Bioethics Subcommittee (permit 11014). We followed the guidelines included in the current Spanish legislation on protection for animals used in research and other scientific aims (RD 53/2013) and the current European Union Directive 2014/11/EU on protection for animals used in experimentation and other scientific aims.
The National Influenza Center in Madrid (Instituto de Salud Carlos III) and other regional laboratories from different Spanish regions, constituted the ReLEG network included in the Spanish Influenza Surveillance System (SISS), which monitored the circulation of influenza viruses each influenza season as a part of the countrywide surveillance. The viruses described in this study have been detected within this surveillance activity. An informed consent is not needed for this study since the patients from whom these viruses were isolated were anonymized.
Cell culture and mouse model experiments performed with recombinant viruses bearing mutations detected in a fatal case of IAV were performed in BSL2+ conditions and in a biological insulator in BSL2+ animal facilities, respectively.
Cell lines used in this study were canine kidney MDCK (ATCC), human lung epithelium A549 (ATCC) [45] and human embryonic kidney HEK293T (ATCC) cells [46].
Viruses used in the present study were selected according to the following criteria for the patients from whom the viruses were isolated. Patients included in the severe/fatal cohort were influenza A(H1N1)pdm09 confirmed cases, aged over 4 and under 65, admitted to intensive care unit (ICU) and with the information related to risk factors reflected in their clinical history. Those patients who developed highly severe disease did not display any comorbidities associated to severe influenza A virus infection, and deceased patients presented some comorbid conditions. Mild patients were influenza A(H1N1)pdm09 confirmed cases, aged over 4 and under 65, who developed mild disease and were monitored by sentinel medical centers included in the Spanish National Influenza Surveillance System. Selection of cases for this mild cohort was randomly made within the patients who meet the described above criteria and whose isolated virus were from the same Saint-Petersburg phylogenetic lineage as those from the severe/fatal cohort, accordingly to their HA gene. Respiratory samples were collected in virus transport medium (MEM, 200 U/ml penicillin, 200 μg/ml streptomycin, 200 U/ml mycostatin and 0.25% bovine serum albumin fraction V) and delivered to the Spanish National Influenza Center.
All influenza A viruses were isolated at the National Influenza Centre (CNM, ISCIII) from respiratory samples sent for virological characterization by the Spanish Influenza Surveillance System (SISS). The National Influenza Center in Madrid and other regional laboratories constitute the ReLEG network of the SISS, which monitors virus circulation each influenza season as a part of the countrywide surveillance. All viruses from either mild or severe/fatal patients were isolated from the upper respiratory tract, pharyngeal or nasopharyngeal exudates. Semi-confluent monolayers of MDCK cells were used for primary viral isolation. The monolayers were inoculated with 200 μl of homogenized samples; when the cytopathic effect was 75–100%, cultures were harvested and the supernatants used for virus stock generation by inoculation of MDCK cells.
Specific mutations were engineered in expression pCAGGS plasmids derived from the CAL strain using the QuickChangeTM site-directed mutagenesis kit (Stratagene) as recommended by the manufacturer. These materials were developed using the Licensed technology (Kawaoka-P99264US Recombinant Influenza viruses for vaccines and gene therapy).
The recombinant minireplicon assay was performed essentially as described [47]. In brief, cultures of HEK293T cells (2.5 × 106 cells) were transfected with a mixture of plasmids expressing the RNP components (pCMVPA, 2.5 ng; pCMVPB1, 12.5 ng; pCMVPB2, 12.5 ng; and pCMVNP, 500 ng) and a genomic plasmid expressing a viral RNA (vRNA)-like chloramphenicol acetyltransferase reporter gene (pHHCAT, 500 ng) using the calcium phosphate technique [48]. At 20 h posttransfection, total cell extracts were prepared and CAT accumulation determined by enzyme-linked immunosorbent assay (ELISA; GE Healthcare), using purified CAT enzyme as a standard.
Specific mutations were engineered in recombinant virus genomic pHH plasmids derived from the A/H1N1/California/04/2009 strain using the QuickChange site-directed mutagenesis kit (Stratagene) as recommended by the manufacturer. These materials were developed using the Licensed technology (Ref. Kawaoka-P99264US Recombinant Influenza viruses for vaccines and gene therapy). The mutations were rescued into infectious virus by standard techniques [49, 50]. Briefly, to rescue infectious virus from cDNAs, 105 293T HEK cells were cotransfected with a mixture of 12 plasmid DNAs (100 ng each) including (i) 8 genomic plasmids each carrying a viral segment cDNA under the control of the polI promoter and (ii) 4 expression plasmids encoding the three polymerase subunits and the NP. Transfection was carried out at with Lipofectamine Plus (Gibco) under the conditions recommended by the manufacturer. At 16 h post-transfection, transfected cells were plated onto an excess of MDCK cells. When a cytopathic effect was apparent, the supernatant medium was collected and used for plaque assay on MDCK cells to estimate viral titer. The supernatant was used to produce a viral stock at low multiplicity of infection. The identity of rescued mutant viruses was ascertained by sequencing of DNAs derived from the PA and PB2 RNA segments by reverse transcription-PCR (RT-PCR) amplification.
Supernatants of harvested cells inoculated with clinical samples or transfected with plasmids for the generation of recombinant viruses were titred by standard plaque assay. These first passages of every virus were used to inoculate fresh MDCK cells at indicated controlled low multiplicity of infection (0.0001 moi). All viral stocks used for further studies had a viral titer about 107 pfu/ml.
For virus purification, culture supernatants of 10−4 moi-infected MDCK cells were centrifuged (10 min, 3110 g, 4°C). Supernatants were sedimented through a sucrose step gradient (TNE buffer; 50% and 33% sucrose in 50 mM Tris-HCl, 100 mM NaCl, 5 mM EDTA, pH 7.5) (1 h, 274000 g, 4°C). The 50 to 33% interphase was collected, diluted in TNE buffer, and pelleted through a cushion of 33% sucrose in TNE (2 h, 112000 g, 4°C). For purification of viruses isolated from infected mouse lung, the previous protocol was used with modification of the sucrose gradient volume and rotors according to sample volume. For isolation, RNA in purified virions was treated with 0.5% SDS and 200μg/ml proteinase K in TNE (2 h, 37°C), followed by extraction with phenol-chloroform-isoamylalcohol-hydroxyquinolein and ethanol precipitation [51]. DNA was removed by DNAse treatment (Roche) according to manufacturer’s instructions. Quality and quantity of each RNA preparation was monitored using the Agilent 2100 Bioanalyzer (Agilent Technologies) (S2 Table). Appropriate amounts of each sample were analyzed by high-throughput sequencing (see below).
For the detection of viral and cellular proteins, total cell extracts were collected and Western blot assays were performed as described [49]. Antibodies to GAPDH, β-actin (both from Sigma), ISG56, Mx1 (both from Santa Cruz), NP and PB1 [57] were used.
Sequencing for previously described F- and M-IAV isolated during the 2009–2010 [5] influenza season was performed with the Illumina Genome Analyzer IIx using Illumina v5 sequencing chemistry and 36 bp single reads. Base calling was performed using Illumina pipeline version 1.7.0 (within SCS 2.8). All other viruses were sequenced with TruSeq v3 chemistry and 50 bp single reads on an Illumina HiSeq 2000. Total reads in each sample are indicated in S1 Table.
RT-PCR for the PA or PB2 segments was used to determine the presence of DVGs and their relative amount to the full-length RNA of the same viral segment. To detect full-length segments, internal primers were used to amplify a central fragment, which is not present in DVGs. To detect DVGs, the same RNA sample and external primers of the PA segment were used in a separate reaction. Short amplification times were applied for the detection of both, internal fragment corresponding to full-length segment and DVGs, to allow detection of RNAs up to 1000nt in length. The method is illustrated in Fig 2A. The reverse transcription reaction was performed for 30 min (42°C), followed by PCR (35 rounds at 94°C for 30 s, 53°/58°C for 40 s, and 68°C for 40 sec using the Titan-One RT-PCR kit (Roche)). As a specificity control, the primers and RT-PCR conditions for DVGs amplification were used with a plasmid encoding the full-length PA segment, and no amplification product was obtained (S1A Fig). Additionally, primers and amplification conditions for internal fragment corresponding to full-length segment were used with purified DVGs, and no amplification product was obtained (S1A Fig).
DVGs from cell cultured purified virions and from infected mice lung tissues were amplified by RT-PCR (Titan-One RT-PCR kit, Roche) as indicated above. Obtained products were amplified with Taq Polymerase (Sigma) for further cloning into pGEM-T vector using pGEM-T Easy kit (Promega). Selected clones were sequenced by Sanger method and obtained sequences were analyzed to confirm that they corresponded to defective genomes, including the 3′and 5′ends and a large internal deletion of the full-length viral segment.
Cultured human lung alveolar epithelial cells (A549) were infected at 10−3 pfu/cell (low multiplicity of infection; moi) or 3 pfu/cell (high moi). After 1 h, non-bound virus was rinsed off with acidic PBS (pH 5.3) and at various times (hours post-infection; hpi), cell supernatants were collected and used for virus titration by plaque assay.
To evaluate pathogenicity of the viruses, 5 female BALB/c AnNHsd mice (6–7 weeks old) were infected intranasally with different doses (106−102) of each of the recombinant influenza viruses described here, or were mock-infected. The animals were monitored daily for clinical signs and body weights for two weeks. For ethical reasons, mice were euthanized when they presented 25% body weight loss.
For the kinetics experiment, 5 female BALB/c mice (6–7 weeks old) were infected intranasally with a sublethal dose (103 pfu/50μl DMEM) of recombinant PA mut, PB2 mut or PB2/PA mut influenza viruses, or were mock-infected (50μl DMEM). Mice were euthanized at 1, 2, 4 and 7 dpi by CO2 inhalation and necropsied.
Lung samples were homogenized in PBS-0.3%-BSA-penicillin/ streptomycin (100 IU/ml) using an Electronic Douncer (IKA T10 basic, Workcenter). Lung samples were homogenized 1min at max speed at 4°C and debris was pelleted by centrifugation (2000 g, 5 min, 4°C). Viral titer was determined by standard plaque assay on MDCK cells.
Lungs samples were kept in RMPI medium at 4°C. Tissue samples were grinded into very small pieces prior to digestion with 180 μg/ml liberase (Roche) and 40 μg/ml DNase I (Roche) in RMPI medium for 30 minutes at 37°C. Digested fragments were filtrated with 40mm Nylon Cell Strainer (BD Falcon) and resuspended with RMPI- 3%FBS. After centrifugation of samples (1640 rmp, 5 min, 4°C) additional step for erythrocyte lysis were performed. Cell pellet was incubated for 1.5–2 min with 1ml of erythrocyte lysis buffer at RT. Lysis is inhibited by adding 9ml of PBS-5mM EDTA-3%FBS. Samples were then filtrated again with 40mm Nylon Cell Strainer and pelleted by centrifugation (1640rpm, 5 min, 4°C).
Cell suspensions were distributed in 96 wells plate and first incubated with violet LIVE/ DEAD due (Invitrogen) for 30 min at 4°C, washed 2 twice with PBS and then incubated for 15 min at 4°C with Fc block CD16 rat antibody. Samples were analysed by staining cell suspension with one or more fluorochrome-labelled antibodies mix in PBS for 30 minutes at 4°C in the dark. Antibodies used were PerCP-Cy5.5-conjugated CD45 (clon 30-F11) (BioLegend), PeCy7-conjugated CD11b (clon M1/70) (BioLegend), APC-conjugated CD11c (clon N418) (eBIOSCIENCE) and PE-conjugated Ly6G (clon 1A8) (BDBIOSCIENCE). Samples were than fixed by incubation with 4% formaldehyde for 20 min, pelleted by centrifugation (700 rmp, 5 min, 4°C) and washed once with PBS. After centrifugation (700 rmp, 5 min, 4°C), cells were resuspended in 0.4ml PBS and kept at 4°C O/N in the dark. Flow cytometric analysis was performed on a cytometer LSR II (BD Biosciences). Data were analyzed using CellQuestPro software.
Animal lungs were fixed in 10% formalin, embedded in paraffin, sliced into 5 mm thick sections, and stained with hematoxylin and eosin (H&E) by conventional methods.
UCSF Chimera 1.10.2 program was used for structural localization of specific mutations in the influenza virus polymerase. Structure of the single influenza A polymerase under accession 4WSB, or structure of the dimerized form of influenza A polymerase complex in Protein Data Bank (PDB) under accession 3J9B have been used as templates.
Student’s t test and two-way ANOVA were used as indicated in each experiments and Figures. A non-parametric Mann-Whitney U test was applied to estimate the statistical significance of differences between RPM. GraphPad Prism v. 5.00 (www.graphpad.com) was used for analysis.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.